Research On The Technology Of Ring Spinning Machine Break Detection And Special Parts Recognition Based On Deep Learning

The automatic joint of ring spinning is a difficult problem in the textile industry.Nowadays,there is no mature solution based on computer vision for this problem.However,it will be a new technological revolution for the textile industry to do the research and make the application of the automatic joint technology.In order to realize this function,it's necessary to identify the broken condition and the positions of several parts of the spinning during the process of automatic joint.Therefore,this paper proposes an algorithm which can detect the broken yarn of ring spinning frame based on deep learning technology.What's more,an algorithm which is for the identification of the collar,guide hook and front rollers of the ring spinning frame is also proposed.First of all,this paper proposes an image enhancement algorithm of spinning frame twisting section,which preprocesses the image of spinning frame twisting section,and enhances the features of yarn in the image based on gray transformation and Laplace operator.Then,aiming at the problem of yarn breakage,an algorithm of yarn breakage detection based on deep learning is proposed.According to the features in the yarn images,the analysis of real-time and the demand of industrial application,a lightweight classification model based on DBYC(deep broken yarn classification)convolutional neural networks(CNN)is constructed.This model includes: building a lightweight convolution unit CEIDSConv-branch(Channel exchange improved depthwise separable convolution)by improving the depthwise separable convolution and proposing a channel fusion operation which is based on address exchanging,adding attention mechanism to fuse high-level and low-level feature information;proposing a acceleration strategy in the full-connected layer based on cyclic block matrix and hadamard transformation,adding a local response normalization layer to enhance the generalization ability of the model and using a dropout layer to avoid over fitting.This algorithm is deployed on Intel Core i7 8700 k CPU.The experimental results show that the algorithm has a good recognition effect,a generalization ability and fast running speed.An algorithm based on deep learning is proposed in order to identify and detect the position of the ring,guide hook and front rollers on the spinning frame.Firstly,this paper improves Faster R-CNN by using deep residual network(Res Net50)with deeper network and less computation instead of the original VGG16,replacing NMS with Soft-NMS,modifying the number and the size of RPN anchors to adapt the size of spinning frame parts,adding online hard case mining mechanism to solve the problem of imbalance between positive and negative samples and useing migration learning to reduce the training time of the model.Finally,this paper proposes an algorithm consisted of image pyramid and region of interest to accelerate the algorithm.The experimental results show that the accuracy of our algorithm has been improved a lot.