This paper proposes a collaborative multimodal unmanned aerial vehicle (UAV) remote sensing method for the early detection of levee piping in complex environments. Addressing the identification challenges caused by weak piping thermal anomaly signals and strong background interference, this method innovatively integrates infrared image saliency detection with multi-source information verification. First, by combining global and local saliency analysis, suspected low-temperature anomaly areas in the infrared image are enhanced and extracted. Subsequently, adaptive threshold segmentation is employed to achieve preliminary localization of suspicious areas. To suppress false alarms, the algorithm designs a multi-level filtering mechanism, sequentially screening candidate areas based on the physical characteristics and geographical priors of piping, such as minimum area, temperature distribution variance, and terrain conditions. Finally, point cloud intensity images are introduced for spatial overlap analysis, leveraging the complementary nature of infrared and LiDAR data for cross-validation, thereby achieving precise identification of piping areas. Experiments validated based on field data collected from multiple locations show that this method can effectively enhance the robustness and accuracy of piping detection in complex field environments, providing a reliable technical solution for intelligent levee inspection.