Hierarchical Pixel-Wavelength Fusion Network for Progressive Hyperspectral Image Reconstruction

This study introduces the hierarchical pixel-wavelength fusion network (HPWF-Net), a novel hyperspectral reconstruction framework designed to address the challenges of tree species classification and spectral reconstruction in complex forest ecosystems. The proposed method employs a hierarchical pixel-wise and target-wavelength-wise reconstruction strategy, progressively recovering hyperspectral images by dynamically selecting and refining target wavelengths at each layer. HPWF-Net utilizes an enhanced U-Net architecture with multilevel loss supervision and convolutional consistency constraints, enabling accurate reconstruction by supervising intermediate encoder outputs, convolutional regressions, and final decoder predictions. This design preserves consistency with the original multispectral (MS) input while enhancing spectral fidelity and robustness. The model is trained on hyperspectral data from the NEON dataset and applied to Sentinel-2 MS imagery of the Daiyun Mountain National Nature Reserve. Experimental results demonstrate that the reconstructed hyperspectral data improve tree species classification accuracy by 5.9% compared to the original Sentinel-2 MS data. This work contributes to hyperspectral reconstruction and fine-grained ecological monitoring, offering a practical solution for remote sensing applications in forest ecosystems. Furthermore, we offer a Python-based package that enables to replicate our results at https://github.com/DarkFlameMaster02/Hierarchical-Pixel-Wavelength-Fusion-Network-for-Progressive-Hyperspectral-Image-Reconstruction