Comparative Analysis of SVM and ANN Classifiers using Multilevel Fusion of Multi-Sensor Data in Urban Land Classification

Multi-sensor data fusion has recently received remarkably more attraction in urban land classification. The fusion of multi-resolution and multi-sensor remote sensing data can help in comprehending more information about the same land cover features, thereby, enhancing the classification accuracy. In this field of study, a combination of hyperspectral data in a long-wave infrared range and a very high-resolution data in a visible range has been extensively used for exploring the spectral and spatial features for decision level fusion classification. This paper proposes a novel method of integrating the classifier decisions with the additional ancillary information derived from spectral and spatial features for improvement in the classification accuracy of natural and man-made objects in urban land cover. The paper also presents a detailed performance comparative evaluation of two classifiers i.e., support vector machine (SVM) and artificial neural network (ANN) to show the effectiveness of these classifiers. The results obtained from a decision-based multilevel fusion of spectral and spatial information using hyperspectral and visible data have shown improvement in classification accuracy. The results also reveal that the classification accuracy of the SVM classifier is better than ANN in multi-sensor data using decision level fusion of combined feature set analysis.