面向遥感图像舰船识别的轻量化目标检测算法

doi:10.13229/j.cnki.jdxbgxb.20240753

摘要/Abstract

摘要：

针对现有基于深度学习的目标检测算法因模型复杂而不利于星上部署的问题，本文提出了一种面向光学遥感图像舰船识别任务的多尺度特征增强轻量化检测算法（MFLDet）。首先，为减少算法的参数量和计算量，构造轻量化网络架构PG-HGNet作为主干网络，同时构造轻量化跨尺度特征融合网络实现特征交互融合；其次，设计多尺度特征增强模块（MFEM），以适应遥感图像中舰船目标的尺度差异性，提高检测精度；最后，引入MPDIoU边界框损失函数，以适应预测框与真实框长宽比相同但长宽绝对值不同的情况。在数据集HRSC2016上进行的对比实验结果表明，与基准模型相比，MFLDet的参数量和计算量分别下降了55%和23.5%，而平均精确度仅下降0.2个百分点，在降低算法复杂度的同时实现了与精确度的平衡。综合考量，所提方法在轻量化水平和检测精度方面均优于其他对比方法。

关键词: 计算机视觉, 目标检测, 遥感图像, 特征增强

Abstract:

To solve the problem of existing deep learning-based object detection algorithms being unsuitable for onboard deployment due to their complexity， a multi-scale feature enhanced lightweight detection algorithm （MFLDet） for ship recognition tasks in optical remote sensing images is proposed. Firstly， to reduce the algorithm's parameter and computation load， a lightweight network architecture， PG-HGNet， is constructed as the backbone network， and constructs a lightweight cross-scale feature fusion network for interactive feature integration. Secondly， a multi-scale feature enhancement module （MFEM） is designed to accommodate the scale variability of ship targets in remote sensing images， thereby enhancing detection accuracy. Finally， the MPDIoU bounding box loss function is introduced to adjust for cases where the predicted and actual bounding boxes share the same aspect ratio but differ in absolute dimensions. Comparative experiments conducted on the HRSC2016 dataset demonstrate that， compared to the baseline model， MFLDet achieves a 55% reduction in parameters and a 23.5% reduction in computation， with only a 0.2 percentage points decrease in average precision， thus effectively balancing complexity and accuracy. Overall， the proposed method surpasses other comparative methods in terms of both lightweight level and detection precision.

Key words: computer vision, object detection, remote sensing image, feature enhancement

中图分类号:

TP391.4

张帆,王景波,朱杨,刘海英,郑煜,王文华. 面向遥感图像舰船识别的轻量化目标检测算法[J]. 吉林大学学报(工学版), 2026, 56(2): 533-542.

Fan ZHANG,Jing-bo WANG,Yang ZHU,Hai-ying LIU,Yu ZHENG,Wen-hua WANG. Lightweight object detection algorithm for ship recognition in remote sensing image[J]. Journal of Jilin University(Engineering and Technology Edition), 2026, 56(2): 533-542.

图/表 13

图1

图2

图3

图4

图5

图6

图7

表1

图8

表2

表3

表4

表5

参考文献 21

[1]	董超, 刘晶红, 徐芳, 等. 光学遥感图像舰船目标快速检测方法[J]. 吉林大学学报: 工学版, 2019, 49(4):1369-1376.
	Dong Chao, Liu Jing-hong, Xu Fang, et al. Fast ship detection in optical remote sensing images[J].Journal of Jilin University (Engineering and Technology Edition), 2019, 49(4): 1369-1376.
[2]	Liu W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]∥14th European Conference on Computer Vision-ECCV 2016. Cham, Switzerland: Springer Internation Publishing, 2016: 21-37.
[3]	Ren S Q, He K, 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.
[4]	Cai Z W, Vasconcelos N.Cascade R-CNN: delving into high quality object detection[C]∥2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2018: 6154-6162.
[5]	李永刚, 朱卫纲, 黄琼男, 等. 复杂背景下 SAR 图像近岸舰船目标检测[J]. 系统工程与电子技术, 2022, 44(10): 3096-3103.
	Li Yong-gang, Zhu Wei-gang, Huang Qiong-nan, et al. Near shore ship target detection with SAR images in complex background[J].Systems Engineering and Electronics, 2022, 44(10): 3096-3103.
[6]	成倩, 李佳, 杜娟. 基于 YOLOv5 的光学遥感图像舰船目前标检测算法[J]. 系统工程与电子技术, 2023, 45(5): 1270-1276.
	Cheng Qian, Li Jia, Du Juan.Ship target detection algorithm of optical remote sensing image based on YOLOv5[J]. Systems Engineering and Electronics, 2023, 45(5): 1270-1276.
[7]	Yi H, Liu B, Zhao B, et al. Small object detection algorithm based on improved YOLOv8 for remote sensing[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 17: 1734-1747.
[8]	车思文, 汪宇玲. 一种改进的YOLOv7光学遥感图像舰船目标检测算法[J]. 电光与控制, 2024, 31(5): 34-39.
	Che Si-wen, Wang Yu-ling.An improved YOLOv7-based ship target detection algorithm for optical remote sensing images[J]. Electronics Optics & Control, 2024, 31(5): 34-39.
[9]	Han K, Wang Y H, Tian Q, et al. GhostNet: More features from cheap operations[C]∥Proceedings of the 2020 IEEE/CVF Conferencef on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 1580-1589.
[10]	Le H, Borji A. What are the receptive, effective receptive, and projective fields of neurons in convolutional neural networks?[J/OL].[2024-09-11].
[11]	Luo W, Li Y, Urtasun R, et al. Understanding the effective receptive field in deep convolutional neural networks[J/OL]. [2024-09-11].
[12]	Ding X H, Zhang X Y, Han J G, et al. Scaling up your kernels to 31×31: revisiting large kernel design in CNNs[C]∥2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2022: 11953-11965.
[13]	Ma N N, Zhang X Y, Zheng H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design[C]∥Proceedings of the European Conference on Computer Vision. Cham, Switzerland: Springer Interaction Publishing, 2018: 116-131.
[14]	Xie X R, Sun F X, Lin J, et al. Fast-ABC: a fast architecture for bottleneck-like based convolutional neural networks[C]∥2019 IEEE Computer Society Annual Symposium on VLSI. Piscataway, NJ: IEEE, 2019: 1-6.
[15]	Ma S L, Xu Y. MPDIoU: a loss for effcient and accurate bounding box regression[J/OL]. [2024-09-11].
[16]	Liu Z K, Yuan L, Weng L B, et al. A high resolution optical satellite image dataset for ship recognition and some new baselines[C]∥International Conference on Pattern Recognition Applications and Methods. New York: INSTICC, 2017, 2: 324-331.
[17]	Pang Y H, Zhang Y M, Kong Q L, et al. SOCDet: a lightweight and accurate oriented object detection network for satellite on-orbit computing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1-15.
[18]	Cheng G, Wang J B, Li K, et al. Anchor-free oriented proposal generator for object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-11.
[19]	Li H, Zhang S, Li X, et al. DetectorNet: Transformer enhanced spatial temporal graph neural network for traffic prediction[C]∥Proceedings of the 29th International Conference on Advances in Geographic Information Systems. New York: ACM, 2021: 133-136.
[20]	Zhou L, Wei H R, Li H, et al. Arbitrary-oriented object detection in remote sensing images based on polar coordinates[J]. IEEE Access, 2020, 8: 223373-223384.
[21]	Yang J R, Wu B, Zhang H Y, et al. Rotating frame labeling and improved YOLOv7-based algorithm for vehicle object detection[C]∥Proceedings of the 2023 IEEE 6th International Conference on Pattern Recognition and ArtificialIntelligence. Piscataway, NJ: IEEE, 2023: 170-175.

相关文章 15

[1]	李军,杨飞帆,龚胜,周科宇. 基于视觉的轻量化路面异常检测算法[J]. 吉林大学学报(工学版), 2026, 56(2): 488-496.
[2]	霍震,金立生,华强,贺阳. 基于边缘特征引导的智能汽车语义分割方法[J]. 吉林大学学报(工学版), 2025, 55(9): 3032-3041.
[3]	郭志荣,李刚. 基于高斯核密度估计的高速运动目标检测算法[J]. 吉林大学学报(工学版), 2025, 55(8): 2741-2745.
[4]	李文辉,杨晨. 基于对比学习文本感知的小样本遥感图像分类[J]. 吉林大学学报(工学版), 2025, 55(7): 2393-2401.
[5]	徐慧智,郝东升,徐小婷,蒋时森. 基于深度学习的高速公路小目标检测算法[J]. 吉林大学学报(工学版), 2025, 55(6): 2003-2014.
[6]	于营,王春平,寇人可,杨博雄,王雷,赵福军,付强. 多时相高分辨率卫星遥感图像语义分割算法[J]. 吉林大学学报(工学版), 2025, 55(6): 2131-2137.
[7]	侯越,郭劲松,林伟,张迪,武月,张鑫. 分割可跨越车道分界线的多视角视频车速提取方法[J]. 吉林大学学报(工学版), 2025, 55(5): 1692-1704.
[8]	田丽,贾煜辉. 改进YOLOv5s算法的高光谱遥感图像目标检测[J]. 吉林大学学报(工学版), 2025, 55(5): 1742-1748.
[9]	刘广文,赵绮莹,王超,高连宇,才华,付强. 基于渐进递归的生成对抗单幅图像去雨算法[J]. 吉林大学学报(工学版), 2025, 55(4): 1363-1373.
[10]	刘广文,谢欣月,付强,才华,王伟刚,马智勇. 基于时空模板焦点注意的Transformer目标跟踪算法[J]. 吉林大学学报(工学版), 2025, 55(3): 1037-1049.
[11]	程德强,刘规,寇旗旗,张剑英,江鹤. 基于自适应大核注意力的轻量级图像超分辨率网络[J]. 吉林大学学报(工学版), 2025, 55(3): 1015-1027.
[12]	郑利民,陈双,李刚. YOLOv5网络算法下交通监控视频违章车辆多目标检测[J]. 吉林大学学报(工学版), 2025, 55(2): 693-699.
[13]	才华,郑延阳,付强,王晟宇,王伟刚,马智勇. 基于多尺度候选融合与优化的三维目标检测算法[J]. 吉林大学学报(工学版), 2025, 55(2): 709-721.
[14]	李云红,王梅,苏雪平,李丽敏,张富星,郝特吉. 结合注意力与上下文融合的遥感图像道路提取[J]. 吉林大学学报(工学版), 2025, 55(12): 4034-4044.
[15]	曹玉东,廖鑫林,陈鑫,贾旭. 融合深度主动学习的视觉目标检测模型[J]. 吉林大学学报(工学版), 2025, 55(11): 3697-3704.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

实验编号	mAP_0.5/%	mAP_0.5：0.95/%	参数量/M	FLOPs/G
1	89.8	77	1.50	6.0
2	89.7	75.2	1.76	7.3
3	90.2	76.5	1.97	7.4
4	90.9	78.8	1.35	6.2

模型	mAP_0.5/ %	mAP_0.5：0.95/ %	参数量/ M	FLOPs/ G
基线	91.1	78.7	3.01	8.1
+PG-HGNet	89.8	76.1	2.31	6.8
+LW-BiFPN	91.7	79.1	1.99	7.1
+MFEM	92.2	80.7	3.33	8.7
+MPDIoU	91.9	79.8	3.01	8.1
+PG-HGNet+ LW-BiFPN	90.3	77.2	1.29	5.8
+MFEM+ MPDIoU	92.5	81.1	3.33	8.7
+all	90.9	78.8	1.35	6.2

算法	mAP_0.5/ %	mAP_0.5：0.95/ %	参数量/ M	FLOPs/ G
Faster R-CNN	90.2	65.4	136.69	369.7
SSD	64.5	37.7	23.61	60.8
Yolov3-tiny	90.9	74.6	12.13	18.9
Yolov5nano	90.8	77.3	2.50	7.1
Yolov7-tiny	90.6	71.2	6.23	13.9
Yolov10nano	87.1	74.2	2.69	8.2
本文	90.9	78.8	1.35	6.2

算法	mAP_0.5/%	参数量/M	FLOPs/G
SOCDet^［17］	94.28	10.01	16.72
AOPG^［18］	90.58	33.87	164.33
PP-PicoDet-S^［19］	90.89	1.08	13.92
LO-Det^［20］	90.17	6.78	10.98
VOD-YOLOv7^［21］	92.1	7.45	116
本文	90.9	1.35	6.2

算法	mAP_0.5/ %	mAP_0.5：0.95/ %	参数量/ M	FLOPs/ G
Yolov3-tiny	91.3	56.8	12.1	18.9
Yolov5nano	97.1	61	2.5	7.1
Yolov7-tiny	96.4	61.3	6.23	13.9
Yolov8nano	96.8	61.7	3	8.1
Yolov10nano	96.5	61.1	2.69	8.2
本文	96.6	60.9	1.35	6.2