An Improved YOLOv8 Algorithm for TGV Microvia Defect Detection

Résumé fourni par la source

Defects such as Broken, Deformed, and Incomplete generated during Through Glass Via (TGV) microvias fabrication will seriously impact subsequent metallization and other processes. To this end, this study proposes a fast and high-precision TGV microvia defect detection method based on improved YOLOv8. Firstly, the SPD-Conv is introduced into the original Backbone network to improve the detection capability of small target defects in TGV microvia defect images. Secondly, the DAT (Deformable Attention Transformer) attention mechanism is integrated in the Neck part to achieve adaptive focusing of TGV microvia defect areas by dynamically adjusting the weight distribution of the receptive field. Finally, the network structure of the Head layer is optimized, the self-attention mechanism is added, and the SATHead (Self-Attention Head) detection head is proposed to strengthen the correlation between local defect features and global context information. Experimental results show that the improved model has a detection precision of 96.4% for common TGV defects such as Broken, Deformed, and Incomplete, which is 2.9% higher than the basic model. The mAP@ 0.5 reaches 94.3%, which is higher than the original model. At the same time, the model has a detection frame rate of$\mathbf{1 0 6. 2}$per second, which can meet the realtime detection needs of TGV microvias batch manufacturing. This study provides a new idea for fast and high-precision TGV microvia defect detection.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.