Research On Application Of TM Image To Auto Extraction Of Forest Area In Western Sichuan Province

With the rapid development of remote sensing technology, a great deal of remote sensing image data was acquired. How to extract the objects that people interested from the remote image is becoming the problem to which people have paid more and more attention.This thesis mostly investigates the application and methods of digital image processing technology, introduces briefly basic knowledge and actuality of remote sensing. Based on spectrum character of vegetation, how to choose the better extraction methods, perform computer auto classification, and analysis the factors that affect the extraction for vegetation information.In this paper, ENVI is the primary analysis tool, assisting with MapGIS6.1, DaoFu country is the region of study, TaiZhan and MaZi forest farm is the material analysis objects, the basic theory and methods of image classification are adopted, mostly finished forest land, pasture range, plantation, shrub land and unuseful land extraction from ETM+ image, from image preprocessing, spectrum character analysis, multi-spectrum transformation, image classification and post classification.The primary work in the paper as followed:According to image spectrum character, correlation coefficient and so on, how to choose the band combination and image fusion.Based on Geometry correction, image enhance, image pretreatment was finished.Impaired effectively the effect of shadow of landform, with proper vegetation index (VI).Made sure of the parameters and thresholds of different objects with supervised and unsupervised classification.For image classification, the means of minimum distance and K-means unsupervised classification based on statistic pattern were used, and reached certain classification accuracy. For bands selection, the traditional TM432 combination was broken, after analyzing image spectrum character, bands fusion and comparing the TM bands threshold .the standard divide and the correlation coefficient of all bands. The combination TM453 was made certain finally by OIF, and achieve better purpose.Adopting Landsat-7 ETM+ data and the measures in the paper, classified result showedthat the minimum distance method could achieve certain classification accuracy, but it only classified exactly these objects which spectrum diversity is obvious, for these objects which spectrum character alike, the classification accuracy was worse, especially the shrub land and the forest land in TaiZhan forest farm.The result also indicated that in the regions where landform and vegetation types were simple, sub-compartment area was larger, unsupervised classification integrating ground survey also could achieve better accuracy. The disposal of shadow of landform, excepted for using vegetation index(VI), we should do some ground survey and validation. As for remote sensing development presently, we should not substitute completely computer auto classification for the ground survey.