Lightweight Neural Network-based Real-time PCB Defect Detection System

Defect detection is to discover and classify the possible defects in the target images, which is an indispensable part in the intelligent manufacturing industry. In the real environment, the defects of industrial products are usually small or not evenly distributed, and the computing power and storage capacity of the detection equipment are too small, thus leading to poor detection results. Given the present situation, this paper formulates a defect identification system using a lightweight neural network design. Based on the good characteristics of small-scale defect detection of YOLOv5s, we make this network lighter. Reduce the amount of computation while ensuring that the accuracy is still in the acceptable range. Considering the actual situation, we also built a device to simulate a real production line scene and designed the human-computer interaction interface. In order to validate the feasibility, we unite the lightweight network with an interactive interface to carry out experiments.The experimental results demonstrate that the lightweight network achieves faster computing speed and higher accuracy.