A fabric principal component detection method based on a convolution neural network

Abstract

The invention discloses a fabric principal component detection method based on a convolution neural network. The method comprises that firstly, an optical amplification device is used to construct a fabric picture library of each principal component; then, a convolution neural network is constructed by using cavity convolution and depth separable convolution techniques; and finally, the image database is transformed into HSV color space, and then the convolutional neural network is trained to get the network which can be used for fabric principal component detection. According to the present invention, the detection personnel do not need to have the relevant professional knowledge of fabric composition detection, the enlargement multiple of the image is low, pictures are not collected by amicroscope, the detection operation threshold is lowered, and the detection flow is simplified. The convolution neural network designed in the invention can simultaneously realize the detection of the principal components of a plurality of fabrics, the model has strong universality, and the convolution neural network has simple network structure, small network scale, short training time and improved detection accuracy compared with other convolution neural network structures.

Description

technical field [0001] The invention belongs to the technical field of fabric component detection, and in particular relates to a fabric principal component detection method based on a convolutional neural network. Background technique [0002] With the development of the textile industry, the types of textile fabrics are increasing day by day. On the one hand, it is due to the research and development of new textile fibers, which directly provide us with more functional and textured clothing fabrics; The research and development of new blended fabrics meets people's requirements for specific clothing fabrics to the greatest extent on the basis of existing textile fibers. Blended fabrics can give full play to the advantages of different fibers at the same time, and can also increase product types and reduce product costs. Therefore, the emergence of new blended fabrics has become a very important development trend in the textile industry. At the same time, the composition o...

AI Technical Summary

Problems solved by technology

At the same time, the composition of textile fabrics is one of the important indicators to measure the quality of textiles, and its detection difficulty and challenge have also increased; Therefore, it is particularly important to find a fast, convenient, and low-threshold fabric component classification method.

[0003] The traditional detection methods of fabric composition include visual inspection method, microscope observation method, chemical method and physical method. These methods require the inspector to have certain professional knowledge, human subjective factors are greatly affected, and the requirements for professional equipment are also high. The fabric needs to be disassembled, and the chemical reagents required by the chemical method not only pollute the environment, but also damage the health of the tester

[0004] Near-infrared spectral analysis technology uses known composition textiles and its near-infrared spectral information to establish a model, and uses this model to quickly detect unknown textile components. Environmental consistency has high requirements; at the same time, this method is suitable for fabrics with smooth surface, simple structure, and consistent front and back sides, and the structure and uniformity of fabric will also affect the use of this method

[0005] Hu Jueliang et al. proposed in the article "Research on Fabric Classification Based on Bayesian Method" to use Bayesian decision theory to classify fabric images by extracting morphological structure parameters. The area, circumference and moments of each order are used to calculate the circularity of each fiber and input it into the Bayesian model as a parameter to obtain the classification results. The classification accuracy of cotton, linen and silk is 93.2% respectively , 91.5% and 90.2%; however, this method is only for a single fiber, which requires a higher optical magnification of the collected image, is more difficult to operate, the extracted image features are relatively single, the model has poor versatility, and the classification accuracy is low

[0006] In the article "Cotton / flax single fiber recognition based on fiber longitudinal microscopic image", Ying Lebin et al proposed to classify cotton / flax fiber by using the single fiber longitudinal fiber image through the least squares support vector machine classifier. For the longitudinal fiber image of a single fiber, first remove the background of the fiber, and then use the method of combining morphological ratio calculation and background area growth to obtain the target area of ​​the fiber. In order to obtain better filtering, their vertical integral projection sequence is obtained from the area map, binary map and thinning map in the vertical direction of the fiber skeleton, and a total of 6 parameters of the coefficient of variation CV and the average value of the three sequences are extracted. These 6 parameters are used as the characteristic parameters of cotton / flax fiber to train the least squares support vector machine classifier. The test results of the test set show that the classifier has a correct recognition rate of 93.3% for cotton / flax short fiber; but the The method only targets a single fiber, requires high optical magnification of the collected images, and is difficult to operate. It only classifies cotton and flax fibers, and the model has poor versatility.

Images