Defect Detection in Corrugated Cardboard based on Improved YOLOv5s

Corrugated cardboard is a crucial raw material in the production and transportation of cigarettes. However, its surface is prone to defects such as damage. Existing visual general models for detection suffer from issues like large model size, slow processing speed, and low detection accuracy, making them unsuitable for practical deployment on production lines. To address these challenges, this paper proposes an improved defect detection method for corrugated cardboard based on the YOLOv5s.The approach introduces Ghost convolution to reduce model complexity and enhance processing speed. BiFPN is employed for weighted feature fusion, improving the primary detection accuracy. Lastly, the loss function is modified to GIoU to enhance the model's ability to model target box positions. Experimental results demonstrate a 5.7 GFLOPS reduction in computational complexity and a 4.9 FPS improvement in processing speed compared to YOLOv5s. The achieved speed is 83.3 FPS. Additionally, the model's accuracy, measured by mAP@0.5:0.95., shows a 1.4% improvement, reaching 97% for mAP@0.5. The model maintains good detection accuracy while ensuring efficient processing speed, demonstrating practical applicability.