Object-oriented information extraction of high resolution imagery based on multi-classifier fuzzy integration

While high spatial resolution imagery provides more detailed information on ground objects, it increases the intra-class spectral variability. Thus the traditional pixel-based classification approaches are no longer applicable. Aiming at the characteristics of high resolution remote sensing imagery, this paper proposes an object-oriented fuzzy classification method based on multi-classifier integration. On the basis of multiresolution segmentation and the establishment of hierarchical network of image objects, this method first divides the image characteristics into the essential ones and the nonessential ones and obtains the feature descriptions of image objects in different feature spaces using the nearest neighbour classifier and the classifier of membership function. Next, it conducts fuzzification to nearest neighbour classifier by employing the exponential function. Then, it performs fuzzy integration to different recognition results using fuzzy logic. Finally, by defuzzification, it realizes the effective separation to ground objects located in the complex scene. Experimental results prove the effectiveness of this method.