摘要
针对在雾天条件下能见度降低和特征信息的丢失,行人车辆检测效果差的问题,提出了一种基于Yolov8改进的I-Yolov8网络模型。首先引入双向加权特征融合金字塔网络并新增了小目标检测头,以增强对多尺度目标的检测能力更有效地捕捉雾天条件下小目标的特征;其次对空间金字塔池化层进行优化,以提升模型对不同分辨率特征的整合能力;模型还集成了ECA注意力机制,以增强模型对图像关键特征的提取;最后采用WIOUV3损失函数以提高目标边界框的预测精度。在RTTS真实雾天场景数据集上的实验结果表明,改进模型的平均准确率达到76.6%,相较于基线模型提升了5.6%,同时保持了125的帧率,实现了检测精度与速度的平衡。
To address the issue of poor pedestrian and vehicle detection performance under foggy conditions due to reduced visibility and loss of feature information,we propose an improved I-Yolov8 network model based on YOLOv8.Initially,a bidirectional weighted feature fusion pyramid network is introduced,and a small object detection head is newly added to enhance the detection capability for multi-scale targets and more effectively capture the features of small targets under foggy conditions.Subsequently,the spatial pyramid pooling layer is optimized to improve the model’s ability to integrate features of different resolutions.The model also integrates the ECA mechanism to enhance the model’s focus on key image features.Finally,the WIOUV3 loss function is employed to improve the accuracy of object bounding box predictions.Experimental results on the RTTS foggy scene dataset demonstrate that the improved model achieves an average accuracy of 76.6%,a 5.6%increase over the baseline model,while maintaining a frame rate of 125,thus balancing detection accuracy and speed.
作者
赵伟康
李军
ZHAO Weikang;LI Jun(Anhui University of Science and Technoloy,Huainan 232000,China)
出处
《长江信息通信》
2025年第1期91-95,共5页
Changjiang Information & Communications
关键词
雾天
目标检测
金字塔网络
金字塔池化
注意力机制
Fog
Object Detection
Pyramid Network
Spatial Pyramid Pooling
Attention Mechanism
作者简介
赵伟康(1999-),男,湖南永州人,硕士,研究方向:目标检测;李军(1997-),男,安徽安庆人,硕士,研究方向:目标检测。
