Estimation model based on spectral-reflectance data

Abstract

A new estimation model based on BPNN is proposed in the thesis to recognize the characteristics of the spectral-reflectance data rapidly and estimate soil salt content accurately. Take the case of USGS Digital Spectral Library to estimate salt content of different soils, BP neural network model has been optimized, improved and realized in MATLAB. In this experiment, the prediction accuracy of multiple BPNN model reaches 83%, which is superior to traditional BP neural network model and multi-factor orthogonal regression analysis, indicating that the multiple BPNN model is suitable for the complex regression analysis of nonlinear spectral data and the high-precision estimation model of soil salt content. This method could also be used in embedded system, and design real-time testing equipment of soil salt content.