储能科学与技术 ›› 2023, Vol. 12 ›› Issue (2): 593-601.doi: 10.19799/j.cnki.2095-4239.2022.0678

• 储能测试与评价 • 上一篇    下一篇

基于LO-RANSAC的锂电池极片表面痕类缺陷检测

姜宝超1,2(), 曾勇1,2(), 韩有军1,2, 胡跃明1,2   

  1. 1.华南理工大学自动化科学与工程学院
    2.精密电子制造装备教育部工程研究中心/广东省 高端芯片智能封测装备工程实验室，广东 广州 510641
  • 收稿日期:2022-11-16 修回日期:2022-11-21 出版日期:2023-02-05 发布日期:2023-02-24
  • 通讯作者: 曾勇 E-mail:jiangbcy@163.com;auyzeng@scut.edu.cn
  • 作者简介:姜宝超（1998—），男，硕士研究生，研究方向为图像处理和深度学习，E-mail：jiangbcy@163.com；
  • 基金资助:
    国家重大科技专项02专项(2014ZX02503)

Scratch defect detection of lithium battery electrode based on LO-RANSAC algorithm

Baochao JIANG1,2(), Yong ZENG1,2(), Youjun HAN1,2, Yueming HU1,2   

  1. 1.College of Automation Science and Engineering, South China University of Technology
    2.Enginering Research Center for Precision Electronic Manufacturing Equipment, Ministry of Education & Guangdong Provincial Engineering Laboratory for Advanced Chip Intelligent Packaging Equipment, South China University of Technology, Guangzhou 510641, Guangdong, China
  • Received:2022-11-16 Revised:2022-11-21 Online:2023-02-05 Published:2023-02-24
  • Contact: Yong ZENG E-mail:jiangbcy@163.com;auyzeng@scut.edu.cn

RichHTML

33

PDF

209

摘要：

针对锂电池极片表面的痕类缺陷检测准确率低、误检率和漏检率高的问题，提出了一种基于局部最优化的随机抽样一致性(locally optimized random sample consensus， LO-RANSAC)的痕类缺陷检测算法。首先，针对锂电池极片表面存在的椒盐噪声、大噪点多的问题，提出了一种改进的自适应中值滤波和基于连通域的滤波算法。其次，针对检测痕类缺陷准确率达不到预期以及误检率漏检率较高的问题，引入一种局部最优化的RANSAC算法。最后，给出了一种基于LO-RANSAC的痕类缺陷分类方法。实验结果表明：本文所提算法相较于标准RANSAC检测准确率提高了5.9%，相较于基于卷积神经网络算法准确率提高了15%，达到了98.2%；多种算法中本工作算法对于痕类缺陷的检测误检率和漏检率最低；平均检测速度较标准RANSAC算法提高了1.7倍，每秒钟检测的图片数量FPS(frame per second)达到12.49。本工作算法具有较高的检测准确率、较低的误检率及漏检率，检测速度达到实时检测要求，因此可满足锂电池极片表面的痕类缺陷检测需求，解决了锂电池极片表面痕类缺陷自动检测难题。

关键词: 痕类缺陷, 自适应中值滤波, RANSAC, 缺陷检测

Abstract:

An algorithm based on locally optimized RANSAC (locally optimized random sample consensus, LO-RANSAC) is recommended for the problems of low trace defect detection accuracy, high false detection rate, and high missed detection rate of scratch defect detection on the surface of a lithium battery. First, in response to the problem of pepper noise, large noises that exist on the scratch defects of lithium batteries, a filtering algorithm based on improved adaptive median filtering and connected domains filtering is proposed. Secondly, to address the problems that the detection accuracy of detecting trace defects does not meet expectations and the false detection and missed detection rates are high, a locally optimized RANSAC algorithm is introduced. Finally, a scratch defect classification based on the LO-RANSAC algorithm is proposed. The experimental results demonstrate that when compared to the standard RANSAC algorithm, the proposed algorithm's average detection accuracy is increased by 5.9%, when compared to the convolution-based neural network algorithm is increased by more than 15%, reaching 98.2%. Among several algorithms, the algorithm achieves the lowest false positive and false negative rate for trace defects. The average detection speed is 1.7 times faster than the standard RANSAC algorithm, with an FPS (frame per second) of 12.49 images detected per second. The proposed algorithm has a high detection accuracy, a low false detection rate, and missed detection rate, and a detection speed that meets real-time detection requirements, allowing it to meet the detection needs of trace defect on the surface of lithium battery pole pieces and solve the problem of automatic detection of trace defects on the surface of lithium battery pole pieces.

Key words: scratch defect, adaptive median filtering, RANSAC, defect detection

中图分类号: