Three Dimensional Parametric Electromagnetic Part Model based SAR ATR using Convolution Neural Network

Recognizing targets in synthetic aperture radar (SAR) images is an important, yet challenging problem for SAR image interpretation. In traditional methods, recognition is fulfilled through match the measured image to template or model images directly or with the extraction and correspondence of designed features. It may not perform well on unforeseen target types under the extended operation condition (EOC). Three dimensional parametric electromagnetic part model (3-D PEPM) characterizes a target's scattering phenomenon by the combination of representative scattering parts which can be manipulated to simulate a target's scattering in different operation conditions (OCs). Convolution neural network (CNN) learns rich hierarchical features from massive training data automatically. It has shown the state-of-the-art power in many classification and recognition tasks. In this paper, a 3-D PEPM based method is proposed for the recognition of SAR images by using CNN. Firstly, 3-D PEPM is introduced to simulate the scattering of the target in different OCs. Then, the model generated image samples are used to train a CNN model for the recognition. In the test stage, the test image is recognized by the trained model as the class with the highest probability. In the end, experiments using electromagnetic simulated data and chamber measured data verify the effectiveness of the proposed method and show that it is a practical approach for target recognition in EOCs.