Amplitude-Phase CNN-Based SAR Target Classification via Complex-Valued Sparse Image

It is known that a synthetic aperture radar (SAR) image obtained by matched filtering (MF)-based algorithms always suffers from serious noise, sidelobes, and clutters. However, the improvement of quality means complexity SAR system will increase, which affects application image. The introduction sparse signal processing technique into imaging proposes new way to solve this problem. Sparse recovery shows better performance than typical complex with lower sidelobes higher signal-to-noise ratio. As most widely applied field image, target classification relies on high quality. Therefore, novel model based amplitude phase information introduced in article. First, dataset constructed iterative soft thresholding (BiIST) algorithm. Compared regularization-based algorithms, BiIST not only can improve recovered but also obtain nonsparse solution retaining background statistical distribution Then, targets are classified proposed amplitude-phase convolutional neural network (AP-CNN). Typical networks imitate those optical just using data. considering particularity AP-CNN uses both for training, theoretically improves accuracy. Experimental results show outperforms amplitude-based CNN classification, under standard operating conditions (SOCs) extended (EOCs). Results SOC demonstrate accuracy 11.46% 1000 training samples. Even EOC, gap between reach 6.6% case 800 This even if number samples limited, optimal result.