The objective of this research is to investigate the capability and application of hyperspectral imaging technology to provide accurate and quick identification and quantification of building damage. Compared to optical and multispectral sensors, hyperspectral sensors can acquire simultaneous earth images in more than 200 narrow and often contiguous spectral bands through the ultraviolet, visible and infrared portions of the electromagnetic spectrum. To achieve this objective, the suitability of hyperspectral images to differentiate object classes and to assess the damage severity based on observed spectral changes will be determined. Furthermore, hyperspectral bands? sensing capability will be evaluated in the presence of various forms and degrees of cloud cover. The outcomes are expected to 1) expand the body of knowledge in using remotely sensed data for damage detection and quantification; 2) improve our understanding of the progression of wind hazards and their multi-faceted effects on built environment; 3) provide abundant high quality data much needed to calibrate and validate risk models for insurers and public agencies.
The broader impacts of this research are achieved by integrating research in curricular development, enhancement of research infrastructure, increased diversity, and broad dissemination of findings and discoveries. These activities collectively will provide valuable training opportunities for graduate and undergraduate students and will nurture the next generation of experts in wind engineering and remote sensing. They will also contribute to the formulation of efficient and effective strategies for windstorm response, recovery and mitigation that minimize human suffering and economic losses.
The major goal of this project was to investigate the capability and application of hyperspectral imaging technology to provide accurate and quick identification and quantification of building damages. It required high-quality and high-quantity datasets of building damages from multiple sources for modal development and testing and robust models and tools were developed for building extraction, damage identification and visualization. The research team firstly developed regression models to correlate remotely-sensed damages to ground observations based on a number of past tornadoes and hurricanes. Then they built a hyperspectral library of natural and man-made materials for object detection and developed joint spatial-spectral solution for urban structure detection using Hyperion datasets. They also developed algorithms for detecting tornado tracks and building roof damages using aerial imagery. In addition, the team collected and archived large sets of building damage data on 2011 tornado outbreaks in Tuscaloosa, AL and Joplin, MO in an online GIS database. It was found that the capability of satellite-based hyperspectral imagery in classifying land cover proved to be excellent thanks to its extremely high spectral resolution comparing to other sensors such as Landsat. However, its application in individual building detection and post-storm damage assessment was adversely affected by the low spatial solution of 30-meter. Meanwhile, tornado path could be confidentially identified either using change detection applied to before- and after-storm images or with color similarity measures on after-storm images alone. In addition, roof damage to residential buildings could be inferred from the presence of blue tarps readily detectable on RGB images. Intellectual merit: This study resulted in statistical models and computer algorithms for assessing windstorm damage using remote sensed imagery. The scale with which building damages were interpreted was tested and improved through the analysis of remote sensed and ground imagery taken after hurricanes and tornadoes. It served as a key building block for an automatic and integrated decision supporting system for hazard response, recovery and mitigation. Furthermore, it expanded the body of knowledge in using remotely sensed data for damage detection and quantification for various hazards. The examination of past events not only improved our understanding of the progression of wind hazards and their multi-faceted effects on built environment but also provided abundant high quality data much needed to calibrate and validate risk models for insurers and public agencies. The research made contribution to the field of remote sensing by developing a joint spatial-spectral solution for urban structure detection using hyperspectral imagery. Compared with the spatial or spectral methods, the segmentation by joint solution showed little misclassification of bare soil region and preserved most of the residential areas except near the edges of residential areas, where periodic patterns were not strong enough. Broader impacts: The project provided valuable training opportunities to three graduate students. Image processing software licenses were acquired, in junction with satellite and aerial images of multiple hurricanes and tornadoes, greatly enhanced the PI's laboratory capability in research and instruction. It facilitated collaboration of faculty across multiple departments at Texas Tech University (i.e. electrical and computer engineering, construction engineering, and wind engineering) and enabled those involved to work together on other interdisciplinary research. The software would be used to expand the spectral library beyond natural elements so that more signatures of man-made and special materials of interest such as pollutants could be digitalized and distributed on the internet. The results from this project could be used to assess the impact of future windstorms on built environment and support a full range of disaster response and recovery efforts.
