家电科技 ›› 2023, Vol. 0 ›› Issue (zk): 115-121.doi: 10.19784/j.cnki.issn1672-0172.2023.99.028

• 第一部分 优秀论文 • 上一篇    下一篇

基于改进的YOLOv5人体目标检测算法研究及应用

张裕松1, 毛跃辉1,2, 梁博1, 陶梦春1   

  1. 1.珠海格力电器股份有限公司 广东珠海 519070;
    2.珠海艾维普信息技术有限公司 广东珠海 519070
  • 出版日期:2023-12-12 发布日期:2023-12-26
  • 通讯作者: 毛跃辉,E-mail:happy200521@163.com。
  • 作者简介:张裕松,硕士学位。研究方向:图像识别技术研究与产品应用。地址:广东省珠海市香洲区前山金鸡西路789号。E-mail:qingshanyusong@163.com。

Research and application of improved YOLOv5 human target detection algorithm

ZHANG Yusong1, MAO Yuehui1,2, LIANG Bo1, TAO Mengchun1   

  1. 1. Gree Electric Appliances, Inc. of Zhuhai Zhuhai 519070;
    2. Zhuhai Avipu Information Technology Co., Ltd. Zhuhai 519070
  • Online:2023-12-12 Published:2023-12-26

摘要: 空调智能化的场景应用离不开高性能的人体目标检测算法,因而提高红外人体目标的检测准确率以及定位精度是必要的。首先,对YOLOv5进行改进,分别引入SIOU损失函数与CA注意力机制,提高了模型对目标的检测能力。其次,针对家用空调的场景需求和红外图像的特点,设计了一套针对性强的数据增强方法,进一步优化了模型的训练效果。最后,将改进模型应用至空调样机的实际场景识别中,结果表明:改进后的YOLOv5模型在人体目标检测准确率和定位精度方面优于传统方法,具有一定的实用价值。

关键词: 红外传感器, YOLOv5, 算法优化, 数据增强, 空调

Abstract: The application of intelligent air conditioning is inseparable from high-performance human target detection algorithm, so it is necessary to improve the detection accuracy and positioning accuracy of infrared human target. First, YOLOv5 is improved, and SIOU loss function and CA attention mechanism are introduced respectively to improve the detection ability of the model. Secondly, according to the scene requirements of household air conditioning and the characteristics of infrared images, a set of targeted data enhancement method is designed to further optimize the training effect of the model. Finally, the improved model is applied to the actual scene recognition of the air conditioning prototype. The results show that the improved YOLOv5 model is superior to the traditional methods in human target detection accuracy and positioning accuracy, and has certain practical value.

Key words: Infrared sensor, YOLOv5, Algorithm optimization, Data enhancement, Air conditioning

中图分类号: