Zebrafish larvae locomotor activity detection using Convolutional Neural Network (CNN)

Monitoring and understanding fish behavior is crucial to achieve precision in practices. The assessment of fish behavior has always been difficult due to the difference between aquaculture and experimental conditions and much sampling time taken. New technologies have been explored for this reason to get better fish behavior observation. However, most of it are high costing but still have limitations in monitoring the fish behavior. According to the studies, the changes in fish behavior may reflect to the changes of water quality. So, fish behavior can be a great indicator of water quality in aquaculture field. These reactions are also very important in behavioral neuroscience, which study on the physiological, genetic, and developmental mechanisms of behavior in humans and other animals. The zebrafish larvae are used as a model organism to examine the effects of neurotoxin to human behavior. To overcome the limitations, this works aims to develop an algorithm to elucidate the zebrafish larvae locomotor activity using Convolutional Neural Network (CNN). The model used for this project is ssd_mobilenet_v2 fpnlite and the result proved that it could detect the activity of zebrafish larvae with 99.05% of accuracy percentage.