基于改进YOLOv7网络的织物疵点检测
Fabric defect detection based on improved YOLOv7 network
石玉文 1林富生 1宋志峰 1余联庆1
作者信息
- 1. 武汉纺织大学 机械工程与自动化学院,湖北 武汉 430200;三维纺织湖北省工程研究中心,湖北 武汉 430200;湖北省数字化纺织装备重点实验室,湖北 武汉 430200
- 折叠
摘要
针对传统的目标检测方法不能平衡检测精度、预测速度和轻量级部署模型、实时疵点检测,提出一种改进YOLOv7 网络的轻量化检测模型.首先在主干网络引入轻量级卷积Ghost conv,在保证检测精度的同时降低网络参数量,提高对织物疵点的检测效率;其次添加CBAM注意力机制抑制无用信息,增强特征提取能力;最后在回归损失函数处引入新度量方法α-SIoU替换IoU,加速损失函数的自由度,提高网络模型的精确度.实验表明:该检测模型的准确率P达到 96.27%,平均精度mAP值为 83.84%,模型大小仅为 19.10 MB,有效平衡了疵点检测的准确性、实时性与轻量级部署的问题.
Abstract
Aiming at the failure of traditional target detection methods to balance detection accuracy,prediction speed and lightweight deployment model to realize real-time defects detection,a lightweight detection model based on improved YOLOv7 network was proposed.Firstly,lightweight convolution Ghost conv was introduced into the backbone network to reduce the number of network parameters while ensuring the detection accuracy and improve the detection efficiency of fabric defects.Secondly,CBAM attention mechanism was added to suppress useless information and enhance feature extraction ability.Finally,a new measurement method α-SIoU was introduced to replace IoU in the regression loss function to accelerate the degree of freedom of the loss function and improve the accuracy of the network model.The experiment shows that the accuracy P of the detection model reaches 96.27%,the average precision mAP value is 83.84%,the model size is only 19.10 MB,which effectively balances the accuracy,real-time and lightweight deployment of defects detection.
关键词
疵点检测/YOLOv7/轻量化/CBAM注意力机制/Ghost/conv卷积/α-SIoUKey words
effect detection/YOLOv7/light weight/CBAM attention module/ghost convolution/α-SIoU引用本文复制引用
出版年
2024
NETL
NSTL
万方数据