Sea-Surface Ship Detection and Embedded Deployment in SAR Images Based on Improved YOLOv5

Abstract

The detection of sea-surface ships represents a crucial means by which countries can compete for marine resources in the context of the current international strategic environment. The ongoing advancement of deep learning has led to a notable enhancement in the accuracy of target detection tasks involving surface ships, particularly when these are conducted using intelligent algorithms. Concurrently, the depth and width of target detection network models are expanding, accompanied by an increase in complexity. This has resulted in a challenge for deploying models that are limited by their own computational complexity in real surface ship target detection tasks. The field programmable gate array (FPGA) offers a solution to the issues of complex target detection network structures, high computational resource consumption and high storage space requirements. The FPGA's high parallelism and customizability make it an ideal choice for addressing these challenges. In light of the aforementioned background and problems, this paper puts forth an enhanced YOLOv5 algorithm for the detection and identification of SAR targets associated with sea-surface ships. To this end, the network model has been converted from floating-point to 8-bit integer parameter distribution through the optimization and quantification of the target detection network model. This enables the embedded deployment of the target detection and identification algorithm in real scenarios and tasks based on FPGA. The implementation of target detection and recognition algorithms in real scenarios and tasks