Session: 01-06-01: New Advances in Acoustics and Vibration: AI and Machine Learning, New Methods and Materials

Paper Number: 144422

144422 - Monitoring Invasive Insect Species Using Artificial Intelligence 

The spread of non-native insects can have a significant impact on the local ecosystem as they do not have natural predators. These invasive species can potentially damage tree fruit crops, vegetables and arable crops, leading to great economic losses. The Asian citrus psyllid (Diaphorina citri) is a sap-sucking bug that originated in Asia and has spread to other citrus growing regions in the world. It vectors a species of bacteria that causes citrus greening disease and threatens citrus production in the USA. Psyllids locate mates via vibrations conducted back and forth through the branches of citrus trees. A male psyllid begins to locate a nearby female by beating his wings, sending vibrations through the twig or branch on which he sits. The female replies with her own vibrational call. Then the male crawls toward the female. The long-term of this research project is to control or eradicate the psyllids by sending artificial responses to both males and females, to disrupt the localization and lure the males away from the females. Monitoring is an important part of the strategy and plays a crucial role in the mitigation of citrus damage. Traditionally, the monitoring process was managed manually. Sticky traps and lures were used to get information on insect activity, with these traps checked periodically to estimate the type and quantity of the insect. This is a time-consuming and labor-intensive task, which often requires a high level of entomological expertise. The focus of the project is to overcome the challenge and automate the monitoring process, which includes two major parts: 1) detection of the psyllids; and 2) gender classification. The automated detection of these mating-call vibrations through microphones indicates the presence of psyllids. The acquired acoustic signals are transformed into time-frequency (TF) images through time-frequency analysis. This study utilizes the continuous wavelet transform (CWT) for the TF analysis because of its multi-resolution capability that allows for rendering accurate and rich details of the signals. Then, deep learning algorithms are used on the images for classification, the accuracy of which is critical to the success of the control strategy. Being able to differentiate between call genders allows for selective responses. This can be challenging through conventional means because there are not established theories on the mating calls along with their signal characteristics. As a potential solution, deep neural networks are capable of learning and extracting meaningful features in an automatous fashion. A transfer learning approach is adopted in this study by using GoogLeNet, a high-performance, representative deep convolutional neural network (CNN) previously developed for image classification and recognition. Transfer learning applies the knowledge learned while solving one task to a different but related task, i.e., transfers knowledge from one domain to another domain. In this particular case, knowledge learned in image classification is used to differentiate the TF images representing mating calls from males and females. The existing have been carefully and efficiently created and designed by experts, and they have high performance on similar tasks. Generally, deep neural networks require a large amount of data to learn a mapping from an input a desired output. Transfer learning is a quick and effective way to deal with the lack of data. The pre-initiated network still needs to be retrained or fine-tuned by the data of the new task. However, a smaller duration of training is required for the new, but related, problem. There were three classification classes in the preliminary study: male calls, female calls, and no calls (background noise). The retrained GoogLeNet was able to yield a classification accuracy close to 90%.

Presenting Author: Yabin Liao Embry–Riddle Aeronautical University, Prescott

Presenting Author Biography: Dr. Yabin Liao received his B.E. degree in automotive engineering from Tsinghua University, Beijing, China, in 1999. Afterward, he received the M.S.E. degree in electrical engineering in 2004, and Ph.D. degree in mechanical engineering in 2005, both from Arizona State University in Tempe, Arizona, USA. Dr. Liao taught at Arizona State University between 2006 and 2017 before moving to Penn State Erie. After that, he joined Embry-Riddle Aeronautical University in 2021, where he is currently an associate professor of aerospace engineering and teaching courses in space systems engineering, and measurements and instrumentation. His research interests include smart materials, energy harvesting, structural dynamics, signal processing, and machine learning.

Authors:

Aviad Golan Embry-Riddle Aeronautical University
Yabin Liao Embry–Riddle Aeronautical University, Prescott
Seth Mcneill Embry-Riddle Aeronautical University