Machine Learning Based Sea Surface Target Recognition

The primary objective of this research is to apply an enhanced version of the YOLOv8 target recognition algorithm to the identification of ship targets in radar images. The study focuses on refining the algorithmic network to bolster its precision and real-time performance for the recognition of sea surface targets amidst complex electromagnetic conditions. The enhancements to YOLOv8 encompass three main areas: the adoption of the ECIoU loss function in place of the original, a systematic and effective streamlining of the YOLOv8 network, and the incorporation of partial convolution in the residual module, replacing the standard convolution operator. Experimental results demonstrate that the refined YOLOv8-tiny model exhibits superior performance in recognizing sea surface ship targets in distance Doppler images under challenging electromagnetic environments, thereby offering a rapid and accurate solution for this specific recognition problem.