|
Engaging Students of All Disciplines in Environmental Issues: Leveraging Research Grants to Break Down Academic Silos (Part 2)
Dr. Chris Halle
Program Developer, Center for Environmental Inquiry, Sonoma State University
Continued from article in TREE Press: November 2023
CEI partnered with SSU Professor Gurman Gill to apply the latest machine learning techniques (Google Tensor Flow) to separate the false alarms from the “valid” images. This research began as a guided undergraduate class project in computer science, with the class spending the first several weeks of the semester learning about classification algorithms. The computer science students also wrote some simple scripts to help biology student interns to step through the large camera dataset to and classify the wildlife images.
A team of students then spent the rest of the semester developing the computer model. The student team lead continued to develop the manuscript after the conclusion of the semester, leading to publication [Granados, Halle, and Gill, 2020].
One interesting result from the research is that Tensor Flow provides a “confidence limit”, or “threshold”, that can be varied to suit the requirements of a given camera network (Figure 3). Camera networks used for real-time public education and outreach should ideally only transmit pictures of animals to the public. Such networks should use a low threshold, so that any picture that might be a false alarm is excluded from transmission to the public. The tradeoff here is that some true capture events will be classified as false alarms. Camera networks that are used for research have different analysis requirements, and the threshold can be adjusted accordingly.
The real payoff of this study was that it fostered collaboration between biology student interns and the technical students. The technical students learned about the difficulties of field research, and the biology interns overcame their fear of working with technology. The biological interns have all either entered the field of vegetation management, or have joined other environmental research organizations since graduating. The computer science team lead (Granados) has entered graduate school, and is engaged in a similar field of study.
References:
Granados, J., C. Halle, and G. Gill, 2020. Classifying False Alarms in Camera
Trap Images using Convolutional Neural Networks, in Computational Science
and Computational Intelligence, (CSCI) 2020 International Conference, DOI:
10.1109/CSCI51800.2020.00270. https://ieeexplore.ieee.org/document/9458176.
| 
