中文摘要：

针对锂电池生产过程中气胀检测依赖于作业人员感觉和经验判断，主观性强且效率低的情况，提出一种基于机器视觉的气胀锂电池在线检测方法。特定角度和强度光源照射下，气胀电池与合格电池呈现不同的反射光斑分布。基于此特点，提取电池图像反射光斑区域特征作为分类器的输入。为满足在线检测系统实时性要求高的特点，提出粗糙集属性约简与C-SVM相结合的方法建立分类模型。首先基于属性重要度的粗糙集属性约简方法优选样本特征集，然后采用K折交叉验证和网格搜索法对C-SVM进行参数寻优，建立分类模型。实验结果表明，属性约简的锂电池检测方法减少了冗余信息的干扰，降低电池样本特征空间维数，缩短了检测时间，提高在线检测效率，检测精度达到96.7742%，为生产过程中气胀电池的自动化分选提供了一种有效的方法。

英文摘要：

During the production of lithium battery, swol en battery detection mostly depends on worker＆#39; feeling and experience, so the detection result is subjective and ineffective. The swol en lithium battery on-line detecting method based on computer vision was proposed. Under light irradiation with certain angle and intensity, swol en battery and qualified battery show up different light-spots distribution. The light-spots region features on lithium battery image were extracted as the inputs of classification model. The classification model combined rough set attribute reduction with C-SVM was proposed to meet the high requirement of real-time. Firstly, rough set attribute reduction method based on attribute importance was used to optimize features. Then the classification model was built by K-fold cross-validation strategy and grid searching optimizing C-SVM parameters. The experiments show that the lithium battery detecting method based on attribute reduction can reduce data redundancies, decrease the feature dimension of battery samples, shorten the detecting time and improve the detection efficiency on-line. The proposed method can achieve a recognition rate of 96.774 2%, providing an efficiency detection method for swol en battery automated separation in production process.