In response to the problems of low efficiency, high cost, single data type, and potential safety hazards existing in traditional dam leakage detection methods, this study integrates sensors such as thermal infrared cameras, RGB cameras, and LiDAR into lightweight unmanned aerial vehicles to construct a multi sensor integrated system. This system can acquire multi source data of dam structures and their surrounding environments in real time, rapidly, all day long, comprehensively, and non contact. In the data processing stage, the thermal infrared images are first enhanced to improve the image quality, and the multi source data registration is completed. Subsequently, the Hierarchical Context Fusion Network is used, combined with information such as point cloud intensity and river water height, to identify potential leakage areas. Then, the Density Based Spatial Clustering of Applications with Noise algorithm is utilized to optimize the results, and the RGB images are used as an aid to accurately locate the leakage points. By processing the data collected from a certain dam, the reliability of the multi sensor equipment and the multi level suspected hazard detection algorithm is verified. This application can reliably detect and locate suspected hazards, significantly reduce the time and resource costs of dam detection, and simultaneously reduce the safety risks of dam detection personnel.