Strip steel surface defect detection method based on deep learning

Abstract

The invention relates to a strip steel surface defect detection method based on deep learning. According to the method, the local information of the surface of the strip steel is extracted through a defect judgment and defect classification double-flow network model. The comprehensive analysis is conducted in combination with a scale pyramid, a similar heat map is obtained. Finally, the types andpositions of defects are obtained at the same time. The defect judgment and defect classification double-flow network model comprises defect judgment branches and defect classification branches. Compared with the prior art, the method has the advantages of being small in calculation amount, high in calculation efficiency, high in robustness, low in labeling cost, high in precision, low in equipment cost and the like.

Description

technical field [0001] The invention belongs to the technical field of defect detection, and relates to a method for detecting defects on the surface of strip steel, in particular to a method for detecting defects on the surface of strip steel based on deep learning. Background technique [0002] The traditional strip surface defect detection methods are roughly manual detection method and stroboscopic light detection method, both of which are non-automatic detection methods. From 1950 to 1960, the method of manual visual inspection was mainly used for the detection of strip surface defects. Due to the high transmission speed during strip steel production, if a tiny surface defect is encountered, the human naked eye cannot accurately judge the type and level of the defect at all, resulting in a large number of missed and false detections of strip surface defects. Moreover, the factory environment is harsh, and there is usually noise and dust. Workers working in this environ...

AI Technical Summary

Problems solved by technology

[0005] (1) High labeling cost: To deal with any defect, it is necessary to construct countless image samples, and to use a box to mark the defect in each image

This method will have three problems: first, it requires a huge labor cost; second, most steel plate defects are textured and have no clear target boundary, so it is extremely difficult to mark; third, different defects may be mixed with each other, blending with each other, which can have a lot of negative effects on model optimization

[0006] (2) Slow speed: Faster-RCNN's feature extraction framework based on VGG-16 has tens of thousands of parameters, which runs slowly on embedded devices and cannot be applied to real-time occasions

[0007] (3) Insufficient generalization ability: This method does not take into account the priority of the judgment of the presence or absence of defects and the judgment of defect types, and analyzes at the same level

However, based on the means of traditional machine vision, the degree of connection with artificial intelligence is not enough, and it is impossible to take advantage of the accuracy, generalization and robustness of big data, the Internet of Things and artificial intelligence in large-scale data processing.

Images