Research On Detection Method Of Multi-parameter Synchronous Inversion Of Surface Water Based On Hyperspectral Data

A large part of human’s living water comes from surface water.Nowadays,due to the joint action of natural environment and human activities,surface water pollution has become a major problem in the world.Surface water resource distribution is very unbalanced in our country and water resource possession per capita is seriously deficient.With the growth of population,the vigorous development of industry and agriculture,and the rapid advancement of urbanization,a large amount of sewage is discharged into rivers and lakes without standardized treatment,which makes the problem of surface water pollution increasingly serious,greatly affecting the industrial and fishery production and people’s livelihood security,and is not conducive to the sustainable development of ecological system.Traditional water quality testing mainly adopts chemical analysis method,which has defects such as cumbersome operation,high environmental requirements such as temperature and pressure,the need to add one or more reagents,waiting time for reaction,insufficient reaction,etc.Improper operation will cause harm to the body of researchers.Hyperspectral data has a relatively high resolution and can provide more abundant and continuous spectral data information,which plays an important role in the detection of water quality parameter concentration and evaluation of water quality.However,most of the current researches collect hyperspectral data through ground object spectrometer or airborne hyperspectral imaging spectroscopy system,which limits the spatial location of water quality detection,and few researches use hyperspectral data for water quality detection indoors.In order to expand the application range of hyperspectral water quality analysis technology and make technical reserve for the development of hyperspectral water quality analysis equipment in laboratory environment,this paper takes 124 surface water samples from 23 stations in Jilin Province as research objects and takes physicochemical analysis results as reference to collect hyperspectral data of water samples in laboratory.Inversion models for chemical oxygen demand（COD）,total phosphorus（TP）,total nitrogen（TN）and ammonia nitrogen（NH₃-N）were constructed.The specific research contents are as follows:（1）Investigate the detection methods of chemical oxygen demand,total phosphorus,total nitrogen and ammonia nitrogen commonly used at home and abroad,and the research progress of hyperspectral imaging technology in the field of water quality detection in recent years.（2）The water samples used in the experiment were sorted out,hyperspectral images were taken and hyperspectral data were extracted,abnormal sample point data were removed,and the reflectivity data was preprocessed by derivative method.The remaining 121 samples were reasonably divided into 91 training set data and 30 verification set data,so that the ratio was about 3:1.（3）Pearson correlation analysis method was used to analyze the correlation between spectral reflectance of data samples and corresponding physicochemical analysis results.For each water quality parameter,bands with absolute correlation coefficients greater than 0.3 in the first derivative spectrum and second derivative spectrum were selected as characteristic spectra,and the corresponding spectral reflectance data were taken as derivative characteristic spectral data.（4）Study the basic principle of least square support vector machine,select particle swarm optimization algorithm to optimize the penalty factor C and the width of radial basis kernel function（?）²,and set appropriate parameter values for the algorithm.Three data indexes,namely determination coefficient R²,root mean square error RMSE and relative percent deviation RPD,were selected as the evaluation criteria of the model.（5）PSO-LSSVM regression algorithm was used to establish inversion models based on five kinds of hyperspectral data including original spectrum,first derivative full spectrum,second derivative full spectrum,first derivative characteristic spectrum and second derivative characteristic spectrum for four water quality parameters.The predicted values were compared with the measured values of physical and chemical analysis.The results show that the prediction effect of first derivative characteristic spectrum modeling is good for COD.For TP,both the model established by the first derivative spectral data and the model established by the second derivative spectral data have merits.For TN,the prediction results are good when the second derivative characteristic spectrum is used for modeling.For NH₃-N,the first derivative characteristic spectrum modeling results are good.At the same time,the prediction results of the four parameters show that the prediction ability of the model using the original spectral data is poor,while the prediction accuracy of the model can be improved by using the derivative spectral data.Moreover,the model based on extracting characteristic spectrum has better prediction ability than that based on full spectrum data.In summary,the principle and process of constructing COD,TP,TN and NH₃-N models with hyperspectral data are discussed in detail in this paper.By comparing and analyzing the prediction results,it is proved that modeling with derivative spectral data and characteristic spectral data can further improve the prediction accuracy of the models,and the feasibility of conducting hyperspectral water quality detection in the laboratory is preliminatively demonstrated.It provides a new method and idea for surface water quality multi-parameter detection.