A Method of SAR Image Automatic Target Recognition Based on Convolution Auto-Encode and Support Vector Machine

In this paper, a method of Synthetic Aperture Radar (SAR) image Automatic Target Recognition (ATR) based on Convolution Auto-encode (CAE) and Support Vector Machine (SVM) is proposed. Using SVM replaces the traditional softmax as classifier CAE model to classify feature vectors extracted by model, which solves problem that less effective in nonlinear case. Since can only solve binary classification problem, order realize class objectives, were designed achieve input samples. After unsupervised training for CAE, coding layer connected with form network. extract features data an method, advantage improve accuracy object recognition. At same time, high-accuracy identification key targets required some special cases. A new initialization proposed, initializes network parameters pretraining changes weights different loss function obtain better extraction, so it ensure good multitarget recognition ability while realizing high targets.