2024/01/18 by Fischer, Manfred M., Staufer-Steinnocher, Petra
paper · doi:10.57938/abae873a-b82e-492a-873f-bc1dd658d42c
Pattern recognition in urban areas is one of the most challenging issues in <br/>classifying satellite remote sensing data. Parametric pixel-by-pixel classification <br/>algorithms tend to perform poorly in this context. This is because urban areas <br/>comprise a complex spatial assemblage of disparate land cover types - including <br/>built structures, numerous vegetation types, bare soil and water bodies. Thus, <br/>there is a need for more powerful spectral pattern recognition techniques, <br/>utilizing pixel-by-pixel spectral information as the basis for automated urban <br/>land cover detection. This paper adopts the multi-layer perceptron classifier <br/>suggested and implemented in [5]. The objective of this study is to analyse the <br/>performance and stability of this classifier - trained and tested for supervised <br/>classification (8 a priori given land use classes) of a Landsat-5 TM image <br/>(270 x 360 pixels) from the city of Vienna and its northern surroundings <br/>- along with varying the training data set in the single-training-site case. <br/>The performance is measured in terms of total classification, map user's and <br/>map producer's accuracies. In addition, the stability with initial parameter <br/>conditions, classification error matrices, and error curves are analysed in some <br/>detail. (authors' abstract)