A Study on the Application of Three-Dimensional Scanning Data Analysis Method Based on Convolutional Neural Network in Machining Inspection

Abstract

In this paper, the mechanical workpiece point cloud data is obtained by 3D scanning technology, and the data preprocessing is realized by the extended Gaussian image clustering algorithm to align the point cloud data and other operations. Subsequently, the sliding least squares method is used to upsample the point cloud, combined with bootstrap filtering to denoise the data image and image sharpening and other processing. The improved Faster R-CNN processing detection method is designed by combining the multi-scale feature extraction structure, which improves the efficiency of machining detection by dimensionality reduction of the feature map of the previous layer, then feature extraction, and stitching the features extracted from different scale convolution kernels together. The test results show that the improved Faster R-CNN algorithm has obvious improvement in the aspects of large part size difference, occlusion, and accuracy, and can effectively solve the point cloud accurate alignment problem of local deformation of mechanical parts with strong robustness. At the same time, the performance of this model is significantly better than other comparative models in the process of target monitoring of construction machinery, and the accuracy of its AP50, AP75 and mAP indicators reaches more than 84%. In addition, the method in this paper can automatically screen the conformity of parts in machining inspection, which greatly reduces the workload.