Year: 2026
Type: article
Abstract: Camouflaged object detection remains difficult when targets share similar colors and textures with their surroundings, particularly when only RGB imagery is available. This study examines whether non-visible spectral information can improve detection under such conditions. We propose a Boundary-Driven Guidance Network (BDGNet) for multispectral camouflaged object detection. BDGNet combines spectral-spatial prior estimation, prior-guided hierarchical feature modulation, and progressive boundary-aware decoding to use spatial and spectral information while reducing background interference. Experiments on the MCOD benchmark show that BDGNet outperforms the compared methods. Band-subset analysis also shows a clear benefit from information beyond RGB: the $$F_\beta$$ score increases from 0.776 with RGB input to 0.791 with Red Edge/NIR input and to 0.845 with the full multispectral input. These results indicate that multispectral observations can complement RGB imagery and improve target delineation in visually ambiguous scenes.
Source: Scientific Reports
Authors: Binfen Ding, Dandan Liu, Huiling Ni, Jieyu An
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236018727
