Research On The Method Of Extracting And Identifying Three-Dimensional Fluorescence Spectral Features Of Typical Red Tide Algae

With the continuous development of the marine economy,the degree of eutrophication of water bodies is becoming more and more serious.As one of the major marine disasters,toxic and harmful red tide hinders economic development,destroys ecological stability and even affects human life and health.Therefore,it is of great significance and application prospect to realize the early prediction of toxicity category and category of red tide algae.This paper investigates the current detection methods of red tide algae,aiming at the problems existing in the traditional detection methods of red tide algae.Combined with spectral analysis and machine learning algorithm,the three-dimensional fluorescence spectra of typical species of red tide algae were measured,and the spectrum library of red tide algae was established.On this basis,the identification model of toxicity categories and categories of living red tide algae was constructed.The main content of the paper is as follows.Firstly,the generation and application principle of 3D fluorescence spectrum is briefly described,the method of realizing 3D fluorescence data management of red tide algae by using database is introduced,the machine learning algorithm of 3D fluorescence data analysis of red tide algae is analysed,and the theoretical feasibility of this study is demonstrated.Secondly,the structure of the spectral library of red tide algae was studied.My SQL database and pyqt toolbox are used to design user information,algae inventory and data management module.The user information module is used for login and registration.The algae species list module is used to retrieve and add information about red tide algae.The data management module is used to retrieve and store 3D fluorescence spectral data of red tide algae.This database can facilitate more scientists to use 3D fluorescence data of red tide algae.Thirdly,the toxicity classification of red tide algae was investigated.Regularization pre-treatment method and principal component analysis were used to achieve feature extraction of spectral data.Decision tree analysis method was used to construct toxicity identification models of single species and mixed red tide algae.The integrated algorithm of gradient lifting decision tree was used to optimize the models and verify the reproducibility.Finally,the identification method of red tide algae was investigated.The three-dimensional fluorescence spectra of Dinoflagellata,Diatomata,Chrysoflagellata,Cryptophyta and Needellaria were plotted using a parallel factor analysis algorithm.The two-dimensional correlation coefficients between samples and standard spectra were calculated to realize the identification of single species and dominant species of mixed red tide algae and to verify the reproducibility.