Near infrared face recognition based on YOLOv8 improvement
Chengliang Song, Qizhi Zhang, Qiong Liu, Wei Liu · 2025
In this paper, a near-infrared face recognition method based on an improved YOLOv8 network is proposed for the needs of nighttime face recognition. The backbone network of YOLOv8 is reconstructed using a lightweight FasterNe network to reduce redundant computations and memory accesses, and extract defective features more efficiently; and the SEAttention attention mechanism is introduced to improve the recognition accuracy and inference speed while reducing the consumption of computational resources. The experimental results show that the improved model maintains a higher accuracy while making the computational complexity and resource requirements smaller, relatively consumes less computational resources, significantly improves the efficiency, and is suitable for deployment and application in resource-constrained environments. The innovation of this research lies in combining lightweight network and attention mechanism to effectively address the challenges of infrared face recognition in dark environments, which provides a feasible technical solution for the application of security surveillance systems.