摘要
传统的目标检测算法及策略已经难以满足目标检测中数据处理的效率、性能、速度和智能化等各个方面要求。深度学习通过对大脑认知能力的研究和模仿以实现对数据特征的分析处理,具有强大的视觉目标检测能力,成为了当前目标检测的主流算法。首先回顾了传统目标检测的发展以及存在的问题;其次介绍以R-CNN为代表的结合region proposal和卷积神经网络(CNN)分类的目标检测框架(R-CNN、SPP-NET、Fast R-CNN、Faster R-CNN);然后介绍以YOLO算法为代表的将目标检测转换为回归问题的目标检测框架(YOLO、SSD);最后对深度学习的目标检测算法存在的问题做出总结,以及未来的发展方向。
The traditional target detection algorithm and strategy has been difficult to meet the target detection of data processing efficiency, performance, speed and intelligence and other aspects. Depth learning through the study of brain cognitive ability and imitation to achieve the analysis of data characteristics of the treatment,with a strong visual target detectioncapabilities, has become the current target detection of the mainstream algorithm. Firstly, the developmentand problems of traditional target deteetionare reviewed; Secondly, the target detection framework which combines region proposal and CNNclassification with R-CNN is introduced（R-CNN, SPP-NET, Fast R-CNN, Faster R-CNN） ; Then, the target detection framework is introduced, which is based on YOLO（YOLO, SSD）algorithm;Finally, this paper makes a summary of the problemsexisting in the target detection algorithm of deep learning and the development of the future.
出处
《电子测量技术》
2017年第11期89-93,共5页
Electronic Measurement Technology
基金
国家自然科学基金(61201444)资助
关键词
深度学习
卷积神经网络
目标检测
deep learning
convolutional neural network
object detection
