Research On SAR Image Classification And Recognition Based On Machine Learning

Synthetic Aperture Radar (SAR) can work in all weather, all day and night. Also, its ability to detect targets through cloud and plant covers makes SAR an important technique in both military and civil use. Single-polarization SAR gains its advantage in simpler structure and smaller computational work. Machine learning is a research hotspot in artificial intelligence area. It’s being widely used and achieved great success in image processing area. In this decade, deep learning, as a part of machine learning, has gained more and more attention with the growth spurts of computation ability and becoming one of the most active fields of Computer Vision. This paper proposes an innovative method in which deep learning algorithm being used in SAR image recognition application.This paper comes up with a SAR image classification system based on machine learning method, and a recognition system based on deep learning method. Experiments conducted to verify the proposed scheme. This paper starts with the analysis of SAR imaging model, SAR image features as well as several de-noise algorithms with the comparing experiments. Robust feature extraction algorithms are necessities concerning the interference of speckles. This paper uses Hu moment and SIFT features based on different scenes to extract SAR image features. Experiments are followed to show the robustness of proposed methods. Then, this paper presents an image classification system based on SVM for complex background situation. Spatial pyramid model and Histogram intersection kernel were brought into the system design. The experiment results verify the proposed design. Finally, this paper proposes a SAR image recognition system based on Convolutional Neural Network. MSTAR data is used to test and analyze this system. The result indicates that over 95% accuracy has been achieved in recognition experiment.