Three-cone visual angle fusion method and system for improving surface defect detection precision and medium

Abstract

The invention provides a three-cone view angle fusion method and system for improving surface defect detection precision and a medium. The method comprises the following steps: step 1, shooting a three-cone view angle graph of a sample; 2, preparing a defect-free graph corresponding to the sample, sending the defect-free graph to a feature matching model, and extracting a first feature graph; 3, sending the three-cone view angle image into a feature matching model in a queuing manner, and extracting a second feature image; 4, sending the first feature map and the second feature map into a homography matrix network to obtain a conversion matrix; 5, performing downsampling on the second feature map; 6, sending the down-sampled picture into a conversion matrix to obtain a conversion graph; 7, carrying out channel level fusion on the transformation graph to form a three-channel fusion graph; and 8, sending the three-channel fusion image into a detection model, and learning to obtain depth information. Based on the three-channel fusion, feature matching and other methods, the problems of background interference and concave-convex defects are solved, and the surface defect detection precision can be effectively improved.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a three-cone perspective fusion method, system and medium for improving the detection accuracy of surface defects. Background technique [0002] In the quality inspection of industrial surface defects, uneven defects such as lack of material, excess material, and cracks are often easily confused. On a flat metal surface, excess material is a protruding part, and lack of material and cracks are recessed parts, so in In 2D, protruding polys and concaves look very similar. Most of the pictures to be detected taken by ordinary cameras are two-dimensional images, so this leads to the fact that the algorithm can only obtain the two-dimensional information (length and width information) of the defect, and cannot obtain the depth information of the image, so the lack of depth information is the reason. It is the main reason for inaccurate and confusing test results. [...

AI Technical Summary

Problems solved by technology

Most of the pictures to be detected taken by ordinary cameras are two-dimensional images, so this leads to the fact that the algorithm can only obtain the two-dimensional information (length and width information) of the defect, and cannot obtain the depth information of the image, so the lack of depth information is the reason. It is the main reason for inaccurate test results and confusion

[0003]However, the speed, accuracy and cost of the depth camera cannot meet the actual needs of the factory. However, due to the complexity of the background texture, the reconstruction results often fail, resulting in lower detection accuracy. Multi-light source image reconstruction itself relies more on a careful parameter adjustment process, so it is not suitable for batch production. engineering

[0004]In the field of industrial quality inspection, accuracy is a very strict indicator, but due to the lack of depth information in the image, it will often lead to serious accuracy degradation

In the prior art, multi-light source image reconstruction is a method to solve the lack of depth information in images. However, due to the reflective properties of metal surfaces, different background textures, and the selection of camera angles, the current multi-light source image reconstruction technology not a mature solution

Images