Stereo matching is a pivotal task in computer vision,enabling precise depth estimation from stereo image pairs,yet it encounters challenges in regions with reflections,repetitive textures,or fine structures.In this pa...Stereo matching is a pivotal task in computer vision,enabling precise depth estimation from stereo image pairs,yet it encounters challenges in regions with reflections,repetitive textures,or fine structures.In this paper,we propose a Semantic-Guided Parallax Attention Stereo Matching Network(SGPASMnet)that can be trained in unsupervised manner,building upon the Parallax Attention Stereo Matching Network(PASMnet).Our approach leverages unsupervised learning to address the scarcity of ground truth disparity in stereo matching datasets,facilitating robust training across diverse scene-specific datasets and enhancing generalization.SGPASMnet incorporates two novel components:a Cross-Scale Feature Interaction(CSFI)block and semantic feature augmentation using a pre-trained semantic segmentation model,SegFormer,seamlessly embedded into the parallax attention mechanism.The CSFI block enables effective fusion ofmulti-scale features,integrating coarse and fine details to enhance disparity estimation accuracy.Semantic features,extracted by SegFormer,enrich the parallax attention mechanism by providing high-level scene context,significantly improving performance in ambiguous regions.Our model unifies these enhancements within a cohesive architecture,comprising semantic feature extraction,an hourglass network,a semantic-guided cascaded parallax attentionmodule,outputmodule,and a disparity refinement network.Evaluations on the KITTI2015 dataset demonstrate that our unsupervised method achieves a lower error rate compared to the original PASMnet,highlighting the effectiveness of our enhancements in handling complex scenes.By harnessing unsupervised learning without ground truth disparity needed,SGPASMnet offers a scalable and robust solution for accurate stereo matching,with superior generalization across varied real-world applications.展开更多
Scattering obscures information carried by waves by producing speckle patterns,posing a fundamental challenge across diverse fields,from microscopy to astronomy.Although machine learning has recently shown promise in ...Scattering obscures information carried by waves by producing speckle patterns,posing a fundamental challenge across diverse fields,from microscopy to astronomy.Although machine learning has recently shown promise in speckle analysis,existing approaches are hindered by their dependence on large,labeled datasets—a significant bottleneck in many real-world applications.Here,we introduce speckle unsupervised recognition and evaluation(SURE),a groundbreaking unsupervised learning strategy for speckle recognition that eliminates the need for labeled training data.SURE's distinctive feature lies in its ability to extract invariant features through advanced clustering algorithms to enable direct classification of high-level information from speckle patterns without prior knowledge.We demonstrate the transformative potential of this approach in two key applications:(1)a noninvasive glucose monitoring system that accurately tracks glucose concentrations over time without extensive calibration and(2)a high-throughput communication system using multimode fibers,achieving improved performance in dynamic environments.In addition,we showcase SURE's unprecedented capability to classify objects hidden behind obstacles using scattered light,further broadening its scope.This versatile approach opens new frontiers in biomedical diagnostics,quantum network decoupling,and remote sensing,unlocking a transformative new paradigm for extracting information from seemingly random optical patterns.展开更多
In modern industrial production,foreign object detection in complex environments is crucial to ensure product quality and production safety.Detection systems based on deep-learning image processing algorithms often fa...In modern industrial production,foreign object detection in complex environments is crucial to ensure product quality and production safety.Detection systems based on deep-learning image processing algorithms often face challenges with handling high-resolution images and achieving accurate detection against complex backgrounds.To address these issues,this study employs the PatchCore unsupervised anomaly detection algorithm combined with data augmentation techniques to enhance the system’s generalization capability across varying lighting conditions,viewing angles,and object scales.The proposed method is evaluated in a complex industrial detection scenario involving the bogie of an electric multiple unit(EMU).A dataset consisting of complex backgrounds,diverse lighting conditions,and multiple viewing angles is constructed to validate the performance of the detection system in real industrial environments.Experimental results show that the proposed model achieves an average area under the receiver operating characteristic curve(AUROC)of 0.92 and an average F1 score of 0.85.Combined with data augmentation,the proposed model exhibits improvements in AUROC by 0.06 and F1 score by 0.03,demonstrating enhanced accuracy and robustness for foreign object detection in complex industrial settings.In addition,the effects of key factors on detection performance are systematically analyzed,providing practical guidance for parameter selection in real industrial applications.展开更多
To avoid the laborious annotation process for dense prediction tasks like semantic segmentation,unsupervised domain adaptation(UDA)methods have been proposed to leverage the abundant annotations from a source domain,s...To avoid the laborious annotation process for dense prediction tasks like semantic segmentation,unsupervised domain adaptation(UDA)methods have been proposed to leverage the abundant annotations from a source domain,such as virtual world(e.g.,3D games),and adapt models to the target domain(the real world)by narrowing the domain discrepancies.However,because of the large domain gap,directly aligning two distinct domains without considering the intermediates leads to inefficient alignment and inferior adaptation.To address this issue,we propose a novel learnable evolutionary Category Intermediates(CIs)guided UDA model named Leci,which enables the information transfer between the two domains via two processes,i.e.,Distilling and Blending.Starting from a random initialization,the CIs learn shared category-wise semantics automatically from two domains in the Distilling process.Then,the learned semantics in the CIs are sent back to blend the domain features through a residual attentive fusion(RAF)module,such that the categorywise features of both domains shift towards each other.As the CIs progressively and consistently learn from the varying feature distributions during training,they are evolutionary to guide the model to achieve category-wise feature alignment.Experiments on both GTA5 and SYNTHIA datasets demonstrate Leci's superiority over prior representative methods.展开更多
Many fields,such as neuroscience,are experiencing the vast prolife ration of cellular data,underscoring the need fo r organizing and interpreting large datasets.A popular approach partitions data into manageable subse...Many fields,such as neuroscience,are experiencing the vast prolife ration of cellular data,underscoring the need fo r organizing and interpreting large datasets.A popular approach partitions data into manageable subsets via hierarchical clustering,but objective methods to determine the appropriate classification granularity are missing.We recently introduced a technique to systematically identify when to stop subdividing clusters based on the fundamental principle that cells must differ more between than within clusters.Here we present the corresponding protocol to classify cellular datasets by combining datadriven unsupervised hierarchical clustering with statistical testing.These general-purpose functions are applicable to any cellular dataset that can be organized as two-dimensional matrices of numerical values,including molecula r,physiological,and anatomical datasets.We demonstrate the protocol using cellular data from the Janelia MouseLight project to chara cterize morphological aspects of neurons.展开更多
Single-pixel imaging(SPI)is a prominent scattering media imaging technique that allows image transmission via one-dimensional detection under structured illumination,with applications spanning from long-range imaging ...Single-pixel imaging(SPI)is a prominent scattering media imaging technique that allows image transmission via one-dimensional detection under structured illumination,with applications spanning from long-range imaging to microscopy.Recent advancements leveraging deep learning(DL)have significantly improved SPI performance,especially at low compression ratios.However,most DL-based SPI methods proposed so far rely heavily on extensive labeled datasets for supervised training,which are often impractical in real-world scenarios.Here,we propose an unsupervised learningenabled label-free SPI method for resilient information transmission through unknown dynamic scattering media.Additionally,we introduce a physics-informed autoencoder framework to optimize encoding schemes,further enhancing image quality at low compression ratios.Simulation and experimental results demonstrate that high-efficiency data transmission with structural similarity exceeding 0.9 is achieved through challenging turbulent channels.Moreover,experiments demonstrate that in a 5 m underwater dynamic turbulent channel,USAF target imaging quality surpasses traditional methods by over 13 dB.The compressive encoded transmission of 720×720 resolution video exceeding 30 seconds with great fidelity is also successfully demonstrated.These preliminary results suggest that our proposed method opens up a new paradigm for resilient information transmission through unknown dynamic scattering media and holds potential for broader applications within many other scattering media imaging technologies.展开更多
Climate downscaling is used to transform large-scale meteorological data into small-scale data with enhanced detail,which finds wide applications in climate modeling,numerical weather forecasting,and renewable energy....Climate downscaling is used to transform large-scale meteorological data into small-scale data with enhanced detail,which finds wide applications in climate modeling,numerical weather forecasting,and renewable energy.Although deeplearning-based downscaling methods effectively capture the complex nonlinear mapping between meteorological data of varying scales,the supervised deep-learning-based downscaling methods suffer from insufficient high-resolution data in practice,and unsupervised methods struggle with accurately inferring small-scale specifics from limited large-scale inputs due to small-scale uncertainty.This article presents DualDS,a dual-learning framework utilizing a Generative Adversarial Network–based neural network and subgrid-scale auxiliary information for climate downscaling.Such a learning method is unified in a two-stream framework through up-and downsamplers,where the downsampler is used to simulate the information loss process during the upscaling,and the upsampler is used to reconstruct lost details and correct errors incurred during the upscaling.This dual learning strategy can eliminate the dependence on high-resolution ground truth data in the training process and refine the downscaling results by constraining the mapping process.Experimental findings demonstrate that DualDS is comparable to several state-of-the-art deep learning downscaling approaches,both qualitatively and quantitatively.Specifically,for a single surface-temperature data downscaling task,our method is comparable with other unsupervised algorithms with the same dataset,and we can achieve a 0.469 dB higher peak signal-to-noise ratio,0.017 higher structural similarity,0.08 lower RMSE,and the best correlation coefficient.In summary,this paper presents a novel approach to addressing small-scale uncertainty issues in unsupervised downscaling processes.展开更多
The original monitoring data from aero-engines possess characteristics such as high dimen-sionality,strong noise,and imbalance,which present substantial challenges to traditional anomalydetection methods.In response,t...The original monitoring data from aero-engines possess characteristics such as high dimen-sionality,strong noise,and imbalance,which present substantial challenges to traditional anomalydetection methods.In response,this paper proposes a method based on Fuzzy Fusion of variablesand Discriminant mapping of features for Clustering(FFD-Clustering)to detect anomalies in originalmonitoring data from Aircraft Communication Addressing and Reporting System(ACARS).Firstly,associated variables are fuzzily grouped to extract the underlying distribution characteristics and trendsfrom the data.Secondly,a multi-layer contrastive denoising-based feature Fusion Encoding Network(FEN)is designed for each variable group,which can construct representative features for each variablegroup through eliminating strong noise and complex interrelations between variables.Thirdly,a featureDiscriminative Mapping Network(DMN)based on reconstruction difference re-clustering is designed,which can distinguish dissimilar feature vectors when mapping representative features to a unified fea-ture space.Finally,the K-means clustering is used to detect the abnormal feature vectors in the unifiedfeature space.Additionally,the algorithm is capable of reconstructing identified abnormal vectors,thereby locating the abnormal variable groups.The performance of this algorithm was tested ontwo public datasets and real original monitoring data from four aero-engines'ACARS,demonstratingits superiority and application potential in aero-engine anomaly detection.展开更多
Time series anomaly detection is crucial in finance,healthcare,and industrial monitoring.However,traditional methods often face challenges when handling time series data,such as limited feature extraction capability,p...Time series anomaly detection is crucial in finance,healthcare,and industrial monitoring.However,traditional methods often face challenges when handling time series data,such as limited feature extraction capability,poor temporal dependency handling,and suboptimal real-time performance,sometimes even neglecting the temporal relationships between data.To address these issues and improve anomaly detection performance by better capturing temporal dependencies,we propose an unsupervised time series anomaly detection method,VLT-Anomaly.First,we enhance the Variational Autoencoder(VAE)module by redesigning its network structure to better suit anomaly detection through data reconstruction.We introduce hyperparameters to control the weight of the Kullback-Leibler(KL)divergence term in the Evidence Lower Bound(ELBO),thereby improving the encoder module’s decoupling and expressive power in the latent space,which yields more effective latent representations of the data.Next,we incorporate transformer and Long Short-Term Memory(LSTM)modules to estimate the long-term dependencies of the latent representations,capturing both forward and backward temporal relationships and performing time series forecasting.Finally,we compute the reconstruction error by averaging the predicted results and decoder reconstruction and detect anomalies through grid search for optimal threshold values.Experimental results demonstrate that the proposed method performs superior anomaly detection on multiple public time series datasets,effectively extracting complex time-related features and enabling efficient computation and real-time anomaly detection.It improves detection accuracy and robustness while reducing false positives and false negatives.展开更多
In recent years,the heterogeneous SAR image classification task of"training on simulated data and testing on measured data"has garnered increasing attention in the field of Synthetic Aperture Radar Automatic...In recent years,the heterogeneous SAR image classification task of"training on simulated data and testing on measured data"has garnered increasing attention in the field of Synthetic Aperture Radar Automatic Target Recognition(SAR-ATR).Although current mainstream domain adaptation methods have made significant breakthroughs in addressing domain shift problems,the escalating model complexity and task complexity have constrained their deployment in real-world applications.To tackle this challenge,this paper proposes a domain adaptation framework based on linear-kernel Maximum Mean Discrepancy(MMD),integrated with a near-zero-cost pseudo-label denoising technique leveraging deep feature clustering.Our method completely eliminates the need for data augmentation and handcrafted feature design,achieving endto-end pseudo-label self-training.Competitive performance is demonstrated across three typical scenarios in the SAMPLE dataset,with the highest accuracy of 98.65%achieved in ScenarioⅢ.The relevant code is available at:https://github.com/TheGreatTreatsby/SAMPLE_MMD.展开更多
Detecting sitting posture abnormalities in wheelchair users enables early identification of changes in their functional status.To date,this detection has relied on in-person observation by medical specialists.However,...Detecting sitting posture abnormalities in wheelchair users enables early identification of changes in their functional status.To date,this detection has relied on in-person observation by medical specialists.However,given the challenges faced by health specialists to carry out continuous monitoring,the development of an intelligent anomaly detection system is proposed.Unlike other authors,where they use supervised techniques,this work proposes using unsupervised techniques due to the advantages they offer.These advantages include the lack of prior labeling of data,and the detection of anomalies previously not contemplated,among others.In the present work,an individualized methodology consisting of two phases is developed:characterizing the normal sitting pattern and determining abnormal samples.An analysis has been carried out between different unsupervised techniques to study which ones are more suitable for postural diagnosis.It can be concluded,among other aspects,that the utilization of dimensionality reduction techniques leads to improved results.Moreover,the normality characterization phase is deemed necessary for enhancing the system’s learning capabilities.Additionally,employing an individualized approach to the model aids in capturing the particularities of the various pathologies present among subjects.展开更多
Supervised learning-based rail fastener anomaly detection models are limited by the scarcity of anomaly samples and perform poorly under data imbalance conditions.However,unsupervised anomaly detection methods based o...Supervised learning-based rail fastener anomaly detection models are limited by the scarcity of anomaly samples and perform poorly under data imbalance conditions.However,unsupervised anomaly detection methods based on diffusion models reduce the dependence on the number of anomalous samples but suffer from too many iterations and excessive smoothing of reconstructed images.In this work,we have established a rail fastener anomaly detection framework called Diff-Fastener,the diffusion model is introduced into the fastener detection task,half of the normal samples are converted into anomaly samples online in the model training stage,and One-Step denoising and canonical guided denoising paradigms are used instead of iterative denoising to improve the reconstruction efficiency of the model while solving the problem of excessive smoothing.DACM(Dilated Attention Convolution Module)is proposed in the middle layer of the reconstruction network to increase the detail information of the reconstructed image;meanwhile,Sparse-Skip connections are used instead of dense connections to reduce the computational load of themodel and enhance its scalability.Through exhaustive experiments onMVTec,VisA,and railroad fastener datasets,the results show that Diff-Fastener achieves 99.1%Image AUROC(Area Under the Receiver Operating Characteristic)and 98.9%Pixel AUROC on the railroad fastener dataset,which outperforms the existing models and achieves the best average score on MVTec and VisA datasets.Our research provides new ideas and directions in the field of anomaly detection for rail fasteners.展开更多
Patients in intensive care units(ICUs)require rapid critical decision making.Modern ICUs are data rich,where information streams from diverse sources.Machine learning(ML)and neural networks(NN)can leverage the rich da...Patients in intensive care units(ICUs)require rapid critical decision making.Modern ICUs are data rich,where information streams from diverse sources.Machine learning(ML)and neural networks(NN)can leverage the rich data for prognostication and clinical care.They can handle complex nonlinear relation-ships in medical data and have advantages over traditional predictive methods.A number of models are used:(1)Feedforward networks;and(2)Recurrent NN and convolutional NN to predict key outcomes such as mortality,length of stay in the ICU and the likelihood of complications.Current NN models exist in silos;their integration into clinical workflow requires greater transparency on data that are analyzed.Most models that are accurate enough for use in clinical care operate as‘black-boxes’in which the logic behind their decision making is opaque.Advan-ces have occurred to see through the opacity and peer into the processing of the black-box.In the near future ML is positioned to help in clinical decision making far beyond what is currently possible.Transparency is the first step toward vali-dation which is followed by clinical trust and adoption.In summary,NNs have the transformative ability to enhance predictive accuracy and improve patient management in ICUs.The concept should soon be turning into reality.展开更多
Many existing immune detection algorithms rely on a large volume of labeled self-training samples,which are often difficult to obtain in practical scenarios,thus limiting the training of detection models.Furthermore,n...Many existing immune detection algorithms rely on a large volume of labeled self-training samples,which are often difficult to obtain in practical scenarios,thus limiting the training of detection models.Furthermore,noise inherent in the samples can substantially degrade the detection accuracy of these algorithms.To overcome these challenges,we propose an immune generation algorithm that leverages clustering and a rebound mechanism for label propagation(LP-CRI).The dataset is randomly partitioned into multiple subsets,each of which undergoes clustering followed by label propagation and evaluation.The rebound mechanism assesses the model’s performance after propagation and determines whether to revert to its previous state,initiating a subsequent round of propagation to ensure stable and effective training.Experimental results demonstrate that the proposed method is both computationally efficient and easy to train,significantly enhancing detector performance and outperforming traditional immune detection algorithms.展开更多
Skin cancer,a severe health threat,can spread rapidly if undetected.Therefore,early detection can lead to an advanced and efficient diagnosis,thus reducing mortality.Unsupervised classification techniques analyse exte...Skin cancer,a severe health threat,can spread rapidly if undetected.Therefore,early detection can lead to an advanced and efficient diagnosis,thus reducing mortality.Unsupervised classification techniques analyse extensive skin image datasets,identifying patterns and anomalies without prior labelling,facilitating early detection and effective diagnosis and potentially saving lives.In this study,the authors aim to explore the potential of unsupervised learning methods in classifying different types of skin lesions in dermatoscopic images.The authors aim to bridge the gap in dermatological research by introducing innovative techniques that enhance image quality and improve feature extraction.To achieve this,enhanced super-resolution generative adversarial networks(ESRGAN)was fine-tuned to strengthen the resolution of skin lesion images,making critical features more visible.The authors extracted histogram features to capture essential colour characteristics and used the Davies-Bouldin index and silhouette score to determine optimal clusters.Fine-tuned k-means clustering with Euclidean distance in the histogram feature space achieved 87.77% and 90.5% test accuracies on the ISIC2019 and HAM10000 datasets,respectively.The unsupervised approach effectively categorises skin lesions,indicating that unsupervised learning can significantly advance dermatology by enabling early detection and classification without extensive manual annotation.展开更多
Reliable electricity infrastructure is critical for modern society,highlighting the importance of securing the stability of fundamental power electronic systems.However,as such systems frequently involve high-current ...Reliable electricity infrastructure is critical for modern society,highlighting the importance of securing the stability of fundamental power electronic systems.However,as such systems frequently involve high-current and high-voltage conditions,there is a greater likelihood of failures.Consequently,anomaly detection of power electronic systems holds great significance,which is a task that properly-designed neural networks can well undertake,as proven in various scenarios.Transformer-like networks are promising for such application,yet with its structure initially designed for different tasks,features extracted by beginning layers are often lost,decreasing detection performance.Also,such data-driven methods typically require sufficient anomalous data for training,which could be difficult to obtain in practice.Therefore,to improve feature utilization while achieving efficient unsupervised learning,a novel model,Densely-connected Decoder Transformer(DDformer),is proposed for unsupervised anomaly detection of power electronic systems in this paper.First,efficient labelfree training is achieved based on the concept of autoencoder with recursive-free output.An encoder-decoder structure with densely-connected decoder is then adopted,merging features from all encoder layers to avoid possible loss of mined features while reducing training difficulty.Both simulation and real-world experiments are conducted to validate the capabilities of DDformer,and the average FDR has surpassed baseline models,reaching 89.39%,93.91%,95.98%in different experiment setups respectively.展开更多
In the dynamic scene of autonomous vehicles,the depth estimation of monocular cameras often faces the problem of inaccurate edge depth estimation.To solve this problem,we propose an unsupervised monocular depth estima...In the dynamic scene of autonomous vehicles,the depth estimation of monocular cameras often faces the problem of inaccurate edge depth estimation.To solve this problem,we propose an unsupervised monocular depth estimation model based on edge enhancement,which is specifically aimed at the depth perception challenge in dynamic scenes.The model consists of two core networks:a deep prediction network and a motion estimation network,both of which adopt an encoder-decoder architecture.The depth prediction network is based on the U-Net structure of ResNet18,which is responsible for generating the depth map of the scene.The motion estimation network is based on the U-Net structure of Flow-Net,focusing on the motion estimation of dynamic targets.In the decoding stage of the motion estimation network,we innovatively introduce an edge-enhanced decoder,which integrates a convolutional block attention module(CBAM)in the decoding process to enhance the recognition ability of the edge features of moving objects.In addition,we also designed a strip convolution module to improve the model’s capture efficiency of discrete moving targets.To further improve the performance of the model,we propose a novel edge regularization method based on the Laplace operator,which effectively accelerates the convergence process of themodel.Experimental results on the KITTI and Cityscapes datasets show that compared with the current advanced dynamic unsupervised monocular model,the proposed model has a significant improvement in depth estimation accuracy and convergence speed.Specifically,the rootmean square error(RMSE)is reduced by 4.8%compared with the DepthMotion algorithm,while the training convergence speed is increased by 36%,which shows the superior performance of the model in the depth estimation task in dynamic scenes.展开更多
Background Zonal application maps are designed to represent field variability using key variables that can be translated into tailored management practices.For cotton,zonal maps for crop growth regulator(CGR)applicati...Background Zonal application maps are designed to represent field variability using key variables that can be translated into tailored management practices.For cotton,zonal maps for crop growth regulator(CGR)applications under variable-rate(VR)strategies are commonly based exclusively on vegetation indices(VIs)variability.However,VIs often saturate in dense crop vegetation areas,limiting their effectiveness in distinguishing variability in crop growth.This study aimed to compare unsupervised framework(UF)and supervised framework(SUF)approaches for generat-ing zonal application maps for CGR under VR conditions.During 2022-2023 agricultural seasons,an UF was employed to generate zonal maps based on locally collected field data on plant height of cotton,satellite imagery,soil texture,and phenology data.Subsequently,a SUF(based on historical data between 2020-2021 to 2022-2023 agricultural seasons)was developed to predict plant height using remote sensing and phenology data,aiming to replicate same zonal maps but without relying on direct field measurements of plant height.Both approaches were tested in three fields and on two different dates per field.Results The predictive model for plant height of SUF performed well,as indicated by the model metrics.However,when comparing zonal application maps for specific field-date combinations,the predicted plant height exhibited lower variability compared with field measurements.This led to variable compatibility between SUF maps,which utilized the model predictions,and the UF maps,which were based on the real field data.Fields characterized by much pronounced soil texture variability yielded the highest compatibility between the zonal application maps produced by both SUF and UF approaches.This was predominantly due to the greater consistency in estimating plant development patterns within these heterogeneous field environments.While VR application approach can facilitate product savings during the application operation,other key factors must be considered.These include the availability of specialized machinery required for this type of applications,as well as the inherent operational costs associated with applying a single CGR product which differs from the typical uniform rate applications that often integrate multi-ple inputs.Conclusion Predictive modeling shows promise for assisting in the creation of zonal application maps for VR of CGR applications.However,the degree of agreement with the actual variability in crop growth found in the field should be evaluated on a field-by-field basis.The SUF approach,which is based on plant heigh prediction,demonstrated potential for supporting the development of zonal application maps for VR of CGR applications.However,the degree to which this approach aligns itself with the actual variability in crop growth observed in the field may vary,necessi-tating field-by-field evaluation.展开更多
Defect detection based on computer vision is a critical component in ensuring the quality of industrial products.However,existing detection methods encounter several challenges in practical applications,including the ...Defect detection based on computer vision is a critical component in ensuring the quality of industrial products.However,existing detection methods encounter several challenges in practical applications,including the scarcity of labeled samples,limited adaptability of pre-trained models,and the data heterogeneity in distributed environments.To address these issues,this research proposes an unsupervised defect detection method,FLAME(Federated Learning with Adaptive Multi-Model Embeddings).The method comprises three stages:(1)Feature learning stage:this work proposes FADE(Feature-Adaptive Domain-Specific Embeddings),a framework employs Gaussian noise injection to simulate defective patterns and implements a feature discriminator for defect detection,thereby enhancing the pre-trained model’s industrial imagery representation capabilities.(2)Knowledge distillation co-training stage:a multi-model feature knowledge distillation mechanism is introduced.Through feature-level knowledge transfer between the global model and historical local models,the current local model is guided to learn better feature representations from the global model.The approach prevents local models from converging to local optima and mitigates performance degradation caused by data heterogeneity.(3)Model parameter aggregation stage:participating clients utilize weighted averaging aggregation to synthesize an updated global model,facilitating efficient knowledge consolidation.Experimental results demonstrate that FADE improves the average image-level Area under the Receiver Operating Characteristic Curve(AUROC)by 7.34%compared to methods directly utilizing pre-trained models.In federated learning environments,FLAME’s multi-model feature knowledge distillation mechanism outperforms the classic FedAvg algorithm by 2.34%in average image-level AUROC,while exhibiting superior convergence properties.展开更多
Side-channel analysis(SCA)has emerged as a research hotspot in the field of cryptanalysis.Among various approaches,unsupervised deep learning-based methods demonstrate powerful information extraction capabilities with...Side-channel analysis(SCA)has emerged as a research hotspot in the field of cryptanalysis.Among various approaches,unsupervised deep learning-based methods demonstrate powerful information extraction capabilities without requiring labeled data.However,existing unsupervised methods,particularly those represented by differential deep learning analysis(DDLA)and its improved variants,while overcoming the dependency on labeled data inherent in template analysis,still suffer from high time complexity and training costs when handling key byte difference comparisons.To address this issue,this paper introduces invariant information clustering(IIC)into SCA for the first time,and thus proposes a novel unsupervised learning-based SCA method,named IIC-SCA.By leveraging mutual information maximization techniques for automatic feature extraction of power leakage data,our approach achieves key recovery through a single training session,eliminating the prohibitive computational overhead of traditional methods that require separate training for all possible key bytes.Experimental results on the ASCAD dataset demonstrate successful key extraction using only 50000 training traces and 2000 attack traces.Furthermore,compared with DDLA,the proposed method reduces training time by approximately 93.40%and memory consumption by about 6.15%,significantly decreasing the temporal and resource costs of unsupervised SCA.This breakthrough provides new insights for developing low-cost,high-efficiency cryptographic attack methodologies.展开更多
基金supported by the National Natural Science Foundation of China,No.62301497the Science and Technology Research Program of Henan,No.252102211024the Key Research and Development Program of Henan,No.231111212000.
文摘Stereo matching is a pivotal task in computer vision,enabling precise depth estimation from stereo image pairs,yet it encounters challenges in regions with reflections,repetitive textures,or fine structures.In this paper,we propose a Semantic-Guided Parallax Attention Stereo Matching Network(SGPASMnet)that can be trained in unsupervised manner,building upon the Parallax Attention Stereo Matching Network(PASMnet).Our approach leverages unsupervised learning to address the scarcity of ground truth disparity in stereo matching datasets,facilitating robust training across diverse scene-specific datasets and enhancing generalization.SGPASMnet incorporates two novel components:a Cross-Scale Feature Interaction(CSFI)block and semantic feature augmentation using a pre-trained semantic segmentation model,SegFormer,seamlessly embedded into the parallax attention mechanism.The CSFI block enables effective fusion ofmulti-scale features,integrating coarse and fine details to enhance disparity estimation accuracy.Semantic features,extracted by SegFormer,enrich the parallax attention mechanism by providing high-level scene context,significantly improving performance in ambiguous regions.Our model unifies these enhancements within a cohesive architecture,comprising semantic feature extraction,an hourglass network,a semantic-guided cascaded parallax attentionmodule,outputmodule,and a disparity refinement network.Evaluations on the KITTI2015 dataset demonstrate that our unsupervised method achieves a lower error rate compared to the original PASMnet,highlighting the effectiveness of our enhancements in handling complex scenes.By harnessing unsupervised learning without ground truth disparity needed,SGPASMnet offers a scalable and robust solution for accurate stereo matching,with superior generalization across varied real-world applications.
基金supported by the National Natural Science Foundation of China(Grant Nos.11934011,12074339,62075194,U21A6006,62202418,and U21B2004)the National Key Research and Development Program of China(Grant Nos.2019YFA0308100,2023YFB2806000,and 2022YFA1204700)+4 种基金the Strategic Priority Research Program of Chinese Academy of Sciences(Grant No.XDB28000000)the Leading Innovation and Entrepreneurship Team in Zhejiang Province(Grant No.2020R01001)the Open Program of the State Key Laboratory of Advanced Optical Communication Systems and Networks at Shanghai Jiao Tong University(Grant No.2023GZKF024)the Fundamental Research Funds for the Central Universities,the Information Technology Center and State Key Lab of CAD&CG at the Zhejiang University,the Zhejiang Provincial Key Laboratory of Information Processing,Communication and Networking(IPCAN)the National Institutes of Health(NIH)(Grant Nos.R01GM127696,R01GM152633,R21GM142107,and R21CA269099)。
文摘Scattering obscures information carried by waves by producing speckle patterns,posing a fundamental challenge across diverse fields,from microscopy to astronomy.Although machine learning has recently shown promise in speckle analysis,existing approaches are hindered by their dependence on large,labeled datasets—a significant bottleneck in many real-world applications.Here,we introduce speckle unsupervised recognition and evaluation(SURE),a groundbreaking unsupervised learning strategy for speckle recognition that eliminates the need for labeled training data.SURE's distinctive feature lies in its ability to extract invariant features through advanced clustering algorithms to enable direct classification of high-level information from speckle patterns without prior knowledge.We demonstrate the transformative potential of this approach in two key applications:(1)a noninvasive glucose monitoring system that accurately tracks glucose concentrations over time without extensive calibration and(2)a high-throughput communication system using multimode fibers,achieving improved performance in dynamic environments.In addition,we showcase SURE's unprecedented capability to classify objects hidden behind obstacles using scattered light,further broadening its scope.This versatile approach opens new frontiers in biomedical diagnostics,quantum network decoupling,and remote sensing,unlocking a transformative new paradigm for extracting information from seemingly random optical patterns.
文摘In modern industrial production,foreign object detection in complex environments is crucial to ensure product quality and production safety.Detection systems based on deep-learning image processing algorithms often face challenges with handling high-resolution images and achieving accurate detection against complex backgrounds.To address these issues,this study employs the PatchCore unsupervised anomaly detection algorithm combined with data augmentation techniques to enhance the system’s generalization capability across varying lighting conditions,viewing angles,and object scales.The proposed method is evaluated in a complex industrial detection scenario involving the bogie of an electric multiple unit(EMU).A dataset consisting of complex backgrounds,diverse lighting conditions,and multiple viewing angles is constructed to validate the performance of the detection system in real industrial environments.Experimental results show that the proposed model achieves an average area under the receiver operating characteristic curve(AUROC)of 0.92 and an average F1 score of 0.85.Combined with data augmentation,the proposed model exhibits improvements in AUROC by 0.06 and F1 score by 0.03,demonstrating enhanced accuracy and robustness for foreign object detection in complex industrial settings.In addition,the effects of key factors on detection performance are systematically analyzed,providing practical guidance for parameter selection in real industrial applications.
基金Australian Research Council Project(FL-170100117).
文摘To avoid the laborious annotation process for dense prediction tasks like semantic segmentation,unsupervised domain adaptation(UDA)methods have been proposed to leverage the abundant annotations from a source domain,such as virtual world(e.g.,3D games),and adapt models to the target domain(the real world)by narrowing the domain discrepancies.However,because of the large domain gap,directly aligning two distinct domains without considering the intermediates leads to inefficient alignment and inferior adaptation.To address this issue,we propose a novel learnable evolutionary Category Intermediates(CIs)guided UDA model named Leci,which enables the information transfer between the two domains via two processes,i.e.,Distilling and Blending.Starting from a random initialization,the CIs learn shared category-wise semantics automatically from two domains in the Distilling process.Then,the learned semantics in the CIs are sent back to blend the domain features through a residual attentive fusion(RAF)module,such that the categorywise features of both domains shift towards each other.As the CIs progressively and consistently learn from the varying feature distributions during training,they are evolutionary to guide the model to achieve category-wise feature alignment.Experiments on both GTA5 and SYNTHIA datasets demonstrate Leci's superiority over prior representative methods.
基金supported in part by NIH grants R01NS39600,U01MH114829RF1MH128693(to GAA)。
文摘Many fields,such as neuroscience,are experiencing the vast prolife ration of cellular data,underscoring the need fo r organizing and interpreting large datasets.A popular approach partitions data into manageable subsets via hierarchical clustering,but objective methods to determine the appropriate classification granularity are missing.We recently introduced a technique to systematically identify when to stop subdividing clusters based on the fundamental principle that cells must differ more between than within clusters.Here we present the corresponding protocol to classify cellular datasets by combining datadriven unsupervised hierarchical clustering with statistical testing.These general-purpose functions are applicable to any cellular dataset that can be organized as two-dimensional matrices of numerical values,including molecula r,physiological,and anatomical datasets.We demonstrate the protocol using cellular data from the Janelia MouseLight project to chara cterize morphological aspects of neurons.
基金supported by the Natural Science Foundation of China Project(No.62525102).
文摘Single-pixel imaging(SPI)is a prominent scattering media imaging technique that allows image transmission via one-dimensional detection under structured illumination,with applications spanning from long-range imaging to microscopy.Recent advancements leveraging deep learning(DL)have significantly improved SPI performance,especially at low compression ratios.However,most DL-based SPI methods proposed so far rely heavily on extensive labeled datasets for supervised training,which are often impractical in real-world scenarios.Here,we propose an unsupervised learningenabled label-free SPI method for resilient information transmission through unknown dynamic scattering media.Additionally,we introduce a physics-informed autoencoder framework to optimize encoding schemes,further enhancing image quality at low compression ratios.Simulation and experimental results demonstrate that high-efficiency data transmission with structural similarity exceeding 0.9 is achieved through challenging turbulent channels.Moreover,experiments demonstrate that in a 5 m underwater dynamic turbulent channel,USAF target imaging quality surpasses traditional methods by over 13 dB.The compressive encoded transmission of 720×720 resolution video exceeding 30 seconds with great fidelity is also successfully demonstrated.These preliminary results suggest that our proposed method opens up a new paradigm for resilient information transmission through unknown dynamic scattering media and holds potential for broader applications within many other scattering media imaging technologies.
基金supported by the following funding bodies:the National Key Research and Development Program of China(Grant No.2020YFA0608000)National Science Foundation of China(Grant Nos.42075142,42375148,42125503+2 种基金42130608)FY-APP-2022.0609,Sichuan Province Key Tech nology Research and Development project(Grant Nos.2024ZHCG0168,2024ZHCG0176,2023YFG0305,2023YFG-0124,and 23ZDYF0091)the CUIT Science and Technology Innovation Capacity Enhancement Program project(Grant No.KYQN202305)。
文摘Climate downscaling is used to transform large-scale meteorological data into small-scale data with enhanced detail,which finds wide applications in climate modeling,numerical weather forecasting,and renewable energy.Although deeplearning-based downscaling methods effectively capture the complex nonlinear mapping between meteorological data of varying scales,the supervised deep-learning-based downscaling methods suffer from insufficient high-resolution data in practice,and unsupervised methods struggle with accurately inferring small-scale specifics from limited large-scale inputs due to small-scale uncertainty.This article presents DualDS,a dual-learning framework utilizing a Generative Adversarial Network–based neural network and subgrid-scale auxiliary information for climate downscaling.Such a learning method is unified in a two-stream framework through up-and downsamplers,where the downsampler is used to simulate the information loss process during the upscaling,and the upsampler is used to reconstruct lost details and correct errors incurred during the upscaling.This dual learning strategy can eliminate the dependence on high-resolution ground truth data in the training process and refine the downscaling results by constraining the mapping process.Experimental findings demonstrate that DualDS is comparable to several state-of-the-art deep learning downscaling approaches,both qualitatively and quantitatively.Specifically,for a single surface-temperature data downscaling task,our method is comparable with other unsupervised algorithms with the same dataset,and we can achieve a 0.469 dB higher peak signal-to-noise ratio,0.017 higher structural similarity,0.08 lower RMSE,and the best correlation coefficient.In summary,this paper presents a novel approach to addressing small-scale uncertainty issues in unsupervised downscaling processes.
基金co-supported by the National Science and Technology Major Project,China(No.J2019-I-0001-0001)the National Natural Science Foundation of China(No.52105545)。
文摘The original monitoring data from aero-engines possess characteristics such as high dimen-sionality,strong noise,and imbalance,which present substantial challenges to traditional anomalydetection methods.In response,this paper proposes a method based on Fuzzy Fusion of variablesand Discriminant mapping of features for Clustering(FFD-Clustering)to detect anomalies in originalmonitoring data from Aircraft Communication Addressing and Reporting System(ACARS).Firstly,associated variables are fuzzily grouped to extract the underlying distribution characteristics and trendsfrom the data.Secondly,a multi-layer contrastive denoising-based feature Fusion Encoding Network(FEN)is designed for each variable group,which can construct representative features for each variablegroup through eliminating strong noise and complex interrelations between variables.Thirdly,a featureDiscriminative Mapping Network(DMN)based on reconstruction difference re-clustering is designed,which can distinguish dissimilar feature vectors when mapping representative features to a unified fea-ture space.Finally,the K-means clustering is used to detect the abnormal feature vectors in the unifiedfeature space.Additionally,the algorithm is capable of reconstructing identified abnormal vectors,thereby locating the abnormal variable groups.The performance of this algorithm was tested ontwo public datasets and real original monitoring data from four aero-engines'ACARS,demonstratingits superiority and application potential in aero-engine anomaly detection.
基金support from the Fundamental Research Funds for Central Public Welfare Research Institutes(SK202324)the Central Guidance on Local Science and Technology Development Fund of Hebei Province(236Z0104G)+1 种基金the National Natural Science Foundation of China(62476078)the Geological Survey Project of China Geological Survey(G202304-2).
文摘Time series anomaly detection is crucial in finance,healthcare,and industrial monitoring.However,traditional methods often face challenges when handling time series data,such as limited feature extraction capability,poor temporal dependency handling,and suboptimal real-time performance,sometimes even neglecting the temporal relationships between data.To address these issues and improve anomaly detection performance by better capturing temporal dependencies,we propose an unsupervised time series anomaly detection method,VLT-Anomaly.First,we enhance the Variational Autoencoder(VAE)module by redesigning its network structure to better suit anomaly detection through data reconstruction.We introduce hyperparameters to control the weight of the Kullback-Leibler(KL)divergence term in the Evidence Lower Bound(ELBO),thereby improving the encoder module’s decoupling and expressive power in the latent space,which yields more effective latent representations of the data.Next,we incorporate transformer and Long Short-Term Memory(LSTM)modules to estimate the long-term dependencies of the latent representations,capturing both forward and backward temporal relationships and performing time series forecasting.Finally,we compute the reconstruction error by averaging the predicted results and decoder reconstruction and detect anomalies through grid search for optimal threshold values.Experimental results demonstrate that the proposed method performs superior anomaly detection on multiple public time series datasets,effectively extracting complex time-related features and enabling efficient computation and real-time anomaly detection.It improves detection accuracy and robustness while reducing false positives and false negatives.
文摘In recent years,the heterogeneous SAR image classification task of"training on simulated data and testing on measured data"has garnered increasing attention in the field of Synthetic Aperture Radar Automatic Target Recognition(SAR-ATR).Although current mainstream domain adaptation methods have made significant breakthroughs in addressing domain shift problems,the escalating model complexity and task complexity have constrained their deployment in real-world applications.To tackle this challenge,this paper proposes a domain adaptation framework based on linear-kernel Maximum Mean Discrepancy(MMD),integrated with a near-zero-cost pseudo-label denoising technique leveraging deep feature clustering.Our method completely eliminates the need for data augmentation and handcrafted feature design,achieving endto-end pseudo-label self-training.Competitive performance is demonstrated across three typical scenarios in the SAMPLE dataset,with the highest accuracy of 98.65%achieved in ScenarioⅢ.The relevant code is available at:https://github.com/TheGreatTreatsby/SAMPLE_MMD.
基金FEDER/Ministry of Science and Innovation-State Research Agency/Project PID2020-112667RB-I00 funded by MCIN/AEI/10.13039/501100011033the Basque Government,IT1726-22+2 种基金by the predoctoral contracts PRE_2022_2_0022 and EP_2023_1_0015 of the Basque Governmentpartially supported by the Italian MIUR,PRIN 2020 Project“COMMON-WEARS”,N.2020HCWWLP,CUP:H23C22000230005co-funding from Next Generation EU,in the context of the National Recovery and Resilience Plan,through the Italian MUR,PRIN 2022 Project”COCOWEARS”(A framework for COntinuum COmputing WEARable Systems),N.2022T2XNJE,CUP:H53D23003640006.
文摘Detecting sitting posture abnormalities in wheelchair users enables early identification of changes in their functional status.To date,this detection has relied on in-person observation by medical specialists.However,given the challenges faced by health specialists to carry out continuous monitoring,the development of an intelligent anomaly detection system is proposed.Unlike other authors,where they use supervised techniques,this work proposes using unsupervised techniques due to the advantages they offer.These advantages include the lack of prior labeling of data,and the detection of anomalies previously not contemplated,among others.In the present work,an individualized methodology consisting of two phases is developed:characterizing the normal sitting pattern and determining abnormal samples.An analysis has been carried out between different unsupervised techniques to study which ones are more suitable for postural diagnosis.It can be concluded,among other aspects,that the utilization of dimensionality reduction techniques leads to improved results.Moreover,the normality characterization phase is deemed necessary for enhancing the system’s learning capabilities.Additionally,employing an individualized approach to the model aids in capturing the particularities of the various pathologies present among subjects.
基金funded by the National Natural Science Foundation of China,grant number 52272385 and 52475085.
文摘Supervised learning-based rail fastener anomaly detection models are limited by the scarcity of anomaly samples and perform poorly under data imbalance conditions.However,unsupervised anomaly detection methods based on diffusion models reduce the dependence on the number of anomalous samples but suffer from too many iterations and excessive smoothing of reconstructed images.In this work,we have established a rail fastener anomaly detection framework called Diff-Fastener,the diffusion model is introduced into the fastener detection task,half of the normal samples are converted into anomaly samples online in the model training stage,and One-Step denoising and canonical guided denoising paradigms are used instead of iterative denoising to improve the reconstruction efficiency of the model while solving the problem of excessive smoothing.DACM(Dilated Attention Convolution Module)is proposed in the middle layer of the reconstruction network to increase the detail information of the reconstructed image;meanwhile,Sparse-Skip connections are used instead of dense connections to reduce the computational load of themodel and enhance its scalability.Through exhaustive experiments onMVTec,VisA,and railroad fastener datasets,the results show that Diff-Fastener achieves 99.1%Image AUROC(Area Under the Receiver Operating Characteristic)and 98.9%Pixel AUROC on the railroad fastener dataset,which outperforms the existing models and achieves the best average score on MVTec and VisA datasets.Our research provides new ideas and directions in the field of anomaly detection for rail fasteners.
文摘Patients in intensive care units(ICUs)require rapid critical decision making.Modern ICUs are data rich,where information streams from diverse sources.Machine learning(ML)and neural networks(NN)can leverage the rich data for prognostication and clinical care.They can handle complex nonlinear relation-ships in medical data and have advantages over traditional predictive methods.A number of models are used:(1)Feedforward networks;and(2)Recurrent NN and convolutional NN to predict key outcomes such as mortality,length of stay in the ICU and the likelihood of complications.Current NN models exist in silos;their integration into clinical workflow requires greater transparency on data that are analyzed.Most models that are accurate enough for use in clinical care operate as‘black-boxes’in which the logic behind their decision making is opaque.Advan-ces have occurred to see through the opacity and peer into the processing of the black-box.In the near future ML is positioned to help in clinical decision making far beyond what is currently possible.Transparency is the first step toward vali-dation which is followed by clinical trust and adoption.In summary,NNs have the transformative ability to enhance predictive accuracy and improve patient management in ICUs.The concept should soon be turning into reality.
基金granted by Key Project of Beijing Municipal Social Science Foundation(No.15ZHA004)Key Project of Beijing Municipal Social Science Foundation and Beijing Municipal Education Commission Social Science Program(No.SZ20231123202).
文摘Many existing immune detection algorithms rely on a large volume of labeled self-training samples,which are often difficult to obtain in practical scenarios,thus limiting the training of detection models.Furthermore,noise inherent in the samples can substantially degrade the detection accuracy of these algorithms.To overcome these challenges,we propose an immune generation algorithm that leverages clustering and a rebound mechanism for label propagation(LP-CRI).The dataset is randomly partitioned into multiple subsets,each of which undergoes clustering followed by label propagation and evaluation.The rebound mechanism assesses the model’s performance after propagation and determines whether to revert to its previous state,initiating a subsequent round of propagation to ensure stable and effective training.Experimental results demonstrate that the proposed method is both computationally efficient and easy to train,significantly enhancing detector performance and outperforming traditional immune detection algorithms.
文摘Skin cancer,a severe health threat,can spread rapidly if undetected.Therefore,early detection can lead to an advanced and efficient diagnosis,thus reducing mortality.Unsupervised classification techniques analyse extensive skin image datasets,identifying patterns and anomalies without prior labelling,facilitating early detection and effective diagnosis and potentially saving lives.In this study,the authors aim to explore the potential of unsupervised learning methods in classifying different types of skin lesions in dermatoscopic images.The authors aim to bridge the gap in dermatological research by introducing innovative techniques that enhance image quality and improve feature extraction.To achieve this,enhanced super-resolution generative adversarial networks(ESRGAN)was fine-tuned to strengthen the resolution of skin lesion images,making critical features more visible.The authors extracted histogram features to capture essential colour characteristics and used the Davies-Bouldin index and silhouette score to determine optimal clusters.Fine-tuned k-means clustering with Euclidean distance in the histogram feature space achieved 87.77% and 90.5% test accuracies on the ISIC2019 and HAM10000 datasets,respectively.The unsupervised approach effectively categorises skin lesions,indicating that unsupervised learning can significantly advance dermatology by enabling early detection and classification without extensive manual annotation.
基金supported in part by the National Natural Science Foundation of China under Grant 62303090,U2330206in part by the Postdoctoral Science Foundation of China under Grant 2023M740516+1 种基金in part by the Natural Science Foundation of Sichuan Province under Grant 2024NSFSC1480in part by the New Cornerstone Science Foundation through the XPLORER PRIZE.
文摘Reliable electricity infrastructure is critical for modern society,highlighting the importance of securing the stability of fundamental power electronic systems.However,as such systems frequently involve high-current and high-voltage conditions,there is a greater likelihood of failures.Consequently,anomaly detection of power electronic systems holds great significance,which is a task that properly-designed neural networks can well undertake,as proven in various scenarios.Transformer-like networks are promising for such application,yet with its structure initially designed for different tasks,features extracted by beginning layers are often lost,decreasing detection performance.Also,such data-driven methods typically require sufficient anomalous data for training,which could be difficult to obtain in practice.Therefore,to improve feature utilization while achieving efficient unsupervised learning,a novel model,Densely-connected Decoder Transformer(DDformer),is proposed for unsupervised anomaly detection of power electronic systems in this paper.First,efficient labelfree training is achieved based on the concept of autoencoder with recursive-free output.An encoder-decoder structure with densely-connected decoder is then adopted,merging features from all encoder layers to avoid possible loss of mined features while reducing training difficulty.Both simulation and real-world experiments are conducted to validate the capabilities of DDformer,and the average FDR has surpassed baseline models,reaching 89.39%,93.91%,95.98%in different experiment setups respectively.
基金funded by the Yangtze River Delta Science and Technology Innovation Community Joint Research Project(2023CSJGG1600)the Natural Science Foundation of Anhui Province(2208085MF173)Wuhu“ChiZhu Light”Major Science and Technology Project(2023ZD01,2023ZD03).
文摘In the dynamic scene of autonomous vehicles,the depth estimation of monocular cameras often faces the problem of inaccurate edge depth estimation.To solve this problem,we propose an unsupervised monocular depth estimation model based on edge enhancement,which is specifically aimed at the depth perception challenge in dynamic scenes.The model consists of two core networks:a deep prediction network and a motion estimation network,both of which adopt an encoder-decoder architecture.The depth prediction network is based on the U-Net structure of ResNet18,which is responsible for generating the depth map of the scene.The motion estimation network is based on the U-Net structure of Flow-Net,focusing on the motion estimation of dynamic targets.In the decoding stage of the motion estimation network,we innovatively introduce an edge-enhanced decoder,which integrates a convolutional block attention module(CBAM)in the decoding process to enhance the recognition ability of the edge features of moving objects.In addition,we also designed a strip convolution module to improve the model’s capture efficiency of discrete moving targets.To further improve the performance of the model,we propose a novel edge regularization method based on the Laplace operator,which effectively accelerates the convergence process of themodel.Experimental results on the KITTI and Cityscapes datasets show that compared with the current advanced dynamic unsupervised monocular model,the proposed model has a significant improvement in depth estimation accuracy and convergence speed.Specifically,the rootmean square error(RMSE)is reduced by 4.8%compared with the DepthMotion algorithm,while the training convergence speed is increased by 36%,which shows the superior performance of the model in the depth estimation task in dynamic scenes.
文摘Background Zonal application maps are designed to represent field variability using key variables that can be translated into tailored management practices.For cotton,zonal maps for crop growth regulator(CGR)applications under variable-rate(VR)strategies are commonly based exclusively on vegetation indices(VIs)variability.However,VIs often saturate in dense crop vegetation areas,limiting their effectiveness in distinguishing variability in crop growth.This study aimed to compare unsupervised framework(UF)and supervised framework(SUF)approaches for generat-ing zonal application maps for CGR under VR conditions.During 2022-2023 agricultural seasons,an UF was employed to generate zonal maps based on locally collected field data on plant height of cotton,satellite imagery,soil texture,and phenology data.Subsequently,a SUF(based on historical data between 2020-2021 to 2022-2023 agricultural seasons)was developed to predict plant height using remote sensing and phenology data,aiming to replicate same zonal maps but without relying on direct field measurements of plant height.Both approaches were tested in three fields and on two different dates per field.Results The predictive model for plant height of SUF performed well,as indicated by the model metrics.However,when comparing zonal application maps for specific field-date combinations,the predicted plant height exhibited lower variability compared with field measurements.This led to variable compatibility between SUF maps,which utilized the model predictions,and the UF maps,which were based on the real field data.Fields characterized by much pronounced soil texture variability yielded the highest compatibility between the zonal application maps produced by both SUF and UF approaches.This was predominantly due to the greater consistency in estimating plant development patterns within these heterogeneous field environments.While VR application approach can facilitate product savings during the application operation,other key factors must be considered.These include the availability of specialized machinery required for this type of applications,as well as the inherent operational costs associated with applying a single CGR product which differs from the typical uniform rate applications that often integrate multi-ple inputs.Conclusion Predictive modeling shows promise for assisting in the creation of zonal application maps for VR of CGR applications.However,the degree of agreement with the actual variability in crop growth found in the field should be evaluated on a field-by-field basis.The SUF approach,which is based on plant heigh prediction,demonstrated potential for supporting the development of zonal application maps for VR of CGR applications.However,the degree to which this approach aligns itself with the actual variability in crop growth observed in the field may vary,necessi-tating field-by-field evaluation.
基金supported in part by the National Natural Science Foundation of China under Grants 32171909,52205254,32301704the Guangdong Basic and Applied Basic Research Foundation under Grants 2023A1515011255,2024A1515010199+1 种基金the Scientific Research Projects of Universities in Guangdong Province under Grants 2024ZDZX1042,2024ZDZX3057the Ji-Hua Laboratory Open Project under Grant X220931UZ230.
文摘Defect detection based on computer vision is a critical component in ensuring the quality of industrial products.However,existing detection methods encounter several challenges in practical applications,including the scarcity of labeled samples,limited adaptability of pre-trained models,and the data heterogeneity in distributed environments.To address these issues,this research proposes an unsupervised defect detection method,FLAME(Federated Learning with Adaptive Multi-Model Embeddings).The method comprises three stages:(1)Feature learning stage:this work proposes FADE(Feature-Adaptive Domain-Specific Embeddings),a framework employs Gaussian noise injection to simulate defective patterns and implements a feature discriminator for defect detection,thereby enhancing the pre-trained model’s industrial imagery representation capabilities.(2)Knowledge distillation co-training stage:a multi-model feature knowledge distillation mechanism is introduced.Through feature-level knowledge transfer between the global model and historical local models,the current local model is guided to learn better feature representations from the global model.The approach prevents local models from converging to local optima and mitigates performance degradation caused by data heterogeneity.(3)Model parameter aggregation stage:participating clients utilize weighted averaging aggregation to synthesize an updated global model,facilitating efficient knowledge consolidation.Experimental results demonstrate that FADE improves the average image-level Area under the Receiver Operating Characteristic Curve(AUROC)by 7.34%compared to methods directly utilizing pre-trained models.In federated learning environments,FLAME’s multi-model feature knowledge distillation mechanism outperforms the classic FedAvg algorithm by 2.34%in average image-level AUROC,while exhibiting superior convergence properties.
文摘Side-channel analysis(SCA)has emerged as a research hotspot in the field of cryptanalysis.Among various approaches,unsupervised deep learning-based methods demonstrate powerful information extraction capabilities without requiring labeled data.However,existing unsupervised methods,particularly those represented by differential deep learning analysis(DDLA)and its improved variants,while overcoming the dependency on labeled data inherent in template analysis,still suffer from high time complexity and training costs when handling key byte difference comparisons.To address this issue,this paper introduces invariant information clustering(IIC)into SCA for the first time,and thus proposes a novel unsupervised learning-based SCA method,named IIC-SCA.By leveraging mutual information maximization techniques for automatic feature extraction of power leakage data,our approach achieves key recovery through a single training session,eliminating the prohibitive computational overhead of traditional methods that require separate training for all possible key bytes.Experimental results on the ASCAD dataset demonstrate successful key extraction using only 50000 training traces and 2000 attack traces.Furthermore,compared with DDLA,the proposed method reduces training time by approximately 93.40%and memory consumption by about 6.15%,significantly decreasing the temporal and resource costs of unsupervised SCA.This breakthrough provides new insights for developing low-cost,high-efficiency cryptographic attack methodologies.
