Research on Lightweight Model-Based Leather Defect Edge Inference Mechanism

Real-time and accurate detection of production defects is one of the most critical aspects of the production process. This paper proposes a model lightweight-based edge inference mechanism for leather defect detection. By collecting thousands of samples of leather defects and creating a self-made dataset. Then train the model through the YOLO algorithm. Constraining the size of the channel factor γ and adding an L1 regular term about the γ to the target equation, so that it can be thinned and automatically pruned during training process. The pruning model is retrained on the self-made dataset by the weight of the pruning model, and the gradient descent algorithm is used to fine-tune the model to restore the accuracy, and then deployed on the edge device to perform inference detection on the real-time production picture of leather. The detection results are transmitted to the industrial control host computer for storage and screening, achieving real-time detection, classification, and display of defects in leather production. Compared with traditional defect detection systems, this paper achieves faster and more effective leather defect detection in the production process.