基于改进Faster R-CNN的红外舰船目标检测算法

顾佼佼; 李炳臻; 刘克; 姜文志

基于改进Faster R-CNN的红外舰船目标检测算法

1.
海军航空大学岸防兵学院，山东烟台 264001
2.
中国人民解放军95668部队，云南昆明 650000

详细信息

作者简介:
顾佼佼（1984-），男，博士，讲师，主要研究方向：人工智能深度学习技术

通讯作者:
李炳臻（1996-），男，硕士，主要研究方向：深度学习技术。E-mail：libingzhen123456@163.com

中图分类号: TP399
计量
- 文章访问数: 624
- HTML全文浏览量: 248
- PDF下载量: 120
- 被引次数: 0
出版历程
- 收稿日期: 2020-06-11
- 修回日期: 2020-07-06
- 刊出日期: 2021-02-20

Infrared Ship Target Detection Algorithm Based on Improved Faster R-CNN

1.
Naval Aviation University, Coast Guard Academy, Yantai 264001, China
2.
Unit 95668 People's Liberation Army of China, Kunming 650000, China

摘要

摘要: 针对Faster R-CNN算法中对于红外舰船目标特征提取不充分、容易出现重复检测的问题，提出了一种基于改进Faster R-CNN的红外舰船目标检测算法。首先通过在主干网络VGG-16中依次引出三段卷积后的3个特征图，将其进行特征拼接形成多尺度特征图，得到具有更丰富语义信息的特征向量；其次基于数据集进行Anchor的改进，重新设置Anchor boxes的个数与尺寸；最后优化改进后Faster R-CNN的损失函数，提高检测算法的整体性能。通过对测试数据集进行分析实验，结果表明改进后的检测算法平均精确度达到83.98%，较之于原Faster R-CNN，精确度提升了3.95%。
- 深度学习 /
- 目标检测 /
- 舰船目标 /
- 红外图像 /
- Faster R-CNN
Abstract: To solve the problem of insufficient feature extraction and repeated detection of infrared ship targets by the Faster R-CNN algorithm, a ship target detection algorithm based on an improved Faster R-CNN is proposed. First, three feature graphs are drawn from the backbone network, VGG-16, after a three-segment convolution, and the features are spliced to form a multi-scale feature graph to obtain a feature vector with richer semantic information; second, the Anchor is improved based on the dataset, and the number and size of the Anchor boxes are reset; finally, the loss function of the improved Faster R-CNN is optimized to improve the feature extraction ability of the target. An analysis of the experimental results on the test dataset demonstrates that the average accuracy of the improved detection algorithm was 83.98%, which is 3.95% higher than that of the original Faster RCNN.
- deep learning /
- target detection /
- ship target /
- infrared image /
- Faster R-CNN

HTML全文

图 1 Faster R-CNN网络结构图

Figure 1. Faster R-CNN network structure diagram

下载: 全尺寸图片幻灯片

图 2 VGG-16网络参数列表图

Figure 2. VGG-16 network parameter list diagram

下载: 全尺寸图片幻灯片

图 3 RPN网络结构示意图

Figure 3. Schematic diagram of RPN network structure

下载: 全尺寸图片幻灯片

图 4 Anchor示意图

Figure 4. Anchor schematic diagram

下载: 全尺寸图片幻灯片

图 5 Bounding box regression示例说明

Figure 5. Bounding box regression example description

下载: 全尺寸图片幻灯片

图 6 不同层级卷积后特征图对比

Figure 6. Comparison of characteristic graphs after convolution at different levels

下载: 全尺寸图片幻灯片

图 7 改进后网络结构图

Figure 7. Improved network structure diagram

下载: 全尺寸图片幻灯片

图 8 特征拼接后特征图

Figure 8. Feature map after feature stitching

下载: 全尺寸图片幻灯片

图 9 改进后的Anchor示意图

Figure 9. Improved Anchor schematic diagram

下载: 全尺寸图片幻灯片

图 10 数据增强示例图

Figure 10. Sample diagram of data enhancement

下载: 全尺寸图片幻灯片

图 11 改进的Faster R-CNN损失函数曲线

Figure 11. Improved Faster R-CNN loss function curve

下载: 全尺寸图片幻灯片

图 12 红外舰船图像检测结果

Figure 12. Detection result of infrared ship image

下载: 全尺寸图片幻灯片

图 13 改进前后Faster R-CNN在红外舰船测试集上的R-P曲线

Figure 13. R-P curves of Faster R-CNN on infrared ship test set before and after improvement

下载: 全尺寸图片幻灯片

图 14 Faster R-CNN改进前后红外目标检测效果对比

Figure 14. Comparison of infrared target detection effect before Faster R-CNN improvement

下载: 全尺寸图片幻灯片

表 1 分类结果判别表

Table 1. Classification result discriminant table

Real situation	Discriminant result
Real situation	Positive example	Counter example
Positive example	TP(True positive example)	FN(False Counter example)
Counter example	FP(False positive example)	TN(True Counter example)

下载: 导出CSV

表 2 改进前后算法性能对比

Table 2. Comparison of algorithm performance before and after improvement

Model name	AP/%	mAP/%	Time/s
Faster R-CNN	80.03	80.03	0.3128
Improved Faster R-CNN	83.98	83.98	0.3384

下载: 导出CSV

参考文献(18)

[1]	施泽浩, 赵启军. 基于全卷积网络的目标检测算法[J]. 计算机技术与发展, 2018(5): 55-58. doi: 10.3969/j.issn.1673-629X.2018.05.013 SHI Zehao, ZHAO Qijun. Target detection algorithm based on full convolution network[J]. Computer Technology and Development, 2018(5): 55-58. doi: 10.3969/j.issn.1673-629X.2018.05.013
[2]	Uijlings J R R, Sande K E A V D, Gevers T, et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013, 104(2): 154-171. doi: 10.1007/s11263-013-0620-5
[3]	Girshick R. Fast R-CNN[C]//Proceedings of ICCV, 2015, 2015: 1440-1448.
[4]	Kaiming H, Georgia G, Piotr D, et al. Mask R-CNN[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 18(11): 1-1.
[5]	Ren S, He K, Girshick R, et al. Faster r-cnn: towards real-time object detection with region proposal networks[C]//IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[6]	Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]//Proceedings of CVPR, 2015: 779-788.
[7]	Redmon J, Farhadi A. YOLO9000: Better, faster, stronger[C]//IEEE Conference on Computer Vision and Pattern Recognition, 2017, 12(7): 6517-6525.
[8]	LIU W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]//Computer Vision-ECCV 2016, Cham: Springer, 2016, 9905: 21-37.
[9]	FU C Y, LIU W, Ranga A, et al. DSSD: deconvolutional single shot detector[C]//2017 IEEE International Conference on Computer Vision, 2017: 2999-3007.
[10]	Redmon J, Farhadi A. YOLO v3: An incremental improvement [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2018, 24(9): 2458-2476.
[11]	李慕锴, 张涛, 崔文楠. 基于YOLOv3的红外行人小目标检测技术研究[J]. 红外技术, 2020, 42(2): 176-181. http://hwjs.nvir.cn/article/id/hwjs202002012 LI Muyi, ZHANG Tao, CUI Wennan. Research on infrared pedestrian small target detection technology based on YOLO v3[J]. Infrared Technology, 2020, 42(2): 176-181. http://hwjs.nvir.cn/article/id/hwjs202002012
[12]	崔少华, 李素文, 黄金乐, 等. 改进的CNN用于单帧红外图像行人检测的方法[J]. 红外技术, 2020, 42(3): 238-244. http://hwjs.nvir.cn/article/id/hwjs202003006 CUI Shaohua, LI Suwen, HUANG Jinle, et al. Improved CNN square method for human detection of single-frame infrared image[J]. Infrared Technology, 2020, 42(3): 238-244. http://hwjs.nvir.cn/article/id/hwjs202003006
[13]	向涛. 一种基于显著区域提取的红外图像舰船目标检测方法[J]. 电讯技术, 2020, 60(7): 50-56. XIANG Tao. A ship target detection method based on salient region extraction in infrared image[J]. Telecommunication Technology, 2020, 60(7): 50-56.
[14]	邢莎, 吉林, 雍杨, 等. 基于梯度统计特性的自动红外舰船目标检测[J]. 数字技术与应用, 2013(10): 66-68, 70. https://www.cnki.com.cn/Article/CJFDTOTAL-SZJT201310047.htm XING Sha, JI Lin, YONG Yang, et al. Automatic infrared ship target detection based on gradient statistics[J]. Digital Technology and Applications, 2013(10): 66-68, 70. https://www.cnki.com.cn/Article/CJFDTOTAL-SZJT201310047.htm
[15]	宫剑, 吕俊伟, 刘亮, 等. 红外偏振图像的舰船目标检测[J]. 光谱学与光谱分析, 2020, 40(2): 586-594. https://www.cnki.com.cn/Article/CJFDTOTAL-GUAN202002052.htm GONG Jian, LV Junwei, LIU Liang, et al. Ship target detection based on infrared polarization image[J]. Spectroscopy and Spectral Analysis, 2020, 40(2): 586-594. https://www.cnki.com.cn/Article/CJFDTOTAL-GUAN202002052.htm
[16]	吴天舒, 张志佳, 刘云鹏. 基于改进SSD的轻量化小目标检测算法[J]. 红外与激光工程, 2018(7): 37-43. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201807007.htm WU Tianshu, ZHANG Zhijia, LIU Yunpeng. Lightweight small beacon detection algorithm based on improved SSD[J]. Infrared and Laser Engineering, 2018(7): 37-43. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201807007.htm
[17]	WANG Zijie J, Robert Turko, Omar Shaikh, et al. CNN explainer: learning convolutional neural networks with interactive visualization[J]. Journal of Sensors, 2019(6): 1-13. http://ieeexplore.ieee.org/document/8941872/
[18]	Zeiler M D, Fergus R. Visualizing and understanding convolutional networks[C]// Proceedings of the European Conference on Computer Vision, Berlin, Heidelberg: Springer, 2014, 8689: 818-833.