Study On The Identification And Classification Of Animal Blood Fluorescence Spectra

In recent years,animal blood identification has many practical needs in the fields of import and export,food safety and health.It is urgent for a fast and accurate method that can adapt to large-scale animal blood recognition.At present,the existing research results of spectral technology in blood detection and recognition mainly focus on the spectral analysis of human blood or animal blood.There is relatively little research on the recognition of blood spectrum between different species.This paper combines the statistical machine learning method to study the identification and classification of the blood fluorescence spectra of 4 kinds of animals,pigeons,chickens,mice and sheep.The main research work includes the following aspects:1.By means of statistical methods,we set a statistical description on the 4 kinds of samples,including concentration,dispersion and distribution of data,we use ShapiroWilk test method and normal method of correlation coefficient of fluorescence spectra intensity were analyzed for normality test and correlation,At the same time,we use wavelet soft threshold denoising method for spectral intensity characteristics the noise to different decomposition levels of inquiry.2.In view of the blood fluorescence spectra of 4 kinds of animals,the recognition method based on support vector machine is discussed.In the process of feature extraction,feature extraction is done by combining distinguishing statistics,principal component analysis and MIV algorithm.In the process of recognition and classification,parameters are optimized by combining support vector machine multiple classifier and particle swarm optimization algorithm.Finally,the cross validation of different wavelet threshold denoising layers is discussed.3.An animal blood fluorescence spectrum recognition method based on deep belief network is discussed for 4 kinds of animal blood fluorescence spectra.First of all,the original spectral intensity features are used as network input features,and the network parameters are initialized.And then,the optimal number of network layers is determined by cross-validation method,the iteration number is adjusted by model reconstruction error.Finally,the method is compared with the recognition method based on support vector machine,and the effect of wavelet threshold denoising decomposition layers on the recognition performance of the model is investigated.