A Novel Method for Automated Identification and Prediction of Invasive Species Growth Using Deep Learning

Overview

Alien invasive species (AIS) cause habitat destruction, lower crop productivity, climate change, and significant losses in global biodiversity. Global efforts to control rapid expansion of AIS have cost $1.2 trillion. With current AIS response/detection, 42% of threatened/endangered species continue to be at risk. This project aims to automate the detection and prediction of AIS growth using machine-learning-based classification and geospatial prediction models. 2-Dimensional Convolutional Neural Networks (CNNs) were developed, employing transfer learning architectures, Generative Adversarial Networks, and hyper-parameter tuning algorithms. The 2D-CNNs can identify 114 high-impact AIS and native species with 93.52% accuracy. Furthermore, using 152,657,384 3-Dimensional data points from AIS scans, 3D-CNNs were developed. Utilizing the Stanford PointNet segmentation architecture to detect invasive genera in aquatic, heat, and ambient conditions at various growth stages, the 3D-CNNs achieved a 97.78% validation accuracy, encompassing over 75 detectable AIS. Additionally, geospatial LSTMs were created, using climatic and AIS clusters/predatory data to accurately predict suitable locations of spread for AIS in the future. A total of 840 geospatial heat projection maps were created with a negligible training loss of 0.0143. The 2D-CNNs and LSTMs were also deployed to a mobile app. Several blinded field studies were conducted to validate the model results, with an average top-5 accuracy of 91%. The creation of state-of-the-art multidimensional classification/prediction models allows expansion into areas like dynamic LIDAR/aerial detection of AIS growth. Overall, this research offers an inexpensive, scalable, and previously unreported solution to the global AIS crisis.

Awards (4)

• Drexel University: Full tuition scholarship
• Association for the Advancement of Artificial Intelligence: Honorable Mention
• Association for the Advancement of Artificial Intelligence: AAAI Membership for the School Libraries of All 8 Winners (in-kind award / part of 1st-3rd prize and honorable mentions' prize)
• U.S. Agency for International Development: Second Award Agriculture and Food Security

Competition history

• ISEF 2022 Earth and Environmental Sciences · Entry EAEV022 · Atlanta, Georgia, United States

Resources

• ISEF abstract page ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.