深度学习典型目标检测算法的改进综述  

Review on Improvement of Typical Object Detection Algorithms in Deep Learning

作　　者：王鑫鹏 王晓强[1] 林浩 李雷孝 杨艳艳 孟闯 高静[4] WANG Xinpeng;WANG Xiaoqiang;LIN Hao;LI Leixiao;YANG Yanyan;MENG Chuang;GAO Jing(College of Information Engineering,Inner Mongolia University of Technology,Hohhot 010080,China;College of Computer Science and Engineering,Tianjin University of Technology,Tianjin 300384,China;College of Data Science and Application,Inner Mongolia University of Technology,Hohhot 010080,China;College of Computer and Information Engineering,Inner Mongolia Agricultural University,Hohhot 010011,China)

机构地区：[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

出　　处：《计算机工程与应用》2022年第6期42-57,共16页Computer Engineering and Applications

基　　金：内蒙古自治区关键技术攻关计划(2019GG273);内蒙古自治区科技成果转化专项(2020CG0073,2021CG0033);内蒙古自治区科技重大专项(2019ZD015,2019ZD016);内蒙古自治区科技计划(2020GG0104)。

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。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.

关 键 词：深度学习 目标检测 数据增强 网络结构 损失计算 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]