Optimization Research Of Random Forest Classification Based On Landsat-8 OLI Remote Sensing Images

Remote sensing images classification plays an important role in land structure planning and land resources exploration.Random Forest classification has good robustness,high stability,and wide range of applicability.In recent years,it has attracted more and more attention.How to improve the classification accuracy of Random Forest and optimize the overall Random Forest classification process has become a research hotspot.In response to this hot issue,this paper conducts Random Forest classification optimization research from three aspects: classification feature set optimization,classification sample selection optimization,and classifier parameter s optimization.The specific research work is as follows:Firstly,based on Landsat-8 OLI remote sensing images,taking Tianjin urban area as the experimental area,and fully considering the advantages of the filter feature selection method and the wrapper feature selection method,this paper proposes Random Forest-Particle Swarm Optimization(RF-PSO)combined feature optimization method,it is used to optimize feature set.Secondly,the influencing factors affecting the classification accuracy of Random Forest are studied,including the optimization of sample size and the optimization of classifier parameters.This paper uses classification accuracy as evaluation index to select optimal parameters.Finally,the optimized pixel-based Random Forest classification and object-oriented Random Forest classification experiments are carried out,and the Random Forest classification is experimentally compared with other classificatio n methods.The experimental results show that the RF-PSO combined feature optimization method proposed in this study improves the classification accuracy;the optimization results of classification samples and the classifier parameters show that when the number of samples is 30 times the number of bands,the number of Random Forest training trees is 300,the minimum node sample is 2,and the minimum impurity is 0.3,the Random Forest classification has the best classification accuracy.The overall classification accuracy reaches 80.45% and the Kappa coefficient is 0.8310,which is higher than other classification methods such as maximum likelihood and neural network.The optimization of the Random Forest classification process has a significant effect on the improvement of remote sensing images classification accuracy.