Classification Of Remote Sensing Images With Multiple Classifier Ensembles

This paper summarizes and analyzes remote sensing image classification technology both at home and abroad,and finds that the current classification accuracy of remote sensing images is an important precondition for application of land use change,environmental monitoring and disaster monitoring.Based on the summary of traditional classification methods,current remote sensing classification methods(1)The traditional classification of remote sensing images is relatively mature and widely used,but different classification accuracy of different classifiers,image classification accuracy and classification of data preprocessing,sample selection,Band selection,classification algorithm and many other factors are closely related to many factors,classification accuracy has great uncertainty,and consider the classification of the whole process to improve the classification accuracy of the mature technology has not yet appeared.(2)Artificial intelligence method and machine learning technology can solve different sample selection,different classification methods of base classifiers,subjectivity of samples,uncertainty of results and fuzziness,etc.,which can greatly improve the classification Precision,machine learning technology will become the hot research field of remote sensing.In order to solve the problem that the accuracy of the traditional remote sensing image classification method is limited,different classifiers have great differences in the classification precision of the same classification task,and there is a certain subjectivity,lack of uniformity of the classification results and even different classification results of the classifier.This paper attempts to introduce integrated learning into remote sensing Image classification,the use of voting methods,the right of evidence,random forest three methods to achieve multi-classifier classification,classifier combination and accuracy evaluation,through the classifier combination theory to explore the classifier combination software implementation method using ArcGIS ModelBuilder tool under the platform integrates voting class and evidence theory classifier into a set of remote sensing image classification tools through ENVI and ARCGIS integrated development technology to achieve the convenience of remote sensing image classification and model sharing,The At the same time,based on the realization of the heterogeneous classifier combination,the open source software EnMAP-Box is applied to explore the homogeneous classifier combination and precision evaluation technology.This paper takes the classification of land use in Xining City of Qinghai Province as an example to classify the classification of land use in the Qinghai Lake Basin as an example to obtain the following conclusions:(1)Voting method and evidence DS combination theory as multi-classification The two methods of image integration can improve the image classification and evaluate the accuracy,which has a good application prospect.(2)EnMP-Box-based random forest classification can effectively integrate the homogeneous decision tree classifier to achieve the accuracy evaluation,which can be used not only in the field of hyperspectral but also in multi-spectral image classification.(3)At the same time,due to the limited research of the classifier's discrepancy measurement,the paper only does a limited research on the uncertainty of the system integration,the classifier selection of the multi-classifier combination,the number of classifiers combined,the combination,After the combination of performance optimization needs further study.