Integrated Convolution Network for ISAR Imaging and Target Recognition

Abstract

Recently, inverse synthetic aperture radar (ISAR) image recognition using deep learning (DL) technology is developed rapidly. However, the imaging and recognition processing is independent of each other, and the recognition network cannot fully capture target features from the radar data. Accordingly, this article proposes an integrated convolution network for ISAR imaging and target recognition, named IITR-Net. In the scheme, a DL imaging module is designed for ISAR imaging instead of using the traditional imaging algorithms, which can be cascaded with the recognition network. Thus, the proposed IITR-Net can realize the end-to-end training using the echo data as input. Moreover, the joint backpropagation process is derived for learnable parameters of the imaging module. In the experimental analysis, the proposed IITR-Net can achieve higher classification accuracy than current recognition frameworks. It implies that the IITR-Net can learn more deep features of the target, which improves the performance of recognition.