Earth & Space Science

Permanent URI for this collection

https://hdl.handle.net/10315/27626

Browse

Now showing 1 - 11 of 11

Open Access
Active Reinforcement Learning for the Semantic Segmentation of Images Captured by Mobile Sensors
(2023-03-28) Jodeiri Rad, Mahya; Armenakis, Costas
Neural Networks have been employed to attain acceptable performance on semantic segmentation. To perform well, many supervised learning algorithms require a large amount of annotated data. Furthermore, real-world datasets are frequently severely unbalanced, resulting in poor detection of underrepresented classes. The annotation task requires time-consuming human labor. This thesis investigates the use of a reinforced active learning as region selection method to reduce human labor while achieving competitive results. A Deep Query Network (DQN) is utilized to identify the best strategy to label the most informative regions of the image. A Mean Intersection over Union (MIoU) training performance equivalent to 98% of the fully supervised segmentation network was achieved with labeling only 8% of dataset. Another 8% of labelled dataset was used for training the DQN. The performance of all three segmentation networks trained with regions selected by Frequency Weighted Average (FWA) IoU is better in comparison with baseline methods.
Open Access
Architectural Heritage Exploration and Visualization Using Interactive and Immersive Technologies
(2024-07-18) Albourae, Abdullah Taha; Armenakis, Costas
The Historic District of Jeddah (HDJ) of the Kingdom of Saudi Arabia, is facing many challenges to attain conservation and sustainability, due to natural disasters and the absence of appropriate digital recordings and documentation. Previous studies on world heritage conservation attempted to address these challenges while adopting the immersive and visual technologies separately or via merging one of them with another to conserve the heritage sites (e.g. BIM/GIS integration or generating GIS/BIM, integrating VR with BIM, or GIS). However, few of these studies handled the case of the historic heritage of KSA. This research aims to fill this gap by utilizing all these technologies together to create an immersive visualization system for heritage sites. The researcher proposes a novel method using immersive technologies (VR, AR, MR) and visualization methods (3D modeling) combined with Heritage Building Information Modelling (HBIM) and Heritage Geographic Information Systems (HGIS) for comprehensive conservation of cultural heritage sites in Saudi Arabia. The proposed method involves three phases: a) HBIM and HGIS generation: This involves creating 3D architectural models and spatial data using industry-standard collection tools, b) HBIM and HGIS Integration: The data from both phases is merged into a single georeferenced system for unified analysis, and c) Interactive User Experience: An interactive user environment is developed using a game engine, allowing users to explore the 3D model and access historical information. This research offers a multi-disciplinary approach to cultural heritage conservation, combining innovative informatics with technology for documentation, communication, and user engagement. The findings will enable users to virtually explore the heritage sites and track changes over time. This immersive experience fosters a deeper understanding and appreciation of the architectural heritage. The proposed method can be applied to other cultural heritage sites globally. Future studies could explore using 3D viewers for information retrieval and potentially using CityGML, a rich 3D geospatial data format, for even more comprehensive HBIM-HGIS integration. This would significantly contribute to the advancement of world heritage conservation efforts.
Open Access
Automatic Alignment of 3D Multi-Sensor Point Clouds
(2018-03-01) Persad, Ravi Ancil; Armenakis, Costas
Automatic 3D point cloud alignment is a major research topic in photogrammetry, computer vision and computer graphics. In this research, two keypoint feature matching approaches have been developed and proposed for the automatic alignment of 3D point clouds, which have been acquired from different sensor platforms and are in different 3D conformal coordinate systems. The first proposed approach is based on 3D keypoint feature matching. First, surface curvature information is utilized for scale-invariant 3D keypoint extraction. Adaptive non-maxima suppression (ANMS) is then applied to retain the most distinct and well-distributed set of keypoints. Afterwards, every keypoint is characterized by a scale, rotation and translation invariant 3D surface descriptor, called the radial geodesic distance-slope histogram. Similar keypoints descriptors on the source and target datasets are then matched using bipartite graph matching, followed by a modified-RANSAC for outlier removal. The second proposed method is based on 2D keypoint matching performed on height map images of the 3D point clouds. Height map images are generated by projecting the 3D point clouds onto a planimetric plane. Afterwards, a multi-scale wavelet 2D keypoint detector with ANMS is proposed to extract keypoints on the height maps. Then, a scale, rotation and translation-invariant 2D descriptor referred to as the Gabor, Log-Polar-Rapid Transform descriptor is computed for all keypoints. Finally, source and target height map keypoint correspondences are determined using a bi-directional nearest neighbour matching, together with the modified-RANSAC for outlier removal. Each method is assessed on multi-sensor, urban and non-urban 3D point cloud datasets. Results show that unlike the 3D-based method, the height map-based approach is able to align source and target datasets with differences in point density, point distribution and missing point data. Findings also show that the 3D-based method obtained lower transformation errors and a greater number of correspondences when the source and target have similar point characteristics. The 3D-based approach attained absolute mean alignment differences in the range of 0.23m to 2.81m, whereas the height map approach had a range from 0.17m to 1.21m. These differences meet the proximity requirements of the data characteristics and the further application of fine co-registration approaches.
Open Access
Deep Learning-Enhanced Autonomous Aerial and Ground Robotics Using UWB and Lidar in GNSS-Denied Environments
(2024-11-07) Arjmandi, Zahra; Sohn, Gunho; Armenakis, Costas
Over the last decade, advancements in Unmanned Aerial Vehicle (UAV) technology and Artificial Intelligence (AI) have led to significant improvements in navigation and positioning, yet widespread adoption remains limited due to challenges in integrating various technologies and ensuring reliable real-time data processing. This thesis addresses these issues by developing a comprehensive framework that merges advanced data collection platforms, deep learning algorithms, and novel fusion methods to enhance UAV positioning accuracy and reliability. A central contribution of this research is the creation of the Q-Drone Ultra-Wideband (UWB) benchmark dataset. This dataset, generated from a UAV equipped with five UWB sensors across five diverse environments (indoor, outdoor, and semi-outdoor) over a distance of 4 km, provides a standardized benchmark for testing UAV positioning systems. It enables researchers to develop and validate algorithms under varied conditions, supporting advancements in UAV navigation and positioning research. The thesis also introduces an incremental smoothing approach, integrating high-rate and low-rate UWB measurements with inertial data within a unified pose graph framework. This method, using an "add-after-eliminating" strategy, reduces Mean Absolute Error (MAE) by 0.2 meters compared to baseline multilateration methods and achieves a 0.3-meter MAE reduction compared to two-factor pose graph methods. Further, the DeepCovPG framework is developed, combining a Variational Autoencoder (VAE) with a Long Short-Term Memory (LSTM) network to predict and incorporate dynamic covariances into the pose graph. This approach results in a 48% reduction in Root Mean Square Error (RMSE) and a 51% reduction in Range Covariance RMSE, with notable improvements of 0.41 meters in tunnels and 0.23 meters in fields. The framework also achieves a 26% reduction in multilateration RMSE and a 32% reduction in multilateration Covariance RMSE. Additionally, the thesis explores Light Detection and Ranging (LiDAR)-based positioning and proposes the INAF fusion method. This method dynamically selects relevant information from geometric and AI-based odometry techniques, improving accuracy by 3.90% over direct fusion methods and 0.25% over attention-based fusion methods. The INAF fusion method demonstrates enhanced adaptability to various driving conditions, improving accuracy in both straight and dynamic environments.
Open Access
Geo-information identification for exploring non-stationary relationships between volcanic sedimentary Fe mineralization and controlling factors in an area with overburden in eastern Tianshan region, China
Zhao, Jie; Cheng, Qiuming; Javis, Gary T.; Armenakis, Costas
GIS-based spatial analysis has been a common practice in mineral exploration, by which mineral potentials can be delineated to support following sequences of exploration. Mineral potential mapping is generally composed of geo-information extraction and integration. Geological anomalies frequently indicate mineralization. Volcanic sedimentary Fe deposits in eastern Tianshan mineral district, China provide an example of such an indication. However, mineral exploration in this area has been impeded by the desert coverage and geo-anomalies indicative to the presence of mineralization are often weak and may not be efficiently identified by traditional exploring methods. Furthermore, geological guidance regarding to spatially non-stationary relationships between Fe mineralization and its controlling factors were not sufficiently concerned in former studies, which limited the application of proper statistics in mineral exploration. In this dissertation, geochemical distributions associated with controlling factors of the Fe mineralization are characterized by various GIS-based spatial analysis methods. The singularity index mapping technique is attempted to separate geochemical anomalies from background, especially in the desert covered areas. Principal component analysis is further used in integrating the geochemical anomalies to identify geo-information of geological bodies or geological activities associated with Fe mineralization. In order to delineate mineral potentials, spatially weighted principal component analysis with more geological guidance is tried to integrate these identified controlling factors. At the end, as the first time been introduced to mineral exploration, a geographically weighted regression method is currently attempted investigate spatially non-stationary interrelationships presented across the space. Based on the results, superimposition of these controlling factors can be qualitatively and quantitatively summarized that provides a constructive geo-information to Fe mineral exploration in this area. From the practices in this dissertation, GIS-based mineral exploration will not only be efficient in mapping mineral potentials but also be supportive to strategies making of following mineral exploration. All of these experiences can be suggested to future mineral exploration in the other regions.
Open Access
Image-Based Spatial Change Detection Using Deep Learning
(2023-12-08) Bousias- Alexakis, Evangelos; Armenakis, Costas
Image change detection is an invaluable tool in monitoring and understanding the built and physical environments and supporting decision-making. Many recent research approaches for automatic change detection have been based on Deep Learning (DL) techniques and especially on variations of Convolutional Neural Network (CNN) architectures. CNNs have achieved notable success thanks to their great representational capacity, straightforward training, and state-of-the-art performance in visual tasks. Nevertheless, CNNs, like most DL approaches, still face limitations relating to their reliance on extensive labelled datasets, the localization accuracy and detail of the predicted outputs, and the notable model performance degradation when the target data have different characteristics from the training data. This research contributes to the development and evaluation of novel DL methods and algorithms for automated image-based spatial change detection. It investigates novel architectures for enhanced model performance and ways to address the limited availability of labelled data and improve model generalizability. Two main approaches are investigated: (i) a direct approach, where both instances are fed into the DL model that is trained to output the prediction of the change map, and (ii) a post-classification approach, where change detection is performed based on each epoch’s semantic segmentation predictions. In both cases, novel, enhanced CNN architectures have been proposed that leverage the semantic information of objects’ boundaries to improve the accuracy of the model’s predictions. Furthermore, training frameworks inspired by self-ensembling and the Mean Teacher method were developed for semi-supervised learning and domain adaptation, attenuating the models’ reliance on large, labelled training datasets and improving their generalization performance. We evaluated the proposed approaches on multiple datasets using the precision, recall, F1 score, and Intersection over Union (IoU) metrics. The results indicate that both boundary-enhanced approaches lead to consistent, albeit marginal, benefits between 1 and 2 %. Notably, the proposed semi-supervised training framework for direct CD approximately matches the performance of the fully supervised approach while using only 20% of the available training labels. In domain adaptation, our post-classification approach significantly outperforms typical supervised methods, with the most notable gains in recall rate (>22%) and IoU (12.6%). These findings highlight the effectiveness of our techniques.
Open Access
Individual tree delineation and species identification in deciduous and mixed Canadian forests using high spatial resolution airborne LiDAR and image data
Li, Jili; Hu, Baoxin; Sohn, Gunho; Armenakis, Costas
Analysis of individual trees in forests is of great value for the monitoring and sustainable management of forests. For the past decade, remote sensing has been a useful tool for individual tree analysis. However, accuracies of individual tree analysis remain insufficient because of the inadequate spatial resolution of most remote sensing data and unsophisticated methods. The improvement of individual tree analysis becomes feasible because of recent advances in LiDAR (Light Detection And Ranging) and airborne image sensing technologies. However, it is challenging to fully exploit and utilize small-footprint LiDAR data and high spatial resolution imagery for detailed tree analysis. This dissertation presents a number of effective methods on individual tree crown delineation and species classification to improve individual tree analysis with advanced remote sensing data. The individual tree crown delineation is composed of a five-step framework, which is unique in its automated determination of dominant crown sizes in a given forest scene and its determination of the number of trees in a segment based on LiDAR profiles. This framework correctly delineated 74% and 72% of the tree crowns in two plots with mixed-wood and deciduous trees, respectively. The study on individual tree species classification is focused on developing novel LiDAR and image features to characterize tree structures. First of all, coniferous and deciduous trees are classified. Features are extracted from LiDAR data to characterize crown shapes and vertical profiles of individual trees, followed by the C4.5 decision tree classification algorithm. Furthermore, groups of new LiDAR features are developed to characterize the internal structures of a tree. Important features are selected via a genetic algorithm and utilized in the multi-species classification based on linear discriminant analysis. An overall accuracy of 77 .5% is obtained for an investigation on 1, 122 sample trees in natural forests. In addition, statistical features based on gray-level co-occurrence matrix (GLCM) and structural texture-features derived from the local binary pattern (LBP) method are proved to be useful to improve the species classification using high spatial resolution aerial image. The research demonstrates that LiDAR data and high spatial resolution images can be used to effectively characterize tree structures and improve the accuracy and efficiency of individual tree species identification.
Open Access
Map-Based Localization for Unmanned Aerial Vehicle Navigation
(2018-03-01) Li-Chee-Ming, Julien Francois; Armenakis, Costas
Unmanned Aerial Vehicles (UAVs) require precise pose estimation when navigating in indoor and GNSS-denied / GNSS-degraded outdoor environments. The possibility of crashing in these environments is high, as spaces are confined, with many moving obstacles. There are many solutions for localization in GNSS-denied environments, and many different technologies are used. Common solutions involve setting up or using existing infrastructure, such as beacons, Wi-Fi, or surveyed targets. These solutions were avoided because the cost should be proportional to the number of users, not the coverage area. Heavy and expensive sensors, for example a high-end IMU, were also avoided. Given these requirements, a camera-based localization solution was selected for the sensor pose estimation. Several camera-based localization approaches were investigated. Map-based localization methods were shown to be the most efficient because they close loops using a pre-existing map, thus the amount of data and the amount of time spent collecting data are reduced as there is no need to re-observe the same areas multiple times. This dissertation proposes a solution to address the task of fully localizing a monocular camera onboard a UAV with respect to a known environment (i.e., it is assumed that a 3D model of the environment is available) for the purpose of navigation for UAVs in structured environments. Incremental map-based localization involves tracking a map through an image sequence. When the map is a 3D model, this task is referred to as model-based tracking. A by-product of the tracker is the relative 3D pose (position and orientation) between the camera and the object being tracked. State-of-the-art solutions advocate that tracking geometry is more robust than tracking image texture because edges are more invariant to changes in object appearance and lighting. However, model-based trackers have been limited to tracking small simple objects in small environments. An assessment was performed in tracking larger, more complex building models, in larger environments. A state-of-the art model-based tracker called ViSP (Visual Servoing Platform) was applied in tracking outdoor and indoor buildings using a UAVs low-cost camera. The assessment revealed weaknesses at large scales. Specifically, ViSP failed when tracking was lost, and needed to be manually re-initialized. Failure occurred when there was a lack of model features in the cameras field of view, and because of rapid camera motion. Experiments revealed that ViSP achieved positional accuracies similar to single point positioning solutions obtained from single-frequency (L1) GPS observations standard deviations around 10 metres. These errors were considered to be large, considering the geometric accuracy of the 3D model used in the experiments was 10 to 40 cm. The first contribution of this dissertation proposes to increase the performance of the localization system by combining ViSP with map-building incremental localization, also referred to as simultaneous localization and mapping (SLAM). Experimental results in both indoor and outdoor environments show sub-metre positional accuracies were achieved, while reducing the number of tracking losses throughout the image sequence. It is shown that by integrating model-based tracking with SLAM, not only does SLAM improve model tracking performance, but the model-based tracker alleviates the computational expense of SLAMs loop closing procedure to improve runtime performance. Experiments also revealed that ViSP was unable to handle occlusions when a complete 3D building model was used, resulting in large errors in its pose estimates. The second contribution of this dissertation is a novel map-based incremental localization algorithm that improves tracking performance, and increases pose estimation accuracies from ViSP. The novelty of this algorithm is the implementation of an efficient matching process that identifies corresponding linear features from the UAVs RGB image data and a large, complex, and untextured 3D model. The proposed model-based tracker improved positional accuracies from 10 m (obtained with ViSP) to 46 cm in outdoor environments, and improved from an unattainable result using VISP to 2 cm positional accuracies in large indoor environments. The main disadvantage of any incremental algorithm is that it requires the camera pose of the first frame. Initialization is often a manual process. The third contribution of this dissertation is a map-based absolute localization algorithm that automatically estimates the camera pose when no prior pose information is available. The method benefits from vertical line matching to accomplish a registration procedure of the reference model views with a set of initial input images via geometric hashing. Results demonstrate that sub-metre positional accuracies were achieved and a proposed enhancement of conventional geometric hashing produced more correct matches - 75% of the correct matches were identified, compared to 11%. Further the number of incorrect matches was reduced by 80%.
Open Access
Mineral exploration modeling and singularity analysis for geological feature recognition and mineral potential mapping in southeastern Yunnan mineral district, China
Wang, Wenlei; Cheng, Qiuming; Jarvis, Gary; Armenakis, Costas
Nowadays, with the development in construction of geo-exploratory datasets and data processing techniques, mineral exploration modeling for recognition of mineralization associated geological features and mapping of mineral potentials become more dependent on GIS-based analysis and geo-information from multi-source datasets. Geological, geochemical and geophysical data as three main sources of geo-information in support of mineral exploration have long been employed in many researches. Spatial distributions of geological bodies or controlling factors associated with mineralization were frequently interpreted from these datasets. However, former characterizations of the controlling factors were simply focused on their location information; concerns on spatial variations of their geological signatures and controlling effects on mineralization were not sufficiently emphasized. Therefore, through a series of newly developed GIS-based manipulations, current study intends to demonstrate a comprehensive mineral exploration modeling process for more explicit recognition of controlling factors and their interactions on mineralization and delineation of hydrothermal mineral potentials in southeastern Yunnan mineral district, China. The hydrothermal mineralization as a nonlinear geo-process is accompanied with anomalous energy release and material accumulation in a narrow spatial-temporal interval. Simultaneously, it is a cascade process associated with various geological activities (e.g., magmatism, tectonism, etc.). Knowledge of these associated geo-activities is consequently beneficial to the exploration of hydrothermal mineralization. As the key point of this study, the singularity index mapping method in the context of fractal/multifractal efficient in separating geo-anomalies from both strong and weak background is applied to characterize variations of geological signatures of three controlling factors (i.e., granitic intrusions, faults and the Gejiu formation). With the guidance of multidisciplinary approaches, these geo-information derived from multi-source datasets is further integrated to produce the potential map. In comparison with traditionally used methods, the newly depicted predictor maps enhance weak geo-anomalies hidden within a strong variance of background. In addition, three geo-information integration methods including RGB composition, the principal component analysis and the weights of evidence method are implemented. By the weights of evidence method, the qualitatively and quantitatively interpretable result possessing advantages of the other two methods, simultaneously, is accepted as the final result of currently proposed mineral exploration modeling. Summarized experiences through this study will not only support future exploration in the study area, but also benefit the work in other areas.
Open Access
Modeling Vessel Behaviours By Clustering Ais Data Using Optimized DBSCAN
(2021-07-06) Han, Xuyang; Armenakis, Costas; Mardkheh, Amaneh Jadidi
Today, maritime transportation represents substantial international trade. Sustainable development of marine transportation requires systematic modeling and surveillance for maritime situational awareness. In this research thesis, we present an enhanced density-based spatial clustering (DBSCAN) method to model vessel behaviors. The proposed methodology enhances the DBSCAN clustering performance by integrating the Mahalanobis Distance metric that considers the correlations of the points representing the locations of the vessels. The clustering method is applied to historical Automatic Identification System (AIS) data and generates an action recommendation tool and a model for detecting vessel trajectory anomalies. Two case studies present outcomes from the openly available Gulf of Mexico AIS data, and Saint Lawrence Seaway and Great Lakes AIS licensed data acquired from ORBCOMM (a maritime AIS data provider). This research proposes a framework for modeling AIS data, an algorithm for generating a clustering model of the vessels' trajectories, and a model for detecting vessel trajectory anomalies such as unexpected stops, deviations from regulated routes, or inconsistent speed. This work's findings demonstrate the applicability and scalability of the proposed method for modeling more water regions, contributing to situational awareness, vessel collision prevention, safe navigation, route planning, and detection of vessel behavior anomalies for auto-vessels development.
Open Access
Vision-based Navigation and Mapping Using Non-central Catadioptric Omnidirectional Camera
(2017-07-27) Khurana, Manas; Armenakis, Costas
Omnidirectional catadioptric cameras find their use in navigation and mapping, owing to their wide field of view. Having a wider field of view, or rather a potential 360 degree field of view, allows the user to see and move more freely in the navigation space. A catadioptric camera system is a low cost system which consists of a mirror and a camera. A calibration method was developed in order to obtain the relative position and orientation between the two components so that they can be considered as one monolithic system. The position of the system was determined, for an environment using the conditions obtained from the reflective properties of the mirror. Object control points were set up and experiments were performed at different sites to test the mathematical models and the achieved location and mapping accuracy of the system. The obtained positions were then used to map the environment.

Browse

Browsing Earth & Space Science by Author "Armenakis, Costas"

Results Per Page

Sort Options