Rapid Detection of Landslides for the Timely Response of Disaster Mitigation and Relief

Abstract

Landslides on the Tibetan Plateau can induce disaster chains, which can extend the damage from one point to an area. At present, the timely treatment of landslides after they have occurred is the only feasible way to avoid the expansion of the disaster. Searching on foot is still the primary method of locating landslides. However, this method is inefficient and time-consuming. To address this problem, this paper designs a one-shot landslide detector (AOSLD) to quickly locate sudden landslide hazards on the Tibetan plateau. AOSLD introduces decoupled head and anchor-free modules based on YOLO-V3. Mixup and Moscia are used to enhance the robustness of the model. The results show that AOSLD can simultaneously achieve multitype, multiobjective, and multiscale landslide detection. AOSLD outperforms 16 kinds of state-of-the-art methods. Moreover, AOSLD is deployed to smartphones, which dramatically improves landslide detection and the efficiency of translating landslide research results into applications. AOSLD is also combined with the F-16 aerial simulator to detect sudden landslide hazards on the Tibetan plateau. To our knowledge, this is the first time that a landslide detection model has been deployed to multiple application vectors for the rapid detection of landslides in the Tibetan Plateau region. Those are important for avoiding the expansion of disaster chains on the Tibetan plateau, protecting life and property in downstream areas, and promoting the translation of landslide research results into applications.