Underwater Pile Foundation Defect Detection Method Based on Diffusion Probabilistic Model and Improved PointMLP

To detect damage in underwater pile foundations, we propose a new method based on the diffusion probability model and improved PointMLP. First, PCA-ICP registration is carried out for the point cloud data from different stations using a sonar system. A variety of filtering algorithms and the Random Sample Consensus (RANSAC) method are employed to obtain a complete point cloud of the pile foundation. The pile foundation defect point cloud is generated and enhanced based on the diffusion probability model. The feature attention mechanism is added to the PointMLP, and then the improved PointMLP is trained to identify the defect of the pile foundation. In our study, the point cloud of a wharf pile foundation was collected, and the experimental results effectively identified the damage to the pile foundation. Up to 95% accuracy was achieved for the calculated volume. The volume error of the damage was 0.0756 m³, with an accuracy of 95.238%. Thus, this method can provide technical support for detecting underwater pile foundation defects and avoiding the occurrence of major accidents.