LPFI-Net: A Fusion Network for LiDAR Prior and Frequency Interaction in Hyperspectral and LiDAR Classification

Abstract

The integration of multisource information from hyperspectral imagery (HSI) and light detection and ranging (LiDAR) provides reliable data support for land cover classification. The heterogeneity between HSI and LiDAR data poses significant challenges for cross-modal feature alignment, while the rich structural priors inherent in LiDAR data remain underutilized. Therefore, significant feature alignment learning between HSI and LiDAR, together with the embedding of LiDAR structural priors, can substantially improve land cover classification performance. Based on these, we propose the LiDAR prior and frequency interaction network (LPFI-Net). LPFI-Net employs frequency-domain feature learning to achieve flexible HSI-LiDAR salient feature alignment. A frequency-domain interaction mechanism, together with a frequency-invertible enhancement module, is introduced to strengthen cross-band representations and improve HSI-LiDAR feature reconstruction. In the spatial–frequency fusion stage, a LiDAR structural prior-guided mechanism transforms spatial information into quantifiable priors, enabling more effective cross-modal feature enhancement. Experiments on the Houston2013, MUUFL, and Augsburg datasets demonstrate the effectiveness of LPFI-Net, with code available online.