@inproceedings{ParyaviMachineVisionTools2021,
 abstract = {An incipient population of Coconut Rhinoceros Beetle (CRB; Oryctes rhinoceros) was discovered on the island of Oahu in late 2013, posing a significant threat to the iconic palm trees and on the verge of migrating onto the other Hawaiian Islands and US Mainland. CRB surveillance and capture with traps deployed in remote and undeveloped areas are critical to contain and eradicate these invasive insects, but manual monitoring is laborious and resource-intensive. To improve the program's scalability, we are developing machine vision tools for trap-based surveillance systems to enable automatic CRB detection and UAV-based aerial imagery systems to allow automatic detection and characterization of CRB damage to palm trees. Both cloud-based and edge-based architectures are considered. The evaluations show that communication of trap images to a domain and inferencing on the cloud using effective ConvNets like MobileNet with Transfer learning for beetle detection is highly accurate (99%) but bandwidth-intensive. Alternatively, an inference made locally on limited microcontroller memory using TensorFlow Lite, with inference results communicated through low data rate radio transmission, demonstrates reduced but acceptable detection accuracy (80 to 90%) in the current prototypes. We plan to implement the latter approach on distributed wireless networks to automate CRB trap surveillance and analysis in remote areas.},
 author = {Paryavi, Mohsen and Jenkins, M. Daniel and Melzer, MIchael and Ghorbani, Reza and Chun, IL Yong and Zheng, Yao},
 booktitle = {2021 Hawai'i Conservation Conference},
 copyright = {All rights reserved},
 month = {July},
 title = {Machine Vision Tools for Delimiting Distribution of Coconut Rhinoceros Beetle on the Island of Oahu},
 year = {2021}
}