采用神经网络架构搜索的遥感影像分割方法

doi:10.19665/j.issn1001-2400.2021.05.007

摘要/Abstract

摘要：

由于传统的深度卷积神经网络分割高分辨率遥感影像需人工设计网络架构,过度依赖专家经验,耗时费力,且网络泛化能力较差,因此,提出一种资源平衡型部分通道采样的神经网络架构搜索方法。首先,在网络架构参数中添加资源平衡项,提升搜索算法稳定性,同时减小剪枝过程中产生的更新不平衡和离散化误差;其次,选择部分通道进行搜索空间的混合操作,以节省计算资源,提升搜索效率,缓解网络过拟合;最后,根据高分辨率遥感影像地物复杂、分布离散及空间范围广等特点,引入Gumbel-Softmax Trick方法从非连续概率分布进行采样,以提高采样效率。在WHUBuilding数据集上MIoU语义分割评价指标达到90.93%,在GID数据集上MIoU语义分割评价指标达到69.53%,优于SegNet、U-Net、Deeplab v3+、NAS-HRIS等网络模型。实验结果表明,新方法能高效地自动搜索出分割高分辨率遥感影像的网络架构,具有分割精度高、计算资源占用率低的特点。

关键词: 深度学习, 高分辨率遥感影像, 神经网络架构搜索, 影像分割, 卷积神经网络

Abstract:

High-resolution remote sensing image segmentation based on the traditional deep convolutional neural network needs hand-crafted architectures,which is excessively dependent on expert experience,time-consuming and laborious,and the network generalization ability is poor.A neural network architecture search method of resource balanced partial channel sampling is presented in this study.First,the resource-balanced strategy is added to the network architecture parameter to minimize the updating imbalances and discretization discrepancy during pruning,so the stability of the search algorithm is improved.Second,the partial channel is sampled for the mixed operation in search space,which can effectively reduce the computing cost,improve the search efficiency and alleviate the problem of network overfitting.Finally,according to the characteristic of complex features,the discrete distribution and the wide spatial range of high resolution remote sensing images,the Gumbel-Softmax trick is introduced to improve the sampling efficiency and make the sampling process backpropagate.The proposed method can achieve 90.93% and 69.53% MIoU on the WHUBuilding and GID dataset,respectively,which outperforms the prior work like SegNet,U-Net,Deeplab v3+ and NAS-HRIS.Experimental results show that this proposed method can help search the architecture for high resolution remote sensing image segmentation efficiently and automatically,and has the advantages of a high segmentation accuracy and low computing resources.

Key words: deep learning, high-resolution remote sensing, network architecture search, image segmentation, convolutional neural network

中图分类号:

TP391

周鹏,杨军. 采用神经网络架构搜索的遥感影像分割方法[J]. 西安电子科技大学学报, 2021, 48(5): 47-57.

ZHOU Peng,YANG Jun. Semantic segmentation of remote sensing images based on neural architecture search[J]. Journal of Xidian University, 2021, 48(5): 47-57.

图/表 10

图1

图2

表1

图3

图4

图5

表2

图6

表3

表4

参考文献 22

[1]	王春艳, 徐爱功, 孙川, 等. 融入空间关系的GMM全色高分辨率遥感影像监督分割方法[J]. 电子与信息学报, 2017, 39(5):1071-1078.
	WANG Chunyan, XU Aigong, SUN Chuan, et al. Surpervised Segmentation Algorithm Based on GMM with Spatial Relationship for High Resolution Ranchromatic Remote Sensing Image[J]. Journal of Electronics & Information Technology, 2017, 39(5):1071-1078.
[2]	刘如意, 宋建锋, 权义宁, 等. 一种自动的高分辨率遥感影像道路提取方法[J]. 西安电子科技大学学报, 2017, 44(1):100-105.
	LIU Ruyi, SONG Jianfeng, QUAN Yining, et al. Automatic Road Extraction Method for High-Resolution Remote Sensing Images[J]. Journal of Xidian University, 2017, 44(1):100-105.
[3]	杨军, 王恒亮. 结合巴氏系数和灰度共生矩阵的遥感影像分割[J]. 遥感信息, 2019, 34(3):19-25.
	YANG Jun, Wang Hengliang. Segmentation of Remote Sensing Image Combined with Bhattacharyya Coefficient and Gray Level Co-Occurrence Matrix[J]. Remote Sensing Information, 2019, 34(3):19-25.
[4]	汪梓艺, 苏育挺, 刘艳艳, 等. 一种改进DeeplabV3网络的烟雾分割算法[J]. 西安电子科技大学学报, 2019, 46(6):52-59.
	WANG Ziyi, SU Yuting, LIU Yanyan, et al. Algorithm for Segmentation of Smoke Using the Improved DeeplabV3 Network[J]. Journal of Xidian University, 2019, 46(6):52-59.
[5]	杨军, 裴剑杰. 一种改进的隐马尔可夫随机场遥感影像分割算法[J]. 遥感技术与应用, 2018, 33(5):857-865.
	YANG Jun, PEI Jianjie. Improved HMRF from Remote Sensing Image Segmentation[J]. Remote Sensing Technology and Application, 2018, 33(5):857-865.
[6]	ELSKEN T, METZEN J, HUTTER F. Neural Architectures Search:A Survey[J]. Journal of Machine Learning Research, 2019, 20(55):1-21.
[7]	ZOPH B, LE Q V. Neural Architecture Search with Reinforcement Learning(2017)[C/OL]. [2017-02-15]. https://openreview.net/pdf?id=r1Ue8Hcxg.
[8]	REAL E, AGGARWAL A, HUANG Y, et al. Regularized Evolution for Image Classifier Architecture Search[C]// Proceedings of the AAAI Conference on Artifical Intellgence.Hawaii:AAAI, 2019:4780-4789.
[9]	LIU H, SIMONYAN K, YANG Y. DARTS:Differentiable Architecture Search (2019)[C/OL]. [2019-04-23]. https://arxiv.org/abs/1806.09055
[10]	LIANG H, ZHANG S, SUN J, et al. DARTS+:Improved Differentiable Architecture Search with Early Stopping (2020)[J/OL]. [2020-10-20]. https://arxiv.org/abs/1909.06035.
[11]	CHU X X, WANG X X, ZHANG B et al. DARTS-:Robustly Stepping Out of Performance Collapse without Indicators (2021)[C/OL]. [2021-01-15]. https://arxiv.org/abs/2009.01027.
[12]	CHEN X, XIE L, WU J, et al. Progressive Differentiable Architecture Search:Bridging the Depth Gap Between Search and Evaluation[C]// Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision.Piscataway:IEEE, 2019: 1294-1303.
[13]	XU Y, XIE L, ZHANG X, et al. PC-DARTS:Partial Channel Connections for Memory-Efficient Differentiable Architecture Search (2019)[C/OL]. [2019-07-12]. https://arxiv.org/abs/1907.05737v1.
[14]	LI G, QIAN G, DELGADILLO I C, et al. SGAS:Sequential Greedy Architecture Search[C]// Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2020:1617-1627.
[15]	RONNEBERGER O, FISCHER P, BROX T. U-Net:Convolutional networks for biomedical image segmentation[C]// Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention.Heidelberg:Springer, 2015:234-241.
[16]	CHEN L C, ZHU Y, PAPANDREOU G, et al. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation[C]// Proceedings of the European conference on Computer Vision.Heidelberg:Springer, 2018:833-851.
[17]	BADRINARAYANAN V, KENDALL A, CIPOLLA R. SegNet:A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(12):2481-2495.
[18]	LIU C X, CHEN L C, SCHROFF F, et al. Auto-DeepLab:Hierarchical Neural Architecture Search for Semantic Image Segmentation[C]// Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2019:82-92.
[19]	ZHANG M W, JING W P, LIN J B, et al. NAS-HRIS:Automatic Design and Architecture Search of Neural Network for Semantic Segmentation in Remote Sensing Images[J]. Sensors, 2020, 20(18):5292-5307. doi: 10.3390/s20185292
[20]	HERRMANN C, BOWEN R S, ZABIH R. Channel Selection Using Gumbel Softmax[C]// Proceedings of the European Conference on Computer Vision.Heidelberg:Springer, 2020:241-257.
[21]	JI S P, WEI S Q, LU M. Fully Convolutional Networks for Multisource Building Extraction from an Open Aerial and Satellite Imagery Data Set[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(1):574-586. doi: 10.1109/TGRS.2018.2858817
[22]	TONG X Y, XIA G S, LU Q, et al. Learning Transferable Deep Models for Land-Use Classification with High-Resolution Remote Sensing Images (2019)[J/OL]. [2019-11-20]. https://arxiv.org/abs/1807.05713.

结构	参数/M	PA/%	F1/%	MIoU/%	搜索时间/h	训练时间/h	Epochs
SegNet	29.44	97.77	88.96	84.51		7.4
U-Net	23.36	98.30	93.56	88.41		6.2
DeepLab v3+	13.40	98.09	94.47	90.20		4.0
NAS-HRIS	0.19	98.52	93.77	90.44	12.1	16.4	30
RBPC-NAS	0.21	98.88	93.82	90.93	13.5	11.4	60

结构	参数/M	PA/%	F1/%	MIoU /%	搜索时间/h	训练时间/h	Epochs
SegNet	29.44	79.96	71.50	63.19		18.3
U-Net	23.36	80.37	73.71	64.66		13.2
DeepLab v3+	13.40	82.42	71.83	63.82		14.9
NAS-HRIS	0.19	88.48	78.35	67.03	10.6	19.5	30
RBPC-NAS	0.21	90.39	80.96	69.53	11.4	13.6	60

结构	参数/M	PA/%	F1/%	MIoU/%	搜索时间/h	训练时间/h	Epochs
NAS-HRIS	0.19	98.52	93.77	90.44	12.1	16.4	30
RBPC-NAS_A	0.26	98.76	93.80	90.79	14.2	18.6	60
RBPC-NAS_B	0.14	98.60	93.78	90.68	10.1	13.5	60
RBPC-NAS	0.21	98.88	93.82	90.93	13.5	11.4	60

结构	参数/M	PA/%	F1/%	MIoU /%	搜索时间/h	训练时间/h	Epochs
NAS-HRIS	0.19	88.48	78.35	67.03	10.6	19.5	30
RBPC-NAS_A	0.26	89.87	80.47	68.81	12.3	21.4	60
RBPC-NAS_B	0.14	89.45	80.17	68.24	8.7	15.5	60
RBPC-NAS	0.21	90.39	80.96	69.53	11.4	13.6	60