Defect Detection Algorithm Of Fiber Coil Winding Based On Machine Vision

The optical fiber coil wound by ring-shaped multi turn fiber is the core component of interferometric optical fiber gyroscope.It propagates two opposite light waves to produce Sagnac phase difference.The quality of optical fiber coil directly affects the measurement accuracy of fiber-optic gyroscope.Therefore,real-time monitoring of fiber coil winding system has great practical value both in theory and engineering.The traditional monitoring of optical fiber winding quality by manual visual inspection can not meet the requirements of high precision and high efficiency in modern production.As a non-contact measurement method,machine vision has the advantages of high detection accuracy,high speed,low cost and high safety.It has been widely used in industrial detection.Therefore,based on machine vision technology,focusing on the practical problem of optical fiber coil winding defect detection,this dissertation adopts image analysis,model construction and other methods to carries out the following research work:(1)By analyzing the defect characteristics,a fast defect detection method based on wavelet transform is proposed.Firstly,the morphological method is used for image preprocessing to obtain the ideal region for subsequent analysis.Then,the projection transform and ability of discrete wavelet transform to signal localization time-frequency analysis is used to locate the defect.Experiments show that this method achieves high accuracy in fiber coil defect detection task.Aiming at the periodic complex texture features of fiber coil images,a defect segmentation method based on texture features and low rank matrix representation model is proposed.Firstly,a texture feature library is constructed according to a variety of feature descriptors to encode the texture information of the defect image.On this basis,the matrix is decomposed into a low rank matrix representing highly redundant information and a sparse matrix corresponding to the significant defect part,and then the defect information and defect location are separated,which has achieved good performance in defect segmentation.(2)In order to identify the types of defects,a classification model combining support vector machine and adaptive genetic algorithm is constructed.An adaptive adjustment scheme of crossover probability and mutation probability is proposed to optimize the classifier parameters.Experiments show that this method has a high recognition rate for optical fiber winding defects.(3)In order to solve the problem that defect detection and type identification is conducted in two phase in traditional detection methods,according to the advanced achievements of current deep learning,this dissertation explores how to apply deep learning to optical fiber coil quality monitoring to obtain better detection performance,and puts forward an end-to-end defect detection and classification integration method based on deep learning.Through the model improvement,the problem of weak detection ability of the model for small-size targets is improved,and the accuracy of defect detection is improved.In order to enhance the robustness of the network,the strategies of data expansion and network fine-tuning are adopted to prevent the phenomenon of over fitting.According to the requirements of lightweight in practical industrial applications,sparse training and channel pruning strategies are adopted to further simplify the model,speed up and meet the real-time requirements,which provides an effective algorithm and reference for automatic detection in the actual production process of optical fiber coil.To sum up,guided by the actual needs and problems of visual quality inspection in the process of optical fiber winding,this dissertation builds a defect visual inspection system,puts forward a variety of defect detection methods that can be used for optical fiber coil image,and carries out a large number of experiments.Results show that the method proposed in this dissertation is effective in solving the problem of optical fiber winding defect detection.It has important theoretical significance and engineering value for improving the automation level of optical fiber coil production in China.