High Resolution Earth Observation Research Articles

Using image segmentation models to analyse high-resolution earth observation data: new tools to monitor disease risks in changing environments.

In the near future, the incidence of mosquito-borne diseases may expand to new sites due to changes in temperature and rainfall patterns caused by climate change. Therefore, there is a need to use recent technological advances to improve vector surveillance methodologies. Unoccupied Aerial Vehicles (UAVs), often called drones, have been used to collect high-resolution imagery to map detailed information on mosquito habitats and direct control measures to specific areas. Supervised classification approaches have been largely used to automatically detect vector habitats. However, manual data labelling for model training limits their use for rapid responses. Open-source foundation models such as the Meta AI Segment Anything Model (SAM) can facilitate the manual digitalization of high-resolution images. This pre-trained model can assist in extracting features of interest in a diverse range of images. Here, we evaluated the performance of SAM through the Samgeo package, a Python-based wrapper for geospatial data, as it has not been applied to analyse remote sensing images for epidemiological studies. We tested the identification of two land cover classes of interest: water bodies and human settlements, using different UAV acquired imagery across five malaria-endemic areas in Africa, South America, and Southeast Asia. We employed manually placed point prompts and text prompts associated with specific classes of interest to guide the image segmentation and assessed the performance in the different geographic contexts. An average Dice coefficient value of 0.67 was obtained for buildings segmentation and 0.73 for water bodies using point prompts. Regarding the use of text prompts, the highest Dice coefficient value reached 0.72 for buildings and 0.70 for water bodies. Nevertheless, the performance was closely dependent on each object, landscape characteristics and selected words, resulting in varying performance. Recent models such as SAM can potentially assist manual digitalization of imagery by vector control programs, quickly identifying key features when surveying an area of interest. However, accurate segmentation still requires user-provided manual prompts and corrections to obtain precise segmentation. Further evaluations are necessary, especially for applications in rural areas.

Experimental demonstration of scene-based cophasing in optical synthetic aperture imaging using the SPGD algorithm

Large-aperture telescopes based on optical synthetic aperture imaging are investigated for recent high-resolution spaceborne observations. An enabling technique of aperture synthesis is a cophasing method to suppress a piston-tip-tilt error between sub-apertures. This paper proposes a scene-based cophasing technique using the stochastic parallel gradient descent (SPGD) algorithm, assuming application to high-resolution Earth observation. A significant advantage of the SPGD algorithm is a model-less cophasing capability based on extended scenes, but the simultaneous scene-based piston-tip-tilt correction between multiple apertures has not been demonstrated. In this paper, we developed a tabletop synthetic aperture imaging system with 37 sub-apertures and demonstrated extended-scene-based piston-tip-tilt control by optimizing applied voltages to 111 actuators simultaneously. The demonstration experiments used not only static scenes but also a time-varying dynamic scene for observation targets. In every measurement, the proposed scene-based approach reduced the initially defined piston-tip-tilt errors, and the image sharpness significantly improved, although the correction rate in the dynamic scene observation was slower. Finally, this paper discusses the influence of scene dynamics on image-based cophasing.

Land Surface Temperature Retrieval From GF5-01A Based on Split- Window Algorithm

Abstract. Land Surface temperature (LST) is a core parameter in the energy exchange between the surface and the atmosphere, and the use of thermal infrared remote sensing can realize the wide-range, fast, and accurate acquisition of surface temperature. GF5-01A is an important part and the final satellite of the major special project on high-resolution Earth observation system which is equipped with a wide-area thermal infrared imager with a resolution of 100 meters and a width of 1,500 kilometers. In this paper, based on the GF5-01A WTI spectral response function, combined with the TIGR2000 atmospheric profiles data and the ASTER spectral library, the data simulation was carried out by using the atmospheric radiative transfer model MODTRAN 5.2, and then constructed the split-window algorithm. Then, the method proposed in this paper was validated and evaluated using Landsat 8/9 temperature products and measured surface temperature data from SURFRAD sites acquired on the same day. The results show that the RMSE between the GF5-01A retrieved LST and the Landsat8/9 retrieved LST is between 1.27–2.24K, and the Bias is between −2.08–1.12K. The RMSE is between 0.68–2.64K and the bias is between −0.68–1.49K compared to the measured surface temperature. The split-window algorithm of GF5-01A proposed in this paper can meet the requirements of thermal infrared remote sensing monitoring and has enormous potential and value.

Land surface modeling informed by earth observation data: toward understanding blue–green–white water fluxes in High Mountain Asia

ABSTRACT Mountains are important suppliers of freshwater to downstream areas, affecting large populations in particular in High Mountain Asia (HMA). Yet, the propagation of water from HMA headwaters to downstream areas is not fully understood, as interactions in the mountain water cycle between the cryo-, hydro- and biosphere remain elusive. We review the definition of blue and green water fluxes as liquid water that contributes to runoff at the outlet of the selected domain (blue) and water lost to the atmosphere through vapor fluxes, that is evaporation from water, ground, and interception plus transpiration (green) and propose to add the term white water to account for the (often neglected) evaporation and sublimation from snow and ice. We provide an assessment of models that can simulate the cryo-hydro-biosphere continuum and the interactions between spheres in high mountain catchments, going beyond disciplinary separations. Land surface models are uniquely able to account for such complexity, since they solve the coupled fluxes of water, energy, and carbon between the land surface and atmosphere. Due to the mechanistic nature of such models, specific variables can be compared systematically to independent remote sensing observations – providing vital insights into model accuracy and enabling the understanding of the complex watersheds of HMA. We discuss recent developments in spaceborne earth observation products that have the potential to support catchment modeling in high mountain regions. We then present a pilot study application of the mechanistic land surface model Tethys & Chloris to a glacierized watershed in the Nepalese Himalayas and discuss the use of high-resolution earth observation data to constrain the meteorological forcing uncertainty and validate model results. We use these insights to highlight the remaining challenges and future opportunities that remote sensing data presents for land surface modeling in HMA.

A Digital Twin of the terrestrial water cycle: a glimpse into the future through high-resolution Earth observations

Climate change is profoundly affecting the global water cycle, increasing the likelihood and severity of extreme water-related events. Better decision-support systems are vital to accurately predict and monitor water-related environmental disasters and optimally manage water resources. These must integrate advances in remote sensing, in situ, and citizen observations with high-resolution Earth system modeling, artificial intelligence (AI), information and communication technologies, and high-performance computing. Digital Twin Earth (DTE) models are a ground-breaking solution offering digital replicas to monitor and simulate Earth processes with unprecedented spatiotemporal resolution. Advances in Earth observation (EO) satellite technology are pivotal, and here we provide a roadmap for the exploitation of these methods in a DTE for hydrology. The 4-dimensional DTE Hydrology datacube now fuses high-resolution EO data and advanced modeling of soil moisture, precipitation, evaporation, and river discharge, and here we report the latest validation data in the Mediterranean Basin. This system can now be explored to forecast flooding and landslides and to manage irrigation for precision agriculture. Large-scale implementation of such methods will require further advances to assess high-resolution products across different regions and climates; create and integrate compatible multidimensional datacubes, EO data retrieval algorithms, and models that are suitable across multiple scales; manage uncertainty both in EO data and models; enhance computational capacity via an interoperable, cloud-based processing environment embodying open data principles; and harness AI/machine learning. We outline how various planned satellite missions will further facilitate a DTE for hydrology toward global benefit if the scientific and technological challenges we identify are addressed.

Seasonal dynamics of fallow and cropping lands in the broadacre cropping region of Australia

Fallowing is an important strategy for enhancing soil health, water harvesting and crop yields, thus improving sustainability and reducing production risks in dryland farming systems in Australia. However, accurate data regarding the location, frequency, extent, and duration of fallow land is not readily available at high spatio-temporal resolutions before and during a cropping season. As a result, continental maps depicting seasonal dynamics of fallow and cropping lands and their responses to climate change and human activities remain missing. This study set out to develop an automated approach to discriminate fallow areas from cropping areas within Australian broadacre cropping region. This was done by applying a mathematical algorithm of turning point detection technique to a curve fitted vegetation index time series from 2016 to 2022, derived from high spatial (10 m) and temporal (5-days) resolution Sentinel-2 data. Specifically, we aimed to derive: (i) 5-day revisit fallow/cropping map layers; (ii) key attributes of the fallow and cropping periods (e.g., start, end, duration); (iii) cropping intensity and area. The proposed approach yielded a high and significant overall accuracy, surpassing 92% with substantial balanced accuracy (>0.9) against >3000 fields for individual winter and summer seasons across Australia. In addition, this study's temporal and geographical outputs, encompassing seasonal attributes, cropping intensity and cropping areas, robustly enhanced the understanding of seasonal cropping dynamics and historical shifts in dryland farming practices. Findings derived from this analysis aligned well with extreme climatic events and changes in crop activities adopted in response to such events. Future research will aim to extract key seasonal dynamics attributes influenced by cropping practices, enhancing the conversion of high-resolution earth observation data into actionable industry knowledge.

Multisource remote sensing image fusion processing in plateau seismic region feature information extraction and application analysis – An example of the Menyuan Ms6.9 earthquake on January 8, 2022

Abstract A 6.9 magnitude earthquake hit Menyuan County, Haibei Prefecture, Qinghai Province, at 01:45 PM Beijing time on January 8, 2022 (17:45 PM GMT time on January 7, 2022). To explore the magnitude of the earthquake deformation and the affected area, this work combined optical remote sensing interpretation, interferometric synthetic aperture radar (InSAR) coseismic deformation extraction, and field surveys for research and analysis. Relying on the high-resolution Earth observation system of the Qinghai Remote Sensing Center for Natural Resources, high-resolution GF1D, GF2, and TRIPLESAT optical remote sensing images were acquired immediately after the earthquake. The airborne triangulation encryption method was used to carry out orthographic correction, fusion, and mosaic processing of digital orthophoto map (DOM) and digital surface model (DSM) images, and first-hand optical remote sensing images of the disaster areas were obtained. Based on differential InSAR (D-InSAR), small baseline subset InSAR (SBAS-InSAR) and lifting rail fusion methods, the coseismic deformation field and deformation rate of the lifting rail direction were obtained by using Sentinel-1A data processing before and after the earthquake. Combined with optical interpretation, InSAR deformation, and field investigation, the results show that the deformation trend of the line of sight (LOS) images to the north and south of the ascending and descending orbits show an obvious opposite trend. The surface shape variables are −50 to 45 cm and −65 to 72 cm, respectively, and the deformation rate before the earthquake reached 25 cm/year. The deformation field characteristics show that the earthquake was mainly due to thrust, and the coseismic deformation field fractured along the WNW‒ESE direction with a length of approximately 33 km. The areas affected by 10 mm, 20 cm, and 50 cm deformation magnitudes in the whole earthquake area were 975.14, 321.10, and 38.55 km2, respectively. Within 20 km, there were two main affected townships, namely, Sujitan Township and Huangcheng Mongolian Township. Within 50 km, there were four main affected towns and townships, namely, Sujitan Township, Mongolian Township of the Imperial city, Qingshizui town, and Haomen town.

Incorporating high-resolution climate, remote sensing and topographic data to map annual forest growth in central and eastern Europe

To enhance our understanding of forest carbon sequestration, climate change mitigation and drought impact on forest ecosystems, the availability of high-resolution annual forest growth maps based on tree-ring width (TRW) would provide a significant advancement to the field. Site-specific characteristics, which can be approximated by high-resolution Earth observation by satellites (EOS), emerge as crucial drivers of forest growth, influencing how climate translates into tree growth. EOS provides information on surface reflectance related to forest characteristics and thus can potentially improve the accuracy of forest growth models based on TRW. Through the modelling of TRW using EOS, climate and topography data, we showed that species-specific models can explain up to 52 % of model variance (Quercus petraea), while combining different species results in relatively poor model performance (R2 = 13 %). The integration of EOS into models based solely on climate and elevation data improved the explained variance by 6 % on average. Leveraging these insights, we successfully generated a map of annual TRW for the year 2021. We employed the area of applicability (AOA) approach to delineate the range in which our models are deemed valid. The calculated AOA for the established forest-type models was 73 % of the study region, indicating robust spatial applicability. Notably, unreliable predictions predominantly occurred in the climate margins of our dataset. In conclusion, our large-scale assessment underscores the efficacy of combining climate, EOS and topographic data to develop robust models for mapping annual TRW. This research not only fills a critical void in the current understanding of forest growth dynamics but also highlights the potential of integrated data sources for comprehensive ecosystem assessments.

RESEARCH ON COOPERATION STRATEGY BASED ON SATELLITE REMOTE SENSING DATA SERVICE AND TECHNOLOGY APPLICATION BETWEEN CHINA AND ASEAN

Abstract. Remote sensing (RS) and earth observing technology are flourished with the development of a series of high-resolution earthobservation satellites. As the improvement of China’s earth observation data acquisition capability, one critical issue is put on the agenda that is what kind of models and techniques can promote the future data processing into a new level in terms of service model, massive data processing, development methods, business models, resource sharing, and system sustainability (1). In order to embed domestic satellite advantages into global world and provide increasingly more Chinese wisdom and solution to support the ASEAN regional development, China highlights the propositions on ASEAN-China remote sensing cooperation. On November 1, 2022, the ASEAN-China Satellite Remote Sensing Application Centre (hereinafter referred to as the ACSAC) was officially inaugurated in Beijing. The ACSAC will mainly focus on the establishment of a system and mechanism for a series of substantive operation, promoting the comprehensive sharing of China's land and ocean satellite data, comprehensively rich the international application scenarios of Chinese satellites, promoting applications in multiple fields and carrying out satellite remote sensing application promotion and typical demonstration, continuing to carry out technical exchanges and training with mutual demands and development. At present, ACSAC has stepped into the substantive construction stage. It will give full play to the advantages of China's natural resources land and ocean satellite remote sensing resources, to form a multi-scale, full coverage data resource support, jointly carry out the construction of the ASEAN-China satellite remote sensing data network platform based on the cloud environment, and to jointly build an integrated service open portal for land and ocean satellite data products to ASEAN region. Focusing on the core construction content of ACSAC, this paper systematically collects, summarizes and analyse the data and application needs of satellite remote sensing in ASEAN countries, finds out the current needs to meet the gap, and puts forward some thoughts and ideas on cooperation mechanism, data sharing, customized products and application demonstration of advanced products, aiming to lay a solid foundation for the substantial construction of ACSAC.

On-orbit geometric calibration and preliminary accuracy verification of GaoFen-14 (GF-14) optical two linear-array stereo camera

ABSTRACT The GaoFen-14 (GF-14) satellite is China’s most recent high-resolution earth observation satellite system. It is equipped with a two linear-array stereo camera and is intend for topographic mapping at 1:10,000-scale without ground control points (GCPs). The technical parameters of payloads will change once the satellite enters orbit. As a result, strict on-orbit geometric calibration is necessary. This study performs the on-orbit geometric calibration of the GF-14 stereo camera using ground calibration data. The exterior orientation errors are corrected by a comprehensive bias matrix, internal orientation errors are described by a fifth order polynomial. The results of Yinchuan calibration field show that the planar accuracy is better than 286 m (RMS) before calibration, and improved to 2.11 m (front camera) and 1.51 m (back camera) by external orientation, and further improved to 1.46 m (front camera) and 0.99 m (back camera) after internal calibration. The front intersection accuracy reaches 0.67 m in plane (RMS) and 1.10 m in elevation (RMS), respectively. Checking by multiple global check points (CKPs), the planar and elevation RMS reaches 2.38 m (5.09 m in CE90) and 2.08 m (3.43 m in LE90), respectively.

Predicting Continental Scale Malaria With Land Surface Water Predictors Based on Malaria Dispersal Mechanisms and High-Resolution Earth Observation Data.

Despite malaria prevalence being linked to surface water through vector breeding, spatial malaria predictors representing surface water often predict malaria poorly. Furthermore, precipitation, which precursors surface water, often performs better. Our goal is to determine whether novel surface water exposure indices that take malaria dispersal mechanisms into account, derived from new high-resolution surface water data, can be stronger predictors of malaria prevalence compared to precipitation. One hundred eighty candidate predictors were created by combining three surface water malaria exposures from high-accuracy and resolution (5m resolution, overall accuracy 96%, Kappa Coefficient 0.89, Commission and Omission error 3% and 13%, respectively) water maps of East Africa. Through variable contribution analysis a subset of strong predictors was selected and used as input for Boosted Regression Tree models. We benchmarked the performance and Relative Contribution of this set of novel predictors to models using precipitation instead of surface water predictors, alternative lower resolution predictors, and simpler surface water predictors used in previous studies. The predictive performance of the novel indices rivaled or surpassed that of precipitation predictors. The novel indices substantially improved performance over the identical set of predictors derived from the lower resolution Joint Research Center surface water data set (+10% R 2, +17% Relative Contribution) and over the set of simpler predictors (+18% R 2, +30% Relative Contribution). Surface water derived indices can be strong predictors of malaria, if the spatial resolution is sufficiently high to detect small waterbodies and dispersal mechanisms of malaria related to surface water in human and vector water exposure assessment are incorporated.

Space polymers erosion in simulated LEO and VLEO environments

The VLEO satellites, that are orbiting at an altitude of 200-300 km and that provide advantages for high-resolution Earth observation and atmospheric science, are becoming important not only for science, but also for agriculture, forestry, and oceans observations, for military, national security and the future of satellite businesses. We evaluated, for the first time, the performance and defined the erosion yields (Ey) for a selected number of main space hydrocarbon, fluorinated, and perfluorinated polymers in the ground-based simulator facility at conditions, imitating the LEO and VLEO environments. We also compared the results with the data for polymers erosion in LEO and with the previously established predictive model, developed for polymers and composites in LEO environment. The obtained data indicated that the predictive model can be extended and used, with some corrections for the predictions of space polymers erosion in the VLEO environment.

An In-Memory Data-Cube Aware Distributed Data Discovery Across Clouds for Remote Sensing Big Data

Abstract: The recent proliferation of high-resolution Earth observation data, along with the advent of open-data initiatives, has led to the availability of petabyte-scale Earth observation datasets accessible for free. However, this abundance of big data presents significant challenges for regional to global spatio-temporal analysis. The traditional approach of "downloadpreprocess-store-analyze" introduces excessive data downloading overhead and computational barriers, hindering efficient analysis. The Earth observation data cube (EODC) paradigm offers a solution by revolutionizing the storage and management of spatio-temporal RS data. Yet, the presence of multiple EODC solutions has resulted in "information silos," making the sharing and joint use of RS data across EODCs difficult.

Detecting the long-term spatiotemporal crop phenology changes in a highly fragmented agricultural landscape

Trends in crop phenometrics reveal the influence of climate variability and change on crop growth and development. However, the trends are less clear in fragmented tropical smallholder landscapes, because there is high spatial and temporal variability in crop phenology. Frequent historical and high spatial resolution (≤30 m) Earth observations are needed to track changes in crop phenology in fragmented landscapes but are often unavailable. The spatial–temporal gap can be closed by integrating infrequent high spatial resolution Earth observations with low spatial/high temporal resolution observations through data fusion. We fused 30 m resolution Landsat and 250 m resolution MODIS imagery to investigate trends in crop phenology from 2000 to 2020 in a fragmented agricultural landscape of Ethiopia. We used the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM) that had recently been modified for application in fragmented agricultural landscapes. We used the non-parametric Mann–Kendall test for crop phenology trend analysis. Crop phenology based on Landsat–MODIS fusion was compared to MODIS-based crop phenology without fusion. We found data fusion yielded a smaller magnitude of changes in the start of season (SOS: -0.2-day-y−1) and end of season (EOS: -0.50-day-y−1) compared to MODIS SOS (-0.5-day-y−1) and EOS (1.38-day-y−1) due to MODIS-related mixing with the surrounding natural vegetation in the fragmented agricultural landscape. EOS showed a faster rate of change compared to SOS over the 21-years. The Landsat and MODIS fusion captured spatial variation in the timing and magnitude of change specific to crops and their growing environment, which has implications for adaptation strategies. Our results highlight the importance of long-term data fusion to improve the spatial dimension of crop phenology time series analysis. Integrating time series land cover maps into the data fusion processing chain could further improve long-term data fusion for crop phenology trend analysis.

Predicting continental scale malaria with land surface water predictors based on malaria dispersal mechanisms and high-resolution earth observation data

Predicting continental scale malaria with land surface water predictors based on malaria dispersal mechanisms and high-resolution earth observation data

A first Chinese building height estimate at 10 m resolution (CNBH-10 m) using multi-source earth observations and machine learning

Building height is a crucial variable in the study of urban environments, regional climates, and human-environment interactions. However, high-resolution data on building height, especially at the national scale, are limited. Fortunately, high spatial-temporal resolution earth observations, harnessed using a cloud-based platform, offer an opportunity to fill this gap. We describe an approach to estimate 2020 building height for China at 10 m spatial resolution based on all-weather earth observations (radar, optical, and night light images) using the Random Forest (RF) model. Results show that our building height simulation has a strong correlation with real observations at the national scale (RMSE of 6.1 m, MAE = 5.2 m, R = 0.77). The Combinational Shadow Index (CSI) is the most important contributor (15.1%) to building height simulation. Analysis of the distribution of building morphology reveals significant differences in building volume and average building height at the city scale across China. Macau has the tallest buildings (22.3 m) among Chinese cities, while Shanghai has the largest building volume (298.4 108 m3). The strong correlation between modelled building volume and socio-economic parameters indicates the potential application of building height products. The building height map developed in this study with a resolution of 10 m is open access, provides insights into the 3D morphological characteristics of cities and serves as an important contribution to future urban studies in China.

Litter on the streets - solid waste detection using VHR images

ABSTRACT Failures in urban areas’ solid waste management lead to clandestine garbage dumping and pollution. This affects sanitation and public human hygiene, deteriorates quality of life, and contributes to deprivation. This study aimed to test a combination of machine learning, high-resolution earth observation and GIS data to detect diverse categories of residual waste on the streets, such as sacks and construction debris. We conceptualised five different classes of solid waste from image interpretation: “Sure”, “Half-sure”, “Not-sure”, “Dispersed”, and “Non-garbage”. We tested a combination of k-means-based segmentation and supervised random forest to investigate the capabilities of automatic classification of these waste classes. The model can detect the presence of solid waste on the streets and achieved an accuracy of up from 73.95%–95.76% for the class “Sure”. Moreover, a building extraction using an EfficientNet deep-learning-based semantic segmentation allowed masking the rooftops. This improved the accuracy of the classes “Sure” and “Non-garbage”. The systematic evaluation of all parameters considered in this model provides a robust and reliable method of solid waste detection for decision-makers. These results highlight areas where insufficient waste management affects the citizens of a given city.

Signal-to-noise ratio model in Python for high-resolution space-borne electro-optic imagers

We present a signal-to-noise ratio (SNR) simulator design for high-resolution Earth observation imagers with time-delay and integration (TDI) imaging sensors. SNR is one of the space-borne electro-optic imagers’ driving design and performance evaluation parameters. TDI line imaging sensors are used to meet these imagers’ SNR requirements by increasing effective exposure time during imaging. Determining the optimum number of TDI stages to meet the SNR requirements of the mission is critical for the imaging sensor selection, optical design of the telescope, laboratory radiometric tests, and imaging in orbit. The SNR simulator is based on an end-to-end image chain calculation following the light pathway from the sun to the imager. The simulator comprises radiance, telescope, sensor, and image modules written in the open-access Python environment. The radiance module uses PcModWin/MODerate resolution atmospheric TRANsmission output data of the atmospheric radiance calculation input for the simulator. The simulator computes the SNR for the given telescope and imaging sensor model and simulates associated images to show the radiometric performance. A high-resolution electro-optic imager model is used to demonstrate the performance and versatility of the simulator for the case studies. Experimental simulator validation is demonstrated for the images acquired by the Göktürk-2 space-borne imager.

Multi-Scale Feature Map Aggregation and Supervised Domain Adaptation of Fully Convolutional Networks for Urban Building Footprint Extraction

Automated building footprint extraction requires the Deep Learning (DL)-based semantic segmentation of high-resolution Earth observation images. Fully convolutional networks (FCNs) such as U-Net and ResUNET are widely used for such segmentation. The evolving FCNs suffer from the inadequate use of multi-scale feature maps in their backbone of convolutional neural networks (CNNs). Furthermore, the DL methods are not robust in cross-domain settings due to domain-shift problems. Two scale-robust novel networks, namely MSA-UNET and MSA-ResUNET, are developed in this study by aggregating the multi-scale feature maps in U-Net and ResUNET with partial concepts of the feature pyramid network (FPN). Furthermore, supervised domain adaptation is investigated to minimise the effects of domain-shift between the two datasets. The datasets include the benchmark WHU Building dataset and a developed dataset with 5× fewer samples, 4× lower spatial resolution and complex high-rise buildings and skyscrapers. The newly developed networks are compared to six state-of-the-art FCNs using five metrics: pixel accuracy, adjusted accuracy, F1 score, intersection over union (IoU), and the Matthews Correlation Coefficient (MCC). The proposed networks outperform the FCNs in the majority of the accuracy measures in both datasets. Compared to the larger dataset, the network trained on the smaller one shows significantly higher robustness in terms of adjusted accuracy (by 18%), F1 score (by 31%), IoU (by 27%), and MCC (by 29%) during the cross-domain validation of MSA-UNET. MSA-ResUNET shows similar improvements, concluding that the proposed networks when trained using domain adaptation increase the robustness and minimise the domain-shift between the datasets of different complexity.

Decoupling Semantic and Edge Representations for Building Footprint Extraction From Remote Sensing Images

Very high-resolution (VHR) earth observation systems provide an ideal data source for man-made structure detection such as building footprint extraction. Manually delineating building footprints from the remotely sensed VHR images, however, is laborious and time-intensive; thus, automation is needed in the building extraction process to increase productivity. Recently, many researchers have focused on developing building extraction algorithms based on the encoder-decoder architecture of convolutional neural networks. However, we observe that this widely adopted architecture cannot well preserve the precise boundaries and integrity of the extracted buildings. Moreover, features obtained by shallow convolutional layers contain irrelevant background noises that degrade building feature representations. This paper addresses these problems by presenting a feature decoupling network (FD-Net) that exploits two essential building information from the input image, including semantic information that concerns building integrity and edge information that improves building boundaries. The proposed FD-Net improves the existing encoder-decoder framework by decoupling image features into the edge subspace and the semantic subspace; the decoupled features are then integrated by a supervision-guided fusion process considering the heterogeneity between edge and semantic features. Furthermore, a lightweight and effective global context attention module is introduced to capture contextual building information and thus enhance feature representations. Comprehensive experimental results on three real-world datasets confirm the effectiveness of FD-Net in large-scale building mapping. We applied the proposed method to various encoder-decoder variants to verify the generalizability of the proposed framework. Experimental results show remarkable accuracy improvements with less computational cost.