摘要
面对雨天场景中语义模糊与密集雨滴造成的遮挡问题,文章在YOLOv11框架上引入空间辅助通道注意力机制进行改进。首先,通过多语义特征分解充分挖掘目标在不同语义层次上的特征表达;其次,结合空间注意力机制从模糊及受干扰图像中提取关键特征,减弱雨滴与反光噪声对特征提取的影响;再次,引入通道自注意力机制以降低通道间语义差异,从而增强目标与背景的区分能力;最后,通过协作融合机制整合多层特征,提高整体检测性能。实验结果表明,改进算法在雨天场景下的mAP@50和mAP@50-95分别达到27.5%与16.2%,较原YOLOv11模型提高1.4%和0.7%。为进一步验证该算法的普适性,文章在PASCAL VOC数据集上对YOLO系列模型进行了对比。结果显示,改进算法在20个类别中的18个类别上获得更优检测性能,证明其在复杂雨天环境下具有更高的鲁棒性与适用性。
In the face of semantic ambiguity and occlusion caused by dense raindrops in rainy weather scenes,this article introduces a spatial assisted channel attention mechanism on the YOLOv11 framework for improvement.Firstly,fully explore the feature expressions of the target at different semantic levels through multi semantic feature decomposition.Secondly,by combining spatial attention mechanisms to extract key features from blurry and disturbed images,the impact of raindrops and reflective noise on feature extraction can be reduced.Again,introducing channel self attention mechanism to reduce semantic differences between channels,thereby enhancing the ability to distinguish between targets and backgrounds.Finally,by integrating multiple layers of features through collaborative fusion mechanisms,the overall detection performance is improved.The experimental results show that the improved algorithm performs well in rainy weather scenarios mAP@50 and mAP@50-95 Reaching 27.5%and 16.2%respectively,an increase of 1.4%and 0.7%compared to the original YOLOv11 model.To further validate the universality of the algorithm,the article compared the YOLO series models on the PASCAL VOC dataset.The results showed that the improved algorithm achieved better detection performance in 18 out of 20 categories,demonstrating its higher robustness and applicability in complex rainy environments.
作者
郝培男
张颖
宁慧聪
闫田子
胡旭如
HAO Peinan;ZHANG Ying;NING Huicong;YAN Tianzi;HU Xuru(School of Computer Science and Information Engineering,Anyang Institute of Technology,Anyang,Henan 455000,China;College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China;Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology,Tianjin 300350,China;Beijing Xiaomi Mobile Software Co.,Ltd.,Beijing 100085,China)
基金
安阳工学院博士启动基金项目:基于大数据与知识图谱的可解释人工智能研究(BSJ2023012)
教育部实验教学和教学实验室建设研究项目:基于“三四三”模式的电子信息类实验实践教学体系改革与实践(SYJX2024-022)
安阳工学院2025年校级“十百千"项目:低空经济背景下智能机器人研究型教学创新实践平台
2025年安阳市人民政府调研课题:新质生产力驱动下安阳市民营经济高质量发展路径研究
2025年安阳市人民政府调研课题:大数据与人工智能赋能安阳市农业经济高质量发展路径研究
安阳工学院2025年度“百千万英才提升计划(2.0)项目:产教学研融合视域下人工智能创业复合型人才培育模式研究。
