1 School of Earth Science and Resources, Chang’an University, Xi’an 710054, China
2 State Key Laboratory of Tibetan Plateau Earth System, Environment and Resources, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing 100101, China
3 Shaanxi Key Laboratory of Land Consolidation, School of Land Engineering, Chang’an University, Xi’an 710054, China
Show less
文章历史+
Received
Published
2024-09-15
2026-09-25
Issue Date
2026-07-08
PDF (13282K)
摘要
石冰川空间分布制图对于研究寒旱区水文地质和气候变化具有重要意义。除野外实地调查和目视解译等传统手段外,深度学习与高分辨率自然彩色影像(RGB)相结合被应用于青藏高原的石冰川编目(TPRoGI)。然而,由于经典深度学习网络只接受三波段影像作为输入,拥有锐利的边界信息和丰富的光谱信息的近红外波段(NIR)尚未被应用于石冰川识别。因此,本次研究通过改进DeeplabV3+(IDNet),使其具备同时从RGB和NIR中提取和融合特征的能力。基于祁连山地区的Sentinel-2影像和TPRoGI训练IDNet,模型的准确率、精准度、召回率、特异度和mIoU(mean Intersection over Union)分别为0.783 0、0.783 0、0.784 0、0.783 5和0.691 6,各项指标均衡且稳定性高。IDNet模型在祁连山地区识别出459个被TPRoGI遗漏的石冰川,证明了基于IDNet和Sentinel-2 RGB-NIR影像识别石冰川可行,并能有效提升石冰川编目工作的效率和精度。
Abstract
Understanding the distribution of rock glaciers is crucial for studying the hydrogeology and climate change in cold and arid regions. Apart from traditional methods such as field surveys and visual interpretation, deep learning (DL) applied to high-resolution natural color (RGB) remote sensing imagery has been used to compile the Tibetan Plateau Rock Glacier Inventory (TPRoGI). However, although the near-infrared (NIR) band provides sharp edge features and rich spectral information that are useful for identifying rock glaciers, it has rarely been utilized in DL-based recognition models due to the three-band input restriction of typical DL networks. In this study, an improved DeepLabV3+ network (IDNet) was designed to simultaneously extract and fuse features from both RGB and NIR bands. The IDNet was trained using Sentinel-2 imagery and the rock glacier labels from the TPRoGI in the Qilian Mountains (QLMs), yielding a well-trained model that achieved an accuracy of 0.7830, a precision of 0.7830, a recall of 0.7840, a specificity of 0.7835, and a mean Intersection over Union (mIoU) of 0.6916. The IDNet model further identified 459 rock glaciers in the QLMs that were missed by the TPRoGI, demonstrating the feasibility and effectiveness of combining the IDNet with RGB and NIR bands for rock glacier recognition. This approach effectively enhances the efficiency and accuracy of rock glacier inventory compilation.
基于此,本次研究改进了CDNet,以HRNetV2-W48(High-Resolution Network Version 2-Wide 48,以下称为“HRNetV2”)为主干网络,继承了经典的编码-解码结构的同时添加了双编码器,可以同时从RGB和NIR波段中提取特征,并在解码阶段进行特征融合。使用Sentinel-2和TPRoGI训练改进的DeeplabV3+(IDNet),并对祁连山地区未包含在TPRoGI中的石冰川进行了补充识别,进一步基于石冰川的标准指南进行人工确认,评价NIR识别石冰川的可行性。
利用相同的数据集和超参数,训练了包括IDNet,CDNet,改进的CDnet,成熟的MSNet[54]。通过五个评价指标对比不同模型之间的性能。如前文所述,改进CDNet使其具备处理六波段影像的方法有两种:①将CDNet输入层的通道参数从3改为6,称为MICDNet(Modified the Input Layer of Classical DeeplabV3+ Network)。需要说明的是,因为MICDNet的结构与预训练模型HRNetV2的结构不匹配,不能进行迁移学习;②在CDNet之前添加一层卷积核尺寸为1×1的卷积操作,保持样本高宽不变的前提下,将6波段的影像降维成3通道的数据,称为ADCDNet(Added Convolution before Classical DeeplabV3+ Network)。MSNet设计的初衷就是处理多光谱影像,并且已经证实了比RTFNet(处理RGB和热红外的数据)[55]、MUFNet(处理多光谱遥感数据)[56]、MFNet(处理多光谱数据)[57]的性能更加优越,所以和MSNet对比的结果基本能代表了IDNet的性能。所有网络的训练结果见表1。
AZÓCARG F, BRENNINGA. Hydrological and geomorphological significance of rock glaciers in the dry andes, chile (27°-33°S)[J]. Permafrost and Periglacial Processes, 2010, 21(1): 42-53.
[2]
BAIB, YUEP, ZHANGQ, et al. Changing characteristics of ecosystem and water storage under the background of warming and humidification in the qilian mountains, china[J]. Science of The Total Environment, 2023, 893: 164959.
[3]
JONESD B, HARRISONS, ANDERSONK, et al. Mountain rock glaciers contain globally significant water stores[J]. Scientific Reports, 2018, 8(1): 2834.
[4]
JONESD, HARRISONS, ANDERSONK, et al. The distribution and hydrological significance of rock glaciers in the nepalese himalaya[J]. Global and Planetary Change, 2018, 160: 123-142.
[5]
ARENSONL, HOELZLEM, SPRINGMANS. Borehole deformation measurements and internal structure of some rock glaciers in switzerland[J]. Permafrost and Periglacial Processes, 2002, 13(2): 117-135.
HARRINGTONJ S, MOZILA, HAYASHIM, et al. Groundwater flow and storage processes in an inactive rock glacier[J]. Hydrological Processes, 2018, 32(20): 3070-3088.
[8]
WINKLERG, WAGNERT, PAURITSCHM, et al. Identification and assessment of groundwater flow and storage components of the relict schöneben rock glacier, niedere tauern range, eastern alps (austria)[J]. Hydrogeology Journal, 2016, 24(4): 937-953.
[9]
HUMLUMO. The climatic significance of rock glaciers[J]. Permafrost and Periglacial Processes, 1998, 9(4): 375-395.
[10]
HUMLUMO. Rock glacier appearance level and rock glacier initiation line altitude: a methodological approach to the study of rock glaciers[J]. Arctic and Alpine Research, 1988, 20(2): 160.
[11]
SORGA. Impacts of climate change on glaciers, rock glaciers and water availability in the tien shan, central asia[D/OL]. Genève: Université de Genève, 2013[2023-10-10].https://archive-ouverte.unige.ch/unige:39554.
[12]
JONESD B, HARRISONS, ANDERSONK. Mountain glacier-to-rock glacier transition[J]. Global and Planetary Change, 2019, 181: 102999.
[13]
HARRISC, ARENSONL U, CHRISTIANSENH H, et al. Permafrost and climate in europe: monitoring and modelling thermal, geomorphological and geotechnical responses[J]. Earth-Science Reviews, 2009, 92(3/4): 117-171.
[14]
POTTERN. Ice-cored rock glacier, galena creek, northern absaroka mountains, wyoming[J]. Geological Society of America Bulletin, 1972, 83(10): 3025.
[15]
BARSCHD, FIERZH, HAEBERLIW. Shallow core drilling and bore-hole measurements in the permafrost of an active rock glacier near the grubengletscher, wallis, swiss alps[J]. Arctic and Alpine Research, 1979, 11(2): 215-228.
[16]
HAEBERLIW, HOELZLEM, KÄÄBA, et al. Ten years after drilling through the permafrost of the active rock glacier murtèl, eastern swiss alps: answered questions and new perspectives[C]// PERMAFROST-Seventh International Conference (Proceedings), Yellowknife (Canada), Collection Nordicana No 55, 1998: 403-410.
[17]
BOLCHT, GORBUNOVA P. Characteristics and origin of rock glaciers in northern tien shan (kazakhstan/kyrgyzstan)[J]. Permafrost and Periglacial Processes, 2014, 25(4): 320-332.
[18]
SCHMIDM-O, BARALP, GRUBERS, et al. Assessment of permafrost distribution maps in the hindu kush himalayan region using rock glaciers mapped in google earth[J]. The Cryosphere, 2015, 9(6): 2089-2099.
[19]
JONESD B, HARRISONS, ANDERSONK, et al. Rock glaciers and mountain hydrology: a review[J]. Earth-Science Reviews, 2019, 193: 63-90. http://hdl.handle.net/10871/36795.
[20]
ERHARTERG H, WAGNERT, WINKLERG, et al. Machine learning - an approach for consistent rock glacier mapping and inventorying: example of austria[J]. Applied Computing and Geosciences, 2022, 16: 100093.
[21]
FENGM, XUJ, WANGJ, et al. Identifying rock glacier in western china using deep learning and satellite data[C]// AGU Fall Meeting Abstracts, 2019.
[22]
AGUERAM. Automatic detection of rock glaciers from very-high resolution satellite imagery using convolutional neural networks and object-based image analysis[D]. Post box 7800, N-5020 Bergen, Norway: University of Bergen, 2021.
[23]
BERTONEA, BARBOUXC, BODINX, et al. Incorporating insar kinematics into rock glacier inventories: insights from 11 regions worldwide[J]. The Cryosphere, 2022, 16(7): 2769-2792.
MARCERM. Rock glaciers automatic mapping using optical imagery and convolutional neural networks[J]. Permafrost and Periglac Process, 2020, 31(4): 561-566.
[26]
ROBSONB A, BOLCHT, MACDONELLS, et al. Automated detection of rock glaciers using deep learning and object-based image analysis[J]. Remote Sensing of Environment, 2020, 250: 112033.
[27]
HUY, LIUL, HUANGL, et al. Mapping and characterizing rock glaciers in the arid west kunlun of china[J]. ESS Open Archive, 2022: 1-37.
[28]
SUNZ, HUY, LIUL, et al. Mapping and inventorying rock glaciers on the tibetan plateau from planet basemaps using deep learning[C/OL]// EGU General Assembly Conference Abstracts. [2023-12-10]. https://meetingorganizer.copernicus.org/EGU23/EGU23-6816.html.
[29]
SUNZ, HUY, LIUL, et al. Mapping rock glaciers on the tibetan plateau from planet basemaps using deep learning[C]// AGU Fall Meeting Abstracts, 2022.
[30]
SUNZ, HUY, RACOVITEANUA, et al. TPRoGI: a comprehensive rock glacier inventory for the tibetan plateau using deep learning[R/OL]. ESSD - Ice/Permafrost, 2024[2024-03-18].https://essd.copernicus.org/preprints/essd-2024-28/.
[31]
JIANGJ, FENGX, LIUF, et al. Multi-spectral rgb-nir image classification using double-channel cnn[J]. IEEE Access, 2019, 7: 20607-20613.
[32]
CARVALHOO L F D, DE CARVALHO JÚNIORO A, ALBUQUERQUEA O D, et al. Instance segmentation for large, multi-channel remote sensing imagery using mask-rcnn and a mosaicking approach[J]. Remote Sensing, 2020, 13(1): 39.
[33]
HANW, LIJ, WANGS, et al. Geological remote sensing interpretation using deep learning feature and an adaptive multisource data fusion network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-14.
[34]
LINL, LIUL, LIUM, et al. DEDNet: dual-encoder DeeplabV3+ network for rock glacier recognition based on multispectral remote sensing image[J]. Remote Sens. 2024, 16: 2603.
[35]
LOUP, WUT, CHENJ, et al. Recognition of thaw slumps based on machine learning and uavs: a case study in the qilian mountains, northeastern qinghai-tibet plateau[J]. International Journal of Applied Earth Observation and Geoinformation, 2023, 116: 103163.
[36]
HUZ, YAND, FENGM, et al. Enhancing mountainous permafrost mapping by leveraging a rock glacier inventory in northeastern tibetan plateau[J]. International Journal of Digital Earth, 2024, 17(1): 2304077.
[37]
JIANGZ, HUETEA, DIDANK, et al. Development of a two-band enhanced vegetation index without a blue band[J]. Remote Sensing of Environment, 2008, 112(10): 3833-3845.
[38]
GILABERTM A, GONZÁLEZ-PIQUERASJ, GARCI’A-HAROF J, et al. A generalized soil-adjusted vegetation index[J]. Remote Sensing of Environment, 2002, 82(2/3): 303-310.
WUH, LIANGC, LIUM, et al. Optimized hrnet for image semantic segmentation[J]. Expert Systems with Applications, 2021, 174: 114532.
[41]
WANGJ, SUNK, CHENGT, et al. Deep high-resolution representation learning for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(10): 3349-3364.
[42]
LIL, TIANT, LIH, et al. SE-hrnet: a deep high-resolution network with attention for remote sensing scene classification[C/OL]// IGARSS 2020-2020 IEEE International Geoscience and Remote Sensing Symposium. Waikoloa, HI, USA: IEEE, 2020: 533-536[2024-05-23]. https://ieeexplore.ieee.org/document/9324633/.
[43]
SEONGS, CHOIJ. Semantic segmentation of urban buildings using a high-resolution network (hrnet) with channel and spatial attention gates[J]. Remote Sensing, 2021, 13(16): 3087.
[44]
ZHANGX, LIUQ, CHANGH, et al. High-resolution network with transformer embedding parallel detection for small object detection in optical remote sensing images[J]. Remote Sensing, 2023, 15(18): 4497.
[45]
YANGX, FANX, PENGM, et al. Semantic segmentation for remote sensing images based on an ad-hrnet model[J]. International Journal of Digital Earth, 2022, 15(1): 2376-2399.
[46]
HEK, ZHANGX, RENS, et al. Deep residual learning for image recognition[C/OL]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE, 2016: 770-778[2024-02-27]. http://ieeexplore.ieee.org/document/7780459/.
[47]
SUR, XUD, SHENGL, et al. PCG-tal: progressive cross-granularity cooperation for temporal action localization[J]. IEEE Transactions on Image Processing, 2021, 30: 2103-2113.
[48]
CHENL C, PAPANDREOUG, KOKKINOSI, et al. DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(4): 834-848.
[49]
HUANGL, LUOJ, LINZ, et al. Using deep learning to map retrogressive thaw slumps in the beiluhe region (tibetan plateau) from cubesat images[J]. Remote Sensing of Environment, 2020, 237: 111534.
[50]
RGIK.Towards standard guidelines for inventorying rock glaciers:baseline concepts (version 4.2.2)[J/OL]. IPA Action Group Rock glacier inventories and Kinematics, 2022, 13. https://bigweb.unifr.ch/Science/Geosciences/Geomorphology/Pub/Website/IPA/Guidelines/V4/210801_Baseline_Concepts_Inventorying_Rock_Glaciers_V4.2.1.pdf.
[51]
BRARDINONIF, SCOTTIR, SAILERR, et al. Evaluating sources of uncertainty and variability in rock glacier inventories[J]. Earth Surface Processes and Landforms, 2019, 44(12): 2450-2466.
[52]
BARONIC, CARTONA, SEPPIR. Distribution and behaviour of rock glaciers in the adamello-presanella massif (italian alps)[J]. Permafrost and Periglacial Processes, 2004, 15(3): 243-259.
[53]
JOHNSONG, CHANGH, FOUNTAINA. Active rock glaciers of the contiguous united states: geographic information system inventory and spatial distribution patterns[J]. Earth System Science Data, 2021, 13(8): 3979-3994.
SUNY, ZUOW, LIUM. RTFNet: RGB-thermal fusion network for semantic segmentation of urban scenes[J]. IEEE Robotics and Automation Letters, 2019, 4(3): 2576-2583.
[56]
XUF, SHANGZ, WUQ, et al. MUFNet: toward semantic segmentation of multi-spectral remote sensing images[C/OL]// 2021 4th Artificial Intelligence and Cloud Computing Conference. Kyoto Japan: ACM, 2021: 39-46[2024-02-18]. https://dl.acm.org/doi/10.1145/3508259.3508265.
[57]
HAQ, WATANABEK, KARASAWAT, et al. MFNet: towards real-time semantic segmentation for autonomous vehicles with multi-spectral scenes[C/OL]// 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Vancouver, BC: IEEE, 2017: 5108-5115[2024-02-27]. http://ieeexplore.ieee.org/document/8206396/.
[58]
JONESD B, HARRISONS, ANDERSONK, et al. Rock glaciers represent hidden water stores in the himalaya[J]. Science of the Total Environment, 2021, 793: 145368.
[59]
WANGQ, ATKINSONP M. Spatio-temporal fusion for daily sentinel-2 images[J]. Remote Sensing of Environment, 2018, 204: 31-42.