Change Detection In High-Resolution Remote Sensing Images Based On Multiple Classifier Ensemble

The wealth of detailed information in high- resolution remote sensing images makes it possible to detect changes in a finer scale for Land Use/Land Cover(LU/LC). Change detection has been become a hot topic in the fields of high-resolution remote sensing. There are existing two kinds of change detection methods(supervised methods and unsupervised methods) for high-resolution remote sensing images, and also they have their merits and disadvantages as well as sphere of application. The combination of the two methods can improve the accuracy of change detection and the advancetage of the algorithm. Ensemble learning can comprehensively utilize the advantages of multiple classifiers to improve the generalization and classification accuracy. Therefore, it provides a novel research direction for supervised change detection.In order to improve the accuracy of change detection in high-resolution remote sensing images, two multiple classifier ensemble systems are proposed to detect the change information. The main nts are as followed:(1) The methods not only extract the texture features, morphological features, spectral information etc., but also utilize the spatial information to impose constrains on the initial change map which effectively decrease the “salt-and-pepper noise”.(2) In order to increase the level of automation, a method of training sample selection based on unsupervised change detection is proposed. This method uses two groups of thresholds instead of one threshold to enhance the accaracy of the selected training samples.(3) Two multiple classifier systems are constructed to detect the changes from different high-resolution remote sensing images. Based on the features of highresolution remote sensing images, three classifiers are chosen as the base classifiers. Different ensemble strategies are proposed to improve the performance of multiple classifier system.