基于融合特征与支持向量机的控制图模式识别

为了提高控制图模式识别的精度, 将控制图模式的原始特征与形状特征相融合得到分类特征, 并采用支持向量机进行模式分类的控制图模式识别。融合所得特征既保持了控制图模式的原始特征所蕴涵的模式全局特性信息, 又通过引入形状特征对部分易混淆模式的局部几何特性进行强化, 使不同模式间的区分度得到有效提高; 而以支持向量机作为模式分类器保证方法在高维度特征和小样本条件下也能获得较好的识别性能。仿真实验结果表明所提方法的识别精度相比其他几种基于形状特征的控制图模式识别方法有明显提高。

Recognition of control chart patterns based on feature fusion with support vector machine

In order to improve the accuracy of control chart patterns recognition (CCPR), this paper proposed a new CCPR method, which extracted shape features from control chart pattern to make them fused with the raw features, i. e. the raw data of control chart pattern, and then based on that to execute pattern classification with support vector machine (SVM). The fusion of features effectively enhanced the discrimination between different patterns by means of strengthening the local geometrical properties of confusable patterns with shape features as well as keeping the global property information contained in raw features of each control chart pattern. Moreover, the used of SVM as classifier ensures this method a well recognition performance even under a condition of high dimension feature and small training sample number. The results of simulation experiments demonstrate that the proposed method can get an improved recognition accuracy compared with several other CCPR methods based on shape features.