基于anchor-free的光学遥感舰船关重部位检测算法

张冬冬; 王春平; 付强

doi:10.13700/j.bh.1001-5965.2022.0450

基于anchor-free的光学遥感舰船关重部位检测算法

doi: 10.13700/j.bh.1001-5965.2022.0450

陆军工程大学石家庄校区电子与光学工程系，石家庄 050003

详细信息

通讯作者:
E-mail：1418748495@qq.com

中图分类号: TP753
计量
- 文章访问数: 132
- HTML全文浏览量: 95
- PDF下载量: 6
- 被引次数: 0
出版历程
- 收稿日期: 2022-05-31
- 录用日期: 2022-08-08
- 网络出版日期: 2022-08-18
- 整期出版日期: 2024-04-29

Ship’s critical part detection algorithm based on anchor-free in optical remote sensing

Department of Electronic and Optical Engineering, People Liberation Army Engineering University-Shijiazhuang, Shijiazhuang 050003, China

More Information

Corresponding author: E-mail：1418748495@qq.com

摘要

摘要:
针对基于深度学习的遥感舰船检测算法存在精细化程度不足、检测效率低的问题，提出一种基于anchor-free的光学遥感舰船关重部位检测算法。所提算法以全卷积的单阶段目标检测(FCOS)算法为基准，在主干网络中引入全局上下文模块，提高网络的特征表达能力；为更好地描述目标的方向性，在预测阶段构建了具有方向表征能力的回归分支；对中心度函数进行优化，使其具备方向感知和自适应能力。实验结果表明：在自建舰船关重部位数据集和HRSC2016上，所提算法的平均精度(AP)比FCOS算法有显著提升；与其他算法相比，所提算法在检测速度和检测精度上均表现优越，具有较高的检测效率。
- 深度学习 /
- 遥感图像 /
- anchor-free /
- 舰船检测 /
- 关重部位检测 /
- 全卷积单阶段检测
Abstract:
Low detection effectiveness and inadequate refinement plague the existing deep learning-based remote sensing ship detection technique. To address the above problems, an optical remote sensing ship critical part detection algorithm based on anchor-free is proposed. The proposed algorithm takes fully convolutional one-stage object detection (FCOS) as the benchmark algorithm and introduces a global context module in the backbone network to improve the feature representation capability of the network. In the prediction step, a regression branch with orientation representation capabilities is built to more accurately describe the orientation of targets. The centrality function is optimized to make it direction-aware and adaptive. The experimental results show that the average precision (AP) of the proposed algorithm is significant improved over FCOS algorithm on the self-built ship critical part dataset and HRSC2016, respectively. Compared with other algorithms, the proposed algorithm has superior performance in both detection speed and detection accuracy and has high detection efficiency.
- deep learning /
- remote sensing /
- anchor-free /
- ship detection /
- critical part detection /
- fully convolutional one-stage object detection

HTML全文

图 1 FCOS算法^[21]的整体框架

Figure 1. Overall architecture of FCOS algorithm^[21]

下载: 全尺寸图片幻灯片

图 2 GCB结构

Figure 2. Structure of GCB

下载: 全尺寸图片幻灯片

图 3 嵌入GCB的残差结构

Figure 3. Residual structure of GCB embedded

下载: 全尺寸图片幻灯片

图 4 有无GCB情况下的特征可视化

Figure 4. Visualization of features with and without GCB

下载: 全尺寸图片幻灯片

图 5 改进后的检测头结构

Figure 5. Improved detection heads structure

下载: 全尺寸图片幻灯片

图 6 “中心到角点”的预测策略示意图

Figure 6. Diagram of “center to corner” prediction strategy

下载: 全尺寸图片幻灯片

图 7 不同中心度的权重热图

Figure 7. Heat map for weights of different center-ness

下载: 全尺寸图片幻灯片

图 8 CP-Ship数据集和HRSC2016^[24]数据集图像样例

Figure 8. Image examples of CP-Ship dataset and HRSC2016^[24]

下载: 全尺寸图片幻灯片

图 9 不同算法的P-R曲线

Figure 9. The P-R curves of different algorithms

下载: 全尺寸图片幻灯片

图 10 不同数据集上检测的可视化结果

Figure 10. Visualization of detection results on different datasets

下载: 全尺寸图片幻灯片

表 1 CP-Ship数据集和HRSC2016^[24]数据集详细信息

Table 1. Details of CP-Ship dataset and HRSC2016^[24] dataset

数据集	图像数量		关重部位数量 (CP-Ship)	舰船数量 (CP-Ship)
数据集	CP-Ship	HRSC2016^[24]	关重部位数量 (CP-Ship)	舰船数量 (CP-Ship)
训练集	812	849	2295	2342
测试集	203	212	561	634
总计	1015	1680	2856	2976

下载: 导出CSV

表 2 消融实验结果

Table 2. Experiment results of ablation

算法	嵌入GCB	改进回归分支	定向自适应中心度	AP/%
算法1				63.04
算法2	√			64.96
算法3		√		64.34
算法4	√	√		66.51
算法5	√	√	√	68.56

下载: 导出CSV

表 3 不同算法在CP-Ship测试集上的定量结果

Table 3. Quantitative results of different algorithms on the CP-Ship test set

算法	主干网络	图像大小/像素	锚框类型	TP	FP	AP/%	FPS	Paras/10⁶
Faster R-CNN^[7]	ResNet-50	800×608	水平框	411	152	65.78	19.2	41.12
Rotated Faster R-CNN	ResNet-50	800×608	旋转框	425	138	68.60	11.9	41.12
RetinaNet^[11]	ResNet-50	800×608	水平框	387	229	60.41	23.8	36.1
Rotated RetinaNet	ResNet-50	800×608	旋转框	388	162	60.61	13.9	36.13
R3Det^[18]	ResNet-50	800×608	旋转框	393	146	64.28	11.3	41.58
CornerNet^[18]	Hourglass-104	511×511	无锚框	419	1274	59.73	3.0	200.95
CenterNet^[19]	ResNet-18	512×512	无锚框	392	136	63.23	68.7	14.21
YOLOX-L^[20]	CSPDarkNet	640×640	无锚框	441	189	72.71	27.8	54.15
VarifocalNet^[29]	ResNet-50	800×608	无锚框	400	262	62.98	20.3	32.48
BBAVectors^[30]	ResNet-50	800×608	无锚框	448	198	68.31	14.5
SASM^[31]	ResNet-50	800×608	无锚框	397	214	63.85	13.5	36.6
FCOS^[21]	ResNet-50	800×608	无锚框	384	101	63.04	24.1	31.84
本文算法	ResNet-50+GCB	800×608	无锚框	417	89	68.56	21.3	32.49

下载: 导出CSV

表 4 不同算法在HRSC2016^[24]测试集上的定量结果

Table 4. Quantitative results of different algorithms on HRSC2016^[24] test set

算法	主干网络	图像大小	锚框类型	TP	FP	AP/%	FPS
Faster R-CNN^[7]	ResNet-50	800×608	水平框	557	93	84.33	20.6
Rotated Faster R-CNN	ResNet-50	800×608	旋转框	558	199	81.82	16.4
RetinaNet^[11]	ResNet-50	800×608	水平框	543	151	81.11	22.7
Rotated RetinaNet	ResNet-50	800×608	旋转框	467	66	68.10	21.2
R3Det^[18]	ResNet-50	800×608	旋转框	535	118	82.80	16.0
CornerNet^[18]	Hourglass-104	511×511	无锚框	521	2128	60.71	1.9
CenterNet^[19]	ResNet-18	512×512	无锚框	541	328	75.25	46.6
YOLOX-L^[20]	CSPDarkNet	640×640	无锚框	567	142	87.09	27.8
VarifocalNet^[29]	ResNet-50	800×608	无锚框	558	144	85.04	20.8
BBAVectors^[30]	ResNet-50	800×608	无锚框	561	219	86.19	15.8
SASM^[31]	ResNet-50	800×608	无锚框	540	431	80.40	20.0
FCOS^[21]	ResNet-50	800×608	无锚框	510	76	78.06	25.7
本文算法	ResNet-50+GCB	800×608	无锚框	558	52	84.73	22.5

下载: 导出CSV

参考文献(33)

[1]	KANJIR U, GREIDANUS H, OŠTIR K. Vessel detection and classification from spaceborne optical images: A literature survey[J]. Remote Sensing of Environment, 2018, 207: 1-26. doi: 10.1016/j.rse.2017.12.033
[2]	刘俊琦, 李智, 张学阳. 可见光遥感图像海面目标检测技术综述[J]. 计算机科学, 2020, 47(3): 116-123. LIU J Q, LI Z, ZHANG X Y. Review of maritime target detection in visible bands of optical remote sensing images[J]. Computer Science, 2020, 47(3): 116-123 (in Chinese).
[3]	李晓斌, 江碧涛, 杨渊博, 等. 光学遥感图像目标检测技术综述[J]. 航天返回与遥感, 2019, 40(4): 95-104. doi: 10.3969/j.issn.1009-8518.2019.04.011 LI X B, JIANG B T, YANG Y B, et al. A survey on object technology in optical remote sensing images[J]. Spacecraft recovery & Remote Sensing, 2019, 40(4): 95-104 (in Chinese). doi: 10.3969/j.issn.1009-8518.2019.04.011
[4]	张财广, 熊博莅, 匡纲要. 光学卫星遥感图像舰船目标检测综述[J]. 电波科学学报, 2020, 35(5): 637-647. ZHANG C G, XIONG B L, KUANG G Y. A survey of ship detection in optical satellite remote sensing images[J]. Chinese Journal of Radio Science, 2020, 35(5): 637-647 (in Chinese).
[5]	张永胜, 计一飞, 董臻. 时-空变化的背景电离层对星载合成孔径雷达方位向成像的影响分析[J]. 电子与信息学报, 2021, 43(10): 2781-2789. ZHANG Y S, JI Y F, DONG Z. Research on background ionospheric impacts imposed by spatio-temporal variations on spaceborne synthetic aperture radar azimuth imaging[J]. Journal of Electronics & Information Technology, 2021, 43(10): 2781-2789 (in Chinese).
[6]	GIRSHICK R. Fast R-CNN[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2015: 1440-1448.
[7]	REN S 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
[8]	HE K M, GKIOXARI G, DOLLÁR P, et al. Mask R-CNN[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 2980-2988.
[9]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot MultiBox detector[C]//Proceedings of the European Conference on Computer Vision. Berlin: Springer, 2016: 21-37.
[10]	REDMON J, FARHADI A. YOLO9000: Better, faster, stronger[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2017: 6517-6525.
[11]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 2999-3007.
[12]	史文旭, 江金洪, 鲍胜利. 基于特征融合的遥感图像舰船目标检测算法[J]. 光子学报, 2020, 49(7): 57-67. SHI W X, JIANG J H, BAO S L. Ship detection method in remote sensing image based on feature fusion[J]. Acta Photonica Sinica, 2020, 49(7): 57-67 (in Chinese).
[13]	吴止锾, 李磊, 高永明. 遥感图像舰船检测的旋转卷积集成YOLOv3模型[J]. 计算机工程与应用, 2019, 55(22): 146-151. WU Z H, LI L, GAO Y M. Rotation convolution ensemble YOLOv3 model for ship detection in remote sensing images[J]. Computer Engineering and Applications, 2019, 55(22): 146-151 (in Chinese).
[14]	YANG X, YAN J C, FENG Z M, et al. R3Det: Refined single-stage detector with feature refinement for rotating object[EB/OL]. (2020-12-08) [2020-12-10]. http://arxiv.org/abs/1908.05612.pdf.
[15]	赵微, 惠斌, 张玉晓. 红外舰船目标的要害点检测算法[J]. 红外与激光工程, 2014, 43(1): 48-52. ZHAO W, HUI B, ZHANG Y X. Detection algorithms of aim points of IR warship images[J]. Infrared and Laser Engineering, 2014, 43(1): 48-52 (in Chinese).
[16]	李晨瑄, 顾佼佼, 王磊, 等. 多尺度特征融合的Anchor-Free轻量化舰船要害部位检测算法[J]. 北京航空航天大学学报, 2022, 48(10): 2006-2019. LI C X, GU J J, WANG L, et al. Warship’s vital parts detection algorithm based on lightweight anchor-free network with multi-scale feature fusion[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(10): 2006-2019 (in Chinese).
[17]	张长伦, 张翠文, 王恒友, 等. 基于注意力机制的NMS在目标检测中的研究[J]. 电子测量技术, 2021, 44(19): 82-88. ZHANG C L, ZHANG C W, WANG H Y, et al. Research on non-maximum suppression based on attention mechanism in object detection[J]. Electronic Measurement Technology, 2021, 44(19): 82-88 (in Chinese).
[18]	LAW H, DENG J. CornerNet: Detecting objects as paired keypoints[C]//Proceedings of the European Conference on Computer Vision. Berlin: Springer, 2018: 765-781.
[19]	DUAN K W, BAI S, XIE L X, et al. CenterNet: Keypoint triplets for object detection[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2020: 6568-6577.
[20]	GE Z, LIU S T, WANG F, et al. YOLOX: Exceeding YOLO series in 2021[EB/OL]. (2021-08-06) [2021-08-10]. http://arxiv.org/abs/2107.08430.pdf.
[21]	TIAN Z, SHEN C H, CHEN H, et al. FCOS: A simple and strong anchor-free object detector[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(4): 1922-1933.
[22]	CAO Y, XU J R, LIN S, et al. GCNet: Non-local networks meet squeeze-excitation networks and beyond[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision Workshop. Piscataway: IEEE Press, 2020: 1971-1980.
[23]	LANG S, VENTOLA F, KERSTING K. DAFNe: A one-stage anchor-free approach for oriented object detection[EB/OL]. (2022-05-30) [2022-06-02]. http://arxiv.org/abs/2109.06148.pdf.
[24]	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]//Proceedings of the 6th International Conference on Pattern Recognition Applications and Methods. Porto: Science and Technology Publications, 2017.
[25]	MA Y, CHAI L, JIN L Z, et al. AVS-YOLO: Object detection in aerial visual scene[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2022, 36(1): 2250004. doi: 10.1142/S0218001422500045
[26]	GAO X, RAM S, PHILIP R C, et al. Selecting post-processing schemes for accurate detection of small objects in low-resolution wide-area aerial imagery[J]. Remote Sensing, 2022, 14(2): 255. doi: 10.3390/rs14020255
[27]	LI K, WAN G, CHENG G, et al. Object detection in optical remote sensing images: A survey and a new benchmark[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 159: 296-307. doi: 10.1016/j.isprsjprs.2019.11.023
[28]	XIA G S, BAI X, DING J, et al. DOTA: A large-scale dataset for object detection in aerial images[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 3974-3983.
[29]	ZHANG H Y, WANG Y, DAYOUB F, et al. VarifocalNet: An IoU-aware dense object detector[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2021: 8510-8519.
[30]	YI J R, WU P X, LIU B, et al. Oriented object detection in aerial images with box boundary-aware vectors[C]//Proceedings of the 2021 IEEE Winter Conference on Applications of Computer Vision. Piscataway: IEEE Press, 2021: 2149-2158.
[31]	HOU L P, LU K, XUE J A, et al. Shape-adaptive selection and measurement for oriented object detection[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Washton D. C.: AAAI, 2022, 36(1): 923-932.
[32]	YANG Y, PAN Z X, HU Y X, et al. CPS-det: An anchor-free based rotation detector for ship detection[J]. Remote Sensing, 2021, 13(11): 2208. doi: 10.3390/rs13112208
[33]	ZHANG X R, WANG G C, ZHU P, et al. GRS-det: An anchor-free rotation ship detector based on gaussian-mask in remote sensing images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(4): 3518-3531. doi: 10.1109/TGRS.2020.3018106