基于深度学习的轮胎表面缺陷检测方法研究

打开文本图片集

关键词：轮胎检测；图像增强；深度学习；数据集

中图分类号：TP391.4 文献标志码：A 文章编号：1003-5168（2025）24-0034-06

DOI： 10.19968/j.cnki.hnkj.1003-5168.2025.24.006

Research on Tire Surface Defect Detection Method Based on Deep Learning

FU Yuehai'SUN Cheng³ZHOU Enquan³YANG Liyong2

（1.Hangzhou Qianjiang Civil Defense Equipment Co.，Ltd.，Hangzhou 3112Oo，China; 2.Haian-Shanghai Jiao

Tong University Intelligent Equipment Research Institute，Nantong 2266O0， China; 3.Hua'ao Technology （Suzhou） Co.，Ltd.， Suzhou 215000， China）

Abstract：[Purposes]To address challenges such as blurred weak target features and strong interference in complex scenes during tire manufacturing surface defect detection，this study proposes a collaborative detection framework integrating image enhancement with deep learning optimization，aiming to improve defect recognition accuracy，anti-interference capability，and industrial real-time performance.[Methodsl In the first stage，a multi-scale Retinex-CLAHE joint enhancement algorithm is developed. Through the synergistic interaction of ilumination compensation and adaptive histogram equalization， over-exposure suppression and dark-area detail enhancement are achieved， improving weak defect contrast by 61% . The second stage leverages an improved YOLOv1O architecture，integrating a channel attention mechanism and a dynamic adaptive threshold module to construct a lightweight dualbranch network.Combined with knowledge distilation techniques，the model parameters are compressed to 34% of the baseline model.[Findings] Experimental results demonstrate that the optimized model achieves a mAP of 89.7% ，an F1 score of O.85，a precision of O.85，and a recall of O.90 on real production line data.[Conclusions] The study validates the universality of the multi-modal enhancement and model lightweight collaborative strategy，providing a new paradigm that balances precision and efciency for industrial weak and small target detection.This framework facilitates intelligent quality inspection process upgrades for enterprises， reducing manual re-inspection costs and propeling tire manufacturing toward full-process automated detection.

Keywords：tire detection; image enhancement; deep learning;dataset

0 引言

轮胎表面缺陷是制造过程中的关键质量问题，直接关乎行车安全与产品可靠性。（剩余4944字）