Hyperspectral Inversion Of Soil Organic Matter Based On Improved Particle Swarm Optimization Neural Network

As one of the important indexes of soil fertility evaluation,soil organic matter plays an important role in agricultural productivity evaluation and agricultural land planning.Traditional soil nutrient detection method is time-consuming and laborious.It is important to obtain the soil organic matter content data quickly and accurately and use it as the basis of soil fertility analysis and precise fertilization.With the development of hyperspectral technology,the rapid prediction of soil nutrient content by using hyperspectral data has become one of the research hotspots.This paper takes Qixing farm,Fujin city,Jiamusi city,Heilongjiang province as the research area,soil samples from a cultivated land collected in June 2019 were taken as research objects,soil spectral data were measured in the laboratory using a portable ground object hyperspectral meter,ASD FiledSpec4,the organic matter content obtained by laboratory chemical analysis was used as the reference value for hyperspectral prediction.The obtained hyperspectral data were preprocessed,spectral characteristics and organic matter content were comprehensively analyzed,and the characteristic band range of organic matter content was roughly determined.The characteristic bands of organic matter content were extracted by correlation coefficient method and multi-step method,a partial least squares regression model based on two feature band extraction methods and 14 spectral transforms was established,it is concluded that the correlation coefficient method can extract the characteristic band range of organic matter content to a certain extent,but the location of the extracted characteristic band is only the result of the single correlation analysis,which is insufficient to reflect the influence between the bands.There are certain limitations in determining the characteristic band solely depending on the size of the correlation and using it as an independent variable to establish the model.However,the model based on multiple step method and first order derivative of logarithm of reflectance has higher accuracy,which is the main data base of modeling.In view of the inadequacy of particle swarm optimization(pso)in balancing global search and local search in the whole iteration process,the inertia weight,one of the main parameters in the process of particle velocity and position update,is improved based on previous researches,an inertial weight adjustment strategy based on individual optimal fitness value is proposed,which increases the diversity of inertial weight,balances the ability of global search and local search of particle swarm optimization algorithm,and improves the optimization performance of the algorithm,The algorithm is combined with the three-layer BP neural network model to enhance the global search capability of the network,so that the prediction accuracy of the model is higher and the generalization ability is stronger.In this paper,a hyperspectral prediction model of soil organic matter was selected by using soil hyperspectral data,various feature extraction methods and modeling methods,and through the evaluation of model accuracy,which provided a new technical process for the rapid and accurate prediction of soil composition by hyperspectral data and the development of precision agriculture,and provided a new idea for the application and development of hyperspectral remote sensing.