The Fishvision Project
An innovative artificial intelligence system for monitoring and analyzing fish behavior, developed to support marine research and the conservation of aquatic ecosystems
Project Objectives
A cloud-based management tool using hybrid neural networks (CRNNs) to monitor fish health and welfare in offshore IMTA farms. Real-time data ensures dynamic adaptation to environmental and climate factors
Advanced models integrate physiological, behavioral, and molecular data to optimize feeding protocols, reduce waste, and improve fish growth and sustainability—aligned with the One Health approach
A digital fish farming platform combining underwater cameras and sensors to monitor fish behavior, environmental impact, and waste management—enabling evidence-based decisions and circular aquaculture
Methodology and Approach
Integrated Data Collection & Cloud Management
FishVision leverages underwater sensors, thermal cameras, and real-time monitoring tools within offshore IMTA cages to collect physiological, behavioral, environmental, and molecular data. All data are processed through a cloud platform for centralized and continuous management
CRNN-Powered Precision Feeding & Monitoring
A convolutional recurrent neural network (CRNN) models fish behavior and health, enabling adaptive feeding and farm management based on environmental changes, stressors, and historical datasets. This ensures welfare-driven decisions, reduces waste, and enhances traceability from hatch to harvest
From Welfare Index to Farm-to-Fork Traceability
FishVision implements a digital “Precision Livestock Farming” tool that monitors fish welfare and environmental interactions in real time. Every lot is tracked end-to-end according to UNI EN ISO 22005, including movements, feeding quality, and health events
Scalable, Sustainable, and Smart Aquaculture
Tested in real conditions at the Gargano Pesca Consortium, the system supports early disease detection, waste reduction, and environmental resilience. Designed for national and international scalability, FishVision combines AI innovation with measurable ecological and social benefits
Project Timeline
WP1 – Cage Installation & Sensor System Setup
Implementation of experimental cages with pre/post photographic documentation, including net replacement by technical staff. Two lots of 100,000 juvenile gilthead seabream (Sparus aurata) and European sea bass (Dicentrarchus labrax) are seeded and officially documented. Selection of monitoring parameters and preliminary lab analyses (chemical-physical and proteomic) are conducted. A multi-parameter monitoring system is installed—comprising underwater stereoscopic and thermal cameras. Two sampling sessions are performed for calibration, comparing lab data with manual instruments. The system is tested and technical staff are trained
WP2 – Data Cleaning & Real-World Dataset Setup
Cleaning of historical aquaculture data to ensure high-quality, reliable inputs for neural network training. The process includes removal of duplicates and anomalies, imputation of missing values, and configuration of AI parameters. Cleaned historical data are compared with real-time environmental data (April–May), CGP datasets, stakeholder input, and public records. QR factorization is used to validate the experimental setup in real farming conditions
WP3 – AI Quantum Livestock Pattern & Model Design
In collaboration with the University of Camerino, key parameters are identified and encoded to evaluate biotic and abiotic stress, fish responsiveness to feeding stimuli, behavioral patterns, and environmental thresholds—contributing to the development of the Welfare Score Index (WSI). The University of Molise leads the design of the AI model, defining objectives, required capabilities, and performance metrics for offshore smart farming, aligned with GR13, ESG, and LCA standards
WP4 – CRNN Design, Implementation & Training for FishVision
The University of Molise completes AI model development up to TRL 5, defining the RNN architecture, activation functions, and parameterization. Data from underwater sensors, stereoscopic and thermal cameras are collated, encoded, and normalized for training and testing. The neural network is then trained, validated, and tested to evaluate performance
WP5 – Cloud Platform Design & Implementation for FishVision
Development of the cloud-based management platform hosting the AI model. Activities include requirements gathering, prototyping, data architecture planning, database and network infrastructure setup, backend logic development, API implementation and deployment, and integration of the AI model. UI/UX design and backend testing complete the platform setup
WP6 – Dissemination, Communication & Training
All project partners engage in dissemination activities through scientific publications, conferences, public events, and an open data system for transparent access to results. FishVision promotes digital farming (smart feeding and precision livestock) as a climate-resilient and sustainable model for offshore aquaculture. The model, tested in Southern Italy, is designed for replicability across Europe. A certified training course for biologists and veterinarians is also included
WP7 – Project Management & Financial Reporting
Ongoing internal monitoring of administrative and financial compliance with the executive plan. Quarterly and final cost reporting is carried out in accordance with Article 10 of Rectoral Decree No. 513 dated 08/11/2023