Ryan Perroy1, Roberto Rodriguez2, David Benitez3, Timo Sullivan1 and Erica Ta1
1Spatial Data Analysis & Visualization Laboratory, University of Hawaiʻi at Hilo, Hilo, HI, USA
2APHIS-PPQ-CPHST Mission Laboratory, United States Department of Agriculture, Edinburg, TX, USA
3Division of Natural Resources Management, National Park Service, Hawaiʻi Volcanoes National Park, HI, USA
Despite decades of active management and widespread cooperative conservation efforts (Duffy & Martin 2019), native forests and ecosystems across Hawaiʻi continue to experience major and growing disruptions from alien plant invasions. These impacts are now compounded by widespread ʻŌhiʻa (Metrosideros polymorpha) mortality on Hawaiʻi and Kauaʻi islands due to introduced fungal pathogens associated with rapid ‘ōhi’a death (ROD; Keith et al. 2015). Regular monitoring to detect incipient invasions and other forest changes is a basic requirement for effectively managing these challenges (Müllerová et al. 2017). Although Hawaiʻi has a great need for this type of monitoring, there is a deficit of available repeat high-resolution (<10 cm) imagery over the islands, even with recent improvements in small unmanned aerial systems (sUAS), manned aviation, and satellite imaging systems. In areas where appropriate imagery datasets are available, bottlenecks associated with tedious manual assessment by trained analysts limit the utility of these data for the detection and identification of targeted plant species over large areas. Here we present work exploring (1) the benefits and challenges of adding consumer-grade (digital single-lens reflex (DSLR) and GoPro) camera systems to acquire useful imagery during conservation helicopter operations in Hawai‘i, and (2) results from the new automated computer vision convolutional neural network (CNN) classifiers for the detection of Miconia calvescens DC and other invasive species of interest (Figure 1). Our trained CNN algorithms were applied to directories of raw geotagged sUAS and helicopter imagery to identify targets of interest; real-world coordinates of the identified targets were then estimated from image EXIF and XMP metadata and trigonometric functions. While the deficit of high-resolution imagery in Hawai‘i remains an issue, we can now routinely map thousands of priority hectares across the state and our automated classifiers can analyze hundreds of images per hour, greatly reducing the processing bottleneck and freeing analysts to examine a much smaller number of curated images.
Duffy D. C. & Martin C. (2019) Cooperative natural resource and invasive species management in Hawai’i. In: Veitch C. R., Clout M. N., Martin A. R., Russell J. C. & West C. J. (eds) Island invasives: scaling up to meet the challenge, p. 497–502. Occasional Paper SSC no. 62. IUCN, Gland, Switzerland.
Keith L. M., Hughes R. F., Sugiyama L. S., Heller W. P., Bushe B. C. & Friday J. B. (2015) First report of Ceratocystis wilt on ˋŌhiˋa (Metrosideros polymorpha). Plant Dis. 99: 1276.
Müllerová J., Brůna J., Bartaloš T., Dvořák P., Vítková M. & Pyšek P. (2017) Timing is important: unmanned aircraft vs. satellite imagery in plant invasion monitoring. Front. Plant Sci. 8: 887.