Land Cover Classification: Comparison between Fuzzy and Boolean ClassifierS

Abstract

Production of Land Use/Land cover maps is very important for environmental monitoring and development. Image classification using either hard and/or soft classifiers is crucial in the production of these maps. While fuzzy classification is suitable for modelling vagueness due to mixed pixels in the land cover, Boolean on the other hand is suitable for modelling land cover with well-defined boundary. The analyst’s choice of image classifier is a very important decision in image classification as this determines the classification output. Using Landsat5 TM of 1984, Landsat 4TM of 1992 and Landsat7 ETM+ of 2000 satellite images, this research looks at the comparison between soft (Fuzzy) and hard (Boolean) classifiers. The Landsat ETM+2000 of a 15m spatial resolution was resampled to a 30m pixel size so that the three images would be of the same pixel size to effectively carry out pixel-to-pixel analysis. Due to the nature of the landscape and bearing in mind that land cover responds differently to various Landsat spectral bands, three band combinations (image bands 2, 3, and 4) were considered for the classification. The images were classified into four (4) different land spectral classes by employing the fuzzy membership function and maximum likelihood classification tools in Idrisi Taiga 16 software. The results obtained shows that the spatial distribution of the modelled land cover classes for both Fuzzy and Boolean is basically the same which buttresses the performance level of both models. The major difference of the two models lies in the output; while fuzzy shows a subtle representation according to degree of membership function of each land cover class, the Boolean on the other hand represented the land cover types with a well-defined boundary. Also, summation of the fuzzy land cover areas is not equal to the size of the study; 108% in 1984, 107% in both 1992 and year 2000 are unlike the Boolean with 100%.