A Hybrid Remote Sensing Pipeline for Landslide and Soil Scar Mapping

This paper presents a hybrid pipeline for monitoring terrain degradation in satellite imagery, focusing on landslide and soil-scar mapping. A deep segmentation network (FCDU-Net) identifies candidate scar regions, followed by a supervised patch-level classifier that validates them using spectral and textural features. Both stages use RGB, NDVI, and GLCM descriptors selected via ReliefF. Across 30 runs on 7,092 patches, the approach improved the mean F1-score (0.864 to 0.905) and the area under the curve (0.955 to 0.983) compared with segmentation alone. The gains were statistically significant according to Welch's t-test, with a p-value below 0.001 and a Cohen's d of 1.35, supporting explicit stage separation.