Genting Highlands landslide mapping using deep learning / Siti Nuha Amisyah Sappe

Using methods of deep learning, this thesis conducts an extensive study on landslide mapping in the Genting Highlands with the goal of creating an automated tool to identify land locations most suitable for development by evaluating probable landslide hazards. The research makes use of Digital Elevation Models (DEMs) and SPOT-7 satellite data to classify and analyse different landslide formations in the training area in accordance with Varnes1 1978 categorization. In order to identify possible landslides, this study trains two models Mask R-CNN and YOLO v3 in Kundasang and then applies them to the Genting Highlands. The approach comprises data acquisition from many sources, including open source DEMs and SPOT-7 satellite images, and then processing the data, including mosaicking, georeferencing, and manual landslide delineation. Deep learning methodologies are used, which include model training, object labelling, and accuracy assessment with metrics including F1-score, precision, and recall. The outcomes show how well Mask R-CNN and YOLO v3 can detect landslide-prone regions; comprehensive landslide maps for the Genting Highlands are produced. These results emphasise the advantages and disadvantages of each model and offer important insights into the effectiveness and dependability of automated landslide detection techniques. According to the study's findings, combining deep learning models with conventional remote sensing methods greatly improves the accuracy and efficiency of landslide detection, which helps with risk management and preventive measures in mountainous areas and provides a useful tool for land-use planning and development in landslide-prone areas.