Research On Rapid Determination Of Organic Matter Concentration In Aquaculture Water Using Multi-source Spectral Data Fusion

Aquaculture has become one of the fastest growing food production industry, which has made a great contribution for the protection of the food supply and economic growth. However, the water pollution in aquaculture has caused great impact on the socio-economic and food security and hampered the sustainable development of aquaculture seriously. Chemical oxygen demand (COD) is an important indicator to measure the degree of organic pollution in water bodies, the conventional determination methods need long analysis time, consume reagents, and have secondary pollution problems.In recent years, the use of ultraviolet spectrum and near-infrared spectrum technology for water quality determination have been widely studied, because spectrum technology goes fast, low-cost, non-destructive. It has been widely used in food, medicine, chemical industry and environment monitoring and other fields.In this study, water quality analysis in aquaculture is the research object, the application of ultraviolet and near-infrared spectrum combined with chemometrics analysis methods and data fusion technology were used to establish a rapid determination method of COD in aquaculture, provide a theoretical and technical basis for aquaculture real-time monitoring. In the current situation of serious water pollution in aquaculture, the state requires to achieve healthy aquaculture, the research on the rapid COD determination in aquaculture water has very important significance. Meanwhile, it provides research ideas for the rapid determination of other parameters of water quality in aquaculture. The main conclusions are as follows:(1)The rapid determination of COD based on ultraviolet spectral data was achieved, presented the original spectral data preprocessing-effective wavelengths selection-establish linear and nonlinear spectral prediction model, established full-band ultraviolet spectrum COD prediction models in aquaculture water based on partial least squares(PLS), least squares support vector machine(LS-SVM) and back propagation artificial neural network(BP-ANN); successive projection algorithm(SPA) and uninformative variable elimination(UVE) algorithm were used to select effective wavelengths of COD in aquaculture water, established prediction model based on effective wavelengths of COD in aquaculture using PLS, multiple linear regression(MLR), LS-SVM and BP-ANN. The optimal model for COD prediction is SPA-LS-SVM(SNV), in the prediction set of sample, the coefficient of determination R2was0.89, root mean square error of prediction(RMSEP) is15.21mg/L.(2)The determination of COD in aquaculture water using near-infrared spectrum was achieved, established COD prediction models based on PLS, LS-SVM and BP-ANN. Nonlinear modeling LS-SVM and BP-ANN got better prediction results than linear modeling PLS approach for using near-infrared for COD determination. SPA and UVE algorithm were used to select effective wavelengths for COD determination in aquaculture, established prediction models based on effective wavelengths using PLS, MLR, LS-SVM and BP-ANN combined with near-infrared spectrum. The optimal model for COD prediction is SPA-BP-ANN(MSC), in the prediction set of sample, R2was0.87, RMSEP is16.01mg/L.(3)The rapid determination of COD in aquaculture based on a multi-source of low-level data fusion (LLDF) and the middle-level data fusion (MLDF) data fusion combined with ultraviolet spectrum and near-infrared spectrum was achieved, prediction results of COD based on LLDF fusion of multi-source spectral data and PLS model is worse than that of model using ultraviolet spectrum and near-infrared spectrum respectively; prediction results of COD based on LLDF fusion of multi-source spectral data and LS-SVM model is better than that of model using ultraviolet spectrum and near-infrared spectrum respectively; prediction results of COD based MLDF fusion of multi-source spectral data and PLS, LS-SVM and BP-ANN model is better than that of model using ultraviolet spectrum and near-infrared spectrum respectively, the best prediction model was based on LS-SVM and SPA combined with MLDF fusion of multi-source spectral data, in the prediction set of sample, R2was0.91, RMSEP is13.58mg/L.The above results achieved the fast and precise determination of COD in aquaculture water quality. They also provided theoretical basis for sensors development for aquaculture water quality analysis, which has a promising application prospect.