Research and Design of YOLOv5-Based Industrial Defective Product Inspection System

Industrial defective product detection is an important quality control tool, which is of great significance to ensure the safety and reliability of products. In this paper, we propose a design of an automatic industrial defective product monitoring system based on YOLOv5 model and FPGA hardware platform for industrial defective product detection task. This paper first introduces the principles and loss functions related to the YOLOv5 model, and then elaborates the overall system architecture design, including hardware selection, hardware platform construction, optimization, quantization, mapping and verification steps of the YOLOv5 model. The system designed in this paper integrates deep learning and FPGA development techniques to configure and map the ZCU102 FPGA embedded platform so that it can load and execute the YOLOv5 model, and communicate and interact with external devices to realize an end-to-end industrial defective product detection system.

I. Introduction

In industrial production, product quality inspection is an important part to ensure product qualification rate and reduce cost. Traditional methods of industrial defective product inspection mainly rely on manual vision or special equipment, which have some drawbacks, such as manual vision is easily affected by subjective factors, fatigue, and environment, while special equipment requires high cost and maintenance. With the development of deep learning technology, target detection methods based on YOLOv5 networks have achieved remarkable results in natural scenes, and these methods can also be applied to industrial salvage detection to improve the accuracy and efficiency of detection. The purpose of this paper is to study and design a yolov5-based industrial defective product detection system, which can perform fast, accurate and stable defective product detection for different types and sizes of industrial products, and achieve automated and intelligent quality control.