Abstract: Object-oriented classification method which is suitable for the high resolution remote sensing images can make full use of the spectral and spatial information of remote sensing images.In this study,the information extraction method of object-oriented was compared to maximum likelihood method based on 2012 ZY-3 satellite high resolution remote sensing image.The results show that the optimal segmentation scale of coniferous and broad-leaves and coniferous and broad-leaves mixed forest is 105,water and building is 65 according to improved local variance method.In object-oriented image analysis,the optimal segmentation scale of different land types was selected by improved local variance method,the ground features of different types were extracted in multi-scale level.The accuracy of high resolution remote sensing image information extraction based on object-oriented image analysis technology was 90.3%,kappa coefficient is 0.82;the accuracy of high resolution remote sensing image information extraction based on maximum likelihood method was 77.6%,kappa coefficient is 0.71;the overall accuracy of the object-oriented image analysis technology is improved by12.7% and Kappa coefficient increases by 11%.It shows obviously that the object-oriented image analysis technology can be applied to domestic high resolution image information extraction.This paper also attempts to extract land use information by remote sensing images in the investigation of forest resources.

Key words: ZY3 Satellite, object-oriented, optimal segmentation scale, fuzzy classification, maximum likelihood, accuracy assessment