Feature Detection in the Environmental Sciences.
详细信息
文摘
The rise of global average temperature,known as global warming,necessitates the analysis of large quantities of environmental science data. This dissertation concerns the extraction and visualization of features in these types of data sets,specifically forestry measurements and wind power simulations,and methods which may be used to represent them. A feature can be defined as a representation of an entity of interest that occurs in data,and which is related to phenomena in the real world. In forestry,data associated with trees is often captured using a remote sensing technique called airborne Light Detection And Ranging LiDAR),where large tracts of land are measured with a laser scanner and represented as a discrete point cloud. One may use a sophisticated computer vision framework to detect and delineate tree crowns as features,as managed forests are now seen as viable carbon sinks that can mitigate additional climate change. This dissertation discusses a shape-fitting approach that is applied to the crown detection problem without requiring an error-prone rasterization step,and contrasts it with other techniques. This work also applies feature detection to simulated wind farms,as wind energy is becoming an important source of electricity production. Unfortunately,the performance of wind farms is negatively affected by turbulence,usually in the form of rotational flow features known as vortices. This dissertation addresses how vortices are relevant to the performance of simulated wind turbines,how they may be portrayed in an illustrative manner,and how their topological properties may be depicted using glyphs. In summary,this thesis contributes features detection frameworks in LiDAR and wind farm simulation data. The former contribution finds tree crowns and calculates tree attributes such as height and radius. It compares favorably to trees manually segmented by domain scientists or trees measured in a related field survey. The latter provides an intuitive visualization system that provides insight to the causes of poor wind turbine performance,and is improved upon via non-photo realistic and topology-based rendering techniques.