多源数据融合的泛时空特征水稻深度学习提取

doi:10.19788/j.issn.2096-6369.000010

摘要/Abstract

摘要：

传统基于时序遥感影像的水稻物候期特征提取方法要求有较高的时间分辨率，受成像条件制约而较难满足；由于不同水稻种植区域环境条件不同，基于单一影像的深度学习水稻种植区域提取方法泛化能力较差。本文选取时相相近的光学和合成孔径雷达（Synthetic Aperture Radar，SAR）数据，削弱水稻种植区影像时空信息差异。通过泛时空特征融合有效地利用光学数据空间特征信息和SAR数据后向散射信息，采用双结构网络模型提取水稻特征。实验表明，基于多源数据融合的泛时空特征水稻深度学习提取方法在三江平原和肥西县水稻数据集上训练模型验证集总体测试精度为95.66%，Kappa系数为0.8805。该模型在南昌市区域水稻提取结果与实际地块边界符合较好，总体提取精度为86.78%，证明了泛时空特征模型的泛化能力和实用性。

关键词: 泛时空特征, SAR数据, 光学数据, 特征融合, 深度学习, 水稻提取

Abstract:

Traditional methods of rice phenological phase feature extraction based on time-series remote sensing images require high temporal resolution, which is difficult to meet due to imaging conditions. Due to the different environmental conditions in different rice growing regions, the rice planting area extraction method based on single image has poor generalization ability. In this paper, similar optical and Synthetic Aperture Radar (SAR) data were selected to reduce the spatiotemporal information differences in rice planting area images. The spatial feature information of optical data and backscatter information of SAR data were effectively used to extract rice features by using a two-structure network model through pan-spatio-temporal feature fusion. Experiments show that the overall test accuracy of the training model validation set on the rice datasets of Sanjiang Plain and Feixi County is 95.66%, and the Kappa coefficient is 0.8805. The results of rice extraction in Nanchang City were in good agreement with the actual field boundaries, and the overall extraction accuracy was 86.78%, which proved the generalization ability and practicability of the pan-temporal feature model.

Key words: pan-temporal characteristics, SAR data, optical data, feature fusion, deep learning, rice extraction

杜家宽, 李雁飞, 孙嗣文, 刘继东, 江腾达. 多源数据融合的泛时空特征水稻深度学习提取[J]. 农业大数据学报, 2024, 6(1): 56-67.

DU JiaKuan, LI YanFei, SUN SiWen, LIU JiDong, JIANG TengDa. Pan-spatiotemporal Feature Rice Deep Learning Extraction Based on Multi-source Data Fusion[J]. Journal of Agricultural Big Data, 2024, 6(1): 56-67.

图/表 17

图3

图1

图2

表1

表2

表3

图4

图5

图6

图7

图8

图9

图10

图11

表4

表5

图12

参考文献 23

[1]	谭深, 吴炳方, 张鑫. 基于Google Earth Engine与多源遥感数据的海南水稻分类研究[J]. 地球信息科学学报, 2019, 21(6):937-947. doi: 10.12082/dqxxkx.2019.180423.
[2]	陈雨思, 李丹, 黎臻, 等. 多时相MODIS影像的黑龙江省水稻种植面积提取[J]. 农业工程学报, 2020, 36(23):201-208.
[3]	李永帅, 齐修东. 基于Google Earth Engine三江平原水田提取研究[J]. 地理空间信息, 2021, 19(12):77-80+6.
[4]	Dong J, Xiao X, Menarguez M A, et al. Mapping paddy rice planting area in northeastern Asia with Landsat 8 images, phenology-based algorithm and Google Earth Engine[J]. Remote Sensing of Environment, 2016, 185: 142-154. https://doi.org/10.1016/j.rse.2016.02.016. doi: 10.1016/j.rse.2016.02.016 pmid: 28025586
[5]	黄侠, 姜雪芹, 张灿, 等. 红边波段在水稻生育期识别中的应用研究[J]. 测绘地理信息, 2023, 48(3):87-90.
[6]	Li P, Feng Z, Jiang L, et al. Changes in rice cropping systems in the Poyang Lake Region, China during 2004-2010[J]. Journal of Geographical Sciences, 2012, 22: 653-668. https://doi.org/10.1007/s11442-012-0954-x. doi: 10.1007/s11442-012-0954-x
[7]	黄青, 吴文斌, 邓辉. 2009年江苏省冬小麦和水稻种植面积信息遥感提取及长势监测[J]. 江苏农业科学, 2010(6):508-511.
[8]	Yang H J, Pan B, Li N, et al. A systematic method for spatio-temporal phenology estimation of paddy rice using time series Sentinel-1 images[J]. Remote Sensing of Environment, 2021, 259: 112394. https://doi.org/10.1016/j.rse.2021.112394. doi: 10.1016/j.rse.2021.112394
[9]	Son N T, Chen C F, Chen C R, et al. A phenological object-based approach for rice crop classification using time-series Sentinel-1 Synthetic Aperture Radar (SAR) data in Taiwan[J]. International Journal of Remote Sensing, 2021, 42(7): 2722-2739. https://doi.org/10.1080/01431161.2020.1862440. doi: 10.1080/01431161.2020.1862440
[10]	Pan B, Zheng Y, Shen R, et al. High resolution distribution dataset of double-season paddy rice in China[J]. Remote Sensing, 2021, 13(22): 4609. https://doi.org/10.3390/rs13224609. doi: 10.3390/rs13224609
[11]	于飞, 吕争, 隋正伟, 等. 基于特征优选的多时相SAR数据水稻信息提取方法[J]. 农业机械学报, 2023, 54(3):259-265+327.
[12]	Zhang L, Zhang L, Du B. Deep learning for remote sensing data: A technical tutorial on the state of the art[J]. IEEE Geoscience and Remote Sensing Magazine, 2016, 4(2): 22-40. https://doi.org/10.1109/MGRS.2016.2540798. doi: 10.1109/MGRS.2016.2540798
[13]	闫利, 徐青, 刘异. 基于注意力网络的遥感影像植被提取方法[J]. 测绘地理信息, 2021, 46(S1):44-48.
[14]	Alami M M, Mansouri L E, Imani Y, et al. Crop mapping using supervised machine learning and deep learning: a systematic literature review[J]. International Journal of Remote Sensing, 2023, 44(8): 2717-2753. https://doi.org/10.1080/01431161.2023.2205984. doi: 10.1080/01431161.2023.2205984
[15]	黄晓涵, 黄文龙. 基于深度学习的多时序遥感影像水稻提取研究[J]. 地理空间信息, 2023, 21(8):61-64+113.
[16]	Fu T, Tian S, Ge J. R-Unet: A Deep Learning Model for Rice Extraction in Rio Grande do Sul, Brazil[J]. Remote Sensing, 2023, 15(16): 4021. https://doi.org/10.3390/rs15164021. doi: 10.3390/rs15164021
[17]	蔡耀通, 刘书彤, 林辉, 等. 基于多源遥感数据的CNN水稻提取研究[J]. 国土资源遥感, 2020, 32(4):97-104.
[18]	Li X, Zhang G, Cui H, et al. MCANet: A joint semantic segmentation framework of optical and SAR images for land use classification[J]. International Journal of Applied Earth Observation and Geoinformation, 2022, 106: 102638. https://doi.org/10.1016/j.jag.2021.102638. doi: 10.1016/j.jag.2021.102638
[19]	Wu D, Zhao J, Wang Z. AM-PSPNet: Pyramid Scene Parsing Network Based on Attentional Mechanism for Image Semantic Segmentation[C]// ICPCSEE Steering Committee.Abstracts of the 8th International Conference of Pioneering Computer Scientists,Engineers and Educators(ICPCSEE 2022) Part I. Department of Electronic Engineering,Heilongjiang University; 2022:1. https://doi.org/10.26914/c.cnkihy.2022.077315.
[20]	Sun K, Xiao B, Liu D, et al. Deep high-resolution representation learning for human pose estimation.[J]. Corr, 2019, abs/1902.09212. https://doi.org/10.48550/arXiv.1902.09212.
[21]	Dominic Masters, Carlo Luschi. Revisiting small batch training for deep neural networks[Z]. ARXIV, 2018, 1804.07612.
[22]	伍光和, 王乃昂, 胡双熙, 等. 自然地理学(第四版)[M]. 北京: 高等教育出版社, 2008.
[23]	Olofsson P, Foody G M, Herold M, et al. Good practices for estimating area and assessing accuracy of land change[J]. Remote Sensing of ENVIRONMENT, 2014, 148: 42-57. https://doi.org/10.1016/j.rse.2014.02.015. doi: 10.1016/j.rse.2014.02.015

精度指标	OA	IoU	Kappa
HRnet	0.9326	0.7981	0.8621
Ours	0.9566	0.8086	0.8805
提升	0.0240	0.0105	0.0184