Research on graphite ball surface defect detection algorithm based on YOLOv5

Graphite balls are essential in various industries, including casting, lubricating materials, refractory materials, batteries, and more. With expanding industrial applications, detecting surface defects on graphite balls has become crucial to ensure product quality and performance. Manual detection methods are often inefficient and error-prone, failing to meet modern industrial demands. This research focuses on using the YOLOv5 model for fast and accurate detection of surface defects on graphite balls, enhancing detection efficiency and accuracy. Traditional methods struggle with complex backgrounds, noise interference, and diverse defect morphologies. The study employs the YOLOv5 model with CSPDarknet53 and compares it with SE attention mechanisms, adaptive convolution (SAConv), and BiFormer modules for small target detection. Attention mechanisms improve focus on defect areas, while adaptive convolution enhances feature extraction. The BiFormer module improves accuracy for smaller defects.The final model detects four types of defects: tool marks, sintering marks, scratches, bumps, and risers, providing valuable data for industrial production. This research offers a high-precision solution for graphite ball surface defect detection and serves as a reference for applying neural network models in industrial defect detection.