基于 STM32 与改进 Yolov5s 的铝片生产表面缺陷检测系统设计

打开文本图片集

DOI：10.19981/j.CN23-1581/G3.2026.12.007

中图分类号：TP242.6

文献标志码：A

文章编号：2095-2945（2026）12-0025-04

Abstract： This paper uses OV5640 to collect image data and builds TCP server through uc/OS-II and LWIP protocol， and sends image data to Web front-end through 5G network. The improved Yolov5s model based on Tensorflow framework is used for defect detection and analysis， and the Flask framework is used to build a Web that supports high concurrency rate. OV5640 collects image data and carries out target recognition and then transmits it back to the front end for analysis and viewing， and the sorting system drives the sorting to rotate forward and backward according to the recognition results to sort the aluminum plate in real time， which can achieve intelligent real-time detection. The experiment shows that the FPS value of the system is 14.5， and the network connection packet loss rate is ⩽0.1% . Based on the improvement of the recognition model， its weight only increases by 2 M， the mAP values of IoU of 0.5 and 0.5：0.95 are increased by 1.9% and 2.2% respectively.

Keywords： STM32; Yolov5s; 5G; Flask frame; aluminum sheet defect detection

在铝片生产过程中，其表面易产生擦伤、褶皱、针孔等瑕疵，将直接影响产品的质量和安全性。（剩余4286字）