Collaborative Classification for Woodland Data Using Similar Multi-concentrated Network

With the increasing of the forest area and complexity of tree species, collaborative classification using multi-source remote sensing data has been drawn increasing attention. Fusion of hyperspectral and LiDAR data can improve to acquire a comprehensive information which is conductive to the forest land classification. In this work, a similar multi-concentrate network focusing on the fine classification of tree species, denoted as SMCN, is proposed for woodland data. More specific, a preprocessing stage named pixel screening for data intensity critical control is firstly designed. Then, a similar multi-concentrate network is developed to capture spectral and spatial features from hyperspectral and LiDAR data and make specific connections, respectively. Experimental results validated on Belgian data have favorably demonstrated that the proposed SMCN outperforms other state-of-the-art methods.