Inversion Of Nitrogen In Small Scale Water Bodies Based On UAV Hyperspectral

The study of nitrogen content in water has always been the focus of social attention.As the "capillaries" of rivers and lakes,small scale water bodies have been polluted to some degree under the influence of social development and human activities in recent years.So control of nitrogen pollution has become an urgent problem to be solved.Monitoring is prior to control.The traditional nitrogen monitoring method is to collect water samples from the field and analyze them in the laboratory,which has the disadvantages of high cost,low monitoring frequency and secondary pollution.Hyperspectral technology realizes fast,pollution-free,large-scale and real-time nitrogen monitoring,which is an important development direction of nitrogen monitoring in the future.Based on this,after constructing the spectral research method,this paper takes three distinct areas of Yueyang River as the research object,and three times of UAV hyperspectral image and on-site water sample are collected in the study area.Next,the spectral reflectance is extracted in preprocessed image and calculate by single band and dual band,then the sensitive band and band combination related to nitrogen index are explored by pearson correlation analysis.Finally the univariate regression model,lasso regression model and BP neural network model are established to verify the best model and complete the three regional nitrogen inversion.The results show that:(1)The correlation coefficients of the three nitrogen indexes(total nitrogen,ammonia nitrogen and nitrate nitrogen)after dual band processing are higher than single band processing.Under single band processing,the highest correlation coefficient of nitrogen index in light sewage water is about 0.4,and that in heavy sewage water is about 0.6 except for total nitrogen.Under dual band processing,the highest correlation coefficient of nitrogen index in light sewage water is about 0.5 and that in heavy sewage water is about 0.7.Dual band processing can better reflect the nitrogen-related information in the spectrum,and the nitrogen related response in the spectrum is directly proportional to the nitrogen concentration.(2)In the regression model established by hyperspectral data and nitrogen index,the accuracy of univariate regression model is insufficient,and the fitting degree of BP neural network is the highest.In the univariate regression model,the determination coefficient of light pollution water is very low,and that of heavy pollution water is less than 0.5.According to the regression model established by lasso regression after variable screening,the accuracy of the model of ammonia nitrogen and nitrate nitrogen in heavy sewage water is higher,and the determination coefficient is more than 0.70.The BP neural network model applies the characteristic variables screened by lasso regression as the input layer and the nitrogen index concentration as the output layer,which can well fit the three nitrogen indexes,and the degree of fit is between0.84 and 0.92.(3)Except for the nitrate nitrogen index of heavily polluted water,the performance of BP neural network model is better than lasso regression.In the inversion model of total nitrogen concentration,the average relative error of BP neural network model of light pollution water and heavy pollution water is 8.6% and 5.2% respectively,which is about 10% lower than lasso regression model.In the inversion model of ammonia nitrogen concentration,the average relative error of the BP neural network model is 10.4%,which is about 15% less than the lasso regression model;in the heavy sewage body,the accuracy of the two models is the same,and the average relative error is 7.8% and 8.7%,respectively.In the inversion model of nitrate nitrogen concentration,the average relative error of BP neural network model is 10.8%,which is about 10% less than lasso regression model;in heavy sewage,the prediction of lasso regression model is better than BP neural network model,and the average relative error is8.6%,reducing 0.6%.(4)Combined with the best inversion model and the preprocessed UAV hyperspectral image,the nitrogen index concentration distribution maps of three sampling regions are obtained,and the nitrogen distribution is basically consistent with the actual situation.In area 1,two sites in the upper and middle reaches are affected by human activities,resulting in higher nitrogen concentrations than those in the surrounding areas,and higher nitrogen concentrations causes by agricultural activities in the lower reaches.In area 2,artificial floating islands and riparian ecological plants can purify water quality,but plant residues may cause secondary pollution.Because area 3 is located in the lower reaches of the river,the nitrogen concentration is higher than area 1 and area 2.At the same time,the water body of the tributaries will slightly increase the nitrogen concentration.