Dr. John Lynham is an economist who has worked on interdisciplinary topics ranging from the political economy of marine reserves to the consequences of tsunamis. He is a UCSB alumnus and has an exciting interdisciplinary talk in store that should appeal to everyone at Bren. Come listen about how citizen science and machine learning can be weaved together to identify fish species and move toward low-cost fish stock assessments!
— Alberto Garcia, Bren School PhD Student
This talk will report on the early stages of a project and a series of research papers formed around the goal of using machine learning techniques to extract data from photographs taken by fishers themselves (as part of a large-scale community science project in Indonesia). Leveraging a unique training dataset (hand-labeled photographs of over two million fish, representing >150 different species), we used a segmentation model to cut out images of fish from background features and then used the Inception-ResNet V3 model to identify fish species. The segmentation model was also used to identify background features of uniform length, which were then used as inputs in a random forest model to predict fish length. At present, we are able to correctly segment 99% of fish from their photographs, accurately identify the species of 91% of fish caught, and measure fish length within 2.2 cm of actual length. Perhaps surprisingly, the distribution of length errors is relatively symmetrical for different sizes of fish. This suggests that machine learning methods could be used to perform accurate, rapid, and low-cost fish stock assessments on photographs taken by fishery participants.
John Lynham is a Professor of Economics at the University of Hawaiʻi, where he is also a UHERO Research Fellow, and the Undergraduate Chair for Economics. He has held visiting positions at Udayana University, IÉSEG School of Management, and Stanford University. He has been selected as a Pacific Century Fellow and as a recipient of the Board of Regents’ Medal for Excellence in Teaching. Visit his website here.