随着新能源行业的发展，市场对锂电池的需求量及质量要求越来越高。极片作为锂电池的重要组成，其缺陷会严重影响锂电池性能及使用寿命，甚至导致安全事件。为了避免这些问题，需要对极片进行缺陷检测，并判断缺陷类型及时调整极片生产工艺，防止缺陷再次发生。然而传统目测方式已经无法适应工业的发展需求，因此研究基于机器视觉的极片缺陷检测与分类系统对提高检测效率、降低生产成本具有重要意义。

本文以锂电池极片为研究对象，根据极片制造工艺，分析缺陷产生的原因及缺陷特点，结合机器视觉技术设计极片缺陷检测与分类系统。根据系统设计进行设备选型，完成图像采集。为降低噪声及机械振动对极片图像质量的影响，采用双边滤波和灰度变换对图像进行预处理。使用Sobel边缘检测和自适应阈值算法分离出极片缺陷，并对缺陷目标进行形态学处理，然后标记缺陷完成检测。分析缺陷特征时，使用一种改进的K-Means算法完成SURF特征聚类，并量化表示为BoF-SURF特征。将该特征与灰度特征加权融合，改善光照影响及单个特征对缺陷描述不全面而导致准确率低的问题。最后，将融合特征作为SVM的输入进行缺陷分类，并使用改进粒子群算法优化核参数。通过实验分析，本文算法准确率为94.43%，与两种特征单独用于分类时相比，提高了5.23%~12.2%，具有较好的分类性能。

        在算法研究和功能需求分析的基础上设计系统软件，使其具有缺陷检测与分类、缺陷结果保存以及历史查询等功能。验证结果表明，系统可有效实现极片缺陷的检测与分类，具有一定的可行性。

With the development of the new energy industry, the demand and quality requirement of lithium batteries are getting higher and higher. As an important part of the lithium battery, pole piece quality will affect the performance and service life of lithium battery seriously, and even lead to safety incidents. It is necessary to detect pole piece defect to avoid these problems, at the same time, identify the type of defects to adjust the production process in time to prevent the defect from recurring. However, traditional detection method has been unable to meet the needs of industrial development, so a defect detection and classification system of pole piece  is designed, and it has a great significance to improve the inspection efficiency and reduce production costs.

In this paper, lithium battery pole piece is taken as the research object. According to the manufacturing process of pole piece, the causes and characteristics of defect are analyzed, a detection and classification system of the pole piece defect is designed in combination with machine vision technology. According to the system demand, the type of equipment is determined to acquire pole piece images. Due to the influence of noise and mechanical vibration, the preprocessing method of bilateral filter and grayscale transformation is used to improve the image quality. With the help of Sobel edge detection and adaptive threshold, the pole defect target is isolated, and the image further processed by using morphology, then the detection is completed by marking defect. In the process of studying defect characteristics, an improved K-means algorithm is applied to complete SURF feature clustering, and it quantified as BoF-SURF feature. In order to solve the problem of low classification accuracy caused by the effect of illumination and the incomplete description of defect by a single feature, the BoF-SURF feature and grayscale feature are weighted fusion. Finally, the fusion feature is used as the input of SVM for defect classification, and an improved particle swarm optimization method is adopted to optimize the parameters. Experimental results show that the accuracy of the algorithm in this paper is 94.43%, compared with the two features used separately, the accuracy is increased by 5.23%~12.2%, and it has better classification performance.

       The system software is designed on the basis of algorithm research and functional requirement, it has functions such as defect detection and classification, result storage, and historical query. The verification results show that the system can effectively realize the detection and classification of pole piece defects and have certain feasibility.