Inversion of Wave Height Using SNR and Shadow Probability Features Extracted from Non-Ideal Regions of X-Band Radar Images Based on a PSO-SVR Model

The three-dimensional fast Fourier transform (3D FFT) spectral analysis approach is the mainstream technique for retrieving significant wave height (SWH) from X-band marine radar image sequences by using the extracted wave signal-to-noise ratio (SNR). Nevertheless, an ideal analysis sub-region cannot be stably acquired in nearshore field observations due to terrain occlusion and variations in wave direction, which leads to azimuth distortion of the extracted SNR. The backscatter intensity and shadow probability of sea waves in radar images present strong directional correlations with wave direction. To solve these problems, wave direction is introduced as a correction factor to compensate for distorted SNR and shadow probability values, and a composite feature vector integrating multi-window SNR, shadow probability, and wave direction is constructed as the input of a particle swarm optimization support vector regression (PSO-SVR) model for SWH inversion. Conventional SVR relies on grid search for hyperparameter optimization, which limits inversion accuracy. PSO is adopted to seek suitable hyperparameters of the SVR model. Field radar datasets collected at Pingtan Island, together with synchronous buoy measurements serving as ground truth, are adopted to quantitatively evaluate the proposed method. Experimental results show that the proposed method, using multi-feature combination and the PSO-SVR model, improves inversion precision, which provides an accurate and robust SWH inversion scheme for practical shore-based radar monitoring without ideal analysis regions.