Classification of Water in an Urban Environment by Applying OBIA and Fuzzy Logic to Very High-Resolution Satellite Imagery

When it comes to soil classification, especially if performed in an urban area, the distinction between water and shadows cast by buildings is a well-known problem. Recently the geomatics laboratory of the University of Pavia purchased an image of the WorldView3 satellite of the entire municipal area of Pavia. The work area was a portion of Pavia called "Citta giardino", in which there is a canal in a context characterised by residential buildings with heights ranging from 6 to 20 m, therefore it was necessary to identify a classification strategy capable of distinguishing effectively between the shadows of the buildings and thewater of the canal. The classification was performed on the multispectral image with a Ground Sampling Distance (GSD) of 30 cm, obtained by fusing with the Gram-Schmidt pan-sharpening technique, the panchromatic and the 8-band multispectral image (respectively 30 and 120 GSD) of the study area. A good result was obtained by accurately describing the water class with different membership functions applied to the most relevant band statistics (mainly the mean and standard deviation of the pixels in each object), in fact an overall accuracy of 96.13% was achieved, by comparing it with the ground truth obtained by manually classifying most of the elements present in the scene in the "water" "vegetation" and "impermeable soil" classes.