Multi-scale Segmentation Parameters Study Of Land Cover Element For Geographic National Conditions Surveying

The extraction of information for land cover is an important task in Geographic National Conditions Surveying and the traditional visual and manual interpretation cannot satisfy the demand of this extraction. So an accuracy and efficiency method to extract the information for land cover is one of the important technologies in present. This paper starts with the multi-scale segmentation for high resolution image segmentation, and uses the genetic algorithm to calculate the optimal parameters for image segmentation to solve the problems of the parameters which are highly depend on the manual selection and lack of theoretical basis.The main works of this paper are as following(1)Discuss the influence of the parameters to the segmentation. This part mainly discuss the influences of the scale of segmentation, the differences of spectral/shape and smooth on multi-scale segmentation. And the importance of different objects are given different parameters is demonstrated through experiments.(2)Discuss the optimal segmentation method based on genetic algorithm. Choose some typical objects and use the genetic algorithm to find the optimal parameters for segmentation. After that, different fitness functions are used to do the segmentation experiments and finally the optimal parameters for the present object are found and we gain a segmentation results with high accuracy. This method can guide the extraction of information in segmentation stages and technology for different objects in different areas.(3) Verify the necessity of multi-scale segmentation with optimal segmentation parameters. To demonstrate that the accuracy of segmentation is improved by multi-scale segmentation parameters. Through three methods of fitness value comparing of multi-scale segmentation results and single-scale segmentation results, accuracy assessment of classification feedback, from the companion we found that hierarchical multi-scale segmentation can improve the accuracy of information extracted from land cover which can improve the one. Efficiency and accuracy of Geographic National Conditions Surveying and proved better services for it.