Aerial Orthoimagery Research Articles

Machine learning approach to burned area mapping for Southern California

ABSTRACT Accurate representation of the location and amount of burned areas is vital to the understanding of patterns and impacts of fires. Some extant burned area maps appear to have high commission errors, which lead to an overrepresentation of burned area. The primary research objective of this study was to assess whether region-specific training data used for machine learning routines improve accuracy of burned area products for western San Diego County. We used training data derived from fine-scale aerial orthoimagery to create and compare three training sets, each with a different Landsat scale sub-pixel burn threshold: 20%, 50%, or 80%. Meaning either 20%, 50%, or 80% of a 30 m × 30 m pixel had to burn for the entire pixel to be classified as burned. High-resolution orthoimagery was used for the creation of the training/testing data as well as determining which sub-pixel threshold leads to more accurate burned area representation. These training data were input into a gradient-boosted regression model. We compared the burned area product from the region-specific gradient boosted model (L-GBRM) to the three products: Monitoring Trends in Burn Severity, Fire and Resource Assessment Program, and the Landsat Burned Area product. We found >20% sub-pixel burn threshold of a Landsat pixel yielded the most accurate classification results. We used a 50% sub-pixel burn threshold for the reference data to compare the results to since the burned area associated with it is closely aligned with the high-resolution orthoimagery determined burn area. The L-GBRM was the most accurate product while also mapping the smallest area burned, suggesting that the extant products have relatively high commission errors. Using region-specific training data achieved a higher accuracy than nationwide training data. Looking at sub-pixel burn thresholds for creating a burned area map could prove to make a more accurate map in terms of area burned represented.

Urban Pedestrian Routes' Accessibility Assessment Using Geographic Information System Processing and Deep Learning-Based Object Detection.

The need to establish safe, accessible, and inclusive pedestrian routes is considered one of the European Union's main priorities. We have developed a method of assessing pedestrian mobility in the surroundings of urban public buildings to evaluate the level of accessibility and inclusion, especially for people with reduced mobility. In the first stage of assessment, artificial intelligence algorithms were used to identify pedestrian crossings and the precise geographical location was determined by deep learning-based object detection with satellite or aerial orthoimagery. In the second stage, Geographic Information System techniques were used to create network models. This approach enabled the verification of the level of accessibility for wheelchair users in the selected study area and the identification of the most suitable route for wheelchair transit between two points of interest. The data obtained were verified using inertial sensors to corroborate the horizontal continuity of the routes. The study findings are of direct benefit to the users of these routes and are also valuable for the entities responsible for ensuring and maintaining the accessibility of pedestrian routes.

Mapping environmentally sustainable urban development within six US cities through object-based image change analysis of aerial orthoimagery

The rapid expansion of cities and continuous urban population growth underscores a need for sustainable urban development. Sustainable development is that which addresses human needs, contributes to well-being, is economically viable, and utilizes natural resources at a degree sustainable by the surrounding environmental systems. Urban green spaces, green roofs, and solar panels are examples of environmentally sustainable urban development (ESUD), or development that focuses on environmental impact, but also presents the potential to achieve social and economic sustainability. The aim of this study was to map and compare amounts of ESUD c. 2010 and c. 2019 through an object-based image analysis (OBIA) approach using National Agricultural Imagery Program (NAIP) aerial orthoimagery for six mid- to large-size cities in the USA.The results of this study indicate a hybrid OBIA and manual interpretation approach applied to NAIP orthoimagery may allow for reliable mapping and areal estimation of urban green space and green roof changes in urban areas. The reliability of OBIA-only mapping and estimation of areal extents of existing green roofs, and new and existing solar panels, is inconclusive due to low mapping accuracy and coarse spatial resolution of aerial orthoimagery relative to some ESUD features. The three urban study areas in humid continental climate zones (Dfa) were estimated to have greater areal extent of new and existing urban green space and existing green roofs, but less areal extent of new green roofs and existing solar panels compared to the three study areas in humid subtropical climate zones (Cfa).

Semi- and weak-supervised learning for Norwegian tree species detection

Tree species mapping of Norwegian production forests is a time-consuming process as forest associations largely rely on manual interpretation of earth observation data. Deep learning based image segmentation techniques have the potential to improve automated tree species classification, but a major challenge is the limited quality and availability of training data. Semi-supervised techniques could alleviate the need for training label and weak supervision enables handling coarse-grained and noisy labels. In this study, we evaluated the added value of semi-supervised deep learning methods in a weakly supervised setting. Specifically, consistency training and pseudo-labeling are applied for tree species classification from aerial ortho imagery in Norway. The techniques are generic and relevant for the wider earth observation domain, especially for other land cover segmentation tasks. The results show that consistency training gives a significant performance increase. Pseudo-labeling on the other hand does not, potentially this is due to varying convergence speeds for different classes causing confirmation bias or a partial violation of the cluster assumption.

Land-Cover-Change Detection with Aerial Orthoimagery Using SegNet-Based Semantic Segmentation in Namyangju City, South Korea

In this study, we classified land cover using SegNet, a deep-learning model, and we assessed its classification accuracy in comparison with the support-vector-machine (SVM) and random-forest (RF) machine-learning models. The land-cover classification was based on aerial orthoimagery with a spatial resolution of 1 m for the input dataset, and Level-3 land-use and land-cover (LULC) maps with a spatial resolution of 1 m as the reference dataset. The study areas were the Namhan and Bukhan River Basins, where significant urbanization occurred between 2010 and 2012. The hyperparameters were selected by comparing the validation accuracy of the models based on the parameter changes, and they were then used to classify four LU types (urban, crops, forests, and water). The results indicated that SegNet had the highest accuracy (91.54%), followed by the RF (52.96%) and SVM (50.27%) algorithms. Both machine-learning models showed lower accuracy than SegNet in classifying all land-cover types, except forests, with an overall-accuracy (OA) improvement of approximately 40% for SegNet. Next, we applied SegNet to detect land-cover changes according to aerial orthoimagery of Namyangju city, obtained in 2010 and 2012; the resulting OA values were 86.42% and 78.09%, respectively. The reference dataset showed that urbanization increased significantly between 2010 and 2012, whereas the area of land used for forests and agriculture decreased. Similar changes in the land-cover types in the reference dataset suggest that urbanization is in progress. Together, these results indicate that aerial orthoimagery and the SegNet model can be used to efficiently detect land-cover changes, such as urbanization, and can be applied for LULC monitoring to promote sustainable land management.

Identifying conifer mortality induced by Armillaria root disease using airborne lidar and orthoimagery in south central Oregon

Armillaria root disease causes tree stress and mortality worldwide, including in conifer forests of western North America. Armillaria-induced tree mortality is slower than other disturbances, such as fire, but persistent over time, and therefore difficult to detect across large areas. Hence methods to detect Armillaria-affected trees and identify infested stands are of great value to forest managers. We used high-density light detection and ranging (lidar), high-resolution aerial orthoimagery, and associated field observations to map individual tree health across an Armillaria-affected forest dominated by Abies in south central Oregon. The lidar point cloud was segmented into individual tree objects (polygons representing tree crown extents as viewed from nadir), for which lidar metrics and spectral metrics derived from orthoimagery were computed. Lidar-detected tree objects were paired with 150 field-observed trees with corresponding health measurements, and a random forest classifier was developed that separated trees into: 1) asymptomatic; 2) live, Armillaria-infected; 3) recently killed (≥50% of red needles remaining); and 4) dead (<50% of dead needles remaining) classes with 83% accuracy using lidar and spectral metrics. The classifier was applied to map individual tree health status for 290,964 tree objects across the 1257-ha study area. Approximately 20% of trees were classified as unhealthy including live, infected and 4% of trees were classified as recently killed or dead. We created hotspot maps using the Getis-Ord Gi* statistic and analyzed clustering of tree health spatial patterns using Ripley’s L statistic. Hotspot maps effectively identified clusters of live unhealthy trees and tree mortality; unhealthy and dead trees were found to be significantly spatially clustered at distances of 100–2500 m. We created a dead tree density grid, which we coupled with a lidar-derived canopy gap grid to identify sites and stands affected by root disease. Canopy openings were mapped using a canopy height model with a minimum opening area of ≥ 202 m2. Clusters of mapped dead trees (excluding recently killed trees) intersecting canopy gaps were used to detect sites with root disease-induced mortality. Twenty-seven stands containing long-term plots from previous studies were evaluated for the presence of root disease-induced mortality. All 27 stands were correctly identified with conifer mortality induced by root disease. This approach for detecting dead trees intersecting canopy openings induced by root disease can aid in: i) prioritizing subsequent field data collection; ii) planning silvicultural prescriptions; and iii) assessing management expectations for snags and wildlife habitat where root disease-induced mortality is altering stand structure.

Light Detection and Ranging (LiDAR)-Assisted Detection of Rock Outcrops in Appalachian Hardwood Forests

Abstract The identification of small habitat features embedded within forest ecosystems is a challenge for many wildlife inventory and monitoring programs, especially for those involving rock outcrop specialist taxa. Rock outcrops are often difficult to remotely detect in dense Appalachian hardwood forests, as most outcrops remain hidden under the forest canopy and therefore invisible when relying on aerial orthoimagery to pinpoint habitat features. We investigated the ability for light detection and ranging point cloud data to identify small rock outcrops during the environmental assessment phase of a proposed management project on the Jefferson National Forest in Virginia. We specifically compared this approach with the visual identification of rock outcrops across the same area using aerial orthoimagery. Our light detection and ranging-based approach identified three times as many rock outcrop sites as aerial orthoimagery, resulting in the field verification of four times as many previously unknown populations of green salamanders Aneides aeneus (a rock outcrop specialist amphibian of high conservation concern) than would have been possible if relying on aerial orthoimagery alone to guide surveys. Our results indicate that light detection and ranging-based methods may provide an effective, efficient, and low-error approach that can remotely identify below-canopy rock outcrops embedded within Appalachian forests, especially when researchers lack pre-existing knowledge of local terrain and the location of habitat features.

Rock glaciers throughout the French Alps accelerated and destabilised since 1990 as air temperatures increased

Rock glaciers—ice-rich creeping landforms typical of permafrost mountain ranges—can develop an anomalous landslide-like behaviour called destabilisation. This behaviour is characterised by failure mechanisms (including cracks and crevasses) and increases in displacement rates by one to two orders of magnitude. Existing studies of this phenomenon have been limited to a small number of landforms and short time spans. Here, we systematically investigate the evolution of rock glacier kinematics over the past seven decades for the entire French Alps by combining observations of landform features indicative of the onset of destabilisation with data on displacements rates using aerial orthoimagery. We show that rock glacier velocities have significantly increased since the 1990s, concurrent with the development of destabilisation in 18 landforms that represent 5% of the 337 active rock glaciers. This pattern of activity correlates with rising air temperatures in the region, which suggests that a warming climate may play a role in this process.

A Comparative Study of Deep Learning Approaches to Rooftop Detection in Aerial Images

This paper investigates the deep neural networks for rapid and accurate detection of building rooftops in aerial orthoimages. The networks were trained using the manually labeled rooftop vector data digitized on aerial orthoimagery covering the Kitchener-Waterloo area. The performance of the three deep learning methods, U-Net, Fully Convolutional Network (FCN), and Deeplabv3+ were compared by training, validation, and testing sets in the dataset. Our results demonstrated that DeepLabv3+ achieved 63.8% in Intersection over Union (IoU), 77.8% in mean IoU (mIoU), 74% in precision, and 78% in F1-score. After improving the performance with focal loss, training loss was greatly cut down and the convergence rate experienced a significant growth. Meanwhile, rooftop detection also achieved higher performance, as Deeplabv3+ reached 93.6% in average pixel accuracy, with 65.4% in IoU, 79.0% in mIoU, 77.6% in precision, and 79.1% in F1-score. Lastly, in order to evaluate the effects of data volume, by changing data volume from 100% to 75% and 50% in ablation study, it shows that when data volume decreased, the performance of extraction also got worse, with IoU, mIoU, precision, and F1-score also mostly decreased.

LiDAR-aided hydrogeologic modeling and object-based wetland mapping approach for Pennsylvania

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

Polish Cadastre Modernization with Remotely Extracted Buildings from High-Resolution Aerial Orthoimagery and Airborne LiDAR

Automatic building extraction from remote sensing data is a hot but challenging research topic for cadastre verification, modernization and updating. Deep learning algorithms are perceived as more promising in overcoming the difficulties of extracting semantic features from complex scenes and large differences in buildings’ appearance. This paper explores the modified fully convolutional network U-Shape Network (U-Net) for high resolution aerial orthoimagery segmentation and dense LiDAR data to extract building outlines automatically. The three-step end-to-end computational procedure allows for automated building extraction with an 89.5% overall accuracy and an 80.7% completeness, which made it very promising for cadastre modernization in Poland. The applied algorithms work well both in densely and poorly built-up areas, typical for peripheral areas of cities, where uncontrolled development had recently been observed. Discussing the possibilities and limitations, the authors also provide some important information that could help local authorities decide on the use of remote sensing data in land administration.

Shrub Fractional Cover Estimation and Mapping of San Clemente Island Shrubland Based on Airborne Multispectral Imagery and Lidar Data

The purpose of this study is to map shrub distributions and estimate shrub cover fractions based on the classification of high-spatial-resolution aerial orthoimagery and light detection and ranging (LiDAR) data for portions of the highly disturbed coastal sage scrub landscapes of San Clemente Island, California. We utilized nine multi-temporal aerial orthoimage sets for the 2010 to 2018 period to map shrub cover. Pixel-based and object-based image analysis (OBIA) approaches to image classification of growth forms were tested. Shrub fractional cover was estimated for 10, 20 and 40 m grid sizes and assessed for accuracy. The most accurate estimates of shrub cover were generated with the OBIA method with both multispectral brightness values and canopy height estimates from a normalized digital surface model (nDSM). Fractional cover products derived from 2015 and 2017 orthoimagery with nDSM data incorporated yielded the highest accuracies. Major factors that influenced the accuracy of shrub maps and fractional cover estimates include the time of year and spatial resolution of the imagery, the type of classifier, feature inputs to the classifier, and the grid size used for fractional cover estimation. While tracking actual changes in shrub cover over time was not the purpose, this study illustrates the importance of consistent mapping approaches and high-quality inputs, including very-high-spatial-resolution imagery and an nDSM.

A Deep Learning-Based Solution for Large-Scale Extraction of the Secondary Road Network from High-Resolution Aerial Orthoimagery

Secondary roads represent the largest part of the road network. However, due to the absence of clearly defined edges, presence of occlusions, and differences in widths, monitoring and mapping them represents a great effort for public administration. We believe that recent advancements in machine vision allow the extraction of these types of roads from high-resolution remotely sensed imagery and can enable the automation of the mapping operation. In this work, we leverage these advances and propose a deep learning-based solution capable of efficiently extracting the surface area of secondary roads at a large scale. The solution is based on hybrid segmentation models trained with high-resolution remote sensing imagery divided in tiles of 256 × 256 pixels and their correspondent segmentation masks, resulting in increases in performance metrics of 2.7–3.5% when compared to the original architectures. The best performing model achieved Intersection over Union and F1 scores of maximum 0.5790 and 0.7120, respectively, with a minimum loss of 0.4985 and was integrated on a web platform which handles the evaluation of large areas, the association of the semantic predictions with geographical coordinates, the conversion of the tiles’ format and the generation of geotiff results compatible with geospatial databases.

END-TO-END BUILDING CHANGE DETECTION MODEL IN AERIAL IMAGERY AND DIGITAL SURFACE MODEL BASED ON NEURAL NETWORKS

Abstract. Multi-temporal building change detection is one of the most essential major issues of photogrammetry and remote sensing at current stage, which is of great significance for wide applications as offering real estate indicators as well as monitoring urban environment. Although current photogrammetry methodologies could be applicated to 2-D remote sensing imagery for rectification with sensor parameters, multi-temporal aerial or satellite imagery is not adequate to offer spectral and textual features for building change detection. Alongside recent development of Dense Image Matching (DIM) technology, the acquisition of 3-D point cloud and Digital Surface Model (DSM) has been generally realized, which could be combined with imagery, making building change detection more effective with greater spatial structure and texture information. Over the past years, scholars have put forward vast change detection techniques including traditional and model-based solutions. Nevertheless, existing appropriate methodology combined with Neural Networks (NN) for accurate building change detection with multi-temporal imagery and DSM remains to be of great research focus currently due to the inevitable limitations and omissions of existing NN-based methods, which is of great research prospect. This study proposed a novel end-to-end model framework based on deep learning for pixel-level building change detection from high-spatial resolution aerial ortho imagery and corresponding DSM sharing same resolution, which is from the dataset of Tokyo whole area.

Multi-scale roof characterization from LiDAR data and aerial orthoimagery: Automatic computation of building photovoltaic capacity

Abstract Photovoltaic self-consumption in buildings requires the installation of photovoltaic (PV) systems mostly on roofs, taking the advantage of these building locations for both the available surface area and the certainty of a big amount of annual incoming solar radiation. Since the parameters required for a proper PV system design are mainly geometric: azimuth-orientation, tilt angle and effective dimensions of the different roof slopes; aerial LiDAR data and orthoimagery offered by the National Geographic Agencies became suitable data sources for this purpose, ensuring its availability for any city regardless of its location. This paper presents a novel automatic methodology that combines LiDAR and orthoimage data processing to geometrically characterize roofs at slope level and calculate their PV solar potential. The methodology developed has been validated against results obtained from a higher-resolution aerial 3D point cloud of the roofs under study. Different locations and roof types have been tested in order to confirm the performance of the methodology under different conditions, being able to accurately characterize the geometry of most types of roofs, such as flat roofs, gable or saddle roofs, single pitched roofs and pyramid roofs at city, neighbourhood and building level.

Deep Convolutional Neural Networks for Automated Characterization of Arctic Ice-Wedge Polygons in Very High Spatial Resolution Aerial Imagery

The microtopography associated with ice-wedge polygons governs many aspects of Arctic ecosystem, permafrost, and hydrologic dynamics from local to regional scales owing to the linkages between microtopography and the flow and storage of water, vegetation succession, and permafrost dynamics. Wide-spread ice-wedge degradation is transforming low-centered polygons into high-centered polygons at an alarming rate. Accurate data on spatial distribution of ice-wedge polygons at a pan-Arctic scale are not yet available, despite the availability of sub-meter-scale remote sensing imagery. This is because the necessary spatial detail quickly produces data volumes that hamper both manual and semi-automated mapping approaches across large geographical extents. Accordingly, transforming big imagery into ‘science-ready’ insightful analytics demands novel image-to-assessment pipelines that are fueled by advanced machine learning techniques and high-performance computational resources. In this exploratory study, we tasked a deep-learning driven object instance segmentation method (i.e., the Mask R-CNN) with delineating and classifying ice-wedge polygons in very high spatial resolution aerial orthoimagery. We conducted a systematic experiment to gauge the performances and interoperability of the Mask R-CNN across spatial resolutions (0.15 m to 1 m) and image scene contents (a total of 134 km2) near Nuiqsut, Northern Alaska. The trained Mask R-CNN reported mean average precisions of 0.70 and 0.60 at thresholds of 0.50 and 0.75, respectively. Manual validations showed that approximately 95% of individual ice-wedge polygons were correctly delineated and classified, with an overall classification accuracy of 79%. Our findings show that the Mask R-CNN is a robust method to automatically identify ice-wedge polygons from fine-resolution optical imagery. Overall, this automated imagery-enabled intense mapping approach can provide a foundational framework that may propel future pan-Arctic studies of permafrost thaw, tundra landscape evolution, and the role of high latitudes in the global climate system.

Visualization on the Web of 20 Years of Crop Rotation and Wildlife Co-Evolutions

Abstract. The accumulation of evidences of the effects of intensive agricultural practices against wildlife fauna and flora, and biodiversity in general, has been largely published in scientific papers (Tildman, 1999). However, data serving as sup-port to their conclusions are often kept hidden behind research institutions. This paper presents a data visualization sys-tem opened on the Web allowing citizens to get a comprehensive access to data issued from such kind of research institution, collected for more than 20 years. The Web Information System has been thought in order to ease the comparison of data issues from various databases describing the same object, the agricultural landscape, at different scales and through different observation devices. An interactive visualization is proposed in order to check co-evolution of fauna and flora together with agricultural practices. It mixes aerial orthoimagery produced since 1950 with vectorial data showing the evolutions of agricultural parcels with those of a few sentinel species such as the Montagu's harrier. This is made through a composition of maps, charts and time lines, and specific tools for comparison. A particular concern is given to the observation effort bias in order to show meaningful statistical aggregates.

MCCT: a multi-channel complementary census transform for image classification

Census transformation and its variants have gained popularity in image classification for their simplicity and better performance. To describe a texture pattern, these approaches generally use sign information while comparing neighboring pixels. However, our observation is that sign and magnitude in a single color channel as well as in different color channels hold complementary information where sign component captures texture in an image and the saliency of that texture can be captured by the magnitude component. Considering these issues, a multi-channel complementary census transform (MCCT) is proposed in this paper by combining all of these information to capture more discriminating features. Rigorous experiments on nine different datasets which belong to six different applications such as flower, gender, aerial orthoimagery, event, leaf, indoor and outdoor scene classification demonstrate that MCCT outperforms existing state-of-the-art techniques.

Object-based Image Mapping of Conifer Tree Mortality in San Diego County based on Multitemporal Aerial Ortho-imagery

Two GEOBIA approaches are compared for their effectiveness in mapping dead trees within island montane forests of Southern California: a spatial contextual approach using an artificial neural network classifier, and a segmentation and multi-pixel classification approach. Both approaches are tested with multitemporal aerial orthoimagery having varying spatial resolutions. Spectral transformation inputs are also tested. An object-based accuracy assessment is conducted. Accuracies range between 30 percent to 90 percent for the dead tree class and are significantly higher for the spatial-contextual approach. Inclusion of spectral transforms increased accuracies by 5 percent for the true object-based approach, up to 13 percent for the spatial contextual approach, and reduced commission error up to 10 percent for both approaches. Masking techniques increased accuracies of the spatial contextual approach by 20 percent. With manual editing, the most accurate maps of individual live and dead trees from the spatial contextual approach are suitable for studying spatio-temporal trends in montane conifer mortality.

Object-based Image Mapping of Conifer Tree Mortality in San Diego County based on Multitemporal Aerial Ortho-imagery

Two GEOBIA approaches are compared for their effectiveness in mapping dead trees within island montane forests of Southern California: a spatial contextual approach using an artificial neural network classifier, and a segmentation and multi-pixel classification approach. Both approaches are tested with multitemporal aerial orthoimagery having varying spatial resolutions. Spectral transformation inputs are also tested. An object-based accuracy assessment is conducted. Accuracies range between 30 percent to 90 percent for the dead tree class and are significantly higher for the spatial-contextual approach. Inclusion of spectral transforms increased accuracies by 5 percent for the true object-based approach, up to 13 percent for the spatial contextual approach, and reduced commission error up to 10 percent for both approaches. Masking techniques increased accuracies of the spatial contextual approach by 20 percent. With manual editing, the most accurate maps of individual live and dead trees from the spatial contextual approach are suitable for studying spatio-temporal trends in montane conifer mortality.