Research On Land Use/Land Cover Classification Based On Machine Learning In The Huangshui River Basin

Aiming at the characteristics of varied and complex geomorphic types,crisscross network of ravines and broken terrain in high altitude complicated terrain regions,it is very important to study and find the rapid and effective land use / land cover classification method for obtaining and timely updating of land use information.Taking the Huangshui river basin located in the transitional zone between the Loess Plateau and the Qinghai-Tibet Plateau as a case study area,the objective of this study is to explore a kind of effective information extraction method from comparison of four kinds machine learning methods for complicated terrain regions.Based on Landsat 8 OLI satellite data,DEM and combined with various thematic features,on the basis of geographical division of the study area,artificial neural network,decision tree,support vector machine and random forest four machine learning methods for land use information extraction were used to obtain land use data,and confusion matrix was constructed to evaluate classification accuracy.The results showed that the classification accuracies of random forest and decision tree are obviously higher than those of support vector machine and artificial neural network.The random forest method has the highest classification accuracy,the overall classification accuracy is 85.65%,the Kappa coefficient is 0.84.In the classification efficiency,the running speed of the random forest and decision tree is higher than that of the support vector machine and the artificial neural network.However,the decision tree classification method is time-consuming and time-consuming in the formulation of classification rules,so the random forest method has more advantages.Comprehensive classification accuracy and classification efficiency of two aspects,the random forest method is more suitable for complex terrain land use / land cover classification.In addition,the overall classification accuracy and Kappa coefficient of the random forest method in the Naoshan area were 87.27% and 0.84,the overall classification accuracy and Kappa coefficient of the random forest method in the Qianshan area were 85.94% and 0.83,the overall classification accuracy and Kappa coefficient of the random forest method in the Chuanshui area were 84.58% and 0.82 respectively,all achieved better classification results.It is proved that the geographical division of the study area has a positive effect on the classification of remote sensing images in complex terrain area.And comparing the classification accuracy of different geographical subareas under the same method,the trend of the areas of Naoshan> Qianshan> Chuanshui is presented.According to the distribution characteristics of land use / land cover types in each geographical subarea,the complexity of the study area is affected Classification accuracy of the important factors.In the study,NDVI,MNDWI,NDBI,DEM,Aspect and other characteristic parameters were used in the study to select and apply to the different geographical subareas to obtain better classification results.In the 4 kinds of classification methods in machine learning,both in the random forest classification accuracy and performance in classification speed is excellent,compared with other methods shows that the random forest has more advantages in processing multidimensional data,embodies the characteristics of random forests with high precision and high efficiency and high stability etc..Based on the above classification,Random forest classification method was chose to further classify Landsat 8 fusion data from panchromatic 15 meter and multispectral 30 meter image,the overall classification accuracy is 86.49% and the Kappa coefficient is 0.85.The overall classification accuracies of Naoshan,Qianshan and Chuanshui areas increased by 0.46%,0.94% and 0.96%,respectively,compared with those of non-fusion images,it shows that the fused image with higher resolution is more excellent in the detail expression.This advantage is more prominent in the regions with more complicated land use / land cover types.The above shows that it is an effective way to improve the classification accuracy of land use / land cover classification in complex topography by band blending of Landsat 8 remote sensing data.Based on the above classification,it is found that the method of random forest can be used to obtain the better classification results in the remote sensing image classification of the Huangshui River Basin by means of integrated data fusion,geographical division and selection of appropriate characteristic parameters.