An object-based classification approach for high-spatial resolution Imagery

With the recent availability of commercial high resolution remote sensing multispectral imagery from sensors such as IKONOS and QuickBird, we can’t get the accuracy of land-cover classification expected using pixel-based approach. In this paper, we bring about object-based approach combined with the nearest neighbor to classify the QuickBird image of LianYungang city. Firstly, the image is segmented into object feature, we make the shape feature and contextual relation feature join the new feature space which is used to classify. And then we compare the classification of object-based approach accuracy with the nearest neighbor method of classification result, we can draw a conclusion that the method of classification in this paper can recognize geo-types much better. And the overall accuracy is 92.19%; the coefficient of Kappa is 0.8835. Salt and pepper effect is decreased effectively. The result indicates that the approach of land-cover classification combined object features with the nearest neighbor approach supplies another new technique for interpreting high resolution remote sensed imagery.