The Study Of Scale Effects In Typical Objects' Feature Extraction From Remote Sensing Image

As the most basic feature of remote sensing image, scale describes the information level of images. Images of the same feature will express different information if the scale varies. How to make efficient use of the relationship between images'scale and the precision of feature extraction to calculate out the appropriate scale that can fully express the features is an important direction in today's scale researches of remote sensing images.Taking Jiu Yuan Drainage as the test area, SPOT5 remote sensing image as data source with a resolution of 2.5m, based on scale transformation, a deep study on feature extraction and appropriate scale calculation of multi-scaled images is carried out. Taking five typical objects as investigated sample, the typical features were extracted and appropriate scales were calculated, respectively. The results obtained by this paper are as follows:(1)After evaluating the images obtained by five classical scale transformation methods with scale effect index, the image converted by cubic convolution method still retains the spectral property, clarity and the enough amount of information compared with the original image, so it can be used as the basic data to extract features and calculate the appropriate scale.(2) Compared the spectral property, shape property and texture information of the typical object sample selected in images of different scale with the help of TM image in this region, we can find that as the image's resolution becomes lower, the typical properties of the samples are less and less, the internal structure information is more and more simple, the gray level is less and less, the shape is more generalized, the samples'shape complexity is lower and lower.(3) In terms of the special features of Jiu Yuan Drainage, terrace, water, forest, residence and irrigated land were selected as five types of investigated samples, whose appropriate scale were studied. The five features'appropriate scale were calculated by use of variant function method and the scale range to express the test region appropriately was calculated by local variance method. Taking the phenomenon that the objects'shape feature becomes more generalized as the image's resolution is lower into consideration, the integrated shape index was developed to calculate the appropriate scale to express the typical objects'shape features.