Undaria pinnatifida (Harvey) Suringar, native to north-western Asia, is a prolific invasive seaweed species that has established across much of New Zealand, competing and co-existing with native seaweed species. Remote sensing could be used to map both invasive and native seaweeds in New Zealand. The aim of this study is to evaluate the hyperspectral differences (and the wavelengths at which they differ) between New Zealand seaweed species at broad taxonomic levels and also between Undaria pinnatifida and other common New Zealand seaweeds. ASD Handheld2 VNIR spectroradiometer with a spectral range of 325 to 1075 nm was used to collect hyperspectral data of the common New Zealand native and invasive seaweed species. Two supervised classification methods such as partial least-square discriminant analysis (PLS-DA) for wavelength selection/classification and random forest for validating the results from PLS-DA were implemented. The seaweed species were separable at broad taxonomic level with accuracies greater than 85% for all three taxonomic groups using PLS-DA. The influential wavelengths in discrimination were 572, 633 to 635, 640, 657 to 661, 687, 704 to 714, 741, 742, 749, and 750 nm. Some of the influential wavelengths were consistent with pigment absorption peaks unique to red and brown seaweeds. Undaria pinnatifida showed differences from native brown seaweed species in the visible (574 nm) and near-infrared (716 to 721, 750 nm) region of the electromagnetic spectrum. It was classified with an accuracy of 97.7% and 90.7% using random forest and PLS-DA, respectively.

Moderate Resolution Imaging Spectroradiometer (MODIS) on Aqua has been in operation providing continuous global observations for science research and applications since 2002. The long-term stability of thermal emissive bands (TEBs) of Aqua MODIS was monitored through inter-comparisons with measurements by hyperspectral or multi-spectral infrared sensors or through vicarious monitoring over cold targets such as Dome-C and deep convective clouds. The radiative transfer modeling (RTM)-based simulation models are developed to perform the long-term monitoring of the stability of Aqua MODIS TEBs. Long-term European Centre for Medium-Range Weather Forecasts global atmospheric reanalysis data are used as inputs to the RTM simulation. By confining ocean in low to middle latitudes as the area of interest, the long-term stabilities of Aqua MODIS TEBs are monitored through observation minus background (O-B) brightness temperature (BT) bias between MODIS measurements and simulations from the RTM. In general, the RTM-based O-B analysis shows that the long-term stability of Aqua MODIS TEBs are maintained well. It is shown that Aqua MODIS surface bands are all radiometrically stable with yearly BT bias drifts <0.005  K  /  year for B20, B22, and B23 and ∼0.01  K  /  year for B31 and B32. The carbon dioxide (CO₂) absorption channels of Aqua MODIS, e.g., B33 to B36, are stable with a BT bias drift <0.005  K  /  year. It is also found that after accounting for the long-term growth of greenhouse gas N₂O, the O-B bias trends for B24 and B25 are quite stable with yearly BT bias drifts around 0.0002 and 0.0055  K  /  year, respectively. The stabilities of the moisture-sensitive channels B27 to B29 and the ozone channel B30 of Aqua MODIS are also evaluated and the remnant small variations in the O-B bias trending of these channels are further discussed. The RTM-based inter-comparison of MODIS TEB measurements with global atmospheric and climate re-analysis data provides validation of the long-term radiometric stability of Aqua MODIS TEBs. The analysis in this paper also demonstrates that the comparison of well-calibrated MODIS TEB measurements with RTM simulation helps quantify the long-term impacts of global greenhouse gas concentration growth on the BT decreases in the MODIS greenhouse gas-sensitive channels.

Land use/land cover (LULC) change and climate change are thought to be closely related and mutually influential, especially in contexts where land is converted for urban expansion or agriculture. We represent a first attempt to specify the relationship between LULC change and dryness in a region of Vietnam that is profoundly affected by climate change. Using the temperature–vegetation dryness index (TVDI), we specified the relationships and changes underway in Vietnam’s Ba river basin, one of the largest river systems in the South Central Coast. Using Google Earth Engine, we extracted land use data from Landsat images and calculated TVDI values from Moderate Resolution Imaging Spectroradiometer (MODIS) data for 2000 to 2019. We found, first, that agricultural area and deforestation rose by 7.2% and 2.4% annually, respectively. These changes were driven by economic development, rising crop prices, illegal logging, wildfires, and emergence of new agricultural areas. Second, areas classified in the driest TVDI intervals (dry and very dry) occupied 57% of the basin in 2019, 70% of which was agricultural lands and 20% other (mainly urban and bare lands). These driest land categories expanded at an average rate of 0.08% to 0.1% per year. Moreover, 90% of areas classified as “very wet” and “wet” were forest. We observed a strong relationship between LULC change and TVDI. Climate change and LULC change thus appear to be propelling the basin’s climate toward increasing dryness, suggesting the need for policies to reduce the agricultural area and expand forests while developing more adaptive and sustainable livelihoods.

In the field of offshore oil slicks at the sea surface, radar or optical imagery can provide much useful information. Regarding optical imagery, detection relies on the differences between the reflectance of water alone and oil-covered water. As soon as the thickness of the oil layer is large enough to induce a spectral difference between the reflectance of water and oil-covered water, spectral indices are good candidates for detection. Thereafter, identification and quantification require extended optical and physical knowledge regarding oil properties. To this end, various products have been characterized in the laboratory. Hyperspectral reflectance and transmission measurements have been performed for various pure or emulsified products. This has enabled the definition of two theoretical thickness boundaries: a minimal thickness leading to a noticeable spectral attenuation effect on the reflectance between oil and water and a maximal thickness after which the spectral reflectance no longer evolves with thickness. For some products, these theoretical minimal and maximal thicknesses may not be realistic values for slicks at the sea surface, meaning that they will never be reached. However, they do give a thickness range in which thickness assessment could be performed with optical imagery and modeling. To go deeper into the understanding of oil behavior at sea, experimentation has also been done using a large tank filled with sea water, thus providing more realistic spectral signatures than in the laboratory. From this experiment, the thickness of a small amount of oil freely spread on a plane water surface has been assessed and compared with the boundaries established using the laboratory measurements. The database of spectral signatures built for product identification has been extended with the thickness boundaries and the experimental thicknesses for quantification purpose. This database has been successfully used to identify a product detected on an airborne hyperspectral image and to assess the corresponding oil slick volume, in the case of the NOFO2015 emulsion.

Surface subsidence and its secondary effects caused by underground coal mining continually threaten the safety and property of residents in mining areas. The commonly used total station and leveling approach is time-consuming and laborious, and the spatial range is limited. In this study, an FEIMA D2000 quadrotor unmanned aerial vehicle (UAV) was used to conduct 4-cm  /  pixel photogrammetry in four trial periods over a surface influence area (3.75  km²) of a working mine face. The performance comparison experiments from five point-cloud filtering algorithms in two sub-regions of the study area show that the ATIN algorithm provides the best filtering for this study area. Coal mining subsidence monitoring in mountainous area based on time-series UAV photogrammetry technology is proposed. Using this approach, the ATIN algorithm was used to filter the overall point cloud data in the study area to obtain the digital elevation model (DEM). The surface dynamic subsidence basin was determined with two phases of DEM subtraction. The results show that, compared with the measured leveling data in the field in the same period, the average root mean square error is 165 mm, and the maximum subsidence monitoring accuracy reaches 98%.

A technique that utilizes quasi-deep convective clouds (qDCC) for the calibration assessment of the thermal emissive bands (TEB) on remote sensing instruments has been proven viable. The Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) and Suomi-National Polar-orbiting Partnership (S-NPP) and NOAA-20 (N20) Visible Infrared Imaging Radiometer Suite (VIIRS) TEB calibration uses a nonlinear algorithm with nonlinear coefficients that rely on on-orbit blackbody (BB) warm-up and cool-down operations for updates. However, a limited BB temperature range affects the calibration’s accuracy, particularly for cold scenes. The deep convective clouds (DCC) core, one of the coldest Earth scenes, is suitable for MODIS calibration assessments, more specifically, for the evaluation of the offset term’s effect in its TEB quadratic calibration function. Moreover, nighttime qDCC measurements provide the advantage of removing solar reflectance effects during daytime, thus enhancing the assessment’s accuracy for the mid-wave infrared TEB. This qDCC method is applied to the Terra MODIS and VIIRS TEB, and their stabilities are assessed using long-term qDCC trending measurements over the instruments’ missions. The measurements from bands with an 11-μm wavelength are used to identify the DCC pixels. The 11-μm bands, MODIS band 31, and VIIRS bands M15 and I5 are stable throughout the MODIS and VIIRS missions and have shown excellent calibration accuracy and noise performance. Hence, using these bands as references, a normalization method is employed to enhance the accuracy of the stability and consistency assessments. The S-NPP VIIRS TEB show stable trends over the instrument’s mission. The S-NPP-to-N20 VIIRS comparison shows that their TEB measurements are consistent over qDCC. The Terra MODIS TEB also shows stable performance—except for bands 27, 29, and 30. Terra MODIS band 30 shows a large downward trend throughout the mission, whereas bands 27 and 29 show slight, upward drifts. Finally, a calibration correction using qDCC assessments is discussed and intended to be used in a future calibration algorithm collection.

A time-series filtering algorithm named the best slope temporal segmentation filtering (BSTS) algorithm was developed to improve the estimation accuracy of above-ground biomass (AGB) of Pinus densata Mast. In this algorithm, the best slope was used to interrupt the time series. Following interruptions, linear regression was employed to simulate each segment of time series. By splicing the time-series fragments, a new reconstructed time series would be generated. The algorithm was used to reconstruct the annual synthetic Landsat surface reflectance images collection in Shangri-La from 1987 to 2019. Permanent sampling plots of Pinus densata Mast. in 1987, 1992, 1997, 2002, 2007, 2012, and 2017 were collected. Then the multiple linear regression and random forests regression (RFR) were used to estimate the AGB of the Pinus densata Mast. The Landsat time series used before and after filtering were evaluated based on the quantitative analysis. The results with RFR showed that for modeling, the R² of the time series reconstructed by the BSTS is 0.913, which is higher than the data before reconstructed (R²  =  0.869). The root-mean-square error (RMSE) of the reconstructed time series (RMSE  =  35.38  t  /  ha) is significantly lower than the data before reconstructed (RMSE  =  43.43  t  /  ha). For validation, the average prediction accuracy (p) before filtering is 61.92%, and the relative root-mean-square error (rRMSE) is 48.52%; whereas the p after filtering is increased to 70.66%, and the rRMSE is reduced to 35.38%. In summary, the BSTS has improved the accuracy of remote sensing-based estimation of AGB of Pinus densata Mast., which provided a method for large area AGB mapping for Pinus densata Mast. in Shangri-La.

Global navigation satellite system (GNSS) is the standard solution for solving the localization problem in outdoor environments, but its signal might be lost when driving in dense urban areas or in the presence of heavy vegetation or overhanging canopies. Hence, there is a need for alternative or complementary localization methods for autonomous driving. In recent years, exteroceptive sensors have gained much attention due to significant improvements in accuracy and cost-effectiveness, especially for 3D range sensors. By registering two successive 3D scans, known as scan matching, it is possible to estimate the pose of a vehicle. This work aims to provide in-depth analysis and comparison of the state-of-the-art 3D scan matching approaches as a solution to the localization problem of autonomous vehicles. Eight techniques (deterministic and probabilistic) are investigated: iterative closest point (with three different embodiments), normal distribution transform, coherent point drift, Gaussian mixture model, support vector-parametrized Gaussian mixture and the particle filter implementation. They are demonstrated in long path trials in both urban and agricultural environments and compared in terms of accuracy and consistency. On the one hand, most of the techniques can be successfully used in urban scenarios with the probabilistic approaches that show the best accuracy. On the other hand, agricultural settings have proved to be more challenging with significant errors even in short distance trials due to the presence of featureless natural objects. The results and discussion of this work will provide a guide for selecting the most suitable method and will encourage building of improvements on the identified limitations.

Deramp space-time adaptive processing (Deramp-STAP) algorithm has been validated in its effectiveness in reducing the required channels to enable ground moving target indication (GMTI) for multi-channel synthetic aperture radar (SAR) systems. However, the parameter estimation for the moving target, including the critical radial velocity, is not accurate enough, and the radial velocity is always approximated by the equivalent velocity obtained by the Deramp operation. For the indication and imaging of ground moving targets, the accurate estimation of the radial velocity is of importance to improve the signal-clutter-noise-ratio of STAP and estimate the initial azimuth position of the moving target. A method for GMTI and the corresponding parameter estimation using Deramp-STAP is proposed. Compared with the traditional Deramp-STAP algorithms, the proposed algorithm utilizes the equivalent velocity estimated by Radon transform to remove the linear range migration before STAP and figure out where the moving target is located in the azimuth frequency domain. To estimate the radial velocity more accurately, the fast Fourier transform along the spatial direction is performed on the pixel where the moving target is located in the azimuth frequency domain after the Deramp operation, and then the radial velocity can be obtained through the spatial spectrum analysis. Finally, the estimated radial velocity is applied to construct the azimuth compression function for the moving target imaging, and then the initial azimuth position can be estimated. Several simulation experiments have been implemented to validate the effectiveness of the proposed algorithm.

Globally, the unprecedented increase in population in many cities has led to rapid changes in urban landscape, which requires timely assessments and monitoring. Accurate determination of built-up information is vital for urban planning and environmental management. Often, the determination of the built-up area information has been dependent on field surveys, which is laborious and time-consuming. Remote sensing data are the only option for deriving spatially explicit and timely built-up area information. There are few spectral indices for built-up areas and often not accurate as they are specific to impervious material, age, colour, and thickness, especially using higher resolution images. The objective of this study is to test the utility of a new built-up extraction index (NBEI) using WorldView-2 (WV-2) to improve built-up material mapping irrespective of material type, age, and color. The new index was derived from spectral bands such as green, red edge, NIR1, and NIR2 bands that profoundly explain the variation in built-up areas on WV-2 image. The result showed that NBEI improves the extraction of built-up areas with high accuracy [area under the receiver operating characteristic curve,   (  AUROC  )    =    ∼  0.82] compared to the existing indices such as built-up area index (AUROC  =    ∼  0.73), built-up spectral index (AUROC  =    ∼  0.78), red edge/green index (AUROC  =    ∼  0.71) and WorldView-Built-up Index (WV-BI) (AUROC  =    ∼  0.67). The study demonstrated that the new built-up index could extract built-up areas using high-resolution images. The performance of NBEI could be attributed to the fact that it is not material-specific, and would be necessary for urban area mapping.

For their high maneuverability, unmanned aerial vehicles (UAVs) are widely used in object detection, including the detection of ships. However, object detection in aerial images taken by UAV remains a challenge due to the arbitrary shooting perspectives and small proportion of targets. Existing anchor-based detectors, whose performance could be easily affected by the aspect ratios and scales of anchor boxes, could get into difficulties in handling candidate targets with wide shape variations. We propose an efficient anchor-free detector to replace a set of predefined anchor boxes. Specifically, guided attention module, embedded in the feature pyramid structure, is put forward to help low-level feature maps acquire the guiding information of high-level feature maps in the multi-scale fusion stage. Then an intersection-over-union (IoU) prediction head is added to predict the IoU for each predicted box. The output from IoU prediction and classification branches is then evaluated to dynamically generate soft labels without sacrificing the effiency in an attempt to improve the performance of the proposed detector. The results of extensive experiments demonstrate that the performance of our proposed detector is better than that of several current mainstream detectors.

Hyperspectral images (HSIs) contain rich spectral information and spatial information. How to apply these two information types and fully combine the correlation between them remains a challenge worthy of further research and discussion. In this study, a multi-branch-multi-scale residual fusion network (MB-MS-RFN) for HSI classification is proposed. First, a 3D multi-branch-multi-scale convolution residual network, which can acquire image features of different scale in the training process and consider the correlation between spectral information and spatial information, is developed. Instead of deepening the network, the multi-branch structure widens the network horizontally to obtain more accurate classification. Finally, the different levels of HSI features are fused to obtain better classification results. Several experiments have been carried out to verify the proposed framework, and the results have demonstrated that the proposed MB-MS-RFN framework can improve the classification performance of HSIs. The performance of the MB-MS-RFN was evaluated using the Indian Pines, Pavia University, and Kennedy Space Center datasets; the performances’ overall accuracies were 99.66%, 99.92%, and 99.97%, respectively. The results from a series of experiments confirm that the proposed method offers several advantages in classification accuracy compared with five other methods.

Crop segregation in optical remote sensing cannot always be easily achieved among similar species since their spectral response can be similar. We aimed to distinguish fields of two C4 gramineae with similar plant structure and crop calendar, maize (Zea mays) and sorghum (Sorghum bicolor), in an agricultural area of the state of Guanajuato, in central México. One hundred and forty-three fields of these two crops were identified and their average spectral signatures were derived at different phenological stages, using digital information from 11 Sentinel 2A/B images, corresponding to the 2019 spring–summer cycle. We also calculated the temporal profile of six spectral indices for each crop, to evaluate seasonal variations and conducted a cropland classification, using the random forests algorithm and different combinations of dates and indices. Results show that segregation of maize and sorghum fields can be achieved due to small differences in the near-infrared region of their spectral signatures and disparities in the rate of decay during the senescence stage. Classification results also confirmed the separability of these two crops, obtaining a very high overall accuracy and kappa values (98.43% and 97.07%, correspondingly) based on the reflectance of nine images and only three spectral indices: the normalized difference vegetation index, the normalized difference water index, and the structure insensitive pigment index. This investigation demonstrates that the combined use of time series of high-resolution images and spectral indices can contribute to distinguish fields of similar crop species.

Livestock in the Amazon is mainly limited by the lack of forage biomass production in the dry season. The sensitivity of pasture to drought depends on pasture management and on certain natural factors that enable the grass cover to stay green for longer under low rainfall. The sustainability of livestock production systems depends on identifying these natural factors to make better use of their forage production potential without consuming additional water resources or resorting to deforestation. We built an index of pasture dependence on rainfall at the beginning of the dry season by calculating the slope of the linear regression between the MODIS enhanced vegetation index and monthly cumulative rainfall obtained from the Tropical Rainfall Measuring Mission. The index was calculated in Paragominas municipality. A 15-year time series (2000 to 2014) was used to assess the sensitivity of pasture to drought. A map of geomorphological units was overlaid on the index map to check whether the soil and the landform influence drought sensitivity at a large scale. Drought resistance was shown to be higher in lowland areas with loamy sand with a shallow groundwater table and on plateaus covered by fertile sedimentary clay. The escarpments with a ferruginous duricrust and mottled clay were shown to be the most sensitive to drought. These results underline the importance of natural factors in pasture productivity in tropical humid areas.

Net surface radiation defines the availability of radiation energy on and near the surface to drive many physical and physiological processes such as latent heat, sensible heat fluxes, and evapotranspiration. One of the prime challenges of modeling radiation budget is estimation of net longwave radiation. Incoming or downwelling longwave radiation (LW_in) flux is one of the two key components of net longwave radiation. Its estimation in cloudy conditions has always been a challenge due to lack of instrumentation and regular measurements at different spatial scales. In this study, two artificial neural network (ANN) multi-layer perceptron (MLP) models were developed for LW_in flux estimation under cloudy-sky during daytime and nighttime using half-hourly flux measurements over different agro-climatic settings and several atmospheric parameters from measurements, satellite-based observations, and model outputs. A comparative evaluation was made between existing or newly developed multivariate linear regression (MVR) models and ANN-based models. The latter set of models were found to be superior to the best MVR model during both daytime and nighttime. The ANN models were found to have consistent performance across different sites and cloud types except less accuracy in sub-humid or humid climate and in deep convection cloud. The ANN models showed overall accuracies of 2.7% and 3.3% of measured mean and R² of 0.86 and 0.85 for daytime and nighttime, respectively, when compared with independent data of in-situ measurements.

In the remote sensing image processing field, cloud and snow detection for high-resolution sensors and cloud and snow morphology in different latitudes of the world is challenging. A deep learning training model (Softmax) was developed to improve the accuracy of cloud and snow identification from Gaofen-1 and Pakistan Remote Sensing Satellite-1 images. First, more than 1800 scenes remote sensing images in various regions over the world are collected. Next, the texture details and spectral information of the objects are extracted. Finally, the Softmax model is applied to process the features to obtain the final cloud and snow masks. The cloud and snow detection results are evaluated by performing statistical analysis. The overall accuracy for cloud detection reaches 92.64% (kappa coefficient = 0.83) and for snow detection reaches 93.94% (kappa coefficient = 0.8). The algorithm is not only accurate but also computationally efficient. It is of great importance for image processing in ground segment and corresponding applications.

We report results of lidar studies of the atmospheric aerosols present above the city of Sofia during four episodes of wintertime Saharan dust intrusions in the last decade. The optical and microphysical properties of the detected aerosols, as well as the aerosol layering and dynamics, are retrieved and characterized. By combining lidar results with in situ PM₁₀ concentration measurement data and air-transport modeling/forecast data, the effects of mixing and interaction of desert aerosols with local ones are analyzed as dependent on the altitude range, phase, and intensity of the dust load events. Included in the analysis are also meteorological radiosonde data, which allow us to ascertain the direct and strong influence of the desert air and aerosols on the local atmospheric conditions and meteorological parameters and thus, indirectly, on essential components of the local weather, such as the seasonal thermal regime and the regimes of cloud formation and precipitation. Using online resources, color maps of air temperature anomalies in the region for the days of lidar measurements are also provided based on climatological data over the three preceding decades. The results presented are in conformity with previously published statistical data of other authors and indicate that the wintertime intrusions of warm air masses from North Africa carrying Saharan dust to Europe, and particularly to the Balkans, rare and atypical until recently, nowadays show a trend of increasing in terms of frequency and intensity, pointing to deviations in the seasonal regime of the involved intercontinental air circulation systems.

Forest change detection is crucial for sustainable forest management. The changes in the forest area due to deforestation (such as wild fires or logging due to development activities) or afforestation alter the total forest area. Additionally, it impacts the available stock for commercial purposes, climate change due to carbon emissions, and biodiversity of the forest habitat estimations, which are essential for disaster management and policy making. In recent years, foresters have relied on hand-crafted features or bi-temporal change detection methods to detect change in the remote sensing imagery to estimate the forest area. Due to manual processing steps, these methods are fragile and prone to errors and can generate inaccurate (i.e., under or over) segmentation results. In contrast to traditional methods, we present AI-ForestWatch, an end to end framework for forest estimation and change analysis. The proposed approach uses deep convolution neural network-based semantic segmentation to process multi-spectral space-borne images to quantitatively monitor the forest cover change patterns by automatically extracting features from the dataset. Our analysis is completely data driven and has been performed using extended (with vegetation indices) Landsat-8 multi-spectral imagery from 2014 to 2020. As a case study, we estimated the forest area in 15 districts of Pakistan and generated forest change maps from 2014 to 2020, where major afforestation activity is carried out during this period. Our critical analysis shows an improvement of forest cover in 14 out of 15 districts. The AI-ForestWatch framework along with the associated dataset will be made public upon publication so that it can be adapted by other countries or regions.

Urbanization at the expense of the natural environment has been increasing in Turkey in recent years. The toll it takes on the ecosystem has an adverse impact on local weather systems and natural resources. These rapid ecosystem changes are mostly observed in the western parts of Turkey. For a reliable prediction of sustainable development planning, the normalized difference vegetation index (NDVI) can be used as the main index for the description of both urbanization and land class. This study proposes the optimum adaptive neural-fuzzy inference systems (ANFIS) and hybrid Wavelet-ANFIS (WANFIS) models to estimate NDVI variation for certain seasonal (summer and winter) data in a grid area of 20  ×  20  km² centered at the Kandilli region of Istanbul, Turkey (28° 57′ 53″ E and 41° 01′ 07″ N). The calculated NDVI values were obtained using the WANFIS model and compared with the Moderate Resolution Imaging Spectroradiometer Observations. The results reveal that the mean absolute percentage error of NDVI values are calculated as 1.5% and 0.7% for the winter and summer test datasets, respectively. The coefficient of determination (R²) values of the winter and summer test datasets are 0.977 and 0.991, respectively. Further, when compared with ANFIS, the results demonstrate that the WANFIS model exhibits a better approximation for the estimation of NDVI variation in the summer than in the winter.

Cloud thermodynamic phase (CTP) classification is a vital piece of information for determining how clouds affect climate and electromagnetic wave propagation. We describe a passive shortwave infrared (SWIR) three-channel polarimeter for CTP remote sensing. This instrument adds polarimetric sensitivity to the traditional radiance ratio method that capitalizes on the difference in spectral absorption between liquid water and ice. The ground-based polarimeter was used to measure the first three Stokes parameters in three SWIR spectral bands alongside a dual-polarization lidar system for validation. Additionally, we operated the polarimeter in division-of-time and division-of-aperture modes to explore the severity of polarization artifacts that arose from fast-moving clouds. Although temporal smearing of polarization signatures was identified, the impacts were found to be minimal. Using a surface fitting technique, the radiance ratio method was directly compared with a method that combined radiance ratios with the S₁ Stokes parameter. We found that the addition of polarimetry improved cloud phase classification ability from ∼73  %   to 95%.

The destruction created by cyclones depends upon their intensity. Ocean wind is the crucial parameter to understand and forecast the intensity of cyclones. During cyclones, due to dynamic ocean conditions and the limited data availability, a neural network with the regularization approach is used to retrieve cyclonic wind (CW) speed. Several optimization algorithms are compared to get a robust neural network. The Bayesian regularization with the Levenberg–Marquardt optimized network (BNN) is found suitable to develop the geophysical model to retrieve CW speed using Cyclone Global Navigation Satellite System (CYGNSS) measurements. To select the suitable observables for BNN, sensitivity analysis of CYGNSS observables is carried out. The root-mean-square difference between retrieved CW speed and the airborne radiometer data is found to be 4.35  ms^−  1, which is smaller than the value quoted in the literature for CYGNSS CW speed. Further, a rigorous analysis is also done to find out the effect of the rain on the retrieved wind speed. Independent validation of our approach is carried out using Soil Moisture Active Passive (SMAP) radiometer high-wind data with a special case of category-5 cyclone Lorenzo. The results support that the proposed algorithm agrees well with the SMAP CW data.

The variability of the remote sensing reflectance, R_rs, now routinely retrieved from ocean color (OC) and high spatial resolution sensors, is often used to characterize water variability due to changes in inherent optical properties of the water body. At the same time, R_rs is partially variable because of uncertainties in its retrieval in the process of atmospheric correction. Using data from SNPP-VIIRS and Landsat-8 OLI sensors, the contribution of the main components to the variance of R_rs due to its spatial variability is determined based on a model in which variances were considered proportional to the mean values of the corresponding components. It is shown that there is practically no spatial variability in the open ocean waters and water variability is proportional to the spatial resolution of the sensor in coastal waters. Variances due to surface effects, inaccuracies of aerosol models, and sunglint can contribute significantly to R_rs variance, which characterizes R_rs spatial variability, with variances due to the water variability itself often being significantly smaller.

Mobile laser scanning (MLS), which can quickly collect a high-resolution and high-precision point cloud of the surroundings of a vehicle, is an appealing technology for three-dimensional (3D) urban scene analysis. In this regard, the classification of MLS point clouds is a common and core task. We focus on pointwise classification, in which each individual point is categorized into a specific class by applying a binary classifier involving a set of local features derived from the neighborhoods of the point. To speed up the neighbor search and enhance feature distinctiveness for pointwise classification, we exploit the topological and semantic information in the raw data acquired by light detection and ranging (LiDAR) and recorded in scan order. First, a two-dimensional (2D) scan grid for data indexing is recovered, and the relative 3D coordinates with respect to the LiDAR position are calculated. Subsequently, a set of local features is extracted using an efficient neighbor search method with a low computational complexity independent of the number of points in a point cloud. These features are further merged to produce a variety of binary classifiers for specific classes via a GentleBoost supervised learning algorithm combining decision trees. The experimental results on the Paris-rue-Cassette database demonstrate that the proposed approach outperforms the state-of-the-art methods with a 10% improvement in the F₁ score, whereas it uses simpler geometric features derived from a spherical neighborhood with a radius of 0.5 m.

The efficient and rapid extraction of information about the growth period of the Zanthoxylum bungeanum maxim is one of the prerequisites for analyzing and mastering its growth trend. The 5-cm high-resolution visible-light red, green, blue (RGB) image of the Zanthoxylum bungeanum growing period is obtained based on the low-altitude aerial photography of a quad-rotor unmanned aerial vehicle (UAV) platform. The color space of the R-, G-, and B-channels is converted into the H-, S-, L-, and V-components, and the sensitivity of seven feature vectors to target features is analyzed. The combination of the R-band, H-component, and S-component (which distinguish ground objects very differently) is obtained as the identification feature. Motivated by this, the color space index (CSI) is built by three sensitive component features, which are segmented by the maximum class variance method (OTSU), filtered by majority filtering analysis, and processed by clump to obtain spots containing only the plants. The number of single plants completely separated was calculated by ArcGIS10.2 software, and the average area of a single plant was obtained. The number of conjoined plants was obtained by dividing the cluster by the average area of a single plant. The number of separated complete plants and cluster divisions is counted to get the final total amount of plants and compared with the excess green mainstream index (EXG), visible-light band difference vegetation index (VDVI), and normalized green red difference index (NGRDI). The results indicate the following: (1) the reflectance overlapping phenomenon between ground objects is usually encountered during the construction of the index using only the RGB channel features of the visible-light band. However, the H-, S-, L-, and V-components are merged when the original image features are changed by the color space, in which the images containing only the RGB band are enriched and more conducive to finding sensitive features to distinguish between features; (2) in the CSI extraction results of three experimental research areas and one test verification area, the overall accuracy exceeded 90% (kappa index over 0.90), the false extraction rate and leakage extraction rate were <10  %  , and the results showed that the CSI had an effect on the identification and extraction of Zanthoxylum bungeanum; and (3) when the growth environment of Zanthoxylum bungeanum is disturbed by weeds, the CSI has more advantages than the EXG and VDVI.

A synthetic aperture radar (SAR) target recognition method is proposed using monogenic components as basic features. The monogenic signal is employed to decompose original SAR images into multi-scale components. Considering the redundancy and possible indiscrimination in the monogenic components, the nonlinear correlation information entropy (NCIE) is adopted as the criteria for the selection of valid components. The subset of monogenic components with the highest NCIE is chosen and classified by joint sparse representation (JSR). Using the inner correlations of the selected components, JSR could improve the overall reconstruction precision thus enhancing the recognition performance. Experiments are proceeded on the moving and stationary target acquisition and recognition dataset under the standard operating condition and several extended operating conditions, including configuration variances, depression angle variances, noise corruption, and partial occlusion. The results validate the superior effectiveness and robustness of the proposed method over several existed SAR target recognition methods.

Wetlands for the entire state of Pennsylvania (119  ,  282  km²) were mapped using object-based image analysis (OBIA). The OBIA combined vector output from logistic statistical models informed by hydrogeologic variables (e.g., hydrology facets, soils, and climate) with LiDAR-derived terrain variables, high-resolution aerial orthoimagery, and land-cover data through a data-fusion approach. Wetland presence data and hydrogeologic features in a geographic information system were used to develop moderate-resolution (10  m²) maximum-entropy models (Maxent) both for wetlands dominated by low-lying emergent vegetation and woody vegetation, respectively. Terrain variables describing slope, landscape wetness, surface elevation, climate, precipitation, and poorly drained soils were most strongly associated with occurrences of wetlands in this modeling procedure. Maxent derived models accurately predicted validation wetlands: Area under curve of 0.91 for woody wetlands and 0.93 for emergent wetlands. High-resolution (1  m²) mapping was then performed in the OBIA by creating objects based on a compound topographic index (CTI) and leaf-off true-color orthoimagery and then classifying them using a combination of the CTI, imagery, and 10-m² statistical models. An accuracy assessment found that the OBIA correctly classified 81.5% of validation locations compared to only 59.3% for National Wetlands Inventory (NWI), although commission rates were higher for OBIA (15.6%) versus NWI (3.4%). The user’s accuracy for OBIA was 82.8% compared to 78.1% for NWI. Our work suggests advances in remote sensing techniques and a data-fusion approach may facilitate improved wetland mapping accuracy while reducing processing time relative to traditional approaches.

Sparsity and low rankness are essential properties of real-world data. The sparsity attribute mentions that the data contain very few non-zero elements. On the other hand, the low-rank property characterizes redundancy in the data. The concept of sparsity and low rankness is central to numerous signal processing tasks such as signal inversion, data compression, and noise removal. With the advent of multi-sensor and multi-dimensional data, we frequently encounter multi-way data or tensors with sparse and low-rank structure. Prevalent vector or matrix-based sparsity estimation criterias cannot conveniently characterize the sparsity structure of multi-way data. Hence, we require efficient criteria to quantify the sparsity structure of multi-way data. Sparsity plays a crucial aspect in hyperspectral image processing tasks such as unmixing and denoising. Hyperspectral images contain an inherent multi-way structure; we perform multi-way sparsity estimation in hyperspectral image processing. We introduce an efficient criterion to estimate multi-way sparsity, facilitating signal inversion, and noise removal tasks. The proposed sparsity criteria combine the sparsity of the core tensor obtained by Tucker tensor decomposition, the low rankness of the tensor, and the approximated rank. Since the core tensor obtained by tucker decomposition is highly sparse, the core tensor sparsity reflects the sparsity of the whole tensor indirectly. On the other hand, the nuclear norm and Stein’s unbiased risk estimate quantifies the low rankness of the tensor. Unlike other approaches, our proposed measure takes both low rankness and sparsity attribute into account. In addition, our proposed sparsity measure also satisfies some of the desirable properties of ideal sparsity measures. We demonstrate the efficacy of our proposed sparsity quantification measure in applications such as hyperspectral image denoising, hyperspectral unmixing tasks as demonstrated in the real image experiments.

Vehicle detection is an important method for understanding high-resolution remote sensing images. Deep convolutional neural network (DCNN)-based methods have improved many computer vision tasks and have achieved state-of-the-art results in many object detection datasets. Object detection of remote sensing images has been radically changed by the introduction of DCNN. Considering correlation between the scale distribution of objects and spatial resolution of remote sensing images, we propose an improved vehicle detection method based on a YOLOv3 model. A multi-scale clustering anchor box generation algorithm is proposed to obtain the anchor box parameters that match the resolution of each layer of the feature pyramid of model. This allows us to get more accurate anchor parameters. Focal loss is introduced into the default loss function to reduce the weight of negative samples, which were easily classified, that focus the model training process on samples that are difficult to classify. For the imbalance problem of positive and negative samples in the detection method based on the prior anchor box, focal loss is used to focus the model training process on samples that are difficult to classify. The experiment is performed on a dataset consisting of remote sensing images obtained from Worldview-3, and the results show that compared with the basic YOLOv3 algorithm, the average accuracy of vehicle detection is improved by 8.44%. The accuracy of vehicle detection of high-resolution remote sensing images is significantly improved while maintaining the speed of single-stage target detection. This approach is tested on an xView dataset consisting of remote sensing images obtained from Worldview-3. In addition, through using the proposed method, the average precision of vehicle detection increased by 8.44%. The experimental results show that the proposed method can be used for object detection in high-resolution remote sensing images effectively, and this method can significantly improve the performance of the model without sacrificing inference speed.

Deep learning opened new possibilities for remote sensing image analysis using multiple neural nets layers. We introduce a hybrid pixel-based model that allows improving the unsupervised training with stacked autoencoders (SAE) by inserting convolutional neural networks (CNN) in the encoding and decoding steps. Inclusion of the convolution in the encoding and decoding steps allows a feature-based description of the pixel’s hyperspectral signature, suitable to perform an initial unsupervised classification. As one-dimensional (1D) filters are applied, the processing effort is lower than when using two-dimensional-CNN. Finally, to adapt the classifier to the desired classes, the parameters of the net are adjusted using training samples and fine-tuning followed by logistic regression using the softmax activation function. This combination explores the potential of both, autoencoders (AE) and convolutional nets, providing an alternative for the classification of hyperspectral data. To evaluate the performance of the proposed approach, it was compared to traditional machine learning algorithms such as support vector machine, artificial neural networks, CNN, and SAE. The results show that the use of the SAE-1DCNN method is more effective in terms of hyperspectral classification accuracy and more efficient in computational complexity and that it can be an alternative for hyperspectral data classification.

The two-dimensional (2D) resolution is poor due to the narrow transmitting bandwidth and the limited observation angle in monostatic ISAR imaging. A multiradar fusion imaging method based on fast linearized Bregman iteration (FLBI) algorithm is proposed to improve the 2D resolution of the ISAR imaging. First, the sparsity of the ISAR imaging echo data is exploited to establish the multiradar fusion ISAR imaging model based on sparse representation, which can be converted into a one-dimensional sparse vector reconstruction problem. Then, a sparse reconstruction method based on FLBI is proposed to solve the sparse representation problem with large scales and achieve the ISAR fusion imaging. Combined with the weighted back-adding residual and condition number optimization of the sensing matrix, the FLBI algorithm can further accelerate the iterative convergence speed. The proposed algorithm only involves matrix–vector multiplications and componentwise shrinkages, which greatly improves the imaging efficiency. Finally, the simulation results show that the proposed method can effectively improve the iterative convergence speed and achieve the better 2D ISAR fusion imaging.

We focus on the range migration (RM) and Doppler frequency migration (DFM) corrections in the long-time coherent integration, and a fast detection method based on two-dimensional trilinear autocorrelation function is proposed for the maneuvering target with jerk motion. This proposed method can integrate the echoes’ energy into peaks in a three-dimensional parameter space coherently and estimate the target’s radial range, acceleration, and jerk simultaneously by the peak detection technique. Then through the estimations of radial range, acceleration, and jerk, the radial velocity can be obtained through one-dimensional parameter searching. Finally, RM and DFM can be compensated simultaneously, and the target can be detected through the constant false alarm technique. This proposed method can strike a good balance between the computational complexity and detection performance. Experiments with the simulation and real measured radar data are conducted to verify the proposed method.

Coastal monitoring is an essential feature for the sustainable management of naturally vulnerable areas; however, data acquisition is a tedious task. We aim to identify an efficient method of automatic shoreline monitoring based on high water level detection using very high-resolution Pleiades images and taking as the pilot zone the Mostaganem coastline (Algeria). Through a comparative study between classification methods based on pixel- and object-based image analyses (PBIA and OBIA, respectively), algorithmic development and optimizing was conducted on two machine learning (ML) classifiers: random forest (RF), and support vector machine (SVM), and two segmentation algorithms: multiresolution (MRS) and meanshift (MSS). These classification methods yielded six different shorelines that were validated using an in-situ GPS survey shoreline acquired on the same day as the Pleiades image. The results showed that the OBIA generated a shoreline with a 5% to 25% better accuracy than that of PBIA using the same ML algorithm. Within the OBIA approach, MRS generated a shoreline with 20% higher accuracy compared to MSS, suggesting the importance of segmentation possessing. The RF based on MRS was the method that produced the shoreline at the best accuracy, where 55.5% of the extracted shoreline was within 1 pixel of the in situ shoreline. This method was successfully shown to be a good alternative for shoreline monitoring of sandy microtidal coasts, offering to coastal managers a reliable tool to complete the data and efficiently manage the coastal erosion.

Activation functions play increasingly important roles in deep learning. The activation functions give the neural network the ability to learn complex patterns by introducing non-linearity. Complex-valued activation functions for complex-valued neural networks are extended from real-valued activation functions like sigmoid, tanh, rectified linear unit, and exponential linear unit. Recently, the most widely used complex-valued activation function is the real-imaginary-type activation function (RIAF), which applies separate real-valued activation functions on both the real and the imaginary parts of a neuron. This separate activation method ignores the internal relationship between the real and imaginary parts of the complex input and loses the integrity of activation. In this work, a complex-valued activation function called Gaussian-type activation function (GTAF) is proposed to activate the real and imaginary parts of a neuron jointly and enhance the learning ability and learning speed. GTAF consists of an ordinary real-valued activation function and a Gaussian function and is suitable for any given complex-valued neural network. Two polar-SAR image classification experiments on complex-valued convolutional neural networks show that GTAF converges faster than RIAF and has higher accuracy.

Staggered synthetic aperture radar (SAR), which operates with pulse repetition interval (PRI) variation, deals with the blind range problem caused by constant PRI in conventional high-resolution wide-swath SAR imaging. Due to the variation of PRI, there appear periodic loss of pulses and nonuniform sampling in azimuth. However, the exiting reconstruction methods mostly correct range curvature after resampling the nonuniform samplings into a uniform grid. It inevitably introduces the errors and thus affects the effectiveness of the correction. To solve the nonuniform sampling problem and eliminate the errors, our study proposes a reconstruction algorithm [reconstruction algorithm with modified second-order keystone transform (RMSKT)] with modified second-order keystone transform in staggered SAR mode. Moreover, to provide a priori knowledge required by the reconstruction, i.e., the nonuniform sampling distribution, which depends on PRI variation, our study first designs an optimal PRI variation scheme and optimizes the sequence parameters. With the designed PRIs sequence, the proposed RMSKT algorithm can achieve nonuniform range curvature correction and data reconstruction jointly. The proposed RMSKT algorithm shows significant improvements over the traditional reconstruction methods in terms of the peak-sidelobe ratio, the integral sidelobe ratio, and the azimuth ambiguity-to-signal ratio. Simulations and experiments on raw data generated in staggered mode are performed to verify the effectiveness of the proposed method.

The semantic segmentation of optical satellite remote sensing images is more challenging than that of natural images, owing to the considerable differences in the texture, shape, topology, and scale of ground features in different areas and the coexistence of dense and sparse arrangements. To alleviate these difficulties and ensure a high accuracy, a semantic segmentation framework named the self-attention and edge enhancement network (SE²Net) is constructed considering two aspects. First, because the self-attention mechanism can capture more useful semantic information by modeling large neighborhood correlations, we embed a self-attention module known as spatial expectation maximization attention (SEMA) in the considered network. Second, the Laplace operator is adopted to explore the significant edge information to design an edge enhancement module (EEM). Finally, both SEMA and EEM are embedded in the proposed SE²Net, thereby forming an end-to-end network. To validate the performance of the proposed approach, we construct a semantic segmentation dataset (SSD) using Tian-Hui 1 satellite images and conduct extensive experiments on both the SSD and the gaofen image dataset (GID). The results demonstrate the superiority of the proposed method over other state-of-the-art approaches and the effectiveness of the constructed SSD.

In recent years, many target detection methods for hyperspectral images based on sparse representation (SR) theory have been proposed and achieved good results. However, these methods still have some deficiencies. Specifically, these methods usually give the same weight to the pixels in the neighborhood when they use the spatial information, and obtain the spectral average features to replace the spectral features of the central pixel. However, the influence of each pixel in the neighborhood of the central pixel may be different. In addition, when the dictionary learning method is used for target detection, its performance depends on the constructed dictionary. To solve these problems, an innovative target detection method based on spatial-spectral joint weighted dictionary learning with an online updating mechanism is proposed. First of all, different from traditional SR method, this method integrates the spatial adaptive neighborhood information into the sparse coding stage to realize the spatial-spectral weighted joint SR. Second, the online dictionary updating mechanism and a two-step optimization method are used together to update the dictionary and sparse coding coefficients alternately, which can learn the optimal dictionary set in real time. The experimental results show that the proposed method has better performance when compared with four state-of-the-art dictionary learning-based methods for hyperspectral target detection.

Preprocessing is a major area of interest in the field of hyperspectral endmember extraction, for it can provide a few high-quality candidates for fast endmember extraction without sacrificing endmember accuracy. We propose a superpixel-guided preprocessing (SGPP) algorithm to accelerate endmember extraction based on spatial compactness and spectral purity analysis. The proposed SGPP first transforms a hyperspectral image into low-dimension data using principal component analysis. SGPP then utilizes the superpixel method, which normally has linear complexity, to segment the first three components into a set of superpixels. Next, SGPP transforms low-dimension superpixels into noise-reduced superpixels and calculates their spatial compactness and spectral purity based on Tukey’s test and data convexity. SGPP finally retains a few high-quality pixels from each superpixel with high spatial compactness and spectral purity indices for subsequent endmember identification. Based on the spectral angle distance, root-mean-square error, and speedup, experiments are conducted on synthetic and real hyperspectral datasets, and they indicate that SGPP is superior to current state-of-the-art preprocessing techniques.

In the final feature map obtained using a convolutional neural network for remote sensing image segmentation, there are great differences between the feature values of the pixels near the edge of the block and those inside the block; ensuring consistency between these feature values is the key to improving the accuracy of segmentation results. The proposed model uses an edge feature branch and a semantic feature branch called the edge assistant feature network (EFNet). The EFNET model consists of one semantic branch, one edge branch, one shared decoder, and one classifier. The semantic branch extracts semantic features from remote sensing images, whereas the edge branch extracts edge features from remote sensing images and edge images. In addition, the two branches extract five-level features through five sets of feature extraction units. The shared decoder sets up five levels of shared decoding units, which are used to further integrate edge features and deep semantic features. This strategy can reduce the feature differences between the edge pixels and the inner pixels of the object, obtaining a per-pixel feature vector with high inter-class differentiation and intra-class consistency. Softmax is used as the classifier to generate the final segmentation result. We selected a representative winter wheat region in China (Feicheng City) as the study area and established a dataset for experiments. The comparison experiment included three original models and two models modified by adding edge features: SegNet, UNet, and ERFNet, and edge-UNet and edge-ERFNet, respectively. EFNet’s recall (91.01%), intersection over union (81.39%), and F1-Score (91.68%) were superior to those of the other methods. The results clearly show that EFNET improves the accuracy of winter wheat extraction from remote sensing images. This is an important basis not only for crop monitoring, yield estimation, and disaster assessment but also for calculating land carrying capacity and analyzing the comprehensive production capacity of agricultural resources.

Because of its high altitude, the Qinghai-Tibetan Plateau (QTP) serves as a de facto ecological barrier between China and Southeast Asia. Studying the phenological characteristics of vegetation on QTP and their response to climate change can help us to understand how climate change can impact on highland terrestrial ecosystems and how to improve predicting highland vegetation phenology. Using long-term NDVI data we derived from the MODIS spectral reflectance product (MOD09A1) and in situ climate data in QTP, we investigated (1) the trends of temporal and spatial variations in the phenological characteristics of the highland vegetation, such as the start of growing season (SOS), middle time of growing season (MOS), end of growing season (EOS), and length of growing season (LEN), and (2) the responses of SOS, MOS, and EOS to asymmetry in diurnal warming, i.e., asymmetry in the preseason daily daytime maximum temperature (T_max) and daily nighttime minimum temperature (T_min). Results showed that (1) the vegetation phenology in QTP displayed a salient zonal distribution pattern. From the northwest to the southeast, the vegetation SOS and MOS gradually advanced, EOS decreased, and LEN extended; (2) T_min warmed up faster than T_max, which indicated that daytime warming and nighttime warming were asymmetrical; (3) the increase in preseason T_max and that in preseason T_min played different roles in SOS, MOS, and EOS; and (4) the preseason T_min showed a stronger control on vegetation phenology than the preseason T_max. For instance, a 1°C increase in preseason T_min advanced SOS by 7.07d (p  <  0.05) and MOS by 6.80d (p  <  0.05), and delayed EOS by 6.70d (p  <  0.05). While a 1°C increase in preseason T_max delayed the SOS, MOS, and EOS by 5.12d, 4.84d, and 1.04d, respectively.

Terrestrial laser scanning technology has developed rapidly, and substantial data have been accumulated in dynamic forest monitoring. Point cloud data of standing trees not only provide tree parameters but also show three-dimensional tree structure. Selection of the key parameters from the point cloud data is a prerequisite for volume estimation of standing trees. Our study collected three phases of data over 5 years from Liriodendron chinense plantation forest. A series of the height-related characteristic parameters were extracted from the scanned points of each tree stems, including a proposed new parameter and the height cumulative percentage (H_z  %). The upper diameter accuracy obtained by multi-station scanning is high, and the correlation coefficient with manually measured data is 0.9864. The shape of the upper tree trunk extracted by the point cloud is equivalent to that of the sample trees (height of 10 to 20 m) with points at H_25  % and H_50  % of the height. These two parameters also show a high correlation with volume. Results show that H_25  % can better associated with tree volume, with R² at 0.951, 0.957, and 0.901 at three stages, respectively. The volume dynamic change calculated by model 2 is linearly correlated with the rate in point cloud restoration, the intercept is −0.081, and the slope is 1.14. Compared with previous researches, the volume model established based on point cloud hierarchical parameters in this study could be used for monitoring the dynamic volume changes in Liriodendron forest. The H_z value extracted from multi-station scanning point cloud data could be used to represent the dynamic forest structure. The results of this study contribute to further development of terrestrial laser scanning-based modeling and estimation methods for individual tree and forest growth, thereby improving the accuracy of forest inventories estimation and providing better tools for decision-making processes.

A wide variety of swarm intelligence algorithm-based approaches have been recently developed for selecting the near-optimal bands from hyperspectral images (HSIs). However, such methods [including the nondominated sorting genetic algorithm (NSGA)] for HSIs pixel classification are limited by the lack of effective initialization and directional evolution. This research proposed a successive projections algorithm (SPA) and individual repair operation for NSGA (named Sr-NSGA) for band selection. Specifically, Sr-NSGA used the SPA to initialize the population and construct repair sequences that optimize new generated individuals in evolution. Meanwhile, with the guidance of two mutually restricted fitness functions, i.e., average mutual information and classification accuracy, Sr-NSGA searched for the near-optimal band set in an iterative way. Different combinations obtained by SR-NSGA and three effective band selection methods were tested and compared on the Botswana, KSC, and Indian Pines datasets. The results show that Sr-NSGA yielded better performance than the other three methods. Furthermore, support vector machine was used as the classifier for the pixel classification of the Indian Pines dataset to test Sr-NSGA. Experimental result show that Sr-NSGA achieves an overall accuracy of 95.57% and adapts to different classifiers.

The Earth’s surface changes continuously due to several natural and humanmade factors. Efficient change detection (CD) is useful in monitoring and managing different situations. The recent rise in launched hyperspectral platforms provides a diversity of spectrum in addition to the spatial resolution required to meet recent civil applications requirements. Traditional multispectral CD algorithms hardly cope with the complex nature of hyperspectral images and their high dimensionality. To overcome these limitations, a CD deep convolutional neural network (CNN) semantic segmentation-based workflow was proposed. The proposed workflow is composed of four main stages, namely preprocessing, training, testing, and evaluation. Initially, preprocessing is performed to overcome hyperspectral image noise and the high dimensionality problem. Random oversampling (ROS), deep learning, and bagging ensemble were incorporated to handle imbalanced dataset. Also, we evaluated the generality and performance of the original UNet model and four variants of UNet, namely residual UNet, residual recurrent UNet, attention UNet, and attention residual recurrent UNet. Three hyperspectral CD datasets were employed in performance assessment for binary and multiclass change cases; all datasets suffer from class imbalance and small region of interest size. Recurrent residual UNet presented the best performance in both accuracy and inference time. Overall, the obtained results imply that deep CNN segmentation models can be utilized to implement efficient CD for hyperspectral imageries.

Accurate information on the spatial distribution of crops is of great significance for scientific research and production practices. Such accurate information can be extracted from high-spatial-resolution optical remote sensing images. However, acquiring these images with a wide coverage is difficult. We established a model named multispectral super-resolution generative adversarial network (MS_SRGAN) for generating high-resolution 4-m images using Gaofen 1 wide-field-view (WFV) 16-m images. The MS_SRGAN model contains a generator and a discriminator. The generator network is composed of feature extraction units and feature fusion units with a symmetric structure, and the attention mechanism is introduced to constrain the spectral value of the feature map during feature extraction. The generator loss introduces feature loss to describe the feature difference of the image. This is realized using pre-trained discriminator parameters and a partial discriminator network. In addition to realizing feature loss, the discriminator network, which is a simple convolutional neural network, also realizes adversarial loss. Adversarial loss can provide some fake high frequency details to the generator to get a more sharpened image. In the Gaofen 1 WFV image test, the performance of MS_SRGAN was compared with that of Bicubic, EDSR, SRGAN, and ESRGAN. The results show that the spectral angle mapper (3.387) and structural similarity index measure (0.998) of MS_SRGAN are higher than those of the other models. In addition, the image obtained by MS_SRGAN is more realistic; its texture details and color distribution are closer to the reference image to a greater extent.

Cloud detection in satellite images is a vital step for cloud/land recognition, cloud/snow discrimination, and cloud shadow removal. Accurate cloud detection plays an important role in land resource management, environmental pollution monitoring, and land target recognition. Deep learning (DL) algorithms have shown great progress in cloud detection. However, as the complexity of the DL-based model increases, cloud detection efficiency decreases. DL-based cloud detection models are unable to successfully balance the performance-efficiency tradeoff. In our study, a multi-dimensional and multi-grained dense cascade forest (MDForest) is proposed for multi-spectral cloud detection. MDForest is a deep forest structure that automatically extracts low-level and high-level features from satellite cloud images end-to-end; a multi-dimensional and multi-grained scanning mechanism is introduced to capture the spectral information of multi-spectral satellite images while enhancing the representation learning ability of cascade forest. The experimental results on the HJ-1A/1B dataset show that MDForest improves the performance of cloud detection and possesses a good inference efficiency compared with DL-based cloud detection methods, which makes the proposed MDForest satisfy the application where good performance and high efficiency are both required.

This erratum clarifies issues with several figures in the paper.