软包装锂离子电池的表面凸点缺陷检测  

Surface bump defect detection for pouch Li-ion battery

作　　者：曾臻 王宏博[1,2] 王正家 何涛 ZENG Zhen;WANG Hong-bo;WANG Zheng-jia;HE Tao(Hubei key Laboratory of Modern Manufacturing Quality Engineering,Wuhan,Hubei 430068,China;Hubei University of Technology,School of Mechanical Engineering,Wuhan,Hubei 430068,China)

机构地区：[1]现代制造质量工程湖北省重点实验室,湖北武汉430068 [2]湖北工业大学机械工程学院,湖北武汉430068

出　　处：《电池》2023年第3期284-288,共5页Battery Bimonthly

摘　　要：软包装锂离子电池铝塑膜外包装凹凸不平且存在反光现象,表面图像的凸点缺陷辨识度低,传统方法很难进行准确识别。分析软包装锂离子电池凸点缺陷图像特征及视觉检测系统,在频域上对图像采用高斯滤波器进行预处理,以达到去除噪声和对缺陷区域图像增强的效果。用逆傅里叶变换将图像从频域转回空间域,最后将处理后的图像导入基于语义分割方法的深度学习模型,进行凸点缺陷检测。对400组缺陷样本进行测试,结果表明:所提方法对软包装锂离子电池缺陷检测的准确率达到95.75%;而未经频域图像增强方法处理的检验准确率仅为44.00%。检测结果准确率得到提高,说明该方法能对软包装锂离子电池表面图像的低辨识度凸点缺陷进行检测,具有一定的实用价值。Because of the uneven and reflective appearance of the aluminum-plastic film outer packaging of the pouch Li-ion battery,the identification of the bump defect of the surface image was low,which was difficult to be accurately identified by traditional methods.The image features of bump defects of pouch Li-ion battery and visual detection system were analyzed.Gaussian filter was used to preprocess the image in frequency domain to achieve the effect of removing noise and image enhancement of defect area.Inverse Fourier transform was used to transfer the image from frequency domain to space domain.Finally,the processed image was imported into the deep learning model based on semantic segmentation method for bump defect detection.400 groups of defect samples were tested,the results showed that the defect detection accuracy of pouch Li-ion battery by proposed method reached 95.75%.The detection accuracy without processing with frequency domain image enhancement method was only 44.00%.The detection results had been significantly improved,which proved that the method could detect the low recognition bump defects of the surface image of the pouch Li-ion battery and had a certain practical value.

关 键 词：频域 图像增强 软包装锂离子电池 凸点 缺陷检测 

分 类 号：TM912.9[电气工程—电力电子与电力传动]