基于改进DeepLabV3+的轮胎缺陷检测  

Tire Defect Detection Based on Improved DeepLabV3+

作　　者：孙英伟 张岩[1] SUN Yingwei;ZHANG Yan(College of Electromechanical Engineering,Qingdao University of Science&Technology,Qingdao 266061)

机构地区：[1]青岛科技大学机电工程学院,青岛266061

出　　处：《计算机与数字工程》2025年第2期535-539,544,共6页Computer & Digital Engineering

摘　　要：论文提出了一种基于改进DeepLabV3+的轮胎缺陷检测模型。该模型不仅能够定位缺陷的位置,还可以表征缺陷的几何形状。首先,采用ECA通道注意力机制聚焦轮胎缺陷特征,使骨干网络有效提取缺陷的特征信息。然后,通过对空洞空间金字塔池化(ASPP)的改进,提出了一种链式空洞金字塔池化(CAPP)。CAPP在空洞卷积分支前采用链式连接的策略,更加充分地提取特征的多尺度信息。经过实验验证表明,该检测模型能够应对各种类型的轮胎射线图像中的缺陷,并且具有较高的检测准确率。This paper proposes a tire defect detection model based on the improved DeepLabV3+.The model can not only lo⁃cate the location of the defect,but also characterize the geometry of the defect.Firstly,the ECA channel attention mechanism is used to focus on the tire defect features,so that the backbone network can effectively extract the feature information of the defects.Then,a Chained Atrous Pyramid Pooling(CAPP)is proposed by improving the Atrous Spatial Pyramid Pooling(ASPP).CAPP adopts a chain connection strategy before the dilated convolution branch to more fully extract the multi-scale information of features.The experimental verification shows that the detection model can deal with defects in various types of tire radiographic images,and has a high detection accuracy.

关 键 词：缺陷检测 注意力机制 多尺度特征提取 轮胎射线图像 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] TQ336.1[化学工程—橡胶工业]