Multi-band Image Classification with Ultra-Lean Complex-Valued Models

Multi-band aerial images are invaluable for remote sensing applications. Paired with modern deep learning methods, this modality has great potential utility in humanitarian assistance and disaster recovery efforts, paired with modern deep learning methods. However, state-of-the-art deep learning methods require large-scale annotations like ImageNet, and there are no equivalent multi-band image datasets. As an alternative to transfer learning on such data with few annotations, we apply complex-valued co-domain symmetric models to classify real-valued multi-band images. Our extensive experimentation on 8-band xView data shows that our ultra-lean model trained on xView from scratch without data augmentations can outperform ResNet with data augmentation and modified transfer learning on xView. Our work is the first to demonstrate the value of complex-valued deep learning on real-valued multi-band spectral data.