Copper Nodule Defect Detection in Industrial Processes Using Deep Learning

Abstract

Copper electrolysis is a crucial process in copper smelting. The surface of cathodic copper plates is often affected by various electrolytic process factors, resulting in the formation of nodule defects that significantly impact surface quality and disrupt the downstream production process, making the prompt detection of these defects essential. At present, the detection of cathode copper plate nodules is performed by manual identification. In order to address the issues with manual convex nodule identification on the surface of industrial cathode copper plates in terms of low accuracy, high effort, and low efficiency in the manufacturing process, a lightweight YOLOv5 model combined with the BiFormer attention mechanism is proposed in this paper. The model employs MobileNetV3, a lightweight feature extraction network, as its backbone, reducing the parameter count and computational complexity. Additionally, an attention mechanism is introduced to capture multi-scale information, thereby enhancing the accuracy of nodule recognition. Meanwhile, the F-EIOU loss function is employed to strengthen the model’s robustness and generalization ability, effectively addressing noise and imbalance issues in the data. Experimental results demonstrate that the improved YOLOv5 model achieves a precision of 92.71%, a recall of 91.24%, and a mean average precision (mAP) of 92.69%. Moreover, a single-frame detection time of 4.61 ms is achieved by the model, which has a size of 2.91 MB. These metrics meet the requirements of practical production and provide valuable insights for the detection of cathodic copper plate surface quality issues in the copper electrolysis production process.