Research On Inversion Method Of Forestland Leaf Area Index Based On Remote Sensing Image

Leaf area index(LAI)is an important indicator to describe vegetation growth status.Timely and accurate estimation of LAI is of great significance to the construction of ecosystem and crop growth models,as well as the detection of agricultural and forestry environment.Compared with traditional methods of LAI acquisition,remote sensing technology has become the main method to obtain large-scale LAI due to its advantages of fast,accurate,large observation range,non-destructive and objective.This article selects part of Maoershan as the research area,selects domestic GF-1 multi-spectral satellite remote sensing image as the research data.The forestland LAI inversion method based on remote sensing image is divided into two processes,including remote sensing image classification and forestland LAI inversion.The specific research results are as follows:(1)Data sets were prepared,including pre-processing remote sensing image data,extraction of six vegetation cover indices,construction of PROSIAL model to obtain LAI values,and classification of remote sensing images of the research area according to spectral characteristics.(2)Chaos initialization and nonlinear convergence factor adjustment strategies are used to improve the swarm intelligence optimization algorithm.(3)The improved grey wolf optimization was used to search the optimal parameter combination of support vector machine.The SVM model and IGWO-SVM model were constructed to class the remote sensing images of research area.The experimental results show that the overall accuracy of SVM model is 98.48%,and the Kappa coefficient is 0.9652.The overall classification accuracy of IGWO-SVM model is 98.77%,and the Kappa coefficient is0.9719.SVM model has high classification accuracy,and IGWO algorithm can further optimize the performance of SVM model.(4)The improved spotted hyena optimizer was used to replace the weight and bias updating process of BP neural network.BP neural network and ISHO-BP neural network were constructed to invert the forestland LAI.The results show that the optimal structure of BP neural network is the network with three-layer and 10 neural.The root mean square error of BP neural network is 0.1392,the R~2of test set is 0.593.The root mean square error of ISHO-BP neural network is 0.1268,the R~2of test set is 0.79.ISHO algorithm can jump out of the local optimal solution and improve the inversion ability of BP neural network.