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

The rapid development of high-resolution remote sensing technology enables the wide application of high-resolution remote sensing image in various fields and provides more possibilities for high-resolution remote sensing image information extraction technology development. The traditional technologies of information extraction are mainly rely on spectral features of the image, and fail to utilize the texture, color, shape and other features of the image, which not only cause a waste of image resources, but also fall short of the desired accuracy. So the traditional supervised classification methods cannot fully meet the actual application requirements. In order to extract surface mulching information faster and more accurately, the texture, color and shape of the image must be put into use. Therefore, this paper uses the object-oriented image analysis method for surface mulching information extraction research in Guandu District of Kunming city. Research work and results are as follows:(1)To determine the optimal segmentation scale. Images of the study area were used Max area method and the mean variance method two kinds of optimal segmentation scale algorithm, and analyzing the results of the two algorithms that, in order to verify the optimum scale of the object, from each segmentation scale of key features for random sampling, and classification of the outcome of each scale of segmentation accuracy evaluation.(2)Image feature extraction.Discovery and mining knowledge characteristics of surface features and the establishment of object classification and recognition rules, make full use of the features of the sample points, and we get the best feature combination each kind of feature information extraction, spectral characteristics are studied, defined feature parameters with segmentation scale increasing variation.(3)The experiment of information extraction in the study area, to determine the optimal segmentation scale of each feature, select the best feature combination, establish the classification hierarchy and database of fuzzy rule.(4)By using supervised classification, object-oriented nearest neighbor method and fuzzy classification results were compared and analyzed. Experimental results show that using the object-oriented fuzzy classification method of high-resolution image classification and accuracy higher. Object oriented fuzzy classification accuracy and kappa coefficient were 93%, 0.920, object-oriented nearest neighbor classification accuracy and kappa coefficient were 85%, 0.819, were higher than those of the supervision classification accuracy(total classification accuracy is 83%, kappa coefficient were 0.797).