Search

Unsupervised band selection (BS) is a crucial research direction in the domain of hyperspectral image (HSI) processing. In recent years, volume-based criteria have garnered considerable attention, with the volume-gradient-based BS (VGBS) algorithm being particularly notable. However, we have identified that VGBS inherently suffers from the local extremum problem due to its reliance on the original volume gradient formula, which only permits the removal of a single band per iteration. To address this issue, we introduce the concept of joint volume gradient (JVG) through a novel determinant formula for the high-order mixed product expansion. We then propose an enhanced version of VGBS, termed JVG-based BS (JVGBS), which allows for the simultaneous deletion of multiple bands. Moreover, a simplified objective function of JVG is developed to mitigate the high computational complexity associated with calculating volume metrics when a small number of bands is removed at once. Regarding the complexity imposed by the large cardinality of traversing matrix column combinations, we provide an exemplary algorithm employing groupwise strategies to achieve rapid computational acceleration. Experimental results on Gaofen-5 and publicly available hyperspectral datasets demonstrate that the proposed algorithms have rather superior performance against state-of-the-art competitors in terms of both computational complexity and classification accuracy.

Accurate extraction of urban impervious surface (UIS) is essential for urban planning and environmental monitoring. However, multispectral remote sensing data for UIS extraction suffers from the inter-class spectral confusions, e.g. UIS and bare soil, and intra-class variations of sub-class UIS. Hyperspectral and full/dual-polarization synthetic aperture radar (full/dual PolSAR) data provide opportunities for reducing such confusions and have potential for fine UIS mapping, i.e., roads, buildings, and grounds. In this study, we first investigated the hyperspectral data (Gaofen-5) capability to reduce the intra/inter-class misclassification in comparison with multispectral data (Landsat-8). Then, we explored contributions of synergistically using full and dual PolSAR (ALOS-2 and Sentinel-1) with hyperspectral and multispectral data using optical-SAR sparse representation classification (OSSRC). Results showed that both the hyperspectral and the SAR polarization features helped better delineation between UIS and bare soil, and sub-class UIS (roads and buildings). The relative contribution of PolSAR was higher in multispectral data than in hyperspectral data, with full PolSAR contributed significantly. The combined hyperspectral and full PolSAR data using OSSRC delivered the best result, with an overall accuracy higher than 90%. The results indicate the promising capability of synergizing hyperspectral and full/dual PolSAR data for improving UIS extraction from advanced satellite data.

Accurate extraction of urban impervious surface (UIS) is essential for urban planning and environmental monitoring. However, multispectral remote sensing data for UIS extraction suffers from the inter-class spectral confusions, e.g. UIS and bare soil, and intra-class variations of sub-class UIS. Hyperspectral and full/dual-polarization synthetic aperture radar (full/dual PolSAR) data provide opportunities for reducing such confusions and have potential for fine UIS mapping, i.e., roads, buildings, and grounds. In this study, we first investigated the hyperspectral data (Gaofen-5) capability to reduce the intra/inter-class misclassification in comparison with multispectral data (Landsat-8). Then, we explored contributions of synergistically using full and dual PolSAR (ALOS-2 and Sentinel-1) with hyperspectral and multispectral data using optical-SAR sparse representation classification (OSSRC). Results showed that both the hyperspectral and the SAR polarization features helped better delineation between UIS and bare soil, and sub-class UIS (roads and buildings). The relative contribution of PolSAR was higher in multispectral data than in hyperspectral data, with full PolSAR contributed significantly. The combined hyperspectral and full PolSAR data using OSSRC delivered the best result, with an overall accuracy higher than 90%. The results indicate the promising capability of synergizing hyperspectral and full/dual PolSAR data for improving UIS extraction from advanced satellite data. [ABSTRACT FROM AUTHOR]

Landscape mapping and pattern analysis of fine-grained urban functional zones (UFZ) are of great research value in the field of urban planning and urban development assessment. Previous studies employing multispectral images have largely enabled only coarse-grained land-use classification of urban areas. By contrast, hyperspectral images can provide very detailed spectral features and have shown some potential to facilitate fine-grained classification of urban areas. In this article, we evaluate and analyze the classification of fine-grained UFZ in the central city of Wuhan, Hubei, China, using GaoFen-5 (GF-5) hyperspectral satellite imagery. We first compare the performance of hyperspectral data (GF-5) and classical multispectral data (Landsat 8) for the classification of fine-grained functional zones of cities by employing two classical classification algorithms and two deep learning methods. We also propose the deep learning-based spectral-spatial unified networks combined with a fully connected conditional random field (SSUN-CRF) algorithm for fine-grained UFZ mapping to enable better landscape pattern analysis. We then analyze the landscape pattern of the main urban areas of Wuhan by combining ten landscape indicators based on the precise UFZ classification results. Experimental results illustrate the followings. First, compared to multispectral images, hyperspectral images can allow for a more accurate UFZ classification. Second, deep learning classification algorithms can better exploit hyperspectral image data, with the SSUN-CRF algorithm, in particular, being able to achieve an overall accuracy of 93.86% and a Kappa coefficient of 92.08%. Finally, the landscape pattern analysis demonstrates that hyperspectral remote sensing imagery shows significant potential in mapping fine-grained UFZ. It is beneficial for further urban study and planning.

Toxic metals have attracted great concern worldwide due to their toxicity and slow decomposition. Although metal concentrations can be accurately obtained with chemical methods, it is difficult to map metal distributions on a large scale due to their inherently low efficiency and high cost. Moreover, chemical analysis methods easily lead to secondary contamination. To address these issues, 110 topsoil samples were collected using a soil sampler, and positions for each sample were surveyed using a global navigation satellite system (GNSS) receiver from a coal mine in northern China. Then, the metal contents were surveyed in a laboratory via a portable X-ray fluorescence spectroscopy (XRF) device, and GaoFen-5 (GF-5) satellite hyperspectral images were used to retrieve the spectra of the soil samples. Furthermore, a Savitzky–Golay (SG) filter and continuous wavelet transform (CWT) were selected to smooth and enhance the soil reflectance. Competitive adaptive reweighted sampling (CARS) and Boruta algorithms were utilized to identify the feature bands. The optimum two-stage method, consisting of the random forest (RF) and ordinary kriging (OK) methods, was used to infer the metal concentrations. The following outcomes were achieved. Firstly, both zinc (Zn) (68.07 mg/kg) and nickel (Ni) (26.61 mg/kg) surpassed the regional background value (Zn: 48.60 mg/kg, Ni: 19.5 mg/kg). Secondly, the optimum model of RF, combined with the OK (RFOK) method, with a relatively higher coefficient of determination (R2) (R2 = 0.60 for Zn, R2 = 0.30 for Ni), a lower root-mean-square error (RMSE) (RMSE = 12.45 mg/kg for Zn, RMSE = 3.97 mg/kg for Ni), and a lower mean absolute error (MAE) (MAE = 9.47 mg/kg for Zn, MAE = 3.31mg/kg for Ni), outperformed the other four models, including the RF, OK, inverse distance weighted (IDW) method, and the optimum model of RF combined with IDW (RFIDW) method in estimating soil Zn and Ni contents, respectively. Thirdly, the distribution of soil Zn and Ni concentrations obtained from the best-predicted method and the GF-5 satellite hyperspectral images was in line with the actual conditions. This scheme proves that satellite hyperspectral images can be used to directly estimate metal distributions, and the present study provides a scientific base for mapping heavy metal spatial distribution on a relatively large scale. [ABSTRACT FROM AUTHOR]

The DPC is a multiangle sensor that detects atmospheric parameters. However, the retrieval of high-precision and high-spatial-resolution aerosol products from the DPC remains a great challenge due to the ill-posed nature of the problem. Thus, a novel aerosol optical depth (AOD) retrieval algorithm was proposed using visible surface reflectance relationships (VISRRs). The VISRR algorithm accounts for the surface anisotropy and needs neither a shortwave infrared band nor a surface reflectance database that can retrieve AOD over dark and bright land cover. Firstly, moderate-resolution imaging spectroradiometer (MODIS) surface reflectance (MYD09) products were used to derive the preceding surface reflectance relationships (SRRs), which are related to surface types, scattering angle, and normalized difference vegetation index (NDVI). Furthermore, to solve the problem of the NDVI being susceptible to the atmosphere, an innovative method based on an iterative atmospheric correction was proposed to provide a realistic NDVI. The VISRR algorithm was then applied to the thirteen months of DPC multiangle data over the China region. AOD product comparison between the DPC and MODIS showed that they had similar spatial distribution, but the DPC had both high spatial resolution and coverage. The validation between the ground-based sites and the retrieval results showed that the DPC AOD performed best, with a Pearson correlation coefficient (R) of 0.88, a root mean square error (RMSE) of 0.17, and a good fraction (Gfrac) of 62.7%. Then, the uncertainties regarding the AOD products were discussed for future improvements. Our results revealed that the VISRR algorithm is an effective method for retrieving reliable, simultaneously high-spatial-resolution and full-surface-coverage AOD data with good accuracy. [ABSTRACT FROM AUTHOR]

The hyperspectral, multispectral, and synthetic aperture radar (SAR) remote sensing images provide complementary advantages in high spectral resolution, high spatial resolution, and geometric and polarimetric properties, generally. How to effectively integrate cross-modal information to obtain a high spatial resolution hyperspectral image with the characteristics of the SAR is promising. However, due to divergent imaging mechanisms of modalities, existing SAR and optical image fusion techniques generally remain limited due to the spectral or spatial distortions, especially for complex surface features such as coastal wetlands. This paper provides, for the first time, an efficient multi-resolution collaborative fusion method for multispectral, hyperspectral, and SAR images. We improve generic multi-resolution analysis with spectral-spatial weighted modulation and spectral compensation to achieve minimal spectral loss. The backscattering gradients of SAR are guided to fuse, which is calculated from saliency gradients with edge preserving. The experiments were performed on ZiYuan-1 02D (ZY-1 02D) and GaoFen-5B (AHSI) hyperspectral, Sentinel-2 and GaoFen-5B (VIMI) multispectral, and Sentinel-1 SAR images in the challenging coastal wetlands. Specifically, the fusion results were comprehensively tested and verified on the qualitative, quantitative, and classification metrics. The experimental results show the competitive performance of the proposed method. [ABSTRACT FROM AUTHOR]

Land cover classification (LCC) of heterogeneous mining areas is important for understanding the influence of mining activities on regional geo-environments. Hyperspectral remote sensing images (HSI) provide spectral information and influence LCC. Convolutional neural networks (CNNs) improve the performance of hyperspectral image classification with their powerful feature learning ability. However, if pixel-wise spectra are used as inputs to CNNs, they are ineffective in solving spatial relationships. To address the issue of insufficient spatial information in CNNs, capsule networks adopt a vector to represent position transformation information. Herein, we combine a clustering-based band selection method and residual and capsule networks to create a deep model named ResCapsNet. We tested the robustness of ResCapsNet using Gaofen-5 Imagery. The images covered two heterogeneous study areas in Wuhan City and Xinjiang Province, with spatially weakly dependent and spatially basically independent datasets, respectively. Compared with other methods, the model achieved the best performances, with averaged overall accuracies of 98.45 and 82.80% for Wuhan study area, and 92.82 and 70.88% for Xinjiang study area. Four transfer learning methods were investigated for cross-training and prediction of those two areas and achieved good results. In summary, the proposed model can effectively improve the classification accuracy of HSI in heterogeneous environments. [ABSTRACT FROM AUTHOR]

Soil salinization has hampered the achievement of sustainable development goals (SDGs) in many countries worldwide. Several countries have recently launched hyperspectral remote sensing satellites, opening new avenues for accurate soil-salinity monitoring. Among them, Gaofen-5 (GF-5) from China has a high comprehensive performance, including a spectral resolution of 5 nm, 330 bands, and signal-to-noise ratio of 700. However, the potential of GF-5 for estimating soil salinity is not well understood. In this study, we proposed a strategy that includes bootstrap methods, fractional order derivative (FOD) techniques and decision-level fusion models to exploit the soil salinity diagnostic information and reduce estimation uncertainty in the Ebinur Lake oasis in northwestern China. The results showed that the GF-5 data were suitable for assessing soil salinity. The FOD technique enhanced the correlation between soil salinity and spectra, identified more diagnostic bands, improved the accuracy of soil salinity estimation, and reduced model uncertainty. The low-order FOD outperformed the high-order FOD. The spectra processed by the 0.9 order derivative were the most correlated with soil salinity (r = −0.76). The model driven by the 0.8 order derivative produced the optimal estimated model (R2 = 0.95, root mean square error (RMSE) = 3.20 dS m−1 and a ratio of performance to interquartile distance (RPIQ) = 5.96). The model driven by the 0.8 order derivative had less uncertainty than the models based on the original and integer-order derivative (first- and second- derivatives) spectra. This study provides a reference for estimating soil salinity from GF-5 data using the proposed framework with low uncertainty and high accuracy. GF-5 data have great potential for assessing environmental problems and facilitating further SDGs.

Aerosol spatial distribution obtained from the satellite sensor is a critical point to understand regional aerosol environment, anthropogenic aerosol emissions, and global climate change. In this study, the performance of aerosol optical depth (AOD) retrieval from the Directional Polarimetric Camera (DPC)/GaoFen-5 by using the Generalized Retrieval of Atmosphere and Surface Properties (GRASP) algorithm was evaluated on a global basis for the first time. The results showed that the DPC GRASP/Model scheme, which used several aerosol-type mixings, achieved good performance. By compared with AERONET observations, the correlation coefficient (R), normalized mean square error, and Expect Error (EE%) were 0.8982, 0.1008, and 83.16 %, respectively. The scattering angle, number of averaged pixels in retrieval units, and radiative and polarized fitting residuals showed impacts on the results of AOD retrieval in the DPC GRASP/Model. From the most of AERONET sites, the R and EE% were larger than ~0.9 and ~80 %. Compared with MODIS products, the spatial and temporal variations of AOD could be caught by the DPC observations with the GRASP/Model, and compared with the MODIS Dark Target algorithm, the DPC GRASP/Model AOD also showed a good performance. The above findings validated the ability of DPC sensor to monitor aerosols. It would contribute to the development of aerosol parameter retrieval from multi-angular polarized sensors in the future. [ABSTRACT FROM AUTHOR]

Surface temperature (ST) plays a great role in urban heat island effect, environment monitoring, earth resources monitoring, and water balance at local and global scales. The Chinese Gaofen-5 (GF5) satellite can capture Earth's thermal infrared information for use in national high-resolution Earth observations. In this article, the nonlinear split-window algorithm and the refined generalized split-window method are used to retrieve sea surface temperature (SST) and land surface temperature (LST) from GF5 observations. For different atmospheric and surface conditions, the algorithm coefficients are calculated using a statistical regression method from the numerical values simulated with the atmospheric radiative transfer model MODTRAN 5.0. The simulation results show that the root mean square error (RMSE) for SST and LST retrieval ranges from 0.09 K to 0.46 K and 0.19 K to 0.69 K, respectively, with increasing water vapor content (WVC). To validate the retrieved STs, Moderate Resolution Imaging Spectroradiometer (MODIS) STs extracted from the MYD11A1 product are used. Note that the RMSEs of both the LST and SST are less than 3.3 K. The RMSE for SST retrieval varies from 1.2 K to 1.45 K, with a mean value of 1.33 K; the RMSE for LST retrieval ranges from 1.57 K to 3.3 K, with a mean value of 2.41 K.

Toxic metals have attracted great concern worldwide due to their toxicity and slow decomposition. Although metal concentrations can be accurately obtained with chemical methods, it is difficult to map metal distributions on a large scale due to their inherently low efficiency and high cost. Moreover, chemical analysis methods easily lead to secondary contamination. To address these issues, 110 topsoil samples were collected using a soil sampler, and positions for each sample were surveyed using a global navigation satellite system (GNSS) receiver from a coal mine in northern China. Then, the metal contents were surveyed in a laboratory via a portable X-ray fluorescence spectroscopy (XRF) device, and GaoFen-5 (GF-5) satellite hyperspectral images were used to retrieve the spectra of the soil samples. Furthermore, a Savitzky–Golay (SG) filter and continuous wavelet transform (CWT) were selected to smooth and enhance the soil reflectance. Competitive adaptive reweighted sampling (CARS) and Boruta algorithms were utilized to identify the feature bands. The optimum two-stage method, consisting of the random forest (RF) and ordinary kriging (OK) methods, was used to infer the metal concentrations. The following outcomes were achieved. Firstly, both zinc (Zn) (68.07 mg/kg) and nickel (Ni) (26.61 mg/kg) surpassed the regional background value (Zn: 48.60 mg/kg, Ni: 19.5 mg/kg). Secondly, the optimum model of RF, combined with the OK (RFOK) method, with a relatively higher coefficient of determination (R2) (R2 = 0.60 for Zn, R2 = 0.30 for Ni), a lower root-mean-square error (RMSE) (RMSE = 12.45 mg/kg for Zn, RMSE = 3.97 mg/kg for Ni), and a lower mean absolute error (MAE) (MAE = 9.47 mg/kg for Zn, MAE = 3.31mg/kg for Ni), outperformed the other four models, including the RF, OK, inverse distance weighted (IDW) method, and the optimum model of RF combined with IDW (RFIDW) method in estimating soil Zn and Ni contents, respectively. Thirdly, the distribution of soil Zn and Ni concentrations obtained from the best-predicted method and the GF-5 satellite hyperspectral images was in line with the actual conditions. This scheme proves that satellite hyperspectral images can be used to directly estimate metal distributions, and the present study provides a scientific base for mapping heavy metal spatial distribution on a relatively large scale.

The DPC is a multiangle sensor that detects atmospheric parameters. However, the retrieval of high-precision and high-spatial-resolution aerosol products from the DPC remains a great challenge due to the ill-posed nature of the problem. Thus, a novel aerosol optical depth (AOD) retrieval algorithm was proposed using visible surface reflectance relationships (VISRRs). The VISRR algorithm accounts for the surface anisotropy and needs neither a shortwave infrared band nor a surface reflectance database that can retrieve AOD over dark and bright land cover. Firstly, moderate-resolution imaging spectroradiometer (MODIS) surface reflectance (MYD09) products were used to derive the preceding surface reflectance relationships (SRRs), which are related to surface types, scattering angle, and normalized difference vegetation index (NDVI). Furthermore, to solve the problem of the NDVI being susceptible to the atmosphere, an innovative method based on an iterative atmospheric correction was proposed to provide a realistic NDVI. The VISRR algorithm was then applied to the thirteen months of DPC multiangle data over the China region. AOD product comparison between the DPC and MODIS showed that they had similar spatial distribution, but the DPC had both high spatial resolution and coverage. The validation between the ground-based sites and the retrieval results showed that the DPC AOD performed best, with a Pearson correlation coefficient (R) of 0.88, a root mean square error (RMSE) of 0.17, and a good fraction (Gfrac) of 62.7%. Then, the uncertainties regarding the AOD products were discussed for future improvements. Our results revealed that the VISRR algorithm is an effective method for retrieving reliable, simultaneously high-spatial-resolution and full-surface-coverage AOD data with good accuracy.

The hyperspectral, multispectral, and synthetic aperture radar (SAR) remote sensing images provide complementary advantages in high spectral resolution, high spatial resolution, and geometric and polarimetric properties, generally. How to effectively integrate cross-modal information to obtain a high spatial resolution hyperspectral image with the characteristics of the SAR is promising. However, due to divergent imaging mechanisms of modalities, existing SAR and optical image fusion techniques generally remain limited due to the spectral or spatial distortions, especially for complex surface features such as coastal wetlands. This paper provides, for the first time, an efficient multi-resolution collaborative fusion method for multispectral, hyperspectral, and SAR images. We improve generic multi-resolution analysis with spectral-spatial weighted modulation and spectral compensation to achieve minimal spectral loss. The backscattering gradients of SAR are guided to fuse, which is calculated from saliency gradients with edge preserving. The experiments were performed on ZiYuan-1 02D (ZY-1 02D) and GaoFen-5B (AHSI) hyperspectral, Sentinel-2 and GaoFen-5B (VIMI) multispectral, and Sentinel-1 SAR images in the challenging coastal wetlands. Specifically, the fusion results were comprehensively tested and verified on the qualitative, quantitative, and classification metrics. The experimental results show the competitive performance of the proposed method.

Land cover classification (LCC) of heterogeneous mining areas is important for understanding the influence of mining activities on regional geo-environments. Hyperspectral remote sensing images (HSI) provide spectral information and influence LCC. Convolutional neural networks (CNNs) improve the performance of hyperspectral image classification with their powerful feature learning ability. However, if pixel-wise spectra are used as inputs to CNNs, they are ineffective in solving spatial relationships. To address the issue of insufficient spatial information in CNNs, capsule networks adopt a vector to represent position transformation information. Herein, we combine a clustering-based band selection method and residual and capsule networks to create a deep model named ResCapsNet. We tested the robustness of ResCapsNet using Gaofen-5 Imagery. The images covered two heterogeneous study areas in Wuhan City and Xinjiang Province, with spatially weakly dependent and spatially basically independent datasets, respectively. Compared with other methods, the model achieved the best performances, with averaged overall accuracies of 98.45 and 82.80% for Wuhan study area, and 92.82 and 70.88% for Xinjiang study area. Four transfer learning methods were investigated for cross-training and prediction of those two areas and achieved good results. In summary, the proposed model can effectively improve the classification accuracy of HSI in heterogeneous environments.

Environmental trace gas monitoring instrument (EMI) onboard GaoFen-5 was launched in May 2018 and has successfully operated on-orbit for more than a year. EMI contains four grating spectrometers, covering wavelengths in the range 240–710 nm with a spectral resolution of 0.3–0.5 nm, and enables one-day global coverage. For EMI on-orbit calibration, two onboard solar diffusers (SD), one surface reflectance aluminum diffuser, and one quartz volume diffuser (QVD) are used to measure solar spectra. The solar spectra are used to perform accurate spectral and radiometric calibrations. EMI on-orbit spectral calibration contains wavelength calibration and instrument spectral response function (ISRF), both of which are the key quantities in trace gas retrievals based on differential optical absorption spectroscopy (DOAS) analysis. The wavelength calibration is performed using the Fraunhofer lines in the solar spectrum, and the ISRF parameters are obtained by fitting high-resolution and EMI-measured solar spectra. Based on the known solar irradiation and characteristic of SD, SD radiance can be calculated and is used for EMI on-orbit radiometric calibration. The radiometric calibration also determines the absolute Earth reflectance spectra that are used as inputs for atmospheric retrieval algorithms. For EMI on-orbit radiometric monitoring, aluminum diffuser is used as reference SD to monitor QVD degradation. An internal white light source is used to detect pixel performance and monitor radiometric throughput. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Central China Normal University is the property of Huazhong Normal University and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The environmental trace gas monitoring instrument (EMI) is a space-borne imaging spectrometer onboard GaoFen-5, which was launched in May 2018, covering wavelengths in the range of 240–710 nm to measure NO2, O3, HCHO, and SO2. An advanced EMI-2 instrument with a higher spatial resolution and sufficient signal-to-noise is currently planned for launch on the GaoFen-5(02) satellite in 2021. The EMI-2 instrument bidirectional scattering distribution function (BSDF) is obtained from the absolute irradiance and radiance calibration on-ground. Based on EMI-2 earth and sun optical paths, the key factors of BSDF parameters are introduced. An NIST-calibrated 1000 W FEL quartz tungsten halogen lamp and a 2D turntable are adopted for the absolute irradiance calibration. A large aperture integrating sphere system is used for the absolute radiance calibration. Based on absolute irradiance and radiance calibration functions, the BSDF parameters are obtained, with accuracy of 4.9% for UV1, 4.3% for UV2, 4.1% for VIS1, and 4.2% for VIS2. The on-ground measurement results show that the reflectance spectrum can be calculated from BSDF parameters. On-orbit application of the EMI-2 instrument BSDF are also discussed.

Recent developments in hyperspectral satellites have dramatically promoted the wide application of large-scale quantitative remote sensing. As an essential part of preprocessing, cloud detection is of great significance for subsequent quantitative analysis. For Gaofen-5 (GF-5) data producers, the daily cloud detection of hundreds of scenes is a challenging task. Traditional cloud detection methods cannot meet the strict demands of large-scale data production, especially for GF-5 satellites, which have massive data volumes. Deep learning technology, however, is able to perform cloud detection efficiently for massive repositories of satellite data and can even dramatically speed up processing by utilizing thumbnails. Inspired by the outstanding learning capability of convolutional neural networks (CNNs) for feature extraction, we propose a new dual-branch CNN architecture for cloud segmentation for GF-5 preview RGB images, termed a multiscale fusion gated network (MFGNet), which introduces pyramid pooling attention and spatial attention to extract both shallow and deep information. In addition, a new gated multilevel feature fusion module is also employed to fuse features at different depths and scales to generate pixelwise cloud segmentation results. The proposed model is extensively trained on hundreds of globally distributed GF-5 satellite images and compared with current mainstream CNN-based detection networks. The experimental results indicate that our proposed method has a higher F1 score (0.94) and fewer parameters (7.83 M) than the compared methods.

Gaofen-5 carries six instruments aimed at various land and atmosphere applications, and it's an important unit of China High-resolution Earth Observation System. As Gaofen-5's thermal infrared payload is similar to that of ASTER, which is widely used in mineral exploration, application of Gaofen-5's thermal infrared data is discussed regarding its capability in mineral classification and silica content estimation. First, spectra of silicate, carbonate, sulfate minerals from a spectral library are used to conduct spectral feature analysis on Gaofen-5's thermal infrared emissivities. Spectral indices of band emissivities are proposed, and by setting thresholds of these spectral indices, it can classify three types of minerals mentioned above. This classification method is tested on a simulated Gaofen-5 emissivity image. With samples acquired from the study area, this method is proven to be feasible. Second, with band emissivities of silicate and their silica content from the same spectral library, correlation models have been tried to be built for silica content inversion. However, the highest correlation coefficient is merely 0.592, which is much lower than that of correlation model built on ASTER thermal infrared emissivity. It can be concluded that GF-5's thermal infrared data can be utilized in mineral classification but not in silica content inversion. [ABSTRACT FROM AUTHOR]

Land surface temperature (LST) is a key variable in the study of the energy exchange between the land surface and the atmosphere. Among the different methods proposed to estimate LST, the quadratic split-window (SW) method has achieved considerable popularity. This method works well when the emissivities are high in both channels. Unfortunately, it performs poorly for low land surface emissivities (LSEs). To solve this problem, assuming that the LSE is known, the constant in the quadratic SW method was calculated by maintaining the other coefficients the same as those obtained for the black body condition. This procedure permits transfer of the emissivity effect to the constant. The result demonstrated that the constant was influenced by both atmospheric water vapour content (W) and atmospheric temperature (T0) in the bottom layer. To parameterize the constant, an exponential approximation between Wand T0 was used. A LST retrieval algorithm was proposed. The error for the proposed algorithm was RMSE = 0.70 K. Sensitivity analysis results showed that under the consideration of NEDT = 0.2 K, 20% uncertainty in W and 1% uncertainties in the channel mean emissivity and the channel emissivity difference, the RMSE was 1.29 K. Compared with AST 08 product, the proposed algorithm underestimated LST by about 0.8 K for both study areas when ASTER L1B data was used as a proxy of Gaofen-5 (GF-5) satellite data. The GF-5 satellite is scheduled to be launched in 2017. [ABSTRACT FROM AUTHOR]

[Display omitted] • The Gaofen-5 satellite hyperspectral data were used for alteration mineral mapping. • Advanced argillic, argillic, phyllic, and propylitic zones were mapped in detail. • High content of Al-rich white mica (alunite + pyrophyllite + kaolinite ± dickite) is an important indication of hydrothermal alteration. Remote sensing technology is an exploration method that can directly map distribution of alteration minerals or zones associated with different types of mineral systems. Hyperspectral remote sensing can identify key pathfinder minerals and compositional variations within minerals. The Duolong porphyry-epithermal ore district, which has proven reserves of approximately 20 million tons of Cu and 420 tons of Au, is part of an emerging world-class metallogenic area in the western part of the Bangong Co-Nujiang metallogenic belt, Tibet, China. The Gaofen-5 hyperspectral (GF-HS) reflectance data were used to map the distribution patterns of hydrothermal alteration and to locate areas potentially containing mineral deposits in the Duolong ore district (DOD). Based on the mixture tuned matched filtering (MTMF) technique and the absorption feature wavelength positions of the minerals, an alteration map was generated using GF-HS data for mineral mapping in the DOD with sparse vegetation. The widely distributed advanced argillic alteration (alunite + pyrophyllite + kaolinite ± dickite), phyllic alteration (high content of Al-rich white mica), propylitic alteration (Mg-rich chlorite), and limited argillic alteration (kaolinite ± dickite ± white mica) were mapped in detail. The distribution patterns of alteration in known deposits were fairly consistent with the mineralogical information obtained in previous studies. In addition, we identified potential mineralization areas with characteristics similar to the Duobuza, Bolong and Naruo porphyry deposits. [ABSTRACT FROM AUTHOR]

The growing production and use of plastics are becoming a serious progressive issue and people pay increasing attention to the effects of plastics on ecosystems and human health. The availability of hyperspectral data from space sensors inspired us to study the feasibility to detect and identify different types of plastics in aircraft -, Goafen-5 (GF-5) and PRISMA satellite data by means of deep -, and machine learning models trained with spectral signatures. In this context, various inhouse and public spectral libraries are used to create a comprehensive database with mixed pixels of different plastic and non-plastic materials. The endmembers of plastic types involved in this study are polyethylene (PE), polypropylene (PP), polyvinyl chloride (PVC), polyethylene terephthalate (PET) and polystyrene (PS), covering 95% of the global production. Additionally, some important varieties of industrial plastics types such as acrylonitrile butadiene styrene (ABS), ethylene vinyl acetate (EVA), polyamide (PA), polycarbonate (PC), and polymethyl methacrylate (PMMA) were included in the investigations. Different samples with varying optical properties (color, brightness, transmissivity) have been selected for each plastic type. As non-plastic materials we have chosen spectra of vegetation, rocks, soils and minerals contained in the public US libraries (ECOSTRESS and USGS). The number of spectra for the training of the deep learning and machine learning models was enlarged by a random linear mixing method and the resulting database was separated into a training and a test group for subsequent multi-label classification. Algorithms selected are a convolutional neural network (CNN), random forest (RF) and support vector machine (SVM). To investigate the transferability to any hyperspectral image data obtained by air-, and spacecraft sensors, we opted for a unification of the spectral response functions (SRF) and the spectral sampling intervals of all data. Validation is accomplished based on the test group of the spectral database, and tested by controlled laboratory and aircraft experiments recorded over surfaces with varying background materials. Results are further analyzed for the influence of different noise quantities and abundance levels. The performance of the three models is roughly balanced for the validation of the spectral data with an overall accuracy of 97%, 96%, and 95% for the CNN, RF, and SVM, models respectively. In the controlled lab experiments, various accuracy indicators, such as the recall rates and the comprehensive metrics F1-score, OA, and Kappa suggest the RF classifier as the most robust one, followed by the SVM and CNN models. As for the evaluation of the aircraft data from controlled experiments, the RF further outperforms the other two models, behaving most robustly and reliably against conditions with unknown plastics and unknown background surfaces. Thus, the RF was used to classify the ten types of plastics mentioned above in one GF-5 and two PRISMA satellite recordings of the same area. In comparison of both sensor systems, the RF produced high quality and transferable results for detecting plastic mainly related to greenhouses, sport fields, photovoltaic constructions and industrial sites that are discussed in detail in this paper. • Machine learning models are used to identify distinct plastics in hyperspectral data. • Instead of regions of interest, spectral libraries were used to train the classifiers. • The RF outperforms the CNN and SVM models and generates transferable results. • Results were validated by controlled experimental lab setups and by aircraft data. • Identification of ten distinct plastic materials in Gaofen-5 and PRISMA satellite data. [ABSTRACT FROM AUTHOR]

Urban land surface temperature (ULST) is a key surface feature parameter in urban heat island studies. However, the geometry and adjacency effects are usually neglected in conventional land surface temperature (LST) retrieval methods. In this study, considering the geometry and adjacency effects, a new urban canopy multiple-scattering thermal radiative transfer (UCM-RT) model and an urban temperature-emissivity separation (UTES) algorithm were developed for ULST retrieval from Chinese GaoFen-5 (GF-5) satellite thermal infrared (TIR) images. The UCM-RT model and the UTES algorithm incorporate multiple scattering within the urban building canopy and consider the thermal radiance contribution from adjacent pixels and the atmosphere. Two schemes were designed to evaluate the geometry and adjacency effects quantitatively in this study: (1) evaluating their impact on ground temperature and top of atmosphere (TOA) brightness temperature for GF-5 four TIR bands from the simulation dataset, and (2) comparing the retrieval temperature difference between the UTES algorithm and conventional temperature-emissivity separation (TES) algorithm (without considering geometry and adjacency effects). For a specific simulation situation, the magnitude of the geometric effects on urban ground temperatures were found to be approximately 3.2 K, 4.2 K, 4.0 K, and 3.9 K for the four GF-5 TIR bands, while the effects on the TOA brightness temperature reached 2.4 K, 3.3 K, 3.3 K, and 3.0 K for those bands, respectively. The results show that the geometry effects have a significant influence on the thermal radiative transfer process. The temperature error of the UTES algorithm was generally up to 0.8 K, much lower than that of the conventional TES algorithm (~2.2 K) in urban surface. The results indicate that the UTES algorithm is suitable for ULST retrieval, and the retrieval temperature difference between the UTES algorithm and conventional TES algorithm ranges from 0.1 K to 2.1 K, especially for dense building pixels. Additionally, the thermal radiation from adjacent pixels in high-spatial-resolution TIR images has significant influence on ULST retrieval. The comparison between the ULST algorithm for urban land surface and the conventional TES algorithm for the flat surface indicated that the three-dimensional structure inside the pixel has a significant influence on the ULST retrieval. The findings of this study indicate that the geometry and adjacency effects must be considered in ULST retrieval for highly accurate results. [Display omitted] • An UCM-RT model taking into account geometry and adjacency effects was deduced. • Accuracy of Urban TES algorithm based on urban effective emissivity is about 0.8 K. • The magnitude of geometry and adjacency effects on urban LST is up to 2.1 K. • Application of high spatial resolution TIR image in urban LST retrieval was analyzed. [ABSTRACT FROM AUTHOR]

Recent developments in hyperspectral satellites have dramatically promoted the wide application of large-scale quantitative remote sensing. As an essential part of preprocessing, cloud detection is of great significance for subsequent quantitative analysis. For Gaofen-5 (GF-5) data producers, the daily cloud detection of hundreds of scenes is a challenging task. Traditional cloud detection methods cannot meet the strict demands of large-scale data production, especially for GF-5 satellites, which have massive data volumes. Deep learning technology, however, is able to perform cloud detection efficiently for massive repositories of satellite data and can even dramatically speed up processing by utilizing thumbnails. Inspired by the outstanding learning capability of convolutional neural networks (CNNs) for feature extraction, we propose a new dual-branch CNN architecture for cloud segmentation for GF-5 preview RGB images, termed a multiscale fusion gated network (MFGNet), which introduces pyramid pooling attention and spatial attention to extract both shallow and deep information. In addition, a new gated multilevel feature fusion module is also employed to fuse features at different depths and scales to generate pixelwise cloud segmentation results. The proposed model is extensively trained on hundreds of globally distributed GF-5 satellite images and compared with current mainstream CNN-based detection networks. The experimental results indicate that our proposed method has a higher F1 score (0.94) and fewer parameters (7.83 M) than the compared methods. [ABSTRACT FROM AUTHOR]