| Biological tissue is the basic component of all life structures.It has a wide variety and complex internal components.The effective identification of different kinds of biological tissue is of great significance for human disease diagnosis and the safe intake of animal derived food.Due to the similarity and difference of element components in different kinds of biological tissues,how to extract element characteristics from the same kind of tissues for micro difference recognition is the main research direction at present.Because the conventional identification methods have the problems of long time-consuming,large sample damage,high experience dependence and low quantification,a new rapid detection method is urgently needed.Laser induced breakdown spectroscopy is a spectral technology based on element detection.It has the advantages of fast,non-contact,simultaneous detection of multiple elements and no sample pretreatment.In the LIBS detection process,a laser emits a pulse to the sample surface to generate plasma.The plasma spectral information is collected by the spectrometer,and the element content information in the sample is characterized by spectral data.In this paper,LIBS technology is used to detect different biological tissues to achieve the purpose of rapid identification of different kinds of biological tissues.Because the surface of biological tissue sample is uneven,cracked and soft at room temperature,LIBS experiment needs to accurately control the position of laser focus relative to the sample surface to realize the accurate control of laser pulse direction and laser focusing.To solve this problem,this paper introduces the stepper motor three-dimensional motion platform control system,and proposes a laser focusing system based on auxiliary laser points.On the basis of completing the system construction,according to the characteristics of complexity,specificity and high dimension of biological tissue spectral data,combined with machine learning method,Extract useful information from complex spectral data and establish a more robust analysis model.Optimize and improve the algorithm to improve its analysis performance.The main research contents of this paper include the following aspects:1.According to the requirements of biological tissue sample observation and precise control of sample position,a LIBS detection platform suitable for biological tissue detection is built.The experimental parameters of each device are optimized according to the plasma temporal and spatial characteristics of biological tissue samples,and the operation timing of the experimental instrument is accurately matched combined with the timing control system.At the same time,the selection and construction of three-dimensional motion platform(image acquisition module and laser focusing system,etc.)are also carried out,focusing on the analysis of the size of protrusions,depressions,cracks and fat areas on the surface of biological tissue samples,and the precise control of laser pulse path and the optimization of defocus amount of focusing system are carried out according to its morphological characteristics and combined with the stepping motor translation table,Finally,the relative standard deviation of characteristic spectrum and signal-to-noise ratio of biological tissue samples are calculated to evaluate the spectral repetition rate and signal quality.2.In order to evaluate the generalization of LIBS detection platform for biological tissue samples,lung adjacent tissues,tumor tissues and normal tissues were selected as connective tissue samples;Beef tissue,pork tissue and mutton tissue were tested as muscle tissue samples.According to the characteristics of baseline drift,high noise and different scales of biological tissue spectral data,the front-end processing of biological tissue spectral data is carried out through spectral baseline deduction,noise reduction and standardization,and the feature vector of biological tissue spectral data is extracted through spectral line recognition and principal component analysis.In view of the problems of unclear clustering and overlapping information in the main scattered point maps of adjacent tissue normal tissue and beef pork mutton tissue samples,the random forest algorithm is selected to calculate the corresponding spectral data feature vector weight.The spectral lines of muscle tissue spectral data greater than the average value of spectral line weight are KI766.5nm、Ca II393.4nm、Na I588.59nm、Ca II396.9nm is used as the input of the model,and the characteristic weights of the spectral lines are 0.45,0.17,0.15 and 0.12 respectively.KI766.5nm、Na I589.59nm、Ca II393.4nm were used as model inputs in the spectral data of adjacent tissues normal tissues,and the weights of characteristic spectral lines were 0.35,0.28 and 0.16 respectively.So as to effectively extract and fully mine the spectral data information of biological tissue.3.Select Naive Bayesian,Bayesian Network and BPNN models to gradually optimize the recognition accuracy of muscle tissue spectral data,and get the best recognition effect of BP neural network model.The recognition accuracy of beef,mutton and Pork Tissue spectral data has reached 99.8%,99.5% and 99% respectively.The overall accuracy,kappa coefficient and root mean square error of the model are better than naive Bayesian and Bayesian network models.4.SVM,LSSVM and PSO-LSSVM are selected to identify the spectral data of tumor normal tissue and adjacent normal tissue,and the LSSVM after particle swarm optimization is the best.The recognition accuracy of tumor normal tissue spectral data is99.6%,and that of adjacent normal tissue spectral data is 98.9%.The overall accuracy,kappa coefficient and root mean square error of the model are better than SVM and LSSVM models.This paper realizes the high-precision identification of biological tissue through LIBS technology,which indicates that LIBS technology has certain value in the field of biological tissue research,and is of great significance for biological tissue detection in rapid identification and real-time online analysis. |
PDF Full Text Request