《计算机应用》唯一官方网站

• •    下一篇

基于边界极限点特征的改进YOLOv3目标检测

李克文杨建涛*,黄宗超   

  1. 中国石油大学(华东) 计算机科学与技术学院,山东 青岛 266580
  • 收稿日期:2021-11-24 修回日期:2022-03-16 接受日期:2022-04-01 发布日期:2022-06-17 出版日期:2022-06-17
  • 通讯作者: 杨建涛
  • 作者简介:李克文(1969—),男,山东东营人,教授,博士,CCF高级会员,主要研究方向:人工智能、机器学习、数据挖掘;杨建涛(1998—),男,甘肃武威人,硕士研究生,主要研究方向:计算机视觉、图像处理、深度学习;黄宗超(1994—),男,山东聊城人,博士研究生,主要研究方向:深度学习、大数据处理、故障智能检测。

Improved YOLOv3 target detection based on boundary limit point features

  • Received:2021-11-24 Revised:2022-03-16 Accepted:2022-04-01 Online:2022-06-17 Published:2022-06-17

摘要: 在目标检测场景中存在目标数量众多、目标尺度较小、目标高度重叠等问题致使目标检测精度低、难度大,为了提升目标检测的精度,尽可能避免漏检误检情况,提出了一种基于边界极限点特征的改进YOLOv3目标检测算法。首先,引入一种边界增强算子Border,从边界的极限点中自适应地提取边界特征来增强已有点特征,提高目标定位准确度;然后,通过增加目标检测尺度,细化特征图,增强特征图深、浅层语义信息融合,进一步提高目标检测精度;最后,基于目标检测中目标实例特性及改进网络模型引入完全交并比(CIoU)函数对原YOLOv3损失函数进行改进,提高检测框收敛速度以及检测框召回率。实验结果表明,改进后的YOLOv3目标检测算法与原YOLOv3目标检测算法相比,检测精度有较大提升,且检测速度与原算法相近,能有效提高模型对目标的检测能力。

关键词: 目标检测, YOLOv3算法, 细化特征图, 多尺度检测, 损失函数

Abstract: In the target detection scene, there are a large number of targets, small target scales, and high overlap of targets, which make target detection low accuracy and difficult. In order to improve the accuracy of target detection and avoid missed detection and false detection as much as possible, a new improved You Only Look Once version3(YOLOv3) target detection algorithm based on boundary limit point features was proposed. Firstly, a boundary enhancement operator Border was introduced, which adaptively extracted boundary features from the limit points of the boundary to enhance the point features and improve the accuracy of target positioning. Then, the accuracy of target detection was further improved by increasing the target detection scale, refining the feature map, and enhancing the fusion of the feature image deep and shallow semantic information. Finally, based on the target instance characteristics in target detection and the improved network model, the Complete Intersection over Union (CIoU) function was introduced to improve the original YOLOv3 loss function to improve the convergence speed and recall rate of the detection frame. Experimental results show that, the improved YOLOv3 target detection algorithm has a greater detection accuracy compared with the original YOLOv3 target detection algorithm, and the detection speed is similar to the original algorithm, which can effectively improve the target detection ability of models.

Key words: target detection, You Only Look Once version3 (YOLOv3) algorithm, refinement feature map, multi-scale detection, loss function

中图分类号: