Boosting Small Object Detection via the Pixel-level Fusion of Panchromatic and Hyperspectral Imagery

Small object detection from panchromatic (PAN) and hyperspectral imagery (HSI) remains challenging because the two modalities differ substantially in spatial resolution and spectral representation. Most existing approaches process image fusion and object detection as separate stages, which increases computational cost and may propagate errors from the fusion stage to the detector. To address this issue, we propose a multi-level supervised framework for small object detection using PAN and HSI. First, an interactive loss function is introduced to enable joint training and to improve both detection performance and interpretability by injecting task-related supervision into the network. Second, a mechanism-guided image-fusion component combines ratio-transform-based fusion with deep feature learning. Finally, a CNN-transformer-based detection component with multiple spatial-spectral attention projection modules is constructed to extract spatial and spectral information effectively. Experiments on datasets collected by the Tiangong-1 and EO-1 satellites show that the proposed method outperforms several state-of-the-art approaches.