The Research On Comparing Detection Methods Of Land Use Changes In Mountain Forest Areas Based On TM Images

Based on two-sets of TM image data, This thesis compared the technical method of extracting changing information of forest land in mountainous regions,trying to offer technology reference for making full use of free Landsat series satellites images to extract Changing Information of forest land in mountainous regions.(1)Do geometric precision correction to images which have done standard terrain correction processing, but had errors. the remote sensing image of the study area which is mountain. And radiation calibration and atmospheric radiation correction were carried out for the images. Because the elevation difference of the study area is not big, the shadow area was mainly the forest land, to improve extraction accuracy of the forest land change information, so the shadow area will be divided into woodland shadow area in the land use/cover classification system.(2)In order to improve the extracting accuracy of Changing information of forest land, with full consideration of the separability of land classes in images of TM and the national standards of the division of forest land types, drew a large number of sample area in class according to the actual circumstances of the land use/cover In the studied area, determined to the separability between classes based on the calculated value of Jeffries–Matusita, combined with the classes of less value, determined the classification system of two periods of Land Use/Land Cover in the studied area.(3)In order to make full use of the information of image of TM, selected optimal band combination based on the value of optimal index calculated to extracted feature information and original bands. Acquired classification results of photo of two-temporal of optimal bands combination and TM543 bands combination respectively with support vector machine(SVM) and maximum likelihood, Compared and analysised the evaluation result of accuracy, find: the land information extraction effect of two periods of optimal bands combination and TM543 bands combination is similar; the overall classification accuracy and kappa coefficient of Maximum likelihood was lower than those of support vector machine(SVM); At the same time considered the image classification results,selected classification results of support vector machine(SVM) of the optimal band combination images in 2004 and TM543 band combination image in 2009 as the land class information sources of extracting forest land change information.(4)To make qualitative evaluation on different images of various analysis methods which based on the pixels according to the extraction Effect of the change Information of forest land, then find: the difference Image of first component of the Principal Component Analysis could extract lots of the change Information of land obviously and wrong information was rare; original TM7 band different image could extract most of the change information of land and also wrong information was rare.(5)The accuracy of extracting change information of forest land of post-classification comparison is higher than classification result of combination image of difference images of analysis methods based on the pixels, the combination image of difference images of analysis methods based on the pixels is sensitive to the change information of forest land,but it is easy to mistake in classification. The accuracy of no change forestland, from forest land to cutting-blank, from forest land to burned area and from forestland to construction land is 96.87 %, 75.47%, 76.92 % and 83.33%.(6)This research indicates that detect automatically the variations that from forest land to cutting-blank, from forest land to burned area and from forest land to construction land by using two-temporal TM images data is applicable when the required accuracy is not very high, the result could be the reference for grasping the change information of forest land in regional scale.