基于卷积神经网络的目标检测算法
基于卷积神经网络的目标检测算法(论文15000字)
摘要:本文是基于卷积神经网络的目标检测算法,区别于传统的目标检测方法,它是一种基于深度学习的检测方法,可以自主学习提取海量数据的特征,且由于卷积神经网络本身参数共享的特性,使得减少了人的工作量,大大提高了运算效率。目前存在各种版本的基于卷积神经网络的目标检测算法,诸如RCNN、Fast RCNN以及Faster RCNN等,本文主要是针对Faster RCNN算法进行配置,循序渐进尝试各种方法进行实验。最终,实验证实,相比传统的目标检测方法,基于卷积神经网络的目标检测算法有着更好的检测精度,以及更高的检测效率,但也有不足之处。
关键词:深度学习;卷积神经网络;目标检测
Convolution Neural Network Features for Object Detection
Abstract:This paper is based on the convolution neural network target detection algorithm, different from the traditional target detection method.It is a kind of detection method based on depth learning, which can learn and extract the characteristics of massive data.And because of the characteristics of the parameters shared by the convolution neural network, it reduces the human workload and greatly improves the operation efficiency.At present, there are various versions of target detection algorithms based on convolution neural networks, such as RCNN, Fast RCNN and Faster RCNN. In this paper, we focus on the Faster RCNN algorithm, and try various methods to conduct experiment.Atlast, it is proved that the target detection algorithm based on convolution neural network has better detection accuracy and higher detection efficiency than traditional target detection method.But there are also shortcomings [资料来源:Doc163.com]
Key words:Depth study;Convolution neural network;Target Detection
目录
1引言 4
1.1 研究背景及意义 4
1.2 国内外研究现状 4
1.3 基于卷积神经网络的目标检测算法 4
2关于卷积神经网络 5
2.1神经网络 5
2.2卷积神经网络简介 7
2.2.1数据输入层 7
2.2.2卷积计算层 8
2.2.3激励层 8
2.2.4池化层 9
2.2.5全连接层 9
3基于卷积神经网络的几种目标检测算法的比较及相关拓展 10
3.1基于卷积神经网络的目标检测算法 10
3.1.1关于RCNN 10
3.1.2关于SPPNet————对rcnn的一次提速 13
3.2FAST-RCNN 15
3.2.1关于RoI pooling layer 16
3.2.2关于预训练 16 [资料来源:http://doc163.com]
3.2.3关于微调(Fine-tuning) 16
3.2.4关于多任务损失 16
3.2.5关于小批量抽样 17
3.3 FASTER-RCNN 17
3.3.1关于RPN 19
3.3.2训练RPN 20
3.4基于回归方法的深度学习目标检测算法————YOLO和SSD 21
3.4.1YOLO 21
3.4.2SSD 22
4关于FASTER-RCNN的实验 22
4.1实验配置 22
4.2实验结果 23
4.2.1设计的网络结构1 23
4.2.2 参考AlexNet,VGG-Net,ResNet设计第二个网络结构 24
4.2.3将Relu改为PRelu 26
4.2.4增加Batch Normalization层 26
4.2.5引入Dropout操作 27
4.2.6使用ZFNet网络模型 27
5总结 35
参考文献 35
致谢 36