Image-Based Monitoring in Aquaculture

Problem and Objectives

Aquaculture production systems require reliable information on stock size, growth, health status, and animal behavior. In practice, however, key management decisions are often based on manual sampling or indirect assumptions. These methods are time-consuming, cause stress to the animals, and provide only limited representative data for large or dense populations.

Additionally, underwater environments pose significant challenges: limited visibility, fluctuating lighting conditions, high stocking densities, and continuous movement. This creates a need for robust, non-invasive, and continuously deployable monitoring systems that provide valid and scalable decision-making data under real production conditions.

Against this backdrop, a camera-based, AI-supported monitoring system for aquatic organisms is being developed. It automatically captures stock, health, and behavioral data and translates them into management-relevant metrics.

Underwater Image Processing and System Infrastructure

Data acquisition is carried out using specially developed underwater camera systems that can operate autonomously or via cable. These systems are equipped with embedded computing units and support both edge and cloud-based analysis.

They can be deployed in various production environments, including recirculating aquaculture systems (RAS), flowing water structures such as fish passages, and net pens in marine aquaculture. The hardware and software architecture is modular and designed for long-term operation under harsh environmental conditions.

Automated Stock Monitoring

A key focus is on AI-based detection, classification, and measurement of aquatic organisms. Computer vision algorithms identify individuals within dense schools, distinguish (sub-)species, and capture length and volume parameters.

This enables continuous determination of stock size, length distributions, and biomass. The methods are designed to deliver robust results even under occlusion, turbidity, and varying lighting conditions.

Physical Health Indicators

In addition to quantitative stock assessment, visually detectable health parameters are integrated. These include growth and posture parameters, dimension-based measurements (height, width, length), and the detection of externally visible anomalies such as injuries in the gill cover or tail fin areas.

Continuous monitoring of these features enables early identification of deviations at both individual and group levels, providing an objective basis for animal welfare-oriented management decisions.

Behavioral Analysis and Movement Parameters

Movement and behavioral patterns are also analyzed automatically. Time series-based evaluations capture swimming speed, movement direction, vertical distribution within the field of view, and schooling dynamics.

Deviations from typical movement patterns may indicate stress, suboptimal environmental conditions, or emerging health issues. Behavioral analysis thus becomes an integral part of a holistic monitoring system.

Rearing Environment and Feeding Context

The monitoring system also incorporates parameters of the rearing environment, including feed input and feed utilization, and links them to growth and behavioral data.

By combining biological and management-related information, a data-driven basis is created for optimizing feeding strategies, stocking densities, and harvesting times.

Specific Applications: Shrimp Farming

The methodological components are transferable (e.g., detection and tracking) but require species-specific adaptation. In shrimp farming, AI-based approaches are used for automated counting, length distribution analysis, and condition assessment. Additionally, the fill level of the digestive tract can be visually analyzed as an indicator of feed intake and metabolic status.

This enables monitoring even in highly dynamic and small-scale production systems.

Funding: