Multispectral Object Detection
Overview
Multispectral object detection studies how to combine complementary sensor bands—such as RGB, thermal infrared, near-infrared, and hyperspectral imagery—to detect objects that may be hard to see in any one band. Common approaches use dual-stream detectors with feature fusion, cross-modal attention, learned alignment, or reliability-aware routing; newer work also adapts vision-language and pretrained vision models to spectral inputs. Benchmarks such as FLIR, LLVIP, M3FD, and DroneVehicle test performance in settings including night-time driving, aerial imagery, and small-object detection. A central challenge is that sensors can be misaligned, noisy, or unavailable, while richer fusion can increase computation and still fail to generalize across datasets and conditions.
Latest papers
43 papers4 papers in the last four weeks, down 33% on the four weeks before. 0.0% of all new papers.