HFMIDet: Hierarchical Feature Fusion-Guided Multidimensional Infrared Pedestrian Detection Network

Infrared images are favored for their exceptional anti-interference capabilities. However, challenges like low resolution and a scarcity of detailed textures can impede the effective recognition of multi-scale object information in infrared imaging. To address these issues, we designed the Hierarchical Feature Fusion Guided Multidimensional Infrared Pedestrian Detection Network (HFMIDet), focusing on extracting richer details and global features within complex scenes. Firstly, we design a hierarchical feature fusion network, which uses multi-scale fusion module to achieve cross-layer feature combination and multi-level information fusion module to achieve multi-level feature fusion, so as to enhance the ability of the model to perceive the target location. In addition, the frequency-spatial feature enhancement module aims to effectively suppress background noise by combining frequency and spatial domain feature information, so that the network can extract effective object shape and global frequency information even in complex backgrounds. Finally, we use the adaptive denoising transformer head for the final detection task, while Focal CIoU is used to perform the loss feedback task. Experimental results on three public datasets of infrared pedestrians show that our model can achieve superior performance and outperform many state-of-the-art methods.