AMFF-YOLO: Low-Light Object Detection Method Based on Adaptive Multi-Scale Feature Fusion

In low-light environments, images captured by color cameras often suffer from severe degradation, including low contrast, strong noise, and blurred edges. This paper presents a low-light object detection method based on adaptive multiscale feature fusion. An adaptive feature enhancement module, guided by a chain-of-thought prompting mechanism, is integrated to mitigate the effects of degraded visual information. A crossstage spatial fusion module is introduced to refine shallow features through precise spatial alignment, effectively suppressing background interference and improving edge discrimination. Furthermore, a bidirectional cross-hierarchical multi-branch enhanced pyramid structure is designed to enable bidirectional information flow, thereby enhancing both global perception and local detail representation. Experimental results on multiple lowlight datasets confirm the effectiveness of the proposed method. On the ExDark dataset, it improves precision by 1.1%, recall by 0. 5%, mAP@ 0.5 by 1.8%, and mAP@ 0.5: 0.95 by 1.9%. On the DarkFace dataset, mAP @ 0.5 and mAP @ 0.5: 0.95 are improved by 17.5% and 8.5%, respectively. On the Nikon dataset, gains of 5.4% and 3.8% are achieved. These results demonstrate the method's strong performance and robustness in low-light object detection scenarios.