Detection of Speckle Defects in Alumina Ceramic Balls Based on MobileNetV3-YOLOv5

Aiming at the current alumina ceramic ball speckle defect detection algorithm in practical applications, such as low detection accuracy and slow detection speed, a fast detection method of ceramic ball surface defects based on the improved YOLOv5 is proposed. The ceramic ball image data is first obtained using the camera, then processed to standardize the image size, rotated, added noise, and subjected to other data enhancement operations. The data is then annotated using data annotation software to create the dataset, which is then used to improve the backbone network of the improved YOLOv5s using MobileNetV3. When parameters are checked, it is discovered that the upgraded network model's detection effect is noticeably better than YOLOv5's, with an average identification rate of more than 64% and a detection time of roughly 203ms. A thorough investigation reveals that the method's spot defect detection effect for alumina ceramic ball production is superior, and it has some practical utility for the batch production line.