Cigarette defect detection based on independent feature extraction constraints

Abstract

The detection of appearance defects in cigarettes is crucial in the field of industrial defect detection. Most existing detection methods achieve defect detection by utilizing deep learning to learn feature representations of various types of defects. However, due to the complexity and randomness of the production and processing process, the differences between defects in different cigarettes are not significant, which greatly affects the detection performance. Therefore, to address this issue, this article proposes a cigarette defect detection method based on independent feature extraction constraints, called IFEC. The core idea of this method is to extract independent features of different defect categories to increase the differences between features of different categories, enhance the distinguishability of features, and achieve accurate cigarette defect detection. In IFEC, an independence feature extraction constraint module is proposed that constrains the network to extract highly independent defect features of different categories through feature decoupling and feature decorrelation. The sufficient experimental results indicate that the proposed IFEC has superior detection performance compared to existing detection methods.