Research on Classification Method of Medium Resolution Remote Sensing Image Based on Machine Learning

Abstract

Surface information extraction is an important link in geographic situation monitoring and environmental protection and plays an important role in the global sustainable development strategy. In this paper, the resolution remote sensing image in Landsat8 OLI was selected as the main data source. Aiming at the problem of the lack of classified sample data, the global land cover data in 2015 and 2017 were optimized and treated as the prior knowledge of classification. The maximum likelihood method, Support Vector Machine (SVM) and Random Forest (RF) Machine learning methods, as well as deep learning methods based on VGGNET-16 and RESNET-18 models, were used to compare and study surface information extraction methods in the Yellow River Delta region. The results show that the above method is highly feasible. Based on the feature optimization, the overall classification accuracy of RF and SVM models in the machine learning method is high, and the classification accuracy of RF and SVM models is up to 87.3% and 86%. In the deep learning algorithm, the classification accuracy of VGGNET-16 and RESNET-18 models is greatly improved compared with the machine learning method. The classification accuracy of RESNET-18 is up to 94.1%, and the Kappa coefficient is 0.91. The research method in this paper has good applicability and popularization value in the classification of medium resolution remote sensing objects.