Aerial Imagery Research Articles

Black-box adversarial patch attacks using differential evolution against aerial imagery object detectors

Adversarial patch attacks are commonly used to attack aerial imagery object detectors. However, most existing methods are designed for white-box settings, which are impractical in real world scenarios. In this work, we propose a novel framework for black-box adversarial patch attacks against aerial imagery object detectors using differential evolution (DE). Specifically, we first introduce a new dimensionality reduction strategy to address the high-dimensionality curse in DE optimization problems. Then, we design three universal fitness functions to help DE to select better individuals within a limited computational budget. Finally, we conduct extensive experiments on the Dataset for Object Detection in Aerial images (DOTA) against You Only Look Once (YOLOv3, YOLOv4), and Faster Region-based Convolutional Network (Faster R-CNN), utilizing attack success rate (ASR) and average precision (AP) to quantitatively evaluate the attack efficiency. Results on YOLOv3 for the plane category indicate that our proposed strategy achieves at least a 20.97% improvement in ASR compared to existing dimensionality reduction schemes. Moreover, experiments conducted on detectors with different architectures reveal inconsistent adversarial robustness across categories. To the best of our knowledge, this is the first work to explore the use of DE in black-box adversarial patch attacks against aerial imagery object detectors. Our code has been released at https://github.com/tang-agui/Bbox-Att-Detector.

Multi‐decadal changes in vegetation and land use inferred from different data sources: Two case studies at the regional scale in the Slovenian karst

AbstractFew studies have used aerial imagery or various statistical records to analyse temporal land use and vegetation dynamics over large areas. This study presents an innovative approach that merges national‐scale data to provide temporal and spatial digital information on land use and vegetation cover change at high resolution. The work includes 326 historical aerial photographs from 1957 and 1980, derived digital terrain models, digital orthophotos (from 1994 to 1997), current land use (available from 2006), and forestry databases. Using two case studies in Slovenia, a detailed procedure for data preparation, processing (including orthorectification, on‐screen digitisation, visual photointerpretation), and result generation is presented. The used multiple sources showed great potential for the acquisition, mapping and quantifying selected land use categories. The study provided very important material for land use change assessment and implementation of the national documentary heritage. Although the processing methods were time‐consuming, they proved to be very suitable for delineating polygons of change with a spatial resolution of at least 1 m. The results show that the forest in the Rižana catchment has increased by 30 percentage points in the last 60 years and that in the Unica catchment 2/3 of the forest was damaged by large‐scale disturbances in 2014–2018.

Artificial intelligence-based smart agricultural systems for saffron cultivation with integration of Unmanned Aerial Vehicle imagery and deep learning approaches

This paper presents an integrated Unmanned Aerial Vehicle (UAV) imagery system and artificial intelligence-based smart farming for saffron cultivation. This work is about monitoring the growth of saffron flowers through UAV imagery every week or month during the pre-harvesting time while the percentage of the covered area by leaf and saffron flowers is calculated. This is crucial for predicting saffron growing areas over cultivated regions. From the vertical viewing of UAV imagery from above, it is problematic from the bird’s eye perspective to identify the particular saffron-growing areas of things, so the multi-class classification of flowers (subjects) recognition system is proposed. So, the implementation of this work is divided into four phases. The first two phases (Phase 1 and Phase 2) are about the algorithms for Saffron growing region detections and predictions using the modified You Only Look Once (YOLO) higher version model followed by statistical analysis based region corner detections. The outcomes of the first two phases are used to form a database of saffron flowers with different scales, orientations, and deformations. This is further used to build a two-class classification model in the third phase (Phase 3) that discriminates saffron vs. non-saffron flowers. In the fourth phase, the two-class classification model is extended to a multi-class flower recognition model for recognizing 900 flower images database available in the largest social media website dedicated exclusively to gardening in the Phase 4. Extensive experiments are conducted, and the performances are compared with some existing state-of-the-art methods that show the superiority of the proposed system.

SDMNet: Spatially dilated multi-scale network for object detection for drone aerial imagery

Multi-scale object detection is a preeminent challenge in computer vision and image processing. Several deep learning models that are designed to detect various objects miss out on the detection capabilities for small objects, reducing their detection accuracies. Intending to focus on different scales, from extremely small to large-sized objects, this work proposes a Spatially Dilated Multi-Scale Network (SDMNet) architecture for UAV-based ground object detection. It proposes a Multi-scale Enhanced Effective Channel Attention mechanism to preserve the object details in the images. Additionally, the proposed model incorporates dilated convolution, sub-pixel convolution, and additional prediction heads to enhance object detection performance specifically for aerial imaging. It has been evaluated on two popular aerial image datasets, VisDrone 2019 and UAVDT, containing publicly available annotated images of ground objects captured from UAV. Different performance metrics, such as precision, recall, mAP, and detection rate, benchmark the proposed architecture with the existing object detection approaches. The experimental results demonstrate the effectiveness of the proposed model for multi-scale object detection with an average precision score of 54.2% and 98.4% for VisDrone and UAVDT datasets, respectively.

Insights into the Effects of Tile Size and Tile Overlap Levels on Semantic Segmentation Models Trained for Road Surface Area Extraction from Aerial Orthophotography

Studies addressing the supervised extraction of geospatial elements from aerial imagery with semantic segmentation operations (including road surface areas) commonly feature tile sizes varying from 256 × 256 pixels to 1024 × 1024 pixels with no overlap. Relevant geo-computing works in the field often comment on prediction errors that could be attributed to the effect of tile size (number of pixels or the amount of information in the processed image) or to the overlap levels between adjacent image tiles (caused by the absence of continuity information near the borders). This study provides further insights into the impact of tile overlaps and tile sizes on the performance of deep learning (DL) models trained for road extraction. In this work, three semantic segmentation architectures were trained on data from the SROADEX dataset (orthoimages and their binary road masks) that contains approximately 700 million pixels of the positive “Road” class for the road surface area extraction task. First, a statistical analysis is conducted on the performance metrics achieved on unseen testing data featuring around 18 million pixels of the positive class. The goal of this analysis was to study the difference in mean performance and the main and interaction effects of the fixed factors on the dependent variables. The statistical tests proved that the impact on performance was significant for the main effects and for the two-way interaction between tile size and tile overlap and between tile size and DL architecture, at a level of significance of 0.05. We provide further insights and trends in the predictions of the extensive qualitative analysis carried out with the predictions of the best models at each tile size. The results indicate that training the DL models on larger tile sizes with a small percentage of overlap delivers better road representations and that testing different combinations of model and tile sizes can help achieve a better extraction performance.

Livestock Detection and Counting in Kenyan Rangelands Using Aerial Imagery and Deep Learning Techniques

Accurate livestock counts are essential for effective pastureland management. High spatial resolution remote sensing, coupled with deep learning, has shown promising results in livestock detection. However, challenges persist, particularly when the targets are small and in a heterogeneous environment, such as those in African rangelands. This study evaluated nine state-of-the-art object detection models, four variants each from YOLOv5 and YOLOv8, and Faster R-CNN, for detecting cattle in 10 cm resolution aerial RGB imagery in Kenya. The experiment involved 1039 images with 9641 labels for training from sites with varying land cover characteristics. The trained models were evaluated on 277 images and 2642 labels in the test dataset, and their performance was compared using Precision, Recall, and Average Precision (AP0.5–0.95). The results indicated that reduced spatial resolution, dense shrub cover, and shadows diminish the model’s ability to distinguish cattle from the background. The YOLOv8m architecture achieved the best AP0.5–0.95 accuracy of 39.6% with Precision and Recall of 91.0% and 83.4%, respectively. Despite its superior performance, YOLOv8m had the highest counting error of −8%. By contrast, YOLOv5m with AP0.5–0.95 of 39.3% attained the most accurate cattle count with RMSE of 1.3 and R2 of 0.98 for variable cattle herd densities. These results highlight that a model with high AP0.5–0.95 detection accuracy may struggle with counting cattle accurately. Nevertheless, these findings suggest the potential to upscale aerial-imagery-trained object detection models to satellite imagery for conducting cattle censuses over large areas. In addition, accurate cattle counts will support sustainable pastureland management by ensuring stock numbers do not exceed the forage available for grazing, thereby mitigating overgrazing.

AerialFormer: Multi-Resolution Transformer for Aerial Image Segmentation

When performing remote sensing image segmentation, practitioners often encounter various challenges, such as a strong imbalance in the foreground–background, the presence of tiny objects, high object density, intra-class heterogeneity, and inter-class homogeneity. To overcome these challenges, this paper introduces AerialFormer, a hybrid model that strategically combines the strengths of Transformers and Convolutional Neural Networks (CNNs). AerialFormer features a CNN Stem module integrated to preserve low-level and high-resolution features, enhancing the model’s capability to process details of aerial imagery. The proposed AerialFormer is designed with a hierarchical structure, in which a Transformer encoder generates multi-scale features and a multi-dilated CNN (MDC) decoder aggregates the information from the multi-scale inputs. As a result, information is taken into account in both local and global contexts, so that powerful representations and high-resolution segmentation can be achieved. The proposed AerialFormer was benchmarked on three benchmark datasets, including iSAID, LoveDA, and Potsdam. Comprehensive experiments and extensive ablation studies show that the proposed AerialFormer remarkably outperforms state-of-the-art methods.

Potential for Events Similar to the Deadly West Salt Creek Landslide, Grand Mesa Area, Colorado

ABSTRACT The deadly West Salt Creek Landslide of 2014, one of the largest landslides to occur in the United States in historical time, was surprising in its suddenness and length of runout. Its source area, the edge of Grand Mesa in western Colorado, was potentially the source of other similar landslides in the past, based on previous mapping in the region. The purpose of this research was to evaluate the potential for similar events by detailed mapping of several canyons flanking West Salt Creek, using aerial imagery and age-dating techniques not available to previous mappers. We conclude that, while the West Salt Creek Landslide has the potential to reactivate in a similar manner to the 2014 movement, the potential for a similar event in the flanking canyons is low. Mapping and semi-quantitative geomorphic methods suggest that the deposits present in several canyons on either side of West Salt Creek are glacial or stream deposits, with one canyon containing debris-flow deposits. Our interpretations are based on the morphology of deposits in the canyons, the presence of specific landform features, measurements of mobility index and longitudinal profiles, and comparison with regional analogs. Furthermore, other mass movements in the region that are of similar magnitude had either slower movement or much shorter runout. The large block at the head of West Salt Creek and the earlier block that shattered and mobilized in 2014 are also unique features locally, distinguishing this setting from others, but also indicating the potential for a future similar event.

DiRecNetV2: A Transformer-Enhanced Network for Aerial Disaster Recognition

The integration of Unmanned Aerial Vehicles (UAVs) with artificial intelligence (AI) models for aerial imagery processing in disaster assessment, necessitates models that demonstrate exceptional accuracy, computational efficiency, and real-time processing capabilities. Traditionally Convolutional Neural Networks (CNNs), demonstrate efficiency in local feature extraction but are limited by their potential for global context interpretation. On the other hand, Vision Transformers (ViTs) show promise for improved global context interpretation through the use of attention mechanisms, although they still remain underinvestigated in UAV-based disaster response applications. Bridging this research gap, we introduce DiRecNetV2, an improved hybrid model that utilizes convolutional and transformer layers. It merges the inductive biases of CNNs for robust feature extraction with the global context understanding of Transformers, maintaining a low computational load ideal for UAV applications. Additionally, we introduce a new, compact multi-label dataset of disasters, to set an initial benchmark for future research, exploring how models trained on single-label data perform in a multi-label test set. The study assesses lightweight CNNs and ViTs on the AIDERSv2 dataset, based on the frames per second (FPS) for efficiency and the weighted F1 scores for classification performance. DiRecNetV2 not only achieves a weighted F1 score of 0.964 on a single-label test set but also demonstrates adaptability, with a score of 0.614 on a complex multi-label test set, while functioning at 176.13 FPS on the Nvidia Orin Jetson device.

Transfer learning may explain pigeons’ ability to detect cancer in histopathology

Pigeons' unexpected competence in learning to categorize unseen histopathological images has remained an unexplained discovery for almost a decade (Levensonet al2015PLoS One10e0141357). Could it be that knowledge transferred from their bird's-eye views of the earth's surface gleaned during flight contributes to this ability? Employing a simulation-based verification strategy, we recapitulate this biological phenomenon with a machine-learning analog. We model pigeons' visual experience during flight with the self-supervised pre-training of a deep neural network on BirdsEyeViewNet; our large-scale aerial imagery dataset. As an analog of the differential food reinforcement performed in Levensonet al's study 2015PLoS One10e0141357), we apply transfer learning from this pre-trained model to the same Hematoxylin and Eosin (H&E) histopathology and radiology images and tasks that the pigeons were trained and tested on. The study demonstrates that pre-training neural networks with bird's-eye view data results in close agreement with pigeons' performance. These results support transfer learning as a reasonable computational model of pigeon representation learning. This is further validated with six large-scale downstream classification tasks using H&E stained whole slide image datasets representing diverse cancer types.

Agricultural land abandonment linked to pipe collapse and gully development: Reconstruction from archival SfM and LiDAR datasets

Land use and land cover changes such as agricultural land abandonment along with soil erosion are perceived to be important factors that add to land degradation worldwide. In this study, we used aerial imagery, archival-Structure from Motion (SfM) reconstruction and airborne-LiDAR surveys to reconstruct land abandonment along with pipe collapse and gully development in a semi-arid Mediterranean catchment area of the Ebro Valley (Spain) during 1957–2021. Agricultural land dynamics were analysed from the study of planform changes to field crops. Land degradation associated to pipe collapse and gully development was inferred from the geomorphic change detection deduced from the comparison of topographic models obtained from archival-SfM and LiDAR datasets. Through spatial correlation analysis, we identified that higher, steeper, and more hydrologically connected field crop areas were the first to be abandoned. Additionally, the extent of pipe collapses and gullying processes was significantly correlated with the degree of land abandonment over the duration of the study period. Our findings reveal that approximately 20 % of agricultural lands within the study area were abandoned during said study period, with up to 15 % of the abandoned area directly impacted by pipe collapse and gullying processes. Erosion rates associated with these processes within the catchment, implied erosion areas of 3.9 ha and 1.5 ha for the period 1977–2009 and 2009–2021, ranging between 120 Mg/ha yr−1 and 203.2 Mg/ha yr−1, respectively. Our study highlights that abandonment in said area is predominantly conditioned by land degradation resulting from pipe collapses and gully development. The recognition and protection of piping and gully affected areas as geodiversity sites is presented as an alternative at a local level to mitigate the economic impacts of soil degradation. Understanding the effects that pipe collapse and gully development have on arable lands over long-term study periods (i.e., >50 years) is essential in order to shed light on the interconnected factors influencing land productivity and agricultural sustainability. This ultimately guides us to make informed policy decisions that mitigate these detrimental effects.

Russian olive distribution and invasion dynamics along the Powder River, Montana and Wyoming, USA

The invasive shrub, Russian olive (Elaeagnus augustifolia), is widely established within riparian areas across North America and eastern Europe. Limited information on its distribution and invasion dynamics in northern regions has hampered understanding and management efforts. Given this lack of spatial and ecological information we worked with local stakeholders and developed two main objectives: (1) map the distribution of Russian olive along the Powder River (Montana and Wyoming, United States) as of 2020 with field data and remote sensing; and (2) relate that distribution to environmental variables to understand its habitat suitability and community/invasion dynamics. Field data showed Russian olive has reached near equal canopy cover (18.3%) to native Plains cottonwood (Populus deltoides; 19.1%) and has a broader distribution. At the watershed scale, we modeled Russian olive distribution using field surveys, ocular sampling of aerial imagery, and spectral variables from Sentinel-2 MultiSpectral Instrument using a random forest model (RMSE = 15.42, R2 = 0.64). A statistical model linking the resulting Russian olive percent cover detection map to environmental variables for the entire watershed indicated Russian olive cover increased with flow accumulation and decreased with elevation, and was associated with poorer soil types. We attribute the success of Russian olive to its broad habitat suitability combined with changing hydrologic conditions favoring it over natives. The maps of Russian olive cover along the Powder River and its main tributaries in northern Wyoming and southern Montana revealed regions of the watershed with high and low cover, which can guide landscape-scale management prioritization. This study provides a repeatable Russian olive detection method due to the use of Sentinel-2 imagery that is available worldwide and provides insight into Russian olive’s ecological relationships and success with relevance for management across areas with similar environmental conditions.

Historic Flood Events and Current Flood Hazard in Ulaanbaatar City, Central Mongolia

Climate change will increase the frequency of extreme weather events, alter rainfall patterns, and exacerbate flood disasters in Ulaanbaatar City. Here we combine aerial and satellite imagery with cadastral data, to scrutinize the historical trajectory of rainfall patterns and flood disasters in Ulaanbaatar over the past six decades. The study focusses on the causative factors behind historical floods, current flood conditions, the geographical distribution of floods, land ownership in floodprone areas, and the spatial allocation of fences and buildings based on social conditions. Over the last 60 years, Ulaanbaatar received a total of 16,780 mm of precipitation, with a staggering 80.5% of this total occurring during the summer season. Over this period, the city has endured about ten significant flood disasters. The most severe and destructive events occurred in 1966, 1982, 1994, 2003, 2009, and 2023 as river basins and mountain flash floods. These flood events claimed at least 220 lives, affected around 46,000 households, and caused economic losses of ca. 3.3 million U.S. dollars. Our study identifies several flood hazard areas along the Tolgoit, Selbe, Uliastai, and Tuul River valleys, which define a flood buffer zone extending 200 m from their banks, encompassing 59 khoroos of 7 districts in Ulaanbaatar. There are 27,970 fences and 12,887 buildings in the 200 m buffer zone, which is 66.5% of all fence unit area, and 46.3% of the total building, situated within the identified flood risk areas. In response to these findings, we emphasize the urgent need for comprehensive long-term strategy for sustainable flood management based on disaster resilence.

Adaptive and automatic aerial image restoration pipeline leveraging pre-trained image restorer with lightweight Fully Convolutional Network

Earth observation and remote sensing necessitate the use of aerial imaging, and unmanned aerial vehicles (UAVs) are widely utilized for this purpose. Although UAVs can be costly, using inexpensive components could result in issues such as noise, defocusing, and motion blur. Utilizing advanced deep learning algorithms, these images can be effectively restored. However, this necessitates the use of physical effort since each image must be restored manually. Our work introduces a new pipeline that can automatically categorize the degradation types of aerial images and subsequently apply the necessary restoration techniques on degraded images. A lightweight adaptive degradation classifier model was proposed to handle input from images of varying sizes and provide guidance to a image restoration model for achieving the desired task. The degradation classifier is a Fully Convolutional Network capable of categorizing images of any size into 3 distinct classes. It offers guidance to the subsequent image restorer. The restoration model is a pre-trained transformer-based model that efficiently restores different types of image degradations. In addition, we have implemented a pipeline loss to assess the performance of our complete system. To build the system, a dataset is created by extracting cropped patches from high-resolution images obtained through Google Earth. Subsequently, the clean patches were artificially deteriorated by applying different filters to replicate the image degradations. The degradation classifier achieved a mean accuracy of 96.71% across 6 different test scenarios, with a standard deviation of 2.97. The mean pipeline loss was 24.95, with a standard deviation of 1.79. Furthermore, we showcased the efficacy of our pipeline in a practical implementation involving aerial imagery, specifically in the domain of vehicle detection. We conducted training on an object detection model and then evaluated the detection performance on both degraded and automatically restored images. Our experiments demonstrate the efficient performance of our method in automatically restoring degraded aerial images, even though the restoration models were not specifically trained on aerial images. In addition, we examined current limitations and difficulties in creating such systems and offered comprehensive perspectives on future advancements.

Evolution, sedimentation and thermal state of the emerging pro‐glacial lakes at Witenwasserengletscher, Switzerland

AbstractAs glaciers retreat worldwide, local basal depressions are exposed where new pro‐glacial lakes are forming. Although the formation of such new lakes has been mapped widely, the impact of their interaction with the glacier on sediment dynamics and the thermal state is rarely studied. At the example of Witenwasserengletscher, Switzerland, we use historic aerial imagery and a bathymetric survey to investigate in detail the interaction of glacier retreat, lake formation and sedimentation. Three lakes have emerged since the mid‐1990s with mean depths between 1.25 and 6 m. The deepest lake lost contact to the ice in 2009 with a delta forming from mobilized sub‐glacial sediment. Phases of direct contact of the glacier with the lakes are found to increase terminus retreat and affect the thermal state and sedimentation. In summer, lake water temperatures in these small ice‐contact lakes stay close to 0°C, whereas the disconnected eastern lake shows temperatures consistently over 6°C. Temperature profiles from 2021 show that after losing ice contact, warm sediment‐rich glacial water forms an underflow that is breaking through the summertime stratification with warm water over cold water. In this case, density stratification is dominated by suspended sediment rather than temperature. Sedimentation shows a high multiannual variability and is dependent on a multitude of factors, ultimately on the routing of glacial streams. These tend to migrate towards the centre of the valley as the glacier retreats. Our study shows that during deglaciation lake evolution, sediment redistribution and the thermal state are highly dynamic and pro‐glacial lakes strongly affect the sediment evolution and may thereby impact on the ecosystem in pro‐glacial streams.

Rice Response to Spermine Foliar Application and Its Association with Aerial Imagery Monitoring Under Water Stress Conditions

Rice is the most consumed food in the world, mainly in Asia and Africa. Malaysia is the second-largest rice importer in Southeast Asia after Indonesia. However, rice yield is limited by water stress. One alternative for a quicker strategy to mitigate water stress is through a combination of foliar spermine application and efficient rice management practices via image monitoring techniques using drone technology. The present study was aimed at evaluating the effects of spermine on rice physiological response and its association with aerial imagery and yield under reproductive during reproductive stage under water stress. The experiment was carried out under greenhouse conditions using a two-factorial randomized complete block design (RCBD), with foliar spermine treatment as the first factor and water stress as the second factor. Physiological parameters showed significantly higher tiller number per pot and photosynthesis rate by 29% and 31%, respectively. Correspondingly, the Normalised Difference Vegetation Index (NDVI) using aerial imagery monitoring showed an increased value in spermine treatments by 2% compared to control. Furthermore, NDVI readings and photosynthetic rate were positively correlated linearly with R2= 0.51. Interestingly, spermine treatments alleviated water stress effects by 40%, 17% and 12% in grain weight per pot, grain number per panicle and percentage filled grain. Biomass partitioning in roots improved by 44% in spermine treatments, even under water stress, due to an efficient translocation of assimilates. In conclusion, spermine foliar application significantly improved growth, grain filling and rice yield production, which was also supported by NDVI values using aerial imagery monitoring.

Habitat changes in response to pressures in the Verlorenvlei Estuarine Lake, South Africa

Verlorenvlei is a globally important RAMSAR wetland on the arid west coast of South Africa. A prolonged drought between 2016 and 2021 and increasing competition for water from the agricultural sector resulted in extremely low water levels. We used historical aerial and satellite imagery from 1942 and rainfall and water level data for the past 50 years, to assess habitat changes within the estuarine functional zone of the Verlorenvlei Estuarine Lake. Prior to the drought, lake water levels and water surface area remained stable (1113 ± 27 ha (SE)). Since then, there has been a 64% decrease in open water area, exposing 193 ha of sandbanks, of which 190.9 ha were hyper-sulfidic pyrite-rich. The water in the lower lake was hypersaline ( >100), and in the middle, it was acidic (pH<3). The low water level plus sediment and nutrient input from surrounding agriculture resulted in a localised increase in reeds. Additional pressures, such as fires, have reduced the above-ground biomass of reeds and sedges, potentially altering surface morphology and reducing stored carbon. Despite flooding and filling up in June 2023, the lake remained in an acidic state (3.9–4.3). Similar low-lake level, hypersaline and acidic conditions are predicted to become more common under future climate change scenarios where aridity and extreme weather events are anticipated. Inflow of fresh water into the estuary and control of farming practices are required to keep the Verlorenvlei in a functional state, with long-term monitoring necessary to assess the ecological condition in response to restoration actions.

Impact of Tile Size and Tile Overlap on the Prediction Performance of Convolutional Neural Networks Trained for Road Classification

Popular geo-computer vision works make use of aerial imagery, with sizes ranging from 64 × 64 to 1024 × 1024 pixels without any overlap, although the learning process of deep learning models can be affected by the reduced semantic context or the lack of information near the image boundaries. In this work, the impact of three tile sizes (256 × 256, 512 × 512, and 1024 × 1024 pixels) and two overlap levels (no overlap and 12.5% overlap) on the performance of road classification models was statistically evaluated. For this, two convolutional neural networks used in various tasks of geospatial object extraction were trained (using the same hyperparameters) on a large dataset (containing aerial image data covering 8650 km2 of the Spanish territory that was labelled with binary road information) under twelve different scenarios, with each scenario featuring a different combination of tile size and overlap. To assess their generalisation capacity, the performance of all resulting models was evaluated on data from novel areas covering approximately 825 km2. The performance metrics obtained were analysed using appropriate descriptive and inferential statistical techniques to evaluate the impact of distinct levels of the fixed factors (tile size, tile overlap, and neural network architecture) on them. Statistical tests were applied to study the main and interaction effects of the fixed factors on the performance. A significance level of 0.05 was applied to all the null hypothesis tests. The results were highly significant for the main effects (p-values lower than 0.001), while the two-way and three-way interaction effects among them had different levels of significance. The results indicate that the training of road classification models on images with a higher tile size (more semantic context) and a higher amount of tile overlap (additional border context and continuity) significantly impacts their performance. The best model was trained on a dataset featuring tiles with a size of 1024 × 1024 pixels and a 12.5% overlap, and achieved a loss value of 0.0984, an F1 score of 0.8728, and an ROC-AUC score of 0.9766, together with an error rate of 3.5% on the test set.

Using deep learning for enrichment of heritage BIM: Al Radwan house in historic Jeddah as a case study

Building information modeling (BIM) can greatly improve the management and planning of historic building conservation projects. However, implementing BIM in the heritage has many challenges, including issues with modeling irregular features, surveying data occlusions, and a lack of predefined libraries of parametric objects. Indeed, surface features can be manually distinguished and segmented depending on the level of human involvement during data scanning and BIM processing. This requires a significant amount of time and resources, as well as the risk of making too subjective decisions. To address these bottlenecks and improve BIM digitization of building geometry, a novel deep learning based scan-to-HBIM workflow is used during the recording of the historic building in historic Jeddah, Saudi Arabia, a UNESCO World Heritage site. The proposed workflow enables access to laser scanner and unmanned aerial vehicle imagery data to create a complete integrated survey using high-resolution imagery acquired independently at the best position and time for proper radiometric information to depict the surface features. By employing deep learning with orthophotos, the method significantly improves the interpretation of spatial weathering forms and façade degradation. Additionally, an HBIM library for Saudi Hijazi architectural elements is created, and the vector data derived from deep learning-based segmentation are accurately mapped onto the HBIM geometry with relevant statistical parameters. The findings give stakeholders an effective tool for identifying the types, nature, and spatial extent of façade degradation to investigate and monitor the structure.

Assessing controls on sedimentation rates and sediment organic carbon accretion in beaver ponds

Beaver dams trap sediment, promote channel-floodplain connectivity, modify biogeochemical cycling and organic carbon (OC) storage, and influence geomorphic form. Beaver-related sediment accumulation has been investigated at longer timescales (e.g., > 1000 years) and shorter timescales (< 10 years), but we lack information on sedimentation and sediment-associated OC accretion rates over multiple decades in relatively persistent beaver ponds (10–100 years old). We coupled field surveys of 45 beaver ponds with historical aerial imagery and radiometric dating with 7Be, 210Pb, and 14C to calculate sedimentation rates, mean sediment depth, and sediment OC content at two study sites in the southern Rocky Mountains, USA. Sedimentation rates in beaver ponds (median = 5.7 cm yr−1, mean = 11.6 cm yr−1) decreased with pond age. Incised, single threaded reaches had greater variability in mean sediment depth compared to less incised reaches. In less incised reaches, mean sediment depth and beaver dam height increased with pond age, indicating more stable dams and depositional environments. Sediment OC content within beaver ponds (median = 0.8 %, mean = 1.7 %) increased with finer sediment grain size distributions. Sediment OC accretion rates in ponds ranged between 0.13 and 23 Mg C ha −1 per year. We used Monte Carlo simulations to estimate it would take ∼100 years or more of uninhibited beaver activity for deposition to laterally reconnect adjacent terraces in the incised study reaches, a common objective within many stream restoration projects. Our findings show that beaver ponds in complex, multi-threaded reaches better retain fine sediment over longer timescales, highlighting the need to incorporate geomorphic context when considering whether beaver can help restore incised river channels and floodplain connectivity, retain fine sediment, and store OC on the landscape.