Effective Segmentation of X-Band Marine Radar Images Using SegFormer

X-band marine radar imaging has been widely used for ocean surface monitoring and parameter estimation. However, the quality of radar images can be significantly affected by various meteorological conditions, leading to inaccurate estimation of ocean surface parameters. In this paper, a novel approach is proposed for segmenting X-band marine radar images into four distinct regions using the SegFormer model, an advanced semantic segmentation architecture that employs a hierarchical transformer encoder to capture global context and a lightweight all-multilayer perception (MLP) decoder to enable high-resolution segmentation. The four regions include areas with clear wave signatures, rain-contaminated areas, low backscatter areas, and wind-dominated rain areas. The experimental data collected by a shipborne Decca radar system is used to validate the effectiveness of the proposed method, and a SegNet-based segmentation approach is utilized for comparison. Experimental results demonstrate that the SegFormer-based method outperforms the SegNet-based method, with an improvement of 0.5% in mean segmentation accuracy.