应用卷积神经网络的细纱断纱锭位识别

针对图像检测细纱断头过程中难以精确获取断头纱线位置的难题，研究了基于图像处理的细纱断纱锭位探测系统。安装在巡回小车上的工业相机，在细纱机前巡回过程中拍摄纺纱段的图像后，对细纱机台大梁上的锭位标识字符进行定位切割、形态学处理、字符分割后得到清晰的字符图像，再借助卷积神经网络对字符图像进行识别，并输出断纱锭位编号。通过对神经网络的结构筛选表明，卷积层的特征映射为4、亚采样层池化矩阵的大小为 2、迭代次数为 300 时，神经网络的准确率达97%以上，识别1 幅图像锭位用时1.152 s。实验结果表明，该系统可识别细纱断头纱线位置并输出。

Position recognition of spinning yarn breakage based on convolution neural network

Aiming at the problem that the position of the broken yarn is difficult to be obtained in the process of detecting the yarn breakage, the spindles detection system based on image processing is studied. An industrial camera mounted on a roving car was used to record the image of the spinning section during the roving of the spinning frame. To get a clear character image, the mark of spindles on the beam of the spinning machine was identified by position cutting, morphological processing, character segmentation. And then the character image was classified by convolution neural network to export the number of broken yarn spindle. By mapping the structure of the neural network, it is shown that the feature map of the convolution layer is 4, the size of the sub-sampling pool pooling matrix is 2, the number of iterations is 300, the accuracy rate of the neural network is over 97%, and the identification ofan image spindle uses1.152 s. The system could recognize the positionof yarn breakage and output the signals.