江汉大学学报(自然科学版) ›› 2023, Vol. 51 ›› Issue (1): 89-96.doi: 10.16389/j.cnki.cn42-1737/n.2023.01.011

• 计算机科学 • 上一篇    

基于改进 YOLOv3 的道路目标检测

朱仕宁,胡晓斌,彭太乐*   

  1. 淮北师范大学 计算机科学与技术学院,安徽 淮北 235000
  • 发布日期:2023-02-21
  • 通讯作者: 彭太乐
  • 作者简介:朱仕宁(1997— ),男,硕士生,研究方向:深度学习、目标检测。*通信作者:彭太乐(1974— ),男,教授,博士,研究方向:机器学习、图像处理、智能信息处理等。

Discussion on Present Situation and Patterns of Greening Under Viaducts in Wuhan City

ZHU Shining,HU Xiaobin,PENG Taile*   

  1. School of Computer Science and Technology,Huaibei Normal University,Huaibei 235000,Anhui,China
  • Published:2023-02-21
  • Contact: PENG Taile

摘要: 针对 YOLOv3 在道路目标检测中漏检率高和检测精度低的问题,提出一种基于改进YOLOv3 的道路目标检测方法。通过将原有 YOLOv3 的 3 个特征尺度增至 4 个,从而提升了对于小目标的检测准确率。使用 CIoU 损失函数提高模型的准确性,利用 K-Means++聚类算法对道路目标重新聚类,得到新的候选框。在 BDD100K 数据集上的验证结果表明,改进的 YOLOv3算法在降低漏检率和提高检测精度方面效果较好。

关键词: 道路目标检测, YOLOv3, K-Means++

Abstract: Aiming at the problems of high missed detection rate and low detection accuracy of YOLOv3 in road target detection,this paper proposed a road target detection method based on improved YOLOv3. By increasing the three feature scales of the original YOLOv3 to four,the object accuracy for small targets was improved. Secondly,we used the CIoU loss function to improve the model's accuracy and the K-Means++ clustering algorithm to recluster the road targets to get new candidate boxes. This paper verified the effect of the improved YOLOv3 algorithm on the BDD100K data set. The experimental results showed that the improved YOLOv3 algorithm had achieved good detection results in reducing the missed detection rate and improving the detection accuracy.

Key words: road target detection, YOLOv3, K-Means++

中图分类号: