New Detection and Classification Algorithms for Mapping Woody Weeds from UAV Data

Summary

Woody weed infestations over large farmland regions pose a difficult and costly land management problem. The extent of an infestation or the presence of individual invasive trees can be difficult to detect from conventional remote-sensing data sources such as satellite images (due to their limited spatial resolution), making it difficult to rely on them for planning intelligent survey, control and eradication strategies. 
This project is part of a larger body of research to demonstrate a low flying autonomous Unmanned Aerial Vehicle (UAV) as a low cost solution for remotely sensing weed infestations (and potentially other spatially distributed natural phenomenon). The proposed system consists of a robotic aircraft and ground station to acquire data, and a framework of computer analysis algorithms to process the data and provide detections, maps and statistics. The preceding project (B.NBP.0474) was concerned with the development and testing of the robotic UAV system over a three year period from 2008-2010. The project concluded in field trials in 2009 and 2010, demonstrating small, low-cost UAVs operating effectively in a rugged farmland environment. 
The successful flight trials provided a number of large-scale aerial image datasets (consisting of high resolution red-green-blue image tiles with accompanying navigational information), and initial concepts were presented for classifying and mapping based on this data. In this ensuing project, development has been focused on the analysis algorithms with the goal of extracting high level scientific information from the UAV sensor data. Three key research areas have been pursued.