基于深度学习的生产建设项目扰动图斑自动提取

doi:10.16843/j.sswc.2023.01.015

Abstract
Figure/Table
References
Related Citation (15)

Download: PDF (10320 KB) HTML (1 KB)
Export: BibTeX | EndNote (RIS)

Abstract [Background] Rapid acquiring the disturbance range of production and construction projects is critical for improving the efficiency of soil and water conservation supervision and inspection, and for controlling man-made soil erosion. The plots for production and construction projects are numerous and scattered, but remote sensing technology can achieve timely and accurate acquisition of their spatial distribution. At present, the remote sensing interpretation of disturbed patches in production and construction projects is mainly based on human-computer interaction with less automatic extraction. The existing algorithms are highly specialized and low in regional applicability, which makes it difficult applying in engineering.[Methods] The central and eastern parts of Gansu province were used as the research area, and deep learning algorithms were applied to carry out research on automatic extraction of disturbance areas. Based on GF-1 and GF-6 satellite images, the disturbance characteristics of production and construction projects were summarized, and bare land and newly added construction land were determined as the characteristic features detecting in production and construction projects. Selected 1 156 disturbances totally for the patch samples, then the enhanced DeepLabV3+ model and the U-Net model were utilized to construct an automatic extraction model of disturbance patches. Based on automatic extraction, the minimum recognition area was set to reduce the interference of mini patches. The extraction effects of different models and different minimum recognition areas were quantitatively evaluated by indicators such as quantity recall rate, quantity accuracy rate, area recall rate, and area accuracy rate.[Results] 1) The DeepLabV3+ model had fewer erroneous image spots. The quantity accuracy rate was more than 15% than that by the U-Net model, and the area accuracy rate was slightly higher than the U-Net model. 2) In Chengguan district and Zhengning county, the DeepLabV3+ model had fewer missing image spots, and the quantity recall rate was more than 5% higher than that of the U-Net model. The quantity recall rate of the DeepLabV3+ model in Baiyin district was slightly lower than that of the U-Net model. 3) Large patches extraction of these two models was better. The area accuracy rate was higher than the quantity accuracy rate, and the area recall rate was higher than the quantity recall rate. 4) As the minimum recognition area increased, the area accuracy of the model increased and the area recall rate decreased. 5) When the minimum recognition area was 1 hm², utilizing the DeepLabV3+ model generated the best extraction effect. The average quantity accuracy rate was 78.31%, the quantity recall rate was 61.69%, the area accuracy rate was 88.82%, and the area recall rate was 80.60%.[Conclusions] In summary, DeepLabV3+ model has better overall classification effect. After setting an appropriate minimum recognition area, it can balance the accuracy and recall of the automatic extraction results. At the same time, the model has good extraction results in different districts and counties, which indicates that the deep learning algorithm has strong regional applicability and can be applied to a wide range of remote sensing supervision work.

Key words： production and construction projects deep learning DeepLabV3+ U-Net

Received: 21 July 2021

PACS:	S157
	TP79

	Service

	E-mail this article
	Add to my bookshelf
	Add to citation manager
	E-mail Alert
	RSS
	Articles by authors
	KANG Rui
	QIU Xinling
	MA Hongcai
	SONG Pengyu

Cite this article:

KANG Rui,QIU Xinling,MA Hongcai, et al. Automatic extraction of disturbed patches in production and construction projects based on deep learning[J]. SSWC, 2023, 21(1): 128-138.

URL:

http://journal12.magtechjournal.com/Jweb_stbc/EN/10.16843/j.sswc.2023.01.015 OR http://journal12.magtechjournal.com/Jweb_stbc/EN/Y2023/V21/I1/128

[1]	李智广,王敬贵.生产建设项目"天地一体化"监管示范总体实施方案[J].中国水土保持,2016(2):14. LI Zhiguang,WANG Jinggui.Overall implementation plan for supervision and demonstration of "integration of heaven and earth" in production and construction projects[J].Soil and Water Conservation in China,2016(2):14.
[2]	徐晓伟,李晓松,李周.基于高分影像的生产建设项目水土保持监管应用研究[J].甘肃科学学报,2019,31(1):92. XU Xiaowei, LI Xiaosong, LI Zhou. Application of production construction project water and soil conservation monitoring based on images with high resolution[J].Journal of Gansu Sciences, 2019,31(1):92.
[3]	毕永清. 高寒地区生产建设项目扰动区的遥感监测方法研究[D].成都:西南交通大学,2017:31. BI Yongqing. Study on the remote sensing monitoring method for disturbance area of production and construction project in arctic-alpine region[D].Chengdu:Southwest Jiaotong University, 2017:31.
[4]	亢庆,姜德文,扶卿华,等.基于最优尺度的生产建设扰动图斑识别[J].中国水土保持科学,2017,15(6):126. KANG Qing, JIANG Dewen, FU Qinghua, et al. On the identification of construction disturbance patches based on optimal segmentation scale[J].Science of Soil and Water Conservation, 2017, 15(6):126.
[5]	高志强,史明昌,杨文涛,等.面向水土保持监管的黄土高原区生产建设项目地块快速提取[J].中国水土保持科学,2017,15(3):81. GAO Zhiqiang, SHI Mingchang, YANG Wentao, et al. Rapid extraction of land parcels from soil and water conservation supervision-oriented productive and constructive projects in the Loess Plateau[J]. Science of Soil and Water Conservation,2017,15(3):81.
[6]	郑远攀,李广阳,李晔.深度学习在图像识别中的应用研究综述[J].计算机工程与应用,2019,55(12):20. ZHENG Yuanpan, LI Guangyang, LI Ye. Survey of application of deep learning in image recognition[J]. Computer Engineering and Applications, 2019, 55(12):20.
[7]	LIU Man, FU Bolin, XIE Shuyu, et al. Comparison of multi-source satellite images for classifying marsh vegetation using DeepLabV3 Plus deep learning algorithm[J]. Ecological Indicators, 2021, 125(11):107562.
[8]	CHEN Liangchieh, ZHU Yukun, PAPANDREOU G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]. Munich, Germany:ECCV,2018:833.
[9]	WANG Junqiang, LI Jiansheng, ZHOU Huachun, et al. Typical element extraction method of remote sensing image based on DeeplabV3+ and CRF[J]. Computer Engineering, 2019,45(10):260.
[10]	张鑫禄,张崇涛,戴晨光,等.基于DeepLabv3架构的高分辨率遥感图像分类[J].海洋测绘,2019,39(2):40. ZHANG Xinlu, ZHANG Chongtao, DAI Chenguang, et al. High resolution remote sensing image classification based on DeepLabV3 architecture[J]. Hydrographic Surveying and Charting, 2019,39(2):40.
[11]	傅晓梦. 基于深度学习的高分辨率遥感影像语义分割方法研究[D].南京:南京理工大学,2019:35. FU Xiaomeng. Semantic segmentation of high resolution remote sensing image based on deep learning[D]. Nanjing:Nanjing University of Science and Technology,2019:35.
[12]	王蓝玉. 基于Deeplab V3+网络的遥感地物图像语义分割研究[D].哈尔滨:哈尔滨工业大学,2020:31. WANG Lanyu. Research on semantic segmentation of remote sensing images of the ground objects based on DeeplabV3+ network[D]. Harbin:Harbin Institute of Technology,2020:31.
[13]	薛磊. 基于改进U-Net的多光谱遥感影像分类方法研究[D].吉林:吉林大学,2018:30. XUE Lei. Research on multi-spectral remote sensing image classification based on improved U-Net[D]. Jilin:Jilin University,2018:30.
[14]	许慧敏. 基于深度学习U-Net模型的高分辨率遥感影像分类方法研究[D].成都:西南交通大学,2018:32. XU Huimin. Method research of high resolution remote sensing imagery classification based on U-Net model of deep learning[D].Chengdu:Southwest Jiaotong University,2018:32.
[15]	杨瑞,祁元,苏阳.深度学习U-Net方法及其在高分辨卫星影像分类中的应用[J].遥感技术与应用,2020,35(4):767. YANG Rui,QI Yuan,SU Yang. U-Net neural networks and its application in high resolution satellite image classification[J].Remote Sensing Technology and Application,2020,35(4):767.
[16]	WAGNER F H, SANCHEZ A, TARABALKA Y, et al. Using the U-Net convolutional network to map forest types and disturbance in the Atlantic rainforest with very high resolution images[J]. Remote Sensing in Ecology & Conservation, 2019.
[17]	WEI Yang, LIU Xiaofang. The application of deep convolution neural network to building extraction in remote sensing images[J]. World Scientific Research Journal, 2020, 6(3):136.
[18]	TAN Mingxing, LE Q V. EfficientNet:Rethinking model scaling for convolutional neural networks[C]. Los Angeles,USA:ICML 2019.