Deep learning-based visual detection method for packing missing strips

Abstract

The invention discloses a deep learning-based visual detection method for packing missing strips. Through the design and construction of a Fast-AnoGAN algorithm model, anomaly detection of cigarette box images is achieved. The Fast-AnoGAN algorithm model includes a generation confrontation network module and an abnormality score calculation module. The generation confrontation network module is combined with a WGAN-GP algorithm model for design and comprises a generator network and a discriminator network. Normal images are used for model training. Feature distribution of normal samples is learned. The abnormality score calculation module is configured to calculate an abnormal score value corresponding to the generated image of the to-be-detected picture input training. Anomaly detection is implemented by determining whether the abnormal score value is greater than a defined abnormality discrimination threshold. The invention has the problems of effectively solving the sample imbalancein the deep learning model training, satisfies the dual requirements of real-time and accuracy of product detection under high-speed production, and has the advantages of adaptability and robustnessto the industrial production environment.

Description

technical field [0001] The invention belongs to the field of industrial product production detection, in particular to a deep learning-based visual detection method for packing missing bars. Background technique [0002] In the production of cigarettes, due to the possibility of skewing and less pushing of the tobacco rods during the process of conveying and pushing the tobacco rods into the cigarette box, the phenomenon of missing rods occurs during the packing process. Sealing machine packing is the last process. If the abnormality of missing bars cannot be detected in time and accurately, it will cause products with serious quality defects to enter the market, which will have a very bad impact on the credibility of cigarette companies. [0003] With the continuous advancement of Industry 4.0, the automation level of the cigarette manufacturing industry has been continuously improved. In view of the problem of missing rods in boxes in cigarette production, it is difficult ...

AI Technical Summary

Problems solved by technology

[0006] However, in the cigarette box image data, there are very few abnormal images, and there are far more normal images than abnormal images

Existing deep learning anomaly detection models are often established based on balanced data, which is difficult to apply in the case of unbalanced data with very limited abnormal images

Images