In response to the scarcity of infrared aircraft samples and the tendency of traditional deep learning to overfit,a few-shot infrared aircraft classification method based on cross-correlation networks is proposed.This...In response to the scarcity of infrared aircraft samples and the tendency of traditional deep learning to overfit,a few-shot infrared aircraft classification method based on cross-correlation networks is proposed.This method combines two core modules:a simple parameter-free self-attention and cross-attention.By analyzing the self-correlation and cross-correlation between support images and query images,it achieves effective classification of infrared aircraft under few-shot conditions.The proposed cross-correlation network integrates these two modules and is trained in an end-to-end manner.The simple parameter-free self-attention is responsible for extracting the internal structure of the image while the cross-attention can calculate the cross-correlation between images further extracting and fusing the features between images.Compared with existing few-shot infrared target classification models,this model focuses on the geometric structure and thermal texture information of infrared images by modeling the semantic relevance between the features of the support set and query set,thus better attending to the target objects.Experimental results show that this method outperforms existing infrared aircraft classification methods in various classification tasks,with the highest classification accuracy improvement exceeding 3%.In addition,ablation experiments and comparative experiments also prove the effectiveness of the method.展开更多
How to recognize targets with similar appearances from remote sensing images(RSIs) effectively and efficiently has become a big challenge. Recently, convolutional neural network(CNN) is preferred in the target classif...How to recognize targets with similar appearances from remote sensing images(RSIs) effectively and efficiently has become a big challenge. Recently, convolutional neural network(CNN) is preferred in the target classification due to the powerful feature representation ability and better performance. However,the training and testing of CNN mainly rely on single machine.Single machine has its natural limitation and bottleneck in processing RSIs due to limited hardware resources and huge time consuming. Besides, overfitting is a challenge for the CNN model due to the unbalance between RSIs data and the model structure.When a model is complex or the training data is relatively small,overfitting occurs and leads to a poor predictive performance. To address these problems, a distributed CNN architecture for RSIs target classification is proposed, which dramatically increases the training speed of CNN and system scalability. It improves the storage ability and processing efficiency of RSIs. Furthermore,Bayesian regularization approach is utilized in order to initialize the weights of the CNN extractor, which increases the robustness and flexibility of the CNN model. It helps prevent the overfitting and avoid the local optima caused by limited RSI training images or the inappropriate CNN structure. In addition, considering the efficiency of the Na¨?ve Bayes classifier, a distributed Na¨?ve Bayes classifier is designed to reduce the training cost. Compared with other algorithms, the proposed system and method perform the best and increase the recognition accuracy. The results show that the distributed system framework and the proposed algorithms are suitable for RSIs target classification tasks.展开更多
With limited number of labeled samples,hyperspectral image(HSI)classification is a difficult Problem in current research.The graph neural network(GNN)has emerged as an approach to semi-supervised classification,and th...With limited number of labeled samples,hyperspectral image(HSI)classification is a difficult Problem in current research.The graph neural network(GNN)has emerged as an approach to semi-supervised classification,and the application of GNN to hyperspectral images has attracted much attention.However,in the existing GNN-based methods a single graph neural network or graph filter is mainly used to extract HSI features,which does not take full advantage of various graph neural networks(graph filters).Moreover,the traditional GNNs have the problem of oversmoothing.To alleviate these shortcomings,we introduce a deep hybrid multi-graph neural network(DHMG),where two different graph filters,i.e.,the spectral filter and the autoregressive moving average(ARMA)filter,are utilized in two branches.The former can well extract the spectral features of the nodes,and the latter has a good suppression effect on graph noise.The network realizes information interaction between the two branches and takes good advantage of different graph filters.In addition,to address the problem of oversmoothing,a dense network is proposed,where the local graph features are preserved.The dense structure satisfies the needs of different classification targets presenting different features.Finally,we introduce a GraphSAGEbased network to refine the graph features produced by the deep hybrid network.Extensive experiments on three public HSI datasets strongly demonstrate that the DHMG dramatically outperforms the state-ofthe-art models.展开更多
To study the scene classification in the Synthetic Aperture Radar （SAR） image, a novel method based on kernel estimate, with the Maxkov context and Dempster-Shafer evidence theory is proposed. Initially, a nonpaxame...To study the scene classification in the Synthetic Aperture Radar （SAR） image, a novel method based on kernel estimate, with the Maxkov context and Dempster-Shafer evidence theory is proposed. Initially, a nonpaxametric Probability Density Function （PDF） estimate method is introduced, to describe the scene of SAR images. And then under the Maxkov context, both the determinate PDF and the kernel estimate method axe adopted respectively, to form a primary classification. Next, the primary classification results are fused using the evidence theory in an unsupervised way to get the scene classification. Finally, a regularization step is used, in which an iterated maximum selecting approach is introduced to control the fragments and modify the errors of the classification. Use of the kernel estimate and evidence theory can describe the complicated scenes with little prior knowledge and eliminate the ambiguities of the primary classification results. Experimental results on real SAR images illustrate a rather impressive performance.展开更多
Automatic image classification is the first step toward semantic understanding of an object in the computer vision area.The key challenge of problem for accurate object recognition is the ability to extract the robust...Automatic image classification is the first step toward semantic understanding of an object in the computer vision area.The key challenge of problem for accurate object recognition is the ability to extract the robust features from various viewpoint images and rapidly calculate similarity between features in the image database or video stream.In order to solve these problems,an effective and rapid image classification method was presented for the object recognition based on the video learning technique.The optical-flow and RANSAC algorithm were used to acquire scene images from each video sequence.After the selection of scene images,the local maximum points on comer of object around local area were found using the Harris comer detection algorithm and the several attributes from local block around each feature point were calculated by using scale invariant feature transform （SIFT） for extracting local descriptor.Finally,the extracted local descriptor was learned to the three-dimensional pyramid match kernel.Experimental results show that our method can extract features in various multi-viewpoint images from query video and calculate a similarity between a query image and images in the database.展开更多
This paper presents a novel efficient semantic image classification algorithm for high-level feature indexing of high-dimension image database. Experiments show that the algorithm performs well. The size of the train ...This paper presents a novel efficient semantic image classification algorithm for high-level feature indexing of high-dimension image database. Experiments show that the algorithm performs well. The size of the train set and the test set is 7 537 and 5 000 respectively. Based on this theory, another ground is built with 12,000 images, which are divided into three classes: city, landscape and person, the total result of the classifications is 88.92%, meanwhile, some preliminary results are presented for image understanding based on semantic image classification and low level features. The groundtruth for the experiments is built with the images from Corel database, photos and some famous face databases.展开更多
The document image segmentation is very useful for printing, faxing and data processing. An algorithm is developed for segmenting and classifying document image. Feature used for classification is based on the histogr...The document image segmentation is very useful for printing, faxing and data processing. An algorithm is developed for segmenting and classifying document image. Feature used for classification is based on the histogram distribution pattern of different image classes. The important attribute of the algorithm is using wavelet correlation image to enhance raw image's pattern, so the classification accuracy is improved. In this paper document image is divided into four types; background, photo, text and graph. Firstly, the document image background has been distingusished easily by former normally method;secondly, three image types will be distinguished by their typical histograms, in order to make histograms feature clearer, each resolution's HH wavelet subimage is used to add to the raw image at their resolution. At last, the photo, text and praph have been devided according to how the feature fit to the Laplacian distrbution by 2 and L . Simulations show that classification accuracy is significantly improved. The comparison with related shows that our algorithm provides both lower classification error rates and better visual results.展开更多
The aim of modulation classification （MC） is to identify the modulation type of a commtmication signal. It plays an important role in many cooperative or noncooperative communication applications. Three spectrogram-...The aim of modulation classification （MC） is to identify the modulation type of a commtmication signal. It plays an important role in many cooperative or noncooperative communication applications. Three spectrogram-based modulation classification methods are proposed. Their recognition scope and performance are investigated or evaluated by theoretical analysis and extensive simulation studies. The method taking moment-like features is robust to frequency offset while the other two, which make use of principal component analysis （PCA） with different transformation inputs, can achieve satisfactory accuracy even at low SNR （as low as 2 dB）. Due to the properties of spectrogram, the statistical pattern recognition techniques, and the image preprocessing steps, all of our methods are insensitive to unknown phase and frequency offsets, timing errors, and the arriving sequence of symbols.展开更多
为精准识别与分类不同花期杭白菊,满足自动化采摘要求,该研究提出一种基于改进YOLOv8s的杭白菊检测模型-YOLOv8s-RDL。首先,该研究将颈部网络(neck)的C2f(faster implementation of CSP bottleneck with 2 convolutions)模块替换为RCS-O...为精准识别与分类不同花期杭白菊,满足自动化采摘要求,该研究提出一种基于改进YOLOv8s的杭白菊检测模型-YOLOv8s-RDL。首先,该研究将颈部网络(neck)的C2f(faster implementation of CSP bottleneck with 2 convolutions)模块替换为RCS-OSA(one-shot aggregation of reparameterized convolution based on channel shuffle)模块,以提升骨干网络(backbone)特征融合效率;其次,将检测头更换为DyHead(dynamic head),并融合DCNv3(deformable convolutional networks v3),借助多头自注意力机制增强目标检测头的表达能力;最后,采用LAMP(layer-adaptive magnitude-based pruning)通道剪枝算法减少参数量,降低模型复杂度。试验结果表明,YOLOv8s-RDL模型在菊米和胎菊的花期分类中平均精度分别达到96.3%和97.7%,相较于YOLOv8s模型,分别提升了3.8和1.5个百分点,同时权重文件大小较YOLOv8s减小了6 MB。该研究引入TIDE(toolkit for identifying detection and segmentation errors)评估指标,结果显示,YOLOv8s-RDL模型分类错误和背景检测错误相较YOLOv8s模型分别降低0.55和1.26。该研究为杭白菊分花期自动化采摘提供了理论依据和技术支撑。展开更多
基金Supported by the National Pre-research Program during the 14th Five-Year Plan(514010405)。
文摘In response to the scarcity of infrared aircraft samples and the tendency of traditional deep learning to overfit,a few-shot infrared aircraft classification method based on cross-correlation networks is proposed.This method combines two core modules:a simple parameter-free self-attention and cross-attention.By analyzing the self-correlation and cross-correlation between support images and query images,it achieves effective classification of infrared aircraft under few-shot conditions.The proposed cross-correlation network integrates these two modules and is trained in an end-to-end manner.The simple parameter-free self-attention is responsible for extracting the internal structure of the image while the cross-attention can calculate the cross-correlation between images further extracting and fusing the features between images.Compared with existing few-shot infrared target classification models,this model focuses on the geometric structure and thermal texture information of infrared images by modeling the semantic relevance between the features of the support set and query set,thus better attending to the target objects.Experimental results show that this method outperforms existing infrared aircraft classification methods in various classification tasks,with the highest classification accuracy improvement exceeding 3%.In addition,ablation experiments and comparative experiments also prove the effectiveness of the method.
基金supported by the National Natural Science Foundation of China(U1435220)
文摘How to recognize targets with similar appearances from remote sensing images(RSIs) effectively and efficiently has become a big challenge. Recently, convolutional neural network(CNN) is preferred in the target classification due to the powerful feature representation ability and better performance. However,the training and testing of CNN mainly rely on single machine.Single machine has its natural limitation and bottleneck in processing RSIs due to limited hardware resources and huge time consuming. Besides, overfitting is a challenge for the CNN model due to the unbalance between RSIs data and the model structure.When a model is complex or the training data is relatively small,overfitting occurs and leads to a poor predictive performance. To address these problems, a distributed CNN architecture for RSIs target classification is proposed, which dramatically increases the training speed of CNN and system scalability. It improves the storage ability and processing efficiency of RSIs. Furthermore,Bayesian regularization approach is utilized in order to initialize the weights of the CNN extractor, which increases the robustness and flexibility of the CNN model. It helps prevent the overfitting and avoid the local optima caused by limited RSI training images or the inappropriate CNN structure. In addition, considering the efficiency of the Na¨?ve Bayes classifier, a distributed Na¨?ve Bayes classifier is designed to reduce the training cost. Compared with other algorithms, the proposed system and method perform the best and increase the recognition accuracy. The results show that the distributed system framework and the proposed algorithms are suitable for RSIs target classification tasks.
文摘With limited number of labeled samples,hyperspectral image(HSI)classification is a difficult Problem in current research.The graph neural network(GNN)has emerged as an approach to semi-supervised classification,and the application of GNN to hyperspectral images has attracted much attention.However,in the existing GNN-based methods a single graph neural network or graph filter is mainly used to extract HSI features,which does not take full advantage of various graph neural networks(graph filters).Moreover,the traditional GNNs have the problem of oversmoothing.To alleviate these shortcomings,we introduce a deep hybrid multi-graph neural network(DHMG),where two different graph filters,i.e.,the spectral filter and the autoregressive moving average(ARMA)filter,are utilized in two branches.The former can well extract the spectral features of the nodes,and the latter has a good suppression effect on graph noise.The network realizes information interaction between the two branches and takes good advantage of different graph filters.In addition,to address the problem of oversmoothing,a dense network is proposed,where the local graph features are preserved.The dense structure satisfies the needs of different classification targets presenting different features.Finally,we introduce a GraphSAGEbased network to refine the graph features produced by the deep hybrid network.Extensive experiments on three public HSI datasets strongly demonstrate that the DHMG dramatically outperforms the state-ofthe-art models.
基金the National Nature Science Foundation of China (60372057).
文摘To study the scene classification in the Synthetic Aperture Radar （SAR） image, a novel method based on kernel estimate, with the Maxkov context and Dempster-Shafer evidence theory is proposed. Initially, a nonpaxametric Probability Density Function （PDF） estimate method is introduced, to describe the scene of SAR images. And then under the Maxkov context, both the determinate PDF and the kernel estimate method axe adopted respectively, to form a primary classification. Next, the primary classification results are fused using the evidence theory in an unsupervised way to get the scene classification. Finally, a regularization step is used, in which an iterated maximum selecting approach is introduced to control the fragments and modify the errors of the classification. Use of the kernel estimate and evidence theory can describe the complicated scenes with little prior knowledge and eliminate the ambiguities of the primary classification results. Experimental results on real SAR images illustrate a rather impressive performance.
文摘Automatic image classification is the first step toward semantic understanding of an object in the computer vision area.The key challenge of problem for accurate object recognition is the ability to extract the robust features from various viewpoint images and rapidly calculate similarity between features in the image database or video stream.In order to solve these problems,an effective and rapid image classification method was presented for the object recognition based on the video learning technique.The optical-flow and RANSAC algorithm were used to acquire scene images from each video sequence.After the selection of scene images,the local maximum points on comer of object around local area were found using the Harris comer detection algorithm and the several attributes from local block around each feature point were calculated by using scale invariant feature transform （SIFT） for extracting local descriptor.Finally,the extracted local descriptor was learned to the three-dimensional pyramid match kernel.Experimental results show that our method can extract features in various multi-viewpoint images from query video and calculate a similarity between a query image and images in the database.
文摘This paper presents a novel efficient semantic image classification algorithm for high-level feature indexing of high-dimension image database. Experiments show that the algorithm performs well. The size of the train set and the test set is 7 537 and 5 000 respectively. Based on this theory, another ground is built with 12,000 images, which are divided into three classes: city, landscape and person, the total result of the classifications is 88.92%, meanwhile, some preliminary results are presented for image understanding based on semantic image classification and low level features. The groundtruth for the experiments is built with the images from Corel database, photos and some famous face databases.
文摘The document image segmentation is very useful for printing, faxing and data processing. An algorithm is developed for segmenting and classifying document image. Feature used for classification is based on the histogram distribution pattern of different image classes. The important attribute of the algorithm is using wavelet correlation image to enhance raw image's pattern, so the classification accuracy is improved. In this paper document image is divided into four types; background, photo, text and graph. Firstly, the document image background has been distingusished easily by former normally method;secondly, three image types will be distinguished by their typical histograms, in order to make histograms feature clearer, each resolution's HH wavelet subimage is used to add to the raw image at their resolution. At last, the photo, text and praph have been devided according to how the feature fit to the Laplacian distrbution by 2 and L . Simulations show that classification accuracy is significantly improved. The comparison with related shows that our algorithm provides both lower classification error rates and better visual results.
文摘The aim of modulation classification （MC） is to identify the modulation type of a commtmication signal. It plays an important role in many cooperative or noncooperative communication applications. Three spectrogram-based modulation classification methods are proposed. Their recognition scope and performance are investigated or evaluated by theoretical analysis and extensive simulation studies. The method taking moment-like features is robust to frequency offset while the other two, which make use of principal component analysis （PCA） with different transformation inputs, can achieve satisfactory accuracy even at low SNR （as low as 2 dB）. Due to the properties of spectrogram, the statistical pattern recognition techniques, and the image preprocessing steps, all of our methods are insensitive to unknown phase and frequency offsets, timing errors, and the arriving sequence of symbols.
文摘为精准识别与分类不同花期杭白菊,满足自动化采摘要求,该研究提出一种基于改进YOLOv8s的杭白菊检测模型-YOLOv8s-RDL。首先,该研究将颈部网络(neck)的C2f(faster implementation of CSP bottleneck with 2 convolutions)模块替换为RCS-OSA(one-shot aggregation of reparameterized convolution based on channel shuffle)模块,以提升骨干网络(backbone)特征融合效率;其次,将检测头更换为DyHead(dynamic head),并融合DCNv3(deformable convolutional networks v3),借助多头自注意力机制增强目标检测头的表达能力;最后,采用LAMP(layer-adaptive magnitude-based pruning)通道剪枝算法减少参数量,降低模型复杂度。试验结果表明,YOLOv8s-RDL模型在菊米和胎菊的花期分类中平均精度分别达到96.3%和97.7%,相较于YOLOv8s模型,分别提升了3.8和1.5个百分点,同时权重文件大小较YOLOv8s减小了6 MB。该研究引入TIDE(toolkit for identifying detection and segmentation errors)评估指标,结果显示,YOLOv8s-RDL模型分类错误和背景检测错误相较YOLOv8s模型分别降低0.55和1.26。该研究为杭白菊分花期自动化采摘提供了理论依据和技术支撑。
