A new report led by Dr angus Carnegie of NSW DPI has shown the potential of using new technologies in combination could save significant time and money for urban forest biosecurity surveillance. This study evaluated high-resolution airborne ArborCam imagery for tree species classification in a complex urban environment. An object instance segmentation of tree crowns was achieved using a deep learning CNN algorithm; one model for all trees; a second model running on Pinus and Platanus. These two dominant tree genera, Pinus and Platanus, were examined in a single local government area in Sydney, Australia and are hosts of high priority tree pests for Australia. Results were promising, and currently may satisfy requirements for the location of sentinel trees for early-detection surveillance. Further work is needed to expand the suite of tree species, and develop the technology for emergency response surveillance.
Find the full report on the project page.