FS-CGNet: Frequency Spectral–Channel Fusion and Cross-Scale Global Aggregation Network for Hyperspectral Image Classification

Hyperspectral image (HSI) classification is a challenging task due to the high-dimensional attribute of the data and the scarcity of labeled samples. Many existing models struggle to simultaneously capture both local structural features and global patterns, limiting their ability to handle complex HSI tasks to some extent. Furthermore, they often fail to efficiently process multiscale and multifrequency information, resulting in suboptimal performance. To address these challenges, we propose FS-CGNet, a novel framework that integrates frequency spectral–channel fusion and cross-scale global aggregation networks for HSI classification. FS-CGNet introduces three key components: 1) a frequency-domain spectral feature extraction module (FDSFM) that utilizes techniques like fast Fourier transform (FFT) to capture multiscale and multifrequency information, improving the ability to discriminate local and global features; 2) a multiscale spectral–channel fusion attention module (MSFAM) that adaptively emphasizes critical spectral channels, enhancing classification accuracy and robustness; and 3) a cross-scale global aggregation module (CGAM) that aggregates features from multiple scales, ensuring optimal utilization of cross-scale information and improving overall classification performance. Comprehensive experiments on four benchmark HSI datasets (Pavia University, Salinas, WHU-Hi-LongKou, and Xiong An) demonstrate that FS-CGNet achieves the state-of-the-art performance, comparable to several leading methods in the field. The code of this work is available at https://github.com/wanxiaoqing/FS-CGNet