TL;DR
This paper introduces a novel system for commercial aquaculture that uses directed water currents to guide fish, like hybrid striped bass, through a channel without physical handling. This approach leverages the fish's natural rheotactic behavior to enable automated sorting and health monitoring, aiming to improve fish welfare and reduce economic losses associated with stress and injury.
What Problem Does This Paper Address?
The current practices in commercial aquaculture, particularly for species like hybrid striped bass, often involve physically handling fish for routine tasks such as grading, inspection, and measurement. This manual handling is labor-intensive, stressful for the fish, and significantly increases the risk of injury, leading to a higher incidence of unmarketable fish and financial losses for producers. Furthermore, malformed fish, which may suffer prolonged starvation due to feeding difficulties, are often identified too late for humane intervention. Existing regulations for animal welfare in the U.S. largely exclude farmed fish, leaving their welfare largely unaddressed. This paper aims to bridge this gap by proposing a system that enhances fish welfare and economic efficiency through automation and a non-invasive approach.
Methods
The core of the proposed system relies on rheotaxis, the natural behavior of fish to orient and move upstream in response to water currents. The researchers developed a prototype system consisting of a tank divided into two regions by a channel. Directed water currents are generated by wave makers to encourage fish to move from a starting region to a destination region through this channel.
Key methodological aspects include:
- Prototype Design: A multi-stage prototype was developed, starting with a wider channel and a single wave maker, and evolving to a narrower channel with two wave makers to improve flow control and reduce turbulence in the destination region. This iterative design process was informed by initial observations of fish behavior, such as "flow refuging."
- Stimulus Control: The system uses directed water currents as the primary stimulus, avoiding aversive stimuli used in prior work. This approach aims to promote voluntary movement and allow fish to "consent" to participation.
- Computer Vision for Measurement: In parallel with the movement study, the paper also discusses a preliminary feasibility study on using computer vision to automatically measure fish length. Two methods were explored: a traditional "Pixel Counting" algorithm and a neural network-based approach.
- Experimental Design: The study involved controlled trials with juvenile striped bass. Fish were divided into groups, and their movement between regions was recorded during both control (wave makers off) and experimental (wave makers on) periods. The order of these periods was reversed between groups to account for potential novelty effects.
Datasets
The study utilized a dataset of 40 juvenile hybrid striped bass for the rheotaxis experiments. These fish were approximately 15 cm in length and were spawned in Spring 2024.
For the computer vision component, the paper mentions collecting "ground truth" length measurements for a population of "XX fish" (the exact number is not specified). This data was used to train and validate the computer vision models.
No new benchmark datasets were introduced; rather, the study generated data from its experimental setup for analysis.
Key Insights and Ideas
- Rheotaxis as a Viable Sorting Mechanism: The study demonstrates that directed water currents can effectively guide striped bass through a channel, suggesting rheotaxis is a powerful, non-invasive tool for automated fish movement and sorting.
- Voluntary Movement and Animal Agency: By leveraging innate behavior and avoiding aversive stimuli, the system promotes voluntary movement, aligning with principles of animal agency and consent in animal-computer interaction.
- Reduced Behavioral Variability: Fish exhibited less behavioral variability and spent more time in the destination region during experimental periods, indicating comfort and compliance with the directed current.
- Early Detection of Malformations: The system's ability to move fish into a controlled observation point opens avenues for early detection of malformed individuals, enabling timely and humane intervention.
- Synergy of Rheotaxis and Computer Vision: The combination of directed fish movement and computer vision for automated sizing and health monitoring presents a promising pathway towards fully automated, ethical aquaculture systems.
Key Results
- Movement Success: The directed water currents led to a significantly higher average number of fish in the destination region during experimental periods compared to control periods (t = -38.9529, p = 4.37 × 10⁻²⁷¹).
- Reduced Variance: The experimental periods showed lower variance in fish distribution within the tank compared to control periods, indicating that fish tended to move to the destination region and stay there rather than moving back and forth.
- Computer Vision Accuracy: The neural network model for estimating fish length achieved an average error of 5.17%, significantly outperforming the traditional pixel counting method (16.19% error).
- Acclimation Period Effectiveness: Fish generally showed no significant changes in appetite or behavior during the 4-day acclimation period, with only one unexplained death in a trial round.
Limitations and Future Work
Limitations:
- Sample Size and Scalability: The study was conducted with a small number of fish and in a prototype tank that is smaller than typical commercial housing. The scalability of the system to larger operations remains to be tested.
- Prototype Refinement: While the second-generation prototype showed improvement, further refinement of the "one-way" channel design is needed to fully prevent fish from regressing to the starting region.
- Limited Computer Vision Data: The computer vision study for length estimation used an unspecified number of fish ("XX fish"), and further work is needed to expand on these preliminary results.
Future Work:
- Closed-Loop Control and Scale: Developing fully autonomous sizing, grading, and sorting systems that operate at a larger scale.
- One-Way Channel Design: Incorporating principles from commercial fish traps to create more robust one-way channels.
- Precise Flow Analysis: Optimizing water current usage with more precise analysis to improve directional control.
- Advanced Computer Vision: Expanding computer vision capabilities to include real-time analysis of traits like coloration and physical abnormalities for more sophisticated sorting.
- Quantifying Welfare Improvements: Rigorously measuring indicators of acute and long-term stress to compare welfare outcomes with traditional methods.
- Cost-Benefit Analysis: Conducting a more thorough cost analysis to demonstrate the financial benefits for aquaculture farmers.
- Disease and Malformation Detection: Incorporating disease or malformation detection into the computer vision system.
Summary updated