Dr. Zhiyong Hu

Sea levels are rising globally. To better assess the impacts of sea level rise and coastal inundation on Texas gulf coast, we need to generate coastal marsh substrate as accurate as possible for the coupled hydrodynamic-marsh model (Hydro-MEM) to the region. After marsh is identified and removed from lidar point clouds, we need generate marsh substrate from neighboring points. The experiment is to find robust and fast algorithms for interpolating underlying surface of coastal marsh. First, we evaluated four commonly used geospatial interpolation methods for lidar points. Then we put forward a multiple-scale raster approach which adopts the principle of Voronoi diagram and inverse distance weights interpolation, the experiments show that it can run much fast than point methods with satisfactory accuracy.

Fully Convolutional Network (FCN), which can adopt various Convolutional Neural Networks (CNN), are now increasingly being used in remote sensing communities. CNN are improved constantly either in accuracy or by reducing parameters for a given equivalent accuracy. This paper investigates five widely used CNNs (AlexNet, VGG16, ResNet, SqueezeNet, and a pruned VGG16) in the context of FCN for coastal beach classification of imagery acquired by Unmanned Aerial Vehicles (UAV). Our experiments show that (1) not every CNN is suitable to FCN for semantic segmentations of images though each CNN approximately achieved an equivalent accuracy for image labeling; (2) band reduced pruning of existing CNN has the least impact on implementation and accuracy. To examine the capability of convolutional layers capturing semantic features, this paper also carries out beach classification experiments using hypercolumn methods with VGG16.

As the centers of knowledge, discovery, and intellectual exploration, US universities provide appealing cybersecurity targets. Cyberattack origin patterns and relationships are not evident until data is visualized in maps and tested with statistical models. The current cybersecurity threat detection software utilized by University of North Florida's IT department records large amounts of attacks and attempted intrusions by the minute. This paper presents GIS mapping and spatial analysis of cybersecurity attacks on UNF. First, locations of cyberattack origins were detected by geographic Internet Protocol (GEO-IP) software. Second, GIS was used to map the cyberattack origin locations. Third, we used advanced spatial statistical analysis functions (exploratory spatial data analysis and spatial point pattern analysis) and R software to explore cyberattack patterns. The spatial perspective we promote is novel because there are few studies employing location analytics and spatial statistics in cyberattack detection and prevention research.

This study applied some decent spatial statistical methods and GIS technology to the investigation of the spatial relationship between the locations of DUI Arrests and alcohol retailers in Pensacola, Florida. After exploratory data analysis and testing spatial association based on bivariate J Function, DUI arrests point pattern shows a strong clustering trend around alcohol retainers. A conditional intensity function based on inhomogeneous Poisson process was given to model the spatial association between two of them, and the formula of model was fitted in statistics software called R. According to apply the fitted model, the probability in each position within this area where DUI Arrest might happen was predicted and plotted, which might be helpful to guide government agencies to layout locations of the alcohol retailers.

This paper presents methodology using dasymetric mapping from remotely sensed imagery, geographic information system (GIS), spatial analysis and spatial statistics to explore relationship between asthma and air pollution in the Pensacola metropolitan region of Florida. Health outcome indicators thought to be sensitive to increased exposure of airborne environmental hazards are mortality and morbidity rates for total population asthma patients. Environmental data for the time around the year 1999 include point source pollution sites and emissions, traffic count with emission estimates, and a Landsat ETM+ image. Standardized mortality/morbility ratios (SMRs) were used as dependent variables for the analysis. A centroid map was created from the zip code map with each centroid assigned the corresponding SMR values. Then spatial interpolation using the Kriging method was used to generate continuous SMR surfaces. An emission or point count based kernel density raster map was created from each of the air pollution maps. A raster layer ‘greenness’ was extracted using tasseled cap transformation from the Landsat ETM+ image. The dasymetric mapping technique was employed to limit the analysis and modeling to the area where human activities occur. The ETM+ image was classified into a thematic land use/cover map and the developed area extracted. A road network was combined with the developed area to generate a buffer (buffer distance = 1.5 km). A random sample with enough number of points was generated across the study area and 505 points were found within the developed area and the buffer. Data values at these sample points were extracted and used for statistical modeling. Two spatial autoregressive models (spatial error and spatial lag) were fitted. Both models show relationship between the asthmas outcome indicators and air pollution (positive) and ‘greenness’ (negative).