High-performance Implementation Methods for SAR Imaging and Target Recognition Based on Reconfigurable Computing

Synthetic aperture radar (SAR) is widely used in both military and civilian fields for imaging and target recognition tasks. However, SAR imaging and target recognition tasks involve large image sizes, whose performance faces severe limitations due to hardware resources. Focused on the emerging field of reconfigurable computing technology, a high-performance implementation for SAR imaging and target recognition systems based on reconfigurable computing chips is proposed in this paper. Reconfigurable computing chips employ reconfigurable control techniques to implement various computation and data paths, combining flexibility with high energy efficiency. The chirp scaling algorithm and YOLOv3-tiny neural network are selected to construct the system algorithm kernel in this paper. Based on the large size character of SAR images, a multi-core parallel and memory allocation scheme is proposed in the imaging phase, and an image segmentation strategy and multi-core parallel scheme are proposed in the target recognition phase. Experimental validation confirms significant performance enhancement achieved by the imaging and target recognition system presented in this paper; achieving a single-image imaging time of 66.8 ms for 1 000×1 000-sized images, outperforming the Intel i5-12500′s 115 ms; and achieving a recognition time of 31.3 ms for 480×480-sized images, surpassing that of the Jetson nano′s 147 ms.