The current deep learning models for braced excavation cannot predict deformation from the beginning of excavation due to the need for a substantial corpus of sufficient historical data for training purposes.To addres...The current deep learning models for braced excavation cannot predict deformation from the beginning of excavation due to the need for a substantial corpus of sufficient historical data for training purposes.To address this issue,this study proposes a transfer learning model based on a sequence-to-sequence twodimensional(2D)convolutional long short-term memory neural network(S2SCL2D).The model can use the existing data from other adjacent similar excavations to achieve wall deflection prediction once a limited amount of monitoring data from the target excavation has been recorded.In the absence of adjacent excavation data,numerical simulation data from the target project can be employed instead.A weight update strategy is proposed to improve the prediction accuracy by integrating the stochastic gradient masking with an early stopping mechanism.To illustrate the proposed methodology,an excavation project in Hangzhou,China is adopted.The proposed deep transfer learning model,which uses either adjacent excavation data or numerical simulation data as the source domain,shows a significant improvement in performance when compared to the non-transfer learning model.Using the simulation data from the target project even leads to better prediction performance than using the actual monitoring data from other adjacent excavations.The results demonstrate that the proposed model can reasonably predict the deformation with limited data from the target project.展开更多
In radiology,magnetic resonance imaging(MRI)is an essential diagnostic tool that provides detailed images of a patient’s anatomical and physiological structures.MRI is particularly effective for detecting soft tissue...In radiology,magnetic resonance imaging(MRI)is an essential diagnostic tool that provides detailed images of a patient’s anatomical and physiological structures.MRI is particularly effective for detecting soft tissue anomalies.Traditionally,radiologists manually interpret these images,which can be labor-intensive and time-consuming due to the vast amount of data.To address this challenge,machine learning,and deep learning approaches can be utilized to improve the accuracy and efficiency of anomaly detection in MRI scans.This manuscript presents the use of the Deep AlexNet50 model for MRI classification with discriminative learning methods.There are three stages for learning;in the first stage,the whole dataset is used to learn the features.In the second stage,some layers of AlexNet50 are frozen with an augmented dataset,and in the third stage,AlexNet50 with an augmented dataset with the augmented dataset.This method used three publicly available MRI classification datasets:Harvard whole brain atlas(HWBA-dataset),the School of Biomedical Engineering of Southern Medical University(SMU-dataset),and The National Institute of Neuroscience and Hospitals brain MRI dataset(NINS-dataset)for analysis.Various hyperparameter optimizers like Adam,stochastic gradient descent(SGD),Root mean square propagation(RMS prop),Adamax,and AdamW have been used to compare the performance of the learning process.HWBA-dataset registers maximum classification performance.We evaluated the performance of the proposed classification model using several quantitative metrics,achieving an average accuracy of 98%.展开更多
Many applications,including security systems,medical diagnostics,and human-computer interfaces,depend on eye gaze recognition.However,due to factors including individual variations,occlusions,and shifting illumination...Many applications,including security systems,medical diagnostics,and human-computer interfaces,depend on eye gaze recognition.However,due to factors including individual variations,occlusions,and shifting illumination conditions,real-world scenarios continue to provide difficulties for accurate and consistent eye gaze recognition.This work is aimed at investigating the potential benefits of employing transfer learning to improve eye gaze detection ability and efficiency.Transfer learning is the process of fine-tuning pre-trained models on smaller,domain-specific datasets after they have been trained on larger datasets.We study several transfer learning algorithms and evaluate their effectiveness on eye gaze identification,including both Regression and Classification tasks,using a range of deep learning architectures,namely AlexNet,Visual Geometry Group(VGG),InceptionV3,and ResNet.In this study,we evaluate the effectiveness of transfer learning-basedmodels against models that were trained fromscratch using eye-gazing datasets on grounds of various performance and loss metrics such as Precision,Accuracy,and Mean Absolute Error.We investigate the effects of different pre-trainedmodels,dataset sizes,and domain gaps on the transfer learning process,and the findings of our study clarify the efficacy of transfer learning for eye gaze detection and offer suggestions for the most successful transfer learning strategies to apply in real-world situations.展开更多
Deep Learning-based systems for Finger vein recognition have gained rising attention in recent years due to improved efficiency and enhanced security.The performance of existing CNN-based methods is limited by the pun...Deep Learning-based systems for Finger vein recognition have gained rising attention in recent years due to improved efficiency and enhanced security.The performance of existing CNN-based methods is limited by the puny generalization of learned features and deficiency of the finger vein image training data.Considering the concerns of existing methods,in this work,a simplified deep transfer learning-based framework for finger-vein recognition is developed using an EfficientNet model of deep learning with a self-attention mechanism.Data augmentation using various geometrical methods is employed to address the problem of training data shortage required for a deep learning model.The proposed model is tested using K-fold cross-validation on three publicly available datasets:HKPU,FVUSM,and SDUMLA.Also,the developed network is compared with other modern deep nets to check its effectiveness.In addition,a comparison of the proposed method with other existing Finger vein recognition(FVR)methods is also done.The experimental results exhibited superior recognition accuracy of the proposed method compared to other existing methods.In addition,the developed method proves to be more effective and less sophisticated at extracting robust features.The proposed EffAttenNet achieves an accuracy of 98.14%on HKPU,99.03%on FVUSM,and 99.50%on SDUMLA databases.展开更多
Seismic fault identification is a critical step in structural interpretation,reservoir characterization,and well-drilling planning.However,fault identification in deep fault-karst carbonate formations is particularly ...Seismic fault identification is a critical step in structural interpretation,reservoir characterization,and well-drilling planning.However,fault identification in deep fault-karst carbonate formations is particularly challenging due to their deep burial depth and the complex effects of dissolution.Traditional manual interpretation methods are often labor intensive and prone to high uncertainty due to their subjective nature.To address these limitations,this study proposes a transfer learningebased strategy for fault identification in deep fault-karst carbonate formations.The proposed methodology began with the generation of a large volume of synthetic seismic samples based on statistical fault distribution patterns observed in the study area.These synthetic samples were used to pretrain an improved U-Net network architecture,enhanced with an attention mechanism,to create a robust pretrained model.Subsequently,real-world fault labels were manually annotated based on verified fault interpretations and integrated into the training dataset.This combination of synthetic and real-world data was used to fine-tune the pretrained model,significantly improving its fault interpretation accuracy.The experimental results demonstrate that the integration of synthetic and real-world samples effectively enhances the quality of the training dataset.Furthermore,the proposed transfer learning strategy significantly im-proves fault recognition accuracy.By replacing the traditional weighted cross-entropy loss function with the Dice loss function,the model successfully addresses the issue of extreme class imbalance between positive and negative samples.Practical applications confirm that the proposed transfer learning strategy can accurately identify fault structures in deep fault-karst carbonate formations,providing a novel and effective technical approach for fault interpretation in such complex geological settings.展开更多
Named Entity Recognition(NER)is vital in natural language processing for the analysis of news texts,as it accurately identifies entities such as locations,persons,and organizations,which is crucial for applications li...Named Entity Recognition(NER)is vital in natural language processing for the analysis of news texts,as it accurately identifies entities such as locations,persons,and organizations,which is crucial for applications like news summarization and event tracking.However,NER in the news domain faces challenges due to insufficient annotated data,complex entity structures,and strong context dependencies.To address these issues,we propose a new Chinesenamed entity recognition method that integrates transfer learning with word embeddings.Our approach leverages the ERNIE pre-trained model for transfer learning and obtaining general language representations and incorporates the Soft-lexicon word embedding technique to handle varied entity structures.This dual-strategy enhances the model’s understanding of context and boosts its ability to process complex texts.Experimental results show that our method achieves an F1 score of 94.72% on a news dataset,surpassing baseline methods by 3%–4%,thereby confirming its effectiveness for Chinese-named entity recognition in the news domain.展开更多
The COVID-19 pandemic,which was declared by the WHO,had created a global health crisis and disrupted people’s daily lives.A large number of people were affected by the COVID-19 pandemic.Therefore,a diagnostic model n...The COVID-19 pandemic,which was declared by the WHO,had created a global health crisis and disrupted people’s daily lives.A large number of people were affected by the COVID-19 pandemic.Therefore,a diagnostic model needs to be generated which can effectively classify the COVID and non-COVID cases.In this work,our aim is to develop a diagnostic model based on deep features using effectiveness of Chest X-ray(CXR)in distinguishing COVID from non-COVID cases.The proposed diagnostic framework utilizes CXR to diagnose COVID-19 and includes Grad-CAM visualizations for a visual interpretation of predicted images.The model’s performance was evaluated using various metrics,including accuracy,precision,recall,F1-score,and Gmean.Several machine learning models,such as random forest,dense neural network,SVM,twin SVM,extreme learning machine,random vector functional link,and kernel ridge regression,were selected to diagnose COVID-19 cases.Transfer learning was used to extract deep features.For feature extraction many CNN-based models such as Inception V3,MobileNet,ResNet50,VGG16 and Xception models are used.It was evident from the experiments that ResNet50 architecture outperformed all other CNN architectures based on AUC.The TWSVM classifier achieved the highest AUC score of 0.98 based on the ResNet50 feature vector.展开更多
The electromagnetic pulse valve,as a key component in baghouse dust removal systems,plays a crucial role in the performance of the system.However,despite the promising results of intelligent fault diagnosis methods ba...The electromagnetic pulse valve,as a key component in baghouse dust removal systems,plays a crucial role in the performance of the system.However,despite the promising results of intelligent fault diagnosis methods based on extensive data in diagnosing electromagnetic valves,real-world diagnostic scenarios still face numerous challenges.Collecting fault data for electromagnetic pulse valves is not only time-consuming but also costly,making it difficult to obtain sufficient fault data in advance,which poses challenges for small sample fault diagnosis.To address this issue,this paper proposes a fault diagnosis method for electromagnetic pulse valves based on deep transfer learning and simulated data.This method achieves effective transfer from simulated data to real data through four parameter transfer strategies,which combine parameter freezing and fine-tuning operations.Furthermore,this paper identifies a parameter transfer strategy that simultaneously fine-tunes the feature extractor and classifier,and introduces an attention mechanism to integrate fault features,thereby enhancing the correlation and information complementarity among multi-sensor data.The effectiveness of the proposed method is evaluated through two fault diagnosis cases under different operating conditions.In this study,small sample data accounted for 7.9%and 8.2%of the total dataset,and the experimental results showed transfer accuracies of 93.5%and 94.2%,respectively,validating the reliability and effectiveness of the method under small sample conditions.展开更多
Coffee is an important agricultural commodity,and its production is threatened by various diseases.It is also a source of concern for coffee-exporting countries,which is causing them to rethink their strategies for th...Coffee is an important agricultural commodity,and its production is threatened by various diseases.It is also a source of concern for coffee-exporting countries,which is causing them to rethink their strategies for the future.Maintaining crop production requires early diagnosis.Notably,Coffee Leaf Miner(CLM)Machine learning(ML)offers promising tools for automated disease detection.Early detection of CLM is crucial for minimising yield losses.However,this study explores the effectiveness of using Convolutional Neural Networks(CNNs)with transfer learning algorithms ResNet50,DenseNet121,MobileNet,Inception,and hybrid VGG19 for classifying coffee leaf images as healthy or CLM-infected.Leveraging the JMuBEN1 dataset,the proposed hybrid VGG19 model achieved exceptional performance,reaching 97%accuracy on both training and validation data.Additionally,high scores for precision,recall,and F1-score.The confusion matrix shows that all the test samples were correctly classified,which indicates the model’s strong performance on this dataset,demonstrating that the model is effective in distinguishing between healthy and CLM-infected leaves.This suggests strong potential for implementing this approach in real-world coffee plantations for early disease detection and improved disease management,and adapting it for practical deployment in agricultural settings.As well as supporting farmers in detecting diseases using modern,inexpensive methods that do not require specialists,and utilising deep learning technologies.展开更多
Aerodynamic evaluation under multi-condition is indispensable for the design of aircraft,and the requirement for mass data still means a high cost.To address this problem,we propose a novel point-cloud multi-condition...Aerodynamic evaluation under multi-condition is indispensable for the design of aircraft,and the requirement for mass data still means a high cost.To address this problem,we propose a novel point-cloud multi-condition aerodynamics transfer learning(PCMCA-TL)framework that enables aerodynamic prediction in data-scarce sce-narios by transferring knowledge from well-learned scenarios.We modified the PointNeXt segmentation archi-tecture to a PointNeXtReg+regression model,including a working condition input module.The model is first pre-trained on a public dataset with 2000 shapes but only one working condition and then fine-tuned on a multi-condition small-scale spaceplane dataset.The effectiveness of the PCMCA-TL framework is verified by comparing the pressure coefficients predicted by direct training,pre-training,and TL models.Furthermore,by comparing the aerodynamic force coefficients calculated by predicted pressure coefficients in seconds with the correspond-ing CFD results obtained in hours,the accuracy highlights the development potential of deep transfer learning in aerodynamic evaluation.展开更多
Background:Pneumoconioses,a group of occupational lung diseases caused by inhalation of mineral dust,pose significant health risks to affected individuals.Accurate assessment of profusion(extent of lung involvement)in...Background:Pneumoconioses,a group of occupational lung diseases caused by inhalation of mineral dust,pose significant health risks to affected individuals.Accurate assessment of profusion(extent of lung involvement)in chest radiographs is essential for screening,diagnosis and monitoring of the diseases along with epidemiological classification.This study explores an automated classification system combining U-Net-based segmentation for lung field delineation and DenseNet121 with ImageNet-based transfer learning for profusion classification.Methods:Lung field segmentation using U-Net achieved precise delineation,ensuring accurate region-of-interest definition.Transfer learning with DenseNet121 leveraged pre-trained knowledge from ImageNet,minimizing the need for extensive training.The model was fine-tuned with International Labour Organization(ILO)-2022 version standard chest radiographs and evaluated on a diverse dataset of ILO-2000 version standardized radiographs.Results:The U-Net-based segmentation demonstrated robust performance(Accuracy 94%and Dice Coefficient 90%),facilitating subsequent profusion classification.The DenseNet121-based transfer learning model exhibited high accuracy(95%),precision(92%),and recall(94%)for classifying four profusion levels on test ILO 2000/2011D dataset.The final Evaluation on ILO-2000 radiographs highlighted its generalization capability.Conclusion:The proposed system offers clinical promise,aiding radiologists,pulmonologists,general physicians,and occupational health specialists in pneumoconioses screening,diagnosis,monitoring and epidemiological classification.Best of our knowledge,this is the first work in the field of automated Classification of Profusion in Chest Radiographs of Pneumoconioses based on recently published latest ILO-2022 standard.Future research should focus on further refinement and real-world validation.This approach exemplifies the potential of deep learning for enhancing the accuracy and efficiency of pneumoconioses assessment,benefiting industrial workers,patients,and healthcare providers.展开更多
With the advancements in parameter-efficient transfer learning techniques,it has become feasible to leverage large pre-trained language models for downstream tasks under low-cost and low-resource conditions.However,ap...With the advancements in parameter-efficient transfer learning techniques,it has become feasible to leverage large pre-trained language models for downstream tasks under low-cost and low-resource conditions.However,applying this technique to multimodal knowledge transfer introduces a significant challenge:ensuring alignment across modalities while minimizing the number of additional parameters required for downstream task adaptation.This paper introduces UniTrans,a framework aimed at facilitating efficient knowledge transfer across multiple modalities.UniTrans leverages Vector-based Cross-modal Random Matrix Adaptation to enable fine-tuning with minimal parameter overhead.To further enhance modality alignment,we introduce two key components:the Multimodal Consistency Alignment Module and the Query-Augmentation Side Network,specifically optimized for scenarios with extremely limited trainable parameters.Extensive evaluations on various cross-modal downstream tasks demonstrate that our approach surpasses state-of-the-art methods while using just 5%of their trainable parameters.Additionally,it achieves superior performance compared to fully fine-tuned models on certain benchmarks.展开更多
This study presents an emergency control method for sub-synchronous oscillations in wind power gridconnected systems based on transfer learning,addressing the issue of insufficient generalization ability of traditiona...This study presents an emergency control method for sub-synchronous oscillations in wind power gridconnected systems based on transfer learning,addressing the issue of insufficient generalization ability of traditional methods in complex real-world scenarios.By combining deep reinforcement learning with a transfer learning framework,cross-scenario knowledge transfer is achieved,significantly enhancing the adaptability of the control strategy.First,a sub-synchronous oscillation emergency control model for the wind power grid integration system is constructed under fixed scenarios based on deep reinforcement learning.A reward evaluation system based on the active power oscillation pattern of the system is proposed,introducing penalty functions for the number of machine-shedding rounds and the number of machines shed.This avoids the economic losses and grid security risks caused by the excessive one-time shedding of wind turbines.Furthermore,transfer learning is introduced into model training to enhance the model’s generalization capability in dealing with complex scenarios of actual wind power grid integration systems.By introducing the Maximum Mean Discrepancy(MMD)algorithm to calculate the distribution differences between source data and target data,the online decision-making reliability of the emergency control model is improved.Finally,the effectiveness of the proposed emergency control method for multi-scenario sub-synchronous oscillation in wind power grid integration systems based on transfer learning is analyzed using the New England 39-bus system.展开更多
The Internet of Things(IoT)is an innovation that combines imagined space with the actual world on a single platform.Because of the recent rapid rise of IoT devices,there has been a lack of standards,leading to a massi...The Internet of Things(IoT)is an innovation that combines imagined space with the actual world on a single platform.Because of the recent rapid rise of IoT devices,there has been a lack of standards,leading to a massive increase in unprotected devices connecting to networks.Consequently,cyberattacks on IoT are becoming more common,particularly keylogging attacks,which are often caused by security vulnerabilities on IoT networks.This research focuses on the role of transfer learning and ensemble classifiers in enhancing the detection of keylogging attacks within small,imbalanced IoT datasets.The authors propose a model that combines transfer learning with ensemble classification methods,leading to improved detection accuracy.By leveraging the BoT-IoT and keylogger_detection datasets,they facilitate the transfer of knowledge across various domains.The results reveal that the integration of transfer learning and ensemble classifiers significantly improves detection capabilities,even in scenarios with limited data availability.The proposed TRANS-ENS model showcases exceptional accuracy and a minimal false positive rate,outperforming current deep learning approaches.The primary objectives include:(i)introducing an ensemble feature selection technique to identify common features across models,(ii)creating a pre-trained deep learning model through transfer learning for the detection of keylogging attacks,and(iii)developing a transfer learning-ensemble model dedicated to keylogging detection.Experimental findings indicate that the TRANS-ENS model achieves a detection accuracy of 96.06%and a false alarm rate of 0.12%,surpassing existing models such as CNN,RNN,and LSTM.展开更多
Effective fault identification is crucial for bearings, which are critical components of mechanical systems and play a pivotal role in ensuring overall safety and operational efficiency. Bearings operate under variabl...Effective fault identification is crucial for bearings, which are critical components of mechanical systems and play a pivotal role in ensuring overall safety and operational efficiency. Bearings operate under variable service conditions, and their diagnostic environments are complex and dynamic. In the process of bearing diagnosis, fault datasets are relatively scarce compared with datasets representing normal operating conditions. These challenges frequently cause the practicality of fault detection to decline, the extraction of fault features to be incomplete, and the diagnostic accuracy of many existing models to decrease. In this work, a transfer-learning framework, designated DSCNN-HA-TL, is introduced to address the enduring challenge of cross-condition diagnosis in rolling-bearing fault detection. The framework integrates a window global mixed attention mechanism with a deep separable convolutional network, thereby enabling adaptation to fault detection tasks under diverse operating conditions. First, a Convolutional Neural Network (CNN) is employed as the foundational architecture, where the original convolutional layers are enhanced through the incorporation of depthwise separable convolutions, resulting in a Depthwise Separable Convolutional Neural Network (DSCNN) architecture. Subsequently, the extraction of fault characteristics is further refined through a dual-branch network that integrates hybrid attention mechanisms, specifically windowed and global attention mechanisms. This approach enables the acquisition of multi-level feature fusion information, thereby enhancing the accuracy of fault classification. The integration of these features not only optimizes the characteristic extraction process but also yields improvements in accuracy, representational capacity, and robustness in fault feature recognition. In conclusion, the proposed method achieved average precisions of 99.93% and 99.55% in transfer learning tasks, as demonstrated by the experimental results obtained from the CWRU public dataset and the bearing fault detection platform dataset. The experimental findings further provided a detailed comparison between the diagnostic models before and after the enhancement, thereby substantiating the pronounced advantages of the DSCNN-HA-TL approach in accurately identifying faults in critical mechanical components under diverse operating conditions.展开更多
Calculating the inter-layer ion diffusion barrier, a crucial metric for evaluating the rate performance of 2D electrode materials, is time-consuming using the transition state search approach. A novel electrostatic po...Calculating the inter-layer ion diffusion barrier, a crucial metric for evaluating the rate performance of 2D electrode materials, is time-consuming using the transition state search approach. A novel electrostatic potential distribution image (EPDI) transfer learning method has been proposed to efficiently and accurately predict the lithium diffusion barriers on metal element-doped transition metal dichalcogenide (TMD) surfaces. Through the analysis of the mean electrostatic potential (MEP) around binding sites, a positive correlation between binding energy and MEP in VIB-TMDs was identified. Subsequently, transfer learning techniques were used to develop a DenseNet121-TL model for establishing a more accurate mapping between the binding energy and electrostatic potential distribution. Trained on training sets containing 33% and 50% transition state search calculation results, which save 66% and 50% of the calculation time, respectively, the model achieves accurate predictions of the saddle point binding energy with mean absolute errors (MAEs) of 0.0444 and 0.0287 eV on the testing set. Based on the prediction of saddle point binding energies, we obtained a diffusion minimum energy profile with an MAE of 0.0235 eV. Furthermore, by analyzing the diffusion data, we observed that the diffusion barrier was lowered by 10% on V-doped TiS2 compared to the stoichiometric surface. Our findings are expected to provide new insights for the high-throughput calculation of ion diffusion on 2D materials.展开更多
In the realm of subsurface flow simulations,deep-learning-based surrogate models have emerged as a promising alternative to traditional simulation methods,especially in addressing complex optimization problems.However...In the realm of subsurface flow simulations,deep-learning-based surrogate models have emerged as a promising alternative to traditional simulation methods,especially in addressing complex optimization problems.However,a significant challenge lies in the necessity of numerous high-fidelity training simulations to construct these deep-learning models,which limits their application to field-scale problems.To overcome this limitation,we introduce a training procedure that leverages transfer learning with multi-fidelity training data to construct surrogate models efficiently.The procedure begins with the pre-training of the surrogate model using a relatively larger amount of data that can be efficiently generated from upscaled coarse-scale models.Subsequently,the model parameters are finetuned with a much smaller set of high-fidelity simulation data.For the cases considered in this study,this method leads to about a 75%reduction in total computational cost,in comparison with the traditional training approach,without any sacrifice of prediction accuracy.In addition,a dedicated well-control embedding model is introduced to the traditional U-Net architecture to improve the surrogate model's prediction accuracy,which is shown to be particularly effective when dealing with large-scale reservoir models under time-varying well control parameters.Comprehensive results and analyses are presented for the prediction of well rates,pressure and saturation states of a 3D synthetic reservoir system.Finally,the proposed procedure is applied to a field-scale production optimization problem.The trained surrogate model is shown to provide excellent generalization capabilities during the optimization process,in which the final optimized net-present-value is much higher than those from the training data ranges.展开更多
Skin cancer is the abnormal development of cells on the surface of the skin and is one of the most fatal diseases in humans.It usually appears in locations that are exposed to the sun,but can also appear in areas that...Skin cancer is the abnormal development of cells on the surface of the skin and is one of the most fatal diseases in humans.It usually appears in locations that are exposed to the sun,but can also appear in areas that are not regularly exposed to the sun.Due to the striking similarities between benign and malignant lesions,skin cancer detection remains a problem,even for expert dermatologists.Considering the inability of dermatologists to di-agnose skin cancer accurately,a convolutional neural network(CNN)approach was used for skin cancer diag-nosis.However,the CNN model requires a significant number of image datasets for better performance;thus,image augmentation and transfer learning techniques have been used in this study to boost the number of images and the performance of the model,because there are a limited number of medical images.This study proposes an ensemble transfer-learning-based model that can efficiently classify skin lesions into one of seven categories to aid dermatologists in skin cancer detection:(i)actinic keratoses,(ii)basal cell carcinoma,(iii)benign keratosis,(iv)dermatofibroma,(v)melanocytic nevi,(vi)melanoma,and(vii)vascular skin lesions.Five transfer learning models were used as the basis of the ensemble:MobileNet,EfficientNetV2B2,Xception,ResNeXt101,and Den-seNet201.In addition to the stratified 10-fold cross-validation,the results of each individual model were fused to achieve greater classification accuracy.An annealing learning rate scheduler and test time augmentation(TTA)were also used to increase the performance of the model during the training and testing stages.A total of 10,015 publicly available dermoscopy images from the HAM10000(Human Against Machine)dataset,which contained samples from the seven common skin lesion categories,were used to train and evaluate the models.The proposed technique attained 94.49%accuracy on the dataset.These results suggest that this strategy can be useful for improving the accuracy of skin cancer classification.However,the weighted average of F1-score,recall,and precision were obtained to be 94.68%,94.49%,and 95.07%,respectively.展开更多
Transfer learning is the predominant method for adapting pre-trained models on another task to new domains while preserving their internal architectures and augmenting them with requisite layers in Deep Neural Network...Transfer learning is the predominant method for adapting pre-trained models on another task to new domains while preserving their internal architectures and augmenting them with requisite layers in Deep Neural Network models.Training intricate pre-trained models on a sizable dataset requires significant resources to fine-tune hyperparameters carefully.Most existing initialization methods mainly focus on gradient flow-related problems,such as gradient vanishing or exploding,or other existing approaches that require extra models that do not consider our setting,which is more practical.To address these problems,we suggest employing gradient-free heuristic methods to initialize the weights of the final new-added fully connected layer in neural networks froma small set of training data with fewer classes.The approach relies on partitioning the output values from pre-trained models for a small set into two separate intervals determined by the targets.This process is framed as an optimization problem for each output neuron and class.The optimization selects the highest values as weights,considering their direction towards the respective classes.Furthermore,empirical 145 experiments involve a variety of neural networkmodels tested acrossmultiple benchmarks and domains,occasionally yielding accuracies comparable to those achieved with gradient descent methods by using only small subsets.展开更多
Various and intricate varieties of lung disease have made it challenging for computer aided diagnosis to appropriately segment lung lesions utilizing computed tomography(CT)images.This study integrates transfer learni...Various and intricate varieties of lung disease have made it challenging for computer aided diagnosis to appropriately segment lung lesions utilizing computed tomography(CT)images.This study integrates transfer learning with the attention mechanism to construct a deep learning model that can automatically detect new coronary pneumonia on lung CT images.In this study,using VGG16 pre-trained by ImageNet as the encoder,the decoder was established utilizing the U-Net structure.The attention module is incorporated during each concatenate procedure,permitting the model to concentrate on the critical information and identify the crucial components efficiently.The public COVID-19-CT-Seg-Benchmark dataset was utilized for experiments,and the highest scores for Dice,F1,and Accuracy were 0.9071,0.9076,and 0.9965,respectively.The generalization performance was assessed concurrently,with performance metrics including Dice,F1,and Accuracy over 0.8.The experimental findings indicate the feasibility of the segmentation network proposed in this study.展开更多
基金supported by the National Key Research and Development Program of China(Grant No.2023YFC3009400)the National Natural Science Foundation of China(Grant Nos.42307218 and U2239251).
文摘The current deep learning models for braced excavation cannot predict deformation from the beginning of excavation due to the need for a substantial corpus of sufficient historical data for training purposes.To address this issue,this study proposes a transfer learning model based on a sequence-to-sequence twodimensional(2D)convolutional long short-term memory neural network(S2SCL2D).The model can use the existing data from other adjacent similar excavations to achieve wall deflection prediction once a limited amount of monitoring data from the target excavation has been recorded.In the absence of adjacent excavation data,numerical simulation data from the target project can be employed instead.A weight update strategy is proposed to improve the prediction accuracy by integrating the stochastic gradient masking with an early stopping mechanism.To illustrate the proposed methodology,an excavation project in Hangzhou,China is adopted.The proposed deep transfer learning model,which uses either adjacent excavation data or numerical simulation data as the source domain,shows a significant improvement in performance when compared to the non-transfer learning model.Using the simulation data from the target project even leads to better prediction performance than using the actual monitoring data from other adjacent excavations.The results demonstrate that the proposed model can reasonably predict the deformation with limited data from the target project.
文摘In radiology,magnetic resonance imaging(MRI)is an essential diagnostic tool that provides detailed images of a patient’s anatomical and physiological structures.MRI is particularly effective for detecting soft tissue anomalies.Traditionally,radiologists manually interpret these images,which can be labor-intensive and time-consuming due to the vast amount of data.To address this challenge,machine learning,and deep learning approaches can be utilized to improve the accuracy and efficiency of anomaly detection in MRI scans.This manuscript presents the use of the Deep AlexNet50 model for MRI classification with discriminative learning methods.There are three stages for learning;in the first stage,the whole dataset is used to learn the features.In the second stage,some layers of AlexNet50 are frozen with an augmented dataset,and in the third stage,AlexNet50 with an augmented dataset with the augmented dataset.This method used three publicly available MRI classification datasets:Harvard whole brain atlas(HWBA-dataset),the School of Biomedical Engineering of Southern Medical University(SMU-dataset),and The National Institute of Neuroscience and Hospitals brain MRI dataset(NINS-dataset)for analysis.Various hyperparameter optimizers like Adam,stochastic gradient descent(SGD),Root mean square propagation(RMS prop),Adamax,and AdamW have been used to compare the performance of the learning process.HWBA-dataset registers maximum classification performance.We evaluated the performance of the proposed classification model using several quantitative metrics,achieving an average accuracy of 98%.
文摘Many applications,including security systems,medical diagnostics,and human-computer interfaces,depend on eye gaze recognition.However,due to factors including individual variations,occlusions,and shifting illumination conditions,real-world scenarios continue to provide difficulties for accurate and consistent eye gaze recognition.This work is aimed at investigating the potential benefits of employing transfer learning to improve eye gaze detection ability and efficiency.Transfer learning is the process of fine-tuning pre-trained models on smaller,domain-specific datasets after they have been trained on larger datasets.We study several transfer learning algorithms and evaluate their effectiveness on eye gaze identification,including both Regression and Classification tasks,using a range of deep learning architectures,namely AlexNet,Visual Geometry Group(VGG),InceptionV3,and ResNet.In this study,we evaluate the effectiveness of transfer learning-basedmodels against models that were trained fromscratch using eye-gazing datasets on grounds of various performance and loss metrics such as Precision,Accuracy,and Mean Absolute Error.We investigate the effects of different pre-trainedmodels,dataset sizes,and domain gaps on the transfer learning process,and the findings of our study clarify the efficacy of transfer learning for eye gaze detection and offer suggestions for the most successful transfer learning strategies to apply in real-world situations.
文摘Deep Learning-based systems for Finger vein recognition have gained rising attention in recent years due to improved efficiency and enhanced security.The performance of existing CNN-based methods is limited by the puny generalization of learned features and deficiency of the finger vein image training data.Considering the concerns of existing methods,in this work,a simplified deep transfer learning-based framework for finger-vein recognition is developed using an EfficientNet model of deep learning with a self-attention mechanism.Data augmentation using various geometrical methods is employed to address the problem of training data shortage required for a deep learning model.The proposed model is tested using K-fold cross-validation on three publicly available datasets:HKPU,FVUSM,and SDUMLA.Also,the developed network is compared with other modern deep nets to check its effectiveness.In addition,a comparison of the proposed method with other existing Finger vein recognition(FVR)methods is also done.The experimental results exhibited superior recognition accuracy of the proposed method compared to other existing methods.In addition,the developed method proves to be more effective and less sophisticated at extracting robust features.The proposed EffAttenNet achieves an accuracy of 98.14%on HKPU,99.03%on FVUSM,and 99.50%on SDUMLA databases.
基金support provided by the China Postdoctoral Science Foundation(Grant No.2024M763650)the Excellent Young Scientists Fund Program of SINOPEC Petroleum Exploration and Production Research Institute(Grant No.yk2024010).
文摘Seismic fault identification is a critical step in structural interpretation,reservoir characterization,and well-drilling planning.However,fault identification in deep fault-karst carbonate formations is particularly challenging due to their deep burial depth and the complex effects of dissolution.Traditional manual interpretation methods are often labor intensive and prone to high uncertainty due to their subjective nature.To address these limitations,this study proposes a transfer learningebased strategy for fault identification in deep fault-karst carbonate formations.The proposed methodology began with the generation of a large volume of synthetic seismic samples based on statistical fault distribution patterns observed in the study area.These synthetic samples were used to pretrain an improved U-Net network architecture,enhanced with an attention mechanism,to create a robust pretrained model.Subsequently,real-world fault labels were manually annotated based on verified fault interpretations and integrated into the training dataset.This combination of synthetic and real-world data was used to fine-tune the pretrained model,significantly improving its fault interpretation accuracy.The experimental results demonstrate that the integration of synthetic and real-world samples effectively enhances the quality of the training dataset.Furthermore,the proposed transfer learning strategy significantly im-proves fault recognition accuracy.By replacing the traditional weighted cross-entropy loss function with the Dice loss function,the model successfully addresses the issue of extreme class imbalance between positive and negative samples.Practical applications confirm that the proposed transfer learning strategy can accurately identify fault structures in deep fault-karst carbonate formations,providing a novel and effective technical approach for fault interpretation in such complex geological settings.
基金funded by Advanced Research Project(30209040702).
文摘Named Entity Recognition(NER)is vital in natural language processing for the analysis of news texts,as it accurately identifies entities such as locations,persons,and organizations,which is crucial for applications like news summarization and event tracking.However,NER in the news domain faces challenges due to insufficient annotated data,complex entity structures,and strong context dependencies.To address these issues,we propose a new Chinesenamed entity recognition method that integrates transfer learning with word embeddings.Our approach leverages the ERNIE pre-trained model for transfer learning and obtaining general language representations and incorporates the Soft-lexicon word embedding technique to handle varied entity structures.This dual-strategy enhances the model’s understanding of context and boosts its ability to process complex texts.Experimental results show that our method achieves an F1 score of 94.72% on a news dataset,surpassing baseline methods by 3%–4%,thereby confirming its effectiveness for Chinese-named entity recognition in the news domain.
文摘The COVID-19 pandemic,which was declared by the WHO,had created a global health crisis and disrupted people’s daily lives.A large number of people were affected by the COVID-19 pandemic.Therefore,a diagnostic model needs to be generated which can effectively classify the COVID and non-COVID cases.In this work,our aim is to develop a diagnostic model based on deep features using effectiveness of Chest X-ray(CXR)in distinguishing COVID from non-COVID cases.The proposed diagnostic framework utilizes CXR to diagnose COVID-19 and includes Grad-CAM visualizations for a visual interpretation of predicted images.The model’s performance was evaluated using various metrics,including accuracy,precision,recall,F1-score,and Gmean.Several machine learning models,such as random forest,dense neural network,SVM,twin SVM,extreme learning machine,random vector functional link,and kernel ridge regression,were selected to diagnose COVID-19 cases.Transfer learning was used to extract deep features.For feature extraction many CNN-based models such as Inception V3,MobileNet,ResNet50,VGG16 and Xception models are used.It was evident from the experiments that ResNet50 architecture outperformed all other CNN architectures based on AUC.The TWSVM classifier achieved the highest AUC score of 0.98 based on the ResNet50 feature vector.
基金Supported by National Natural Science Foundation of China(Grant No.51675040)。
文摘The electromagnetic pulse valve,as a key component in baghouse dust removal systems,plays a crucial role in the performance of the system.However,despite the promising results of intelligent fault diagnosis methods based on extensive data in diagnosing electromagnetic valves,real-world diagnostic scenarios still face numerous challenges.Collecting fault data for electromagnetic pulse valves is not only time-consuming but also costly,making it difficult to obtain sufficient fault data in advance,which poses challenges for small sample fault diagnosis.To address this issue,this paper proposes a fault diagnosis method for electromagnetic pulse valves based on deep transfer learning and simulated data.This method achieves effective transfer from simulated data to real data through four parameter transfer strategies,which combine parameter freezing and fine-tuning operations.Furthermore,this paper identifies a parameter transfer strategy that simultaneously fine-tunes the feature extractor and classifier,and introduces an attention mechanism to integrate fault features,thereby enhancing the correlation and information complementarity among multi-sensor data.The effectiveness of the proposed method is evaluated through two fault diagnosis cases under different operating conditions.In this study,small sample data accounted for 7.9%and 8.2%of the total dataset,and the experimental results showed transfer accuracies of 93.5%and 94.2%,respectively,validating the reliability and effectiveness of the method under small sample conditions.
文摘Coffee is an important agricultural commodity,and its production is threatened by various diseases.It is also a source of concern for coffee-exporting countries,which is causing them to rethink their strategies for the future.Maintaining crop production requires early diagnosis.Notably,Coffee Leaf Miner(CLM)Machine learning(ML)offers promising tools for automated disease detection.Early detection of CLM is crucial for minimising yield losses.However,this study explores the effectiveness of using Convolutional Neural Networks(CNNs)with transfer learning algorithms ResNet50,DenseNet121,MobileNet,Inception,and hybrid VGG19 for classifying coffee leaf images as healthy or CLM-infected.Leveraging the JMuBEN1 dataset,the proposed hybrid VGG19 model achieved exceptional performance,reaching 97%accuracy on both training and validation data.Additionally,high scores for precision,recall,and F1-score.The confusion matrix shows that all the test samples were correctly classified,which indicates the model’s strong performance on this dataset,demonstrating that the model is effective in distinguishing between healthy and CLM-infected leaves.This suggests strong potential for implementing this approach in real-world coffee plantations for early disease detection and improved disease management,and adapting it for practical deployment in agricultural settings.As well as supporting farmers in detecting diseases using modern,inexpensive methods that do not require specialists,and utilising deep learning technologies.
基金supported by the Natural Science Foundation of Hunan Province of China(Grant No.2021JJ10045).
文摘Aerodynamic evaluation under multi-condition is indispensable for the design of aircraft,and the requirement for mass data still means a high cost.To address this problem,we propose a novel point-cloud multi-condition aerodynamics transfer learning(PCMCA-TL)framework that enables aerodynamic prediction in data-scarce sce-narios by transferring knowledge from well-learned scenarios.We modified the PointNeXt segmentation archi-tecture to a PointNeXtReg+regression model,including a working condition input module.The model is first pre-trained on a public dataset with 2000 shapes but only one working condition and then fine-tuned on a multi-condition small-scale spaceplane dataset.The effectiveness of the PCMCA-TL framework is verified by comparing the pressure coefficients predicted by direct training,pre-training,and TL models.Furthermore,by comparing the aerodynamic force coefficients calculated by predicted pressure coefficients in seconds with the correspond-ing CFD results obtained in hours,the accuracy highlights the development potential of deep transfer learning in aerodynamic evaluation.
文摘Background:Pneumoconioses,a group of occupational lung diseases caused by inhalation of mineral dust,pose significant health risks to affected individuals.Accurate assessment of profusion(extent of lung involvement)in chest radiographs is essential for screening,diagnosis and monitoring of the diseases along with epidemiological classification.This study explores an automated classification system combining U-Net-based segmentation for lung field delineation and DenseNet121 with ImageNet-based transfer learning for profusion classification.Methods:Lung field segmentation using U-Net achieved precise delineation,ensuring accurate region-of-interest definition.Transfer learning with DenseNet121 leveraged pre-trained knowledge from ImageNet,minimizing the need for extensive training.The model was fine-tuned with International Labour Organization(ILO)-2022 version standard chest radiographs and evaluated on a diverse dataset of ILO-2000 version standardized radiographs.Results:The U-Net-based segmentation demonstrated robust performance(Accuracy 94%and Dice Coefficient 90%),facilitating subsequent profusion classification.The DenseNet121-based transfer learning model exhibited high accuracy(95%),precision(92%),and recall(94%)for classifying four profusion levels on test ILO 2000/2011D dataset.The final Evaluation on ILO-2000 radiographs highlighted its generalization capability.Conclusion:The proposed system offers clinical promise,aiding radiologists,pulmonologists,general physicians,and occupational health specialists in pneumoconioses screening,diagnosis,monitoring and epidemiological classification.Best of our knowledge,this is the first work in the field of automated Classification of Profusion in Chest Radiographs of Pneumoconioses based on recently published latest ILO-2022 standard.Future research should focus on further refinement and real-world validation.This approach exemplifies the potential of deep learning for enhancing the accuracy and efficiency of pneumoconioses assessment,benefiting industrial workers,patients,and healthcare providers.
文摘With the advancements in parameter-efficient transfer learning techniques,it has become feasible to leverage large pre-trained language models for downstream tasks under low-cost and low-resource conditions.However,applying this technique to multimodal knowledge transfer introduces a significant challenge:ensuring alignment across modalities while minimizing the number of additional parameters required for downstream task adaptation.This paper introduces UniTrans,a framework aimed at facilitating efficient knowledge transfer across multiple modalities.UniTrans leverages Vector-based Cross-modal Random Matrix Adaptation to enable fine-tuning with minimal parameter overhead.To further enhance modality alignment,we introduce two key components:the Multimodal Consistency Alignment Module and the Query-Augmentation Side Network,specifically optimized for scenarios with extremely limited trainable parameters.Extensive evaluations on various cross-modal downstream tasks demonstrate that our approach surpasses state-of-the-art methods while using just 5%of their trainable parameters.Additionally,it achieves superior performance compared to fully fine-tuned models on certain benchmarks.
基金funded by Sponsorship of Science and Technology Project of State Grid Xinjiang Electric Power Co.,Ltd.,grant number SGXJ0000TKJS2400168.
文摘This study presents an emergency control method for sub-synchronous oscillations in wind power gridconnected systems based on transfer learning,addressing the issue of insufficient generalization ability of traditional methods in complex real-world scenarios.By combining deep reinforcement learning with a transfer learning framework,cross-scenario knowledge transfer is achieved,significantly enhancing the adaptability of the control strategy.First,a sub-synchronous oscillation emergency control model for the wind power grid integration system is constructed under fixed scenarios based on deep reinforcement learning.A reward evaluation system based on the active power oscillation pattern of the system is proposed,introducing penalty functions for the number of machine-shedding rounds and the number of machines shed.This avoids the economic losses and grid security risks caused by the excessive one-time shedding of wind turbines.Furthermore,transfer learning is introduced into model training to enhance the model’s generalization capability in dealing with complex scenarios of actual wind power grid integration systems.By introducing the Maximum Mean Discrepancy(MMD)algorithm to calculate the distribution differences between source data and target data,the online decision-making reliability of the emergency control model is improved.Finally,the effectiveness of the proposed emergency control method for multi-scenario sub-synchronous oscillation in wind power grid integration systems based on transfer learning is analyzed using the New England 39-bus system.
基金the Deanship of Graduate Studies and Scientific Research at Najran University for supporting the research project through the Group Research,with the project code NU/GP/SERC/13/712。
文摘The Internet of Things(IoT)is an innovation that combines imagined space with the actual world on a single platform.Because of the recent rapid rise of IoT devices,there has been a lack of standards,leading to a massive increase in unprotected devices connecting to networks.Consequently,cyberattacks on IoT are becoming more common,particularly keylogging attacks,which are often caused by security vulnerabilities on IoT networks.This research focuses on the role of transfer learning and ensemble classifiers in enhancing the detection of keylogging attacks within small,imbalanced IoT datasets.The authors propose a model that combines transfer learning with ensemble classification methods,leading to improved detection accuracy.By leveraging the BoT-IoT and keylogger_detection datasets,they facilitate the transfer of knowledge across various domains.The results reveal that the integration of transfer learning and ensemble classifiers significantly improves detection capabilities,even in scenarios with limited data availability.The proposed TRANS-ENS model showcases exceptional accuracy and a minimal false positive rate,outperforming current deep learning approaches.The primary objectives include:(i)introducing an ensemble feature selection technique to identify common features across models,(ii)creating a pre-trained deep learning model through transfer learning for the detection of keylogging attacks,and(iii)developing a transfer learning-ensemble model dedicated to keylogging detection.Experimental findings indicate that the TRANS-ENS model achieves a detection accuracy of 96.06%and a false alarm rate of 0.12%,surpassing existing models such as CNN,RNN,and LSTM.
基金supported by the National Natural Science Foundation of China(12272259)the Key Research and Development Fund of Universities in Hebei Province(2510800601A).
文摘Effective fault identification is crucial for bearings, which are critical components of mechanical systems and play a pivotal role in ensuring overall safety and operational efficiency. Bearings operate under variable service conditions, and their diagnostic environments are complex and dynamic. In the process of bearing diagnosis, fault datasets are relatively scarce compared with datasets representing normal operating conditions. These challenges frequently cause the practicality of fault detection to decline, the extraction of fault features to be incomplete, and the diagnostic accuracy of many existing models to decrease. In this work, a transfer-learning framework, designated DSCNN-HA-TL, is introduced to address the enduring challenge of cross-condition diagnosis in rolling-bearing fault detection. The framework integrates a window global mixed attention mechanism with a deep separable convolutional network, thereby enabling adaptation to fault detection tasks under diverse operating conditions. First, a Convolutional Neural Network (CNN) is employed as the foundational architecture, where the original convolutional layers are enhanced through the incorporation of depthwise separable convolutions, resulting in a Depthwise Separable Convolutional Neural Network (DSCNN) architecture. Subsequently, the extraction of fault characteristics is further refined through a dual-branch network that integrates hybrid attention mechanisms, specifically windowed and global attention mechanisms. This approach enables the acquisition of multi-level feature fusion information, thereby enhancing the accuracy of fault classification. The integration of these features not only optimizes the characteristic extraction process but also yields improvements in accuracy, representational capacity, and robustness in fault feature recognition. In conclusion, the proposed method achieved average precisions of 99.93% and 99.55% in transfer learning tasks, as demonstrated by the experimental results obtained from the CWRU public dataset and the bearing fault detection platform dataset. The experimental findings further provided a detailed comparison between the diagnostic models before and after the enhancement, thereby substantiating the pronounced advantages of the DSCNN-HA-TL approach in accurately identifying faults in critical mechanical components under diverse operating conditions.
基金supported by the National Natural Science Foundation of China(Nos.51974056 and 51474047)the Foundation of the Supercomputing Center of Dalian University of Technology,and the Foundation of the Key Laboratory of Solidification Control and Digital Preparation Technology(Liaoning Province),China.
文摘Calculating the inter-layer ion diffusion barrier, a crucial metric for evaluating the rate performance of 2D electrode materials, is time-consuming using the transition state search approach. A novel electrostatic potential distribution image (EPDI) transfer learning method has been proposed to efficiently and accurately predict the lithium diffusion barriers on metal element-doped transition metal dichalcogenide (TMD) surfaces. Through the analysis of the mean electrostatic potential (MEP) around binding sites, a positive correlation between binding energy and MEP in VIB-TMDs was identified. Subsequently, transfer learning techniques were used to develop a DenseNet121-TL model for establishing a more accurate mapping between the binding energy and electrostatic potential distribution. Trained on training sets containing 33% and 50% transition state search calculation results, which save 66% and 50% of the calculation time, respectively, the model achieves accurate predictions of the saddle point binding energy with mean absolute errors (MAEs) of 0.0444 and 0.0287 eV on the testing set. Based on the prediction of saddle point binding energies, we obtained a diffusion minimum energy profile with an MAE of 0.0235 eV. Furthermore, by analyzing the diffusion data, we observed that the diffusion barrier was lowered by 10% on V-doped TiS2 compared to the stoichiometric surface. Our findings are expected to provide new insights for the high-throughput calculation of ion diffusion on 2D materials.
基金funding support from the National Natural Science Foundation of China(No.52204065,No.ZX20230398)supported by a grant from the Human Resources Development Program(No.20216110100070)of the Korea Institute of Energy Technology Evaluation and Planning(KETEP)。
文摘In the realm of subsurface flow simulations,deep-learning-based surrogate models have emerged as a promising alternative to traditional simulation methods,especially in addressing complex optimization problems.However,a significant challenge lies in the necessity of numerous high-fidelity training simulations to construct these deep-learning models,which limits their application to field-scale problems.To overcome this limitation,we introduce a training procedure that leverages transfer learning with multi-fidelity training data to construct surrogate models efficiently.The procedure begins with the pre-training of the surrogate model using a relatively larger amount of data that can be efficiently generated from upscaled coarse-scale models.Subsequently,the model parameters are finetuned with a much smaller set of high-fidelity simulation data.For the cases considered in this study,this method leads to about a 75%reduction in total computational cost,in comparison with the traditional training approach,without any sacrifice of prediction accuracy.In addition,a dedicated well-control embedding model is introduced to the traditional U-Net architecture to improve the surrogate model's prediction accuracy,which is shown to be particularly effective when dealing with large-scale reservoir models under time-varying well control parameters.Comprehensive results and analyses are presented for the prediction of well rates,pressure and saturation states of a 3D synthetic reservoir system.Finally,the proposed procedure is applied to a field-scale production optimization problem.The trained surrogate model is shown to provide excellent generalization capabilities during the optimization process,in which the final optimized net-present-value is much higher than those from the training data ranges.
文摘Skin cancer is the abnormal development of cells on the surface of the skin and is one of the most fatal diseases in humans.It usually appears in locations that are exposed to the sun,but can also appear in areas that are not regularly exposed to the sun.Due to the striking similarities between benign and malignant lesions,skin cancer detection remains a problem,even for expert dermatologists.Considering the inability of dermatologists to di-agnose skin cancer accurately,a convolutional neural network(CNN)approach was used for skin cancer diag-nosis.However,the CNN model requires a significant number of image datasets for better performance;thus,image augmentation and transfer learning techniques have been used in this study to boost the number of images and the performance of the model,because there are a limited number of medical images.This study proposes an ensemble transfer-learning-based model that can efficiently classify skin lesions into one of seven categories to aid dermatologists in skin cancer detection:(i)actinic keratoses,(ii)basal cell carcinoma,(iii)benign keratosis,(iv)dermatofibroma,(v)melanocytic nevi,(vi)melanoma,and(vii)vascular skin lesions.Five transfer learning models were used as the basis of the ensemble:MobileNet,EfficientNetV2B2,Xception,ResNeXt101,and Den-seNet201.In addition to the stratified 10-fold cross-validation,the results of each individual model were fused to achieve greater classification accuracy.An annealing learning rate scheduler and test time augmentation(TTA)were also used to increase the performance of the model during the training and testing stages.A total of 10,015 publicly available dermoscopy images from the HAM10000(Human Against Machine)dataset,which contained samples from the seven common skin lesion categories,were used to train and evaluate the models.The proposed technique attained 94.49%accuracy on the dataset.These results suggest that this strategy can be useful for improving the accuracy of skin cancer classification.However,the weighted average of F1-score,recall,and precision were obtained to be 94.68%,94.49%,and 95.07%,respectively.
基金supported by the BK21 FOUR project(AI-driven Convergence Software Education Research Program)funded by the Ministry of Education,School of Computer Science and Engineering,Kyungpook National University,Republic of Korea(4120240214871)supported by the New Faculty Start Up Fund from LSU Health Sciences New Orleans,LA,USA.
文摘Transfer learning is the predominant method for adapting pre-trained models on another task to new domains while preserving their internal architectures and augmenting them with requisite layers in Deep Neural Network models.Training intricate pre-trained models on a sizable dataset requires significant resources to fine-tune hyperparameters carefully.Most existing initialization methods mainly focus on gradient flow-related problems,such as gradient vanishing or exploding,or other existing approaches that require extra models that do not consider our setting,which is more practical.To address these problems,we suggest employing gradient-free heuristic methods to initialize the weights of the final new-added fully connected layer in neural networks froma small set of training data with fewer classes.The approach relies on partitioning the output values from pre-trained models for a small set into two separate intervals determined by the targets.This process is framed as an optimization problem for each output neuron and class.The optimization selects the highest values as weights,considering their direction towards the respective classes.Furthermore,empirical 145 experiments involve a variety of neural networkmodels tested acrossmultiple benchmarks and domains,occasionally yielding accuracies comparable to those achieved with gradient descent methods by using only small subsets.
基金the Natural Science Foundation of Zhejiang Province(No.LQ20F020024)。
文摘Various and intricate varieties of lung disease have made it challenging for computer aided diagnosis to appropriately segment lung lesions utilizing computed tomography(CT)images.This study integrates transfer learning with the attention mechanism to construct a deep learning model that can automatically detect new coronary pneumonia on lung CT images.In this study,using VGG16 pre-trained by ImageNet as the encoder,the decoder was established utilizing the U-Net structure.The attention module is incorporated during each concatenate procedure,permitting the model to concentrate on the critical information and identify the crucial components efficiently.The public COVID-19-CT-Seg-Benchmark dataset was utilized for experiments,and the highest scores for Dice,F1,and Accuracy were 0.9071,0.9076,and 0.9965,respectively.The generalization performance was assessed concurrently,with performance metrics including Dice,F1,and Accuracy over 0.8.The experimental findings indicate the feasibility of the segmentation network proposed in this study.
