Research On Object - Oriented Information Extraction Of High - Resolution Remote Sensing Image Land Cover Information

High spatial resolution remote sensing can be clearly and accurately express characteristic boundary, shape, texture, internal structure and spatial relationship of surface features, it has been widely used in various industries for its image. However, with the substantial increase in image resolution, the image inside the spectral differences also increased significantly, how efficient and accurate extraction of high resolution remote sensing image feature information on a major research directions in the field of remote sensing. Object-oriented image analysis technology, compared to traditional pixel-based classification, the higher classification accuracy, it is possible to avoid the "salt and pepper" effect, classification easier vectorization and storage.Thesis on object-oriented image information extraction method of study and research, the use of high resolution image data SPOT6 in Daxing District of Beijing as the study area, based on two studies to improve the study area of automated information extraction and classification accuracy. The first is the image segmentation, proposed a "first over-segmentation, and then optimize segmentation" segmentation strategy is the use of multi-scale object-oriented segmentation of the initial images for "over-segmentation", and then split the difference on the use of spectrum through segmentation result is optimized segmentation, image segmentation by analysis of the experimental results can be seen, the use of image objects obtained in this manner is divided not only meet internal homogeneity of the largest, but also reduces the degree of fragmentation segmentation results. Followed by the image classification, using the same classification of samples, testing samples, segmentation and image objects scale parameter characteristics were supervised classification of nearest neighbor classification, rules of classification threshold separation algorithm, and a combination of the two proposed integrated hierarchical classification Thought classification policy methods used in the study area and land cover classification accuracy analysis. Comparative classification results, the overall accuracy of nearest neighbor classification, classification threshold separating operator and integrated classification by 75.59% and 78.65% up to 90.44%, KAPPA coefficient rose to 0.726 from the 0.688 and 0.876, thus indicating the proposed combination of layered paper Closest classification and threshold classification algorithm ideological separation methods used in combination, can effectively improve the classification accuracy and classification stability, more suitable for high-resolution images of land cover classification.