Leveraging UAVs and Machine Learning for Enhanced Landslide Detection and Risk Management

Landslides pose a significant threat to safety, demanding advanced detection and monitoring techniques to mitigate risks. This study explores an approach that benefits from recent advancements in both data acquisition and analysis methods. UAV-derived ultra-high-resolution digital elevation models and orthomosaic offer an accurate foundation for object-based image analysis to achieve highly accurate landslide detection. The widespread availability of high-resolution digital terrain models from UAVs, combined with advances in computer vision and ultra-high-resolution sensor technology, enables more precise landslide identification and characterization in engineering geology. Focusing on Atali, an extensive cut slope along NH-7, Uttarakhand, India that is frequently vulnerable to rockslides during monsoon seasons, this study highlights the necessity of detailed analysis to prevent and mitigate future hazards. We use the object-based image analysis technique with a thresholding value to segment UAV-derived orthomosaics into meaningful objects based on spectral, spatial, textural, and morphological attributes, providing a more coherent basis for classification. The digital surface model derived from the first UAV survey revealed that the slope is active, and thereafter, temporal flights at the Atali slope were also done. The slope and elevation values change over the monsoon cycle highlights the critical need for detailed analysis. By identifying distinctive landslide patterns within the UAV imagery, the random forest algorithm is used to produce an improved landslide detection technique with fewer false positives and negatives, leading to more reliable safety assessments. The preliminary results indicate the effectiveness of the combined OBIA-machine learning approach in delivering high-accuracy landslide classification. The detected landslide body serves as crucial data for detailed landslide mapping and has the potential to improve decision-making in engineering practices.

Keywords: Digital elevation model, Landslide detection, Machine learning, OBIA, UAV.