作者
Lizong Cao,Qingyou Han,Rong Luo,Li Xu,Weikuan Jia
摘要
In the leather manufacturing industry, the detection of surface defects is crucial for ensuring product quality. Traditional manual inspection methods are subjective, inefficient, and susceptible to environmental influences, and can no longer meet the demands for high efficiency and quality in modern leather production. Therefore, developing a fast, accurate, and automated defect detection system has become an urgent need in the industry. Against this backdrop, this paper conducts an in-depth study and targeted optimization of the YOLOv8 algorithm, proposing a novel wet blue leather surface defect detection model, ACI-Net, to enhance detection accuracy and robustness. To address the challenge of distinguishing defects from similar background textures, this study introduces the ACMix attention module. This module effectively captures long-range dependencies in images, significantly improving the accuracy of defect recognition. The study incorporates the MetaNeXtStage module, which focuses on the effective integration of multi-scale features, enabling the model to precisely identify a wide range of defect sizes, thereby enhancing overall detection performance. Comparative experiments demonstrate that this algorithm surpasses existing models in defect detection, achieving accuracy rates of 86.2%, 99%, and 88.8% for brand, broken hole, and broken surface, respectively, thus meeting the dual requirements for precision and robustness in industrial applications.