Abstract: To address the high computational complexity and poor detection performance of small targets in infrared scenarios, an improved target-detection algorithm, NFZ-YOLOv10, is proposed. By introducing the lightweight network StarNet, the computational load and parameter count of the model were reduced, and the feature extraction capability was enhanced. The neck network was optimized, and the slim-neck paradigm was constructed using the GSConv and VOVGSCSP modules, achieving a deep integration of high-level semantic information and fine-grained spatial features. The NWD loss function was introduced to effectively alleviate the sensitivity of traditional loss functions to positional deviations in small-scale target-detection tasks, thereby significantly improving the detection accuracy of the model. The experimental results showed that the mean average precision of NFZ-YOLOv10 on the self-built dataset reached 94.5%, an increase of 2.7% compared to YOLOv10n. Additionally, the computational cost and the number of parameters of NFZ-YOLOv10 have been reduced by 24.4% and 22.3%, respectively, compared to YOLOv10n. The single-frame image detection time is as low as 4.5 ms, and the model size has been shrunk to 4.9 MB. In summary, NFZ-YOLOv10 achieves a more effective balance between detection accuracy and complexity and has strong application potential in the detection of low-slow-small targets within airport control zones.