The growing popularity of vision transformers (ViTs) in remote sensing image classification is due to their ability to effectively capture long-range dependencies. However, their high computational cost and memory footprint limit their applicability, particularly for small-scale datasets and resource-constrained environments. To address these challenges, we propose the multiscale multihead compact convolutional transformer (MSHCCT), a lightweight yet powerful model that integrates convolutional tokenization with small-scale ViTs to enhance multiscale feature representation while maintaining computational efficiency. Despite a modest increase in parameters and training time, MSHCCT achieves superior classification accuracy and robustness on high-resolution aerial scenes. Importantly, our approach eliminates the need for model pretraining, additional datasets, or multisensor data fusion, ensuring a computationally efficient and practical solution for remote sensing applications. The code will be made publicly available athttps://github.com/aj1365/MSHCCT

Landslides pose a significant threat to both people and the environment worldwide. Several natural or human factors such as earthquakes, volcanic activities, rainfall, and land use trigger landslides. Understanding where and when landslides occur requires accurate information on the timing, pattern, and extent of events. Remote sensing and geospatial methods can provide critical information to detect and monitor landslide activities in time and space and useful insight to support disaster risk assessment and management. This research investigates frequency ratio technique for landslide susceptibility analysis using geospatial data. The method was implemented and tested over the Howard Gap Road area of Polk County, North Carolina, USA. The method successfully predicted the landslide occurred in the area in 2018.

As climate change intensifies the severity of extreme weather, harnessing the protective functions of wetlands becomes increasingly imperative. The southeastern United States, particularly North Carolina, is highly endowed with different wetland classes that act as natural buffers during natural disasters or storms such as Hurricane Matthew and Hurricane Florence in 2016 and 2018 respectively. This research addresses the delineation of the wetland boundaries after Hurricane Florence, emphasizing the pivotal role of wetlands in flood resilience. Building on the Wetland Intrinsic Potential (WIP) tool, the paper employs machine learning to map and delineate wetlands in Southern North Carolina, focusing on Bladen and Wilmington counties. The study integrates LiDAR data, Sentinel-2 imagery, and the National Wetlands Inventory, utilizing hydrographic, imagery, and topographic inputs for accurate wetland mapping. Results showcase high accuracy in predicting wetland and upland locations, contributing to sustainable flood management practices. The research provides valuable insights into the application of machine learning tools, such as WIP, for wetland mapping and flood mitigation in vulnerable regions.

This study investigates an approach to map 3D flood map (i.e., floodwater extent and depth) using UAV high resolution imagery and LiDAR for Hurricane Matthew. We utilize a deep learning approach to map flooded areas from post event UAV images, and then employ spatial statistics to estimate the water depth of the flooded areas leveraging on the DEM. Afterward, an auxiliary dataset is combined with the generated flood depth result to map and analyze flood impact risk in settlement areas within the study area. Our result showed that settlement areas in Grifton exhibit different risk levels from a 3D flood depth perspective. This information could significantly enhance near real-time emergency response strategies, as well as future mitigation initiatives.

Effective disaster management and response require accurate and timely mapping of floodwater extent. Optical images facilitate easier flood identification, but their limitation with cloud cover makes them suitable mainly for post-flood analysis. SAR data, with its ability to penetrate clouds, offers advantages in flood scenarios. This study introduces a unique strategy by merging SAR data and UAV optical images, bridging the spatial-temporal gap between spaceborne and ground-based observations. UAVs provide precise details crucial for calibrating and validating flood routing models. The paper also proposes a deep learning-based approach for flood mapping through an efficient fusion of SAR and optical RGB imagery, contributing to enhanced disaster monitoring and response capabilities.

This research presents a groundbreaking approach to enhance traffic guidance systems by integrating GPT-4 with the Physics-Informed Neural Network-based traffic state estimator (PINN-TSE). Unlike traditional systems, our proposed method minimizes reliance on live data, ensuring robust service delivery despite data gaps. The PINN-TSE model demonstrates high precision with a Mean Absolute Error of less than 4 vehicles/mile in traffic density estimation, even in data-scarce regions. Results highlight its reliability in providing accurate traffic information, especially in areas with sparse conventional sensors or potential data interruptions. Additionally, GPT-4 enhances user interactions, offering not only precise updates but also personalized experiences by understanding and responding to user inquiries. This AI-integrated traffic guidance system surpasses traditional methods in estimation, personalization, and reliability, paving the way for smarter traffic management.

Acquisition of geospatial data by UAV has been acknowledged as an effective method of attaining reliable and quick high-resolution remote sensing data for analysis and decision-making in different applications such as agriculture as it produces timely results, and it is affordable. UAVs coupled with other technological applications such as robotics, computing and deep learning facilitate the execution of precision agriculture. The execution of individual tree identification enables vivid description of the crop peak height and canopy for accurate estimation of issues related to crop growth such as biomass reduction, lodging, and stunted growth. A geospatial intelligence algorithm was developed to determine the lodging state of the crops. First, the optical imagery (RGB) was automatically annotated using maximum likelihood classification algorithm for faster acquisition of ground truth and to verify with the manually annotated image data for deep learning training. The acquired images were split for training of fine-tuned three deep learning methods including U-net, U-net++ and attention-residual Unet. The results show that each model has the tendency to learn patterns in data and predict lodged crops. Results from the analysis were validated by visual assessment of the time-series aerial images and validation data acquired from the ground truth data using k-fold cross validation approach.

High-resolution imagery is increasingly used to detect flooded areas during a crisis situation. The article presents a comparison of four image classification methods for flood extent mapping. The methods include Random Forest (RF), support vector machine (SVM), fully convolutional network (FCN), and normalized difference water index (NDWI). High-resolution UAV imagery collected during Hurricane Matthew (2016) flood events were used to evaluate the classification methods for generating an accurate flood extent map. In this study, a fully convolutional network fine-tuned to segment the inundation areas. RF, SVM, and NDWI are implemented using the same dataset used for mapping flood extents. The results show that the FCN achieved an overall accuracy of 97.72%, followed by NDWI with 96.0%, SVM with 88.9%, and 87.8 % of RF. The results imply that FCN is more efficient than RF, SVM, and NDWI on generating real-time flood extent maps.

Vision-based lane detection for intelligent vehicles is a well-researched problem in the past decades. However, there are still many road conditions in which the lane marking extraction is very challenging. In this paper, a new lane marking extraction algorithm that performs better than the traditional Canny edge detector and Hough transform based techniques is proposed. The proposed system uses bird's eye view image with a 2D Gabor filter for lane marker enhancement followed by a marking extraction approach and a Bezier curve fitting technique. Preliminary test results show that the proposed algorithm works very well on highways and urban roads despite various environmental challenges such as shadows due to trees or bridges, road texture variations, and lighting conditions. In addition, the algorithm can run in real time at a rate of 25 frames per second for images of size 1280×720 pixels. Testing computer has Intel Core i7 processor with 8GB RAM and 3.6GHz frequency.

Geographic information systems (GISs) are widely used for representation, management, and analysis of spatial data in many disciplines. In particular, geoscientists increasingly use these tools for data integration and management purposes in many environmental applications, ranging from water resources management to the study of global warming. Beyond these capabilities, geoscientists need to model and simulate three-dimensional (3D) dynamic fields and readily integrate those results with other relevant spatial information in order to have a better understanding of the environmental problems. However, GISs are very limited for the modeling and simulation of spatial fields, which are mostly 3D and dynamic. These limitations are mainly related to the existing GIS spatial data structures that are static and limited to 2D space. In order to overcome these limitations, we develop and implement a new kinetic 3D spatial data structure based on Delaunay tetrahedralization and a 3D Voronoi diagram to support a 3D dynamic field simulation within GISs. In this article, we describe in detail the different steps from discretization of a 3D continuous field to its numerical integration, based on an event-driven method. For validation of the proposed spatial data structure itself and its potential for the simulation of a dynamic field, two case studies are presented in the article. According to our observations, during the simulation process, the data structure is maintained and the 3D spatial information is managed adequately. Furthermore, the results obtained from both experiments are very satisfactory and are comparable with the results obtained from other existing methods for the simulation of the same dynamic field. To conclude, we discuss the current challenges related to the development of the 3D kinetic data structure itself and its adaptation to 3D dynamic field simulation and suggest some solutions for its improvement.