Search

Detecting and repairing road defects is crucial for road safety, vehicle maintenance, and enhancing tourism on well-maintained roads. However, monitoring all roads by vehicle incurs high costs. With the widespread use of remote sensing technologies, high-resolution satellite images offer a cost-effective alternative. This study proposes a new technique, SDPH, for automated detection of damaged roads from vast, high-resolution satellite images. In the SDPH technique, satellite images are organized in a pyramid grid file system, allowing deep learning methods to efficiently process them. The images, generated as 256 × 256 dimensions, are stored in a directory with explicit location information. The SDPH technique employs a two-stage object detection models, utilizing classical and modified RCNNv3, YOLOv5, and YOLOv8. Classical RCNNv3, YOLOv5, and YOLOv8 and modified RCNNv3, YOLOv5, and YOLOv8 in the first stage for identifying roads, achieving f1 scores of 0.743, 0.716, 0.710, 0.955, 0.958, and 0.954, respectively. When the YOLOv5, with the highest f1 score, was fed to the second stage; modified RCNNv3, YOLOv5, and YOLOv8 detected road defects, achieving f1 scores of 0.957,0.971 and 0.964 in the second process. When the same CNN model was used for road and road defect detection in the proposed SDPH model, classical RCNNv3, improved RCNNv3, classical YOLOv5, improved YOLOv5, classical YOLOv8, improved RCNNv8 achieved micro f1 scores of 0.752, 0.956, 0.726, 0.969, 0.720 and 0.965, respectively. In addition, these models processed 11, 10, 33, 31, 37, and 36 FPS images by performing both stage operations, respectively. Evaluations on geotiff satellite images from Kayseri Metropolitan Municipality, ranging between 20 and 40 gigabytes, demonstrated the efficiency of the SDPH technique. Notably, the modified YOLOv5 outperformed, detecting paths and defects in 0.032 s with the micro f1 score of 0.969. Fine-tuning on TileCache enhanced f1 scores and reduced computational costs across all models. [ABSTRACT FROM AUTHOR]

Deep-learning-based change detection methods have received wide attention, thanks to their strong capability in obtaining rich features from images. However, existing AI-based change detection (CD) methods largely rely on three functionality-enhancing modules, that is, semantic enhancement, attention mechanisms, and correspondence enhancement. The stacking of these modules leads to great model complexity. To unify these three modules into a simple pipeline, we introduce relational change detection transformer (RCDT), a novel and simple framework for remote sensing change detection tasks. The proposed RCDT consists of three major components, a weight-sharing Siamese Backbone to obtain bi-temporal features, a relational cross attention module (RCAM) that implements offset cross attention to obtain bi-temporal relation-aware features, and a features constrain module (FCM) to achieve the final refined predictions with high-resolution constraints. Extensive experiments on four different publicly available datasets suggest that our proposed RCDT exhibits superior change detection performance and great trade-off between performance and complexity compared with other competing methods. The theoretical, methodological, and experimental knowledge of this study is expected to benefit future change detection efforts that involve the cross attention mechanism.

Tree cover is a crucial vegetation structural parameter for simulating ecological, hydrological, and soil erosion processes on the Chinese Loess Plateau, especially after the implementation of the Grain for Green project in 1999. However, current tree cover products performed poorly across most of the Loess Plateau, which is characterized by grasslands with sparse trees. In this study, we first acquired high-accuracy samples of 0.5 m tree canopy and 30 m tree cover using a combination of unmanned aerial vehicle imagery and WorldView-2 (WV-2) imagery. The spectral and textural features derived from Landsat 8 and WV-2 were then used to estimate tree cover with a random forest model. Finally, the tree cover estimated using WV-2, Landsat 8, and their combination were compared, and the optimal tree cover estimates were also compared with current products and tree cover derived from canopy classification. The results show that (1) the normalized difference moisture index using Landsat 8 shortwave infrared and the standard deviation of correlation metric calculated by means of gray-level co-occurrence matrix using the WV-2 near-infrared band are the optimal spectral feature and textural feature for estimating tree cover, respectively. (2) The accuracy of tree cover estimated using only WV-2 is highest (RMSE = 7.44%), indicating that high-resolution textural features are more sensitive to tree cover than the Landsat spectral features (RMSE = 11.53%) on grasslands with sparse trees. (3) Textural features with a resolution higher than 8 m perform better than the combination of Landsat 8 and textural features, and the optimal resolution is 2 m (RMSE = 7.21%) for estimating tree cover, whereas the opposite is observed when the resolution of textural features is lower than 8 m. (4) The current global product seriously underestimates tree cover on the Loess Plateau, and the tree cover calculation using the canopy classification of high-resolution imagery performs worse than the method of directly using remote sensing features. [ABSTRACT FROM AUTHOR]

Modern geoinformation technologies, such as remote sensing satellite missions and classification methods, are becoming increasingly prominent in land cover classification. Due to the emergence of high spatial resolution missions with improved temporal and spectral resolutions, such as Worldview-3, this approach enabled new possibilities in land management. To provide an in-depth analysis of such possibilities, this study reviews methods of land cover classification using WorldView-3 satellite imagery. With 29 different spectral channels and a spatial resolution of 1.2 m, Worldview-3 multispectral satellite images represent the most modern currently publicly available commercial multispectral images. The classification of multispectral images is performed to facilitate the identification and recognition of objects in the images. Analyzed classification methods are: supervised (semi-automatic) classification methods, unsupervised (automatic) classification methods, and object-based classification methods. In order to increase the accuracy in land cover studies, it was determined as necessary to develop automatic methods that rely on a combination of controlled and uncontrolled classification methods. This approach enables the automatic determination of samples for conducting supervised classifications of interest for land management. [ABSTRACT FROM AUTHOR]

Urban land cover classification using high-resolution imagery is important for many applications where detailed and precise urban land cover products are needed. Machine learning algorithms are currently some of the most commonly used methods for classifying high-resolution imagery due to their impressive capabilities. However, the reliability of the land cover products obtained from the classification of high-resolution urban imageries is dependent upon the accuracy of the Machine Learning (ML) classification algorithm used. The need for an appropriate selection of classifiers for urban land cover classification and their applicable settings necessitates the performance comparison of major ML algorithms used for classification. In this study, we compared the performance of three major Machine Learning (ML) classifier algorithms using a high-resolution image dataset of an urban area. The algorithms are Support Vector Machine (SVM), Naïve Bayes, and Ensemble classifiers. The performance of three model types of SVM classifiers namely Medium Gaussian, Linear, and Quadratic SVM, two model types of Naïve Bayes classifiers namely Gaussian and Kernel Naïve Bayes, and three model types of ensemble classifiers namely Bagged Trees, Subspace Discriminant, and RUSBoosted Trees were compared. Performance evaluation was carried out using Confusion Matrix (CM) and Receiver Operating Curves (ROC) plots. Results obtained from the comparison of the three ML classifier algorithms show that the Subspace Discriminant ensemble classifier had the highest accuracy at 85.1%, closely followed by the Medium Gaussian SVM classifier (84.5%) and Gaussian Naïve Bayes classifier (81.5%). This research provides insights into the selection of classifiers for future urban land cover classification and their applicable settings.

Modern geoinformation technologies, such as remote sensing satellite missions and classification methods, are becoming increasingly prominent in land cover classification. Due to the emergence of high spatial resolution missions with improved temporal and spectral resolutions, such as Worldview-3, this approach enabled new possibilities in land management. To provide an in-depth analysis of such possibilities, this study reviews methods of land cover classification using WorldView-3 satellite imagery. With 29 different spectral channels and a spatial resolution of 1.2 m, Worldview-3 multispectral satellite images represent the most modern currently publicly available commercial multispectral images. The classification of multispectral images is performed to facilitate the identification and recognition of objects in the images. Analyzed classification methods are: supervised (semi-automatic) classification methods, unsupervised (automatic) classification methods, and object-based classification methods. In order to increase the accuracy in land cover studies, it was determined as necessary to develop automatic methods that rely on a combination of controlled and uncontrolled classification methods. This approach enables the automatic determination of samples for conducting supervised classifications of interest for land management.

This article develops and examines techniques to enable coherent change detection (CCD) for short-range, high-resolution drone-borne synthetic aperture radar (SAR) systems operating at high frequencies. The potential of using high frequencies at short ranges for fine-resolution imagery and sensitivity to temporal change detection is highlighted, as are the challenges in terms of sensitivity to SAR motion errors. SAR system characteristics for CCD are derived, and the impact of motion errors, which leads to spatial decorrelation and co-registration errors on CCD maps, is discussed. Subsequently, a CCD algorithm able to generate change maps with a >0.75 average coherence value is presented. The validity of the approach is tested through various experimental scenarios. As a result, car tyre marks and human footprints are possible to discern with a drone-borne SAR demonstrator operating at 24 GHz.

Estimating forest biomass is a necessary task to assess the carbon sequestration potential and storage, which serves to evaluate the environmental value of forests. It is possible to develop models for predicting forest biomass by combining remote sensing, field data and statistical models. In this study, we assessed the feasibility of estimating aboveground biomass (AGB) of Rhizophora apiculata plantation forests in Thanh Phu Nature Reserve by combining PlanetScope imagery and field inventory data obtained from 55 sample plots. Furthermore, we investigated the potential to enhance the accuracy of AGB estimation by applying four statistical models, generalized linear model (GLM), generalized exponential model (GEM), support vector machine (SVM) and random forest (RF). Our results showed that incorporating gray—level co-occurrence matrix (GLCM) texture features led to a more robust AGB estimation compared to using only spectral bands or vegetation indices. The RF model achieved the highest accuracy of AGB estimation based on the combination of spectral bands, vegetation indices and the optimum texture features, with an R2 value of 0.77. In addition, the spectral and texture features of the green and near-infrared bands were also important in predicting AGB. Finally, the results indicated that PlanetScope imagery has a great potential for mapping the AGB of R. apiculata plantations in mangrove forests, with relatively good accuracy (e.g., 78.26 and 86.36% for low and high values, respectively). [ABSTRACT FROM AUTHOR]

Estimating the number of trees within a forest stand, i.e., the forest stand density (FSD), is challenging at large scales. Recently, researchers have turned to a combination of remote sensing and machine learning techniques to derive these estimates. However, in most cases, the developed models rely heavily upon additional data such as LiDAR-based elevations or multispectral information and are mostly applied to managed environments rather than natural/mixed forests. Furthermore, they often require the time-consuming manual digitization or masking of target features, or an annotation using a bounding box rather than a simple point annotation. Here, we introduce the Tree Convolutional Row Neural Network (Tree-CRowNN), an alternative model for tree counting inspired by Multiple-Column Neural Network architecture to estimate the FSD over 12.8 m × 12.8 m plots from high-resolution RGB aerial imagery. Our model predicts the FSD with very high accuracy (MAE: ±2.1 stems/12.8 m2, RMSE: 3.0) over a range of forest conditions and shows promise in linking to Sentinel-2 imagery for broad-scale mapping (R2: 0.43, RMSE: 3.9 stems/12.8 m2). We believe that the satellite imagery linkage will be strengthened with future efforts, and transfer learning will enable the Tree-CRowNN model to predict the FSD accurately in other ecozones. [ABSTRACT FROM AUTHOR]

The photovoltaic (PV) industry boom has accelerated the need for accurately understanding the spatial distribution of PV energy systems. The synergy of remote sensing and artificial intelligence presents significant prospects for PV energy monitoring. Currently, numerous studies have focused on extracting rooftop PV systems from airborne or satellite imagery, but their small-scale and size-varying characteristics make the segmentation results suffer from PV internal incompleteness and small PV omission. To address these issues, this study proposed a size-aware deep learning network called Rooftop PV Segmenter (RPS) for segmenting small-scale rooftop PV systems from high-resolution imagery. In detail, the RPS network introduced a Semantic Refinement Module (SRM) to sense size variations of PV panels and reconstruct high-resolution deep semantic features. Moreover, a Feature Aggregation Module (FAM) enhanced the representation of robust features by continuously aggregating deeper features into shallower ones. In the output stage, a Deep Supervised Fusion Module (DSFM) was employed to constrain and fuse the outputs at different scales to achieve more refined segmentation. The proposed RPS network was tested and shown to outperform other models in producing segmentation results closer to the ground truth, with the F1 score and IoU reaching 0.9186 and 0.8495 on the publicly available California Distributed Solar PV Array Dataset (C-DSPV Dataset), and 0.9608 and 0.9246 on the self-annotated Heilbronn Rooftop PV System Dataset (H-RPVS Dataset). This study has provided an effective solution for obtaining a refined small-scale energy distribution database. [ABSTRACT FROM AUTHOR]

The sweet potato is an essential food and economic crop that is often threatened by the devastating sweet potato virus disease (SPVD), especially in developing countries. Traditional laboratory-based direct detection methods and field scouting are commonly used to rapidly detect SPVD. However, these molecular-based methods are costly and disruptive, while field scouting is subjective, labor-intensive, and time-consuming. In this study, we propose a deep learning-based object detection framework to assess the feasibility of detecting SPVD from ground and aerial high-resolution images. We proposed a novel object detector called SPVDet, as well as a lightweight version called SPVDet-Nano, using a single-level feature. These detectors were prototyped based on a small-scale publicly available benchmark dataset (PASCAL VOC 2012) and compared to mainstream feature pyramid object detectors using a leading large-scale publicly available benchmark dataset (MS COCO 2017). The learned model weights from this dataset were then transferred to fine-tune the detectors and directly analyze our self-made SPVD dataset encompassing one category and 1074 objects, incorporating the slicing aided hyper inference (SAHI) technology. The results showed that SPVDet outperformed both its single-level counterparts and several mainstream feature pyramid detectors. Furthermore, the introduction of SAHI techniques significantly improved the detection accuracy of SPVDet by 14% in terms of mean average precision (mAP) in both ground and aerial images, and yielded the best detection accuracy of 78.1% from close-up perspectives. These findings demonstrate the feasibility of detecting SPVD from ground and unmanned aerial vehicle (UAV) high-resolution images using the deep learning-based SPVDet object detector proposed here. They also have great implications for broader applications in high-throughput phenotyping of sweet potatoes under biotic stresses, which could accelerate the screening process for genetic resistance against SPVD in plant breeding and provide timely decision support for production management. [ABSTRACT FROM AUTHOR]

Tree cover is a crucial vegetation structural parameter for simulating ecological, hydrological, and soil erosion processes on the Chinese Loess Plateau, especially after the implementation of the Grain for Green project in 1999. However, current tree cover products performed poorly across most of the Loess Plateau, which is characterized by grasslands with sparse trees. In this study, we first acquired high-accuracy samples of 0.5 m tree canopy and 30 m tree cover using a combination of unmanned aerial vehicle imagery and WorldView-2 (WV-2) imagery. The spectral and textural features derived from Landsat 8 and WV-2 were then used to estimate tree cover with a random forest model. Finally, the tree cover estimated using WV-2, Landsat 8, and their combination were compared, and the optimal tree cover estimates were also compared with current products and tree cover derived from canopy classification. The results show that (1) the normalized difference moisture index using Landsat 8 shortwave infrared and the standard deviation of correlation metric calculated by means of gray-level co-occurrence matrix using the WV-2 near-infrared band are the optimal spectral feature and textural feature for estimating tree cover, respectively. (2) The accuracy of tree cover estimated using only WV-2 is highest (RMSE = 7.44%), indicating that high-resolution textural features are more sensitive to tree cover than the Landsat spectral features (RMSE = 11.53%) on grasslands with sparse trees. (3) Textural features with a resolution higher than 8 m perform better than the combination of Landsat 8 and textural features, and the optimal resolution is 2 m (RMSE = 7.21%) for estimating tree cover, whereas the opposite is observed when the resolution of textural features is lower than 8 m. (4) The current global product seriously underestimates tree cover on the Loess Plateau, and the tree cover calculation using the canopy classification of high-resolution imagery performs worse than the method of directly using remote sensing features.

Soil moisture variability caused by soil erosion, weather extremes, and spatial variations in soil health is a limiting factor for crop growth and productivity. Crop evapotranspiration (ET) is significant for irrigation water management systems. The variability in crop water requirements at various growth stages is a common concern at a global level. In Canada’s Prince Edward Island (PEI), where agriculture is particularly prominent, this concern is predominantly evident. The island’s most prominent business, agriculture, finds it challenging to predict agricultural water needs due to shifting climate extremes, weather patterns, and precipitation patterns. Thus, accurate estimations for irrigation water requirements are essential for water conservation and precision farming. This work used a satellite-based normalized difference vegetation index (NDVI) technique to simulate the crop coefficient (Kc) and crop evapotranspiration (ETc) for field-scale potato cultivation at various crop growth stages for the growing seasons of 2021 and 2022. The standard FAO Penman–Monteith equation was used to estimate the reference evapotranspiration (ETr) using weather data from the nearest weather stations. The findings showed a statistically significant (p < 0.05) positive association between NDVI and tabulated Kc values extracted from all three satellites (Landsat 8, Sentinel-2A, and Planet) for the 2021 season. However, the correlation weakened in the subsequent year, particularly for Sentinel-2A and Planet data, while the association with Landsat 8 data became statistically insignificant (p > 0.05). Sentinel-2A outperformed Landsat 8 and Planet overall. The Kc values peaked at the halfway stage, fell before the maturity period, and were at their lowest at the start of the season. A similar pattern was observed for ETc (mm/day), which peaked at midseason and decreased with each developmental stage of the potato crop. Similar trends were observed for ETc (mm/day), which peaked at the mid-stage with mean values of 4.0 (2021) and 3.7 (2022), was the lowest in the initial phase with mean values of 1.8 (2021) and 1.5 (2022), and grew with each developmental stage of the potato crop. The study’s ET maps show how agricultural water use varies throughout a growing season. Farmers in Prince Edward Island may find the applied technique helpful in creating sustainable growth plans at different phases of crop development. Integrating high-resolution imagery with soil health, yield mapping, and crop growth parameters can help develop a decision support system to tailor sustainable management practices to improve profit margins, crop yield, and quality.

Urban planning within Riyadh, the capital of Saudi Arabia, has been impacted by the presence of informal settlements. An understanding of the spatial distribution of these settlements is essential in developing urban policies. This study used remotely sensed imagery to evaluate and characterize informal settlements within the city, both with and without expert knowledge of the study area (defined as expert knowledge, EK). An informal settlement ontology for four study sites within Riyadh City was developed using an analytical hierarchy process (AHP). Local knowledge was translated into a ruleset to identify and map settlement areas using spatial, spectral, textural, and geometric techniques. These were combined with an object-based image analysis (OBIA) approach. The study demonstrated that combining expert knowledge and remotely sensed data can efficiently and accurately identify informal settlements. Two classified images were produced, one with EK, and one without EK, to investigate how a detailed understanding of local conditions could affect the final image classification. Overall accuracy when using EK was 94%, with a kappa coefficient of 89%, while without EK accuracy was 68% (kappa coefficient of 61%). The final OBIA classes included formal and informal settlements, road networks, vacant blocks, shaded areas, and vegetation. This study demonstrated that local expert knowledge and OBIA helpful in urban mapping. It also indicated the value of integrating a local ontological process during digital image classification. This work provided improved techniques for mapping informal settlements in Middle Eastern cities. [ABSTRACT FROM AUTHOR]

Spartina alterniflora (smooth cordgrass), China's most common invasive species, has posed significant challenges to native plant communities and coastal environments. Monitoring the invasion and dieback process of S. alterniflora by multisource high-resolution imagery is necessary to manage the invasion of the species. Current spatial analyses, however, are insufficient. As a result, we first extracted S. alterniflora by integrating multisource high-resolution images through the multiscale object-oriented classification method, then identified the expansion patterns of S. alterniflora on the seaward side by the landscape expansion index, and conformed the main drivers of S. alterniflora dieback on the landward side in the Jiangsu Dafeng Milu National Nature Reserve. The findings revealed that the area of S. alterniflora decreased in size from 1511.26 ha in 2010 to 910.25 ha in 2020. S. alterniflora continues to grow to the sea and along the tidal creek on the seaward side, with a total increase of 159.13 ha. External isolation expansion patterns accounted for 65.16% of the total expansion patches, with marginal expansion patches accounting for 24.22% and tidal creek-leading expansion patches accounting for 10.62%. While the landward side showed a declining trend, the total area decreased by 852.36 ha, with an annual average change rate of 8.67%. S. alterniflora dieback was negatively related to the number of tidal creeks and positively related to the number of wild Elaphures davidianus and the length of artificial ditches. Our findings provide a scientific foundation for the ecological control of S. alterniflora. Its presence in coastal wetlands inspires evidence-based protection and management strategies to protect the coastal wetland ecosystem. [ABSTRACT FROM AUTHOR]

Pixel-level segmentation with deep learning is widely applied in interpretation tasks of high-resolution remote sensing imagery, such as change detection and object extraction. However, most existing methods focus on designing deep learning model structures, but few consider improving segmentation accuracy in the inference stage with a trained model. The most important advantage of improving model accuracy in the inference stage is that it does not need to retrain models but can improve the detection accuracy by a cost-effective means. In this letter, a novel decision-level fusion method based on the Dempster–Shafer (DS) theory was proposed, namely, DeepDSFusion. As a general method, it can be seamlessly integrated into any other pixel-level segmentation model. In the implementation detail of the DeepDSFusion, first, several classical data augmentation methods, such as rotation transform and scale transform, were adopted to acquire multiscale probability maps. Then, the DS theory was used to fuse multiscale probability maps into a single probability map. Finally, a simple threshold is applied in the single probability map to acquire segmentation results. Three classical pixel-level segmentation tasks, deforestation detection, road extraction, and landcover mapping on high-resolution imagery, prove the effectiveness of DeepDSFusion. [ABSTRACT FROM AUTHOR]

Remote sensing data play a crucial role in precision agriculture and natural resource monitoring. The use of unmanned aerial vehicles (UAVs) can provide solutions to challenges faced by farmers and natural resource managers due to its high spatial resolution and flexibility compared to satellite remote sensing. This paper presents UAV and spectral datasets collected from different provinces in South Africa, covering different crops at the farm level as well as natural resources. UAV datasets consist of five multispectral bands corrected for atmospheric effects using the PIX4D mapper software to produce surface reflectance images. The spectral datasets are filtered using a Savitzky–Golay filter, corrected for Multiplicative Scatter Correction (MSC). The first and second derivatives and the Continuous Wavelet Transform (CWT) spectra are also calculated. These datasets can provide baseline information for developing solutions for precision agriculture and natural resource challenges. For example, UAV and spectral data of different crop fields captured at spatial and temporal resolutions can contribute towards calibrating satellite images, thus improving the accuracy of the derived satellite products. Dataset: Available on request to munghemezuluc@arc.agric.za. The ARC (Agricultural Research Council) is in the process of developing an interface that will allow users to have access to the data directly from our server in line with our policies on data sharing. Therefore, in the meantime, users can request the data. Dataset License: CC BY-NC 4.0 [ABSTRACT FROM AUTHOR]

Proliferation of plastic-covered greenhouse (PCG) farming areas has been unparalleled in recent decades around the globe. Therefore, PCG's proper mapping and continuous monitoring are required to maintain protected farming sustainability and to reduce plastic pollution. Topography is one of the key factors affecting PCG mapping accuracy from satellite imagery. In this research, in order to understand the influence of elevation on PCG mapping accuracy in a Moroccan (with altitude below 200 m) and Vietnamese (an average altitude of 1,500 m) area, five supervised algorithms were applied to high-resolution PlanetScope and RapidEye imagery: maximum likelihood classification (MLC), k-nearest neighbours (k-NN), random forest (RF), support vector machine with linear kernel (SVML) and radial basis kernel (SVMRB). The findings indicated that k-NN (SVMRB) produced the greatest overall accuracy, followed by SVMRB (RF) for Dalat City (Loukkos). The present study also found that RapidEye outperforms PlanetScope in both areas and independently of the machine learning algorithm. Elevation data inclusion significantly improves (1.6% to 6.8%) various satellites' performance, particularly in Dalat City. Thus, this study demonstrated that topography has an impact on PCG mapping, especially in hilly areas. [ABSTRACT FROM AUTHOR]

The tilted viewing nature of the off-nadir aerial images brings severe challenges to the building change detection (BCD) problem: the mismatch of the nearby buildings and the semantic ambiguity of the building facades. To tackle these challenges, we present a multi-task guided change detection network model, named as MTGCD-Net. The proposed model approaches the specific BCD problem by designing three auxiliary tasks, including: (1) a pixel-wise classification task to predict the roofs and facades of buildings; (2) an auxiliary task for learning the roof-to-footprint offsets of each building to account for the misalignment between building roof instances; and (3) an auxiliary task for learning the identical roof matching flow between bi-temporal aerial images to tackle the building roof mismatch problem. These auxiliary tasks provide indispensable and complementary building parsing and matching information. The predictions of the auxiliary tasks are finally fused to the main BCD branch with a multi-modal distillation module. To train and test models for the BCD problem with off-nadir aerial images, we create a new benchmark dataset, named BANDON. Extensive experiments demonstrate that our model achieves superior performance over the previous state-of-the-art competitors. [ABSTRACT FROM AUTHOR]

Drones are an increasingly popular choice for wildlife surveys due to their versatility, quick response capabilities, and ability to access remote areas while covering large regions. A novel application presented here is to combine drone imagery with neural networks to assess mortality within a bird colony. Since 2021, Highly Pathogenic Avian Influenza (HPAI) has caused significant bird mortality in the UK, mainly affecting aquatic bird species. The world’s largest northern gannet colony on Scotland’s Bass Rock experienced substantial losses in 2022 due to the outbreak. To assess the impact, RGB imagery of Bass Rock was acquired in both 2022 and 2023 by deploying a drone over the island for the first time. A deep learning neural network was subsequently applied to the data to automatically detect and count live and dead gannets, providing population estimates for both years. The model was trained on the 2022 dataset and achieved a mean average precision (mAP) of 37%. Application of the model predicted 18,220 live and 3761 dead gannets for 2022, consistent with NatureScot’s manual count of 21,277 live and 5035 dead gannets. For 2023, the model predicted 48,455 live and 43 dead gannets, and the manual count carried out by the Scottish Seabird Centre and UK Centre for Ecology and Hydrology (UKCEH) of the same area gave 51,428 live and 23 dead gannets. This marks a promising start to the colony’s recovery with a population increase of 166% determined by the model. The results presented here are the first known application of deep learning to detect dead birds from drone imagery, showcasing the methodology’s swift and adaptable nature to not only provide ongoing monitoring of seabird colonies and other wildlife species but also to conduct mortality assessments. As such, it could prove to be a valuable tool for conservation purposes.

The proliferation of plastic-covered greenhouse (PCG) farming has resulted in high horticulture crop yields worldwide during the last few decades. A proper and cost-effective PCG monitoring method is necessary for maintaining sustainable horticulture and high-quality agricultural production with less plastic pollution. Remote sensing applications for mapping PCG have received great attention from the scientific community in recent years. In this paper, a comparative study was carried out in two plastic-covered greenhouse areas in Loukkos perimeter in Morocco and Dalat City in Vietnam to test PCG mapping accuracy of high spatial resolution RapidEye and PlanetScope satellite data and to understand the differences in PCG mapping quality due to topographic effects. Medium-resolution Landsat-8 OLI and Sentinel-2 MSI imagery were also applied. Moreover, two classification algorithms—retrogressive plastic greenhouse index (RPGI) and a supervised classification algorithm using random forest (RF)—were used for mapping PCG. The findings reveal that RF outperforms RPGI. Overall, the mapping accuracy achieved exceeded 90% in both study areas, except for the RPGI method using Landsat-8 data (PCG mapping accuracy using Landsat data varied between 87.4 and 89%). Furthermore, PCGs were better detected by PlanetScope data than by RapidEye imagery due to the differences in the spectral range. Better performance in Loukkos perimeter can be explained by the study area's topography; Dalat City and surrounding areas are situated in mountainous terrain. The results obtained from this study indicate that spectral indices can be used as a cost-effective tool for mapping PCG under cloud-free conditions. PCG mapping using RF classifiers resulted in accurate PCG mapping without topographic factors' influence. [ABSTRACT FROM AUTHOR]

Background: Fire management is a crucial part of managing ecosystems. The years since last burn (YSLB) metric is commonly used in fire planning to predict when an area might be suitable to burn; however, this metric fails to account for variable recovery due to climate variability. Aim: The aim of this study was to develop a predictor of when an area may be able to 'carry' fire based on observed patterns of vegetation recovery and fire occurrence that is responsive to climate variability. Methods: Fire history maps and Landsat satellite imagery within the Great Victoria Desert of Australia were used to map vegetation recovery following fire. Burn potential models were then created by calculating the distributions of YSLB and vegetation recovery values for areas that subsequently burnt. Key result: A burn potential model based on vegetation recovery is a better predictor of when an area is likely to burn than a model based on YSLB. Conclusions: A burn potential model based on vegetation recovery provides an evidence-based and dynamic assessment of whether an area is likely to burn. Implications: This approach provides a model that is responsive to climate variability that can assist fire managers in burn planning and assessing fire risk. Fire management is a crucial part of managing many environments. We present a burn potential model developed using satellite imagery that predicts when an area may be able to 'carry' fire. The model accounts for rapid or delayed vegetation cover response since the last burn due to climate variability. [ABSTRACT FROM AUTHOR]

The Bienville National Forest (BNF) in central Mississippi experienced an outbreak of southern pine beetle (SPB) beginning in 2015 and continuing through 2019. To assess the extent of the outbreak and subsequent treatments of impacted areas, high-resolution imagery was obtained from various sources and interpreted to determine the feasibility of this imagery for detecting SPB spots and tracking their spread and treatments. A negative binomial regression model then described the relationship between spot detection, year, and status (i.e., infestation/treatment) and then incidence rate ratios were calculated. The peak active infestation acreage occurred in 2017 with treatment and mitigation measures peaking in 2018. In total, over 4450 hectares (ha) were treated through 2019 in the BNF. Overall, it is possible to discern small areas of active SPB infestation and treatments. If used as a monitoring method, consistently available high-resolution imagery (e.g., from a satellite) provides an effective means of detecting, evaluating, and tracking infestations and related treatments. [ABSTRACT FROM AUTHOR]

The photovoltaic (PV) industry boom has accelerated the need for accurately understanding the spatial distribution of PV energy systems. The synergy of remote sensing and artificial intelligence presents significant prospects for PV energy monitoring. Currently, numerous studies have focused on extracting rooftop PV systems from airborne or satellite imagery, but their small-scale and size-varying characteristics make the segmentation results suffer from PV internal incompleteness and small PV omission. To address these issues, this study proposed a size-aware deep learning network called Rooftop PV Segmenter (RPS) for segmenting small-scale rooftop PV systems from high-resolution imagery. In detail, the RPS network introduced a Semantic Refinement Module (SRM) to sense size variations of PV panels and reconstruct high-resolution deep semantic features. Moreover, a Feature Aggregation Module (FAM) enhanced the representation of robust features by continuously aggregating deeper features into shallower ones. In the output stage, a Deep Supervised Fusion Module (DSFM) was employed to constrain and fuse the outputs at different scales to achieve more refined segmentation. The proposed RPS network was tested and shown to outperform other models in producing segmentation results closer to the ground truth, with the F1 score and IoU reaching 0.9186 and 0.8495 on the publicly available California Distributed Solar PV Array Dataset (C-DSPV Dataset), and 0.9608 and 0.9246 on the self-annotated Heilbronn Rooftop PV System Dataset (H-RPVS Dataset). This study has provided an effective solution for obtaining a refined small-scale energy distribution database.

The sweet potato is an essential food and economic crop that is often threatened by the devastating sweet potato virus disease (SPVD), especially in developing countries. Traditional laboratory-based direct detection methods and field scouting are commonly used to rapidly detect SPVD. However, these molecular-based methods are costly and disruptive, while field scouting is subjective, labor-intensive, and time-consuming. In this study, we propose a deep learning-based object detection framework to assess the feasibility of detecting SPVD from ground and aerial high-resolution images. We proposed a novel object detector called SPVDet, as well as a lightweight version called SPVDet-Nano, using a single-level feature. These detectors were prototyped based on a small-scale publicly available benchmark dataset (PASCAL VOC 2012) and compared to mainstream feature pyramid object detectors using a leading large-scale publicly available benchmark dataset (MS COCO 2017). The learned model weights from this dataset were then transferred to fine-tune the detectors and directly analyze our self-made SPVD dataset encompassing one category and 1074 objects, incorporating the slicing aided hyper inference (SAHI) technology. The results showed that SPVDet outperformed both its single-level counterparts and several mainstream feature pyramid detectors. Furthermore, the introduction of SAHI techniques significantly improved the detection accuracy of SPVDet by 14% in terms of mean average precision (mAP) in both ground and aerial images, and yielded the best detection accuracy of 78.1% from close-up perspectives. These findings demonstrate the feasibility of detecting SPVD from ground and unmanned aerial vehicle (UAV) high-resolution images using the deep learning-based SPVDet object detector proposed here. They also have great implications for broader applications in high-throughput phenotyping of sweet potatoes under biotic stresses, which could accelerate the screening process for genetic resistance against SPVD in plant breeding and provide timely decision support for production management.

Estimating the number of trees within a forest stand, i.e., the forest stand density (FSD), is challenging at large scales. Recently, researchers have turned to a combination of remote sensing and machine learning techniques to derive these estimates. However, in most cases, the developed models rely heavily upon additional data such as LiDAR-based elevations or multispectral information and are mostly applied to managed environments rather than natural/mixed forests. Furthermore, they often require the time-consuming manual digitization or masking of target features, or an annotation using a bounding box rather than a simple point annotation. Here, we introduce the Tree Convolutional Row Neural Network (Tree-CRowNN), an alternative model for tree counting inspired by Multiple-Column Neural Network architecture to estimate the FSD over 12.8 m × 12.8 m plots from high-resolution RGB aerial imagery. Our model predicts the FSD with very high accuracy (MAE: ±2.1 stems/12.8 m2, RMSE: 3.0) over a range of forest conditions and shows promise in linking to Sentinel-2 imagery for broad-scale mapping (R2: 0.43, RMSE: 3.9 stems/12.8 m2). We believe that the satellite imagery linkage will be strengthened with future efforts, and transfer learning will enable the Tree-CRowNN model to predict the FSD accurately in other ecozones.

Digital elevation models (DEMs) provide fundamental data for scientific and engineering applications in lunar exploration missions. Lunar DEMs have been mainly generated by laser altimetry and stereophotogrammetry. Complementarity to stereo photogrammetry, reflection-based surface reconstruction methods, such as shape from shading (SFS), have been studied and applied in lunar DEM reconstruction from a single image. However, this method often suffers from solution ambiguity and instability. In this paper, we propose a generative adversarial network (GAN)-based method that is able to generate high-resolution pixel-scale DEMs from a single image aided by a low-resolution DEM. We have evaluated the accuracy of the reconstructed high-resolution DEMs from 25 LROC NAC images of four regions using LROC NAC DEMs (2 m/pixel) as ground truth. The experimental results demonstrate good accuracy and adaptability to changes in illumination conditions. The root mean square error (RMSE) can reach about 2 m in areas where the degree of elevation variation is less than 100 m, and the RMSE value ranges from around 3 m to 10 m without considering the degree of the elevation variation in large-area reconstruction. As high-resolution monocular images and low-resolution DEMs are available for the entire lunar surface, the proposed GAN-based method has great potential in high-resolution lunar DEM reconstruction for lunar mapping applications. [ABSTRACT FROM AUTHOR]

Spartina alterniflora (smooth cordgrass), China’s most common invasive species, has posed significant challenges to native plant communities and coastal environments. Monitoring the invasion and dieback process of S. alterniflora by multisource high-resolution imagery is necessary to manage the invasion of the species. Current spatial analyses, however, are insufficient. As a result, we first extracted S. alterniflora by integrating multisource high-resolution images through the multiscale object-oriented classification method, then identified the expansion patterns of S. alterniflora on the seaward side by the landscape expansion index, and conformed the main drivers of S. alterniflora dieback on the landward side in the Jiangsu Dafeng Milu National Nature Reserve. The findings revealed that the area of S. alterniflora decreased in size from 1511.26 ha in 2010 to 910.25 ha in 2020. S. alterniflora continues to grow to the sea and along the tidal creek on the seaward side, with a total increase of 159.13 ha. External isolation expansion patterns accounted for 65.16% of the total expansion patches, with marginal expansion patches accounting for 24.22% and tidal creek-leading expansion patches accounting for 10.62%. While the landward side showed a declining trend, the total area decreased by 852.36 ha, with an annual average change rate of 8.67%. S. alterniflora dieback was negatively related to the number of tidal creeks and positively related to the number of wild Elaphures davidianus and the length of artificial ditches. Our findings provide a scientific foundation for the ecological control of S. alterniflora. Its presence in coastal wetlands inspires evidence-based protection and management strategies to protect the coastal wetland ecosystem.

Urban planning within Riyadh, the capital of Saudi Arabia, has been impacted by the presence of informal settlements. An understanding of the spatial distribution of these settlements is essential in developing urban policies. This study used remotely sensed imagery to evaluate and characterize informal settlements within the city, both with and without expert knowledge of the study area (defined as expert knowledge, EK). An informal settlement ontology for four study sites within Riyadh City was developed using an analytical hierarchy process (AHP). Local knowledge was translated into a ruleset to identify and map settlement areas using spatial, spectral, textural, and geometric techniques. These were combined with an object-based image analysis (OBIA) approach. The study demonstrated that combining expert knowledge and remotely sensed data can efficiently and accurately identify informal settlements. Two classified images were produced, one with EK, and one without EK, to investigate how a detailed understanding of local conditions could affect the final image classification. Overall accuracy when using EK was 94%, with a kappa coefficient of 89%, while without EK accuracy was 68% (kappa coefficient of 61%). The final OBIA classes included formal and informal settlements, road networks, vacant blocks, shaded areas, and vegetation. This study demonstrated that local expert knowledge and OBIA helpful in urban mapping. It also indicated the value of integrating a local ontological process during digital image classification. This work provided improved techniques for mapping informal settlements in Middle Eastern cities.

The increasing development of imaging technology has made aerial image analysis one of the most widely used fields in image processing. Building extraction is the basic step in analysing urban structures, detecting construction violations, updating urban geographical divisions, and forecasting natural disasters. In this study, the aim is to automatically segment buildings in high-resolution satellite images using a new hybrid deep learning model, named Gated Residual Supervision Network (GRSNet). GRSNet extends the UNet framework by including three important components, i.e. attention gates (AG), residual units, and deep supervision part, with the main focus on transferring and reusing features. At first, the AG mechanism utilizes channel and spatial fine features to merge them effectively and deep supervision transfers feature details from the depths of the network. Then, residual units retrieve the information at different levels to train the model quickly. Finally, a fully connected classifier recovers features from the input image. GRSNet is evaluated on Massachusetts buildings public dataset and Inria aerial image labeling benchmark. The obtained results show the superiority of the proposed method compared to other deep learning-based state-of-the-art building segmentation methods with an intersection over union (IOU) of 89.86% and an F1-score of 94.53%. [ABSTRACT FROM AUTHOR]

Landslide extraction is one of the most popular topics in remote sensing. Numerous techniques have been proposed to manage the landslide identification problem. However, most aim to extract landslides that have already occurred or delineate the potential landslide manually. It is greatly important to identify and delineate potential landslides automatically, which has not been investigated. In this paper, we propose an automatic identification and delineation method, i.e., object-based image analysis (OBIA) of potential landslides by integrating optical imagery with a deformation map. We applied a deformation map generated by the interferometric synthetic aperture radar (InSAR) technique, rather than the digital elevation model (DEM) for landslide segmentation. Then, we used a classification and regression tree (CART) model with the spectral, spatial, contextual and deformation characteristics for landslide classification. For accuracy assessment, we implemented the evaluation indicators of recall and precision. The proposed method is verified in both specific landslide-prone regions (Jinpingzi and Shuanglongtan landslides) and a large catchment of the Jinsha River, China. By comparing our results with the ones using purely optical imagery, the precision of the Jinpingzi landslide is improved by 14.12%, and the recall and precision of the Shuanglongtan landslide are improved by 3.1% and 3.6%, respectively, and the recall for the large catchment is improved by 9.95%. Our method can improve delineation of potential landslides significantly, which is crucial for landslide early warning and prevention. [ABSTRACT FROM AUTHOR]

Remote sensing data play a crucial role in precision agriculture and natural resource monitoring. The use of unmanned aerial vehicles (UAVs) can provide solutions to challenges faced by farmers and natural resource managers due to its high spatial resolution and flexibility compared to satellite remote sensing. This paper presents UAV and spectral datasets collected from different provinces in South Africa, covering different crops at the farm level as well as natural resources. UAV datasets consist of five multispectral bands corrected for atmospheric effects using the PIX4D mapper software to produce surface reflectance images. The spectral datasets are filtered using a Savitzky–Golay filter, corrected for Multiplicative Scatter Correction (MSC). The first and second derivatives and the Continuous Wavelet Transform (CWT) spectra are also calculated. These datasets can provide baseline information for developing solutions for precision agriculture and natural resource challenges. For example, UAV and spectral data of different crop fields captured at spatial and temporal resolutions can contribute towards calibrating satellite images, thus improving the accuracy of the derived satellite products.

Context: Studies of tree cover and spatial pattern in tropical savannas often focus on coarse-scale effects of climate, disturbance, and herbivory. Local hydrological, topographic, or edaphic factors, however, play an important role in determining resource availability for trees, but we often lack data for these variables at fine spatial resolution. River networks can serve as a proxy for some of these effects. Objectives: Our objective was to characterize the relationship between distance to rivers and tree cover and spatial pattern, and to provide a first-order assessment of the total fraction of tree cover associated with river networks in a savanna ecosystem. Methods: We explored the relationship between tree cover amount and patchiness as a function of distance to rivers in Serengeti National Park, Tanzania. We used segmented regression to identify the proportion of trees most likely to be influenced by river-associated resources. Results: Both tree cover and tree patchiness declined systematically with distance to rivers, with tree cover declining up to about 100 m from river beds. Assuming that trees closer than this threshold are influenced by rivers (e.g., through greater access to groundwater), between a third and half of all Serengeti trees may be classifiable as 'river-associated trees.' Conclusions: Our results suggest that river networks may play a greater role in determining the amount and spatial pattern of tree cover in savanna ecosystems than previously considered, with potentially important consequences for savanna dynamics, and perhaps drought resilience. [ABSTRACT FROM AUTHOR]

Automatic tree crown segmentation from remote sensing data is especially challenging in dense, diverse, and multilayered tropical forest canopies, and tracking mortality by this approach is even more difficult. Here, we examine the potential for combining airborne laser scanning (ALS) with multispectral and hyperspectral data to improve the accuracy of tree crown segmentation at a study site in French Guiana. We combined an ALS point cloud clustering method with a spectral deep learning model to achieve 83% accuracy at recognizing manually segmented reference crowns (with congruence >0.5). This method outperformed a two-step process that involved clustering the ALS point cloud and then using the logistic regression of hyperspectral distances to correct oversegmentation. We used this approach to map tree mortality from repeat surveys and show that the number of crowns identified in the first that intersected with height loss clusters was a good estimator of the number of dead trees in these areas. Our results demonstrate that multisensor data fusion improves the automatic segmentation of individual tree crowns and presents a promising avenue to study forest demography with repeated remote sensing acquisitions.

To solve some problems of high spatial resolution remote sensing images caused by land coverage, building coverage and shading of trees, such as difficult road extraction and low precision, a road extraction method based on multi-task key point constraints is put forward in this article based on Linknet. At the preprocessing stage, an auxiliary constraint task is designed to solve the connectivity problem caused by shading during road extraction from remote sensing images. At the encoding & decoding stage, first, a position attention (PA) mechanism module and channel attention (CA) mechanism module are applied to realize the effective fusion of semantic information in the context during road extraction. Second, a multi-branch cascade dilated spatial pyramid (CDSP) is established with dilated convolution, by which the problem of loss of partial information during information extraction from remote sensing road image is solved and the detection accuracy is further improved. The method put forward in this article is verified through the experiment with public datasets and private datasets, revealing that the proposed method provides better performance than several state of the art techniques in terms of detection accuracy, recall, precision, and F1-score.

High-spatial resolution urban scenes, which contain ecological, social, and environmental zones, are composed of diverse functional areas. Depicting urban scenes from both qualitative and quantitative aspects, i.e., scene classification and scene unmixing, are increasingly important tasks. The purpose of scene classification is to mark a scene as a separate category, and this approach is widely applied to qualitatively define scenes. However, with the rapid development of cities, urban scenes usually show a mixed nature, resulting in urban mixed scenes (UMSs). Traditional scene classification cannot describe the mix of urban scenes. Few studies have focused on the quantitative measurements of UMSs, which aim at discovering and analyzing the proportion of different subscenes. In this article, a scene unmixing framework based on nonnegative matrix factorization for UMSs (UnUMS) is proposed to estimate the mixing ratio of urban scenes. In the UnUMS, a deep feature representation is used for the high-dimensional representation of the image, and then a nonnegative matrix factorization strategy is used to estimate the proportion of UMSs. Finally, the UMS can be described by the proportion of each subscene. To confirm the robust generalizability of UnUMS, the WH-center Mix dataset with a set of small mixed scenes and the WH-hy LMix dataset with a large UMS image are utilized. The root-mean-square error and the newly proposed vector angular distance are utilized to quantitatively evaluate the effect of scene unmixing from different aspects. Experiments with the two datasets demonstrate that UnUMS can accurately estimate the proportions of different scenes for UMS.

Tracking progress in poverty alleviation and promptly identifying the distribution of poor areas are critical for strategic policy interventions, especially for regions with poor statistical systems. The massive satellite imagery and geospatial data provide great opportunities for timely and cost-effective socioeconomic evaluations. However, existing research on poverty identification is mostly based on satellite images, and the potential of combined multi-source geospatial data on poverty identification has not been fully explored. Here, we propose an approach that evaluates how village-level poverty can be identified by integrating high-resolution imagery (HRI), point-of-interest (POI), OpenStreetMap (OSM), and digital surface model (DSM) data. The study area included 338 villages from Yunyang County, located in Hubei Province, central China. We extracted the explanatory variables indicating access to facilities and services, agricultural production conditions, village construction, and the spatial distribution of village settlements from the HRI, POI, OSM, and DSM data. The random forest algorithm was then used to model the relationship between village-level poverty and explanatory variables. The results demonstrated a 54% accuracy in the prediction of village-level poverty; the best prediction performance (72%) was observed for the villages categorized as poor. The built-up land proportion and the time cost to the facilities and services contributed the most to the identification of village-level poverty, while the proxy variables of agricultural production conditions contributed the least. This study provides an approach to village-level poverty identification using satellite imagery and geospatial data and proves that the data employed in this study could identify the poorest areas that are highly coupled with natural geographical conditions and backward public services.

Urban green space (UGS) is directly or indirectly related to the human well-being of urban inhabitants, information on the availability or quantity of UGS is thus very fundamental for policy-makers to conduct sustainable land management. Analysis of UGS patterns and their influencing factors at large scales using high-resolution remotely sensed imagery is still understudied. Our study aimed to map the spatial patterns of UGS coverage (UGSC) in all (i.e., 3,535) urban areas of the contiguous US (CONUS) and uncover the main influencing factors that dominate the spatial patterns. We mapped the UGS cover of each urban area using the one-meter high-resolution remote sensing images provided by the National Agriculture Imagery Program (NAIP) of the US on the Google Earth Engine platform. Then we calculated the UGSC of each urban area and quantified the spatial patterns of UGSC for the CONUS urban areas. We established a random forests model to quantify the impact of the influencing factors on UGSC. The results showed that: (1) UGSC in the CONUS urban areas varied largely from 2.2% in Kayenta, AZ to 89.36% in Ocala Estates, FL, with a mean UGSC of 39.43% (SD = 18.19%); (2) UGSC in humid Eastern US was much higher than that in urban areas with hyper-arid or arid climate classes in Western or Central regions of the US. Yet, UGSC of urban areas with different city sizes dose not vary largely; (3) the climatic factors were the main influencing factors that dominate the spatial patterns of UGSC in urban areas of the CONUS, while the socio-economic and terrain factors play relatively less important roles in shaping the UGSC pattern.

This article develops and examines methods for the production of real-world, very high-resolution imagery using a high-frequency drone-borne synthetic aperture radar (SAR) operating at short ranges. The significance of motion errors which lead to space-invariant/variant phase errors is discussed. Subsequently, an imaging algorithm capable of handling these errors is proposed and presented. The validity of the approach is tested through both simulation and experiment. We present novel short-range, fine-resolution imagery (less than 2 cm in cross-range) of an extended target area generated using a low-cost, drone-borne vehicular frequency-modulated continuous-wave (FMCW) radar operating at 77 GHz, without using a dedicated inertial navigation system (INS) or Global Positioning System (GPS). [ABSTRACT FROM AUTHOR]

High-resolution land cover mapping over large areas is a challenging task due to the lack of high-quality labels. A potential solution is to leverage the existing knowledge contained in the freely available lower-resolution land cover products. However, the relatively low resolution and low accuracy of the products lead to numerous inaccurate labels, which harms the performance of the neural network. This article addresses the challenge by jointly optimizing the network parameters and correcting the noisy labels with a novel online noise correction approach and a synergistic noise correction loss. By incorporating the information entropy as a measurement to determine the probable correct labels, the proposed noise correction approach learns to make effective correction of the noisy labels during training and eventually boosts the performance with a training set containing less noisy labels. Experimental results show that the proposed method can effectively correct the noisy labels and reduce their negative impact on network training. By employing the proposed method, we produce a refined high-resolution (3-m) land cover map from a lower-resolution (10-m) product in China and improve the accuracy from 74.96% (10-m) to 81.32% (3-m). Such an approach that can effectively learn from noisy data sets leads to many potential opportunities for using and magnifying existing knowledge and results. [ABSTRACT FROM AUTHOR]

Abstract Background Cities are social-ecological systems characterized by remarkably high spatial and temporal heterogeneity, which are closely related to myriad urban problems. However, the tools to map and quantify this heterogeneity are lacking. We here developed a new three-level classification scheme, by considering ecosystem types (level 1), urban function zones (level 2), and land cover elements (level 3), to map and quantify the hierarchical spatial heterogeneity of urban landscapes. Methods We applied the scheme using an object-based approach for classification using very high spatial resolution imagery and a vector layer of building location and characteristics. We used a top-down classification procedure by conducting the classification in the order of ecosystem types, function zones, and land cover elements. The classification of the lower level was based on the results of the higher level. We used an object-based methodology to carry out the three-level classification. Results We found that the urban ecosystem type accounted for 45.3% of the land within the Shenzhen city administrative boundary. Within the urban ecosystem type, residential and industrial zones were the main zones, accounting for 38.4% and 33.8%, respectively. Tree canopy was the dominant element in Shenzhen city, accounting for 55.6% over all ecosystem types, which includes agricultural and forest. However, in the urban ecosystem type, the proportion of tree canopy was only 22.6% because most trees were distributed in the forest ecosystem type. The proportion of trees was 23.2% in industrial zones, 2.2% higher than that in residential zones. That information “hidden” in the usual statistical summaries scaled to the entire administrative unit of Shenzhen has great potential for improving urban management. Conclusions This paper has taken the theoretical understanding of urban spatial heterogeneity and used it to generate a classification scheme that exploits remotely sensed imagery, infrastructural data available at a municipal level, and object-based spatial analysis. For effective planning and management, the hierarchical levels of landscape classification (level 1), the analysis of use and cover by urban zones (level 2), and the fundamental elements of land cover (level 3), each exposes different respects relevant to city plans and management.

A key component in constructing a broad-scale, gridded population dataset is fine resolution geospatial data accurately depicting the extent of human activity. Analogous datasets are often developed using a wide range of methods and classification techniques, including the use of spatial features, spectral features, or the coupling of both to identify the presence of man-made structures from high-resolution satellite imagery. By using spatial and textural-based descriptors to generate high-resolution settlement layers for two dissimilar regions at the peak of seasonal disparity, this study attempts to quantify the influence of seasonality on the accuracy of a supervised, multi-scale, feature extraction framework for automated delineation of human settlement. Results generated by numerous models are evaluated against a reference dataset allowing for assessment of seasonal and feature differences in the context of accuracy. Global or regional mapping of human settlement requires the assemblage of high-resolution satellite images with variegated acquisition characteristics (season, sun elevation, off-nadir, etc.) to produce a cloud-free composite image from which features are extracted. Results of this study suggest an emphasis on imagery criteria, in particular acquisition date, could improve classification accuracy when mapping human settlement at scale.

The Bienville National Forest (BNF) in central Mississippi experienced an outbreak of southern pine beetle (SPB) beginning in 2015 and continuing through 2019. To assess the extent of the outbreak and subsequent treatments of impacted areas, high-resolution imagery was obtained from various sources and interpreted to determine the feasibility of this imagery for detecting SPB spots and tracking their spread and treatments. A negative binomial regression model then described the relationship between spot detection, year, and status (i.e., infestation/treatment) and then incidence rate ratios were calculated. The peak active infestation acreage occurred in 2017 with treatment and mitigation measures peaking in 2018. In total, over 4450 hectares (ha) were treated through 2019 in the BNF. Overall, it is possible to discern small areas of active SPB infestation and treatments. If used as a monitoring method, consistently available high-resolution imagery (e.g., from a satellite) provides an effective means of detecting, evaluating, and tracking infestations and related treatments.

Digital elevation models (DEMs) provide fundamental data for scientific and engineering applications in lunar exploration missions. Lunar DEMs have been mainly generated by laser altimetry and stereophotogrammetry. Complementarity to stereo photogrammetry, reflection-based surface reconstruction methods, such as shape from shading (SFS), have been studied and applied in lunar DEM reconstruction from a single image. However, this method often suffers from solution ambiguity and instability. In this paper, we propose a generative adversarial network (GAN)-based method that is able to generate high-resolution pixel-scale DEMs from a single image aided by a low-resolution DEM. We have evaluated the accuracy of the reconstructed high-resolution DEMs from 25 LROC NAC images of four regions using LROC NAC DEMs (2 m/pixel) as ground truth. The experimental results demonstrate good accuracy and adaptability to changes in illumination conditions. The root mean square error (RMSE) can reach about 2 m in areas where the degree of elevation variation is less than 100 m, and the RMSE value ranges from around 3 m to 10 m without considering the degree of the elevation variation in large-area reconstruction. As high-resolution monocular images and low-resolution DEMs are available for the entire lunar surface, the proposed GAN-based method has great potential in high-resolution lunar DEM reconstruction for lunar mapping applications.

Building extraction based on high-resolution remote sensing imagery has been widely used in automatic surveying and mapping. However, few methods have been developed for building instance extraction, i.e., extracting each building’s footprint separately, which is required in a number of applications, such as the smallest unit of a cadastral database. In building instance extraction, there are two challenges: 1) buildings with various scales exist in the imagery and 2) precise building footprints are difficult to extract due to the blurry boundaries. In this article, to solve these problems, a multiscale U-shaped convolutional neural network building instance extraction framework with edge constraint (EMU-CNN) for high-spatial-resolution remote sensing imagery is proposed. The proposed framework consists of three components: 1) a multiscale fusion U-shaped network (MFUN); 2) a region proposal network (RPN); and 3) an edge-constrained multitask network (ECMN). First, in the proposed method, the MFUN includes three parallel branches to learn multiple building features with different scales. The RPN then detects the positions of the building instances, even for buildings that are connected with each other. Moreover, according to the instance positions, the ECMN is proposed to extract a precise mask and suppress overfitting. The experiments conducted on a self-annotated data set and two public data sets (the ISPRS Vaihingen semantic labeling contest data set and the WHU aerial image data set) show that the EMU-CNN method can achieve excellent performance and shows great robustness at different scales. [ABSTRACT FROM AUTHOR]

Landslide extraction is one of the most popular topics in remote sensing. Numerous techniques have been proposed to manage the landslide identification problem. However, most aim to extract landslides that have already occurred or delineate the potential landslide manually. It is greatly important to identify and delineate potential landslides automatically, which has not been investigated. In this paper, we propose an automatic identification and delineation method, i.e., object-based image analysis (OBIA) of potential landslides by integrating optical imagery with a deformation map. We applied a deformation map generated by the interferometric synthetic aperture radar (InSAR) technique, rather than the digital elevation model (DEM) for landslide segmentation. Then, we used a classification and regression tree (CART) model with the spectral, spatial, contextual and deformation characteristics for landslide classification. For accuracy assessment, we implemented the evaluation indicators of recall and precision. The proposed method is verified in both specific landslide-prone regions (Jinpingzi and Shuanglongtan landslides) and a large catchment of the Jinsha River, China. By comparing our results with the ones using purely optical imagery, the precision of the Jinpingzi landslide is improved by 14.12%, and the recall and precision of the Shuanglongtan landslide are improved by 3.1% and 3.6%, respectively, and the recall for the large catchment is improved by 9.95%. Our method can improve delineation of potential landslides significantly, which is crucial for landslide early warning and prevention.

Woody vegetation dynamics in the Sahel have been debated since the great droughts of the 1970s–1980s. Here, we combined high-resolution satellite and aerial imagery, field inventory, and historical botanical records to study woody vegetation trends over the years 1965, 1980, 2008, and 2018 in the Ferlo, the Sahelian sylvo-pastoral zone of Senegal. While tree density has decreased from 1965 (14.8 trees ha−1) to 1980 (13.4 trees ha−1) and 2008 (11.9 trees ha−1), tree density has stabilized in 2018 (12.2 trees ha−1). The relatively moderate decrease in tree density over 50 years characterized by extensive human pressure and droughts, as well as the rather stable woody cover following the drought years after the 1980s, do not support narratives of widespread desertification in this region. However, we observed a shift in the composition of species. While the drought-resistant tree Balanites aegyptiaca showed a stable abundance, Acacia tortilis showed strong increases and other species like Sclerocarya birrea and Combretum glutinosum decreased. In addition, recent field surveys show that the ratio between shrubs and trees has increased towards more shrubs. The observed loss of species diversity combined with the increase of drought-resistant species is in line with current observations for savanna ecosystems in the context of an increased aridity. [ABSTRACT FROM AUTHOR]