Performance Evaluation of Deep Learning Classification Network for Image Features

Abstract

This is the pre-acceptance version, to read the final version please go to IEEE Geoscience and Remote Sensing Letters on IEEE Xplore. In recent years, deep learning has been widely used in SAR ATR and achieved excellent performance on the MSTAR dataset. However, due to constrained imaging conditions, MSTAR has data biases such as background correlation, i.e., background clutter properties have a spurious correlation with target classes. Deep learning can overfit clutter to reduce training errors. Therefore, the degree of overfitting for clutter reflects the non-causality of deep learning in SAR ATR. Existing methods only qualitatively analyze this phenomenon. In this paper, we quantify the contributions of different regions to target recognition based on the Shapley value. The Shapley value of clutter measures the degree of overfitting. Moreover, we explain how data bias and model bias contribute to noncausality. Concisely, data bias leads to comparable signal-toclutter ratios and clutter textures in training and test sets. And various model structures have different degrees of overfitting for these biases. The experimental results of various models under standard operating conditions on the MSTAR dataset support our conclusions. Index Terms-Synthetic aperture radar (SAR), automatic target recognition (ATR), deep learning, data bias, Shapley value, causality 2 Texture refers to the spatial organization of a set of basic elements [7]. As shown in Fig. 1, the scattering point amplitude has different spatial distributions in the target, clutter, and shadow regions. These spatial distributions create texture signatures of different regions in SAR images.

Research on image sentiment analysis technology based on sparse representation

“…By processing the data through singular value decomposition, the original data can be turned into a relatively small dataset, which actually retains the main feature values and reduces noise and partial redundant information without affecting the recognition accuracy, thereby optimizing the data and effectively reducing the experimental time. After reading the public data set of facial expression, the pixel points of the image are read successively according to the size, and the pixel points are copied successively to obtain the pixel data; the corresponding pixel values are stored in the matrix and the singular value matrix is obtained through the SVD Equation (10); Sparse emotion images are obtained according to the top K% singular values selected from the singular value matrix.…”

“…In the formula, Q is a standard orthogonal matrix, that is, QQ T ¼ I, Σ is a diagonal matrix, and the dimension of the above matrix is m � m. λ I is called the eigenvalue, and q i is the column vector of Q. A has an m � n-dimensional real number matrix, and the matrix A is decomposed into the form of Equation (10):…”

“…Firstly, this article proposes an image sentiment analysis method based on Singular Value Decomposition (SVD). Any matrix can be approximately described by the largest k singular values and the corresponding left and right singular vectors, we represent the image data in a matrix, and use SVD of the face expression dataset to obtain a sparse facial expression image; then on this basis, the Single Shot MultiBox Detector (SSD) algorithm [9] is performed to identify sparse facial expression images to obtain face detection results; finally, the face detection results are set into the VGGNet [10, 11] for facial expression classification. On the basis of the above image pre‐processing and convolutional neural network (CNN) model, combined with the K‐Singular Value Decomposition (K‐SVD) [12], we form a Fast Dictionary Learning (FDL) algorithm to efficiently extract the sparse features of the image, and then input the sparse features into the CNN to realize sentiment analysis.…”

Multi-modal data combination strategy based on chest HRCT images and PFT parameters for intelligent dyspnea identification in COPD

“…However, radiomics features are extracted from medical images by specific calculation equations, preset types of images, and preset classes, limiting the forms of radiomics features. Convolutional neural networks (CNN) based on images for classification also rapidly developed ( 23 ). Features extracted from medical images based on the CNN model will compensate for the limitations of radiomics features.…”

Analysis of Nomadic Civilization in Northern Grassland in Plastic Arts Based on Deep Learning

“…These advantages are exploited in the recognition, extraction, and analysis of image features. In particular, for the northern steppe nomadic civilization [11][12][13][14][15], its plastic arts are very rich, including color, texture, shape, and local features, and deep learning can replace tra-ditional methods to build deep and shallow networks as its input and feature recognition, which solves the problem of image feature recognition. At the same time, the convolution idea is introduced to enlarge its features, which is more conducive to feature recognition, extraction, and analysis.…”