Introduction
Feeding is the primary factor determining efficiency and cost in aquaculture and requires optimization to decrease feed loss and maintain fish health. Monitoring of individual and group swimming patterns in response to external factors such as feeding can provide useful information for improving production management. For this reason, intelligent feeding control based on the detection of behavioral changes and growth status has gained increasing attention and efforts are being made to define behavioral indicators to detect satiation levels and control feeding for different species. Here, an automated routine that enables individual fish tracking in sea cages as well as the detection of different swimming patterns is developed. Its potential applications on the detection of the swimming behaviour around feeding times are also presented.
Materials and methods
A group of E. seabass fish of 200 g body weight was reared in a circular polyester cage (40 m diameter, 9 m depth) located at the pilot scale netpen cage farm of HCMR at Souda bay, Crete (certified as an aquaculture facility from the national veterinary authority; code GR94FISH0001). A submerged network camera (Fyssalis V3.1) capturing at 10 fps was used for monitoring and video recording during daylight hours in April 2022. The camera was positioned at 4 m depth using a gyroscopic gimbal stabilizer to ensure it pointed upwards. Manual feeding was performed once daily, between 08:00 to 10:00. A machine learning model for tracking people (DeepSort) was trained and adapted to track fish individually (using OPENCV/Python) and extract their speed and direction. In addition, computer vision techniques that can detect feeding events and group polarized movement were incorporated into the model.
Results
The system is capable of distinguishing between three different movement patterns related to feeding (figure 1). The polarized motion seen before feeding is realized (figure 1 left), motion that resembles feeding behavior but lasts for a limited time period and could be an indicator of feeding or other warning situation (figure 1 middle) and the feeding event where the fish shoal swirls around the feed (figure 1 right).
The system is also capable to detect variations in the activity of the fish. Extraction of the group speed for two consecutive days showed significant variations of the fish speed during the day, suggesting that the fish start being more active during and after feeding and less active in the afternoon. More specifically, the group’s speed increased from 0.2 bd/s to 1.1 bd/s during the feeding window (i.e. from 08:00 to 10:00, figure 2), and it remained at maximum levels until 12:00, two hours after feeding. Then it started decreasing until it reached a minimum speed of 0.2 bd/s at midnight. This pattern is repeated also in the second day.
Conclusion
The presented system can successfully capture speed variations and different swimming patterns related to feeding or other external factors and can be used to provide useful information on the dynamics of the movements and possible critical values that indicate a transition from hungry to satiated states resulting in the better control of the feeding process. Changes in speed can successfully be captured, suggesting that fish start increasing their activity just before feeding something that would indicate anticipation for feeding and decrease their activity in the afternoon. Further studies are required to help us understand the contribution of other factors such as the human presence, or the internal circadian rhythm on the variation of the activity.
Acknowledgments
This work was partly supported by the European Union Horizon 2020 Project iFishIENCi (818036).
References
Antonucci, Francesca, and Corrado Costa. "Precision aquaculture: a short review on engineering innovations." Aquaculture International 28.1 (2020): 41-57.
Føre, Martin, et al. "Modelling growth performance and feeding behaviour of Atlantic salmon (Salmo salar L.) in commercial-size aquaculture net pens: Model details and validation through full-scale experiments." Aquaculture 464 (2016): 268-278.
Georgopoulou, Dimitra Georgios, et al. "Tracking and Analysis of the Movement Behavior of European Seabass (Dicentrarchus labrax) in Aquaculture Systems." Frontiers in Animal Science (2021): 79.
Wojke, Nicolai, Alex Bewley, and Dietrich Paulus. "Simple online and realtime tracking with a deep association metric." 2017 IEEE international conference on image processing (ICIP). IEEE, 2017.
Zhou, C., Xu, D., Lin, K., Sun, C. and Yang, X. 2018. Intelligent feeding control methods in aquaculture with an emphasis on fish: a review. Rev Aquacult, 10: 975-993. https://doi.org/10.1111/raq.12218