Developing a Machine Learning-Based Framework for Roadway Vulnerability and Impact Assessment Using Aerial Imagery

Automated remote sensing methods have not gained widespread usage for damage assessment after hurricane events, especially for low-rise buildings, such as individual houses and small businesses. Hurricane wind, storm surge with waves, and inland flooding have unique damage signatures, further complicating the development of robust automated assessment methodologies. As a step toward realizing automated damage assessment for multi-hazard hurricane events, this paper presents a mono-temporal image classification methodology that quickly and accurately differentiates urban debris from non-debris areas using post-event images. Three classification approaches are presented: spectral, textural, and combined spectral–textural. The methodology is demonstrated for Gulfport, Mississippi, using IKONOS panchromatic satellite and NOAA aerial colour imagery collected after 2005 Hurricane Katrina. The results show that multivariate texture information significantly improves debris class detection performance by decreasing the confusion between debris and other land cover types, and the extracted debris zone accurately captures debris distribution. Additionally, the extracted debris boundary is approximately equivalent regardless of imagery type, demonstrating the flexibility and robustness of the debris mapping methodology. While the test case presents results for hurricane hazards, the proposed methodology is generally developed and expected to be effective in delineating debris zones for other natural hazards, including tsunamis, tornadoes, and earthquakes.

Auto-identification of linear archaeological traces of the Great Wall in northwest China using improved DeepLabv3+ from very high-resolution aerial imagery

This research delves into the application of semantic segmentation in precision agriculture, specifically targeting the automated identification and classification of various irrigation system types within agricultural landscapes using high-resolution aerial imagery. With irrigated agriculture occupying a substantial portion of US land and constituting a major freshwater user, the study's background highlights the critical need for precise water-use estimates in the face of evolving environmental challenges, the study utilizes advanced computer vision for optimal system identification. The outcomes contribute to effective water management, sustainable resource utilization, and informed decision-making for farmers and policymakers, with broader implications for environmental monitoring and land-use planning. In this geospatial evaluation research, we tackle the challenge of intraclass variability and a limited dataset. The research problem centers around optimizing the accuracy in geospatial analyses, particularly when confronted with intricate intraclass variations and constraints posed by a limited dataset. Introducing a novel approach termed "dynamic contrastive learning," this research refines the existing contrastive learning framework. Tailored modifications aim to improve the model's accuracy in classifying and segmenting geographic features accurately. Various deep learning models, including EfficientNetV2L, EfficientNetB7, ConvNeXtXLarge, ResNet-50, and ResNet-101, serve as backbones to assess their performance in the geospatial context. The data used for evaluation consists of high-resolution aerial imagery from the National Agriculture Imagery Program (NAIP) captured in 2015. It includes four bands (red, green, blue, and near-infrared) with a 1-meter ground sampling distance. The dataset covers diverse landscapes in Lonoke County, USA, and is annotated for various irrigation system types. The dataset encompasses diverse geographic features, including urban, agricultural, and natural landscapes, providing a representative and challenging scenario for model assessment. The experimental results underscore the efficacy of the modified contrastive learning approach in mitigating intraclass variability and improving performance metrics. The proposed method achieves an average accuracy of 96.7%, a BER of 0.05, and an mIoU of 88.4%, surpassing the capabilities of existing contrastive learning methods. This research contributes a valuable solution to the specific challenges posed by intraclass variability and limited datasets in the realm of geospatial feature classification. Furthermore, the investigation extends to prominent deep learning architectures such as Segformer, Swin Transformer, Convexnext, and Convolution Vision Transformer, shedding light on their impact on geospatial image analysis. ConvNeXtXLarge emerges as a robust backbone, demonstrating remarkable accuracy (96.02%), minimal BER (0.06), and a high MIOU (85.99%).

Accurate building footprint polygons provide essential data for a wide range of urban applications. While deep learning models have been proposed to extract pixel-based building areas from remote sensing imagery, the direct vectorization of pixel-based building maps often leads to building footprint polygons with irregular shapes that are inconsistent with real building boundaries, making it difficult to use them in geospatial analysis. In this study, we proposed a novel deep learning-based framework for automated extraction of building footprint polygons (DLEBFP) from very high-resolution aerial imagery by combining deep learning models for different tasks. Our approach uses the U-Net, Cascade R-CNN, and Cascade CNN deep learning models to obtain building segmentation maps, building bounding boxes, and building corners, respectively, from very high-resolution remote sensing images. We used Delaunay triangulation to construct building footprint polygons based on the detected building corners with the constraints of building bounding boxes and building segmentation maps. Experiments on the Wuhan University building dataset and ISPRS Vaihingen dataset indicate that DLEBFP can perform well in extracting high-quality building footprint polygons. Compared with the other semantic segmentation models and the vector map generalization method, DLEBFP is able to achieve comparable mapping accuracies with semantic segmentation models on a pixel basis and generate building footprint polygons with concise edges and vertices with regular shapes that are close to the reference data. The promising performance indicates that our method has the potential to extract accurate building footprint polygons from remote sensing images for applications in geospatial analysis.

Photovoltaic panel extraction from very high-resolution aerial imagery using region–line primitive association analysis and template matching

This study investigates the application of deep learning algorithms and high-resolution aerial imagery for individual tree detection in urban areas, using a neighborhood in Mersin, Turkey, as a case study. Employing the DeepForest Python package, we utilize high-resolution (7cm) aerial imagery to detect and map the city's tree population accurately. The results showcase an impressive accuracy rate of 80.87%, demonstrating the potential of deep learning in urban forestry applications and contributing to effective urban planning. The information generated from this study is crucial for conserving urban green spaces, enhancing resilience to climate change, and supporting urban biodiversity. While this research is focused on Mersin, the methods employed are globally adaptable, laying a foundation for further refinement and potential identification of different tree species in future work. This investigation highlights the transformative role of advanced technology in fostering sustainable urban environments.

Transforming the global energy sector from fossil-fuel based to renewable energy sources is crucial to limiting global warming and achieving climate neutrality. The decentralized nature of the renewable energy system allows private households to deploy photovoltaic systems on their rooftops. However, inconsistent data on installed photovoltaic (PV) systems complicate planning for an efficient grid expansion. To address this issue, deep-learning techniques, can support collecting data about PV systems from aerial and satellite imagery. Previous research, however, lacks the consideration for ground truth data-specific characteristics of PV panels. This study aims to implement a semantic segmentation model that detects PV systems in aerial imagery to explore the impact of area-specific characteristics in the training data and CNN hyperparameters on the performance of a CNN. Hence, a U-Net architecture is employed to analyze land use types, rooftop colors, and lower-resolution images. Additionally, the impact of near-infrared data on the detection rate of PV panels is analyzed. The results indicate that a U-Net is suitable for classifying PV panels in high-resolution aerial imagery (10 cm) by reaching F1 scores of up to 91.75% while demonstrating the importance of adapting the training data to area-specific ground truth data concerning urban and architectural properties.

Recent interest in assessing coral reef functions has raised questions about how carbonate production rates have altered over the past few decades of ecological change. At the same time, there is growing interest in quantifying carbonate production on larger reef-scales. Resolving these issues is challenging because carbonate production estimates require three-dimensional survey data, which are typically collected in-situ over small spatial scales. In contrast, data that can be extracted from archive photograph or video imagery and high-resolution aerial imagery are generally planar. To address this disconnect, we collected data on the relationship between linear planar and 3D contour lengths of 62 common Indo-Pacific hard coral genera-morphotypes to establish appropriate conversion metrics (i.e. coral class rugosity values, hereafter termed Rcoral). These conversion values allow planar colony dimensions to be converted to estimates of 3D colony contour length, which can be employed within existing census budget methodologies like ReefBudget to estimate coral carbonate production (G, in kg CaCO3 m−2 yr−1). We tested this approach by comparing in-situ carbonate production data collected using the ReefBudget methodology against estimates derived from converted colony length data from video imagery. The data show a high level of consistency with an error of ~ 10%. We then demonstrate potential applications of the conversion metrics in two examples, the first using time-series (2006 to 2018) photo-quadrat imagery from Moorea, and the second using high-resolution drone imagery across different reef flat habitats from the Maldives. Whilst some degree of error must necessarily be accepted with such conversion techniques, the approach presented here offers exciting potential to calculate coral carbonate production: (1) from historical imagery to constrain past coral carbonate production rates; (2) from high quality aerial imagery for spatial up-scaling exercises; and (3) for use in rapid photograph or video-based assessments along reef systems where detailed surveys are not possible.

Extracting buildings automatically from high-resolution aerial images is a significant and fundamental task for various practical applications, such as land-use statistics and urban planning. Recently, various methods based on deep learning, especially the fully convolution networks, achieve impressive scores in this challenging semantic segmentation task. However, the lack of global contextual information and the careless upsampling method limit the further improvement of the performance for building extraction task. To simultaneously address these problems, we propose a novel network named Efficient Non-local Residual U-shape Network(ENRU-Net), which is composed of a well designed U-shape encoder-decoder structure and an improved non-local block named asymmetric pyramid non-local block (APNB). The encoder-decoder structure is adopted to extract and restore the feature maps carefully, and APNB could capture global contextual information by utilizing self-attention mechanism. We evaluate the proposed ENRU-Net and compare it with other state-of-the-art models on two widely-used public aerial building imagery datasets: the Massachusetts Buildings Dataset and the WHU Aerial Imagery Dataset. The experiments show that the accuracy of ENRU-Net on these datasets has remarkable improvement against previous state-of-the-art semantic segmentation models, including FCN-8s, U-Net, SegNet and Deeplab v3. The subsequent analysis also indicates that our ENRU-Net has advantages in efficiency for building extraction from high-resolution aerial images.

We present a new framework for creating lane-level detailed HD-maps at scale for autonomous vehicles (AVs). In order to overcome scaling challenges of ground survey-based HD-map creation, we propose a number of innovations by leveraging three data sources: high-resolution aerial imagery, aggregated vehicle telemetry, and a navigation map. We first divide map creation problem into several categories based on lane configurations. The road category is predicted in a supervised setting using aerial imagery, which is pre-processed by using aggregated vehicle telemetry without supervision. We utilized the navigation map to guide the process along the road network and used aerial imagery and aggregated vehicle telemetry to extract lane level features at each step. We propose a multi-task convolutional neural network (CNN) to predict road polygons, road-way centerline, number of lanes, lane and road edges using both the aerial imagery and the corresponding aggregated vehicle telemetry. The predicted road features for each image are then stitched along a road segment to construct the road and lane edge polylines, which are then used to predict lane marking and road edge types in a sliding window fashion along the road segment. The extracted features are finally utilized to calculate higher level features for each point in the HD-map. Our experimental results show that the proposed framework works well, offering a flexible solution for creating HD-maps for AVs at scale.

Mental health is largely shaped by the daily environments in which people live their lives, with positive components of mental health emphasising the importance of feeling good and functioning effectively. Promoting mental health relies on broad-based actions over multiple sectors, which can be difficult to measure. Different types of Impact Assessment (IA) frameworks allow for a structured approach to evaluating policy actions on different levels. A systematic review was performed exploring existing IA frameworks relating to mental health and mental wellbeing and how they have been used. A total of 145 records were identified from the databases, from which 9 articles were included in the review, with a further 6 studies included via reference list and citation chaining. Five different mental-health-related IA frameworks were found to be implemented in a variety of settings, mostly in relation to evaluating community actions. A Narrative Synthesis summarised key themes from the 15 included articles. Findings highlight the need for participatory approaches in IA, which have the dual purpose of informing the IA evaluation and advocating for the need to include mental health in policy development. However, it is important to ensure that IA frameworks are user-friendly, designed to be used by laypeople in a variety of sectors and that IA frameworks are operational in terms of time and monetary resources.

Remote sensing is increasingly being used to create large-scale forest descriptions. In New Zealand, where radiata pine ( Pinus radiata ) plantations dominate the forestry sector, the current national forest description lacks spatially explicit information and struggles to capture data on small-scale forests. This is important as these forests are expected to contribute significantly to future wood supply and carbon sequestration. This study demonstrates the development of a spatially explicit, remote sensing-based forest description for the Gisborne region, a major forest growing area. We combined deep learning-based forest mapping using high-resolution aerial imagery with regional airborne laser scanning (ALS) data to map all planted forest and estimate key attributes. The deep learning model accurately delineated planted forests, including large estates, small woodlots, and newly established stands as young as 3-years post planting. It achieved an intersection over union of 0.94, precision of 0.96, and recall of 0.98 on a withheld dataset. ALS-derived models for estimating mean top height, total stem volume, and stand age showed good performance ( R 2 = 0.94, 0.82, and 0.94 respectively). The resulting spatially explicit forest description provides wall-to-wall information on forest extent, age, and volume for all sizes of forest. This enables stratification by key variables for wood supply forecasting, harvest planning, and infrastructure investment decisions. We propose satellite-based harvest detection and digital photogrammetry to continuously update the initial forest description. This methodology enables near real-time monitoring of planted forests at all scales and is adaptable to other regions with similar data availability. • Deep learning model accurately maps planted radiata forests from aerial imagery. • Lidar-based models predict key forest metrics with high accuracy. • Spatially explicit regional forest description captures large and small forests. • Spatial forest description enables stratification and enrichment for analysis. • Remote sensing framework proposed for updatable national forest description.

Mobile Mapping (MM) is a technique to obtain geo-information using sensors mounted on a mobile platform or vehicle. The mobile platform’s position is provided by the integration of Global Navigation Satellite Systems (GNSS) and Inertial Navigation Systems (INS). However, especially in urban areas, building structures can obstruct a direct line-of-sight between the GNSS receiver and navigation satellites resulting in an erroneous position estimation. Therefore, derived MM data products, such as laser point clouds or images, lack the expected positioning reliability and accuracy. This issue has been addressed by many researchers, whose aim to mitigate these effects mainly concentrates on utilising tertiary reference data. However, current approaches do not consider errors in height, cannot achieve sub-decimetre accuracy and are often not designed to work in a fully automatic fashion. We propose an automatic pipeline to rectify MM data products by employing high resolution aerial nadir and oblique imagery as horizontal and vertical reference, respectively. By exploiting the MM platform’s defective, and therefore imprecise but approximate orientation parameters, accurate feature matching techniques can be realised as a pre-processing step to minimise the MM platform’s three-dimensional positioning error. Subsequently, identified correspondences serve as constraints for an orientation update, which is conducted by an estimation or adjustment technique. Since not all MM systems employ laser scanners and imaging sensors simultaneously, and each system and data demands different approaches, two independent workflows are developed in parallel. <br><br> Still under development, both workflows will be presented and preliminary results will be shown. The workflows comprise of three steps; feature extraction, feature matching and the orientation update. In this paper, initial results of low-level image and point cloud feature extraction methods will be discussed as well as an outline of the project and its framework will be given.

Coastal shorelines are complex and valuable environments that are ever changing. Remote sensing technologies such as satellites and aerial imaging sensors provide an aerial perspective for mapping and analyzing shoreline change. This study aims to compare two accessible and widely used remote sensing data sources- MAXAR WorldView 2 satellite (WV2) and National Agriculture Imagery Program (NAIP) aerial imagery, while also identifying shoreline change trends adjacent to North Inlet, South Carolina. We found that the NAIP and WV2 imagery were very similar in their ability to provide a reliable backdrop for digitizing shoreline data, albeit with differing costs and temporal considerations. In the case study application of using the imagery to detect shoreline change at North inlet over a 12-year period, we found that the three northern regions of the study area experienced overall average erosional impacts of −2.1 m, −4.5 m, and −11.3 m per year. The southernmost region experienced an overall average accretional rate of + 10.6 m. These changes were like a few nearby inlet-adjacent shoreline change studies and are considered to be caused by storm and tidal events, natural sediment drift, and other the dynamic influences of the inlet.

[EMBARGOED UNTIL 08/01/2025] The rapid urbanization of landscapes presents significant environmental and ecological challenges, which necessitates the precise mapping of impervious surfaces. This study aims to leverage the capabilities of deep learning, specifically U-Net architectures, for the semantic segmentation of impervious surfaces from high-resolution aerial imagery. Our approach enhances the conventional U-Net architecture to improve its efficiency and accuracy in handling the spatial complexity of urban landscapes. We utilized a dataset comprising aerial images from urban and suburban areas within the city limits of Columbia, Missouri. The datasets are annotated for training, validation, and testing. The models evaluated include U-Net, Residual U-Net, Attention U-Net, Attention Residual U-Net, scSE U-Net, and scSE Residual U-Net. The models were trained and validated under different conditions, with performance metrics calculated for the overall testing datasets. Our results from the 2017 (leaf-on conditions) and 2019 (leaf-off conditions) test sets indicate improvements. The scSE Residual U-Net model achieved the highest IoU of 88.0 percent, precision of 93.6 percent, recall of 93.6 percent, F1 score of 93.6 percent, and pixel accuracy of 94.6 percent. These results underscore the effectiveness of advanced U-Net models in delineating impervious surfaces with high precision and highlight their potential in remote sensing applications for urban planning and environmental monitoring.