摘要
目标检测一直是计算机视觉领域中的研究热点。随着深度学习技术的迅猛发展,基于卷积神经网络的目标检测模型逐渐被广泛关注。文中主要对基于卷积神经网络的目标检测模型的现状进行综述。首先,介绍了目标检测的相关基础,特别罗列了一些目标检测模型中常用的卷积神经网络结构,也介绍了检测模型常用的梯度下降法训练方式。然后,重点从候选区域和回归方法两类对近年来提出的优秀模型进行综述,候选区域一类也创新地使用特征尺度进行区分,说明了多尺度特征能够有效提高小尺度目标检测精度。对于每一类检测模型,根据同一数据集上的检测结果分析这些模型的优势与缺陷,最后根据分析的结果总结一些基于卷积神经网络的目标检测模型的优化方案。
Object detection has always been a research hotspot in the field of computer vision.With the rapid development of deep learning technology,the object detection model based on convolutional neural network is widely concerned.We mainly review the current status of object detection models based on convolutional neural networks.First of all,we introduce the relevant basis of target detection,especially the convolutional neural network structure commonly used in some object detection models,and also introduce the gradient descent training method commonly used in detection models.Then,we summarize the excellent models proposed in recent years from region-based and region-free and compare the test results.The region-based models are distinguished with feature scales intelligently,which shows that multi-scale features can effectively improve the accuracy of small-scale object detection.For each type of detection model,we analyze the advantages and disadvantages of these models based on the results on the same data set.Finally,based on the analysis results,some optimization schemes based on the convolutional neural network are proposed.
作者
许必宵
宫婧
孙知信
XU Bi-xiao;GONG Jing;SUN Zhi-xin(School of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Key Laboratory of Broadband Wireless Communication and Sensor Network Technology,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;School of Modern Posts,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)
出处
《计算机技术与发展》
2019年第12期87-92,共6页
Computer Technology and Development
基金
国家自然科学基金(61373135)
江苏省研究生科研与实践创新计划项目(KYCX17_0775)
关键词
卷积神经网络
目标检测
深度学习
计算机视觉
convolutional neural network
object detection
deep learning
computer vision
作者简介
许必宵(1993-),男,在读硕士,工程师,研究方向为目标检测技术;宫婧,博士,副教授,研究方向为深度学习、计算机视觉等;通信作者:孙知信,博士后,教授,研究方向为信息安全、人工智能与计算机视觉。
