Deep Learning Based Surface Defect Detection of Ceramic Cores

The ceramic valve core is widely used in industry due to its wear resistance, corrosion resistance, and hightemperature resistance. However, its high hardness and brittleness make it prone to damage and cracks during processing, affecting the product's appearance, safety, and service life. Therefore, accurate and efficient surface defect detection is crucial. Although existing detection methods have made some progress, there are still shortcomings in detection speed, computational resource consumption, and hyperparameter tuning. To address these issues, this paper proposes a surface defect detection method for ceramic valve cores based on a lightweight object detection network. The method employs MobileNetV3 to construct a multi-scale feature pyramid, incorporates an anchor-free region proposal network that generates candidate boxes by directly predicting object centers and offsets to achieve real-time defect detection. Additionally, a channel attention mechanism is used to improve classification accuracy. Experimental results demonstrate that this method offers high detection accuracy and low computational resource consumption, making it suitable for defect detection tasks in industrial production. It effectively resolves the challenges of traditional methods in handling the diversity of defects and multiscale characteristics in complex production environments.