Comparison of Support Vector Machine, Artificial Neural Networks, and Maximum Likelihood Methods in Urban Impervious Surfaces Detection Using High Spatial Resolution Image

Impenetrable surfaces are an important factor in monitoring urban development and environmental quality. For this reason, recognizing these levels will be a precondition for achieving sustainable urban development. Although there are various methods for detecting and studying these levels, accurate and cost-effective determination of these levels is still an important challenge for urban researchers. Due to the increasing availability of remote sensing data, in this study using the GeoEye 2009 image, the accuracy of the three most common classification methods, artificial neural network and support vector machine for determining impermeable surfaces in a part of Bandar Abbas city was compared. For this purpose, after performing the necessary preprocessing operations on the image, using the aforementioned algorithms, five classes of street and building (as impermeable surfaces), water body, vegetation and wasteland (as impermeable urban surfaces) for each Three methods were extracted. To evaluate the results, methods of overall accuracy, kappa coefficient, user and producer accuracy were used. The results showed that the support vector machine with 94.7% overall accuracy and kappa coefficient 0.93, compared to artificial neural network method (93.1% overall accuracy and kappa coefficient 0.90) and the most similarity method (with accuracy). The overall accuracy was 92.2% and the kappa coefficient was 0.89). Although the present study showed that the support vector machine method was more accurate, nevertheless, the accuracy of the most similar and artificial neural network methods were accurate and acceptable in determining impermeable surfaces and processing of high spatial resolution images with these methods, Can detect impermeable surfaces.