基于改进的特征提取网络的目标检测算法 下载: 961次
Object Detection Algorithm Based on Improved Feature Extraction Network
天津大学电气自动化与信息工程学院, 天津 300072
图 & 表
图 1. 4种数据增强技巧效果对比图
Fig. 1. Comparison of effects of four data augmentation techniques
下载图片 查看原文
图 2. 翻转、裁剪、旋转方法组合效果对比图
Fig. 2. Comparison of combination effects of flipping, cropping, and rotating methods
下载图片 查看原文
图 3. ResNet、DensNet、双通道网络单元结构节点表示图。(a) ResNet网络单元;(b) DensNe网络单元;(c)双通道网络单元
Fig. 3. Node representations of cell structures of ResNet, DensNet, and two-path networks. (a) ResNet network; (b) DensNet network; (c) two-path network
下载图片 查看原文
图 4. 传统的NMS问题举例图。(a)马;(b)鸟
Fig. 4. Examples of traditional NMS problems. (a) Horses; (b) birds
下载图片 查看原文
图 5. 52层、100层、133层深度的特征提取网络参数量和Top-1错误率趋势图
Fig. 5. Trend of parameter quantity of feature extraction network with Top-1 error rate and 52, 100, and 133 layers
下载图片 查看原文
图 6. 网络增长率为12,18,24,48的特征提取网络的参数量和Top-1错误率趋势图
Fig. 6. Trend of parameter quantity of feature extraction network with Top-1 error rate and network growth rates of 12, 18, 24, and 48
下载图片 查看原文
表 1特征提取网络结构
Table1. Structure of feature extraction network
Layer | Output size | Detail |
---|
Conv1 | 112×112 | 7×7,64,stride 2 | Conv2 | 56×56 | 3×3 max pool,stride 2×α1 | Conv3 | 28×28 | ×α2 | Conv4 | 14×14 | ×α3 | Conv5 | 7×7 | ×α4 |
|
查看原文
表 2不同特征提取网络的复杂性比较
Table2. Comparison of complexity of different feature extraction networks
Feature extraction network | Depth | Parameter /106 |
---|
VGG-16 | 16 | 168 | DensNet(K=48) | 161 | 111 | ResNet | 101 | 150 | Ours(α1α2α3α4=6,8,16,3; K=48) | 100 | 134 |
|
查看原文
表 3IoU阈值、β参数、加权平均对平均值(AP)的影响(0.5、0.6、0.7代表不同的IoU阈值,w代表加权平均)
Table3. Influences of IoU threshold, β parameter, and weighted average on AP (0.5, 0.6, and 0.7 represent different IoU thresholds; w represents weighted average)
Different parameter | A | A | A | A | A | A |
---|
Normal NMS | 44.37 | 44.83 | 39.18 | 39.67 | 29.83 | 30.34 | β=2.5, σ=0.4 | 46.42 | 46.92 | 42.83 | 43.40 | 34.68 | 35.24 | β=1.67, σ=0.6 | 46.58 | 47.11 | 43.30 | 43.79 | 35.21 | 35.76 | β=1.25, σ=0.8 | 45.93 | 46.45 | 41.68 | 42.21 | 33.01 | 33.53 |
|
查看原文
表 4数据增强和改进的NMS机制对准确率的影响
Table4. Influences of data augmentation and improved NMS mechanism on accuracy
Detection framwork | Backbone | Training set | Testing set | mAP /% |
---|
OursNo augmentation No improved NMS | ProposedProposedProposed | VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012 | VOC2007VOC2007VOC2007 | 79.176.678.0 |
|
查看原文
表 5不同epoch对准确率的影响结果
Table5. Influences of different epochs on accuracy
Nums of epoch | Learning rate setting | mAP /% |
---|
0 | No warming up | 78.20 | 2 | 0.01, 0.1 | 78.25 | 345 | 0.001, 0.01, 0.10.0001, 0.001, 0.01, 0.10.00001, 0.0001, 0.001, 0.01, 0.1 | 78.3678.6778.71 |
|
查看原文
表 6不同算法在VOC2007+VOC2012训练集下的测试结果
Table6. Testing results of different algorithms under VOC2007+VOC2012 training sets
Method | Backbone | Training set | Testing set | mAP/% | Frame rate /(frame·s-1) |
---|
Twostage | Fast R-CNNFaster R-CNNFaster R-CNNMR-CNNIONOurs | VGG-16VGG-16ResNet-101ResNet-101VGG-16Proposed | VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012 | VOC2007VOC2007VOC2007VOC2007VOC2007VOC2007 | 70.073.276.478.276.579.1 | 0.5072.400.031.252.10 | Onestage | YOLOYOLOv2SSD321SSD300*DSOD300DSSD513 | GoogleNetDarknet-19ResNet-101VGG-16DS/64-192-48-1ResNet-101 | VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012VOC2007+VOC2012 | VOC2007VOC2007VOC2007VOC2007VOC2007VOC2007 | 63.478.677.177.277.781.5 | 454011.204617.405.50 |
|
查看原文
乔婷, 苏寒松, 刘高华, 王萌. 基于改进的特征提取网络的目标检测算法[J]. 激光与光电子学进展, 2019, 56(23): 231008. Ting Qiao, Hansong Su, Gaohua Liu, Meng Wang. Object Detection Algorithm Based on Improved Feature Extraction Network[J]. Laser & Optoelectronics Progress, 2019, 56(23): 231008.