用于遥感图像变化检测的全尺度特征聚合网络

刘国强; 房胜; 李哲

doi:10.13700/j.bh.1001-5965.2021.0522

用于遥感图像变化检测的全尺度特征聚合网络

doi: 10.13700/j.bh.1001-5965.2021.0522

山东科技大学计算机科学与工程学院, 青岛 266590

基金项目:

山东省自然科学基金 ZR2020MF132

详细信息

通讯作者:
房胜, E-mail: fangsheng@tsinghua.org.cn

中图分类号: TP751
计量
- 文章访问数: 625
- HTML全文浏览量: 209
- PDF下载量: 54
- 被引次数: 0
出版历程
- 收稿日期: 2021-09-06
- 录用日期: 2021-10-01
- 网络出版日期: 2021-10-29
- 整期出版日期: 2022-08-20

A full-scale feature aggregation network for remote sensing image change detection

College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China

Funds:

Shandong Provincial Natural Science Foundation ZR2020MF132

More Information

Corresponding author: FANG Sheng, E-mail: fangsheng@tsinghua.org.cn

摘要

摘要:
变化检测(CD)是遥感的一项重要任务，通常面临许多伪变化和较大的尺度变化。目前的方法主要侧重于对差异特征的建模，忽略了从原始图像中提取足够的信息，影响了特征的识别能力，难以稳定地区分出变化区域。针对以上问题，提出了一种全尺度特征聚合网络(FFANet)来更充分地利用原始图像特征，促使生成的特征表示在语义上更丰富、在空间上更准确，从而提高了网络对小目标和目标边缘的检测性能。同时，拓展了深监督来结合多尺度的预测图，以促使不同对象在更合适的尺度上进行检测，从而提升了网络对对象尺度变化的鲁棒性。在CDD数据集上，相比于基线网络，所提方法仅增加了1.01×10⁶的参数量，就将F₁分数提升了0.034。
- 变化检测(CD) /
- 深监督 /
- 全尺度特征聚合 /
- 多尺度预测 /
- 遥感图像
Abstract:
Change detection (CD) is an important task of remote sensing, always facing many pseudo changes and large scale variations. However, existing methods mainly focus on modeling difference features and neglect extracting sufficient information from the original images, which affects feature discrimination and makes it difficult to distinguish change regions stably. To address these problems, a full-scale feature aggregation network (FFANet) is proposed to make fuller use of the original image features, which drives the generated feature representations to be semantically richer and spatially more precise, thus improving the network's detection performance for small targets and target edges. Deep supervision is also extended to combine multi-scale prediction maps to drive the detection of different objects at more appropriate scales, thus improving the robustness of the network to object scale variations. On the CDD dataset, our proposed method improves the F₁-score by 0.034 compared to the baseline network by increasing the number of parameters by only 1.01×10⁶.
- change detection (CD) /
- deep supervision /
- full-scale feature aggregation /
- multi-scale prediction /
- remote sensing images

HTML全文

图 1 FFANet的网络结构

Figure 1. Network architecture of FFANet

下载: 全尺寸图片幻灯片

图 2 用于全尺度特征融合的编码器

Figure 2. Encoders for full-scale feature fusion

下载: 全尺寸图片幻灯片

图 3 特征图S₃^(N)的构建

Figure 3. Construction of feature map S₃^(N)

下载: 全尺寸图片幻灯片

图 4 用于生成多尺度差异特征的解码器

Figure 4. Decoder for generating multi-scale difference feature

下载: 全尺寸图片幻灯片

图 5 多尺度分类器

Figure 5. Multi-scale classifier

下载: 全尺寸图片幻灯片

图 6 CDD测试集上不同方法的可视化

Figure 6. Visualization of different methods on CDD test set

下载: 全尺寸图片幻灯片

图 7 普通的编码器

Figure 7. Plain encoder

下载: 全尺寸图片幻灯片

图 8 编码器和分类器的对比实验

Figure 8. Comparison experiments of encoder and classifier

下载: 全尺寸图片幻灯片

表 1 CDD和LEVIR数据集上FFANet与其他方法的对比

Table 1. Comparison of FFANet with other methods on CDD and LEVIR datasets

方法	参数量/10⁶	计算量/GFLOPs	CDD			LEVIR
方法	参数量/10⁶	计算量/GFLOPs	P	R	F₁	P	R	F₁
FC-EF	1.35	7.14	0.749	0.494	0.595	0.754	0.730	0.742
FC-Siam-conc	1.55	10.64	0.779	0.622	0.692	0.852	0.736	0.790
FC-Siam-diff	1.35	9.44	0.786	0.588	0.673	0.861	0.687	0.764
IFN	35.72	164.53	0.950	0.861	0.903	0.903	0.876	0.889
DASNet	16.25	113.09	0.914	0.925	0.919	0.811	0.788	0.799
SNUNet	12.03	109.62	0.956	0.949	0.953	0.889	0.874	0.881
FFANet	8.64	28.81	0.962	0.957	0.960	0.925	0.892	0.908
注：GFLOPs指10⁹次浮点运算。

下载: 导出CSV

表 2 CDD数据集上的消融实验

Table 2. Ablation experiments on CDD data set

序号	编码器	分类器	参数量/10⁶	P	R	F₁
①	×	×	7.63	0.955	0.900	0.926
②	√	×	8.64	0.960	0.942	0.951
③	×	√	7.63	0.957	0.943	0.950
④	√	√	8.64	0.962	0.957	0.960

下载: 导出CSV

参考文献(21)

[1]	LEICHTLE T, GEIß C, LAKES T, et al. Class imbalance in unsupervised change detection-A diagnostic analysis from urban remote sensing[J]. International Journal of Applied Earth Observation and Geoinformation, 2017, 60: 83-98. doi: 10.1016/j.jag.2017.04.002
[2]	USEYA J, CHEN S B, MUREFU M. Cropland mapping and change detection: Toward zimbabwean cropland inventory[J]. IEEE Access, 2019, 7: 53603-53620. doi: 10.1109/ACCESS.2019.2912807
[3]	QIAO H J, WAN X, WAN Y C, et al. A novel change detection method for natural disaster detection and segmentation from video sequence[J]. Sensors (Basel), 2020, 20(18): 5076. doi: 10.3390/s20185076
[4]	RONG K, FANG B, CHEN G, et al. Progressive domain adaptation for change detection using season-varying remote sensing images[J]. Remote Sensing, 2020, 12(22): 3815. doi: 10.3390/rs12223815
[5]	VU V T, PETTERSSON M I, MACHADO R, et al. False alarm reduction in wavelength-resolution SAR change detection using adaptive noise canceler[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(1): 591-599. doi: 10.1109/TGRS.2016.2611684
[6]	BOVOLO F, BRUZZONE L. A novel theoretical framework for unsupervised change detection based on CVA in polar domain[C]//2006 IEEE International Symposium on Geoscience and Remote Sensing. Piscataway: IEEE Press, 2006: 379-382.
[7]	DENG J S, WANG K, DENG Y H, et al. PCA-based land-use change detection and analysis using multitemporal and multisensor satellite data[J]. International Journal of Remote Sensing, 2008, 29(15-16): 4823-4838.
[8]	LI W, LU M, CHEN X W. Automatic change detection of urban land-cover based on SVM classification[C]//2015 IEEE International Symposium on Geoscience and Remote Sensing. Piscataway: IEEE Press, 2015: 1686-1689.
[9]	DAUDT R C, SAUX B L, BOULCH A. Fully convolutional Siamese networks for change detection[C]//IEEE International Conference on Image Processing (ICIP). Piscataway: IEEE Press, 2018: 4063-4067.
[10]	ZHANG C Z, PENG Y, TAPETE D, et al. A deeply supervised image fusion network for change detection in high resolution bi-temporal remote sensing images[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 166: 183-200. doi: 10.1016/j.isprsjprs.2020.06.003
[11]	FANG S, LI K Y, SHAO J Y, et al. SNUNet-CD: A densely connected Siamese network for change detection of VHR images[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19: 1-5.
[12]	ZHOU Z, SIDDIQUEE M, TAJBAKHSH N, et al. UNet++: A nested U-Net architecture for medical image segmentation[C]// Deep Learning in Medical Image Analysis (DLMIA) Workshop, 2018: 3-11.
[13]	CHEN H, SHI Z W. A spatial-temporal attention-based method and a new dataset for remote sensing image change detection[J]. Remote Sensing, 2020, 12(10): 1662. doi: 10.3390/rs12101662
[14]	CHEN J, YUAN Z Y, PENG J, et al. DASNet: Dual attentive fully convolutional siamese networks for change detection of high resolution satellite images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2020, 14: 1194-1206.
[15]	RONNEBERGER O, FISCHER P, BROX T. U-Net: Convolutional networks for biomedical image segmentation[C]//Medical Image Computing and Compwter-Assisted Intervertion-MICCAI 2015, 2015.
[16]	SHELHAMER E, LONG J, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640-651. doi: 10.1109/TPAMI.2016.2572683
[17]	LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2017: 936-944.
[18]	BAO T E, FU C Q, FANG T, et al. PPCNET: A combined patch-level and pixel-level end-to-end deep network for high-resolution remote sensing image change detection[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 17(10): 1797-1801. doi: 10.1109/LGRS.2019.2955309
[19]	HUAN R, ZHOU M, XING Y, et al. Change detection with various combinations of fluid pyramid integration networks[J]. Neurocomputing, 2021, 437: 84-94. doi: 10.1016/j.neucom.2021.01.030
[20]	YANG K, LIU Z, LU Q, et al. Multi-scale weighted branch network for remote sensing image classification[C]//2010 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019.
[21]	LEBEDEV M A, VIZILTER Y V, VYGOLOV OLEG, et al. Change detection in remote sensing images using conditional adversarial networks[J]. The International Archives of the photogrammetry, Rewote Sensing and Spatial Information Sciences, 2018, 48(2): 565-571.