Journal of Computer Applications ›› 2020, Vol. 40 ›› Issue (8): 2225-2230.DOI: 10.11772/j.issn.1001-9081.2020010030

• Artificial intelligence • Previous Articles     Next Articles

Object detection of Gaussian-YOLO v3 implanting attention and feature intertwine modules

LIU Dan1, WU Yajuan1, LUO Nanchao2, ZHENG Bochuan3   

  1. 1. School of Computer Science, China West Normal University, Nanchong Sichuan 637002, China;
    2. School of Computer Science and Technology, Aba Teachers University, Aba Sichuan 623002, China;
    3. School of Mathematics and Information, China West Normal University, Nanchong Sichuan 637002, China
  • Received:2020-01-16 Revised:2020-04-01 Online:2020-08-10 Published:2020-04-09
  • Supported by:
    This work is partially supported by the Sichuan Science and Technology Program (2019YFG0299), the Fundamental Research Project of China West Normal University (19B045), the Innovation and Entrepreneurship Project of College Students of China West Normal University (cxcy2018305).

嵌入注意力和特征交织模块的Gaussian-YOLO v3目标检测

刘丹1, 吴亚娟1, 罗南超2, 郑伯川3   

  1. 1. 西华师范大学 计算机学院, 四川 南充 637009;
    2. 阿坝师范学院 计算机科学与技术学院, 四川 阿坝 623002;
    3. 西华师范大学 数学与信息学院, 四川 南充 637009
  • 通讯作者: 郑伯川(1974-),男,四川自贡人,教授,博士,CCF会员,主要研究方向:机器学习、深度学习、计算机视觉。
  • 作者简介:刘丹(1996-),女,四川广安人,硕士研究生,CCF会员,主要研究方向:深度学习、目标检测;吴亚娟(1974-),女,四川大竹人,教授,博士,主要研究方向:图像处理、数值计算;罗南超(1975-),男,四川富顺人,教授,主要研究方向:云计算、数据分析与挖掘、图形图像处理。
  • 基金资助:

Abstract: Wrong object detection may lead to serious accidents, so high-precision object detection is very important in autonomous driving. An object detection method of Gaussian-YOLO v3 combining attention and feature intertwine module was proposed, in which several specific feature maps were mainly improved. First, the attention module was added to the feature map to learn the weight of each channel autonomously, enhancing the key features and suppressing the redundant features, so as to enhance the network ability to distinguish foreground object and background. Second, at the same time, different channels of the feature map were intertwined to obtain more representative features. Finally, the features obtained by the attention and feature intertwine modules were fused to form a new feature map. Experimental results show that the proposed method achieves mAP (mean Average Precision) of 20.81% and F1 score of 18.17% on BDD100K dataset, and has the false alarm rate decreased by 3.5 percentage points, reducing the false alarm rate effectively. It can be seen that the detection performance of the proposed method is better than those of YOLO v3 and Gaussian-YOLO v3.

Key words: Gaussian-YOLO v3, attention mechanism, feature intertwine, autonomous driving, object detection

摘要: 错误的目标检测可能导致严重事故,因此高精度的目标检测在汽车自动驾驶中至关重要。提出了一种嵌入注意力和特征交织模块的Gaussian-YOLO v3目标检测方法。该方法主要对Gaussian-YOLO v3的几个特定特征图进行了改进:首先在特征图中添加注意力模块以自主学习每个通道的权重,增强关键特征、抑制冗余特征,从而加强网络对前景目标和背景的区分能力;其次,同时将特征图的不同通道进行特征交织得到更具表征性的特征;最后,把注意力和特征交织模块分别得到的特征融合构成新的特征图。实验结果表明,所提方法在BDD100K数据集上达到了20.81%的平均精确率均值(mAP)和18.17%的F1分数,使误报率减少了3.5%,意味着误报率得到了有效降低。由此可见,所提方法的检测性能优于YOLO v3和Gaussian-YOLO v3。

关键词: Gaussian-YOLO v3, 注意力机制, 特征交织, 自动驾驶, 目标检测

CLC Number: