Finding Homogeneous Collections in Aerial Object Detection Datasets

Object detection in remote sensing imagery leads to a geo-tagged set of detected objects. Multiple images collected over adjacent region depict the collection of objects for large spatial areas. This work intends to find interesting collections of a single class, i.e., homogeneous, such as a vehicle lot, a terminal of storage tanks, a chain of railway cars, etc., across such geospatial imagery. Finding collections in remote sensing can be useful in urban planning, traffic control, disaster management, etc. This study defines the problem and focuses on exploring techniques for finding patterns of such collections, demonstrating its utility through available datasets such as xView and FAIR1M.