Complementary Detection of Camouflaged Soldiers Using a Diffusion Model

Identifying camouflaged soldiers in warfare is crucial for reducing friendly casualties and gaining substantial tactical advantages. This paper proposes a novel approach for classifying and detecting camouflaged soldiers using synthetic datasets, ensemble models, and the complementary object recognition algorithm. To overcome the challenges of visually identifying camouflaged soldiers, we leverage advanced intelligent recognition technologies. To address the scarcity of datasets featuring camouflaged soldiers, we employ a diffusion-based synthetic data generation method. Specifically, we use DreamBooth to produce large-scale synthetic data from a limited number of real images, effectively expanding the dataset. We utilize this augmented dataset to train ensemble classification models, which combine the strengths of multiple classifiers to achieve improved performance. To further enhance reliability, the proposed algorithm integrates the classification models with object detection models, enabling them to interact and compensate for each other’'s false detections. This synergy significantly improves overall detection accuracy. Experimental results confirm that the proposed method outperforms approaches relying solely on classification or detection models, demonstrating its superior performance in identifying camouflaged soldiers.