نویسندگان
1 استاد دانشکده مهندسی برق و کامپیوتر، دانشگاه تربیت مدرس، دکترای مخابرات
2 دانشجوی دکتری مهندسی برق، دانشگاه علوم و تحقیقات تهران
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
By increasing spatial resolution of remote sensing images, it is possible to apply spatial information in the classification. This leads to improve the accuracy of multi-spectral images classification. One of the methods for incorporation spatial information is un-supervised segmentation which can be implemented through Expectation Maximization(EM) clustering and connected-component labeling. But this clustering method which always is trapped in local optimum. Therefore, a new algorithm is proposed that can solve the mentioned problem and has better performance in multi-spectral images classification. In order to form a spatial-spectral classifier, first a pixel-wised classifier is applied. Then, after using the LPP approach for dimension reduction, the results of pixel -wised classifier and segmentation map, obtained from proposed method, are combined via majority voting. The results of simulation indicate that the presented spectral-spatial classifier leads to so considerable improvement that the accuracy and validity of classification have reached to 88.68% and 80.8%, respectively.
کلیدواژهها [English]
)