Remote Sensing Inversion of TN Concentration in Xiaolangdi Reservoir Optimized Neural Network by Conjugate Gradient Method

To popularize remote sensing monitoring technology of water quality, by measured data of total nitrogen (TN) from three sections in Xiaolangdi Reservoir from 2011 to 2015, together TM satellite data with the difference date in 48h to TN, eighteen groups of TN concentration data and six sets TM images were selected. The reflectance values of the first four bands were extracted after the images were processed. By WEKA software, fifteen band combination of TM images closely related to TN were extracted from 66 band combinations as the input vector, and the measured concentration of TN was the output vector, to create BP neural network. The model convergence speed was quickly after conjugate gradient method was adopted to train the network, the average relative error of prediction was 4.771 percent and less than 5 percent, the accuracy showed strong nonlinear processing ability for neural network due to the image data discrete in time and space. The research results will provide basis for water quality remote sensing monitoring in the reservoir in the future.