CNN-Based Target Detection and Classification When Sparse SAR Image Dataset is Available

• Published in: IEEE journal of selected topics in applied earth observations and remote sensing Vol. 14; pp. 6815 - 6826
• Main Authors: Bi, Hui, Deng, Jiarui, Yang, Tianwen, Wang, Jian, Wang, Ling
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial neural networks; Classification; Clutter; complex approximate message passing (CAMP); Convolutional neural network (CNN); Datasets; Detection; Detectors; Feature extraction; Frameworks; Image classification; Image resolution; Information retrieval; Message passing; Military applications; Neural networks; Object detection; Radar imaging; Radar polarimetry; Regularization; SAR (radar); Sidelobes; Signal processing algorithms; sparse synthetic aperture radar (SAR) image; Synthetic aperture radar; Target detection; target detection and classification; Target recognition
• ISSN: 1939-1404; 2151-1535
• DOI: 10.1109/JSTARS.2021.3093645

Synthetic aperture radar (SAR) is an earth observation technology that can obtain high-resolution image in all-weather and all-time conditions, and hence, has been widely used in civil and military applications. SAR target detection and classification are the key processes for the detailed feature information extraction of the interested target. Compared with traditional matched filtering (MF) recovered result, sparse SAR image has lower sidelobes, noise, and clutter. Thus, it will theoretically has better performance in target detection and classification. In this article, we propose a novel sparse SAR image based target detection and classification framework. This novel framework first obtains the sparse SAR image dataset by complex approximate message passing (CAMP), which is an <inline-formula><tex-math notation="LaTeX">L_1</tex-math></inline-formula>-norm regularization sparse imaging method. Different from other regularization recovery algorithms, CAMP can output not only a sparse solution, but also a nonsparse estimation of considered scene that well preserves the statistical characteristic of the image when protruding the target. Then, we detect and classify the targets by using the convolutional neural network based technologies from the sparse SAR image datasets constructed by the sparse and nonsparse solutions of CAMP, respectively. For clarify, these two kinds of sparse SAR image datasets are named as <inline-formula><tex-math notation="LaTeX">\mathcal {D}_{\rm Sp}</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">\mathcal {D}_{\rm Nsp}</tex-math></inline-formula>. Experimental results show that under standard operating conditions, the proposed framework can obtain 92.60% and 99.29% mAP on Faster RCNN and YOLOv3 by using the <inline-formula><tex-math notation="LaTeX">\mathcal {D}_{\rm Nsp}</tex-math></inline-formula> sparse SAR image dataset. Under extended operating conditions, the mAP value of Faster RCNN and YOLOv3 are 95.69% and 89.91% mAP, respectively. These values based on the <inline-formula><tex-math notation="LaTeX">\mathcal {D}_{\rm Nsp}</tex-math></inline-formula> dataset are much higher than the classified result based on the corresponding MF dataset.