Rule-based land cover classification model : expert system integration of image and non-image spatial data

[摘要] ENGLISH ABSTRACT:Remote sensing and image processing tools provide speedy and up-to-date information on landresources. Although remote sensing is the most effective means of land cover and land use mapping, itis not without limitations. The accuracy of image analysis depends on a number of factors, of which theimage classifier used is probably the most significant. It is noted that there is no perfect classifier, butsome robust classifiers achieve higher accuracy results than others. For certain land cover/uses,discrimination based only on spectral properties is extremely difficult and often produces poor results.The use of ancillary data can improve the classification process. Some classifiers incorporate ancillarydata before or after the classification process, which limits the full utilization of the informationcontained in the ancillary data. Expert classification, on the other hand, makes better use of ancillarydata by incorporating data directly into the classification process.In this study an expert classification model was developed based on spatial operations designed toidentify a specific land cover/use, by integrating both spectral and available ancillary data. Ancillarydata were derived either from the spectral channels or from other spatial data sources such as DEM(Digital Elevation Model) and topographical maps. The model was developed in ERDAS Imagineimage-processing software, using the expert engineer as a final integrator of the different constituentspatial operations. An attempt was made to identify the Level I land cover classes in the South AfricanNational Land Cover classification scheme hierarchy. Rules were determined on the basis of expertknowledge or statistical calculations of mean and variance on training samples. Although rules couldbe determined by using statistical applications, such as the classification analysis regression tree(CART), the absence of adequate and accurate training data for all land cover classes and the fact thatall land cover classes do not require the same predictor variables makes this option less desirable. Theresult of the accuracy assessment showed that the overall classification accuracy was 84.3% and kappastatistics 0.829. Although this level of accuracy might be suitable for most applications, the model isflexible enough to be improved further.

[发布机构] Stellenbosch University