基于BP神经网络的织物表面绒毛质量的检测方法

Method for detecting fluff quality of fabric surface based on BP neural network

为了对起毛工艺后的织物表面绒毛状态进行客观评定,提出了基于BP神经网络的织物表面绒毛质量的检测方法。以光切成像原理采集绒毛轮廓图像,利用自适应图像分割方法对绒毛区域进行分割,将得到的二值图像应用Freeman链码原理提取织物的上边缘轮廓坐标,以此作为BP神经网络的输入对BP神经网络进行训练,并将训练得到的2组权值根据BP神经网络的计算过程进行验证,提出应用激活函数和训练的权值相结合直接计算的方法。应用基于光切成像原理搭建的绒毛织物检测平台,对4种不同颜色和不同起毛工艺加工后的织物进行检测,准确率为93.02%;且权值的计算结果与网络实际计算结果相符合,因此可以直接利用网络训练的权值做矩阵运算,缩短实际检测的时间。

In order to evaluate objectively the fluff state of fabric surface after the raising process,a method for detecting the surface fluff quality of the fabric based on BP neural network was proposed.The fluff contour image was collected by the principle of optical imaging,and the fluff region was segmented by adaptive image segmentation method.The Freeman chain code principle was applied to the obtained binary image to extract the upper edge contour coordinates of the fabric,and the BP neural network was trained as an input of the BP neural network.The two sets of weights obtained through the training were verified according to the calculation process of the BP neural network,and the method of applying the activation function and the weight of the training combined with the direct calculation was proposed.Applying the fluff fabric detection platform built on the principle of optical imaging,the fabrics processed by four different colors and different fluffing processes were tested.The detection accuracy rate reached 93.02%,and the calculation result of the weight is shown consistent with the actual calculation result of the network.This suggests that the weight of the network training may be used directly for the matrix calculation,which can shorten the actual detection time.