Fuzzy aggregation for multimodal remote sensing classification

This paper investigates methods of fusing hyperspectral imagery (HSI) and LiDAR data for urban land use and land cover classification. A variety of fusion methods including combination rules, deep neural networks, and fuzzy aggregation are compared against using any single modality for classification. The experimental results demonstrate that the two fuzzy aggregation methods, the linear order statistic neuron (LOSN) and the Choquet integral (ChI), achieved the best overall and average classification accuracy, respectively. We further discuss how the fuzzy aggregation methods provides advantages with difficult samples and the opportunity to gain network explainability.