Research On Textiles Color Segmentation And Extraction Based On Hyperspectral Imaging Technology

In the textile industry,color is an important feature in evaluating the quality of textiles.At present,color measurement methods such as standard color cards,machine vision,spectrophotometers,etc.have certain limitations in measurement accuracy and efficiency.The use of standard color cards for color comparison is easy to be affected by subjective factors of testers,and the measurement efficiency is low.The digital camera in the machine vision system cannot obtain the spectral information of the color,and the measurement accuracy is low.The spectrophotometer can only obtain the average color of the sample in the aperture,and it cannot measure directly on textiles such as single yarns and printed fabrics.Aiming at the limitations of traditional color measurement methods,a new textile color measurement method which uses the hyperspectral imaging system to obtain the hyperspectral image of textile and then segments and extracts the color information from the image was proposed.The measurement method can obtain fine spectral information and spatial information.It has higher color measurement accuracy and can measure the color of a single yarn and multi-color textiles directly.The main research contents of this paper are as follows:1.Spectral consistency correction of the hyperspectral imaging system:Due to the differences in geometry and colorimetric principles of the hyperspectral imaging system and the spectrophotometer,the measured spectral reflectance is inconsistent.Aiming at this problem,a spectral consistency correction algorithm based on improved R-model was proposed.The idea of the algorithm was to select a bands combination with the best correction accuracy from all bands of spectral reflectance using the Partial Least Squares(PLS)regression.The experimental results showed that the proposed algorithm was better than the traditional correction algorithms in improving the spectral consistency of the hyperspectral imaging system2.Yarn segmentation and color extraction:The limitation of spectrophotometer is that it cannot directly measure the color of a single yarn.It becomes possible for hyperspectral imaging system to measure the color of a single yarn because of its capability of obtaining fine spectral information and spatial information.A yarn segmentation algorithm based on Frechet distance spectral matching was proposed.The algorithm used the difference between spectral curves of a background pixel and a yarn pixel to separate yarn pixels from background pixels in the yarn hyperspectral image by the spectral matching method based on the Frechet distance,so that the single yarn could be separated from the background.The experimental results showed that the proposed algorithm could accurately separate the yarn from the background,and it performed better than other segmentation algorithms in retaining edge information of the yarn.3.Color segmentation and extraction of multi-color yarn woven fabric:To solve the problem of the large computation while directly segmenting hyperspectral image of multi-color yarn woven fabric,a color segmentation algorithm based on Frechet distance space transformation was proposed.The proposed algorithm firstly generated the gray image by using the Frechet distance space transformation,and then the improved watershed algorithm was used to segment the gray image.Finally,the improved K-means clustering algorithm was applied to merge the over-segmented color regions.The experimental results showed that the proposed algorithm could accurately segment the different color regions of multi-color yarn woven fabrics.4.Color segmentation and extraction of printed fabric:Due to the rich color and complex patterns of printed fabrics,the number of colors cannot be directly determined by human eyes.In order to solve the problem,a color segmentation algorithm based on Self-Organizing Maps(SOM)neural network and Density Peak Clustering(DPC)algorithm was proposed.The proposed algorithm firstly used the SOM neural network to initially cluster the dataset,dividing the data with similar color features to the same neuron,and then the DPC algorithm was used to deeply cluster the neurons in the output layer of SOM neural network.Finally,the clustering effectiveness evaluation indexes were applied to determine the optimal number of colors for automatically segmenting the printed fabric.The experimental results showed that the proposed algorithm performed better than other segmentation algorithms in color segmentation and execution time.