Pierce presenting his research during his Honors Thesis Defense

Author: Pierce Helton | Major: Computer Science | Semester: Fall 2022

Since the Summer of 2021, I have worked in Dr. Khoa Luu’s Computer Vision and Image Understanding lab in the CSCE department at the University of Arkansas. Dr. Luu has served as my research mentor for several projects, including my honors thesis. The focus of my research grant dealt with insects and how to measure them in the environment. Dr. Luu and Dr. Ashley Dowling at the department of Entomology and Plant Pathology have teamed up to create an AI system that solves the problem of identifying and classifying insects in real time to aid in decision making with regard to insect management. For the second semester of my research grant, I continued my work on my previous research, focusing on how to accurately quantify these insect populations. Successful completion of my research and the lab project will have a profound impact on crop management and the creation of policy designed to address harmful and beneficial insect populations. This semester, I focused on refining the classification and detection models previously developed in our lab.

Work on this project began with data collection, something that I was tasked with when I first joined the lab. Generally, AI models require large amounts of data for training in order to perform specific tasks, so researchers need to spend time and resources to gather and modify this data. As the project continued, our lab investigated how we could apply AI in a way that would solve the problem of gathering meaningful entomological data for researchers. One of the challenges in working with AI is determining how to deploy systems that can run at a low cost for long periods of time while still providing quick and accurate results. We chose to solve this problem by creating a combined detection and classification framework that could capture images and identify insects in any given frame.

To solve this, our lab has developed a prototype system that can capture images in real time. My research has focused on the software rather than the hardware. Last semester specifically, I worked on developing a classification model that could predict an insect’s species upon receiving an input image. The framework functions by first receiving a raw image. This image is then passed through a detection model to identify insects in the image. Any insects identified are then passed to the classifier to predict species and keep track of the count of these insects. This semester, I needed to work with other students in the lab to figure out how to improve our system.

Our system had several problems that needed to be addressed, and Dr. Luu wanted to start with our detection model. When our system encountered overlapping or unique insects, our detection model could not accurately place bounding boxes over these bugs. Overlapping bounding boxes are suppressed due to the nature of the detection model, so we needed to figure out how to fix this issue by using different kinds of functions to improve these results. In addition, we wanted to train our model to recognize insects that we would likely encounter. Moths are one of the important insects that need to be quantified, so I included a dataset of moths in our training data. I also used the research grant funding to aid in my Honors Thesis.

This semester, I encountered problems with managing the data used in training and testing. Trying to convert different datasets to a common format can be challenging, especially when the data included in each dataset changes. Several times, I had to alter the functions I was using after encountering a case that I did not prepare for. Additionally, I ran out of time to finish the research and had to instead focus on my thesis and the defense presentation. However, this semester’s research and the thesis development taught me valuable lessons. I learned more about developing smart ways to modify data. Having a solid understanding of the input and desired output is important when modifying large amounts of data. I also received valuable feedback regarding my thesis and the presentation from graduate students and Dr. Luu. My time in the lab was time well spent; I enjoyed the work and learned much working under Dr. Luu. In the future, I hope other students can pick up where I left off and continue work on this interesting project.