Hyperspectral remote sensing image classification based on random forest

Abstract

With the development of remote sensing technology and machine learning, the research on hyperspectral remote sensing image classification has also progressed rapidly. In this paper, based on the random forest model, a new classification model of hyperspectral remote sensing images is proposed, which can effectively classify the spectral information and texture information of hyperspectral images. First, the texture features are extracted from the hyperspectral remote sensing image data and superimposed on the original spectral domain data, thus forming a new spectral-space domain data. Then, the model forms an optimal combination of parameters by selecting parameters such as the number of decision trees, maximum depth and minimum number of leaves, and this model has a higher classification accuracy for the dataset than the original random forest model. Experimental results on Indian, KSC and Salinas hyperspectral images show that the method proposed in this paper can solve the problems of hyperspectral data nonlinearity and information redundancy due to the expansion of data and the selection of parameter combinations, thus improving the classification effect of the original random forest and the average accuracy of the classification.