Research Of Textile Fiber Classification Based On Feature Fusion And Convolutional Neural Network

Textiles are one of the main export products of China's light industrial products.Among them,cotton and hemp fiber account for the highest proportion of textiles.Since in practical applications,the calculation of the cotton and hemp content in textiles is usually identified by skilled technicians under microscopes.The method has the disadvantages of high labor cost and the accuracy may be influenced'by human factors.Therefore,the realization of automatic detection and classification of cotton and hemp fibers has practical commercial application prospects and value.In recent years,the develop-ment and application of machine learning is very extensive,not only in data mining,natural language processing,biological pattern recognition,etc.,but more importantly,it is widely used in computer vision.Among them,the support vector machine has the advantages of strong generalization ability and small calculation amount which provides a good reference value for image classification research.Feature extraction of images is an important basis for image recognition technology.In the study of fine-grained classification,proper feature extraction is crucial for the final classification recognition rate.Since cotton and hemp are natural fibers,they are affected by the variety and growth conditions,so the characteristic parameters of the fibers fluctuate greatly.The convolutional neural network has the character-istics of self-learning,self-organization and self-adaptation,and has good learning ability and generalization ability for the characteristics of natural fiber parameter fluctuation.Therefore,this paper completed the following research tasks:Firstly,this paper built an automatic support vector machine classification system for cotton and hemp fibers which is based on HOG feature extraction.The method initially solved the problem of automatic classification of cotton and hemp fibers with a recognition rate reaching at 84.5%.Then,this paper built an automatic support vector machine classification system for cotton and hemp fibers based on features fusion of HOG feature and LBP feature.The method improved the recognition rate by utilizing the complementarity of texture information and edge contour information.The experimental results showed that the system recognition rate of this feature fusion algorithm reached at 89.9%.Compared with the single HOG feature extraction algorithm,this algorithm had a 5.45%improvement in recognition rate.Finally,based on the above research,a deep learning and transfer learning method was introduced.A convolutional neural network algorithm based on MobileNet and transfer learn?ing was proposed.After several adjustments to the convolutional neural network parameters,the experimental results showed that the system recognition rate of this algorithm finally reached at 95.05%.Compared with the previous feature fusion algorithm,the recognition rate of this algorithm was increased by 5.15%.Compared with a single HOG feature extraction algorithm,the recognition rate of the algorithm was increased by 10.55%.The experimental results showed that the convolutional neural network classification algorithm achieved a good automatic classification effect on cotton and hemp fibers.Compared with the traditional feature extraction algorithm and feature fusion algorithm,the classification recognition rate was significantly improved.