最新刊期
卷 28 ， 期 3 ， 2024
-
摘要：Earth is a complex giant system. The spheres of earth, such as the Atmosphere, Hydrosphere, Biosphere, and Lithosphere, are interconnected and interact on each other as subsystems of the giant system. However, achieving a comprehensive and profound understanding of the spatial-temporal processes and interaction mechanisms within Earth system remains an imposing scientific endeavor. One significant obstacle is the limited availability of fine and interior data of spheres that can support groundbreaking researches on critical Earth system science topics and applications of remote sensing observation to the earth. Thus, there is an urgent need for systematic, multi-dimensional, and multi-scale observation information of the interior of the earth’s spheres to effectively address these concerns.Utilizing multiple observing means—including satellites, aircraft, and ground-based systems, along with advanced remote sensing detecting technologies that cover a spectrum from electromagnetic to microwave, laser, and gravity, the physical elements, internal structure, and evolution processes of the atmosphere, hydrosphere, biosphere, and lithosphere can be finely and comprehensively measured. Integration on diverse observing means will greatly enhance the ability to acquire highly accurate and deeply detailed insights into the earth and its subsystems.This paper proposed the concept of Transparent Earth-Observing (TEO) and its composition，which refers to the integration of new transparent-detecting model and advanced remote sensing technologies to acquire comprehensive, dynamic, and multi-dimensional interior information of Earth spheres. Transparent Earth-Observing helps to overcome the limitations of traditional Earth observation methods that primarily focus on the Earth’s surface, facilitating multi-dimensional, penetrative, high-resolution sensing of internal structures and process variation within the atmosphere, forests, oceans, and solid Earth. Transparent Earth-Observing, to a certain extent, symbolizes a new era in earth observation. This paper also put forwards a scheme to establish prototype system of Transparent Earth-Observing, which primarily focuses on design and implementation of airborne transparent observing platform, a novel remote sensing flying laboratory. The system follows a methodology that includes transparent-detecting methods, payload integration, field experiments, parameter retrieval, and validation and assessment. The prototype system of Transparent Earth-Observing enables the integration of multidimensional remote sensing technologies and the consolidation of diverse interior detecting information of the Earth sphere, thereby enhancing the dimension, scale, and density of observation information regarding various physical quantities and broadening our cognitive horizons.In summary, as an exploration research and development direction of the next generation of Earth observation, the Transparent Earth-Observing will support the quantitative understanding of internal structures within Earth spheres and then promote the advancement of Earth System Science. At the same time, it will also stimulate innovation in core areas such as aerospace technology, enhance humanity’s ability to develop and utilize resources, predict and respond to extreme weather events and global environmental changes.关键词：Transparent Earth-Observing (TEO);remote sensing;earth observation;Atmosphere;forest;solid-earth;ocean- 924
- |
- 318
- |
- 0
发布时间：2024-04-08 -
摘要：Lake ice is not only an important part of the cryosphere but also one of the most direct indicators of global climate change. In the context of climate warming and intensified human activities, global lake ice presents a trend of delayed freeze onset date, advanced break onset date, shortened ice cover duration, and thinning ice thickness. This trend tends to last for a long time. Consequently, a series of chain reactions of lake physical hydrology, hydrochemistry, and ecosystem will inevitably be triggered, further heaving the burden of natural environment and habitat construction. Therefore, it is necessary to perform fine-scale monitoring and scientific analysis of spatiotemporal patterns on lake ice variations for further predicting the early warning of global climate change. Toward overcoming the limitation of in-situ surveys, remote sensing technique comes to play a significant role in lake ice monitoring, which can provide large-scale, long time series, and high temporal resolution data for lake ice research. Previous efforts always focus on lake ice and its response to climate change using different remote sensing sensors, parameters, and characteristics. Through reviewing pioneering research, this study presents a general review on the remote sensing data source and methods for lake ice studies as well as spatial and temporal variations of lake ice in global hotspots. This paper first reviews the development of the commonly used remote sensing data sources for lake ice monitoring, which include spaceborne and airborne remote sensing platforms and existing lake ice data products. Then, the methods of lake ice identification and retrieval of lake ice phenology and ice thickness parameters are compared and discussed. Threshold and index-based methods are commonly used in lake ice research. According to the previous studies, this review likewise summaries the research hotspots of lake ice and analyzes the spatial and temporal characteristics of lake ice variations. The research hotspots are mostly distributed in the Northern hemisphere, especially in Northern Europe, North America, and the Tibetan Plateau. In addition, influencing factors of lake ice variations, including climate factors and lake shape attributes, are discussed in this study. Finally, future development directions of lake ice study by remote sensing are discussed as follows: (1) to fully integrate multiple satellite data at medium and high spatial resolution to improve the accuracy of lake ice observations, particularly for small- and medium-sized lakes; (2) to reconstruct the long time series of lake ice phenology and thickness information and predict their future changes based on techniques such as big earth data and machine learning methods; and (3) to focus more on the research of past, present, and future of lake ice variation characteristics in the Tibetan Plateau, which is rather sensitive to climate change and remains largely unexplained. Remote sensing is an effective tool to monitor the variations of lake ice, yet what we should do imperatively is to advance the scientific understanding on climate change impacts and take immediate actions.关键词：lake ice;Lake ice phenology;ice thickness;remote sensing monitoring;climate change- 675
- |
- 298
- |
- 0
发布时间：2024-04-08 -
摘要：Remote sensing time series contain information about the changes and differences in forest composition, structure, and function driven by natural factors and human activities. It provides theoretical support for forest disturbance detection and attribution by integrating spatio-temporal-spectral information from remote sensing time series data, which can effectively improve the understanding of the processes of forest succession, developmental trend, and their driving and response mechanism. This paper systematically reviewed the research progress of forest disturbance detection and attribution by integrating spatio-temporal-spectral information from remote sensing time series.The prerequisite for forest disturbance attribution is the detection of forest disturbance events, and the accuracy of disturbance detection directly affects the accuracy of subsequent attribution. In this paper, current forest disturbance detection methods and techniques are highlighted from multiple perspectives, including data (time series observation frequency selection), features (spectral feature selection, fusion of spatial and temporal feature), and algorithms (multi-algorithm integration, forest low-intensity disturbance detection). From the data perspective, based on the frequency of available observations in different regions, the change detection methods for dense and sparse time series are introduced respectively. From the feature perspective, the spectral response characteristics of forest disturbances are summarized. The change detection strategy of multi-spectral feature integration is introduced to address the problems of change detection based on single spectral features. The fusion of temporal and spatial features for forest disturbance detection is summarized. From the algorithmic perspective, to address issues such as differences in the results of different change detection algorithms and the fact that a single algorithm may not be the most efficient way to describe all conditions, two multi-algorithm integration strategies, parallel and serial, are presented. Based on our analysis of the reasons for the poor detection of low-intensity disturbances (e.g., selective logging, pests and diseases, drought, etc.), progress in research on change detection oriented to mid- and low-intensity disturbances in forests is described.The essence of forest disturbance attribution is a classification problem involving multiple types of forest disturbances. This process identifies disturbance types by utilizing remote sensing features of forest disturbances caused by different driving factors as inputs for classification algorithms. In this paper, we first summarized attribution features as the input of forest disturbance attribution, that is, pre-, mid-, and post-disturbance features in chronological order and temporal, spatial, spectral, and topographic features in feature dimensions. Then, according to the condition of whether disturbance detection occurs before the attribution of disturbances, methods for attributing multiple types of forest disturbance are summarized and compared. These methods are based on the spatio-temporal-spectral and topographic features of remote sensing time series, including the direct method and the two-stage method.Lastly, we analyzed the current problems in forest disturbance monitoring using remote sensing and predicted the future research directions, such as the fusion of spatio-temporal-spectral features, simultaneous detection of forest multi-intensity disturbance, and attribution of forest multi-type disturbance under limited sample conditions. We hope this article becomes a reference for the detection and attribution of changes using an ensemble of spatio-temporal-spectral information from remote sensing time series.关键词：forest disturbance;remote sensing time series;spatio-temporal-spectral information;feature ensemble;disturbances detection;attribution of disturbances- 387
- |
- 392
- |
- 0
发布时间：2024-04-08 -
摘要：Forests are the largest ecosystems on land and play an important role in the global carbon and oxygen cycle. Synthetic aperture Radar tomography (TomoSAR) has the capability to carry out three-dimensional (3-D) imaging of observation targets and obtain information about forest internal structure, which serves an important function in the inversion of forest parameters. This paper will review the imaging methods and applications of TomoSAR over the past two decades and focus on its latest research progress in forest parameter inversion. More importantly, different parameter inversion methods will be systematically compared, and the challenges in TomoSAR forest parameter inversion will be analyzed.First, the mathematical models of TomoSAR in single-polarization and full-polarization mode were introduced. Then, different TomoSAR imaging algorithms were analyzed in detail. The performances of different methods in terms of vertical resolution, radiation accuracy, computational efficiency, and stability were compared. Next, we summarized the progress of TomoSAR in the inversion of forest parameters, such as underlying topography, forest height, and biomass. Finally, this paper analyzed the key challenges faced in the inversion of forest parameters using TomoSAR and predicted the frontier applications of TomoSAR. The P-band TropiSAR 2009 dataset over a test site in Paracou, French Guiana, were used to analyze the performance of different methods.By reviewing the published literature, the theoretical differences between different TomoSAR imaging algorithms were listed. Experiments showed that the Fourier transform method has limited vertical resolution but high radiation accuracy and has been successfully used for biomass estimation. Beamforming spectral estimation method can improve the vertical resolution, but the image quality is seriously degraded when the number of observations is reduced. Compressed sensing and statistical optimization algorithms have sparse imaging capabilities and super-resolution, enabling the fine-grained identification of forest vertical structures. For the estimation of forest underlying topography and forest height, an accurate estimation of canopy scattering center and ground phase center is an important prerequisite. The addition of polarization information is more conducive to the identification of different scattering mechanisms. In biomass estimation, the application of a 3-D structure can significantly improve the accuracy of inversion.The 3-D structure of forests plays an important role in the estimation of forest parameters. TomoSAR can reconstruct the 3-D structure of forests through specific imaging techniques. In general, high-resolution imaging algorithms are beneficial to distinguish and identify scatterers with different heights and are widely used in underlying topography and forest height estimation. However, for biomass estimation, radiation accuracy is more of a concern for researchers. At present, the most critical challenge of TomoSAR is the data processing and application of spaceborne data. The main difficulties include the correction of time decoherence and atmospheric delay errors. In the future, long-wavelength TomoSAR systems will become one of the most important approaches for forest biomass estimation on a global scale.关键词：remote sensing;SAR Tomography;three-dimensional imaging;underlying topography;forest height;forest vertical structure;forest biomass- 985
- |
- 431
- |
- 0
发布时间：2024-04-08
Reviews
-
摘要：The 5 m Optical Satellite 02 (ZY1-02E) was successfully launched on December 26, 2021. Along with the previous model (ZY1-02D), an operational constellation of mid-resolution satellites will be established, forming a large-scale, quantitative, and comprehensive earth observation capability. This satellite was designed to provide 2.5 m panchromatic/10 m multispectral, 30 m hyperspectral, and 16 m thermal infrared image data. The whole project was led by the Ministry of Natural Resources. It will support national monitoring and survey and mapping projects at the scale of 1∶100000—1∶250000. Compared with traditional multispectral data, hyperspectral data can obtain more abundant spectral information of ground objects, which is particularly important for fine-grained survey and monitoring tasks. The development of China’s hyperspectral remote sensing technology is basically synchronized with the international frontier level. However, since the birth of hyperspectral remote sensing technology, the main way to obtain data has relied on human-operated aerial platforms, and for a long period of time, there has been no large-scale acquisition capacity from the data source. Therefore, the hyperspectral payload carried by this satellite is its biggest highlight and has attracted widespread attention from domestic and foreign peers. During the in-orbit test, the Ministry of Natural Resources designed a total of 32 application test items, focusing on land resources, geology and minerals, surveying and mapping geographic information, ocean island monitoring, and other fields. All application items and other key indicators were evaluated in accordance with the relevant standards for natural resource survey and monitoring and technical specifications. The result shows that the satellite system, ground system, and application system have been operating stably and maintaining good consistency between each satellite of the constellation. It meets the data quality requirements of main businesses in various fields, including natural resource supervision and enforcement, geological and mineral resource investigation, ecological restoration project monitoring and evaluation, geospatial information update, coastline change monitoring, and steel industrial capacity monitoring. Hyperspectral and thermal infrared payloads have achieved good application results. This paper will introduce the data characteristics and application capabilities of the satellite, focusing on the evaluation results of some specific applications, especially with hyperspectral and thermal infrared cameras. Hopefully, this paper can provide a reference for the further application of this satellite.关键词：remote sensing;5 m Optical Satellite 02 (ZY1-02E);natural resources;satellite application;visible/near-infrared;hyperspectral;thermal infrared- 294
- |
- 321
- |
- 0
发布时间：2024-04-08 -
摘要：Thermal infrared images are widely used in the field of land resource management, fire detection, and missile warning and surveillance. To improve China’s ecological management and strengthen the supervision of high-energy-consuming enterprises (e.g., steel plant), the requirements for the development of a high-resolution, high-sensitivity, and large-width space thermal infrared camera are proposed. On this basis, the thermal infrared camera installed on the 5 m Optical 02 Satellite was designed and developed. The thermal infrared camera was successfully launched at the Taiyuan Satellite Launch Center on December 26, 2021.The thermal infrared camera adopts an advanced push-broom imaging system and uses an 8192 pixel long-line array long-wave infrared detector to obtain thermal infrared images in the 7.7—10.5 μm spectrum. It chooses an off-axial three-mirror main optical system and a transmission system with free-form surfaces to achieve an 8.6°×1.1° field of view. The absolute distortion of the optical system is less than 1 μm, the average MTF is better than 0.32, and the uniformity of the focal plane illumination is better than 0.98. The focal plane of the thermal infrared camera is composed of eight CCDs. The number of pixels of each CCD is 1024×6×2. Time delay and integration technology is used to improve the SNR ratio. The material deformation matching method is used to ensure the thermal stress and displacement meet the requirement at a low temperature of 80K. The thermal infrared camera also uses cryogenic optical technology to reduce the system background noise and improve the dynamic range of the camera. Three lens close to the focal plane work at 200 K. The cryogenic lens is cooled by a pulse tube cryocooler. At the same time, a calibration mechanism and focusing mechanism are installed inside the thermal infrared camera, which can realize on-orbit radiation calibration and focal plane adjustment. These designs improve the radiometric accuracy and reliability of the camera. After completion of the detailed design, the thermal infrared camera has a ground sampling distance of 15 m and a coverage width of 120 km. The NETD of the thermal infrared camera is better than 0.1 K@300 K. The thermal infrared camera achieves high-resolution, high-sensitivity, and large-width observation simultaneously.After the alignment and ground test of the thermal infrared camera, the average MTF of the entire field of view reaches 0.151, NETD reaches 0.06 K@300 K, dynamic range exceeds 240—340 K, and linearity of the radiometric calibration curve is better than 0.99. After the 5 m Optical 02 Satellite was launched, the on-orbit test was completed. The image geometric performance and radiation performance on-orbit are excellent, and the resolution and width meet the requirements. An analysis of the image data of high-temperature and low-temperature ground objects near 300K indicates that the on-orbit NETD is close to 0.08K, and the absolute radiation calibration accuracy on-orbit is better than 1 K.Compared with the GF-5 02 VIMS launched in China in 2021 and the Landsat 9 TIRS-2 launched in the United States in 2021, the thermal infrared camera onboard the 5 m Optical 02 satellite has greatly improved the resolution and maintained a large width. At the same time, the NETD is close to the best in the world. The thermal infrared camera can well support businesses related to land resources.关键词：remote sensing;Thermal infrared camera;Push-broom imaging;Infrared detector;5 m Optical 02 Satellite- 653
- |
- 265
- |
- 0
发布时间：2024-04-08 -
摘要：Hyperspectral image supervised classification is a crucial and challenging task in remote sensing, as its performance depends heavily on the quantity and quality of labeled samples. However, labeling hyperspectral data is a difficult and time-consuming procedure. This problem results in a limited number of labeled samples in real-world scenarios, rendering the supervised classifiers vulnerable to the issue of overfitting. To address this problem, researchers have sought solutions in the field of unsupervised domain adaptation, utilizing labeled samples from previous images (source domain) to classify new hyperspectral data (target domain). Most existing domain adaptation methods strive to learn domain-invariant features in a new space, but many of them focus on aligning the overall statistics of the two domains without considering the spectral shifts in each class. Other methods attempt to align every class of the source and target domains simultaneously but often overlook the issues of mixture of samples across classes and incorrect sample selection. This may lead to a negative transfer and reduced separability of data. The significant discrepancies across domains will further compound the problem.In this paper, we propose a novel class-independent domain adaptation algorithm that addresses these issues in hyperspectral image classification. Our method first creates an independent subspace for each class and then aligns the corresponding single-class samples of the two domains in those subspaces. The posterior probabilities are learned independently through the aligned samples in each subspace. Then, the posterior probabilities obtained from multiple subspaces are fused to produce the final classification labels, aiming at increasing the confidence of results. Additionally, we use smoothed classification labels as pseudo labels for further iteration and incorporate a strategy for selecting representative samples to enhance subspace performance.Experimental results on two real hyperspectral datasets demonstrate the high classification performance of our proposed method. Compared to the joint domain adaptation algorithm, our method with the nearest neighbor classifier improved the accuracy by 9.56% on the Honghu data and 18.45% on the Wen-County data. Compared to other competitors, our method also has the advantage of generating smoother classification maps with more distinct boundaries of ground objects. These remarkable results stem from the substantial improvement in data separability achieved by our approach, which has been validated through calculations in our experiments.In conclusion, our proposed class-independent domain adaptation algorithm is a promising solution for hyperspectral image classification, providing high performance with reduced risk of overfitting. By aligning samples in the class-independent subspaces and fusing posterior probabilities, our method leads to improved data separability and more accurate classifications. Furthermore, the use of representative sample selection helps mitigate the potential impact of mislabeled samples on class alignments. Thus, our algorithm is able to overcome the limitations of existing domain adaptation methods and achieve improved results. In future work, we plan to extend our method to more complex high-dimensional datasets and incorporate advanced deep learning models. We also intend to evaluate the applicability of our method for real-time hyperspectral image classification, which is a critical requirement for many remote sensing applications. Overall, our research represents a significant advancement in the field of hyperspectral image classification, offering a new approach for solving the challenge of insufficient labeled samples.关键词：remote sensing;hyperspectral image;domain adaptation;classification;class-independent subspace;ZY1-02D;GF-5- 592
- |
- 226
- |
- 0
发布时间：2024-04-08 -
Farmland shelterbelt information extraction based on multispectral image of the ZY1-02E satellite AI导读
摘要：The successful launch and in-a-bit operation of the 5 m Optical Satellite 02 (ZY1-02E) have provided a wealth of remote sensing data for various main businesses within the forestry and grass industry, providing reliable information support for forestry management and ecological services. This study aims to test the application capability of ZY1-02E multispectral data in farmland shelterbelt monitoring, a primary forestry business. Zhangbei County, Hebei Province, serves as the study area. Spectral, vegetation index, and texture feature sets are constructed based on the ZY1-02E multispectral data, and four classification information extraction schemes are designed: (1) spectral features, (2) spectral features + vegetation index, (3) spectral features + texture features, and (4) spectral features + vegetation index + texture features. Random forest algorithm was employed for feature selection, classification information extraction, and validation to evaluate the application potential and effectiveness of the ZY1-02E multispectral data in farmland shelterbelt information extraction. The results show that (1) ZY1-02E multispectral data allow for the accurate extraction of farmland shelterbelt information in the study area, reflecting the actual distribution of farmland shelterbelts to a high degree. Among them, the overall accuracy and Kappa coefficient of Scheme 1 are 0.8371 and 0.7760, respectively; the overall accuracy and Kappa coefficient of Scheme 2 are 0.8440 and 0.7855, respectively; the overall accuracy and Kappa coefficient of Scheme 3 reach 0.8839 and 0.8403, respectively; and Scheme 4 has the highest accuracy, with its overall accuracy and Kappa coefficient being 0.8908 and 0.8499, respectively. (2) The effective use of multiple feature variables can significantly improve the accuracy of farmland shelterbelt information extraction. Regarding the contribution of different features to farmland shelterbelt information extraction, in terms of their contribution to farmland shelterbelt information extraction, the spectral features are the most significant, followed by texture features and vegetation indices. (3) ZY1-02E multispectral data exhibit high accuracy and reliable results for farmland shelterbelt information extraction, which can better meet the needs of protection forest monitoring operations and has considerable potential for application in forest surveys and monitoring thematic operations. In conclusion, this study demonstrates the potential and effectiveness of ZY1-02E multispectral data for extracting farmland shelterbelt information. Using multiple feature variables and the random forest algorithm enables the accurate extraction and validation of farmland shelterbelt information, providing valuable insights for future forest monitoring and management. As more data become available and the application capabilities of the ZY1-02E are further explored, future work can consider integrating multispectral data from different periods and linear features of farmland shelterbelt to enhance the accuracy of information extraction, ultimately achieving more efficient and precise extraction of farmland shelterbelt information.关键词：remote sensing;forestry and grass industry;farmland shelterbelt;feature extraction;application testing;ZY1-02E satellite- 349
- |
- 188
- |
- 0
发布时间：2024-04-08 -
摘要：Spartina alterniflora, an introduced coastal wetland plant, has rapidly propagated and expanded in recent years due to its strong adaptability and tolerance to climate and environment. Owing to the rapid reproduction and expansion of S. alterniflora, it has invaded the ecological niche of native vegetation and caused serious damage to the local ecosystem. Remote sensing technology can realize long-term, large-scale, real-time dynamic and accurate surveys and be effectively applied to the precise monitoring of S. alterniflora. Furthermore, it can provide guidance and a basis for the management of S. alterniflora and the restoration of coastal wetland ecosystems. Wetland management provides accurate, real-time, and dynamic information and technical support.To solve the problem of rapid and precise identification of S. alterniflora, this study proposed a method to construct the growth period of a S. alterniflora index based on ZY1-02D hyperspectral data. On the basis of the characteristics of S. alterniflora and other salt marsh wetland vegetation in near-infrared and short-wavelength infrared bands, the differentially sensitive bands were selected to construct an S. alterniflora index, increasing the spectral difference between S. alterniflora and other salt marsh vegetation in the complex coastal wetland environment. This outcome effectively reduces the problem of difficult vegetation discrimination caused by the phenomenon of different spectra of the same objects and the same spectrum of different objects in surface cover.In this study, two national nature reserves, the Yellow River Delta wetland and Yancheng coastal wetland, were selected as the study area, and ZY1-02D hyperspectral images of September were selected as the data source. First, image preprocessing was performed to obtain the reflectance data of the study area. Second, the differentially sensitive bands were determined, the Growing Period Normalized Difference Spartina Alterniflora Index (GNDSAI) was constructed, and the decision tree was constructed to extract the information of S. alterniflora accurately. Finally, the qualitative and quantitative accuracy of the classification results were evaluated, with Normalized Differential Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Support Vector Machine (SVM), Random Forest (RF), Maximum Likelihood Classifier (MLC), and Artificial Neural Network (ANN) selected for comparative evaluation.Experimental results show that the proposed method has strong regional adaptability in the two study areas. S. alterniflora can be well separated from other wetland salt marsh vegetation in the box plot of the GNDSAI value. The optimal lower threshold of GNDSAI for the Yellow River Delta wetland is 0.4 and that of the Yancheng coastal wetland is 0.27. The average producer’s accuracy and user’s accuracy of S. alterniflora were 92.00% and 91.68%, respectively, which were better than those of the other classification methods. Spectral heterogeneity of S. alterniflora, selection of band combinations, variability of thresholds, and influence and uncertainty of tide levels are discussed in the text.In this study, GNDSAI was proposed using ZY1-02D hyperspectral images. Considering the phenological characteristics of S. alterniflora, GNDSAI was constructed by using the four bands of near-infrared (765 nm), near-infrared (842 nm), short-wave infrared (1644 nm), and short-wave infrared (2216 nm) to enhance the difference between S. alterniflora and other wetland salt marsh vegetation through the calculation of spectral bands. Combined with MNDWI and prior knowledge, the decision tree classification model based on GNDSAI was designed, which realized simple, rapid, and accurate extraction of S. alterniflora, providing a new idea or method for information extraction of S. alterniflora.关键词：remote sensing;spartina alterniflora;vegetation index;spartina alterniflora index;hyperspectral data;ZY1-02D- 177
- |
- 79
- |
- 0
发布时间：2024-04-08
Specifications and Applications of 5 m 02 Satellite
-
Comparative evaluation of simulation methods for the maximum light-use efficiency of vegetation AI导读
摘要：The light-use efficiency model is a parametric model for estimating vegetation productivity based on remote sensing data. Its core parameter, maximum light-use efficiency (LUEmax), was considered a fixed vaLUE for all vegetation types in the early stage. However, it became a parameter that varied with vegetation type since the MODIS-LUE model, and recently, it was considered to require adjusting in time according to the phenological and physiological status of vegetation. Although the current vegetation productivity estimation model based on seasonal dynamic LUEmax parameters showed a relatively higher accuracy, these studies were mostly limited to specific vegetation types or spatial scales. Thus, the applicability of different dynamic LUEmax parameters in a wider range of vegetation types or regions and the differences between geographical areas remain clear. In this paper, we presented a comparative analysis of three typical dynamic LUEmax parameter simulation methods (which are based on the chlorophyll index, leaf area index, and Markov chain Monte Carlo) by using the same dataset (FLUXNET 2015 dataset) and model structure (MODIS-LUE model structure). Results showed that the seasonal variation characteristics of three different dynamic LUEmax parameters differed significantly, generally showing three characteristics of single-peaked, U-shaped, and horizontal fluctuations in different vegetation types. The accuracy of the estimated gross primary productivity (GPP) based on dynamic LUEmax parameters was better than when the original static parameter was used, but it relied on the specific LUEmax parameter simulation method. The Markov chain Monte Carlo method had a good simulation effect on the LUEmax parameter, and its GPP estimation accuracy improved in all vegetation types at all time periods (compared to the original MODIS-LUE static LUEmax, ΔRMSE = 10.9 g/(m2·month), calculated in units of carbon), especially in closed shrub, deciduous needle-leaf forest, and evergreen broadleaf forest. These findings can provide a basis for the uncertainty analysis of light-use-efficiency-based vegetation productivity estimation and the development of new models.关键词：remote sensing;vegetation productivity;gross primary productivity;light use efficiency model;maximum light use efficiency;parameter evaluation- 1167
- |
- 282
- |
- 0
发布时间：2024-04-08 -
Transferring deep convolutional neural network models for generalization mapping of autumn crops AI导读
摘要：Deep Convolutional Neural Networks (DCNNs) have been increasingly applied in remote sensing crop recognition due to their “end-to-end” advantages and efficient extraction of shallow shape details and deep semantic features. However, deep learning models require a large number of labeled samples, which are time-consuming, labor-intensive, and costly to obtain, limiting the 2016—2020 period. The U-net model based on CDL training can be popularized and applied in the United States. First, the overall accuracy of time generalization in the three test areas in the United States from 2016 to 2020 is more than 80%, and the recognition accuracy of corn is higher than that of soybeans. Deep learning models have good transferability in space. Second, for autumn grain in Heihe City, the average recognition accuracy of corn for many years is 3% higher than that of soybean. This is because the corn planting plot is more regular and the planting scale is higher than that of soybean; the overall accuracy of autumn grain identification in a single year is between 69% and 79%. The year classification model is better than the single-year classification model, which may be because the representativeness of the training samples is enhanced with the increase of the number of labeled samples, and the difference in autumn grain planting between China and the United States can be compensated by the expansion of the number of samples. However, the model is migrated to the Heihe region of China. The accuracy of the models is lower than that of the continental United States, which is due to the inconsistency of remote sensing response characteristics due to differences in intercontinental climate and crop planting habits. These, in turn, reduce the generalization performance of the model. The DCNN model is better than random forest algorithm because of the training process driven by big data. The principle of transferring the basic trained crop classification model to map crop distribution timely and accurately has broad prospects for application across a large scale of time and space. The consistency of remote sensing features and phenology of the crops of the test area compared to the training data are fundamental factors that must be carefully considered, as these determine the success of high crop mapping performance. Therefore, it is essential to analyze the prerequisites when transferring the model to other places.关键词：remote sensing;transfer learning;CDL;time-space generalization;soybeans;maize- 108
- |
- 92
- |
- 0
发布时间：2024-04-08
Forestry and Agriculture
-
摘要：Satellite remote sensing retrieval is an important way to solve the problem of obtaining near-real-time high-resolution precipitation information. Fengyun-4A (FY-4A) is outfitted with the Advanced Geosynchronous Radiation Imager (AGRI), which boasts world-leading performance. The dual scanning mirrors of AGRI enable precise 2-D pointing, allowing for minute-level regional scans—a groundbreaking achievement. This advanced instrument can capture high-frequency images of the Earth’s cloud cover in more than 14 spectral bands. It can generate the official FY-4A REGC (Regional Precipitation Estimation Near-real-time Product for China), which is one of the precipitation estimates information that China can independently obtain from satellite remote retrieval. However, the accuracy of FY-4A REGC still lags behind that of IMERG-Early, the counterpart product of the Global Precipitation Measurement (GPM). Currently, the prevailing approach for correcting satellite-derived precipitation products involves constructing linear prior relationship models between historical satellite rainfall estimates and corresponding ground truth measurements, typically obtained from rain gauges or radar systems. When new observational data become available, this relationship is utilized to derive corrected precipitation values. However, linear models struggle to precisely capture the intricate relationship between satellite rainfall estimates and ground truth measurements. We have observed that ensemble learning methods offer nonlinear models that exhibit advantages such as faster training, reduced data requirements, and robust model stability. In this study, a correction method for official FY-4A precipitation estimates is dynamically constructed using an ensemble machine learning method (LightGBM) with FY-4A REGC as the model input and IMERG-Early as the training calibration for the mainland China region. The revised FY-4A precipitation product (FY-4A Adj) was compared with the original FY-4A REGC using the CMPA automatic gauge observations as the ground reference. The Correlation Coefficient (CC), Root Mean Square Error (RMSE), and relative bias (Bias) of FY-4A Adj were found to be improved significantly compared with those of FY-4A REGC. The revised algorithm effectively reduced the significant overestimation of the original FY-4A REGC in southern China. Our investigation revealed that choosing the correct order for training information significantly enhances model accuracy, with this study opting for training order 221. In practical applications, the ensemlde learning model can continually optimize its model parameters and performance by dynamically adjusting to the latest training data in real time. We also conducted a comparative analysis of two classes of methods employing ensemble learning, namely, bagging and boosting. Our findings indicate that the Random Forest method performs better when working with limited data volumes, while LightGBM is the recommended choice for large datasets. In conclusion, the correction method based on ensemble machine learning proposed in this paper can quickly and effectively improve the near-real-time Precipitation estimates of FY-4A REGC. This method provides guidance for producing high-quality satellite precipitation products based on FY-4A.关键词：remote sensing;Fengyun satellite;FY-4A;ensemble learning;near real-time precipitation estimates;precipitation estimates correction- 100
- |
- 64
- |
- 0
发布时间：2024-04-08 -
摘要：Dynamic changes of the coastline reflect the transgression-regression process of the sea–land interaction, which is of great significance to the environmental protection and development planning of coastal areas. On the basis of six periods of Landsat satellite images from 1973 to 2019, this paper obtained coastline data of each period and calculated the changes in their length and type by means of human-machine interactive interpretation.Detailed analysis in terms of Net Shoreline Movement (NSM), End Point Rate (EPR), and Linear Regression Rate (LRR) in the Digital Shoreline Analysis System (DSAS) were conducted to explore the potential driving factors regarding the dynamic changes of the coastline.Results show that from 1973 to 2019, the total length of the Huizhou coastline increased from 224,565 m to 249,656 m, with an average NSM of 185.46 m and an average EPR of 4.04 m/a. The coastline was mainly characterized by erosion from 1973 to 1984 and showed an overall expansion trend afterwards. The expansion amplitude of the coastline exceeded the highest during 1984—1993, with the NSM reaching 100.2 m. In addition, the artificial coastline, which is mainly distributed in Daya Bay Petrochemical Zone, northern Fanhe Bay reclamation and aquaculture Section, Huizhou Port, western Kaozhou Bay reclamation and aquaculture section, and the vicinity of Huangbu Town section, showed the most remarkable change (8.27%—57.45%) among the different coastline types. The main reasons of coastline change include the development of a coastal aquaculture; the construction of ports, industrial areas, and coastal tourist areas; and the expansion of construction land due to population growth and economic development. Some coastline segments, such as the core area of Xunliao Bay and the east and west flanks of Shuangyue Bay, remained stable during the whole period.We conclude that human impact was the main driving factor and that natural factors have little influence in the coastline change during past nearly 50 years. Compared with other coastal zones in mainland China, the rate of change in Huizhou is less remarkable. Judging by the EPR (2.51%) from 2013 to 2019, the change of coastline in Huizhou remained stable. Thus, we estimate that there will be no significant expansion toward the sea in the future.关键词：Huizhou City;coastline;remote sensing;DSAS;dynamic change- 238
- |
- 93
- |
- 0
发布时间：2024-04-08
Atmosphere and Ocean
-
摘要：Satellite laser altimetry technology, with its ability to acquire highly accurate vertical distribution information, possesses unique advantages in the field of Earth observation. Currently, different satellite laser altimetry systems are operating in orbit both domestically and internationally. In recent years, China has launched the GF-7 satellite and ZY3-03 satellite, both equipped with laser altimetry systems, which are primarily employed for acquiring global laser elevation control points. With the steady operation of these satellites in orbit and the continuous acquisition of laser altimetry data, China has, for the first time, formed related data products in the field of laser altimetry, called Satellite Laser Altimetry standard (SLA03) data products for natural resources. To better understand the accuracy level of the products and guide the subsequent application and optimization, a comprehensive accuracy evaluation must be conducted. Based on the high-accurate airborne LiDAR data gathered in plain areas and mountainous regions with forests, thousands of laser points are collected to comprehensively evaluate the terrain height accuracy of the SLA03 data products for natural resources in this paper. After coordinate transformation and data conversion, the SLA03 data and reference data are unified into the same coordinate framework. Then, taking the size of the laser ground spot into consideration, the reference terrain elevations are obtained based on the points classified as ground from the LiDAR data. Multiple accuracy metrics, including overall bias, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and 90th percentile Linear Error (LE90), are utilized for elevation accuracy assessment. Results show that, after eliminating the laser points located at tree canopies or buildings using a proper threshold, the height accuracy of the SLA03 data products from the GF-7 satellite is 0.653 m in RMSE and 1.055 m in LE90 in plain areas, with over 60% under 0.3 m, but decreases to 1.210 m (RMSE) and 2.002 m (LE90) in mountainous regions with forests. The accuracy of the SLA03 data products from the ZY3-03 satellite is 1.312 m (RMSE) and 2.389 m (LE90) in plain areas with more than 50% under 0.5 m, while in the mountainous regions with forests, it declines to 1.661 m (RMSE) and 2.999 m (LE90). The potential for elevation control points in the plains is above 60% for both the GF-7 and ZY3-03 satellites, but additional screening is required before use. The elevation accuracy of SLA03 products from the GF-7 satellite is obviously affected by seasonal factors, which are mainly caused by vegetation growth. Meanwhile, the elevation accuracy of SLA03 products from the ZY3-03 satellite is inferior to that of the GF-7 satellite, hindering the former satellite from effectively distinguishing the impact of seasonal changes in vegetation. The relevant conclusions will guide the effective application of SLA03 products and provide support for the design and parameter argumentation of subsequent satellite laser altimetry systems.关键词：remote sensing;satellite laser altimetry;the GF-7 Satellite;the ZY3-03 Satellite;Elevation Accuracy;the SLA03 Data Product- 136
- |
- 179
- |
- 0
发布时间：2024-04-08 -
摘要：Hyperspectral images (HSIs) have a wealth of continuous spectrum information, covering hundreds of bands from visible light to infrared wavelengths. The data characteristics of HSIs give it a unique advantage in harnessing the inherent attributes of the spectrum in image processing. This advantage is conducive to making full use of spatial and spectral information and detecting targets in the region of interest. However, due to the high dimensionality of hyperspectral data, the complexity of actual scenes, and the limited number of labeled samples, hyperspectral anomaly detection faces the problem of indistinct background and anomalies and low detection accuracy. Therefore, we propose a background memory model for hyperspectral anomaly detection. First, the pseudo background and anomaly vectors are obtained through an unsupervised rough inspection method based on density estimation. Second, we design a background memory generation adversarial network model based on anomalous prominent regular term constraints. Moreover, we expand the distance between the false background and false anomalies in a weak supervision-pseudo-label manner. Thus, the network has a strong background generation ability while the effect on anomaly reconstruction is weakened, which reduces the generalization of background and anomaly reconstruction and enhances the difference and discrimination between background and anomaly. We also perform adversarial learning in the feature domain and image domain to improve sample generation ability, enabling better learning of the distribution of input samples and strengthening the capability to generate background. Finally, a nonlinear background suppression method is introduced to reduce the false alarm rate and further improve the detection accuracy. The experimental results show that our model has a better detection effect on different datasets than other detection algorithms.HSIs have continuous spectral information of hundreds of bands, which make it possible to capture the deep and intrinsic characteristics in a spectrum. However, due to the high dimension of HSI, the complexity of the scene, and the limitation of labeled samples, hyperspectral anomaly detection remains a challenge. To solve the abovementioned problem, we propose a generative adversarial network with anomaly-highlighted regularization and train it in a weakly supervised manner. We aim to separate the anomaly and background vectors to make the difference more obvious and obtain a more accurate detection map.In this paper, we propose a background memory generative adversarial network for hyperspectral anomaly detection. First, we obtain the pseudo background and anomalies through unsupervised coarse detection based on density estimation as the input of the network. Next, to reduce anomaly contamination in background estimation, we impose the constraint of anomaly-highlighted regularization to expand the distance between the background and anomaly. In the weak supervised pseudo labeling training mode, the network can reconstruct background vectors well but gains poor performance for anomaly reconstruction. Besides, there are two discriminators in the latent and reconstruction domains, which aim to improve the ability of background generation and estimation. Finally, we perform nonlinear background suppression on the detection map as post-processing to reduce the false alarm rate.Compared with other new algorithms with good performance, the proposed method has better detection results in both quantitative and qualitative aspects of different datasets. The AUC score of (Pd, Pf) achieves the highest value across different datasets and outperforms other algorithms and has the advantage of an order of magnitude. For example, the AUC score of (Pd, Pf) achieves 0.99771 for the ABU-1 dataset, while the AUC score of (Pf, τ) is 0.00258, which outperforms the second-best algorithm AED with scores of 0.99760 and 0.02230, respectively. The visual results are consistent with the qualitative results as well. The ROC curve locates near the upper left corner. Under the same false alarm rate, the proposed method has the highest accuracy for most datasets, obtaining higher detection probability and lower false alarm rate, and has better detection performance. The box plot likewise reveals that the background and anomaly of this method are more separable.In this paper, we propose a generative adversarial network memorizing background for hyperspectral anomaly detection. Different from our previous work, we obtain the pseudo background and anomaly vector adaptively in an unsupervised manner to solve the problem of the small number of anomaly samples and lack of prior information. Based on weakly supervised pseudo-labeling learning, we aim to model a hyperspectral anomaly and background vector. As a result, the network can reconstruct background data well but performs poorly on anomaly vector reconstruction. We also apply the constraint of highlighting an anomaly regular term in the network to enhance the separability between background and anomaly. Finally, we perform post-processing of nonlinear background suppression to reduce the false alarm rate under the same detection accuracy. Experimental results show that the proposed method can achieve better detection performance than other algorithms on different datasets.关键词：remote sensing;GAN;hyperspectral image;anomaly detection;weakly supervised learning;unsupervised learning- 738
- |
- 136
- |
- 0
发布时间：2024-04-08 -
摘要：Synthetic Aperture Radar (SAR) is a kind of high-resolution imaging sensor, which is able to work under nearly all weather and illumination conditions. SAR plays an important role in earth observation. For single-band, single-polarization SAR image, however, there’s only one complex scalar in each pixel. So that the information contained in such single-channel SAR image could be quite limited, which limits its performance in various applications. Since terrain classification is one of the typical tasks in SAR image interpretation, this paper takes it as an example to demonstrate the problem and gives our solution. To address the above problem, this paper proposes a representation for the spatial information on SAR image—the Directional Context Covariance Matrix (DCCM). DCCM obtains the variance of pixel intensity in several orientations inside the neighborhood in order to make use of the context information. During such process, the target pixel in extended from a complex scalar to a group of matrices, so that its information content is increased. Besides, the matrix form also enables some of the advanced matrix algorithms to be applied to single-channel SAR image. On the basis of it, the DCCM texture feature is derived, which can better represent the texture properties on SAR image and shows better discriminability for different land covers. Then, the texture feature is combined with two traditional classifiers as well as the Convolutional Neural Network (CNN), respectively. Thereafter, a SAR image classification scheme is established. To illustrate the performance of proposed method, terrain classification experiments are carried out on AIRSAR and UAVSAR datasets. Methods based on three commonly used texture features, the gray level co-occurrence matrix (GLCM), Gabor filters and Multilevel Local Pattern Histogram (MLPH) are taken into comparison. On traditional classifiers, the overall classification accuracies are increased by 7% on both datasets. While combining with CNN, the overall accuracies and kappa coefficients are significantly improved with DCCM texture feature than the original SAR data. The proposed feature also shows nice efficiency and better robustness when compared to other texture features. The experiment results indicate that DCCM is an effective representation that is suitable for SAR image. DCCM is efficient, robust and easy-to-use. The proposed DCCM based classification method can improve the classification performance of single-channel SAR image by increasing the pixel information content. Beyond that, DCCM could be a promising method for many other SAR image interpretation tasks.关键词：SAR image;representation;directional context covariance matrix;texture;terrain classification- 105
- |
- 105
- |
- 0
发布时间：2024-04-08 -
摘要：The effects that stray light has on satellite-borne infrared optical systems are embodied in two perspectives: one is degrading the quality of imaging, and the other is decreasing the precision of calibration. Satellite-borne infrared optical systems are situated in an extraneous atmosphere environment, where stray light is simply formed at a specific operating position relative to celestials such as the sun and the moon. Owing to this, the imaging capability of the payload and the precision of calibration rely vitally on the evaluation of stray light. The Visible and InfraRed Radiometer (VIRR), which embarks on FY-3A/B/C satellites, has been operating in orbit for more than 10 years, during which the massive quantity of data acquired is significant for the study of climate change. The long period of the payload’s consistent operation covered various circumstances of illumination. By evaluating the variation of the payload’s response in a long time sequence, the changing trend of the payload’s on-orbit performance can be obtained, which is of great significance for the recalibration of historical data.The imaging optical system of the VIRR is modeled and simulated by means of TracePro, which yields the Point Source Transmittance (PST) curves of different spectral bands in pitch, scan, and yaw dimension, respectively. PST is one of the widely adopted indicators for the characterization of an optical system’s response to stray light. It is defined as the radiance of the detector aroused by an off-axis point source and then normalized to the entrance pupil radiance when the point source is on axis.The experimental stray light measurement of the backup of the on-orbit instrument is conducted as well. Based on the experimental results, a comparative analysis between simulated result and actual measurement is achieved qualitatively. Possible incident angles for which the instrument is susceptible to stray light are found by studying the PST curves. A shield is later mounted on the instrument to demonstrate the effectiveness of the stray light suppression method.To evaluate the on-orbit performance of VIRR under stray light influence, the simulation and analysis of orbital parameters, especially the solar irradiance, are performed as well. Taking into consideration the mutual effects of payloads, the illumination situation during satellite crossing terminator is simulated, and the path through which solar contamination influences the instrument both externally and internally is found.The PST curves of pitch, scan, and yaw dimension of VIRR reveal the geometric profile of the instrument. The peaks of pitch and scan dimension appear at an off-axial angle of 10°, which is consistent with the simulation results. Likewise, there is a peak at a large off-axial angle (~75°) in the pitch dimension, which corresponds to a stray light incident from the motor side. The yaw dimension PST is two to three magnitudes lower than that of the pitch and scan dimension within the range of ±20°. This is due to the blockage of the motor housing. Preliminary results of the magnitude of stray light are also obtained. The transmitting path of stray light is predicted by simulation results and later verified by on-orbit data.This work can provide reference for the design of similar payloads with regard to suppressing stray light and preventing sun contamination. In addition, the results yielded can fundamentally support the recalibration of historical remote sensing data.关键词：remote sensing;Visible and Infra-Red Radiometer (VIRR);stray light suppression;point source transmittance (PST);solar contamination;recalibration- 573
- |
- 100
- |
- 0
发布时间：2024-04-08 -
摘要：GF-7 was successfully launched in Taiyuan in November 2019 and is the first sub-meter high-resolution stereo mapping satellite of China.The first domestic star map fusion star sensor is used on the satellite, with dynamic correction of the angle between the two star sensors, which can greatly improve the accuracy of satellite attitude determination.Based on the overall design of the GF-7 satellite,this paper studies the key geometric processing techniques in the processing of the GF-7 satellite to achieve high-precision positioning of the High Resolution 7 dual line array camera, and verifies the satellite’s geometric level through extensive verifications. Based on attitude measurement data and orbit measurement data, the GF-7 satellite has greatly improved its attitude and orbit determination accuracy by using star map fusion star sensors. And the GF-7 satellite has added a unified time system and a unified attitude orbit model, constructing a rigorous geometric model with multiple platforms and cameras.To address the issue of changes in payload pointing angle during the imaging process of the GF-7 satellite,it is not only necessary to accurately calibrate the geometric distortion curve of the on-orbit camera, but also to obtain the precise light pointing of each CCD imaging detector in the payload to create a high-precision standard image,providing precise internal orientation geometric elements for the production of high-precision standard image products to ensure the internal geometric accuracy of satellite image products.The accuracy of geometric positioning and stereographic mapping products of GF-7 satellite is verified by high precision basic data and field ground measurement points.The high-precision basic data uses the DOM image with a geometric accuracy of better than 1:2000 and the DEM data with a high accuracy of grid and earth accuracy better than 5 m, and the field ground measurement points use 71 ground control points evenly distributed on the image.The verification results show that the uncontrolled positioning accuracy of the camera can be controlled to about 6 meters.After introducing a few ground control points to carry out rigid translation, the plane geometric residual is about 1.4 m, and the elevation residual is about 0.86 meters, which meets the demand of high-precision stereoscopic mapping in China.Based on the practical experience of satellite data processing system construction, this paper mainly combines the load design features of GF-7 dual-array camera,studies the key technologies and methods of GF-7 satellite combined processing, so as to improve the research system in the field of high-resolution optical satellite image geometric processing.Finally, a large number of data are used to verify the accuracy, and the results show that compared with the previous satellites, the geometric calibration accuracy, geometric positioning accuracy and relative geometric accuracy of the front and rear cameras of GF-7 dual-linear array camera have been significantly improved., reaching the current foreign commercial satellite plane elevation geometric accuracy standards. This indicates that the domestic satellite high -precision satellite load platform and ground processing capacity rose to a new step,which is at the international advanced level and has important practical significance for the subsequent development of related satellite work.关键词：GF-7 satellite;stereo mapping satellite;geometric check;Rigorous geometric model;attitude track data processing;star sensors;uncontrolled positioning accuracy;elevation residual;control points- 751
- |
- 168
- |
- 0
发布时间：2024-04-08 -
摘要：Disasters and accidents often occur in remote areas that are inaccessible by conventional transportation. Common map navigation software also fails to provide a passable path. Therefore, only Sport Utility Vehicles (SUVs) can be used in field rescues. However, ground vehicles have limited awareness of their surroundings, so we chose to set up an air-ground cooperative system to overcome these vehicles’ shortcoming. In the air-ground cooperative system, Unmanned Aerial Vehicles (UAVs) can provide environmental images surrounding the SUV for the ground terminal to obtain the navigation path of the vehicle by quickly extracting the type of surface and terrain undulations in the images. To better meet the above application requirement, this paper improves the existing A* algorithm with three main innovations. First, we proposed a passability cost function that integrates surface type and surface elevation information considering the application characteristics of the outdoor surface environment. Second, we proposed a fast path search algorithm based on grid cells given the scale relationship between the resolution of UAV images and the actual vehicle paths. Third, we chose the marginal feature points of the grid for the passability path search in view of the connected surface-type distribution inside the grid cell. This approach improves the search efficiency of the algorithm while taking into account the terrain information inside the grid cell, such that the algorithm could make full use of the detailed information of the UAV images and optimize the calculation. The 3D visualization experiments show that the paths searched by the Grid A* algorithm are more reliable and meet the needs of SUVs. The Grid A* algorithm proposed in this paper, aiming at the high-resolution images obtained by UAV, synthesizes image classification, slope calculation, and other methods to structure a passability cost function so that the paths are more reliable. In addition, the time cost of the algorithm is reduced to 15% of the traditional A* algorithm, which improves the timeliness of emergency rescue in the field.关键词：remote sensing;path planning;Grid A* algorithm;A* algorithm;air-ground coordination;passability cost function- 692
- |
- 161
- |
- 0
发布时间：2024-04-08 -
摘要：The time-series interferometric synthetic aperture radar (InSAR) technology based on distributed scatterers (DSs) makes up for the drawback of the persistent scatterer InSAR technology, which struggles to obtain a sufficient number of deformation monitoring points in low-coherence regions. Given that DS targets are easily affected by temporal and spatial decorrelation factors, leading to the introduction of speckle noise, the phase of DS targets must be optimized. In this paper, a new polarimetric phase optimization method based on eigenvalue decomposition and adaptive filtering (EVD-FPO) is proposed to address these limitations. This method uses Sentinel-1A dual-polarization data amplitude information to identify DS and PS targets and uses eigenvalue decomposition polarimetric phase optimization technology based on the temporal and spatial coherence matrix and adaptive mean filtering technology to improve the phase quality. To prove the feasibility and effectiveness of the proposed method, 17 Sentinel-1A dual-polarization (VV-VH) SAR images are used to evaluate its performance. Experimental results show that the proposed EVD-FPO method can effectively increase the density of coherent target points and improve the phase quality. Compared with the single polarimetric amplitude dispersion method (VV-- 2523
- |
- 170
- |
- 0
发布时间：2024-04-08 -
Cross-domain transfer learning algorithm for few-shot ship recognition in remote-sensing images AI导读
摘要：Cross-domain transfer learning aims to utilize public datasets as source data to improve the recognition accuracy of target data, breaking through the limitation that the category space between source data and target data must be consistent.For the few-shot remote-sensing ship recognition task, existing cross-domain transfer learning algorithms have the disadvantages of transfer category restriction and negative transfer effect. Therefore, a cross-domain transfer learning algorithm based on source data correlation sorting was proposed to solve the above problems.First, the target data were added reversely into the source domain recognition task. According to the variation of the source data recognition accuracy before and after the target data were added, various source data were classified into strong/weak/negative correlation samples, and only the strong correlation samples would be selected. Then, the self-supervised joint learning strategy was adopted to introduce the auxiliary self-supervised angle prediction branch into the classification network in the target domain. The selected strong correlation source samples were added but only into the training of the self-supervised branch, which avoided changing the main classification network structure.Randomly selecting 65 category samples as the source data from miniImageNet and conducting comparative experiments on few-shot ship targets in remote-sensing images yields the following results: 1) When Resnet18 is chosen as the classification network, the performance of the proposed algorithm is better than that of the Fine-tune algorithm, which is widely used in cross-domain transfer learning. Moreover, compared with the recognition algorithm, which only uses the main classification network, the proposed algorithm improves the recognition accuracy of target data from 78.89% to 96.48%. 2) Using different networks to sort correlations for the source data, the selected strong correlation source samples are not exactly the same and their degree of category coincidence is close to 60%. However, they are all helpful to the classification task of the target domain. At the end of this paper, through visualizing the extracted target features, it is verified that the target features extracted by using the proposed algorithm are more abundant and have higher generalization ability.The proposed algorithm has two main advantages. First, the weak/negative correlation source samples are eliminated by correlation sorting, which can avoid the occurrence of negative transfer effect. Second, by introducing the self-supervised angle prediction branch, the information of the strong correlation source samples is fully utilized and the features with more generalization ability are extracted while maintaining the structural integrity of the main classification network.关键词：remote sensing;Ship recognition;few-shot learning;Cross-domain transfer learning;Correlation sorting;self-supervised learning- 883
- |
- 267
- |
- 0
发布时间：2024-04-08
Models and Methods
- 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