Dark area recognition algorithm for SAR images based on superpixel statistical features

The dark areas in SAR images are primarily low backscatter regions dominated by noise signals, characterized by significant texture, low echo energy, and large phase interference errors in the echo signals. Accurately identifying these dark regions is crucial for high-precision SAR image target detection and classification, SAR cross-radiometric calibration of satellite constellations, SAR image filtering, and the generation of digital surface models based on SAR technology. This paper proposes a dark region recognition algorithm for SAR images based on superpixel statistical features(SSF). The algorithm constructs a bimodal distribution curve using the statistical features of superpixels and searches for the region with the minimum amplitude proportion within the bimodal distribution, which serves as a constraint to ensure that the OTSU thresholding method converges to the global optimal solution. This approach addresses the issue of inaccurate threshold detection in traditional algorithms, significantly improving the accuracy of dark region detection in SAR images. The detection results can further be used to identify low backscatter areas, such as shadows, water bodies, and asphalt roads.