摘要
为了提升目标检测的准确性和效率,基于YOLOv7提出了一种多维注意力机制融合的目标检测网络(MAF-Net)。在网络主干部分,采用ODConv替代传统卷积堆叠,融入多维注意力机制并行策略,提升了特征提取能力。在头部网络部分,插入SimAM注意力机制,通过计算特征图的3-D关注权重来优化检测精度。同时,结合Inner-IoU与CIoU,提高了模型精度并优化了训练效率。经MS COCO 2017数据集实验验证,MAF-Net的mAP相较于原模型提升1.3%。综上所述,提出的MAF-Net模型在目标检测领域展现出了优异的性能,具有良好的识别精度、精细的特征提取以及高效的训练效率。
In order to improve the accuracy and efficiency of object detection,a multi-dimensional attention mechanism fusion object detection network(MAF Net)based on YOLOv7 is proposed.In the backbone of the network,ODConv is used instead of traditional convolutional stacking,and a multi-dimensional attention mechanism parallel strategy is incorporated to improve the feature extraction ability.In the head network section,SimAM attention mechanism is inserted to optimize detection accuracy by calculating the 3D attention weight of the feature map.At the same time,combining Inner IoU and CIoU improved model accuracy and optimized training efficiency.After experimental verification on the MS COCO 2017 dataset,the mAP of MAF-Net increased by 1.3%compared to the original model.In summary,the MAF-Net model proposed in this article has shown excellent performance in the field of object detection,with good recognition accuracy,fine feature extraction,and efficient training efficiency.
作者
吴宇杰
王挺
邵士亮
曹风魁
刘军
WU Yujie;WANG Ting;SHAO Shiliang;CAO Fengkui;LIU Jun(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang 110159,China;State Key Laboratory of Robotics,Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110169,China)
出处
《组合机床与自动化加工技术》
北大核心
2025年第5期71-77,共7页
Modular Machine Tool & Automatic Manufacturing Technique
基金
国家自然科学基金联合基金项目(U20A20201)。
