Defect detection semantic segmentation model modeling method and device, medium and equipment

Abstract

The invention discloses a workpiece defect detection semantic segmentation model modeling method. The method comprises the steps that: a plurality of groups of original workpiece point cloud data fortraining are acquired, and workpieces are workpieces containing specified defect types; the defect synthesis image of each workpiece is determined based on the original workpiece point cloud data; theoriginal workpiece point cloud data are converted into a two-dimensional depth map; a first gradient map Scale1 and a second gradient map Scale2 are acquired based on the two-dimensional depth map; the two-dimensional depth map, Scale1 and Scale2 are synthesized based on an RGB three-channel color image, so that a defect synthesis image can be obtained; a defect labeling Label image correspondingto each workpiece is acquired; a semantic segmentation model is constructed, and training is performed based on the multiple groups of workpiece defect synthesis images and the corresponding defect labeling Label images, semantic segmentation model parameters are iterated, so that a convergent semantic segmentation model can be obtained; and after semantic segmentation training is finished, modellayers and corresponding weights are solidified into pb files, and the pb files are exported as pb model files corresponding to the workpiece defect images, so that 3D defect segmentation precision and a detection rate can be effectively improved.

Description

technical field [0001] The invention relates to the technical field of defect detection on industrial sites, in particular to a defect detection semantic segmentation model modeling method, device, medium and equipment. Background technique [0002] In industrial scenarios, there are very strict requirements on the delivery indicators of product parts. These requirements provide a guarantee for the functional role of product parts after they leave the factory. In industrial scenarios, various mechanical, sound, light, and electrical complex environments and numerous processes may damage the appearance of product parts, making them defective product parts. [0003] At present, there are two main methods of product defect detection commonly adopted in the industry: (1) manual detection; (2) detection methods based on machine vision. Among them, manual detection has many shortcomings such as low efficiency, high cost, and manual fatigue; in mass industrial production, defect d...

AI Technical Summary

Problems solved by technology

Since the feature vector extracted from the image of the defect area is similar to the dimension of the feature vector extracted from the adjacent image, the accuracy of defect segmentation is relatively poor, resulting in a low accuracy rate of defect detection.

Images