Lightweight and fast visual detection method for 3C assembly

Wenbai Chen, Genjian Yang, Bo Zhang, Jingchen Li, Yiqun Wang, Haobin Shi

Abstract: Highlights•By utilizing the CSPNet architecture, BiFPN, Ghost convolution, ViT, and other techniques, we propose an advanced lightweight detection network named YOLOv5-GTB.•A novel, lightweight, high-performance module combination combines the Ghost module, the Transformer encoder module, and the Meta-ACON adaptive activation function.•We provide an efficient solution that can be applied in the 3C assembly scenario, which has high precision and fast detection speed. This solution effectively addresses the problems of target detection algorithms occupying a large amount of system resources, low detection accuracy caused by flexible targets and small-scale heterogeneous components in the scene.