College of Science & Engineering
These researchers are using MSI for a variety of projects:
- The first project is statistically significant spatial data mining. Spatial patterns (e.g., spatial hotspots, co-location patterns) are widely used in various domains such as public health, criminology, and biology. Because the spurious patterns detected in these domains may be of large societal and economic influence, statistical significance tests are used to avoid chance patterns. Monte Carlo methods, which are a class of computational methods that rely on repeated random sampling to obtain numerical results, are often used, since the underlying probability distribution is unknown. However, Monte Carlo methods can be computationally expensive. For example, spatial hotspots detection algorithms may be executed 100 or 1,000 times on complete spatial randomness datasets to evaluate the statistical significance of the detected spatial hotspots.
- The second project is to investigate deep learning frameworks for identifying geospatial objects (e.g., building footprints, tree canopies, urban gardens, etc.) in remote sensing datasets (e.g., satellite imagery, LiDAR point cloud). Given a remote sensing dataset from a set of spatial domains and ground truth data of object footprints from a subset of the spatial domains, geospatial object detection aims to find the footprints of the geospatial objects in the rest of the spatial domains (i.e., where ground truth data are not available).
- The group is continuing their study on urban tree detection. Inventories of individual trees provide meaningful information for urban planning, sustainable community planning, natural resource management, etc. In recent years, the invasive emerald ash borer has expanded to many countries (e.g., U.S., Canada, U.K.) and caused tree deaths in the millions. Since ash trees account for more than 20% of the urban forest in many U.S. cities, the cost of this ash borer disease has been estimated to be over 10 billion U.S. dollars. To respond to the threat, many state and city governments have begun to identify and treat (or remove) every individual ash tree in their management zones. In addition, increasingly many cities are eager to acquire and use fine-scale tree inventories (e.g., with locations, canopy sizes, heights) in their green infrastructure management, which is a critical part for sustainable community planning. These researchers will refine the initial approach they explored in 2019 to leverage deep learning to automatically detect individual urban trees using high-resolution remote sensing datasets.
- The group will also expand the work from 2019 for urban garden detection, which is an important component of urban agriculture, equity, and well-being (e.g., enhancing accessibility to fresh vegetables in many communities). In particular, they will analyze new aerial imagery datasets from Fulton county, Georgia, and Wayne County, Michigan. These datasets are large; for example, Wayne county imagery is around 50 GB and requires a large amount of main memory for processing. In addition, a large number of GPGPUs are needed to accelerate the process of data, as well as learning parameters of models such as deep neural networks.