弱特征共焦通道调控水下声呐目标检测

何梦云; 何自芬; 张印辉; 陈光晨; 张枫

doi:10.37188/CO.2024-0031

弱特征共焦通道调控水下声呐目标检测

doi: 10.37188/CO.2024-0031

昆明理工大学机电工程学院, 云南昆明 650500

基金项目: 国家自然科学基金资助项目（No. 62171206，No. 62061022）

详细信息

作者简介:
何梦云(2000—)，女，云南昆明人，硕士研究生，2022年于天津科技大学机械工程学院获得学士学位，现为昆明理工大学机电工程学院硕士研究生，主要从事图像处理、机器视觉及机器智能方面的研究。E-mail：18088400647@163.com

何自芬(1976—)，女，山西阳泉人，博士，教授，硕士生导师，2000年、2005年于西安理工大学分别获得学士和硕士学位，2013年于昆明理工大学获得博士学位，主要从事图像处理和机器视觉等方面的研究。E-mail：zyhhzf1998@163.com

张印辉（1977—），男，河北故城人，博士，教授、博士生导师，2000年和2005年分别于西安理工大学获得学士学位和硕士学位，2010年于昆明理工大学获得博士学位，主要从事图像处理、机器视觉及机器智能等方面的研究。E-mail：zhangyinhui@kust.edu.cn

陈光晨（1997—），男，安徽六安人，博士研究生，2020年于安徽工程大学获得学士学位，2023年于昆明理工大学获得硕士学位，现为昆明理工大学机电工程学院博士研究生。主要从事图像处理、机器视觉、深度学习等方面的研究。E-mail：guangchen_c@yeah.net

张枫（1998—），男，江苏南通人，硕士研究生，2021年于淮阴工学院机械与材料工程学院获得学士学位，现为昆明理工大学机电工程学院硕士研究生，主要从事图像处理、机器视觉及深度学习等方面的研究。E-mail：zf1977497475@163.com

中图分类号: TP394.1;TH691.9
计量
- 文章访问数: 77
- HTML全文浏览量: 36
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2024-02-05
- 录用日期: 2024-04-26
- 网络出版日期: 2024-05-17

Weak feature confocal channel regulation for underwater sonar target detection

Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650500, China

Funds: Supported by the National Natural Science Foundation of China ( No. 62171206,No.62061022)

More Information

Corresponding author: zyhhzf1998@163.com

摘要

摘要:
声呐图像视觉检测是复杂水域资源勘探和水下异物目标探测领域的重要技术之一。针对声呐图像中小目标存在特征微弱和背景信息干扰问题，本文提出弱特征共焦通道调控水下声呐目标检测算法。首先，为提高模型对弱小目标信息捕获和表征能力，设计弱小目标特征激活策略并引入先验框尺度校准机制匹配底层语义特征检测分支以提高小目标检测精度；其次，应用全局信息聚合模块深入挖掘弱小目标全局特征，避免冗余信息覆盖小目标微弱关键特征；最后，为解决传统空间金字塔池化易忽视通道信息的问题，提出共焦通道调控池化模块，保留有效通道域小目标信息并克服复杂背景信息干扰。实验表明，本文模型在水下声呐数据集的九类弱小目标识别上平均检测精度达到83.3%，相较基准提高5.5%，其中铁桶、人体模型和立方体检测精度得到显著提高，分别提高24%、8.6%和7.3%，有效改善水下复杂环境中弱小目标漏检和误检问题。
- 弱小目标检测 /
- 水下声呐图像 /
- 全局信息聚合 /
- 共焦通道调控池化
Abstract:
Visual detection of sonar image is one of the important technologies in the field of resource exploration in complex waters and underwater foreign object target detection. Aiming at the problem of weak features and background information interference of small targets in sonar images, this paper proposes a weak feature confocal channel modulation algorithm for underwater sonar target detection. Firstly, in order to improve the model's ability to capture and characterize the information of weak targets, we design a weak target-specific activation strategy and introduce an a priori frame scale calibration mechanism to match the underlying semantic feature detection branch to improve the accuracy of small target detection; secondly, we apply the global information aggregation module to deeply excavate the global features of weak targets to avoid the redundant information from covering the small target's weak key features; lastly, in order to solve the problem of the traditional space pyramid Finally, in order to solve the problem of traditional spatial pyramid pooling which is easy to ignore the channel information, the confocal channel regulation pooling module is proposed to retain the effective channel domain small target information and overcome the interference of complex background information. Experiments show that the model in this paper achieves an average detection accuracy of 83.3% on nine types of weak targets in the underwater sonar dataset, which is 5.5% higher than the benchmark, among which the detection accuracy of iron bucket, human body model and cube is significantly improved by 24%, 8.6% and 7.3%, respectively, which effectively improves the problem of leakage and misdetection of weak targets in the underwater complex environment.
- weak target detection /
- underwater sonar images /
- global information aggregation /
- confocal channel modulation pooling

HTML全文

图 1 WFCCMNet模型结构图

Figure 1. WFCCMNet model structure

下载: 全尺寸图片幻灯片

图 2 全局信息聚合模块结构图

Figure 2. Structure of global information aggregation module

下载: 全尺寸图片幻灯片

图 3 共焦通道调控池化模块结构图

Figure 3. Structure of confocal channel modulation pooling module

下载: 全尺寸图片幻灯片

图 4 数据实例分布图

Figure 4. Distribution of data instances and a priori frames

下载: 全尺寸图片幻灯片

图 5 检测结果可视化

Figure 5. Visualisation of detection results

下载: 全尺寸图片幻灯片

表 1 改进前后输出分支与先验框对应关系

Table 1. Correspondence between output branches and a priori boxes before and after improvement

改进前		改进后
输出分支	先验框尺度	输出分支	先验框尺度
-	-	P₄	[(116,90),(156,198),(373,326)]
N₃	[(10,13),(16,30),(33,23)]	P₃	[(5,6),(8,14),(15,11)]
N₄	[(30,61),(62,45),(59,119)]	P₂	[(10,13),(16,30),(33,23)]
N₅	[(116,90),(156,198),(373,326)]	P₁	[(30,61),(62,45),(59,119)]

下载: 导出CSV

表 2 目标图像面积占比

Table 2. Target image area percentage

Classes name	Maximum area of target frame(height×width)	Ration
ball	156×142	0.04
circle cage	128×148	0.04
cube	140×164	0.04
cylinder	122×102	0.02
human body	215×216	0.09
metal bucket	168×206	0.07
plane	187×278	0.10
square cage	130×118	0.03
tyre	166×166	0.05

下载: 导出CSV

表 3 改进模块消融实验

Table 3. Improved module ablation experiments

	GFLOPS	mAP50	human body	ball	circle cage	square cage	tyre	metal bucket	cube	cylinder	plane
Baseline	15.8	77.8	81.2	84.9	81.6	78.7	64.8	61.1	84.4	71.8	92.5
+特征激活	18.6	80.1	87.6	82.5	81.1	80.9	64.9	68.8	86.4	77.5	91
+GIAM	19.5	80.9	86.8	83.8	84.1	78.5	64.3	80.1	85.8	74.8	90.3
+CFCRP	23.7	83.3	89.8	85.7	79.1	82.9	73.1	85.1	91.7	71.2	91

下载: 导出CSV

表 4 实验代号与输出分支和先验框尺度对应关系

Table 4. Correspondence of experiment codes with output layers and a priori boxes

实验代号	输出分支	先验框尺度	输出分支	先验框尺度
G1	-	-	N₄	[(30,61),(62,45),(59,119)]
G1	N₃	[(10,13),(16,30),(33,23)]	N₅	[(116,90),(156,198),(373,326)]
G2	P₄	[(5,6),(8,14),(15,11)]	P₂	[(30,61),(62,45),(59,119)]
G2	P₃	[(10,13),(16,30),(33,23)]	P₁	[(116,90),(156,198),(373,326)]
G3	P₄	[(10,13),(16,30),(33,23)]	P₂	[(30,61),(62,45),(59,119)]
G3	P₃	[(5,6),(8,14),(15,11)]	P₁	[(116,90),(156,198),(373,326)]
G4	P₄	[(30,61),(62,45),(59,119)]	P₂	[(10,13),(16,30),(33,23)]
G4	P₃	[(5,6),(8,14),(15,11)]	P₁	[(116,90),(156,198),(373,326)]
G5	P₄	[(116,90),(156,198),(373,326)]	P₂	[(10,13),(16,30),(33,23)]
G5	P₃	[(5,6),(8,14),(15,11)]	P₁	[(30,61),(62,45),(59,119)]

下载: 导出CSV

表 5 检测分支与先验框组合定量实验结果

Table 5. Results of quantitative experiments on the combination of detection branch and a priori frame

实验代号	GFLOPS	mAP50	human body	ball	circle cage	square cage	tyre	metal bucket	cube	cylinder	plane
G1	15.8	77.8	81.2	84.9	81.6	78.7	64.8	61.1	84.4	71.8	92.5
G2	15.9	78.6	88.8	83.7	79.7	82.6	60.6	66.9	87.1	68.5	89.7
G3	15.9	77.5	77.9	82.3	80.7	72.4	63.9	67.6	86.9	73.7	92.1
G4	15.9	79.2	79.9	81.7	78.5	75.2	64.4	74.2	87.3	76.6	95.1
G5	15.9	79.3	89.4	83.2	78.6	83.3	63	71.1	84.3	71.2	90

下载: 导出CSV

表 6 多尺度共焦卷积对比实验

Table 6. Multi-scale confocal convolution comparison experiments

Base Model	共焦卷积核	GFLOPS	mAP50	human body	ball	circle cage	square cage	tyre	metal bucket	cube	cylinder	plane
YOLOv5s +特征激活 +GIAM	-	19.5	80.9	86.8	83.8	84.1	78.5	64.3	80.1	85.8	74.8	90.3
	+1×1	19.3	80.8	85.3	86.1	67.3	85.1	71.3	84.5	89.7	68.8	89.5
	+1×1 +3×3	19.4	81.4	89.4	84.1	78.8	83.3	66.1	87.5	87.1	68.7	87.5
	+1×1 +3×3 +5×5	20.8	80.6	91.8	83.4	75	79.1	67.2	72.3	87.2	80.5	88.8
	+1×1 +3×3 +5×5 +7×7	23.7	83.3	89.8	85.7	79.1	82.9	73.1	85.1	91.7	71.2	91
	+1×1 +3×3 +5×5 +7×7 +9×9	27.6	78.2	85.3	84.9	74	80.2	66.4	78.2	88.6	61.6	84.6

下载: 导出CSV

表 7 空间特征金字塔池化对比实验

Table 7. Spatial feature pyramid pooling comparison experiments

Base Model	空间特征金字塔池化	GFLOPS	mAP50	human body	ball	circle cage	square cage	tyre	metal bucket	cube	cylinder	plane
YOLOv5s +特征激活 +GIAM	+SPPF	19.5	80.9	86.8	83.8	84.1	78.5	64.3	80.1	85.8	74.8	90.3
	+SPP	18.5	81.4	86.1	83.6	81.9	86.8	67.5	73.2	86.4	76.8	90.2
	+SPPFCSPC	23.6	73.9	78.1	79.7	71.6	76.5	60.3	60.4	85.4	62.1	91.2
	+ASPP	25.1	77	78.4	80.1	72.8	77.6	70	76.6	88	61.8	87.9
	+RFB	19.0	80.5	86.8	85.8	78.7	80.4	69.7	80.3	87.2	72.2	83.8
	+SPPCSPC	23.6	75.2	80.3	78.6	73.4	72.8	62.4	71.4	82.4	72.3	83.1
	+CFCRP	23.7	83.3	89.8	85.7	79.1	82.9	73.1	85.1	91.7	71.2	91

下载: 导出CSV

表 8 网络模型对比实验

Table 8. Network model comparison experiment

Model	GFLOPS	mAP50	human body	ball	circle cage	square cage	tyre	metal bucket	cube	cylinder	plane
SSD^[38]	347.1	78.8	89.1	90.9	74.8	76.4	62.3	77.4	81	68.4	89
RetinaNet^[39]	207.9	68.4	89.8	77.5	59.8	74.7	47.4	75.9	78.6	24	87.9
YOLOv5x^[40]	203.9	77.4	76.1	83.8	76.8	72.7	62.3	80	83.8	66.2	95
Faster RCNN^[41]	193.8	71.2	80.4	81.1	77.6	75.7	45	70	81.8	40.4	88.7
YOLOv7^[29]	103.3	60.4	71.5	74.2	57.8	67	48.1	59.8	72.8	35.8	56.4
YOLOv5m^[40]	48	71.3	68.6	81.8	79	61.5	56.7	58.9	85.1	69.3	80.7
DAMO-YOLO^[42]	36	72.7	65.5	81.3	70.2	70.8	38.7	87.8	85.6	71.8	82.2
YOLOv8s^[43]	28.2	81.4	82.3	86.5	81.1	86.6	65	93.7	88.5	53.5	95
YOLOXs^[44]	26.7	82.4	88.1	88	75.9	87.2	71.7	79.3	89.6	71.4	90
YOLOv5s^[40]	15.8	77.8	81.2	84.9	81.6	78.7	64.8	61.1	84.4	71.8	92.5
YOLOv7-tiny^[45]	13.1	64.2	70.3	84.5	46.7	76.8	35.4	66.3	83.3	41.7	72.7
YOLOv3-tiny^[41]	5.57	63.2	65.2	73.6	69.4	63.2	37.5	63.9	76	48	72.1
YOLOv5n^[40]	4.2	72.3	86	79.2	74.5	77.9	59.3	52.8	81.8	58	80.8
WFCCMNet	23.7	83.3	89.8	85.7	79.1	82.9	73.1	85.1	91.7	71.2	91

下载: 导出CSV

参考文献(45)

[1]	王芳. 新时期海洋强国建设形势与任务研究[J]. 中国海洋大学学报(社会科学版),2020(5):11-19. WANG F. Research on the situation and tasks of building a strong maritime power in the new era[J]. Journal of Ocean University of China (Social Sciences), 2020(5): 11-19. (in Chinese).
[2]	CLAY C S, HORNE J K. Acoustic models of fish: the Atlantic cod (Gadus morhua)[J]. The Journal of the Acoustical Society of America, 1994, 96(3): 1661-1668. doi: 10.1121/1.410245
[3]	HARLEY H E, DELONG C M. Echoic object recognition by the bottlenose dolphin[J]. Comparative Cognition & Behavior Reviews, 2008, 3: 46-65.
[4]	谭亦秋. 基于直流超导量子干涉仪的水下铁磁性目标探测技术研究[D]. 长沙: 国防科技大学, 2020. TAN Y Q. Research on detection technology of underwater ferromagnetic target based on DC superconducting quantum interference device[D]. Changsha: National University of Defense Technology, 2020. (in Chinese).
[5]	陈正想, 卢俊杰. 弱磁探测技术发展现状[J]. 水雷战与舰船防护,2011,19(4):1-5,24. CHEN ZH X, LU J J. Current development of weak magnetic detection[J]. Mine Warfare & Ship Self-Defence, 2011, 19(4): 1-5,24. (in Chinese).
[6]	XU W H, YANG J M, WEI H D, et al. A localization algorithm based on pose graph using Forward-looking sonar for deep-sea mining vehicle[J]. Ocean Engineering, 2023, 284: 114968. doi: 10.1016/j.oceaneng.2023.114968
[7]	罗逸豪. 基于深度学习的声呐图像目标检测系统[J]. 数字海洋与水下攻防,2023,6(4):423-428. LUO Y H. Sonar image object detection system based on deep learning[J]. Digital Ocean & Underwater Warfare, 2023, 6(4): 423-428. (in Chinese).
[8]	ISHAK A B. A two-dimensional multilevel thresholding method for image segmentation[J]. Applied Soft Computing, 2017, 52: 306-322. doi: 10.1016/j.asoc.2016.10.034
[9]	ZHANG B M, ZHOU T, SHI ZI F, et al. An underwater small target boundary segmentation method in forward-looking sonar images[J]. Applied Acoustics, 2023, 207: 109341. doi: 10.1016/j.apacoust.2023.109341
[10]	万广南. 基于激光和超声的水下目标探测方法研究[D]. 哈尔滨: 哈尔滨工业大学, 2017. WAN G N. Research on underwater target detection based laser and ultrasound[D]. Harbin: Harbin Institute of Technology, 2017. (in Chinese).
[11]	翟厚曦, 江泽林, 张鹏飞, 等. 一种合成孔径声呐图像目标分割方法[J]. 仪器仪表学报,2016,37(4):887-894. doi: 10.3969/j.issn.0254-3087.2016.04.022 ZHAI H X, JIANG Z L, ZHANG P F, et al. Object segmentation method for synthetic aperture sonar images[J]. Chinese Journal of Scientific Instrument, 2016, 37(4): 887-894. (in Chinese). doi: 10.3969/j.issn.0254-3087.2016.04.022
[12]	杨卫东, 叶长彬, 陈正林, 等. 基于snake算法的声呐图像轮廓提取方法[J]. 压电与声光,2023,45(5):752-758. YANG W D, YE CH B, CHEN ZH L, et al. Image contour extraction method based on snake algorithm[J]. Piezoelectrics & Acoustooptics, 2023, 45(5): 752-758. (in Chinese).
[13]	胡钢. 基于深度学习的水下目标识别和运动行为分析技术研究[D]. 哈尔滨: 哈尔滨工程大学, 2021. HU G. Research on underwater target recognition and motion behavior analysis technology based on deep learning[D]. Harbin: Harbin Engineering University, 2021. (in Chinese).
[14]	DIVYABARATHI G, SHAILESH S, JUDY M V. Object classification in underwater SONAR images using transfer learning based ensemble model[C]. Proceedings of 2021 International Conference on Advances in Computing and Communications, IEEE, 2021: 1-4.
[15]	CHANDRASHEKAR G, RAAZA A, RAJENDRAN V, et al. Side scan sonar image augmentation for sediment classification using deep learning based transfer learning approach[J]. Materials Today: Proceedings, 2023, 80: 3263-3273. doi: 10.1016/j.matpr.2021.07.222
[16]	KONG W Z, HONG J CH, JIA M Y, et al. YOLOv3-DPFIN: a dual-path feature fusion neural network for robust real-time sonar target detection[J]. IEEE Sensors Journal, 2020, 20(7): 3745-3756. doi: 10.1109/JSEN.2019.2960796
[17]	REDMON J, FARHADI A. YOLOv3: an incremental improvement[J]. arXiv preprint arXiv: 1804.02767, 2018. (查阅网上资料, 请核对文献类型及格式) .
[18]	FAN X N, LU L, SHI P F, et al. A novel sonar target detection and classification algorithm[J]. Multimedia Tools and Applications, 2022, 81(7): 10091-10106. doi: 10.1007/s11042-022-12054-4
[19]	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[J]. arXiv preprint arXiv: 2004.10934, 2020. (查阅网上资料, 请核对文献类型及格式) .
[20]	GERG I D, MONGA V. Structural prior driven regularized deep learning for sonar image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 4200416.
[21]	NANDHINI S, ASHOKKUMAR K. An automatic plant leaf disease identification using DenseNet-121 architecture with a mutation-based henry gas solubility optimization algorithm[J]. Neural Computing and Applications, 2022, 34(7): 5513-5534. doi: 10.1007/s00521-021-06714-z
[22]	ZHU X Y, LIANG Y SH, ZHANG J L, et al. STAFNet: swin transformer based anchor-free network for detection of forward-looking sonar imagery[C]. Proceedings of the 2022 International Conference on Multimedia Retrieval, ACM, 2022: 443-450.
[23]	LIU Z, LIN Y T, CAO Y, et al. Swin transformer: hierarchical vision transformer using shifted windows[C]. Proceedings of 2021 IEEE/CVF International Conference on Computer Vision, IEEE, 2021: 10012-10022.
[24]	刘彦磊, 李孟喆, 王宣宣. 轻量型YOLOv5s车载红外图像目标检测[J]. 中国光学(中英文),2023,16(5):1045-1055. doi: 10.37188/CO.2022-0254 LIU Y L, LI M ZH, WANG X X. Lightweight YOLOv5s vehicle infrared image target detection[J]. Chinese Optics, 2023, 16(5): 1045-1055. doi: 10.37188/CO.2022-0254
[25]	朱威, 王立凯, 靳作宝, 等. 引入注意力机制的轻量级小目标检测网络[J]. 光学精密工程,2022,30(8):998-1010. doi: 10.37188/OPE.20223008.0998 ZHU W, WANG L K, JIN Z B, et al. Lightweight small object detection network with attention mechanism[J]. Optics and Precision Engineering, 2022, 30(8): 998-1010. (in Chinese). doi: 10.37188/OPE.20223008.0998
[26]	乔美英, 史建柯, 李冰锋, 等. 改进损失函数的增强型FPN水下小目标检测[J]. 计算机辅助设计与图形学学报,2023,35(4):525-537. QIAO M Y, SHI J K, LI B F, et al. Enhanced FPN underwater small target detection with improved loss function[J]. Journal of Computer-Aided Design & Computer Graphics, 2023, 35(4): 525-537. (in Chinese).
[27]	张艳, 李星汕, 孙叶美, 等. 基于通道注意力与特征融合的水下目标检测算法[J]. 西北工业大学学报,2022,40(2):433-441. doi: 10.3969/j.issn.1000-2758.2022.02.025 ZHANG Y, LI X SH, SUN Y M, et al. Underwater object detection algorithm based on channel attention and feature fusion[J]. Journal of Northwestern Polytechnical University, 2022, 40(2): 433-441. (in Chinese). doi: 10.3969/j.issn.1000-2758.2022.02.025
[28]	LI L, LI Y P, YUE CH H, et al. Real-time underwater target detection for AUV using side scan sonar images based on deep learning[J]. Applied Ocean Research, 2023, 138: 103630. doi: 10.1016/j.apor.2023.103630
[29]	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]. Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, IEEE, 2023: 7464-7475.
[30]	张绍文, 史卫亚, 张世强, 等. 基于加权感受野和跨层融合的遥感小目标检测[J]. 电子测量技术,2023,46(18):129-138. ZHANG SH W, SHI W Y, ZHANG SH Q, et al. Remote sensing small target detection based on weighted receptive field and cross-layer fusion[J]. Electronic Measurement Technology, 2023, 46(18): 129-138. (in Chinese).
[31]	CUI L SH, LV P, JIANG X H, et al. Context-aware block net for small object detection[J]. IEEE Transactions on Cybernetics, 2022, 52(4): 2300-2313. doi: 10.1109/TCYB.2020.3004636
[32]	YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions[J]. arXiv preprint arXiv: 1511.07122, 2015. (查阅网上资料, 请核对文献类型及格式) .
[33]	WANG C Y, LIAO H Y M, WU Y H, et al. CSPNet: a new backbone that can enhance learning capability of CNN[C]. Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, IEEE, 2020: 390-391.
[34]	https://openi.pcl.ac.cn/OpenOrcinus_orca. .
[35]	刘颖, 刘红燕, 范九伦, 等. 基于深度学习的小目标检测研究与应用综述[J]. 电子学报,2020,48(3):590-601. doi: 10.3969/j.issn.0372-2112.2020.03.024 LIU Y, LIU H Y, FAN J L, et al. A survey of research and application of small object detection based on deep learning[J]. Acta Electronica Sinica, 2020, 48(3): 590-601. (in Chinese). doi: 10.3969/j.issn.0372-2112.2020.03.024
[36]	CHEN CH Y, LIU M Y, TUZEL O, et al. R-CNN for small object detection[C]. Proceedings of the 13th Asian Conference on Computer Vision, Springer, 2017: 214-230.
[37]	高新波, 莫梦竟成, 汪海涛, 等. 小目标检测研究进展[J]. 数据采集与处理,2021,36(3):391-417. GAO X B, MO M J CH, WANG H T, et al. Recent advances in small object detection[J]. Journal of Data Acquisition and Processing, 2021, 36(3): 391-417. (in Chinese).
[38]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]. Proceedings of the 14th European Conference, Springer, 2016: 21-37.
[39]	WANG Y Y, WANG CH, ZHANG H, et al. Automatic ship detection based on RetinaNet using multi-resolution Gaofen-3 imagery[J]. Remote Sensing, 2019, 11(5): 531. doi: 10.3390/rs11050531
[40]	JOCHER G, STOKEN A, BOROVEC J, et al. Ultralytics/yolov5: v3.0[Z]. Zenodo, 2020. (查阅网上资料, 请核对文献类型及格式) .
[41]	REN SH Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. doi: 10.1109/TPAMI.2016.2577031
[42]	XU X ZH, JIANG Y Q, CHEN W H, et al. DAMO-YOLO: a report on real-time object detection design[J]. arXiv preprint arXiv: 2211.15444, 2022. (查阅网上资料, 请核对文献类型及格式) .
[43]	HUSSAIN M. YOLO-v1 to YOLO-v8, the rise of YOLO and its complementary nature toward digital manufacturing and industrial defect detection[J]. Machines, 2023, 11(7): 677. doi: 10.3390/machines11070677
[44]	WU Q, ZHANG B, XU CH G, et al. Dense oil tank detection and classification via YOLOX-TR network in large-scale SAR images[J]. Remote Sensing, 2022, 14(14): 3246. doi: 10.3390/rs14143246
[45]	MA L, ZHAO L Y, WANG Z X, et al. Detection and counting of small target apples under complicated environments by using improved YOLOv7-tiny[J]. Agronomy, 2023, 13(5): 1419. doi: 10.3390/agronomy13051419