基于Deeplab V3+模型的高分辨率遥感影像道路损毁信息提取

doi:10.12196/j.issn.1000-3274.2022.02.014

摘要/Abstract

摘要： 破坏性地震发生后, 道路是抗震救灾的“生命线”, 能够第一时间获取震后的道路损毁信息, 就能在很大程度上减少生命和财产的损失。针对目前震害提取方法精度低、人工参与多及用时较长的问题, 提出了基于Deeplab V3+深度学习网络模型的震后单时相高分辨率遥感影像道路震害信息提取方法。该方法充分挖掘了多尺度的上下文信息, 逐步重构空间信息以便更好地捕捉道路边界, 从而提高道路震害信息提取精度。以2013年四川省雅安市芦山7.0级地震为例, 通过选取山区、平原以及城市内部不同等级的道路高分影像样本, 创建模型训练的样本集。为了进一步验证Deeplab V3+模型的效果, 选择目前较为主流的FCN模型与Deeplab V3+模型进行对比。经多次训练和精度分析, 综合考虑模型精度和训练量、训练次数的关系, 最终选择7528对训练样本集、 100000训练次数和总精度达到95%的Deeplab V3+模型作为最佳模型。然后, 将训练后的最佳Deeplab V3+模型应用到提取震后无人机道路损毁信息。将人工目视解译的结果作为模型提取精度评价的标准, 对比分析后Deeplab V3+模型的震后高分遥感影像道路损毁信息提取精度可达88%。结果表明, 基于Deeplab V3+模型的道路震害提取方法具有较高的准确率和普适性, 可在震后道路损毁信息的快速提取中推广应用。

关键词: Deeplab V3+, 高分遥感影像, 道路, 2013年芦山7.0级地震

Abstract: After the earthquake, the road is the “lifeline” of earthquake relief. The first time to obtain the road damage information after the earthquake is to minimize the loss of life and property. Aiming at the problems of current seismic damage extraction methods with low accuracy, multiple manual participation and long time consumption, a method for extracting road seismic damage information from single-phase high-resolution remote sensing images based on Deeplab V3+ deep learning network model is proposed. This method fully excavates multi-scale contextual content information, and gradually reconstructs spatial information to better capture road boundaries, thereby improving the accuracy of road earthquake damage information extraction. In this paper, taking the 2013 Lushan M7.0 earthquake in Ya' an of Sichuan Province as an example, by selecting different grades of road high-scoring image samples in mountains, plains, and cities, creating a sample set of model training. In order to further verify the effect of the Deeplab V3+ model, the current more mainstream FCN model is selected for comparison with the Deeplab V3+ model. After multiple training and accuracy analysis, considering the relationship between model accuracy, training quantity and training times. Deeplab V3+ model with 7528 training samples, 100000 training times and 95% total accuracy was selected as the best model. Then, the best Deeplab V3+ model after training is applied to extract post-earthquake road damage information. The results of manual visual interpretation are used as the standard for evaluating the accuracy of model extraction. After comparative analysis, the accuracy of the road information extraction of post-earthquake high-resolution remote sensing images of Deeplab V3+ model can reach 88%. The results show that the method based on Deeplab V3+ model has the highest accuracy and universality, and can be widely used in the rapid extraction of post-earthquake damaged road information.

Key words: Deeplab V3+, High-resolution remote sensing image, Road, the 2013 Lushan M7.0 earthquake

中图分类号:

TP79

陈丹丹, 窦爱霞, 王鑫. 基于Deeplab V3+模型的高分辨率遥感影像道路损毁信息提取[J]. 地震, 2022, 42(2): 171-189.

CHEN Dan-dan, DOU Ai-xia, WANG Xin. Road Damage Information Extraction from High-Resolution Remote Sensing Images Based on Deeplab V3+ Network Model[J]. EARTHQUAKE, 2022, 42(2): 171-189.

参考文献

[1] 韩玲. 地震灾害遥感特征分析与信息提取技术研究[D]. 成都: 成都理工大学, 2011.
HAN Ling. Study on technologies of remote sensing feature analysis and information extraction of earthquake disaster[D]. Chengdu: Chengdu University of Technology, 2011 (in Chinese).
[2] 陆超. 基于WorldView-2影像的面向对象信息提取技术研究[D]. 杭州: 浙江大学, 2012.
LU Chao. Research on information extraction of WorldView-2 imagery with object-oriented technology[D]. Hangzhou: Zhejiang University, 2012 (in Chinese).
[3] 田卫方. 基于高分辨率遥感影像的震后道路损毁识别研究以鲁甸地震为例[D]. 北京: 北京交通大学, 2015.
TIAN Wei-fang. Research on recognition of earthquake damaged road based on high resolution remote sensing imagery: Taking Ludian earthquake for an example[D]. Beijing: Beijing Jiaotong University, 2015 (in Chinese).
[4] 李强, 张景发. 道路震害遥感提取方法现状与应急应用展望[J]. 地震, 2017, 37(4): 80-92.
LI Qiang, ZHANG Jing-fa. Extraction method of road seismic damage from multi-source remote sensing images and application prospect for earthquake emergency[J]. Earthquake, 2017, 37(4): 80-92 (in Chinese).
[5] 谢翠容, 代侦勇, 杨洋. 改进型算法的高分辨率遥感影响道路灾害信息提取[J]. 测绘与空间地理信息, 2016, 39(9): 105-108.
XIE Cui-rong, DAI Zhen-yong, YANG Yang. The extraction algorithm of disaster information based on high resolution remote sensing image[J]. Geomatics & Spatial Information Technology, 2016, 39(9): 105-108 (in Chinese).
[6] 王岩, 王晓青, 窦爱霞. 面向对象遥感分类方法在震害快速提取中的应用[J]. 国际地震动态, 2012(6): 288.
WANG Yan, WANG Xiao-qing, DOU Ai-xia. Object-oriented remote sensing data classification method application in earthquake damage rapid extraction[J]. Recent Developments in World Seismology, 2012(6): 288 (in Chinese).
[7] 陈世荣, 马海建, 范一大, 等. 基于高分辨率遥感影像的汶川地震道路损毁评估[J]. 遥感学报, 2008, 12(6): 949-955.
CHEN Shi-rong, MA Hai-jian, FAN Yi-da, et al. Road damage assessment from high resolution satellite remote sensing imagery in Wenchuan earthquake[J]. Journal of Remote Sensing, 2008, 12(6): 949-955 (in Chinese).
[8] Sato H P, Harp E L. Interpretation of earthquake-induced landslides triggered by the 12 May 2008, M7.9 Wenchuan earthquake in the Beichuan area, Sichuan Province, China using satellite imagery and Google Earth[J]. Landslides, 2009, 6: 153-159.
[9] Lacroix P, Zavala B, Berthier E, et al. Supervised method of landslide inventory using panchromatic SPORT5 images and application to the earthquake-triggered landslides of Pisco[J]. Remote Sensing, 2013(5): 2590-2616.
[10] 王艳萍, 姜纪沂, 林玲玲. 高分辨率遥感影像中道路震害信息的识别方法[J]. 计算机工程与应用, 2012, 48(3): 173-175+227.
WANG Yan-ping, JIANG Ji-yi, LIN Ling-ling. Road damage detection from high resolution remote sensing[J]. Computer Engineering and Applications, 2012, 48(3): 173-175+227 (in Chinese).
[11] 马海建, 陆楠, 李晓璇. 基于边线的遥感影像道路震害快速提取方法研究[J]. 地震, 2013, 33(2): 71-78.
MA Hai-jian, LU Nan, LI Xiao-xuan. Rapid extraction of earthquake road damages based on sidelines from remote sensing images[J]. Earthquake, 2013, 33(2): 71-78 (in Chinese).
[12] 胡岩峰, 刘波, 李峰, 等. 一种基于统计分析的图像变化检测方法[J]. 光子学报, 2005, 34(1): 146-149.
HU Yan-feng, LIU Bo, LI Feng, et al. An image change detection method using statistical analysis[J]. Acta Photonica Sinica, 2005, 34(1): 146-149 (in Chinese).
[13] 钟家强, 王润生. 一种基于线特征的道路网变化检测算法[J]. 遥感学报, 2007, 11(1): 27-32.
ZHONG Jia-qiang, WANG Run-sheng. A road network change detection algorithm based on linear feature[J]. Journal of Remote Sensing, 2007, 11(1): 27-32 (in Chinese).
[14] 刘明众, 张景发, 李成龙, 等. 高分辨率遥感影像道路震害的快速提取[J]. 地震, 2013, 33(2): 79-86.
LIU Ming-zhong, ZHANG Jing-fa, LI Cheng-long, et al. Rapid extraction of road damages from high resolution remote sensing images[J]. Earthquake, 2013, 33(2): 79-86 (in Chinese).
[15] Mohan R. Deep deconvolutional networks for scene parsing[J]. Computer Science, 2014(1): 1-8.
[16] 韩洁, 李盛阳, 张涛. 基于深度学习的遥感影像城市扩展方法研究[J]. 载人航天, 2017, 23(3): 414-418+426.
HAN Jie, LI Sheng-yang, ZHANG Tao. Research on urban expansion method based on deep learning of remote sensing Image[J]. Manned Space Flight, 2017, 23(3): 414-418+426 (in Chinese).
[17] Cheng G, Yang C Y, Yao X W, et al. When deep learning meets metric learning: Remote sensing Image scene classification via learning discriminative CNNs[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(5): 2811-2821.
[18] Zou Q, Ni L H, Zhang T, et al. Deep learning based feature selection for remote sensing scene classification[J]. IEEE Geoscience and Remote Sensing Letters, 2015, 12(11): 2321-2325.
[19] 雒培磊, 李国庆, 曾怡. 一种改进的基于深度学习的遥感影像拼接方法[J]. 计算机工程与应用, 2017, 53(20): 180-186.
LUO Pei-lei, LI Guo-qing, ZENG Yi. Modified approach to remote sensing image mosaic based on deep learning[J]. Computer Engineering and Applications, 2017, 53(20): 180-186 (in Chinese).
[20] Wang J, Song J W, Chen M Q, et al. Road network extraction: a neural-dynamic framework based on deep learning and a finite state machine[J]. International Journal of Remote Sensing, 2015, 36(12): 3144-3169.
[21] 王港, 陈金勇, 高峰, 等. 基于深度学习的遥感影像基础设施目标检测研究[J]. 无线电工程, 2018(3): 219-224.
WANG Gang, CHEN Jin-yong, GAO Feng, et al. Research on the infrastructure target detection of remote sensing Image based on deep learning[J]. Radio Engineering, 2018(3): 219-224 (in Chinese).
[22] 方茂青. 基于Deeplabv3+的遥感图像城市道路分割算法研究[D]. 武汉: 华中科技大学, 2019.
FANG Mao-qing. Research on urban road segmentation algorithm based on Deeplabv3+ remote sensing image[D]. Wuhan: Huazhong University of Science and Technology, 2019 (in Chinese).
[23] Chen L C, Zhu Y K, George P, et al. Encoder-Decoder with atrous separable convolution for semantic image segmentation[Z]. 2018-8-22. http://link.springer.com/conference/eccv.
[24] Achanta R, Shaji A, Smith K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Trans on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282.
[25] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]. IEEE Conference on Computer Vision and Pattern Recognition, IEEE Computer Society, 2015: 3431-3440.