Film defect detection method based on improved attention mechanism

Abstract

The invention discloses a film defect detection method based on an improved attention mechanism. The method comprises the following steps: reading a film image in real time; inputting the image into anetwork model for reasoning; and judging whether flaws exist in the thin film image or not according to an inference result, if so, automatically marking by using a red rectangular frame, and otherwise, reading the next thin film image. According to the invention, the YOLOv5 network structure is simplified and modified, and the lightweight attention module is added to the network, so that the defect detection accuracy can be effectively improved on the premise of not influencing the speed; the method is used for industrial film defect detection, the quality of film products can be effectivelyimproved, manual intervention is not needed, and labor and time cost is saved.

Description

technical field [0001] The invention belongs to the field of video image processing and pattern recognition in computer vision, and relates to a film defect detection method based on deep learning. Background technique [0002] With the increasing demand for films in the international and domestic film markets and the increasingly fierce competition within the film industry, more and more film manufacturers have begun to use production lines with wider widths and faster production speeds to reduce costs and increase production. efficiency. However, in the actual production process of the film, due to the influence of various factors, defects such as holes, mosquitoes, black spots, crystal spots, scratches, spots, etc. will appear on the surface of the film, which seriously affects the quality of the film and brings manufacturers unnecessary loss. The traditional manual detection method is susceptible to subjective factors and lacks consistency; and human eyes are prone to ...

AI Technical Summary

Problems solved by technology

The two-stage algorithms mainly include Fast R-CNN, Faster R-CNN, etc. First, the candidate region generation network (RPN) is used, and then they are further classified and returned. This type of method has high precision, but the calculation is complicated and the detection time is high. It is too long to meet the real-time requirements of industrial production; single-stage algorithms mainly include SSD and YOLO series. In recent years, the accuracy of single-stage detection algorithms has been continuously improved and has reached industrial-level requirements. The speed has been greatly improved, so it is especially suitable to use YOLOv5 for real-time defect detection system

However, due to the small number of training samples for thin film defects, the small target of thin film defects and the complex characteristics of thin film defects, it is necessary to further improve the YOLOv5 algorithm to meet the requirements of industrial thin film defect detection.

Images