Investigation On Classification Method And Key Algorithms Of Animal Organization Using Laser-induced Breakdown Spectroscopy

Animal tissues have so many categories that their classification is of great significance in reality.Traditional classification techniques of animal tissues are usually based on DNA and protein detection.However,sample processing and DNA or protein extraction require professional personnel operation,which are complex,time-consuming,thus not suitable for in situ detections.The laser-induced breakdown spectroscopy(LIBS)technique can meet the requirements of fast,real-time,online diagnosis by analyzing compositions of tissues coupled with chemometrics classification algorithms.In this thesis,taking the common meat tissues and humanas examples,LIBS technology was applied into the animal tissues classification.The detailed contents are as follows:Firstly,tissue sample preparation and collection methods were investigated.Using 4 common pork tissues,different pretreatment and spectral acquisition methods were applied to handle the meat samples,including paraffin embedding(fixed point acquisition),glass pressing(fixed point collection),glass pressing(scanning acquisition),and automatic slicing(scanning acquisition).Support vector machine(SVM)algorithm based on mapping kernel of radial basis function was used to classify the 4 kinds of tissues,and the effects of different pretreatment and spectral acquisition methods on the classification accuracies were compared.The results showed that the pretreatment method of glass pressing with scanning acquisition can obtain the best classifying results for tissues,in which the corresponding average predicted recognition rate is 93%.Secondly,different classification algorithms were compared to choose the best algorithm.The LIBS spectra of 6 kinds of meat tissues(including Pork,Beef,and Chicken)with the pretreatment of glass pressing were collected by the way of scanning acquisition.Different chemometrics classification algorithms,including supervised algorithms(SVM,Random forest,Partial least squares discriminant analysis,Linear discriminant analysis,Naive Bayesian classifier and Neural network,K-nearest neighbor)and unsupervised algorithms(Principal component analysis and K-means)were used to classify the 6 fresh meat tissues.The results show that the support vector machine algorithm can obtain the best results,and the average prediction recognition rate is 88.44%.Next,combination and optimization of algorithms based on SVM were studied.Using the 6 fresh meat tissues above,the SVM and principal component analysis were combined to classify the tissues.The modeling efficiency was improved from 94.63 s to 57.00 s,and the average prediction recognition rates of the SVM model was increased from 88.44% to 89.11% by reducing the dimensionality and noise of the principal components.Meanwhile,the recognition rate was improved to 90.22% by optimizing the kernel parameters of SVM.In order to solve artificial selection of spectral lines before the classification,the wavelet transform was used to automatically select the spectral lines by use of the threshold value of signal-noise ratio and signal-to-background ratio,thus making the classification algorithm more intelligent.Finally,the results were applied in the classification of medical tumor tissues.SVM coupled with the principal components and SVM were used to classify the normal tissues and the tumor tissues.The average prediction recognition rates were 94.44% and 93.06%,respectively.Moreover,the utilization of principal components makes the modeling time be reduced from 21.55 s to 1.97 s.In a word,LIBS combined with chemometrics classification algorithms can be applied in the classification of animal and medical tissues,which provided reference data for online tissues classification using LIBS.