Ceramic tile surface defect detection based on deep learning

Abstract

Ceramic tiles are widely used in the construction field. In actual production, the difficulty in extracting texture ceramic tile features and the small size of defects lead to low detection accuracy and inefficiency. To solve these problems, we propose a detection method based on improved YOLOv5s. Firstly, the network layer is deepened in the backbone network and the attention mechanism CBAM module is added. Then, a small-scale detection layer is added, and the model is increased from a three-output prediction layer to a four-output prediction layer. Thirdly, the network feature fusion is enhanced in the neck network. Finally, the original convolution is replaced with depthwise separable convolution, and a lightweight ceramic tile detection system is constructed. Experimental results show that our model can solve the problems caused by small defects and insufficient feature information effectively.