Machine Learning Applications of Convolutional Neural Networks and Unet Architecture to Predict and Classify Demosponge Behavior

Biological data sets are increasingly becoming information-dense, making it effective to use a computer science-based analysis. We used convolution neural networks (CNN) and the specific CNN architecture Unet study sponge behavior over time. analyzed large time series of hourly high-resolution still images marine sponge, Suberites concinnus (Demospongiae, Suberitidae) captured between 2012 2015 using NEPTUNE seafloor cabled observatory, off west coast Vancouver Island, Canada. applied semantic segmentation with some modifications, including adapting parts be more applicable three-channel (RGB). Some alterations that made this model successful were dice-loss coefficient, Adam optimizer dropout function after each convolutional layer which provided losses, accuracies dice scores up 0.03, 0.98 0.97, respectively. The was tested five-fold cross-validation. This is first step towards analyzing trends in demosponge an environment experiences severe seasonal inter-annual changes climate. end objective correlate size (activity) seasons years environmental variables collected from same observatory platform. Our work provides roadmap for others who seek cross interdisciplinary boundaries biology science.