Nonlinear transforms have significantly advanced learned image compression(LIC),particularly using residual blocks.This transform enhances the nonlinear expression ability and obtain compact feature representation by ...Nonlinear transforms have significantly advanced learned image compression(LIC),particularly using residual blocks.This transform enhances the nonlinear expression ability and obtain compact feature representation by enlarging the receptive field,which indicates how the convolution process extracts features in a high dimensional feature space.However,its functionality is restricted to the spatial dimension and network depth,limiting further improvements in network performance due to insufficient information interaction and representation.Crucially,the potential of high dimensional feature space in the channel dimension and the exploration of network width/resolution remain largely untapped.In this paper,we consider nonlinear transforms from the perspective of feature space,defining high-dimensional feature spaces in different dimensions and investigating the specific effects.Firstly,we introduce the dimension increasing and decreasing transforms in both channel and spatial dimensions to obtain high dimensional feature space and achieve better feature extraction.Secondly,we design a channel-spatial fusion residual transform(CSR),which incorporates multi-dimensional transforms for a more effective representation.Furthermore,we simplify the proposed fusion transform to obtain a slim architecture(CSR-sm),balancing network complexity and compression performance.Finally,we build the overall network with stacked CSR transforms to achieve better compression and reconstruction.Experimental results demonstrate that the proposed method can achieve superior ratedistortion performance compared to the existing LIC methods and traditional codecs.Specifically,our proposed method achieves 9.38%BD-rate reduction over VVC on Kodak dataset.展开更多
Modeling soil salinity in an arid salt-affected ecosystem is a difficult task when using remote sensing data because of the complicated soil context （vegetation cover, moisture, surface roughness, and organic matter...Modeling soil salinity in an arid salt-affected ecosystem is a difficult task when using remote sensing data because of the complicated soil context （vegetation cover, moisture, surface roughness, and organic matter） and the weak spectral features of salinized soil. Therefore, an index such as the salinity index （SI） that only uses soil spectra may not detect soil salinity effectively and quantitatively. The use of vegetation reflectance as an indirect indicator can avoid limitations associated with the direct use of soil reflectance. The normalized difference vegetation index （NDVI）, as the most common vegetation index, was found to be responsive to salinity but may not be available for retrieving sparse vegetation due to its sensitivity to background soil in arid areas. Therefore, the arid fraction integrated index （AFⅡ） was created as supported by the spectral mixture analysis （SMA）, which is more appropriate for analyzing variations in vegetation cover （particularly halophytes） than NDVI in the study area. Using soil and vegetation separately for detecting salinity perhaps is not feasible. Then, we developed a new and operational model, the soil salinity detecting model （SDM） that combines AFⅡ and SI to quantitatively estimate the salt content in the surface soil. SDMs, including SDM1 and SDM2, were constructed through analyzing the spatial characteristics of soils with different salinization degree by integrating AFⅡ and SI using a scatterplot. The SDMs were then compared to the combined spectral response index （COSRI） from field measurements with respect to the soil salt content. The results indicate that the SDM values are highly correlated with soil salinity, in contrast to the performance of COSRI. Strong exponential relationships were observed between soil salinity and SDMs （R2〉0.86, RMSE〈6.86） compared to COSRI （R2=0.71, RMSE=16.21）. These results suggest that the feature space related to biophysical properties combined with AFII and SI can effectively provide information on soil salinity.展开更多
The critical technical problem of underwater bottom object detection is founding a stable feature space for echo signals classification. The past literatures more focus on the characteristics of object echoes in featu...The critical technical problem of underwater bottom object detection is founding a stable feature space for echo signals classification. The past literatures more focus on the characteristics of object echoes in feature space and reverberation is only treated as interference. In this paper, reverberation is considered as a kind of signal with steady characteristic, and the clustering of reverberation in frequency discrete wavelet transform （FDWT） feature space is studied. In order to extract the identifying information of echo signals, feature compression and cluster analysis are adopted in this paper, and the criterion of separability between object echoes and reverberation is given. The experimental data processing results show that reverberation has steady pattern in FDWT feature space which differs from that of object echoes. It is proven that there is separability between reverberation and object echoes.展开更多
Spectral and spatial features in remotely sensed data play an irreplaceable role in classifying crop types for precision agriculture. Despite the thriving establishment of the handcrafted features, designing or select...Spectral and spatial features in remotely sensed data play an irreplaceable role in classifying crop types for precision agriculture. Despite the thriving establishment of the handcrafted features, designing or selecting such features valid for specific crop types requires prior knowledge and thus remains an open challenge. Convolutional neural networks(CNNs) can effectively overcome this issue with their advanced ability to generate high-level features automatically but are still inadequate in mining spectral features compared to mining spatial features. This study proposed an enhanced spectral feature called Stacked Spectral Feature Space Patch(SSFSP) for CNN-based crop classification. SSFSP is a stack of twodimensional(2 D) gridded spectral feature images that record various crop types’ spatial and intensity distribution characteristics in a 2 D feature space consisting of two spectral bands. SSFSP can be input into2 D-CNNs to support the simultaneous mining of spectral and spatial features, as the spectral features are successfully converted to 2 D images that can be processed by CNN. We tested the performance of SSFSP by using it as the input to seven CNN models and one multilayer perceptron model for crop type classification compared to using conventional spectral features as input. Using high spatial resolution hyperspectral datasets at three sites, the comparative study demonstrated that SSFSP outperforms conventional spectral features regarding classification accuracy, robustness, and training efficiency. The theoretical analysis summarizes three reasons for its excellent performance. First, SSFSP mines the spectral interrelationship with feature generality, which reduces the required number of training samples.Second, the intra-class variance can be largely reduced by grid partitioning. Third, SSFSP is a highly sparse feature, which reduces the dependence on the CNN model structure and enables early and fast convergence in model training. In conclusion, SSFSP has great potential for practical crop classification in precision agriculture.展开更多
Precise control of machining deformation is crucial for improving the manufacturing quality of structural aerospace components.In the machining process,different batches of blanks have different residual stress distri...Precise control of machining deformation is crucial for improving the manufacturing quality of structural aerospace components.In the machining process,different batches of blanks have different residual stress distributions,which pose a significant challenge to machining deformation control.In this study,a reinforcement learning method for machining deformation control based on a meta-invariant feature space was developed.The proposed method uses a reinforcement-learning model to dynamically control the machining process by monitoring the deformation force.Moreover,combined with a meta-invariant feature space,the proposed method learns the internal relationship of the deformation control approaches under different stress distributions to achieve the machining deformation control of different batches of blanks.Finally,the experimental results show that the proposed method achieves better deformation control than the two existing benchmarking methods.展开更多
With the development of large scale text processing, the dimension of text feature space has become larger and larger, which has added a lot of difficulties to natural language processing. How to reduce the dimension...With the development of large scale text processing, the dimension of text feature space has become larger and larger, which has added a lot of difficulties to natural language processing. How to reduce the dimension has become a practical problem in the field. Here we present two clustering methods, i.e. concept association and concept abstract, to achieve the goal. The first refers to the keyword clustering based on the co occurrence of展开更多
Soil moisture is a critical component of the soil-plant-atmosphere continuum(SPAC)in fruit trees.However,highprecision monitoring of orchard soil moisture at the regional scale still remains a challenge.This study pre...Soil moisture is a critical component of the soil-plant-atmosphere continuum(SPAC)in fruit trees.However,highprecision monitoring of orchard soil moisture at the regional scale still remains a challenge.This study presents a two-stage feature space model to estimate root zone soil moisture using UAV remote sensing data.The results indicate that the temperature-leaf area index(TLDI)is negatively correlated with soil water content.The upper triangular space performs highly effectively for deep soil moisture inversion,with R2 values ranging from 0.56 to 0.66,RMSE between 0.20 and 0.27,and RPD from 1.25 to 1.50.Conversely,the lower triangular space yields superior results for shallow soil moisture inversion,with R2 values between 0.67 and 0.82,RMSE from 0.15 to 0.19,and RPD between 1.67 and 2.09.The results suggest that the lower triangular space is optimal for shallow soil moisture inversion,while the upper triangular space is more suited for deep soil moisture inversion.This study presents a novel approach for estimating deep soil moisture in orchards,providing a theoretical basis for improving soil moisture management.展开更多
The traditional space target detection methods mainly use the spatial characteristics of the star map to detect the targets, which can not make full use of the time domain information. This paper presents a new space ...The traditional space target detection methods mainly use the spatial characteristics of the star map to detect the targets, which can not make full use of the time domain information. This paper presents a new space moving target detection method based on time domain features. We firstly construct the time spectral data of star map, then analyze the time domain features of the main objects(target, stars and the background) in star maps, finally detect the moving targets using single pulse feature of the time domain signal. The real star map target detection experimental results show that the proposed method can effectively detect the trajectory of moving targets in the star map sequence, and the detection probability achieves 99% when the false alarm rate is about 8×10^(-5), which outperforms those of compared algorithms.展开更多
Using the view point of nonlinear science and the method of selecting numerical features of pattern recognition for reference, the physical and numerical features of precursory ground tilt data are synthetically emplo...Using the view point of nonlinear science and the method of selecting numerical features of pattern recognition for reference, the physical and numerical features of precursory ground tilt data are synthetically employed. The dynamic changes of data series are described with the numerical features in multi dimensional space and their distributive relations instead of an unique factor. The relationship between the ground tilt data and earthquake is examined through recognition and classification.展开更多
Unsupervised methods based on density representation have shown their abilities in anomaly detection,but detection performance still needs to be improved.Specifically,approaches using normalizing flows can accurately ...Unsupervised methods based on density representation have shown their abilities in anomaly detection,but detection performance still needs to be improved.Specifically,approaches using normalizing flows can accurately evaluate sample distributions,mapping normal features to the normal distribution and anomalous features outside it.Consequently,this paper proposes a Normalizing Flow-based Bidirectional Mapping Residual Network(NF-BMR).It utilizes pre-trained Convolutional Neural Networks(CNN)and normalizing flows to construct discriminative source and target domain feature spaces.Additionally,to better learn feature information in both domain spaces,we propose the Bidirectional Mapping Residual Network(BMR),which maps sample features to these two spaces for anomaly detection.The two detection spaces effectively complement each other’s deficiencies and provide a comprehensive feature evaluation from two perspectives,which leads to the improvement of detection performance.Comparative experimental results on the MVTec AD and DAGM datasets against the Bidirectional Pre-trained Feature Mapping Network(B-PFM)and other state-of-the-art methods demonstrate that the proposed approach achieves superior performance.On the MVTec AD dataset,NF-BMR achieves an average AUROC of 98.7%for all 15 categories.Especially,it achieves 100%optimal detection performance in five categories.On the DAGM dataset,the average AUROC across ten categories is 98.7%,which is very close to supervised methods.展开更多
In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.How...In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.However,the difference in the feature space between natural and adversarial examples hinders the accuracy and robustness of the model in adversarial training.This paper proposes a learnable distribution adversarial training method,aiming to construct the same distribution for training data utilizing the Gaussian mixture model.The distribution centroid is built to classify samples and constrain the distribution of the sample features.The natural and adversarial examples are pushed to the same distribution centroid to improve the accuracy and robustness of the model.The proposed method generates adversarial examples to close the distribution gap between the natural and adversarial examples through an attack algorithm explicitly designed for adversarial training.This algorithm gradually increases the accuracy and robustness of the model by scaling perturbation.Finally,the proposed method outputs the predicted labels and the distance between the sample and the distribution centroid.The distribution characteristics of the samples can be utilized to detect adversarial cases that can potentially evade the model defense.The effectiveness of the proposed method is demonstrated through comprehensive experiments.展开更多
Network embedding(NE)tries to learn the potential properties of complex networks represented in a low-dimensional feature space.However,the existing deep learningbased NE methods are time-consuming as they need to tra...Network embedding(NE)tries to learn the potential properties of complex networks represented in a low-dimensional feature space.However,the existing deep learningbased NE methods are time-consuming as they need to train a dense architecture for deep neural networks with extensive unknown weight parameters.A sparse deep autoencoder(called SPDNE)for dynamic NE is proposed,aiming to learn the network structures while preserving the node evolution with a low computational complexity.SPDNE tries to use an optimal sparse architecture to replace the fully connected architecture in the deep autoencoder while maintaining the performance of these models in the dynamic NE.Then,an adaptive simulated algorithm to find the optimal sparse architecture for the deep autoencoder is proposed.The performance of SPDNE over three dynamical NE models(i.e.sparse architecture-based deep autoencoder method,DynGEM,and ElvDNE)is evaluated on three well-known benchmark networks and five real-world networks.The experimental results demonstrate that SPDNE can reduce about 70%of weight parameters of the architecture for the deep autoencoder during the training process while preserving the performance of these dynamical NE models.The results also show that SPDNE achieves the highest accuracy on 72 out of 96 edge prediction and network reconstruction tasks compared with the state-of-the-art dynamical NE algorithms.展开更多
A novel mercer kernel based fuzzy clustering self-adaptive algorithm is presented. The mercer kernel method is introduced to the fuzzy c-means clustering. It may map implicitly the input data into the high-dimensional...A novel mercer kernel based fuzzy clustering self-adaptive algorithm is presented. The mercer kernel method is introduced to the fuzzy c-means clustering. It may map implicitly the input data into the high-dimensional feature space through the nonlinear transformation. Among other fuzzy c-means and its variants, the number of clusters is first determined. A self-adaptive algorithm is proposed. The number of clusters, which is not given in advance, can be gotten automatically by a validity measure function. Finally, experiments are given to show better performance with the method of kernel based fuzzy c-means self-adaptive algorithm.展开更多
In medical research and clinical diagnosis, automated or computer-assisted classification and retrieval methods are highly desirable to offset the high cost of manual classification and manipulation by medical experts...In medical research and clinical diagnosis, automated or computer-assisted classification and retrieval methods are highly desirable to offset the high cost of manual classification and manipulation by medical experts. To facilitate the decision-making in the health-care and the related areas, in this paper, a two-step content-based medical image retrieval algorithm is proposed. Firstly, in the preprocessing step, the image segmentation is performed to distinguish image objects, and on the basis of the ...展开更多
This paper presents a nonlinear multidimensional scaling model, called kernelized fourth quantifica- tion theory, which is an integration of kernel techniques and the fourth quantification theory. The model can deal w...This paper presents a nonlinear multidimensional scaling model, called kernelized fourth quantifica- tion theory, which is an integration of kernel techniques and the fourth quantification theory. The model can deal with the problem of mineral prediction without defining a training area. In mineral target prediction, the pre-defined statistical cells, such as grid cells, can be implicitly transformed using kernel techniques from input space to a high-dimensional feature space, where the nonlinearly separable clusters in the input space are ex- pected to be linearly separable. Then, the transformed cells in the feature space are mapped by the fourth quan- tifieation theory onto a low-dimensional scaling space, where the sealed cells can be visually clustered according to their spatial locations. At the same time, those cells, which are far away from the cluster center of the majority of the sealed cells, are recognized as anomaly cells. Finally, whether the anomaly cells can serve as mineral potential target cells can be tested by spatially superimposing the known mineral occurrences onto the anomaly ceils. A case study shows that nearly all the known mineral occurrences spatially coincide with the anomaly cells with nearly the smallest scaled coordinates in one-dimensional sealing space. In the case study, the mineral target cells delineated by the new model are similar to those predicted by the well-known WofE model.展开更多
Recognition and counting of greenhouse pests are important for monitoring and forecasting pest population dynamics.This study used image processing techniques to recognize and count whiteflies and thrips on a sticky t...Recognition and counting of greenhouse pests are important for monitoring and forecasting pest population dynamics.This study used image processing techniques to recognize and count whiteflies and thrips on a sticky trap located in a greenhouse environment.The digital images of sticky traps were collected using an image-acquisition system under different greenhouse conditions.If a single color space is used,it is difficult to segment the small pests correctly because of the detrimental effects of non-uniform illumination in complex scenarios.Therefore,a method that first segments object pests in two color spaces using the Prewitt operator in I component of the hue-saturation-intensity(HSI)color space and the Canny operator in the B component of the Lab color space was proposed.Then,the segmented results for the two-color spaces were summed and achieved 91.57%segmentation accuracy.Next,because different features of pests contribute differently to the classification of pest species,the study extracted multiple features(e.g.,color and shape features)in different color spaces for each segmented pest region to improve the recognition performance.Twenty decision trees were used to form a strong ensemble learning classifier that used a majority voting mechanism and obtains 95.73%recognition accuracy.The proposed method is a feasible and effective way to process greenhouse pest images.The system accurately recognized and counted pests in sticky trap images captured under real greenhouse conditions.展开更多
Gene selection (feature selection) is generally pertormed in gene space(feature space), where a very serious curse of dimensionality problem always existsbecause the number of genes is much larger than the number of s...Gene selection (feature selection) is generally pertormed in gene space(feature space), where a very serious curse of dimensionality problem always existsbecause the number of genes is much larger than the number of samples in gene space(G-space). This results in difficulty in modeling the data set in this space and the lowconfidence of the result of gene selection. How to find a gene subset in this case is achallenging subject. In this paper, the above G-space is transformed into its dual space,referred to as class space (C-space) such that the number of dimensions is the verynumber of classes of the samples in G-space and the number of samples in C-space isthe number of genes in G-space. it is obvious that the curse of dimensionality in C-spacedoes not exist. A new gene selection method which is based on the principle of separatingdifferent classes as far as possible is presented with the help of Principal ComponentAnalysis (PCA). The experimental results on gene selection for real data set areevaluated with Fisher criterion, weighted Fisher criterion as well as leave-one-out crossvalidation, showing that the method presented here is effective and efficient.展开更多
Deep Neural Networks(DNNs)have become the tool of choice for machine learning practitioners today.One important aspect of designing a neural network is the choice of the activation function to be used at the neurons o...Deep Neural Networks(DNNs)have become the tool of choice for machine learning practitioners today.One important aspect of designing a neural network is the choice of the activation function to be used at the neurons of the different layers.In this work,we introduce a four-output activation function called the Reflected Rectified Linear Unit(RRe LU)activation which considers both a feature and its negation during computation.Our activation function is"sparse",in that only two of the four possible outputs are active at a given time.We test our activation function on the standard MNIST and CIFAR-10 datasets,which are classification problems,as well as on a novel Computational Fluid Dynamics(CFD)dataset which is posed as a regression problem.On the baseline network for the MNIST dataset,having two hidden layers,our activation function improves the validation accuracy from 0.09 to 0.97 compared to the well-known Re LU activation.For the CIFAR-10 dataset,we use a deep baseline network that achieves 0.78 validation accuracy with 20 epochs but overfits the data.Using the RRe LU activation,we can achieve the same accuracy without overfitting the data.For the CFD dataset,we show that the RRe LU activation can reduce the number of epochs from 100(using Re LU)to 10 while obtaining the same levels of performance.展开更多
In order to effectively utilize the dielectric response characteristics of transformers to diagnose the insulation state,this paper proposes a two-level hybrid optimization method for analyzing time-domain dielectric ...In order to effectively utilize the dielectric response characteristics of transformers to diagnose the insulation state,this paper proposes a two-level hybrid optimization method for analyzing time-domain dielectric response characteristics.The optimization algorithm is based on the combined statistical indicators(CSI)and random forest(RF)theory.The initial feature space set is formed with 23 time-domain characteristics.In the first-level stage,statistical indices correlation,distance,and information indicators are integrated to assess the synthesis score of the characteristics,while highly redundant and lowclass discrimination characteristics are eliminated from the initial space set.In the second-level stage,the Random Forest based outside bagging data theory is introduced to evaluate the least important characteristics,and the characteristics with low importance indices are excluded to obtain the final optimal feature space set.The proposed method is carried out on 82 sets of data from actual dielectric response tests on oil-paper insulation transformers.Finally,the final optimal feature space set,along with several other data sets,is tested via different diagnosis methods.The results show that the optimal feature space set obtained via the proposed method outperforms other feature space sets in terms of better adaptability and diagnosis accuracy.展开更多
In this paper,improvement on man-computer interactive classification of clouds based on hispeetral satellite imagery has been synthesized by using the maximum likelihood automatic clustering(MLAC)and the unit feature ...In this paper,improvement on man-computer interactive classification of clouds based on hispeetral satellite imagery has been synthesized by using the maximum likelihood automatic clustering(MLAC)and the unit feature space classification(UFSC)approaches.The improved classification not only shortens the time of sample-training in UFSC method,but also eliminates the inevitable shortcomings of the MLAC method.(e.g.,1.sample selecting and training is confined only to one cloud image:2.the result of clustering is pretty sensitive to the selection of initial cluster center:3.the actual classification basically can not satisfy the supposition of normal distribution required by MLAC method;4.errors in classification are difficult to be modified.) Moreover,it makes full use of the professionals'accumulated knowledge and experience of visual cloud classifications and the cloud report of ground observation,having ensured both the higher accuracy of classification and its wide application as well.展开更多
基金supported by the Key Program of the National Natural Science Foundation of China(Grant No.62031013)Guangdong Province Key Construction Discipline Scientific Research Capacity Improvement Project(Grant No.2022ZDJS117).
文摘Nonlinear transforms have significantly advanced learned image compression(LIC),particularly using residual blocks.This transform enhances the nonlinear expression ability and obtain compact feature representation by enlarging the receptive field,which indicates how the convolution process extracts features in a high dimensional feature space.However,its functionality is restricted to the spatial dimension and network depth,limiting further improvements in network performance due to insufficient information interaction and representation.Crucially,the potential of high dimensional feature space in the channel dimension and the exploration of network width/resolution remain largely untapped.In this paper,we consider nonlinear transforms from the perspective of feature space,defining high-dimensional feature spaces in different dimensions and investigating the specific effects.Firstly,we introduce the dimension increasing and decreasing transforms in both channel and spatial dimensions to obtain high dimensional feature space and achieve better feature extraction.Secondly,we design a channel-spatial fusion residual transform(CSR),which incorporates multi-dimensional transforms for a more effective representation.Furthermore,we simplify the proposed fusion transform to obtain a slim architecture(CSR-sm),balancing network complexity and compression performance.Finally,we build the overall network with stacked CSR transforms to achieve better compression and reconstruction.Experimental results demonstrate that the proposed method can achieve superior ratedistortion performance compared to the existing LIC methods and traditional codecs.Specifically,our proposed method achieves 9.38%BD-rate reduction over VVC on Kodak dataset.
基金financially supported by the National Basic Research Program of China (2009CB825105)the National Natural Science Foundation of China (41261090)
文摘Modeling soil salinity in an arid salt-affected ecosystem is a difficult task when using remote sensing data because of the complicated soil context （vegetation cover, moisture, surface roughness, and organic matter） and the weak spectral features of salinized soil. Therefore, an index such as the salinity index （SI） that only uses soil spectra may not detect soil salinity effectively and quantitatively. The use of vegetation reflectance as an indirect indicator can avoid limitations associated with the direct use of soil reflectance. The normalized difference vegetation index （NDVI）, as the most common vegetation index, was found to be responsive to salinity but may not be available for retrieving sparse vegetation due to its sensitivity to background soil in arid areas. Therefore, the arid fraction integrated index （AFⅡ） was created as supported by the spectral mixture analysis （SMA）, which is more appropriate for analyzing variations in vegetation cover （particularly halophytes） than NDVI in the study area. Using soil and vegetation separately for detecting salinity perhaps is not feasible. Then, we developed a new and operational model, the soil salinity detecting model （SDM） that combines AFⅡ and SI to quantitatively estimate the salt content in the surface soil. SDMs, including SDM1 and SDM2, were constructed through analyzing the spatial characteristics of soils with different salinization degree by integrating AFⅡ and SI using a scatterplot. The SDMs were then compared to the combined spectral response index （COSRI） from field measurements with respect to the soil salt content. The results indicate that the SDM values are highly correlated with soil salinity, in contrast to the performance of COSRI. Strong exponential relationships were observed between soil salinity and SDMs （R2〉0.86, RMSE〈6.86） compared to COSRI （R2=0.71, RMSE=16.21）. These results suggest that the feature space related to biophysical properties combined with AFII and SI can effectively provide information on soil salinity.
基金Supported by the National Natural Science Foundation of China, under Grant No.51279033.
文摘The critical technical problem of underwater bottom object detection is founding a stable feature space for echo signals classification. The past literatures more focus on the characteristics of object echoes in feature space and reverberation is only treated as interference. In this paper, reverberation is considered as a kind of signal with steady characteristic, and the clustering of reverberation in frequency discrete wavelet transform （FDWT） feature space is studied. In order to extract the identifying information of echo signals, feature compression and cluster analysis are adopted in this paper, and the criterion of separability between object echoes and reverberation is given. The experimental data processing results show that reverberation has steady pattern in FDWT feature space which differs from that of object echoes. It is proven that there is separability between reverberation and object echoes.
基金supported by the National Natural Science Foundation of China (67441830108 and 41871224)。
文摘Spectral and spatial features in remotely sensed data play an irreplaceable role in classifying crop types for precision agriculture. Despite the thriving establishment of the handcrafted features, designing or selecting such features valid for specific crop types requires prior knowledge and thus remains an open challenge. Convolutional neural networks(CNNs) can effectively overcome this issue with their advanced ability to generate high-level features automatically but are still inadequate in mining spectral features compared to mining spatial features. This study proposed an enhanced spectral feature called Stacked Spectral Feature Space Patch(SSFSP) for CNN-based crop classification. SSFSP is a stack of twodimensional(2 D) gridded spectral feature images that record various crop types’ spatial and intensity distribution characteristics in a 2 D feature space consisting of two spectral bands. SSFSP can be input into2 D-CNNs to support the simultaneous mining of spectral and spatial features, as the spectral features are successfully converted to 2 D images that can be processed by CNN. We tested the performance of SSFSP by using it as the input to seven CNN models and one multilayer perceptron model for crop type classification compared to using conventional spectral features as input. Using high spatial resolution hyperspectral datasets at three sites, the comparative study demonstrated that SSFSP outperforms conventional spectral features regarding classification accuracy, robustness, and training efficiency. The theoretical analysis summarizes three reasons for its excellent performance. First, SSFSP mines the spectral interrelationship with feature generality, which reduces the required number of training samples.Second, the intra-class variance can be largely reduced by grid partitioning. Third, SSFSP is a highly sparse feature, which reduces the dependence on the CNN model structure and enables early and fast convergence in model training. In conclusion, SSFSP has great potential for practical crop classification in precision agriculture.
基金This work is supported by National Key R&D Programs of China,No.2021YFB3301302the National Natural Science Foundation of China,No.52175467the National Science Fund of China for Distinguished Young Scholars,No.51925505。
文摘Precise control of machining deformation is crucial for improving the manufacturing quality of structural aerospace components.In the machining process,different batches of blanks have different residual stress distributions,which pose a significant challenge to machining deformation control.In this study,a reinforcement learning method for machining deformation control based on a meta-invariant feature space was developed.The proposed method uses a reinforcement-learning model to dynamically control the machining process by monitoring the deformation force.Moreover,combined with a meta-invariant feature space,the proposed method learns the internal relationship of the deformation control approaches under different stress distributions to achieve the machining deformation control of different batches of blanks.Finally,the experimental results show that the proposed method achieves better deformation control than the two existing benchmarking methods.
文摘With the development of large scale text processing, the dimension of text feature space has become larger and larger, which has added a lot of difficulties to natural language processing. How to reduce the dimension has become a practical problem in the field. Here we present two clustering methods, i.e. concept association and concept abstract, to achieve the goal. The first refers to the keyword clustering based on the co occurrence of
基金supported by the National Natural Science Foundation of China(Grant No.52309050)Key R&D and Promotion Projects in Henan Province(Science and Technology Development)(Grant No.232102110264)+2 种基金Key Scientific Research Projects of Colleges and Universities in Henan Province(Grant No.24B416001)Youth Backbone Teacher Project of Henan University of Science and Technology(Grants No.13450013 and 13450010)Sichuan Province Science and Technology Support Projects(Grants No.2023YFQ0025,2024YFHZ0217,2024ZHCG0101,2024YFHZ0200).
文摘Soil moisture is a critical component of the soil-plant-atmosphere continuum(SPAC)in fruit trees.However,highprecision monitoring of orchard soil moisture at the regional scale still remains a challenge.This study presents a two-stage feature space model to estimate root zone soil moisture using UAV remote sensing data.The results indicate that the temperature-leaf area index(TLDI)is negatively correlated with soil water content.The upper triangular space performs highly effectively for deep soil moisture inversion,with R2 values ranging from 0.56 to 0.66,RMSE between 0.20 and 0.27,and RPD from 1.25 to 1.50.Conversely,the lower triangular space yields superior results for shallow soil moisture inversion,with R2 values between 0.67 and 0.82,RMSE from 0.15 to 0.19,and RPD between 1.67 and 2.09.The results suggest that the lower triangular space is optimal for shallow soil moisture inversion,while the upper triangular space is more suited for deep soil moisture inversion.This study presents a novel approach for estimating deep soil moisture in orchards,providing a theoretical basis for improving soil moisture management.
基金supported by the National High Technology Research and Development Program of China(No.2011AAXXX2035)the Third Phase of Innovative Engineering Projects Foundation of the Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences(No.065X32CN60)
文摘The traditional space target detection methods mainly use the spatial characteristics of the star map to detect the targets, which can not make full use of the time domain information. This paper presents a new space moving target detection method based on time domain features. We firstly construct the time spectral data of star map, then analyze the time domain features of the main objects(target, stars and the background) in star maps, finally detect the moving targets using single pulse feature of the time domain signal. The real star map target detection experimental results show that the proposed method can effectively detect the trajectory of moving targets in the star map sequence, and the detection probability achieves 99% when the false alarm rate is about 8×10^(-5), which outperforms those of compared algorithms.
文摘Using the view point of nonlinear science and the method of selecting numerical features of pattern recognition for reference, the physical and numerical features of precursory ground tilt data are synthetically employed. The dynamic changes of data series are described with the numerical features in multi dimensional space and their distributive relations instead of an unique factor. The relationship between the ground tilt data and earthquake is examined through recognition and classification.
基金This work was supported in part by the National Key R&D Program of China 2021YFE0110500in part by the National Natural Science Foundation of China under Grant 62062021in part by the Guiyang Scientific Plan Project[2023]48-11.
文摘Unsupervised methods based on density representation have shown their abilities in anomaly detection,but detection performance still needs to be improved.Specifically,approaches using normalizing flows can accurately evaluate sample distributions,mapping normal features to the normal distribution and anomalous features outside it.Consequently,this paper proposes a Normalizing Flow-based Bidirectional Mapping Residual Network(NF-BMR).It utilizes pre-trained Convolutional Neural Networks(CNN)and normalizing flows to construct discriminative source and target domain feature spaces.Additionally,to better learn feature information in both domain spaces,we propose the Bidirectional Mapping Residual Network(BMR),which maps sample features to these two spaces for anomaly detection.The two detection spaces effectively complement each other’s deficiencies and provide a comprehensive feature evaluation from two perspectives,which leads to the improvement of detection performance.Comparative experimental results on the MVTec AD and DAGM datasets against the Bidirectional Pre-trained Feature Mapping Network(B-PFM)and other state-of-the-art methods demonstrate that the proposed approach achieves superior performance.On the MVTec AD dataset,NF-BMR achieves an average AUROC of 98.7%for all 15 categories.Especially,it achieves 100%optimal detection performance in five categories.On the DAGM dataset,the average AUROC across ten categories is 98.7%,which is very close to supervised methods.
基金supported by the National Natural Science Foundation of China(No.U21B2003,62072250,62072250,62172435,U1804263,U20B2065,61872203,71802110,61802212)the National Key R&D Program of China(No.2021QY0700)+4 种基金the Key Laboratory of Intelligent Support Technology for Complex Environments(Nanjing University of Information Science and Technology),Ministry of Education,and the Natural Science Foundation of Jiangsu Province(No.BK20200750)Open Foundation of Henan Key Laboratory of Cyberspace Situation Awareness(No.HNTS2022002)Post Graduate Research&Practice Innvoation Program of Jiangsu Province(No.KYCX200974)Open Project Fund of Shandong Provincial Key Laboratory of Computer Network(No.SDKLCN-2022-05)the Priority Academic Program Development of Jiangsu Higher Education Institutions(PAPD)Fund and Graduate Student Scientific Research Innovation Projects of Jiangsu Province(No.KYCX231359).
文摘In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.However,the difference in the feature space between natural and adversarial examples hinders the accuracy and robustness of the model in adversarial training.This paper proposes a learnable distribution adversarial training method,aiming to construct the same distribution for training data utilizing the Gaussian mixture model.The distribution centroid is built to classify samples and constrain the distribution of the sample features.The natural and adversarial examples are pushed to the same distribution centroid to improve the accuracy and robustness of the model.The proposed method generates adversarial examples to close the distribution gap between the natural and adversarial examples through an attack algorithm explicitly designed for adversarial training.This algorithm gradually increases the accuracy and robustness of the model by scaling perturbation.Finally,the proposed method outputs the predicted labels and the distance between the sample and the distribution centroid.The distribution characteristics of the samples can be utilized to detect adversarial cases that can potentially evade the model defense.The effectiveness of the proposed method is demonstrated through comprehensive experiments.
基金National Natural Science Foundation of China,Grant/Award Numbers:62173236,61876110,61806130,61976142,82304204.
文摘Network embedding(NE)tries to learn the potential properties of complex networks represented in a low-dimensional feature space.However,the existing deep learningbased NE methods are time-consuming as they need to train a dense architecture for deep neural networks with extensive unknown weight parameters.A sparse deep autoencoder(called SPDNE)for dynamic NE is proposed,aiming to learn the network structures while preserving the node evolution with a low computational complexity.SPDNE tries to use an optimal sparse architecture to replace the fully connected architecture in the deep autoencoder while maintaining the performance of these models in the dynamic NE.Then,an adaptive simulated algorithm to find the optimal sparse architecture for the deep autoencoder is proposed.The performance of SPDNE over three dynamical NE models(i.e.sparse architecture-based deep autoencoder method,DynGEM,and ElvDNE)is evaluated on three well-known benchmark networks and five real-world networks.The experimental results demonstrate that SPDNE can reduce about 70%of weight parameters of the architecture for the deep autoencoder during the training process while preserving the performance of these dynamical NE models.The results also show that SPDNE achieves the highest accuracy on 72 out of 96 edge prediction and network reconstruction tasks compared with the state-of-the-art dynamical NE algorithms.
文摘A novel mercer kernel based fuzzy clustering self-adaptive algorithm is presented. The mercer kernel method is introduced to the fuzzy c-means clustering. It may map implicitly the input data into the high-dimensional feature space through the nonlinear transformation. Among other fuzzy c-means and its variants, the number of clusters is first determined. A self-adaptive algorithm is proposed. The number of clusters, which is not given in advance, can be gotten automatically by a validity measure function. Finally, experiments are given to show better performance with the method of kernel based fuzzy c-means self-adaptive algorithm.
文摘In medical research and clinical diagnosis, automated or computer-assisted classification and retrieval methods are highly desirable to offset the high cost of manual classification and manipulation by medical experts. To facilitate the decision-making in the health-care and the related areas, in this paper, a two-step content-based medical image retrieval algorithm is proposed. Firstly, in the preprocessing step, the image segmentation is performed to distinguish image objects, and on the basis of the ...
基金supported by National Natural Science Foundation of China (No.40872193)
文摘This paper presents a nonlinear multidimensional scaling model, called kernelized fourth quantifica- tion theory, which is an integration of kernel techniques and the fourth quantification theory. The model can deal with the problem of mineral prediction without defining a training area. In mineral target prediction, the pre-defined statistical cells, such as grid cells, can be implicitly transformed using kernel techniques from input space to a high-dimensional feature space, where the nonlinearly separable clusters in the input space are ex- pected to be linearly separable. Then, the transformed cells in the feature space are mapped by the fourth quan- tifieation theory onto a low-dimensional scaling space, where the sealed cells can be visually clustered according to their spatial locations. At the same time, those cells, which are far away from the cluster center of the majority of the sealed cells, are recognized as anomaly cells. Finally, whether the anomaly cells can serve as mineral potential target cells can be tested by spatially superimposing the known mineral occurrences onto the anomaly ceils. A case study shows that nearly all the known mineral occurrences spatially coincide with the anomaly cells with nearly the smallest scaled coordinates in one-dimensional sealing space. In the case study, the mineral target cells delineated by the new model are similar to those predicted by the well-known WofE model.
基金This work was financially supported by the National Natural Science Foundation of China(Grant No.61601034)and the National Natural Science Foundation of China(Grant No.31871525)The authors acknowledge Kimberly Moravec,PhD,from Liwen Bianji,Edanz Editing China(www.liwenbianji.cn/ac),for editing the English text of a draft of this manuscript.
文摘Recognition and counting of greenhouse pests are important for monitoring and forecasting pest population dynamics.This study used image processing techniques to recognize and count whiteflies and thrips on a sticky trap located in a greenhouse environment.The digital images of sticky traps were collected using an image-acquisition system under different greenhouse conditions.If a single color space is used,it is difficult to segment the small pests correctly because of the detrimental effects of non-uniform illumination in complex scenarios.Therefore,a method that first segments object pests in two color spaces using the Prewitt operator in I component of the hue-saturation-intensity(HSI)color space and the Canny operator in the B component of the Lab color space was proposed.Then,the segmented results for the two-color spaces were summed and achieved 91.57%segmentation accuracy.Next,because different features of pests contribute differently to the classification of pest species,the study extracted multiple features(e.g.,color and shape features)in different color spaces for each segmented pest region to improve the recognition performance.Twenty decision trees were used to form a strong ensemble learning classifier that used a majority voting mechanism and obtains 95.73%recognition accuracy.The proposed method is a feasible and effective way to process greenhouse pest images.The system accurately recognized and counted pests in sticky trap images captured under real greenhouse conditions.
文摘Gene selection (feature selection) is generally pertormed in gene space(feature space), where a very serious curse of dimensionality problem always existsbecause the number of genes is much larger than the number of samples in gene space(G-space). This results in difficulty in modeling the data set in this space and the lowconfidence of the result of gene selection. How to find a gene subset in this case is achallenging subject. In this paper, the above G-space is transformed into its dual space,referred to as class space (C-space) such that the number of dimensions is the verynumber of classes of the samples in G-space and the number of samples in C-space isthe number of genes in G-space. it is obvious that the curse of dimensionality in C-spacedoes not exist. A new gene selection method which is based on the principle of separatingdifferent classes as far as possible is presented with the help of Principal ComponentAnalysis (PCA). The experimental results on gene selection for real data set areevaluated with Fisher criterion, weighted Fisher criterion as well as leave-one-out crossvalidation, showing that the method presented here is effective and efficient.
文摘Deep Neural Networks(DNNs)have become the tool of choice for machine learning practitioners today.One important aspect of designing a neural network is the choice of the activation function to be used at the neurons of the different layers.In this work,we introduce a four-output activation function called the Reflected Rectified Linear Unit(RRe LU)activation which considers both a feature and its negation during computation.Our activation function is"sparse",in that only two of the four possible outputs are active at a given time.We test our activation function on the standard MNIST and CIFAR-10 datasets,which are classification problems,as well as on a novel Computational Fluid Dynamics(CFD)dataset which is posed as a regression problem.On the baseline network for the MNIST dataset,having two hidden layers,our activation function improves the validation accuracy from 0.09 to 0.97 compared to the well-known Re LU activation.For the CIFAR-10 dataset,we use a deep baseline network that achieves 0.78 validation accuracy with 20 epochs but overfits the data.Using the RRe LU activation,we can achieve the same accuracy without overfitting the data.For the CFD dataset,we show that the RRe LU activation can reduce the number of epochs from 100(using Re LU)to 10 while obtaining the same levels of performance.
基金supported by The National Natural Science Foundation of China(61174117)the Foundation of Scientific Research Project of Jinjiang Science and Education Development Center of Fuzhou Univerdity(2019-JJFDKY-33).
文摘In order to effectively utilize the dielectric response characteristics of transformers to diagnose the insulation state,this paper proposes a two-level hybrid optimization method for analyzing time-domain dielectric response characteristics.The optimization algorithm is based on the combined statistical indicators(CSI)and random forest(RF)theory.The initial feature space set is formed with 23 time-domain characteristics.In the first-level stage,statistical indices correlation,distance,and information indicators are integrated to assess the synthesis score of the characteristics,while highly redundant and lowclass discrimination characteristics are eliminated from the initial space set.In the second-level stage,the Random Forest based outside bagging data theory is introduced to evaluate the least important characteristics,and the characteristics with low importance indices are excluded to obtain the final optimal feature space set.The proposed method is carried out on 82 sets of data from actual dielectric response tests on oil-paper insulation transformers.Finally,the final optimal feature space set,along with several other data sets,is tested via different diagnosis methods.The results show that the optimal feature space set obtained via the proposed method outperforms other feature space sets in terms of better adaptability and diagnosis accuracy.
文摘In this paper,improvement on man-computer interactive classification of clouds based on hispeetral satellite imagery has been synthesized by using the maximum likelihood automatic clustering(MLAC)and the unit feature space classification(UFSC)approaches.The improved classification not only shortens the time of sample-training in UFSC method,but also eliminates the inevitable shortcomings of the MLAC method.(e.g.,1.sample selecting and training is confined only to one cloud image:2.the result of clustering is pretty sensitive to the selection of initial cluster center:3.the actual classification basically can not satisfy the supposition of normal distribution required by MLAC method;4.errors in classification are difficult to be modified.) Moreover,it makes full use of the professionals'accumulated knowledge and experience of visual cloud classifications and the cloud report of ground observation,having ensured both the higher accuracy of classification and its wide application as well.
