The Wetlands Research With Remote Sensing Images Based On Object-oriented Classification In Momoge Region, Jilin Province

As one of the most important wetlands resources, Momoge Nature ConservationZone is a site for Aves to find food and to rest when they are migrating. For a longtime, because of various reasons, we have not researched the wetlands in this regionby the numbers. The importance of wetlands is known more widely than before andrecently wetlands in this region have changed remarkably. Using two phases ofremotely sensed images and the technology of 3S, we have applied the object-orientedclassification to the sensitive region—Momoge Conservation Zone, where wetlands isdiffusely situated. At last, we have concluded the change trend of the wetlands in thisregion.With regard to the classification of wetlands, there are various ways ofclassification in the world. Taking different classifications in native and foreigncountries into account, and combining the features of the local place, we haveestablished a classification system of wetlands in Momoge Nature Conservation Zone.We have extracted the wetlands classification of seven classes and the land use/coverclassification of fifteen classes.Object-oriented classification is an approach of image classification basedmulti-features. Being different from methods of traditional classification, which onlyapplies spectrum of single pixel to research, object-oriented classification takes intoaccount contextual information, including texture, shape, area and size, etc, and therelationship between super-objects, sub-objects, neighbor-objects. By adoptingobject-oriented classification, through image segmentation and the constructing offeature space, this dissertation has obtained maps with integrated area. The mapscome closer to the situation in the field. In this research, we have completed wetlandsmaps, land use/cover maps, change map of land use/cover, and vector graphs.This dissertation has compared object-oriented classification with pixel-basedclassification. Choosing a test area in this region for the examination, we haveclassified it with these two approaches and made an accuracy assessment with 200points generated randomly by the computer. The result shows the accuracy ofobject-oriented classification is much better than that of pixel-based classification.The information we extracted from remote sensing images is more accurate and morereliable.Based on the research above, we have drawn a conclusion that this sensitiveregion has been influenced by the main factors of nature and human for decades, andthe area of wetlands decreased rapidly. It is high time that the local governments tookeffective measures to protect environment.