基于YOLOv5的舰船目标及关键部位检测算法

doi:10.12305/j.issn.1001-506X.2022.06.07

摘要/Abstract

摘要：

为进一步提升对可见光图像中水面舰船目标的检测识别成功率, 提出一种基于YOLOv5的舰船目标识别算法。使用基于随机池化方法的空间金字塔池化网络, 运用双向特征金字塔网络进行多尺度特征融合, 采用指数线性单元函数作为激活函数加快网络训练收敛速度, 提升算法鲁棒性, 从而实现了对水面舰船目标和舰船关键部位的快速准确识别。通过在舰船目标及其关键部位数据集上实验验证, 对比多个经典目标检测方法, 在识别准确率上均有不同程度提升, 对比原YOLOv5s模型, 平均精度均值提升3.03%, 速度提升2 FPS, 模型保持了YOLOv5轻量化的特点, 在应用部署上有良好前景。

关键词: YOLOv5, 随机池化, 双向特征金字塔网络, 指数线性单元函数

Abstract:

In order to improve the detection and recognition success rate of the surface warship target in visible light images, an algorithm based on YOLOv5 is proposed. The spatial pyramid pooling network based on stochastic pooling is used for pooling operation, and the bi-directional feature pyramid network is used for feature fusion. At the same time, the exponential linear unit function is used as the activation function to further accelerate the convergence speed and improve the robustness of the model, so as to realize the rapid and accurate recognition of surface ship targets and key parts of the ship. Through the experimental verification on the data set of the ship target and its key parts, compared with the mainstream target detection methods, the recognition accuracy is improved in varying degrees. Compared with the original YOLOv5s model, the mean average precision is improved by 3.03%, and the speed is improved by 2 FPS. The model maintains the lightweight characteristics of YOLOv5 and has a good prospect in application deployment.

Key words: YOLOv5, stochastic pooling, bi-directional feature pyramid network, exponential linear unit function

中图分类号:

TP39

钱坤, 李晨瑄, 陈美杉, 王瑶. 基于YOLOv5的舰船目标及关键部位检测算法[J]. 系统工程与电子技术, 2022, 44(6): 1823-1832.

Kun QIAN, Chenxuan LI, Meishan CHEN, Yao WANG. Ship target and key parts detection algorithm based on YOLOv5[J]. Systems Engineering and Electronics, 2022, 44(6): 1823-1832.

图/表 20

图1

表1

图2

图3

图4

图5

图6

图7

图8

图9

表2

表3

图10

图11

图12

表4

表5

图13

表6

图14

参考文献 36

1	KIM Y. Convolutional neural networks for sentence classification[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1408.5882v2.
2	DALAL N. Histograms of oriented gradients for human detection[C]//Proc. of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2005: 886-893.
3	王瑶, 徐昌, 舒福舟. 基于SVM算法的两种特征提取的图像分类方法分析[J]. 电脑与信息技术, 2019, 27 (6): 18- 20.18-20, 33
	WANG Y , XU C , SHU F Z . Analysis of image classification methods based on two feature extraction based on SVM algorithm[J]. Computer and Information Technology, 2019, 27 (6): 18- 20.18-20, 33
4	ZHU J , ARBOR A , HASTIE T . Multi-class adaBoost[J]. Statistics and its Interface, 2006, 2 (3): 349- 360.
5	苏赋, 吕沁, 罗仁泽. 基于深度学习的图像分类研究综述[J]. 电信科学, 2019, 35 (11): 58- 74.
	SU B , LYU Q , LUO R Z . Review of image classification based on deep learning[J]. Telecommunications Science, 2019, 35 (11): 58- 74.
6	ZOU Z X, SHI Z W, GUO Y H, et al. Object detection in 20 years: a survey[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1905.05055v2.
7	HINTON G E , OSINDERO S , TEH Y W . A fast learning algorithm for deep belief nets[J]. Neural Computation, 2014, 18 (7): 1527- 1554.
8	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580-587.
9	南晓虎, 丁雷. 深度学习的典型目标检测算法综述[J]. 计算机应用研究, 2020, 37 (S2): 15- 21.
	NAN X H , DING L . Review of typical target detection algorithms based on deep learning[J]. Application Research of Computers, 2020, 37 (S2): 15- 21.
10	GIRSHICK R. Fast R-CNN[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1504.08083.
11	REN S Q , HE K M , GIRSHICK R , et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Trans.on Pattern Analysis & Machine Intelligence, 2017, 39 (6): 1137- 1149.
12	CHEN X L, GUPTA A. An implementation of faster RCNN with study for region sampling[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1702.02138v2.
13	DAI J F, LI Y, HE K M, et al. R-FCN: object detection via region-based fully convolutional networks[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1605.06409v2.
14	SINGH B, LI H D, SHARMA A, et al. R-FCN-3000 at 30fps: decoupling detection and classification[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1712.01802v1.
15	HE K M, GKIOXARI G, P DOLLAR, et al. Mask R-CNN[C]// Proc. of the IEEE International Conference on Computer Vision, 2017: 2980-2988.
16	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multiBox detector[C]//Proc. of the European Conference on Computer Vision, 2020, 42(2): 318-327.
17	LIN T Y , GOYAL P , GIRSHICK R , et al. Focal loss for dense object detection[J]. IEEE Trans.on Pattern Analysis & Machine Intelligence, 2020, 42 (2): 318- 327.
18	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 779-788.
19	REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 6517-6525.
20	YU H, ZHANG Z, QIN Z N, et al. Loss rank mining: a general hard example mining method for real-time detectors[C]//Proc. of the International Joint Conference on Neural Networks, 2018.
21	REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1804.02767v1.
22	BOCHKOVSKIV A, WANG C Y, LIAO H. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. [2022-04-22]. https://arxiv.org/abs/2004.10934v1.
23	聂丰英. 基于多特征联合稀疏表示的SAR舰船目标识别方法[J]. 火力与指挥控制, 2020, 45 (10): 34- 38.
	NIE F Y . SAR ship target recognition method using joint sparse representation of multiple features[J]. Fire Control & Command Control, 2020, 45 (10): 34- 38.
24	吴映铮, 杨柳涛. 基于HOG和SVM的船舶图像分类算法[J]. 上海船舶运输科学研究所学报, 2019, 42 (1): 58- 64.
	WU Y Z , YANG L T . Ship image classification by combined use of HOG and SVM[J]. Journal of Shanghai Ship and Shipping Research Institute, 2019, 42 (1): 58- 64.
25	李兆桐, 孙浩云. 基于全卷积神经网络的船舶检测和船牌识别系统[J]. 计算机与现代化, 2019, (12): 72- 77.
	LI Z T , SUN H Y . A ship detection and plate recognition system based on FCN[J]. Computer and Modernization, 2019, (12): 72- 77.
26	段敬雅. 基于深度学习的船舶目标识别算法研究[D]. 广州: 华南理工大学, 2020.
	DUAN J Y. Research on ship recognition algorithm based on deep learning[D]. Guangzhou: South China University of Technology, 2020.
27	曲颖丽. 基于卷积神经网络的船舶识别[D]. 大连: 大连海事大学, 2020.
	QU Y L. Ship recognition based on convolutional neural network[D]. Dalian: Dalian Maritime University, 2020.
28	HE K M , ZHANG X Y , REN S Q , et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Trans.on Pattern Analysis & Machine Intelligence, 2014, 37 (9): 1904- 1916.
29	OUYANG W, WANG X Q, ZENG X G, et al. DeepID-net: deformable deep convolutional neural networks for object detection[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2015.
30	LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8759-8768.
31	LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 936-944.
32	ZEILER M D, FERGUS R. Stochastic pooling for regularization of deep convolutional neural networks[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1301.3557v1.
33	高惠琳. 基于卷积神经网络的军事图像分类[J]. 计算机应用研究, 2017, (11): 323- 325.
	GAO H L . Military image classification based on convolutional neural network[J]. Application Research of Computers, 2017, (11): 323- 325.
34	TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2020.
35	CLEVERT D A, UNTERTHINER T, HOCHREITER S. Fast and accurate deep network learning by exponential linear units (ELUs)[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1511.07289v5.
36	DUAN K W, BAI S, XIE L X, et al. CenterNet: keypoint triplets for object detection[EB/OL]. [2022-04-22]. https://arxiv.org/abs/1904.08189v3.

参数	YOLOv5s	YOLOv5m	YOLOv5l	YOLOv5x
深度倍数	0.33	0.67	1.0	1.33
宽度倍数	0.50	0.75	1.0	1.25
骨干段CSPN数量	133	266	399	41 212
颈部段CSPN数量	1	2	3	4
骨干段Focus和4个CBL中卷积核数量	3 264 128, 256 512	4 896 192, 384 768	64 128 256, 5 121 024	80 160 320, 6 401 280

标注类别	数据集总数	航空母舰	驱逐舰	桅杆	驾驶舱	相控阵雷达天线	舵机舱	民用船只
数量/幅	1 544	636	741	1 476	1 032	405	393	558
各类占比/%	-	41.2	48.0	95.6	66.8	26.2	25.5	36.1

操作系统	Windows10
CPU	i7-9700 3.00 GHz
GPU	NVIDIA 2060 SUPER
GPU显存/GB	8
内存/G	16
加速环境	CUDAv11.0、CuDNNv7.6
显卡驱动版本	27.21.14.617 2
Pytorch版本	1.7
Python版本	3.8.5

算法	图像分辨率	mAP/%	航母	驱逐舰	桅杆	驾驶舱	天线	舵机舱	民船	FPS
HOG+SVM[24]	500×500	64.82	88.4	70.2	73.9	52.6	62.8	66.3	39.5	12
SSD[16]	500×500	72.34	94.1	77.6	81.9	64.4	69.7	64.8	53.9	28
Mask R-CNN[15]	512×384	73.65	96.2	78.9	83.1	66.3	72.5	63.7	54.8	14
CenterNet-Hourglass(simple)[36]	512×512	72.25	94.6	77.2	81.4	62.3	68.5	63.4	51.3	26
改进YOLOv2[26]	416×416	72.20	96.8	77.5	80.2	62.1	67.2	64.5	50.1	24
YOLOv4[22]	512×512	73.78	97.2	81.7	80.3	66.5	66.1	64.2	53.4	24
YOLOv5s	640×640	74.10	97.2	79.4	84.6	65.9	69.2	67.1	55.3	25
本文算法	640×640	77.13	97.3	86.1	84.7	67.8	79.9	63.9	60.2	27

算法	参数量/M	FLOPs/GB
RetinaNet[17]	53.1	127
Mask R-CNN[15]	63.4	164
CenterNet-Hourglass (simple)[36]	95.4	140
改进YOLOv2[26]	107.7	149
YOLOv4[22]	27.6	31
YOLOv5s	7.5	17
Proposed method	7.3	13.2