地表异常遥感探测轻量化大模型方法

doi:10.11947/j.AGCS.2025.20250092

测绘学报 ›› 2025, Vol. 54 ›› Issue (9): 1664-1676.doi: 10.11947/j.AGCS.2025.20250092

地表异常遥感探测轻量化大模型方法

闫凯¹^,²^,³(), 徐健明¹^,², 王桥¹^,²()

^1.北京师范大学卫星应用前沿交叉研究院，北京　100875
^2.北京师范大学地理科学学部，北京　100875
^3.遥感与数字地球全国重点实验室，北京　100875

收稿日期:2025-03-02 修回日期:2025-09-15 出版日期:2025-10-10 发布日期:2025-10-10
通讯作者: 王桥 E-mail:kaiyan@bnu.edu.cn;wangqiao@bnu.edu.cn
作者简介:闫凯（1988—），男，博士，副教授，研究方向为即时遥感、定量遥感。E-mail：kaiyan@bnu.edu.cn
基金资助:
国家自然科学基金重大项目(42192580)

Earth surface anomaly detection based on lightweight large vision model features in remotely sensed imagery

Kai YAN¹^,²^,³(), Jianming XU¹^,², Qiao WANG¹^,²()

^1.Advanced Interdisciplinary Institute of Satellite Applications, Beijing Normal University, Beijing 100875, China
^2.Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China
^3.State Key Laboratory of Remote Sensing and Digital Earth, Beijing 100875, China

Received:2025-03-02 Revised:2025-09-15 Online:2025-10-10 Published:2025-10-10
Contact: Qiao WANG E-mail:kaiyan@bnu.edu.cn;wangqiao@bnu.edu.cn
About author:YAN Kai (1988—), male, PhD, associate professor, majors in real-time remote sensing and quantitative remote sensing. E-mail: kaiyan@bnu.edu.cn
Supported by:
The National Natural Science Foundation of China Major Program(42192580)

摘要/Abstract

摘要：

在全球变化和城市化进程中，自然灾害、环境污染及非法开发等地表异常事件的影响范围与程度不断扩大，使得及时准确的地表异常探测成为国家需求和研究热点。遥感技术具有观测范围大且能周期性重访的优势，成为地表异常探测的关键手段。然而，现有地表异常探测方法多依赖地面大算力运算，涉及复杂的任务规划、卫星运控、星地传输和异常提取等环节，难以实现地表异常的快速识别。此外，已有方法通常面向特定异常类型，依赖专家知识和人工解译，难以用于通用化和智能化业务部署。为此，本文提出一种基于轻量化大模型特征的地表异常遥感自动探测方法，以提升地表异常探测的通用性、即时性和智能化部署能力。该方法包含3个环节：①采用视觉大模型作为特征提取器，以增强特征泛化性和算法通用性，通过高斯混合模型和贝叶斯信息准则实现自动化特征压缩，生成便于星地传输和在轨储存的轻量级先验知识；②利用高效字典查找法实现地表异常分值的快速推理，以适用于计算资源有限的星上环境；③在异常分值基础上，通过提示词和深度分割模型进行地表异常边界提取，提高提取的精细化和自动化程度。试验表明，与传统方法相比，本文方法表现更优且稳定，具有更好的通用性。试验案例中，先验知识库的平均压缩率约为100倍，大幅提升了在轨存储和星地传输更新的能力。同时，本文方法利用提示词和深度分割模型解决了固定阈值用于异常边界提取时自动化程度低、抗噪声能力差的问题，实现了精细化地表异常对象级提取。总体而言，本文方法具有数据存储量小、算力要求低、探测精度高等优势，具备成为通用化地表异常在轨即时探测业务化算法的潜力。

关键词: 地表异常探测, 即时遥感, 视觉大模型, 轻量化特征, 在轨运算

Abstract:

Earth surface anomaly detection (ESAD) has become increasingly vital due to intensifying global change and urbanization, leading to more frequent and severe disasters, pollution, and illegal development. Although remote sensing enables wide-range and periodic ESAD, efficiency remains a concern due to lengthy data transmission, dissemination, and processing procedures. Existing methods are primarily task-specific, relying on expert knowledge and human involvement, hindering generalized and automated deployment. In this context, we introduce a novel method aimed at inherently timely on-orbit detection. This approach is not only characterized by its generalizability and automation capabilities, but lightweight parameters that facilitate ground-satellite data transmission for updates. Our approach comprises three main processes: ①employing large vision models as feature extractors to enhance algorithm universality, with automatic feature compression achieved through Gaussian mixture models and Bayesian information criteria, generating lightweight prior knowledge suitable for ground-satellite transmission and on-orbit storage; ②utilizing an efficient dictionary lookup method for rapid inference of surface anomaly scores, making it applicable to satellite-based environments with limited computational resources; ③extracting surface anomaly boundaries based on anomaly scores using prompt words and deep segmentation models. This generates accurate anomaly boundaries with a threshold-insensitive method, reducing human involvement and promoting automation. Experiments demonstrate that the proposed method outperforms traditional approaches, offering better stability and generalization. In experimental cases, the average compression rate of the prior knowledge base was approximately 100 times, significantly improving the ability for on-orbit storage and ground-satellite data transmission updates. Furthermore, the method largely mitigates the issues of low automation and poor noise resistance associated with fixed thresholds for anomaly boundary extraction, achieving automated object-level surface anomaly extraction. Overall, the proposed ESAD method offers advantages such as small data storage, low computational requirements, and high detection accuracy. These features highlight its potential to become a generalized, on-orbit, real-time ESAD for operational deployment.

Key words: surface anomaly detection, real-time remote sensing, large vision model, lightweight features, on-orbit computation

中图分类号:

P237

闫凯, 徐健明, 王桥. 地表异常遥感探测轻量化大模型方法[J]. 测绘学报, 2025, 54(9): 1664-1676.

Kai YAN, Jianming XU, Qiao WANG. Earth surface anomaly detection based on lightweight large vision model features in remotely sensed imagery[J]. Acta Geodaetica et Cartographica Sinica, 2025, 54(9): 1664-1676.

图/表 9

图1

表1

图2

图3

表2

图4

表3

图5

图6

参考文献 41

[1]	WEI Haishuo, JIA Kun, WANG Qiao, et al. Real-time remote sensing detection framework of the Earth's surface anomalies based on a priori knowledge base[J]. International Journal of Applied Earth Observation and Geoinformation, 2023, 122: 103429.
[2]	Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovern mental Panel on Climate Change[R]. Zurich: IPCC, 2023: 35-115. DOI: . doi: 10.59327/IPCC/AR6-9789291691647
[3]	黄顺祥. 大气污染与防治的过去、现在及未来[J]. 科学通报, 2018, 63(10): 895-919.
	HUANG Shunxiang. The past, present, and future of air pollution prevention and control[J]. Scientific Bulletin, 2018, 63(10): 895-919.
[4]	施斌, 张泰银. 岩土体灾变感知与应用[J]. 中国科学：技术科学, 2023, 53(10): 1639-1651.
	SHI Bin, ZHANG Taiyin. Perception and application of rock and soil mass disasters[J]. Chinese Science: Technical Science, 2023, 53(10): 1639-1651.
[5]	BEHMEL S, DAMOUR M, LUDWIG R, et al. Water quality monitoring strategies: a review and future perspectives[J]. Science of The Total Environment, 2016, 571: 1312-1329.
[6]	李德仁. 从珞珈系列卫星到东方慧眼星座[J]. 武汉大学学报(信息科学版), 2023, 48(10): 1557-1565.
	LI Deren. From the Luojia series satellites to the oriental smart eye constellation[J]. Geomatics and Information Science of Wuhan University, 2023, 48(10): 1557-1565.
[7]	SHAH-HOSSEINI R, SAFARI A, HOMAYOUNI S. Natural hazard damage detection based on object-level support vector data description of optical and SAR Earth observations[J]. International Journal of Remote Sensing, 2017, 38(11): 3356-3374.
[8]	LI Chengfan, ZHANG Zixuan, LIU Lan, et al. A novel deep multi-instance convolutional neural network for disaster classification from high-resolution remote sensing images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 2098-2114.
[9]	SALESKA S R, DIDAN K, HUETE A R, et al. Amazon forests green-up during 2005 drought[J]. Science, 2007, 318(5850): 612.
[10]	ANNIBALLE R, NOTO F, SCALIA T, et al. Earthquake damage mapping: an overall assessment of ground surveys and VHR image change detection after L'Aquila 2009 earthquake[J]. Remote Sensing of Environment, 2018, 210: 166-178.
[11]	SENF C, SEIDL R. Mapping the forest disturbance regimes of Europe[J]. Nature Sustainability, 2021, 4(1): 63-70.
[12]	CASAGLI N, INTRIERI E, TOFANI V, et al. Landslide detection, monitoring and prediction with remote-sensing techniques[J]. Nature Reviews Earth and Environment, 2023, 4(1): 51-64.
[13]	李德仁, 沈欣, 龚健雅, 等. 论我国空间信息网络的构建[J]. 武汉大学学报(信息科学版), 2015, 40(6): 711-715, 766.
	LI Deren, SHEN Xin, GONG Jianya, et al. On construction of China's space information network[J]. Geomatics and Information Science of Wuhan University, 2015, 40(6): 711-715, 766.
[14]	王桥. 地表异常遥感探测与即时诊断方法研究框架[J]. 测绘学报, 2022, 51(7): 1141-1152. DOI: . doi: 10.11947/j.AGCS.2022.20220124
	WANG Qiao. Research framework of remote sensing monitoring and real-time diagnosis of earth surface anomalies[J]. Acta Geodaetica et Cartographica Sinica, 2022, 51(7): 1141-1152. DOI: . doi: 10.11947/j.AGCS.2022.20220124
[15]	李德仁, 王密, 杨芳. 新一代智能测绘遥感科学试验卫星珞珈三号01星[J]. 测绘学报, 2022, 51(6): 789-796. DOI: . doi: 10.11947/j.AGCS.2022.20220184
	LI Deren, WANG Mi, YANG Fang. A new generation of intelligent mapping and remote sensing scientific test satellite Luojia-3 01[J]. Acta Geodaetica et Cartographica Sinica, 2022, 51(6): 789-796. DOI: . doi: 10.11947/j.AGCS.2022.20220184
[16]	王密, 杨芳. 智能遥感卫星与遥感影像实时服务[J]. 测绘学报, 2019, 48(12): 1586-1594. DOI: . doi: 10.11947/j.AGCS.2019.20190454
	WANG Mi, YANG Fang. Intelligent remote sensing satellite and remote sensing image real-time service[J]. Acta Geodaetica et Cartographica Sinica, 2019, 48(12): 1586-1594. DOI: . doi: 10.11947/j.AGCS.2019.20190454
[17]	侯霞, 刘哲绮, 常亦迪, 等. 卫星激光通信技术发展现状与趋势分析[J]. 中国激光, 2024, 51(11): 231-244.
	HOU Xia, LIU Zheqi, CHANG Yidi, et al. Analysis on development status and trend of space laser communication technology[J]. Chinese Journal of Lasers, 2024, 51(11): 231-244.
[18]	王行行, 霍占伟, 牟洪元, 等. 吉林一号卫星激光通信数据传输试验及应用[J]. 国际太空, 2024(2): 42-49.
	WANG Xingxing, HUO Zhanwei, MOU Hongyuan, et al. Laser communication data transmission experiment and application of Jilin-1 satellite[J]. Space International, 2024(2): 42-49.
[19]	YANG Yuanxi, LIU Li, LI Jinlong. Featured services and performance of BDS-3[J]. Science Bulletin, 2021, 66(20): 2135-2143.
[20]	RŮŽIČKA V, VAUGHAN A, DE MARTINI D, et al. RaVÆn: unsupervised change detection of extreme events using ML on-board satellites[J]. Scientific Reports, 2022, 12: 16939.
[21]	李德仁, 沈欣. 我国天基信息实时智能服务系统发展战略研究[J]. 中国工程科学, 2020, 22(2): 138-143.
	LI Deren, SHEN Xin. Research on the development strategy of real time and intelligent space-based information service system in China[J]. Strategic Study of CAE, 2020, 22(2): 138-143.
[22]	GANDHI G M, PARTHIBAN S, THUMMALU N, et al. Ndvi: vegetation change detection using remote sensing and GIS-a case study of Vellore district[J]. Procedia Computer Science, 2015, 57: 1199-1210.
[23]	SU Jinya, COOMBES M, LIU Cunjia, et al. Machine learning-based crop drought mapping system by UAV remote sensing RGB imagery[J]. Unmanned Systems, 2020, 8(1): 71-83.
[24]	XU Yichu, ZHANG Lefei, DU Bo, et al. Hyperspectral anomaly detection based on machine learning: an overview[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022, 15: 3351-3364.
[25]	WANG Wei, ZHANG Baoju, WANG Dan, et al. Anomaly detection based on probability density function with Kullback-Leibler divergence[J]. Signal Processing, 2016, 126: 12-17.
[26]	RŽIČKA V, VAUGHAN A, DE MARTINI D, et al. RaVÆn: unsupervised change detection of extreme events using ML on-board satellites[J]. Scientific Reports, 2022, 12: 16939.
[27]	XU Jianming, YAN Kai, FAN Zaiwang, et al. Toward a novel method for general on-orbit earth surface anomaly detection leveraging large vision models and lightweight priors[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 1-21.
[28]	GAO Bocai. NDWI: a normalized difference water index for remote sensing of vegetation liquid water from space[J]. Remote Sensing of Environment, 1996, 58(3): 257-266.
[29]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
[30]	HUANG Xin, ZHU Tingting, ZHANG Liangpei, et al. A novel building change index for automatic building change detection from high-resolution remote sensing imagery[J]. Remote Sensing Letters, 2014, 5(8): 713-722.
[31]	MORAVEC D, KOMÁREK J, LÓPEZ-CUERVO MEDINA S, et al. Effect of atmospheric corrections on NDVI: intercomparability of Landsat 8, Sentinel-2, and UAV sensors[J]. Remote Sensing, 2021, 13(18): 3550.
[32]	NAGOL J R, VERMOTE E F, PRINCE S D. Effects of atmospheric variation on AVHRR NDVI data[J]. Remote Sensing of Environment, 2009, 113(2): 392-397.
[33]	KIRILLOV A, MINTUN E, RAVI N, et al. Segment anything[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision. Paris: IEEE, 2023: 3992-4003.
[34]	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16x16 words: transformers for image recognition at scale[EB/OL]. [2024-10-23]. https://api.semanticscholar.org/CorpusID:225039882.
[35]	BAHRI Y, DYER E, KAPLAN J, et al. Explaining neural scaling laws[J]. Proceedings of the National Academy of Science, 2024, 121(27): e2311878121.
[36]	HE Kaiming, FAN Haoqi, WU Yuxin, et al. Momentum contrast for unsupervised visual representation learning[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 9726-9735.
[37]	SOM R K. Practical sampling techniques[M]. Boca Raton: CRC press, 1995.
[38]	CARVALHO JÚNIOR O A, GUIMARÃES R F, GILLESPIE A R, et al. A new approach to change vector analysis using distance and similarity measures[J]. Remote Sensing, 2011, 3(11): 2473-2493.
[39]	WARNER T. Hyperspherical direction cosine change vector analysis[J]. International Journal of Remote Sensing, 2005, 26(6): 1201-1215.
[40]	GOLDSTEIN M, DENGEL A. Histogram-based outlier score (HBOS): a fast unsupervised anomaly detection algorithm[J]. KI-2012: Poster and Demo Track, 2012, 1: 59-63.
[41]	ZHAO Y, NASRULLAH Z, LI Z. PyOD: a Python toolbox for scalable outlier detection[J]. Journal of Machine Learning Research, 2019, 20(96): 1-7.

方法类型	变量	实例化方法
低阶变化检测	P	像元值/低阶特征^[22]
	d	向量空间度量
	o（x），τ	像素尺度分析
深度变化检测	P	深度神经网络特征
	d	向量空间度量
	o（x），τ	影像块尺度分析
异常检测	P	像元值/低阶特征的统计模型^[23-24]
	d	概率空间度量^[25]
	o（x），τ	像素尺度分析

数据集类型	经纬度	先验图	异常图	特点
滑坡	43.57°S，170.17°E	2012-03-13 EO-1 TOA（真彩色）	2013-02-05 EO-1 TOA（真彩色）	2013年奥拉基山滑坡，滑坡区域比较有代表性且明显
野火	39.81°N，121.45°E	2017-11 Landsat-8中值合成（真彩色）	2018-11-08 Landsat-8 TOA（真彩色）	2018年加州天堂镇附近火灾事件，同时捕捉到燃烧区、过火区和浓烟，比较能代表野火事件
水体异常	34.64°N，77.07°W	2017-11 Landsat-8中值合成（真彩色）	2018-09-19 Landsat-8 TOA（真彩色）	2018年佛罗伦萨洪水引发的水体异常事件，相较其他事件异常程度较弱
火山1	19.02°N，98.62°W	2023-02 Landsat-8中值合成（标准假彩色）	2024-02-28 Landsat-8 TOA（标准假彩色）	（1）波波卡特佩特火山喷发，具有明显的火山羽流，能作为火山活动的代表样本
火山1	19.02°N，98.62°W	2023-02 Landsat-8中值合成（标准假彩色）	2024-02-28 Landsat-8 TOA（标准假彩色）	（2）采用标准假彩色合成，检验不同波段组合下的工作效果
火山2	56.05°N，160.63°E	2022-11-01 Landsat-8TOA（真彩色）	2023-11-01 Landsat-8 TOA（真彩色）	克柳切夫火山喷发，先验图和异常图在同一天获取，但空间上不匹配，可以测试算法在应用上的灵活性

方法	参数	滑坡	野火	水体异常	火山1	火山2
深度变化检测（余弦距离）	F₁值	0.653	0.925	0.695	0.657	—
深度变化检测（余弦距离）	数据量/KB	4096
深度变化检测（欧氏距离）	F₁值	0.653	0.925	0.695	0.657	—
深度变化检测（欧氏距离）	数据量/KB	4096
孤立森林	F₁值	0.353	0.709	0.572	0.434	0.651
	数据量/KB	208.54	213.20	212.83	209.67	220.14
	压缩率/（%）	19.64	19.21	19.25	19.54	18.61
凝聚法聚类	F₁值	0.672	0.949	0.797	0.779	0.809
	数据量/KB	107	85	42	68	49
	压缩率/（%）	38.28	48.19	97.52	60.24	83.59
提出方法	F₁值	0.970	0.971	0.978	0.814	0.710
	数据量/KB	47	38	32	47	36
	压缩率/（%）	87.15	107.79	128.00	87.15	113.78

地表异常遥感探测轻量化大模型方法

Earth surface anomaly detection based on lightweight large vision model features in remotely sensed imagery

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 9

参考文献 41

相关文章 1

编辑推荐

Metrics

本文评价