Multispectral detection of camouflaged targets in foggy and complex scenes

Multispectral camouflaged target detection (MSCTD) in complex scenes under fog remains a formidable challenge, primarily due to the difficulty of recovering distorted multispectral information and identifying anomalous bands that effectively highlight camouflaged targets. This study effectively addresses these challenges through a systematic approach that integrates an enhanced wavelength-dependent extinction coefficient model with an adaptive anomalous band selection method. We demonstrate the efficacy of the proposed method in detecting anomalies embedded within complex backgrounds under fog conditions. All results are obtained using a YOLOv7 model trained on both a self-constructed multispectral and pseudo-RGB camouflaged tank dataset and an existing multispectral-RGB semantic segmentation of camouflage objects dataset. Importantly, the proposed approach is versatile as it can be seamlessly integrated with advanced RGB-based dehazing techniques through prior knowledge of operational bands. This adaptability underscores its significance as a step forward in foggy MSCTD with broad applications in remote sensing and military reconnaissance.