A method of fur near-infrared spectroscopy identification based on grouping support vector machine

Abstract

The invention discloses a fur near-infrared spectral discrimination method based on a packet support vector machine. The method includes the following steps that (1) near-infrared spectral data of different kinds of fur are collected and preprocessed, and KPCA feature extraction is carried out on the data; (2) a near-infrared spectral data training set of the fur is built; (3) the training set is trained by means of the packet support vector machine to form a fur classifier; (4) fur samples to be discriminated are discriminated by means of the fur classifier. The near-infrared spectral technology is used, the data obtained through spectral preprocessing and feature extraction can capture nuance information of near-infrared spectrums of different kinds of fur, the variety of the fur is discriminated by means of the packet support vector machine, the method has the advantages that detection speed is high, classification accuracy is high and the fur is not damaged, and classification of different kinds of fur can be achieved.

Description

technical field [0001] The invention relates to the field of fur type identification, in particular to a fur near-infrared spectrum identification method based on a grouping support vector machine. Background technique [0002] From ancient times to the present, fur has been regarded as a symbol of wealth and social status. Common furs include fox fur, rex rabbit fur, mink fur, beaver fur, otter fur, raccoon fur, mouse fur, etc. Mink fur has the reputation of "soft gold". Natural fur is composed of the epidermis and its surface densely covered with needle hair, fluff, and coarse hair. However, due to different animal species, the proportion of fur composition is different, so the quality of fur varies. Fur, as a high-grade fabric in clothing, costs tens of thousands or even hundreds of thousands. Driven by profit and the scarcity of fur, there are fakes and shoddy adulteration in the market. It is very necessary to have a unique fur type identification method. [0003] Cu...

AI Technical Summary

Problems solved by technology

However, there are many defects in these two types of methods: (1) the identification time is long; (2) the identification person needs specialized technical knowledge; (3) the process is cumbersome; (4) the identification cost is high, etc.

Liu Xinru used near-infrared spectral analysis and the discriminant method of principal component-Mahalanobis distance to detect camel hair in "Visible / Near-Infrared Diffuse Reflectance Spectral Analysis Technology to Detect Animal Hair Fiber" (Gansu Agricultural University, Master's Thesis, 2013). It can be distinguished from cashmere, homogeneous sheep wool and cashmere, but this method is classified according to the three-dimensional graph of the principal component score. When the principal components of the two samples are superimposed and displayed in the three-dimensional graph, the method is invalid; Wu Guifang et al. in "Based on Principal Component Analysis and Support Vector Machine Cashmere Variety Identification Analysis" (Spectroscopy and Spectral Analysis, June 2009) uses a class of support vector machine methods to classify the remaining classes, and uses the score of the principal component of the sample as The training set of the support vector machine algorithm. There are three problems in this method. One is that it only verifies the effectiveness of identifying cashmere raw material varieties. The other is that ordinary support vector machines cannot effectively mine the intrinsic information of data samples during the learning process. It is a kind of training method for residual classes that takes a long time and will cause "data set skew" results

Images