Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (6): 42-57.DOI: 10.3778/j.issn.1002-8331.2110-0070

• Research Hotspots and Reviews • Previous Articles     Next Articles

Review on Improvement of Typical Object Detection Algorithms in Deep Learning

WANG Xinpeng, WANG Xiaoqiang, LIN Hao, LI Leixiao, YANG Yanyan, MENG Chuang, GAO Jing   

  1. 1.College of Information Engineering, Inner Mongolia University of Technology, Hohhot 010080, China
    2.College of Computer Science and Engineering, Tianjin University of Technology, Tianjin 300384, China
    3.College of Data Science and Application, Inner Mongolia University of Technology, Hohhot 010080, China
    4.College of Computer and Information Engineering, Inner Mongolia Agricultural University, Hohhot 010011, China
  • Online:2022-03-15 Published:2022-03-15



  1. 1.内蒙古工业大学 信息工程学院,呼和浩特 010080
    2.天津理工大学 计算机科学与工程学院,天津 300384
    3.内蒙古工业大学 数据科学与应用学院,呼和浩特 010080
    4.内蒙古农业大学 计算机与信息工程学院,呼和浩特 010011

Abstract: Object detection is one of the most challenging tasks in the field of machine vision, and deep learning is the most mainstream implementation method for object detection. In recent years, the rapid development of deep learning theory and technology has made great progress in object detection algorithms based on deep learning. Scholars have started from data processing, network structure, loss function and other aspects, and a series of improved methods are proposed for object detection algorithms. This article reviews the improvement methods of typical object detection algorithms. The commonly used data sets and performance evaluation indicators are summarized, and the characteristics, advantages and application fields of the data sets are compared. It sorts out the latest improvement ideas of typical deep learning-based object detection algorithms, and discusses, summarizes and compares and analyzes data augmentation, anchor selection, network model construction, prediction anchor selection and loss calculation. Combined with the existing problems, the future research direction of typical object detection algorithms based on deep learning is prospected.

Key words: deep learning, object detection, data augmentation, network structure, loss calculation

摘要: 目标检测是机器视觉领域内最具挑战性的任务之一,深度学习则是目标检测最主流的实现方法。近年来,深度学习理论及技术的快速发展,使得基于深度学习的目标检测算法取得了巨大进展,学者从数据处理、网络结构、损失函数等多方面入手,提出了一系列对于目标检测算法的改进方式。针对典型目标检测算法的改进方式进行综述。归纳了常用数据集和性能评价指标,并对数据集的特点、优势及应用领域进行了对比。梳理了典型的基于深度学习的目标检测算法的最新改进思路,从数据增强、先验框选择、网络模型的构建、预测框的选取及损失计算几个方面分别进行论述、总结与对比分析。结合当前存在的问题,展望了基于深度学习的典型目标检测算法的未来研究方向。

关键词: 深度学习, 目标检测, 数据增强, 网络结构, 损失计算