The Study Of Remote Sensing Classification Method For Large Areas Land Cover Based On Landsat7ETM+Image

Extreme weather has been common occurrence in recently years, which causedgreat harm to countries all around the world. Global change and its influence havebeen attracting increasing attention of human. In order to deal effectively with globalchanging, international community and the relevant countries are dedicated to thestudy of the solutions to the series of scientific problems brought by it, which mainlyfocuses on analyzing objectively the basic facts and objective laws of global changingand predicting its future trends and influence. This paper uses Landsat7ETM+multi-spectral image data and selects an image with the line-column number isP089R082as representative, and the research is based on two points to improve theaccuracy of remote sensing image classification. The first one is to use the effectiveimage classification method such as Maximum Likelihood Classifier (MLC), decisiontree, Support Vector Machine (SVM) and neural network. According to the projectrequirements, the comparison analysis of MLC, SVM and C5.0algorithm baseddecision tree in the research shows that C5.0algorithm based decision tree hasstronger adaptability and higher accuracy, based on which this algorithm is chosen asthe method to produce the large areas land cover production. Secondly use multi-featured image data to add the data source beneficial to classification and to make fulluse of the information of spectrum, slope and various index (K-T transformation,NDVI). The test indicates that multi-featured merged data helps to improve theaccuracy of remote sensing image classification especially when the slope feature cangreatly eliminate the influence of hill shade. At the meantime, while choosing thesamples, GLOVACOVER classification data and related literature materials arecombined to make sure of the type of sample points and that Google Earth highresolution image inspects the sample in time in order to guarantee the quality ofclassification rules.In the process of generating large areas land cover production, a sampleacquisition strategy is proposed, in which, the image is classified by month and therepresentative samples of every image are united into regular samples, and inpost-classification processing the multi-scale segmentation vector of the original image by eCongnation software, is used as an assistant tool so as to promote theprocessing speed and the ultimate precision, and to realize the quick production oflarge areas land cover products.