An improved PCB defect detection algorithm for YOLOv7-tiny

Abstract

In the manufacturing of printed circuit boards, due to production processes and other issues that can easily lead to defects in the circuit board. In order to improve the efficiency of circuit board defect detection, a defect detection algorithm for bare PCB based on improved YOLOv7-tiny is proposed. First, a new ELAN structure, New-ELAN, is proposed to replace the ELAN structure in the Head section, and the three detection heads in the Head section are reduced to two. Next, reconnecting the Neck structure and reducing the number of channels to reduce computation. The experimental results show that: under certain training conditions, the improved YOLOv7-tiny's mAP value reaches 93.9%, which is 4.8% higher than the original model. In addition, the speed and size of the improved model remain essentially the same. The improved model has better detection results.