Polarimetric SAR Targets Detection and Classification with Deep Convolutional Neural Network

Abstract

Target detection and classification are important applications for polarimetric SAR (PolSAR). Advanced deep learning techniques represented by deep convolutional neural network (CNN) have been utilized to enhance the application performance. One current challenge is how to adapt deep CNN model for PolSAR target detection and classification with limited training samples while keeping good generalization performance. This work attempts to contribute to this problem. The core idea is to incorporate expert knowledge of target scattering mechanism interpretation and polarimetric feature mining to assist deep CNN model training and improve the final application performance. A polarimetric-feature-driven deep CNN detection and classification scheme is established. Both classical polarimetric features and hidden polarimetric features in the rotation domain are used to drive the proposed deep CNN model. Comparison studies validate the efficiency and superiority of the proposal. For ship detection application, two methods respectively driven by classical polarimetric features and the selected polarimetric features show relatively good performances, with detection probabilities over 90.8% for Radarsat-2 data. Considering the overall detection and false-alarm probabilities, the proposed polarimetric-feature-driven CNN approach achieves even better performance. For land cover classification, the proposed method achieves the state of the art classification accuracy for the benchmark AIRSAR data. Meanwhile, the convergence speed from the proposed polarimetric-feature-driven CNN approach is about 2.3 times faster than the normal CNN method. For multi-temporal UAVSAR datasets, the proposed scheme achieves comparably high classification accuracy as the normal CNN method for train-used temporal data, while for train-not-used data it obtains average 4.86% higher overall accuracy than the normal CNN method.