Research On Automatical Classification Of Land Use/Land Cover Change In Complicated Terrain Regions

Land use and land cover change (LUCC) is an important component part ofglobal environmental change and one of main causes result in global environmentalchange. In recent years, the land use/land cover changes in regional scale haveattracted more and more extensive attention. As an important means of access to dataon land use, remote sensing image classification enables understanding of changes inwatershed land use/land cover in an comprehensive, efficient, and real-time way,thereby providing data support to the updates, dynamic monitoring, and hydrologicalmodeling of data on land use/land cover in real time.Taking Huangshui River Basin in Qinghai Province as the study area and the2011Landsat5TM images as the source of data, this paper firstly divides the studyarea by the altitude, water and heat conditions into irrigated sub-area, low hillysub-area and hilly sub-area, and then through comparison of the initial classificationresults, it determines the best bands of classification for Huangshui River Basinand carries out band combination, and selects samples for each geographicalsub-area and makes automatic classification of the land use/land cover ofHuangshui Watershed by using the maximum likelihood method and artificial neuralnetwork respectively; secondly, on this basis, according to the distribution ofwatershed land use types and their characteristics, it selectes6bands of TM, DEM,slope, aspect, NDVI, NDBI, MNDWI, principal component analysis bands andhand-painted irrigated land border as characteristic parameters for establishment ofdecision tree and then establishes the decision tree for each geographical sub-area toachieve the automatic classification of watershed. This paper through precisionevaluation determines the most suitable method of classification for HuangshuiWatershed, providing reference for quick access to the data on land use/land coverof Huangshui Watershed in the future. The main conclusions are as follows:(1)The overall accuracy of Maximum likelihood classification method, artificialneural network classification and decision tree classification in the Huangshui basinland use/land cover classification were76.20%,79.44%,84.93%, Kappa index0.72,0.76,0.82. The result indicated that the best method of decision tree classification,artificial neural network classification method was slightly better than the maximumlikelihood classification method, and maximum likelihood classification worst.(2)From the view of producers’ precision, each precision of land use type of thethree classifications varies from each other. First, all the producers’ precisions ofunused land of the three categories are relatively high, and classification accuracy is between86.14%~91.09%, of which the maximum likelihood classification>Artificial Neural Network Classification> decision tree classification. Secondly,producers’ precision of irrigated land, dryland, grassland and water area is this:decision tree classification> Artificial Neural Network Classification> maximumlikelihood classification. But the classification accuracy of the woodland, urban andrural residential construction land and unused land is different from the categoriesmentioned above, which displays like this: woodland: an artificial neural networkclassifier> maximum likelihood classification> decision tree classification; urbanand rural residential land for construction: performance of maximum likelihoodclassification> decision tree classification> Artificial neural network classification.