Information Extraction Base On The Remote Sensing Image Valid Segment

This paper briefly traduce the method of remote sensing image interpretation, and point out the issues of information extraction from low resolution remote sensing image using the traditional methods. The tradition method of information extraction usually only takes into account the spectral feature of the remote sensing image, which is hardly used to the high resolution remote sensing image for information extraction. By contrast, object oriented Segmentation which takes into account the multi-dimention information of the segment unit is more effect to express the object features. However the multiscale-segmentation scale of object oriented is usually hard to get because of the complexity of the numerous types of objects. One segment scale usually result in the problems of over-segmentation and less-segmentation.In order to solve the problems of less-segmentation and over segmentation phenomenon this paper designs a strategy of top-to-down segmentation. Then, the valid unit conception was used to model the valid segmentation of remote sensing image in order to judge whether the segmentation can fit the target objects well. Lastly, proper method on how to get segmentation threshold was given by the analysis of statistic and theoretical experience segmentation. The paper mainly takes into account the level structure, segment spectral and segment adjoin relationship.This paper emphasis on the validation of multiscale-segmentation strategy. In the experiment, a software tool of modeling the eCognition multiscale-segmentation was developed with the ArcObject in order to certify the paper design. The main procedure can be divided into two steps. Firstly, getting the original segmentation data with(Fractal Net Evolution Approach, FNEA) algorithm of eCognition. Then, inputting the segment data to the verification software tool to get the segmentation judging results timely. The experiment results show that the segmentation strategy of this paper can well control the phenomenon of over-segmentation and less-segmentation in the information extration of high resolution remote sensing image.