Year: 2024
Type: conference-paper
Abstract: The quality of semiconductor lead frame exposure products will affect the quality of the finished lead frame, and then affect the life of the chip. Affected by the production process, there will be copper exposure, copper deficiency, foreign body and other defects on the surface of the exposed product, these defects are generally small in size, complex and changeable structure, and it is not easy to distinguish from the background texture, and leakage and misjudgment often occur in the detection process. To solve the above problems, this paper proposes an improved YOLOv8 defect detection algorithm. First, DSCONV was introduced to improve the sensitivity of the model to foreign body defects shaped like hair analogs. Secondly, the SPPFCSPC module is used to replace the SPPF module in the original network Backbone and fuse the feature information of different scales. Finally, in order to further improve the detection accuracy, CBAM, a hybrid attention mechanism with better effect, is added to enhance the ability of the network to represent defect features. The experimental results show that the improved YOLOv8 model has made remarkable progress in improving the detection results, and the MAP value reaches 91.1%.
Authors: Wanyu Deng, Xinlei Shang, Zebin Jiang
Institution: Xi’an University of Posts and Telecommunications
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4411730219