Deep learning based detection of debris flow areas using Mask- RCNN and PCA: insights from the Uluborlu Basin, Türkiye

OpenAlex English

Debris flow is a rapid, displaced flow of material, often occurring on steep slopes, which can cause significant damage to infrastructure, agricultural areas, natural ecosystems, residential areas, and other regions due to environmental and topographic factors.These events are difficult to detect in rugged terrains.This study applies an object-based approach using a Mask-RCNN deep learning algorithm to detect debris flow areas in rugged terrains.Principal Component Analysis (PCA) was performed on parameters extracted from ASTER DEM, climate, and geological data, and the first 8 PC's explaining 92.97% of the total variance were integrated with Sentinel-2A satellite imagery (RGB-NIR bands with 10 m spatial resolution) to identify debris flow areas.The debris flow areas were detected with high performance with 89% Recall, 90% Precision, 92% Accuracy, 81% mIoU and 92% mPA.The results show that integrating PCA-derived components into high-resolution remote sensing data improves the efficiency of the Mask-RCNN method for debris flow detection, especially in rugged terrains and small-scale studies.