New advances in AI-based electronic monitoring (EM) technologies for automatic, real-time catch data collection: the iObserver 2.0
The implementation and fully compliance of the Common Fisheries Policy (CFP) of the EU depends largely on the ability to quantify total catches on board commercial fishing vessels. To this aim, the use of electronic devices is gaining relevance and vision-based electronic monitoring technologies hav...
| Autores: | , , , , , , , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2023 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/392007 |
| Acceso en línea: | https://hdl.handle.net/2117/392007 |
| Access Level: | acceso abierto |
| Palabra clave: | Fishery technology Electronic monitoring Deep learning Image recognition Fish species identification Fish length regression Line scan Tecnologia pesquera Àrees temàtiques de la UPC::Enginyeria agroalimentària::Pesca Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Aprenentatge automàtic |
| Sumario: | The implementation and fully compliance of the Common Fisheries Policy (CFP) of the EU depends largely on the ability to quantify total catches on board commercial fishing vessels. To this aim, the use of electronic devices is gaining relevance and vision-based electronic monitoring technologies have emerged as a more cost-effective and efficient way to monitor fishing activity. In this work, we present the iObserver 2.0, a device that uses Deep Learning image recognition to automatically identify and quantify in real time the entire catch on board fishing vessels. It builds upon two previous prototypes, improving image quality by using line scan technology. Two neural networks are used for fish species segmentation, identification, and length regression tasks. As main results of this disruptive technology, the iObserver 2.0 distinguishes more than twice the number of species than previous version, works with area scan and line scan camera images, and it is evaluated with a test set incorporating more complex images. An experimental fishing survey has been conducted to assess the system’s performance in real-life conditions, showing promising results in terms of total catch registration of target and discard fish species. |
|---|