How AI Is Revolutionizing Fish Health in Aquaculture

Maintaining a healthy environment for aquatic life is a complex challenge. At the heart of modern, sustainable aquaculture is the ability to understand what is happening beneath the water’s surface - long before problems become visible. The Fish Health Monitoring System developed by RiniSoft within the framework of the NESTLER project, addresses this challenge by combining artificial intelligence and computer vision to provide continuous, 24/7 monitoring of fish populations.

Traditional fish monitoring relies heavily on manual inspections. These checks are not only labor-intensive but can also stress the fish. Our system transforms this process through a multi-layered, fully automated approach:

What Is the AI Actually Watching?

The system does not simply “see fish.” It tracks specific, measurable parameters that function as vital signs for the population:

Individual and Group Tracking

Using the Deep SORT algorithm, the AI follows individual fish across video frames, analyzing their trajectories, spacing, and group dynamics.

Physical Growth

By measuring linear dimensions, the system monitors growth rates and developmental trends over time.

Environmental Quality

Water transparency is also assessed. Because clarity is essential for fish health, the AI evaluates image defocus levels to estimate water quality and immersion depth.

Early Warning Signals: Detecting Disease Through Behavior

Changes in behavior are often the earliest indicators of health problems. Our system focuses on two key behavioral metrics to identify risks at an early stage.

1. Reference Speed of Healthy Fish (RSHF)

Healthy fish move with consistent, purposeful speed. The system calculates an average swimming speed to establish a baseline for normal behavior.If this speed decreases by 20% or more, the system triggers an alert, signaling potential gill disease, infection, or stress.

2. Average Diving Depth of Fish (ADDF)

This parameter is especially effective for detecting gastrointestinal (GIT) issues.

The science: Fish with digestive problems may experience buoyancy issues caused by gas accumulation.

The detection: If an unusually high number of fish are detected swimming within 10 cm of the water surface, the system flags a potential nutritional or digestive concern.

Visualisation of classifier performance

Visualizing Activity with Heat Maps

To better understand how fish use their environment, the AI generates daily heat maps of tank activity. These maps divide the space into low-, medium-, and high-density zones.By analyzing thousands of video fragments, farmers can see where fish prefer to feed, rest, or congregate - valuable information for optimizing tank design and feeding strategies.

From Sturgeon to Tilapia: A Flexible Solution

Our initial research focused on sturgeon, whose bottom-dwelling behavior makes them ideal for behavioral analysis. However, the system is highly adaptable.We have successfully transferred these neural networks to Nile tilapia, recalibrating the algorithms to accommodate their smaller size, faster movement, and surface-oriented behavior.

Hardware Designed for Real Farms

Advanced software requires reliable hardware. Our system uses durable, high-clarity Hikvision IP bullet cameras (8 MP), paired with high-performance RiniSoft servers.Processing occurs at a refresh rate of 25 Hz, ensuring that even subtle changes in movement or behavior are captured in real time.

The Future of Sustainable Aquaculture

By shifting from reactive treatment to proactive, AI-driven monitoring, fish farms can reduce mortality, improve growth rates, and maintain healthier stocks.This technology supports a more efficient, ethical, and sustainable future for aquaculture - one where problems are identified early, and fish welfare is continuously protected.