APPLICATION OF BOOSTING TO IMPROVE IMAGE IMAGE CLASSIFICATION ACCURACY IN RICE PARCEL WITH DECISION TREE

摘要

Rice cultivation is a very important part of agriculture in Taiwan. The Taiwanese government employs satellites image to detect rice parcels, as it is a fast and economical method to extract rice parcels. Besides, the development of high-spatial and spectral resolution satellites images, facilitates enhanced results while extracting rice parcels. Thus we employ high-spatial and spectral resolution QuickBird satellite image to classify the rice parcels. However, it is common knowledge that supervised classification uses the training samples to stand for the true things to be classified. When performing image-classification, training samples selected based on the researcher's knowledge are employed and these are seldom properly examined. Invariably, the researcher's subjective knowledge leads to the inclusion of wrong information. Hitherto, this problem has not been satisfactorily addressed. Therefore, apply a new algorithm called "Boosting" to examine the training samples and enhance the classification accuracy. It is a statistics method and can give wrong samples a heavy weight and reduce the right samples weight. Through this method, we can get suitable training samples to correct them to the normal distribution. In the end, we combine one of the decision tree, CART (Classification and Regression Tree), with boosting to illustrate the results. This new method examines the training sample objectively and successfully enhances the accuracy.