最新刊期
卷 26 ， 期 2 ， 2022
-
摘要：Climate warming has become a great challenge for global sustainable development. Under the Paris Agreement, every country must present a climate action plan in five-yearly cycles, a National Determined Contributions (NDC) report will be presented using a standard inventory approach for each country since 2020, and all countries will engage in the global stocktake every five years to assess countries’ NDC progress since 2023. The 49th session of the Intergovernmental Panel on Climate Change (IPCC 49) recommend a ‘top-down’ inversion approach to account greenhouse gas (GHG) emission based on space-borne atmospheric measurements. The European Union, the United States, Japan, and Canada are vigorously developing MVS (Monitoring & Verification Support) capabilities for accounting GHG emissions using satellite remote sensing. Here, we aimed to give a detailed review on the methods and progresses of satellite-based inversion for global stocktaking, and highlight the challenges and perspectives for satellite remote sensing for global stocktaking in China.Firstly, Earth observation for atmospheric GHG, including ground-based observation networks and GHG satellites, were summarized. Compared to ground-based observations, satellite remote sensing has been providing more and more accurate and higher resolution global GHG detection. In the next five years, 13 GHG satellites will be launched, with resolutions ranging from 25 m to 100 km. Secondly, the progresses of satellite remote sensing of ecosystem carbon fluxes were reviewed. There are three kinds of methods to estimate global carbon fluxes, including: the assimilation inversion method (also named as the “top-down” method), that uses atmospheric chemical transmission model and ground-based or satellite observations of atmospheric GHG to invert carbon flux； the modelling simulation method (also named as the “bottom-up” method) that uses the process model to estimate terrestrial and marine ecosystems carbon fluxes; the data-driven machine learning method that uses remote sensing datasets and metrological datasets to model the carbon uptakes of terrestrial and marine ecosystems. However, the uncertainty in the estimation results of all these top-down or bottom-up methods is still huge at regional or global scale. Thirdly, the researches on satellite monitoring of anthropogenic GHG emissions were summarized. Satellite remote sensing has been an important platform for realizing large-scale, long-term observations of anthropogenic GHG emissions. Although the current accuracy of the satellite-based observations does not fully meet the requirements of the global stocktake, satellite remote sensing has become a promising tool for verifying hot-spot, city, national and global anthropogenic emissions. Finally, the current capability of satellite remote sensing to support global carbon monitoring was assessed, and the Chinese carbon satellite future program was proposed. According the preliminary simulations based on Observation System Simulation Experiments (OSSE), the China’s next generation carbon satellite (TanSat-2) are presented. Similar to CO2M project supported by European Union, TanSat-2 will give global accurate retrieval of GHGs (1 ppm for CO2 and 10 ppb for CH4), pollution gases (1.0×1015 molecules/cm2 for NO2, 10% for CO) and solar-induced chlorophyll fluorescence (SIF) (0.25 mw m-2·nm-1·sr-1) with a swath of 1000 km and a resolution 500 m resolution, which will provide unprecedented imaging capabilities for estimating GHG emissions.Satellite remote sensing plays extremely role in build the MVS capability for global stocktake, we provide a reference for the roadmap of the Chinese carbon monitoring program based on the preliminary OSSE simulations. It is absolutely necessary to integrate satellite remote sensing, in-situ observations, big data, carbon assimilation to achieve high precision, high-resolution scientific data on GHG fluxes at hot-spot, regional and global scales, and to effectively distinguish and quantify the flux contributions of anthropogenic GHG emissions and terrestrial carbon sinks /sources.关键词：global warming;carbon stocktaking;carbon emissions;carbon sources and sinks;satellite remote sensing;assimilation- 5396
- |
- 1708
- |
- 28
发布时间：2022-02-28 -
摘要：Quantitative remote sensing is a technique for quantitatively inferring or inverting earth environmental variable from the original remote sensing observations, it is an important step to turn electromagnetic wave signals into easy-to-understand information about surface environment. Traditional quantitative remote sensing methods are mainly model-driven, emphasizing inversion through mathematical or physical models. With the development and popularization of artificial intelligence technology, data-driven methods have gradually received widespread attention. It emphasizes the use of machine learning methods to mine the information contained in remote sensing observation data to achieve the quantitative inversion of geophysical parameters. With the support of powerful computing capacity, the data-driven method has achieved gratifying achievements in many fields of quantitative remote sensing. This article systematically summarizes the principles, characteristics, and applications in quantitative remote sensing field of different types of data-driven models, including regression algorithms, regularization methods, instance-based algorithms, decision tree, Bayesian methods, kernel based algorithms, genetic algorithms, ensemble learning, artificial neural network, and deep learning. Though data-driven models show satisfying retrieval performance in multiple fields, its drawbacks of ignoring the laws of physics and lack of causality have also brought resistance to its development. In this context, coupling the laws of physics and machine learning to develop an inversion framework driven by both models and data has become a new research hotspot. Some pioneering researches have already achieved delightful performance through using machine learning to assist physical models or restricting machine learning with physical laws. Using machine learning techniques to optimize the systematic basis, the sub-model, and the model parameters largely improve the performance of model-driven methods. Meanwhile, integrating physics knowledge into machine learning models through adjusting the training data, modifying the loss function, and constraining the solution space also benefit the improvement of data-driven models. However, there are still great challenges to be broken through. Physical models contain complex mechanisms and rich knowledge, current fusion of data-driven and model-driven methods are quite shallow with very limited amount of physical knowledge being used. A deeper coupling strategy is worth exploring in the future. Besides, the uncertainty, generalization, and transferability of the joint model have not been scientifically evaluated currently, to which attention should be paid. Finally, there are many cases when the training samples were very difficult to obtain, therefore, the applicability of the joint model in the case of small samples is also a problem that needs to be solved urgently. The deep, robust, and generalizable coupling of data-driven and model-driven models is expected in the future.关键词：remote sensing;quantitative remote sensing;model-driven;data-driven;deep learning;machine learning- 2841
- |
- 1137
- |
- 6
发布时间：2022-02-28 -
摘要：Parameter Sensitivity Analysis (SA) is an important research method for Uncertainty Analysis(UA), key parameters identification and parameters optimization in remote sensing, ecological and hydrological models. In this paper, the sensitivity analysis of ecological and hydrological research based on remote sensing is analyzed. The sensitivity analysis methods commonly used in remote sensing ecological hydrology are reviewed, and the advantages and applicable conditions of each SA method are summarized. Parameter sensitivity analysis methods are generally divided into Local Sensitivity Analysis (LSA) and Global Sensitivity Analysis (GSA), also can be divided into variance based, statistics based and graphic based methods from mathematical mechanism. Sobol 'and EFAST are the most reliable and stable global sensitivity methods among the current sensitivity algorithms, which are most suitable for most remote sensing inversion and model. There are many methods for parameter sensitivity analysis, so it is very important to select the appropriate method. The initial setting of sensitivity analysis will also affect the results of the analysis. The sensitivity of parameters varies at different scales, The parameter of remote sensing fluorescence model is also one of the key scientific issues. Parametric sensitivity analysis methods have also promoted the development and use of microwave scattering/radiation models. Parameter sensitivity In the process of remote sensing inversion, the order of importance of parameters can be judged according to the sensitivity order, thus providing prior knowledge for multi-stage inversion. In conclusion, sensitivity analysis can effectively improve the simulation accuracy of hydrological, ecological and growth models driven by remote sensing data, and effectively analyze the uncertainties caused by parameters at different scales. Parameter sensitivity analysis can be judged according to the order of sensitivity so as to provide a priori knowledge for multi-stage inversion in the process of remote sensing inversion. The difference of parameter sensitivity analysis in different scales, different bands and different observation angles, as well as the parameter uncertainty, must be paid attention to and analyzed. The four platforms for sensitivity analysis and uncertainty analysis also are introduced in order to make it more convenient for remote sensing scientists to use parameter sensitivity analysis method. Parameter sensitivity analysis as the prior knowledge of the model promotes the development of uncertainty analysis and parameter optimization. In future studies, Under the framework of Uncertainty and Sensitivity Matrix (USM), it is necessary to pay more attention to the research of multi-stage remote sensing inversion by combining global SA, scale effect of parameter sensitivity index and spatio-temporal heterogeneity of parameter Sensitivity. Meanwhile, the model construction and parameter setting are supported by prior knowledge of parameter sensitivity analysis. Parameter sensitivity analysis should be combined with parameter optimization, data assimilation, spatial analysis and multi-stage inversion to optimize remote sensing inversion and reduce uncertainty. The improvement of computational efficiency and stability of parameter sensitivity analysis is the trend of future research, which requires multi-threaded synchronization, grouping strategy and cloud computing platform.关键词：remote sensing;Parameter sensitivity analysis;parameter optimization;uncertainty analysis;eco-hydrological- 1956
- |
- 880
- |
- 4
发布时间：2022-02-28
Research Progress
-
摘要：The rapid development of remote sensing technology has enabled accumulation of massive earth observation data in recent years. Traditional desktop-based remote sensing platforms (e.g., ERDAS and ENVI) cannot satisfy the current application requirements of remote sensing big data. Google Earth Engine (GEE), as a leading cloud-based remote sensing platform, has not only changed the traditional processing and analysis means of data but also brought new opportunities for the rapid processing and information mining of massive data. To date, many researchers have successfully conducted a large number of works and published many academic papers and reviews based on GEE. However, no one has systematically analyzed how GEE promotes the development of remote sensing science. Therefore, this study aims to explore the innovative changes caused by GEE in the aspect of resources, methods, and applications, compared with the traditional desktop-based platform. (1) In terms of resources, GEE breaks the separation of traditional data, model algorithms, and computing power, realizes cloud deployment of the three, and shows great potential in the rapid processing and analysis of massive data through combining big data and cloud computing together. (2) With regard of methods, innovative remote sensing analysis methods provided by GEE break through the bottleneck of traditional remote sensing technology and promote the technological innovation of data processing and analysis. Thus, the efficiency of massive data processing and information mining greatly improves. (3) In the perspective of applications, GEE not only brings development opportunities for the rapid global-scale long-time analysis but also promotes the rapid sharing of data, algorithms, and products, which further ushers in the era of opening and sharing remote sensing. Systematically summarizing the advantages of GEE will not only help the potential new users understand GEE as well as deepen the existing users’ understanding but also encourage Google developers to perfect and improve GEE while catalyzing new discoveries in the scientific research of the earth system.关键词：Google Earth Engine;cloud-based remote sensing platform;desktop-based remote sensing platform;remote sensing big data;pixel-based analysis method- 2745
- |
- 1002
- |
- 8
发布时间：2022-02-28 -
摘要：In the context of big data Remote Sensing (RS), the development of RS cloud computing platforms has changed the mode of RS traditional data processing and analysis. It also has greatly improved the computing efficiency, which enables it to quickly analyze long-term time-series on the global scale. Although many scholars have conducted related works with RS cloud computing platforms, an objective review on the development and application of RS cloud computing platforms is still lacking. In this study, we retrieved the research literature related to RS cloud computing platforms between January 2011 and April 2021 based on the Web of Science and China National Knowledge Infrastructure. The retrieved data were analyzed in terms of publication volume, collaboration analysis, keyword co-occurrence analysis, and co-citation analysis using bibliometric methods. Results show that (1) the number of studies based on RS cloud computing platforms is increasing. China and the United States are the most active countries in this field, and the Chinese Academy of Sciences (CAS) is the most active institution. (2) The intersection of related disciplines is extensive, and it involves RS, environmental science and ecology, computer science, engineering, electrical and electronics, and other disciplines. Among them, RS is the most researched field using cloud computing platforms, and environmental science and ecology and computer science are more closely connected with other disciplinary fields. (3) At present, Google Earth Engine is a widely used RS cloud computing platform. In addition, Amazon Web Services Cloud, Earth Data Miner (a pioneering earth data mining and analysis system of CAS), PIE-Engine, and other platforms are also in a rapid development stage. (4) Large-scale land cover mapping, land use, vegetation dynamics, and climate change have been the main application areas. Environmental health assessment and research on the impact of human activities on the environment will also be important application areas of the platforms in the future. These results quantitatively demonstrated the development history, research hotspots, and applications of RS cloud computing platforms, which provide a reference for relevant researchers to grasp the development dynamics of the field and explore valuable new research directions.关键词：bibliometric;visualisation;remote sensing;big data;remote sensing cloud computing platform- 1450
- |
- 606
- |
- 0
发布时间：2022-02-28 -
摘要：Satellite remote sensing is the main way for humans to observe the Earth. The commercialization of remote sensing technology has facilitated the rapid growth of the number of global commercial remote sensing satellites. Real-time monitoring is possible. However, the current satellite remote sensing data acquisition has various problems, such as duplication, blindness, untimely, disconnection from user needs, and a large amount of data idle. An effective platform and application model is also lacking to build a bridge between satellite data providers and users. Real-time monitoring of the Earth’s surface can be realized by only directional monitoring and updating of change information. Thus, this study proposes a set of online service cloud platforms for real-time earth change monitoring by combining satellite remote sensing and the Internet. The automatic change detection is the core, which gathers user needs for surface change information to form more high-definition satellite shooting conditions. According to the orbit prediction model, it will quickly push satellite shooting instructions to the nearest satellite data service provider to achieve fixed-point directional data update and ensure that users can view the latest images of the surface of the area of interest at any time. Its core ideas are mainly based on Internet online capture technology, cloud platform automatic change detection technology, and satellite on-orbit real-time processing technology. It aims to analyze, mine, and extract geographic entity change information, realize the collection and update of directional and fixed-point geographic entity data based on change information by pushing satellite shooting instructions in real time, and maintain real-time monitoring services for areas of interest. This study describes the implementation methods in the process of cloud platform construction, builds and validates the basic prototype of the cloud platform, and discusses the new overall architecture design ideas and application areas in the cloud platform. Results show that the platform is useful for the current business The solution to the problems of passivity, singularity, delay, and repetition in remote sensing services provides a design idea from the current cloud platform based on traditional data query and ordering to the fixed-point update service cloud platform based on changing information.关键词：remote sensing;satellite-earth integration;change detection;on-board processing;cloud platform;fixed-point update- 1643
- |
- 422
- |
- 0
发布时间：2022-02-28 -
摘要：With the arrival of remote sensing big data era, numerous remote sensing cloud computing platforms have emerged inland and overseas to rapidly process and analyze massive remote sensing data. The emergence of remote sensing cloud computing platform makes it possible to quickly analyze and apply remote sensing data on a global scale or for longterm sequences. However, currently, there is lacking of remote sensing cloud computing platform with complete functions in domestic, while foreign remote sensing cloud computing platform has insufficient support for domestic satellite data. Based on this situation, we have independently developed a spatiotemporal remote sensing cloud computing platform, PIE (Pixel Information Expert) -Engine Studio. By adopting container cloud technology, this platform integrating data, computing power and technology, can implements on-demand acquisition of remote sensing data and rapid processing of massive data just driven by the script. (1) This study first introduced the system architecture of PIE-Engine Studio, and then described the data storage and access mode. (2) PIE-Engine Studio provides operations for multiple objects such as number, matrix, image, vector, list, dictionary, etc., also machine learning algorithms and some special satellite algorithms. (3) Furthermore, this study illustrated the calculation flow of the platform in detail. Firstly, the user writes a script in the front-end to describe the calculation process of remote sensing data. Click the “Run” button, these codes automatically build the preliminary chained structure call syntax tree. Then the syntax tree is optimized in the back-end through filter the invalid calculation content. The computing tasks are then distributed to the computing services on multiple nodes through the scheduling center. Finally, the resulting visual map layer or data file is returned to the front-end interface triggered by specific front-end requests or operators (print, addLayer, export).(4) At last, an application case is presented, we adopted Landsat 8 data and taking the calculation of Normalized Difference Vegetation Index (NDVI) in the growing season as an example, the calculation results and running time of this platform are compared with Google Earth Engine (GEE). The results show that, due to the limitation of computing resources, the running and export time of this platform are slightly longer than that of GEE, but the spatial distribution of calculation results is consistent, among which about 68% values are distributed between (0.48, 0.77), and 95.33% of the difference between the two results is concentrated between (-0.13, 0.13). It shows that the results are reliable. Therefore, the remote sensing cloud computing platform constructed by this paper, can provide data resources and computing power for research in the field of earth science, and will help promote the development of remote sensing cloud computing platform in China and the application of domestic satellite data in cloud computing platform.关键词：remote sensing;big data;remote sensing cloud computing platform;distributed storage;parallel computing- 3525
- |
- 769
- |
- 6
发布时间：2022-02-28 -
摘要：Intertidal wetlands are an important part of coastal wetlands and have crucial ecological functions, such as maintaining biodiversity and promoting carbon sink. However, intertidal wetlands are severely threatened by coastal erosion, sea level rise, and human activities. Timely and accurately monitoring the status of intertidal wetlands is the basis for achieving the goal of sustainable management of intertidal wetlands. Periodic tidal inundation is one of the largest challenges in mapping intertidal wetlands. The key is to obtain the remote sensing images at the time with the lowest and highest tides rapidly and accurately to accurately extract information of intertidal wetlands. At present, the dense temporal resolution of Sentinel-2 images with revisit interval of 3—5 days offers a great opportunity to capture the lowest and highest tides, which is vital to conduct accurate and robust delineation of intertidal wetlands. Previous efforts on intertidal wetland classification relied on either training samples, manual intervened thresholds or pre and postprocessing. This work aimed to set up an automatic, rapid, and high precision procedure that uses time series Sentinel-2 images to derive intertidal wetland information based on the Google Earth Engine platform.The methodology includes four steps: (1) building a high-quality dense time series image stack; (2) deeply analyzing the time series remote sensing characteristics of different wetland types and selecting appropriate spectral indexes; (3) creating the maximal water surface image, the minimal water surface image, and vegetation difference enhanced image on the basis of the maximum spectral index composite algorithm; and (4) establishing a multilayer automatic decision tree classification model to extract different intertidal wetlands from simple types to complex types by using the Otsu algorithm.The procedure was utilized to classify the intertidal wetlands in the Fujian Zhangjiangkou National Mangrove Nature Reserve in 2020 with an overall accuracy of 96.5% and a kappa coefficient of 0.95. The intertidal wetlands in the Zhangjiangkou Reserve consist of mangrove forest, Spartina alterniflora, and tidal flat, with an area of 82.46, 218.26, and 496.84 hm2, respectively. Abundant tidal flat resources were mainly located on the outer edge of mangrove forest and S. alterniflora. Mangroves were mainly concentrated on the southwest coast of the Zhangjiang River. S. alterniflora was mainly distributed on the south of Zhangjiang River with good integrity, whereas part of them grew on the north side of Zhangjiang River with a banding distribution.The high-quality Sentinel-2 dense time series image stack increases the opportunity to obtain the lowest and highest tide images and provides sufficient phenological information for the classification of mangrove and S. alterniflora. The maximal intertidal water surface can be easily obtained by combining with the modified normalized difference water index maximum value composite image and the method of extracting the largest patch area. The Normalized Difference Vegetation Index (NDVI) maximum value composite image well highlights the difference between tidal flats, water bodies, and vegetated areas. The negative NDVI maximum value composite image plays a positive role in enhancing the characteristic difference between mangrove forest and S. alterniflora. The proposed method can realize the automatic, rapid, and accurate classification of intertidal wetlands, which has important reference value for the accurate classification research of intertidal and other inland wetlands.关键词：tidal flat;wetland;Sentinel-2 imagery;Maximum Spectral Index Composite (MSIC);otsu algorithm (Otsu);Google Earth Engine (GEE)- 1917
- |
- 734
- |
- 6
发布时间：2022-02-28 -
摘要：Traditional water extraction algorithms are mostly based on single-scene remote sensing image of a certain period and cannot show the highly variable characteristics of water bodies over time and space. Although some time series water products have appeared in China and abroad, their spatial resolution and water boundary accuracy still cannot meet the needs of some studies and applications. This paper takes the Yangtze River Basin with complex surface environment as the research area based on the Google Earth Engine (GEE) cloud platform. The Sentinel-2 MSI annual long time series image sets are combined with the “temporal characteristics” of pixels, and a high-precision water extraction algorithm with more universality, operability, and better effect in large-scale environment is proposed. Specifically, an algorithm based on time series image data is combined with multi-index and “temporal characteristic” fusion Digital Elevation Model (DEM). This algorithm selects automated water extraction index, Modified Normalized Difference Water Index (MNDWI), normalized difference vegetation index, and enhanced vegetation index for multi-index logical combination to extract water bodies. Near-infrared band reflectivity value and slope data set generated by SRTM DEM are used to assist in suppressing high reflectivity noise and shadow noise. The accuracy of water bodies in the whole basin is verified with the validation sample points, and the correct extraction rate is more than 96% through visual interpretation. The accuracy evaluation at the subpixel level shows that the mixed edge pixels account for 3.37% of the total pixels, the misclassification error is 0.46%, and the omission error is 0.21%, indicating that the proposed algorithm has a good inhibitory effect on the mixed pixels. Compared with the traditional NDWI and MNDWI water index based on spectral characteristics, the multi-index combined with temporal characteristic algorithm has better effect in suppressing shadow noise. Compared with some existing water products, the proposed algorithm can ensure the integrity of the whole water area and retain the local details of the water body. It has certain advantages in the extraction of small water bodies. Results of the remote sensing extraction of water bodies in the Yangtze River Basin show that the spatial distribution of water bodies in the basin is uneven, and the temporal and spatial changes in various water body types are obvious. From 2017 to 2020, 67.41% of the increase in permanent water bodies is transformed from seasonal water bodies, and the mutual conversion between seasonal water bodies and nonwater bodies is the most obvious. In addition, 74.64% of the increase in seasonal water bodies is converted from nonwater bodies, and 56.25% of the decrease in seasonal water bodies is converted to nonwater bodies. Experimental results show that the proposed algorithm has a certain universal importance in extracting water bodies in different spatiotemporal locations and different environments and can effectively avoid the phenomenon of “the same objects with different spectra” and “the same spectra with different objects” caused by the mixing of water and other ground objects. This algorithm has a good inhibitory effect on complex background noise and has high accuracy and precision.关键词：Google Earth Engine(GEE);Sentinel-2;remote sensing extraction of water bodies;temporal characteristics;Multi-index combination;shadow noise- 2275
- |
- 745
- |
- 8
发布时间：2022-02-28 -
摘要：Intertidal wetlands are the transitional zone between terrestrial and marine ecosystems, and they are of ecological and economic importance. However, intertidal wetlands are severely damaged due to natural causes (e.g., climate change and sea-level rise) and anthropogenic causes (e.g., coastal reclamation and excessive tourism development). Therefore, tracking the spatiotemporal changes of intertidal wetlands is important for scientific management and high-quality development of coastal areas. Compared with traditional surveying methods, remote sensing has better capacity in monitoring intertidal wetlands dynamically on a large scale. Acquiring complete information of intertidal wetland from a single-phase remote sensing image is difficult owing to the influences of cloudy weather and tidal periodic submergence. The problem of extracting the information of the intertidal zone under the influences of dynamic tidal submerge should be solved for the application of remote sensing in coastal areas.In this study, we proposed a combined method using the time-series remote sensing indices and the geographic characteristics in the subtropical intertidal wetland of Fujian Province, China on the basis of the GEE platform. Three main types of intertidal wetlands including high marsh, low marsh, and tidal flat were classified by the following steps. First, water and vegetation indices were utilized to extract water bodies and vegetation from every single image. Second, the water and vegetation frequencies derived from time-series images were used to distinguish the high marsh, low marsh, and tidal flat according to the tidal dynamics and vegetational phonology. Finally, the accuracy of the results was verified by the high-resolution image on Google Earth Pro and in situ data. The results were compared with similar datasets to assess the reliability and robustness of the proposed method.The overall classification accuracy was 97.47%, and the Kappa coefficient was 0.96. The verifications showed misclassifications in the transitional area. The total area of intertidal wetlands in Fujian Province was 1061.3 km2, and the areas of high marsh, low marsh, and tidal flat were 18.1, 137.3, and 905.8 km2, respectively. Intertidal wetlands were concentrated in estuaries and bays. The area of tidal flat decreased from north to south along the coast, but a converse trend of the area of high marsh was observed. The vegetation was mainly distributed in the southern Quanzhou Bay, Jiulongjiang Estuary, and Zhangjiang Estuary, and it was less in northern Fujian. Comparing the results of this study with similar datasets showed that our study improved classification accuracy in the Fujian Province. However, some objective factors such as mixed pixels and clouds could affect the accuracy of the classification.This research developed a method based on the GEE platform and time-series remote sensing indices to classify intertidal wetlands for overcoming the dilemma faced by single-phase remote sensing images in the intertidal zone information extraction. The results showed certain superiority compared with similar datasets during the same period. The method reduced the impact of the year-round cloudy and rainy weather in the subtropical coastal zone and tidal dynamics effectively. The present datasets will provide important basic data and technical supports for the sustainable management and utilization of coastal resources of the region.关键词：Time-series remote sensing data;intertidal wetlands;GEE;Fujian province;phenology;frequency-based adgorithm- 3474
- |
- 611
- |
- 4
发布时间：2022-02-28 -
摘要：Wetland is an important ecosystem on the planet and plays a pivotal role in maintaining global ecological environment security. Traditional wetland monitoring requires a lot of manpower and financial resources due to the unique hydrological characteristics of wetlands, and extracting large-scale wetland information is also difficult. Compared with traditional field surveys, remote sensing technology, which has the advantages of wide observation range, short update cycle, has played an important role in large-scale wetland information extraction. However, remote sensing monitoring of march wetland mainly uses optical images traditionally, which are severely affected by weather such as clouds and rain, making large-scale marsh wetland extraction challenging. The use of radar and terrain data can combine the spectral information and scattering mechanism, which has great potential for marsh wetland information extraction. Nevertheless, there are few studies that evaluate the differences of optical, SAR, and topographic features in importance for the extraction of marsh wetland information. The rise of big data and cloud computing has enabled large-scale and long time series spatial data processing. On the basis of the Google Earth Engine (GEE) cloud platform, this study uses Sentinel-1 synthetic aperture radar data, Sentinel-2 optical data, and terrain data to explore their importance to the extraction of marsh wetland at large scale, and verify the feasibility of JM distance to find the optimal feature combination to the extraction of marsh wetland. Random forest algorithm is also used to extract marsh wetlands in Heilongjiang Basin in 2018. In order to explore the importance of red edge, radar and topographic features and the best features conducive to marsh wetland extraction, six experimental schemes are designed. Scheme one uses the combination of spectral feature, vegetation index and water index. Scheme two uses the combination of spectral feature and red edge feature. Scheme three uses the combination of spectral feature and terrain feature. Scheme four uses the combination of spectral feature and radar feature. Scheme five uses the combination of spectral feature, vegetation index, water index, red edge feature, terrain feature and radar feature. Scheme six uses all features, which are optimized by JM distance. The research shows that (1) Sentinel-2 red edge bands and Sentinel-1 radar bands and terrain data are conducive to marsh wetland information extraction. Compared with vegetation indexes and water indexes, the producer accuracy of marsh increased by 7.56%, 5.04%, and 4.48%. (2) The separation obtained using JM distance shows that the order is red-edge features > other optical features > terrain features > radar features. The marsh wetland in scheme six has the highest mapping accuracy and user accuracy. After using the JM distance to select features, the producer and user accuracies of the marsh wetland increased by 1.45% and 3.02%, respectively. The overall accuracy of the combined random forest algorithm was 91.54%, and the accuracy of marsh extraction was 88%. This study uses the GEE cloud platform, multisource remote sensing data, and machine learning algorithms to accurately, quickly, and efficiently extract large-scale marsh wetland information. This method has great application potential.关键词：remote sensing;Google Earth Engine;marsh;Sentinel-1;Sentinel-2;JM distance;Random Forest;red-edge bands- 2546
- |
- 1803
- |
- 18
发布时间：2022-02-28
Remote Sensing Cloud Computing
-
摘要：The Hughes problem together with strong intra-band correlations and massive data seriously hinders hyperspectral processing and further applications. Dimensionality reduction using band selection can be used to conquer the abovementioned problems and guarantee the application performance of hyperspectral data. In particular, spectral clustering is a typical method for high-dimensional hyperspectral data. This method finds clusters of all hyperspectral bands on the connected graph and selects the representatives. Unfortunately, the regular similarity measures are negatively affected by outliers or noise of hyperspectral data in measuring the similarity of different bands. They could also only represent one feature of band similarity and have respective limitations. Accordingly, the obtained similarity matrix could not represent the full information of band selection required and could not guarantee obtaining aimed bands from spectral clustering. Therefore, we propose a Robust Multifeature Spectral Clustering (RMSC) method to solve the two problems mentioned above and enhance the performance of hyperspectral band selection from spectral clustering.The RMSC combines multiple features of similarity measures for pairwise bands, namely, information entropy, band correlation, and band dissimilarity, to construct the integrated similarity matrix. It utilizes spectral information divergence to quantify the information entropy between pairwise bands. The coefficient correlation is utilized to measure the band correlations and construct the similarity matrix of band correlations. The Laplacian graph is also adopted to construct a similarity matrix and show the dissimilarity between different bands considering the inner clustering structure of all bands. The spectral angle distance matrix is constructed as well to reflect the similarity from the aspects of overall differences. The RMSC regards that each similarity matrix of all four features reflect the underlying true clustering information of all bands and has low-rank property. It formulates the estimation of combined dissimilarity matrix into a low-rank and sparse decomposition problem and utilizes the augmented Lagrangian multiplier to solve it. Thereafter, it implements the regular spectral clustering on the integrated similarity matrix and selects the representative bands from each cluster.Two hyperspectral datasets are used to design four groups of experiments and testify the performance of RMSC. Five state-of-the-art methods, namely, WaluDI, fast density-peak-based clustering, orthogonal projections based band selection, Improved Sparse Spectral Clustering (ISSC) and SC-SID, and support vector machine, are used to quantify the classification accuracy. Experimental results show that RSMC outperforms the five other band selection methods in overall classification accuracy with shorter computational time. The regularization parameter is insensitive to RMSC, and a small candidate could produce high classification accuracy.RMSC is better in selecting representative bands than current spectral clustering such as ISSC. It can also be a good choice in hyperspectral dimensionality reduction.关键词：remote sensing;hyperspectral remote sensing;dimensionality reduction;band selection;spectral clustering;robust multi-feature spectral clustering- 1907
- |
- 1272
- |
- 1
发布时间：2022-02-28 -
摘要：High-resolution remote sensing images are complex and difficult to classify. Deep learning methods can extract more and deeper information of the features, which is suitable for the classification of high-resolution remote sensing images. This paper studies the high-precision classification of impervious ground, buildings, low vegetation, trees, vehicles, and other features in high-resolution images. An E-DeepLab network model is proposed by combining the characteristics of remote sensing multiground feature classification based on DeepLab v3+ network model. The main improvements are as follows: (1) Improving the combination of encoder and decoder modules and using a simple and effective addition connection method. (2) Reducing the single upsampling multiple, increasing the upsampling layer, and improving the tightness of the connection between the encoder and the decoder. (3) Using the improved adaptive weight loss function to automatically adjust the weight of losses. In accordance with the characteristics of the data, a multichannel training method combining digital surface model and normalized difference vegetation index data is proposed. Using the data from the two regions to conduct experiments, the overall extraction accuracy in Potsdam region reaches 93.2%, the extraction accuracy in buildings reaches 97.8%, the overall extraction accuracy in Vaihingen region reaches 90.7%, and the accuracy of building extraction reaches 96.3%. The accuracy of the two regions is significantly better than the original DeepLab v3+ model and other related models. The visual effect of the classification results is extremely close to the standard map by comparing the classification map and the standard marker map. Results show that the E-DeepLab network has good application value in the feature extraction and classification of high-precision remote sensing images.关键词：remote sensing;high-resolution images;deep learning;E-DeepLab;adaptive weight loss function- 1530
- |
- 791
- |
- 4
发布时间：2022-02-28 -
摘要：Deep learning has strong learning ability and has become a widely studied method in the hyperspectral image classification community. However, the deep learning-based classification model requires a large number of training samples to train a good model. Overfitting will occur when the training sample is small. The accuracy of the model on the test set is lower than the accuracy on the training set. Researchers have proposed overfitting suppression methods such as weight decay and dropout to suppress overfitting. However, these methods need to work in a specific environment and have limited suppression effect on overfitting. Thus, this study proposes an overfitting suppression algorithm based on generative adversarial networks to suppress the overfitting phenomenon of the model.First, a spatial neighborhood block for the standard dataset is constructed, and the dataset is divided into labeled, unlabeled, and test samples. Then, the labeled and unlabeled samples are sent to the generative adversarial networks for training. During input, the pixels in the neighborhood block are independently fed into the fully connected network discriminator to extract the spectral features of each pixel. Finally, the spectral features of each pixel are fused by the average pooling, and they connected to the output layer to obtain the classification result. The overfitting is caused by the large value and variance of the network parameters. Thus, the large parameter values enable the model to fit more samples. Therefore, the network is first fitted to the data by labeled samples in each iteration, and then, the optimizer is used to minimize the mean of the high-dimensional features. This process will re-update the network parameters, reduce the value and variance of the parameters, and thus suppress the overfitting.The algorithm was applied to two standard datasets, namely, Indian Pines and Pavia University datasets. The 1% labeled samples were randomly selected for training. The overall classification accuracy rates were 89.61% and 98.79%, which were better than those of several algorithms. Compared with several commonly used overfitting suppression methods such as batch normalization, L2 regularization, and dropout, the proposed overfitting suppression algorithm obtains 5.60% and 3.20% higher results on randomly selected 1% labeled samples from the Indian Pines dataset and randomly selected 0.1% labeled samples from Pavia University dataset.The model of generative adversarial networks designed for the characteristics of hyperspectral data can fully utilize the spectral and spatial features of hyperspectral images. The proposed overfitting suppression algorithm can significantly improve the classification performance of the model. However, the overfitting suppression effect of the algorithm is not obvious when the number of labeled samples is large. Thus, further research is needed.关键词：remote sensing;hyperspectral image classification;small training samples;overfitting;generative adversarial network;spectral-spatial feature;feature extraction- 1343
- |
- 604
- |
- 1
发布时间：2022-02-28
Remote Sensing Intelligent Interpretation
- Postal code：100190
- Tel：010-58887052 Email：nrsb@aircas.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备20021838号-9
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0