Unified Enhancement-detection Frameworks for Higher Precision in Low-light Object Recognition

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

Low-light conditions pose a critical challenge to modern computer vision systems, as diminished illumination often leads to significant drops in detection accuracy.With the increasing demand for reliable visionbased solutions in safety-critical domains, enhancing object detection performance in such environments has become a pressing research priority.In the field of computer vision, effective detection of objects often relies on sufficient illumination, a condition frequently compromised in low-light environments.This study addresses the challenges associated with low-light object detection by introducing an enhanced YOLOv8 model tailored for such conditions.By integrating the ExDark dataset, which is specifically curated for low-light scenarios, and incorporating a dual-pool spatial attention mechanism integrated in the YOLOv8 neck, the model reorganises feature accumulation to diminish the impact of irrelevant features while emphasising critical data.This refined attention strategy improves the detection of small and occluded objects.Experimental results demonstrate significant improvements over the classical YOLOv8 model, achieving a mean Average Precision (mAP) of 0.572, precision of 0.597 and recall of 0.56.These findings validate the efficacy of the approach in enhancing detection accuracy and flexibility in low-light conditions, establishing a strong benchmark for future research and paving the way for real-time applications in surveillance, autonomous vehicles, and cybernetics.

Links and sources

• PDF from original source

Comments