A Homogeneous Pixel Selection Method Based on Vision Transformer and Multitemporal Polarimetric Decomposition SAR Data in Mountainous Areas

Traditional selection of homogeneous pixels (SHP) methods insufficiently use polarization information and rely on extensive statistical computations from large datasets. When applied to mountainous or densely vegetated regions with low coherence, these methods are susceptible to noise interference, which reduces the interpretation reliability. In this study, we propose ViTPolSHP—the Vision Transformer (ViT)-based polarimetric synthetic aperture radar (PolSAR) SHP algorithm for selecting homogeneous pixels in multitemporal PolSAR imagery. The ViTPolSHP algorithm leverages the temporal average Yamaguchi four-component decomposition features for classification and integrates the terrain characteristics and geometric distortions to derive homogeneous pixel sets even with limited datasets, presenting a promising solution for synthetic aperture radar (SAR)/InSAR data processing in complex mountainous terrains.