ARTIFICIAL INTELLIGENCE TO GAIN VALUABLE INSIGHT ON AQUATIC ORGANISMS

Samuel Couture Brochu*, Marnix Faes, Louis-David Coulombe and Julien Roy
 
Xpertsea Solutions Inc
100-1365 Ave Galilee
Quebec, QC, G1P4G4, Canada
samuel.c.brochu@xpertsea.com
 

Data quantity and reliability has been the main driver for yield optimization in agriculture and most livestock industries. In aquaculture however, accurate and reliable data is hard to obtain since counting and sizing small aquatic organisms mostly still relies on manual methods.   These manual methods are time consuming, inaccurate and non-repeatable.  Inconsistency in inventory assessments of aquatic organisms leads to mismanagement of feed and poor production performances for aquaculture producers.  

In recent years, technologies such as computer vision have been explored with moderate success to provide information about aquatic organisms.  However, recent development in artificial intelligence are proving to deliver viable options for efficient development of computer vision based solutions in aquaculture. In this project, an artificial intelligence approach using machine learning and computer vision was used to accurately predict the number of Giant tiger prawn (Penaeus monodon) post-larvae in a production setting. Data was gathered using an electronic device that image samples in optimal conditions. A training framework was then used to train and validate a classifying algorithm based on annotated data.

Once trained, the algorithm could count Giant tiger prawn (Penaeus monodon) post-larvae with more than 97.6% accuracy and 2.1% standard deviation. Other algorithms were also developed combining different technologies for different species and prediction of the size distribution was also implemented in a similar but more complex way.