Industrial defect detection method based on multi-task learning

Abstract

The invention relates to the technical field of industrial defect detection, and discloses an industrial defect detection method based on multi-task learning. A defect classification task is subdivided into two subtasks, a normal/abnormal (ok/ng) dichotomy problem (recorded as task1) and a multi-label classification problem (recorded as task2) of n defect categories are solved by constructing a classification model based on a convolutional neural network (CNN). The classification model is composed of a base model and a head. The base model is responsible for extracting image features of an input image to obtain a corresponding feature image, and the base models of different tasks share network weights by adopting a hardsharing connection mode. The head is an output layer, and two branches are led out from the base model and are respectively used for solving the problems of task1 and task2; the two branches are respectively composed of a full connection layer and a sigmod function, and the probability of the ng category and the category probability of the n defects are output. The method can alleviate the problem that the current industrial defect detection method is easily interfered by imaging conditions, small differences between defects and backgrounds, low image contrast, large scale and appearance changes of defects of the same type and the like, resulting in unstable detection effect.

Description

technical field [0001] The invention relates to the technical field of industrial defect detection, in particular to an industrial defect detection method based on multi-task learning. Background technique [0002] Industrial defect detection is a technology that uses machine vision algorithms to automatically identify defects in images captured by industrial cameras. Specifically, industrial defect detection needs to judge whether there is a defect in the image and identify the type of defect, and then analyze the degree of defect of the industrial product. This technology can be widely used in various industrial fields to replace manual inspection and improve product production efficiency, detection accuracy and stability. [0003] Traditional machine vision-based industrial defect detection methods generally need to select an appropriate imaging scheme (such as bright field imaging, dark field imaging, and hybrid imaging, etc.) defects, and then use image processing alg...

AI Technical Summary

Problems solved by technology

[0006] The embodiment of the present invention provides an industrial defect detection method based on multi-task learning, which is used to solve the problem that the existing traditional industrial defect detection method is vulnerable to imaging conditions, small differences between defects and background, low image contrast, and the same type of defect scale. Interference with factors such as large changes in appearance and appearance, resulting in unstable detection results, which cannot meet the actual use requirements

Images