零件自动分类的BP神经网络实现

Work-piece Pattern Recognition Based on BP Neural Networks

为了提高零件识别速度,事先对零件(模板)进行分类,识别时先判别零件属于哪一类然后再在相应类中进行识别。考虑到工件识别时拍摄的是工件实体的投影图,故提出以三维实体建模零件的生成原理进行分类,即将其分成旋转类、拉伸类、扫掠类、混成类,采用适合分类的BP神经网络实现,并根据零件图像特征选取了均值、三阶矩、一致性、熵、不变矩等特征作为训练样本,并作为神经网络的输入,最后以实例证明了这种方法是切实可行的,且其识别准确率高。

To increase the speed of work-piece identification, a novel identification process was proposed. Firstly the recognized work-pieces were divided into four categories based on the generating grammar of work-pieces. Then the work-pieces and the images were recognized in the stencil gallery matches. BP neural network was used to deal with image pattern classification. The features of the work-piece images such as area ratios, smoothness, consistency, third moment, entropy were choosed as the inputting parameters of the BP neural network to recognize the work-piece. The results of the experiment show that the recognition ratio of the proposed method is high.