基于深度学习的光纤收卷机器视觉自动检测技术

已有的光纤收卷检测方法的泛化能力和环境适应性均较差，无法应用于工业生产.提出基于深度学习的机器视觉方法对收卷过程中的收卷图像进行分类来解决光纤收卷问题.通过考虑光纤收卷时光纤间力的作用，建立了光纤收卷模型，提出了光纤收卷时排线机构的速度表达式.使用相机采集大量光纤收卷图片形成数据集，搭建并训练神经网络模型用于对收卷情况进行分类.通过实验验证，该方法对间隙状态的识别正确率在94.67%以上，叠线状态识别正确率为100%，检测速度高于实际生产绕线速度，是一种可以和控制系统相结合替代人工收卷并实现自动精密绕线的良好方法.

Machine Vision Automatic Inspection Technology of Optical Fiber Winding Based on Deep Learning

Existing optical fiber winding inspection methods have poor generalization ability and environmental adaptability， and cannot be applied to industrial production. A machine vision method based on deep learning was proposed to classify the winding images during the winding process to solve the optical fiber winding problem. By considering the effect of the force between the optical fibers when the optical fiber was winding， the optical fiber winding model was established， and the speed expression of the arranging mechanism was proposed when the optical fiber was winding. The camera was used to collect a large number of optical fiber winding pictures to form a data set， and a neural network model was built and trained to classify the winding situation. Experimental verification showed that the accuracy of this method for gap state recognition is over 94.67%， and the accuracy of overlapped line recognition is 100%. The inspection speed is higher than the actual production winding speed. It is a favorable method that can be combined with the control system to replace manual winding and realize automatic precision winding.