计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (20): 232-242.DOI: 10.3778/j.issn.1002-8331.1908-0121

• 工程与应用 • 上一篇    下一篇

改进YOLOv3网络在图像中评价空气质量

邓益侬,罗健欣,张琦,刘祯,胡琪,金凤林,毕鹏程   

  1. 1.陆军工程大学 指挥控制工程学院,南京 210007
    2.南京小吉狗网络科技有限公司,南京 210000
  • 出版日期:2020-10-15 发布日期:2020-10-13

Improving YOLOv3 Network to Assess Air Quality in Image

DENG Yinong, LUO Jianxin, ZHANG Qi, LIU Zhen, HU Qi, JIN Fenglin, BI Pengcheng   

  1. 1.College of Command & Control Engineering, Army Engineering University of PLA, Nanjing 210007, China
    2.Nanjing Lucky Dog Network Technology Co., Ltd., Nanjing 210000, China
  • Online:2020-10-15 Published:2020-10-13

摘要:

空气质量指数和PM2.5浓度是衡量大气污染程度的两种重要指标。在图像检测算法的框架下,提出了一种基于YOLOv3网络模型的空气质量评价方法。该模型对YOLOv3网络进行了改进,包含多锚点检测机制和卷积投票网两个模块,适用于非固定场景下的空气质量指数评估,以及固定场景下的PM2.5浓度预测。该方法的准确率在2018年全球人工智能应用大赛中得到了总分第3名的成绩,同时基于darknet框架的YOLOv3模型可以达到实时的需求,对空气质量评价的相关研究具有重要的借鉴意义。

关键词: 空气质量, PM2.5, 图像检测, 多锚点, 卷积投票

Abstract:

Air quality index and PM2.5 concentration are two important indicators to measure the degree of air pollution. In the framework of image detection algorithm, this paper proposes an air quality assessment method based on YOLOv3 network model. The YOLOv3 network is improved in the model, which includes two modules:multi-anchor detection mechanism and convolution voting network. The model is suitable for air quality index evaluation in non-fixed scenarios and PM2.5 concentration prediction in fixed scenarios. The accuracy of the method described in this paper has been ranked 3rd in the global AI application contest in 2018. At the same time, the YOLOv3 model based on Darknet framework can meet the real-time requirements, which has important reference significance for the related research of air quality assessment.

Key words: air quality, PM2.5, image detection, multi anchors, convolutional voting