Multilevel Feature Gated Fusion Based Spatial and Frequency Domain Attention Network for Joint Classification of Hyperspectral and LiDAR Data

作者：Cuiping Shi, Zhipeng Zhong, Shihang Ding, Yeqi Lei, Liguo Wang, Zhan Jin · 发表于：IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · 年份：2025 · DOI：10.1109/jstars.2025.3534286 · 被引用次数：8 · 研究领域：Remote-Sensing Image Classification

Hyperspectral images provide rich spectral information, while LiDAR data supplements three-dimensional spatial structural information. The combination of the two can effectively improve the accuracy of land cover classification. However, how to effectively utilize their complementary advantages for cross modal feature fusion and enable the fused joint features to capture global contextual information while maintaining local texture details is a challenge. In addition, most existing joint classification methods based on attention and transformer only perform global modeling in the spatial domain, ignoring the sensitivity of the frequency domain to fine features. In this article, a multilevel feature gated fusion based spatial and frequency domain attention network is proposed for joint classification of hyperspectral and LiDAR data. First, extract multilevel convolutional features from hyperspectral and LiDAR images and adaptively fuse them through a gating mechanism. Then, design an attention module that combines spatial frequency domain to model global fine features. In addition, a carefully designed texture feature extraction module is utilized to further enhance local fine feature extraction. The experimental results on three commonly used datasets show that the classification performance of the proposed method is significantly better than some state-of-the-art methods.