An Improved YOLOv7-Tiny-Based Algorithm for Wafer Surface Defect Detection

Wafer surface defect detection is a critical component in the chip manufacturing process. To address the shortcomings of manual inspection and the limitations of existing machine learning methods, this paper proposes a wafer defect detection algorithm based on an improved YOLOv7-tiny. First, a coordinate attention (CA) module is incorporated into the feature extraction network to enhance the network's ability to learn features at defect locations. Next, a lightweight convolutional module, ghost shuffle convolution (GSConv), is introduced into the feature fusion network to reduce the network's parameter count while maintaining a certain level of detection accuracy. Finally, the loss function is optimized by adopting IoU with minimum points distance (MPDIoU) to address issues such as small sizes and dense distributions. Experiments conducted on a self-constructed dataset show that the improved algorithm achieved a mean Average Precision (mAP) of 90.1%, representing a 3.2% increase over the original algorithm. The model size is only 5.85MB and the detection speed has been effectively enhanced, providing valuable insights for research in industrial real-time detection applications.