Bidirectional Cross-Spectral Attention for Fine-Grained Weed Recognition in UAV Multispectral Imagery
Contributors
Vudddagiri MNSSVKR Gupta
Shiva Shankar Reddy
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Fine-grained recognition of weed species using UAV imagery from the multispectral bands is still a difficult task, as existing approaches do not fully utilize complementary information available in the different bands (RGB, NIR, Red-Edge and SWIR). In this paper, we propose SpectraFuse-Net, a cross-spectral attention framework that includes a multispectral encoder, bidirectional cross-spectral attention, spectral-aware attention, a fine-grained classifier, and Spectral Complementarity Regularisation. We evaluated the model on WeedMap and four custom field datasets, achieving 95.3% species F1, 93.8% mAP@0.5, 92.6% weed recall, and 96.9% crop precision. This is 6.4% higher than the best baseline for species F1 and 4.9% higher for mAP. Ablation results show that bidirectional attention and spectral regularisation significantly improve species-level discrimination.