Deep learning-based landslides detection improved by the gray level co-occurrence matrix of RGB optical images

Landslides, particularly those induced by rainfall, are among the most devastating natural hazards, often causing severe disruptions to transportation infrastructure and hindering emergency rescue operations in mountainous regions. This study presents a method to improve the accuracy of landslide detection using RGB remote sensing imagery by integrating texture features derived from the gray level co-occurrence matrix (GLCM). A hybrid framework was developed that incorporates GLCM-based texture descriptors into deep learning models, including U-Net, DeepLabv3+, and Mask R-CNN, enabling the automatic extraction and recognition of complex spatial patterns. Experimental results demonstrate that the integration of texture features substantially enhances detection performance, with average improvements of 15.2% in intersection over union (IoU), 7.1% in F 1-score, and 9.9% in sensitivity (SE) across the three models compared to RGB-only inputs. These findings confirm the effectiveness of texture-assisted deep learning and highlight its potential as a robust and cost-efficient solution for precise landslide detection, offering significant advantages for geological hazard monitoring and emergency response.