Sustainable fishing – Recognition of fish and fish species

The project was born from an Italian consortium of fishermen who have been working for years to fully respect the marine ecosystem. Their mission is clear: to practice selective fishing, ensuring the quality of their catch and preserving the balance of the waters in which they operate. However, manually selecting species is a slow process, prone to error, and not always compatible with today’s required sustainability standards.

The need: to recognize the species in real time

The consortium wanted an innovative system that could:

  • automatically identify fish species in the Mediterranean Sea;

  • distinguish in real time, from images of the catch, anchovies and sardines – two very similar “blue fish”;

  • minimize human intervention;

  • contribute to a more sustainable and controlled management of catches.

To achieve this, a computer vision system was needed that could not only analyze images but also accurately classify fish.

The project: from the camera to the cloud

The system was built using ADR-Flow, a modular platform designed to support supervised machine learning.

The project is composed of two modules:

Edge Module 

The edge module is a mobile station equipped with a “smart” camera, based on a Raspberry-type mini-computer.

This station can be configured in two operating modes:

  • a surface on which to arrange the fish;

  • a conveyor belt on which to slide them.

The camera analyzes images in real time, applying a specialized neural network for fish recognition.
Each species is highlighted with a specific color, a useful feature for both operators and automated processes.

The station also includes:

  • a user interface (graphical and sound);

  • image flow management;

  • Sending metadata to the cloud module.

Cloud Module

The cloud module consists of a virtual machine dedicated to:

  • receive and archive images from the sea;

  • enable technicians to manually search, review, and categorize photos using visual prompting techniques;

  • support data scientists in the continuous training and improvement of neural networks;

  • provide the system manager with a monitoring dashboard, with graphs and performance indicators.

Once trained, the neural network is fed back to the camera to progressively improve the quality of recognition in the field.

alici e sardine come selezionare le specie ittiche
pesca sostenibile - riconoscimento pesci

The key technologies used

To create a system capable of automatically recognizing fish species and supporting sustainable fishing, it was necessary to integrate several advanced computer vision and machine learning technologies. Below is a list of the main components underlying the developed solution:

  • Computer vision for real-time image processing.

  • Deep learning with specialized models for fish species classification.

  • Visual prompting for faster image annotation by technicians.

  • Edge processing to reduce latency and network dependency.

  • Cloud integration for advanced analytics, training, and monitoring.

Results: Selection leads to sustainable fishing

The experiment gave excellent results:

  • accurate recognition between anchovies and sardines, species that are very similar visually;

  • possibility of automating the selection of the catch;

  • improving the traceability and quality of the entire process;

  • concrete contribution to sustainable fishing, thanks to the reduction of waste and controlled selection.

The consortium now has a system that combines technological innovation and respect for the environment, bringing traditional fishing into the future.