Surface Defect Detection in Sanitary Ceramics Based on Lightweight Object Detection Network

Sanitary ceramic products, such as toilet and wash basin, are widely used in our daily life. ceramics expected to have some excellent physical properties, corrosion resistance, easy cleaning, low water absorption. However, surface defects sanitary inevitable due complex production processes changing environment. Therefore, defect detection must be performed the manufacturing process of ceramics. There many types ceramics, different large differences characteristics scales. Traditional methods with artificially designed features classifiers difficult apply this context. In addition, there few studies on based deep neural networks. article, a lightweight real-time network backbone MobileNetV3 is presented. The proposed achieves multi-scale multi-layer feature pyramid. Combining region proposal anchor-free method, achieved. Finally, head channel attention structure low-level mixed classification strategy perform higher accuracy. Experimental results show that approach at least 22.9% speed improvement 35.0% average precision while reducing memory consumption by 8.4% compared classic one-stage SSD, YOLO V3 two-stage Faster R-CNN methods.