The Research Of Land Use/Cover Change Detection Using High Resolution Remote Sensing Images Based On Image Segments Statistical Analysis

Land resources are the most important resource for human to live on the earth. It cannot only reflect a country or a region's surface environment and geography basicsituation, but also can to a certain extent reflect regional economic development,urbanization and military layout and so on. In recent years, along with thedevelopment of3S technology, geographic information service has been engaged intothe people's daily life in various styles, and the geographical information acquisition,quickly update have been put forward higher demand. Land use/cover situation is thecomprehensive result of the interaction between human and environment, and as thebasic data needed for resource management and geographic information services, theacquisition, analysis and updating of land use/cover information seems particularlyimportant.The macro and real-time features of the remote sensing image data has always made itto be the most important data source for land use/cover and its change detection, andin this century, land use/cover change detection has a further development due to theavailable of the high resolution remote sensing image data. From the research currentsituation, there are several approaches for land use/cover change detection, and theefficiency of each method depends on the change detection requirements andexperimental data characteristics. Up till now there is not a mature framework orsystem for change detection. At present the research focuses are two aspects: themethod of change detection, and the automation of the method. More informationused for image analysis could be gotten from high resolution remote sensing image,compare to the low resolution images. However, on the one hand, the more prominentcontradiction occurs between the increased amount of image data and the lowefficiency, high cost artificial interpretation; on the other hand, the approaches usedfor analyze low resolution image data may not be appropriate for high resolutionimage analysis. So we can say high resolution remote sensing images bring both advantages and challenges for land use/cover change detection. Therefore, this papergo further researches on both methods of change detection and integrity, automationof change detection process using high resolution remote sensing images.More information could be extracted for land use/cover detection using multiplesource data, and these data can be regarded as the prior knowledge which can engageinto the process of image analysis directly. Meanwhile, the final purpose of changedetection using remote sensing data is always for updating existing land use map.Thus, this paper used both remote sensing images and existing land use map forchange detection. Image segments are used as the basic units for image analysisaccounting for the characteristics of high resolution images and the mechanism ofartificial objects identification. Through the matching between remote sensing imagesand the land use map, image segments could be gotten directly by the land use patches.Information contained in the attributes table of land use map can be used for thefollow-up image interpretation.Because of the inconsistence of the existing categories between land use and landcover, the image object obtained directly through matching process could not keep allthe pixels within it homogeneous, which would bring some difficulty to follow-upimage processing. This paper discussed this inconsistence mentioned above, usingmulti-scale image segmentation and data mining methods in order to improve thehomogeneity of the image segment.Next, this paper divided the change detection process into two methods: beforeclassification comparison and post classification and did a discussion, summarizationand experiments for each method. After analyzing both advantages and disadvantagesof these two methods, this paper gives an approach for land use/cover changedetection accounting for class spectral change rules and proves its validity by anexperiment. Meanwhile ROC curve was used for change detection threshold decisionso as to enhance the automation degree of change detection methods.On this basis, through the construction of Markov Random Field graph model andusing statistical method, spectral features of the image segments, spatial relationshipamong the image segments and their neighborhood segments, and timing relationship among segments on different period images are used comprehensively for decidingthe new class properties of changed image segments. The experiments proved theefficiency of this method and meanwhile confirmed the important role of spatial andtiming relationship using for remote sensing image interpretation.Finally, this paper introduced a system framework of high resolution remote sensingimage land use/cover change detection based on the statistical analysis of imagesegments, and also a brief introduction of the main function of this system which havealready achieved.