Feature Level Fusion of Hyperspectral Image and LiDAR Data based on Multi-Objective Particle Swarm Optimization in Classification of Urban Area

Hyperspectral and LiDAR data provide spectral and height information and they have high potential in classification of complex urban area. This paper proposed meta-heuristic method in feature level fusion of them. For this purpose, a comprehensive spectral-spatial-structural feature space is generated based on feature extraction method such as spectral indices, texture analysis, roughness, etc. Previous methods apply just one criterion to evaluate classification performance. However, in the proposed method, three criteria including generalization ability, classification complexity and classes separation are considered. Multi-Objective Particle Swarm Optimization (MOPSO) is implemented to select optimum feature space and Support Vector Machines (SVMs) parameters simultaneously while optimize all three parameters. The obtained results show the proposed method increases classification accuracy up to 11% and 58% respect to hyperspectral imagery and LiDAR data by eliminating 300 features (among 611 feature) and also increasing classes separation.