Unauthorized drone intrusions are regarded as a severe threat to airport airspace security, especially under low-illumination conditions such as nighttime. Infrared thermal imaging is adopted as an important means for nighttime detection and recognition. However, high feature coupling between drones and birds is presented in infrared images, leading to high misjudgment rates and insufficient positioning accuracy in existing systems. To solve these problems, an improved YOLOv8n-based method for drone and bird recognition in infrared images is proposed. Firstly, a Spatial Perception (SP) module is designed to alleviate information loss; the model’s parameter count is reduced and the vanishing feature problem of small targets is mitigated via the complementation of channel and spatial information. Secondly, a dedicated Component Spatial Attention (CSA) module is developed for the significant differences between drones and birds in infrared images, so that the model’s ability to distinguish the two targets is enhanced. Experimental tests were conducted, and the results show that the improved model achieves 96.5% in mAP@0.5 and 52.3% in mAP@0.5:0.95, with an increase of 0.7% and 3.0% respectively compared with the original YOLOv8n model. An effective balance between detection accuracy and parameter count is realized by the proposed model, and the most remarkable improvement is improved most remarkably.