摘要
针对多源干扰环境下输送带纵向撕裂图像质量低和损伤特征信息丢失严重问题,提出一种图像自适应视觉检测方法IAE-YOLOv5。首先,设计IAE模块增强输送带损伤特征信息,改善不稳定光照、雾气等环境干扰导致的图像失真。然后,引入小型网络CNN-PL学习IAE模块各滤波器超参数。最后,通过检测损失联合训练CNN-PL与YOLOv5检测网络,预测合适的滤波器超参数,提升IAE模块处理效果,实现图像自适应增强和检测。实验结果表明:相较于原始YOLOv5,IAE-YOLOv5在自制原始数据集和增强数据集上mAP分别提高1.53%和6.9%。
Aiming at the problem of low image quality and serious loss of damage characteristic information of conveyor belt longitudinal tear under multi-source interference environment,an image-adaptive visual detection method IAE-YOLOv5 was de-veloped.Firstly,the IAE module was designed to enhance the feature information of the tear damage,mitigating the image distor-tion caused by environmental interferences such as unstable lighting and fog.Then,a subnetwork,CNN-PL,was introduced to learn the hyperparameters for filters in IAE module.Finally,by jointly training CNN-PL and YOLOv5 detecting network through detection loss,suitable filter hyperparameters was predicted to improve the processing effect of IAE module,and realize image a-daptive enhancement and detection.The experimental results show that,compared with the original YOLOv5 model,the IAE-YOLOv5 method improves the mean average precision(mAP)by 1.53%and 6.9%on the homemade original dataset and en-hanced dataset,respectively.
作者
沈景轩
王贡献
孙晖
杨泽坤
洪湖城
胡志刚
SHEN Jingxuan;WANG Gongxian;SUN Hui;YANG Zekun;HONG Hucheng;HU Zhigang(School of Transportation and Logistics Engineering,Wuhan University of Technology,Wuhan 430063,China)
出处
《仪表技术与传感器》
CSCD
北大核心
2023年第12期69-74,121,共7页
Instrument Technique and Sensor
基金
海南省自然科学基金面上项目(622MS097)。
关键词
输送带
损伤检测
多源干扰环境
图像自适应增强
联合训练
conveyor belt
damage detection
multi-source interference environment
image-adaptive enhancement
joint training
