Recognition System Of Microscopic Features Of Similar Fibers Based On Deep Learning

My country is the world’s largest producer and exporter of textiles and apparel.The textile industry is one of my country’s important economic pillar industries.The total export volume in 2021 will reach US$315.47 billion,an increase of 8.4% year on year.In the fiber inspection industry for a long time,the manual inspection method based on a microscope is the current mainstream inspection method.The manual inspection method requires inspectors to observe the morphology of fibers under a microscope for a long time,and this method is very dependent on the inspectors’ subjective experience.Moreover,the microscopically similar structure of fibers makes it difficult for researchers to re-identify,and the efficiency is timeconsuming and labor-intensive.With the development of computer disciplines such as image processing technology and deep learning,computer vision technology has been widely used in image recognition and target detection,which not only greatly improves the recognition speed of samples but also improves the recognition of microscopic similar structures of fibers.and the research efficiency of classification algorithms.Therefore,this thesis proposes the problem of microscopic image processing methods for similar fibers in textiles.By constructing an improved neural network model based on Res Net50,the four most used and highly similar fiber structure characteristics in the market are analyzed.Micro-optical detection platform for detection.First,in view of the fiber image collected by the optical fiber platform is susceptible to environmental interference,the fiber image is screened according to the micro-optical platform,and the interference portion is filtered through the image processing technology,and the appropriate preprocessing step extracts the fiber characteristics to facilitate processing into a data set.Herein,we propose corresponding processing techniques based on the collected fiber images,thereby achieving the purpose of protruding fiber image characteristics.Secondly we apply the deep convolutional neural network structure model to improve the structure and propagation algorithm.In this thesis,the classification model suitable for recognition is selected,the input data is set as fiber image,and the output is the type of fiber.According to the actual situation of fiber image,the selected deep neural network optimization algorithm and network structure are adjusted,and the commonly used classification evaluation index is selected to determine the optimized parameter value.Finally,this thesis establishes a deep neural network model through the Python development environment,adjusts the network structure and performance optimization based on the improved method,improves the network recognition performance,and identifies the types of fibers with similar structures.The optimization model is compared with the classical fiber classification algorithm,and the application of deep learning in fiber identification network is explored.The experimental results show that the model in this thesis has more than99% effect on fiber recognition,which improves the recognition efficiency and reduces labor costs and product production costs.The experimental results show that the identification method in this thesis can effectively distinguish similar fiber structures,and the identification efficiency reaches 99.3%.It can avoid problems such as low efficiency and low accuracy caused by cumbersome and complicated manual operation processes in the traditional fiber identification method,and reduce the production cost.A fiber image recognition system based on a convolutional neural network proposed in this thesis can effectively identify fiber types.