Neural network classification with IFSAR and multispectral data fusion

Abstract

The authors present neural network classification results for interferometric SAR (IFSAR) and multispectral imagery data, and describe a classification fusion scheme for the combination of the two classification results to reduce ambiguities and false classification rates. Two multilayer perceptron (MLP) neural networks were developed for the classification of IFSAR and multispectral data, separately. Classes include tree area, road, building, bare earth, water, etc. A classification fusion scheme that examines both the IFSAR and multispectral classification results at a pixel location and decides the fusion class for various cases is then discussed. Classification fusion results, especially the building classification and detection results, are presented. The results show that the scheme is effective in reducing false classification rate for buildings detection in the remotely sensed imagery data.