Ren Bin, Xu Zhaohui, Zhao Junwu, Hao Rujiang, Zhang Jianchao
School of Mechanical Engineering, Shijiazhuang Tiedao University, Shijiazhuang, 050043, Hebei, China.
Sci Rep. 2025 Jul 1;15(1):21873. doi: 10.1038/s41598-025-07348-0.
For object detection in complex dark environments, the existing methods generally have problems such as low detection accuracy, false detection and missed detection. These problems lead to a lack of key information and incomplete context information, which seriously affects the detection effect. Therefore, this paper proposes an object detection method for complex dark environment based on YOLO-AS. First, the Zero-DCES image enhancement module is designed to improve the image quality of the dark environment via adaptive contrast enhancement. Second, a YOLO-AS detection model is constructed, which integrates ECA_ASPP and the SK attention mechanism. The receptive field is expanded by dilated convolution, and dynamic feature detection is realized by combining channel attention, which effectively enhances the ability of multiscale feature expression. Finally, the model is tested on the ExDark dataset and the LOL dataset, and compared with current mainstream models such as YOLOv5s, YOLOv8n and YOLOv11. Experiments show that the proposed method achieves 78.39% map@50 on the ExDark dataset while maintaining a basically unchanged detection speed, which is 5.78% higher than that of the benchmark model, and significantly improves the detection accuracy in complex dark environments.
对于复杂黑暗环境中的目标检测，现有方法普遍存在检测精度低、误检和漏检等问题。这些问题导致关键信息缺失和上下文信息不完整，严重影响检测效果。因此，本文提出一种基于YOLO-AS的复杂黑暗环境目标检测方法。首先，设计了Zero-DCES图像增强模块，通过自适应对比度增强来提高黑暗环境的图像质量。其次，构建了YOLO-AS检测模型，该模型集成了ECA_ASPP和SK注意力机制。通过空洞卷积扩大感受野，并结合通道注意力实现动态特征检测，有效增强了多尺度特征表达能力。最后，在ExDark数据集和LOL数据集上对该模型进行测试，并与YOLOv5s、YOLOv8n和YOLOv11等当前主流模型进行比较。实验表明，该方法在ExDark数据集上实现了78.39%的map@50，同时检测速度基本保持不变，比基准模型高5.78%，显著提高了复杂黑暗环境下的检测精度。
