Research On The Classification Method Of Textile Color Fastness Based On Deep Learning

Color fastness is an extremely important index for controlling the production quality of textiles and inspecting finished products.The current color fastness grade evaluation of textiles is based on the staining of the lining fabric^[1].When evaluating the color difference between the original sample and the sample,the visual method is basically used^[2].This process is really susceptible to the external environment and the personal subjective factors of the assessor.It completely relies on the practical experience of the inspector,and there are great subjective factors^[3].At present,the existing solution is to use instrument measurement method to achieve automatic color fastness rating^[4],which is objective,stable and not affected by personal factors.However,due to various objective issues such as research bottlenecks,sampling,standards,etc.,the instrumental method of rating has not been widely used instead of the visual method,and the measurement results are mostly used as references for manual evaluation rather than direct results.^[5-7].Therefore,it is necessary to explore and research a smarter and more convenient auxiliary rating method for textile color fastness.Improve defects.This article uses image processing technology and deep learning to explore and verify the intelligent rating method of fabric color fastness.It has realized the innovatate combination and intelligent promotiasion of computers'technology and traditional textile examing.The main idea is to make samples and collect images for scenes and problems,and use image processing technology for image preprocessing and segmentation.Under the actual conditions of multiple classifications and small samples,a suitable neural network is used for training and learning to intelligently and automatically give the newly input rating object rating level.The feasibility and efficiency of the staining fastness rating of this research are verified through examples of visual rating and instrument rating.The main research in this article discusses the following:(1)Based on the experimental samples collected by all parties after the textile color fastness rating,the evaluation samples that are missing in the actual rating will be supplemented experimentally,and the sample size will be supplemented,and the lightest and darkest standard samples for each rating level will be made as an image Information input defines a clear range of characteristics.(2)Through literature review and actual shooting experiments,appropriate acquisition parameters are determined to obtain a complete stained lining image,and specific image processing such as multi-angle preprocessing and staining feature image segmentation are used to study textile single fiber stained samples and their collected images,Multi-fiber staining rating samples of textiles and their collected images,summarizing their textile characteristics and the impact on deep learning,from the perspective of image clarity,clear features and other image processing perspectives,textile structure,textile surface characteristics,and experimental operations.Perform pre-processing and featured extraction on the image to ensure the clarity and accuracy of the obtained image,thereby ensuring in-depth learning of the input image information to reduce the influence of irrelevant factors.Establish a three-level classification database according to the type of lining,rating level and neural network sample requirements to form a complete and rich image information system,and obtain a clear structure and hierarchical data information database,which lays a good foundation for end-to-end neural network learning.(3)Design a neural network for the textile color fastness rating scene,constantly search for problems and make adjustments based on literature comparison,summary and trial operation,stitch the corresponding deep learning modules,and the rating level is given in the way of image recognition and classification,based on the exploration of the super deep convolutional neural network.SPP,SVM and other modules are innovatively introduced into unsupervised learning,which solves the practical problems of multi-level classification,small sample size,and inability to extract expert rules in the rating process,and realizes the staining of single fiber.Fastness image feature extraction and recognition and the simulation of the gaze rating process are simple to operate;combine the influence of different raw materials of multi-fiber lining,the characteristics of multiple rating levels and the analysis of the results of single fiber staining rating to construct a new deep learning idea.The creation of level supervision principles and corresponding algorithms and modules realizes the simulation of the multi-fiber lining staining rating scene,the operation is simple and fast,and the deep learning algorithm for the appropriate rating scene is constructed.(4)Carry out examples of textile color fastness ratings and analysis of results.100 groups of single-fiber staining samples and 162 groups of multi-fiber staining samples were actually rated and verified.By comparing the results of computer and visual ratings and instrument ratings,the accuracy of single-fiber staining fastness was within the half-level tolerance range.Up to 98%,the accuracy of multi-fiber staining is 90.12%,the accuracy is higher,and the efficiency is higher than the instrument rating method.The results demonstrate that the selected image processing technology has a good effect on image enhancement and segmentation,and has an auxiliary effect on the subsequent improvement of the accuracy of deep learning;under the conditions of small samples and multiple classifications,deep learning has a strong color for textile staining samples.The rating accuracy rate of the degree is over 90%,which is efficient,objective and accurate.The combination of image processing and deep learning can simplify the rating operation,reduce the work pressure of inspectors,and achieve the effect of improving accuracy and work efficiency.The color fastness rating of textiles can be carried out by applying deep learning.There is no need to change the existing rating process and operation,the operation is simple,the rating results are stable and not affected by subjective factors,the efficiency is higher than the instrument method,and the application is more feasible.The research results of this paper are of high reference value.A new method for grading textile staining color fastness is proposed,which improves the deficiencies of the current methods,provides a better idea for realizing the use of neural networks to replace the human eye grading process,and expands deep learning and textile application scenarios'combination.