基于动态特征选择的遥感图像目标检测算法

陈超; 赵巍

doi:10.13700/j.bh.1001-5965.2021.0300

基于动态特征选择的遥感图像目标检测算法

doi: 10.13700/j.bh.1001-5965.2021.0300

陈超,
赵巍^,

北京航空航天大学电子信息工程学院，北京 100191

详细信息

作者简介:
陈超：男，硕士研究生。主要研究方向：计算机视觉与识别技术

赵巍：女，博士，教授，博士生导师。主要研究方向：图像处理、计算机视觉与识别技术

通讯作者:
E-mail：zhao29wei@263.net

中图分类号: TP391.4
计量
- 文章访问数: 435
- HTML全文浏览量: 112
- PDF下载量: 57
- 被引次数: 0
出版历程
- 收稿日期: 2021-06-04
- 录用日期: 2021-08-13
- 网络出版日期: 2021-09-09
- 整期出版日期: 2023-03-30

Remote sensing target detection based on dynanic feature selection

CHEN Chao,
ZHAO Wei^,

School of Electronic and Information Engineering，Beihang University，Beijing 100191，China

More Information

Corresponding author: E-mail：zhao29wei@263.net

摘要

摘要:
在遥感图像目标检测领域内，旋转物体的检测存在挑战，卷积神经网络在提取信息时会受制于固定的空间结构，采样点无法聚焦于目标；遥感图像尺度变化大，不同物体需要具有不同尺度感受野的特征映射，具有单一尺度感受野的特征映射无法包含所有有效信息。基于此，提出了可变形对齐卷积，根据候选边框调节采样点，并根据特征映射学习采样点的细微偏移，使采样点聚焦于目标，从而实现动态特征选择；同时提出了基于可变形对齐卷积的感受野自适应模块，对具有不同尺度感受野的特征映射进行融合，自适应地调整神经元的感受野。在公开数据集上的大量实验验证了所提算法可以提高遥感图像目标检测的精度。
- 深度学习 /
- 卷积神经网络 /
- 遥感图像目标检测 /
- 特征选择 /
- 感受野
Abstract:
In the field of remote sensing image target detection, there still are challenges in oriented object detection. Convolutional neural network is subject to a fixed spatial structure when extracting information, and sampling locations cannot focus on objects. The scale of the remote sensing image varies greatly, and different objects require receptive fields of different scales to obtain feature map. Meanwhile, feature map with a single-scale receptive field cannot contain all effective information. In response to the first problem, deformable alignment convolution was proposed, which can first adjust the sampling locations according to the region of interest, and further learn slight offsets according to feature map, so that sampling locations can focus on objects and realize dynamic feature selection. For the second question, receptive field adaptive module based on deformable alignment convolution was proposed to fuse feature map with receptive fields of different scales and adaptively adjust the receptive field of neurons. Extensive experiments on public datasets showed that this method can improve the accuracy of remote sensing image target detection.
- deep learning /
- convolutional neural network /
- remote sensing image target detection /
- feature selection /
- receptive field

HTML全文

图 1 DFSNet总体结构

Figure 1. Overall structure of DFSNet

下载: 全尺寸图片幻灯片

图 2 RetinaNet总体结构

Figure 2. Overall structure of RetinaNet

下载: 全尺寸图片幻灯片

图 3 两种输出边框

Figure 3. Two output bounding boxes

下载: 全尺寸图片幻灯片

图 4 核尺寸为3×3的不同卷积的采样点对比

Figure 4. Comparison of sampling locations of different convolutions with kernel size of 3×3

下载: 全尺寸图片幻灯片

图 5 经过不同卷积的舰船采样点

Figure 5. Ship’s sampling locations after different convolutions

下载: 全尺寸图片幻灯片

图 6 动态特征选择层

Figure 6. Dynamic feature selection layer

下载: 全尺寸图片幻灯片

图 7 感受野自适应模块

Figure 7. Receptive field adaptive module

下载: 全尺寸图片幻灯片

图 8 IoU的解释

Figure 8. Explanation of IoU

下载: 全尺寸图片幻灯片

图 9 DOTA数据集可视化结果

Figure 9. Visual results on DOTA

下载: 全尺寸图片幻灯片

表 1 可变形对齐卷积与其他卷积对比

Table 1. Comparison between deformable alignment convolution and other convolutions

方法	mAP/%
标准卷积	71.17
可变形卷积	71.68
对齐卷积	72.45
可变形对齐卷积	73.18

下载: 导出CSV

表 2 DFSNet的消融实验对比

Table 2. Ablation studies of DFSNet

方法	感兴趣区域转换模块	动态特征选择层	感受野自适应模块	mAP
基线				68.05
DFSNet的不同设置	√			71.17
	√	√		73.18
	√	√	√	74.04

下载: 导出CSV

表 3 DFSNet与其他模型在DOTA数据集上的对比结果

Table 3. Comparison of DFSNet and other methods on DOTA

双/单阶段	模型	AP/%					mAP/%
双/单阶段	模型	PL	BD	BR	GTF	SV	mAP/%
双阶段	FR-O^[2]	79.42	77.13	17.70	64.05	35.30	54.72
	RoI Trans former^[8]	88.64	78.52	43.44	75.92	68.81	71.10
	CAD-Net^[20]	87.80	82.40	49.40	73.50	71.10	72.84
	SCRDet^[21]	89.98	80.65	52.09	68.36	68.36	71.89
单阶段	RetinaNet	88.82	81.74	44.44	65.72	67.11	69.57
	DRN^[16]	88.91	80.22	43.52	63.35	73.48	69.90
	R3Det^[22]	89.54	81.99	48.46	62.52	70.48	70.60
	DFSNet	89.12	77.40	52.05	73.47	78.02	74.01

双/单阶段	模型	AP/%					mAP/%
双/单阶段	模型	LV	SH	TC	BC	ST	mAP/%
双阶段	FR-O^[2]	38.02	37.16	89.41	69.64	59.28	58.70
	RoI Trans former^[8]	73.68	83.59	90.74	77.27	81.46	81.35
	CAD-Net^[20]	63.50	76.60	90.90	79.20	73.30	76.70
	SCRDet^[21]	60.32	72.41	90.85	87.94	86.86	79.68
单阶段	RetinaNet	55.82	72.77	90.55	82.83	76.30	75.65
	DRN^[16]	70.69	89.94	90.14	83.85	84.11	83.75
	R3Det^[22]	74.29	77.54	90.80	81.39	83.54	81.51
	DFSNet	79.31	87.35	90.90	85.13	84.90	85.52

双/单阶段	模型	AP/%					mAP/%
双/单阶段	模型	SBF	RA	HA	SP	HC	mAP/%
双阶段	FR-O^[2]	50.30	52.91	47.89	47.40	46.30	48.96
	RoI Trans former^[8]	58.39	53.54	62.83	58.93	47.67	56.27
	CAD-Net^[20]	48.40	60.90	62.00	67.00	62.20	60.10
	SCRDet^[21]	65.02	66.68	66.25	68.24	65.21	66.28
单阶段	RetinaNet	54.19	63.64	63.71	69.73	53.37	60.93
	DRN^[16]	50.12	58.41	67.62	68.60	52.50	59.45
	R3Det^[22]	61.97	59.82	65.44	67.46	60.05	62.95
	DFSNet	60.90	63.83	67.31	67.56	53.35	62.59

下载: 导出CSV

参考文献(22)

[1]	王彦情, 马雷, 田原. 光学遥感图像舰船目标检测与识别综述[J]. 自动化学报, 2011, 37(9): 1029-1039. WANG Y Q, MA L, TIAN Y. State-of-the-art of ship detection and recognition in optical remotely sensed imagery[J]. Acta Automatica Sinica, 2011, 37(9): 1029-1039(in Chinese).
[2]	XIA G S, BAI X, DING J, et al. DOTA: A large-scale dataset for object detection in aerial images[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 3974-3983.
[3]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic seg-mentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2014: 580-587.
[4]	GIRSHICK R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2015: 1440-1448.
[5]	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
[6]	HE K M, GKIOXARI G, DOLLÁR P, et al. Mask R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 2961-2969.
[7]	DAI J F, QI H Z, XIONG Y W, et al. Deformable convolutional networks[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 764-773.
[8]	DING J, XUE N, LONG Y, et al. Learning RoI Transformer for oriented object detection in aerial images[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019: 2849-2858.
[9]	LIU Z K, HU J G, WENG L B, et al. Rotated region based CNN for ship detection[C]//2017 IEEE International Conference on Image Processing(ICIP). Piscataway: IEEE Press, 2017: 900-904.
[10]	MA J Q, SHAO W Y, YE H, et al. Arbitrary-oriented scene text detection via rotation proposals[J]. IEEE Transactions on Multimedia, 2018, 20(11): 3111-3122. doi: 10.1109/TMM.2018.2818020
[11]	HAN J M, DING J, LI J, et al. Align deep features for oriented object detection[EB/OL]. (2021-07-12)[2021-07-12].https://arxiv. org/abs/2008.09397.
[12]	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2017: 2117-2125.
[13]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 2980-2988.
[14]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 770-778.
[15]	LI Y H, CHEN Y T, WANG N Y, et al. Scale-aware trident networks for object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 6054-6063.
[16]	PAN X J, REN Y Q, SHENG K K, et al. Dynamic refinement network for oriented and densely packed object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 11207-11216.
[17]	IOFFE S , SZEGEDY C. Batch Normalization: Accelerating deep network training by reducing internal covariate shift[C]//International Conference on Machine Learning, 2015: 448-456.
[18]	TAN M X, PANG R M, LE Q V. EfficientDet: Scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 10781-10790.
[19]	PASZKE A, GROSS S, MASSA F, et al. PyTorch: An imperative style, high-performance deep learning library[EB/OL]. (2019-12-03)[2021-06-01].https://arxiv. org/abs/11912.01703.
[20]	ZHANG G J, LU S J, ZHANG W. CAD-Net: A context-aware detection network for objects in remote sensing imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(12): 10015-10024. doi: 10.1109/TGRS.2019.2930982
[21]	YANG X, YANG J R, YAN J C, et al. SCRDet: Towards more robust detection for small, cluttered and rotated objects[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 8232-8241.
[22]	YANG X, LIU Q Q, YAN J C, et al. R3Det: Refined single-stage detector with feature refinement for rotating object[EB/OL]. (2020-12-08)[202-06-01].https://arxiv.org/abs/1908.05612v1.