Tools for ecosystem monitoring based on fish detection and classification using deep neural networks

This study explores the transformative impact of artificial intelligence (AI) in ecosystem monitoring, specifically object detection with YOLO (You Only Look Once), emphasising the search for optimal tools and model efficiency. The shift from manual counting to AI-based detection significantly reduc...

Descripción completa

Detalles Bibliográficos
Autores: Prat Bayarri, Oriol, Baños Castelló, Pol|||0000-0003-0780-3255, Martínez Padró, Enoc|||0000-0003-1233-7105, Río Fernández, Joaquín del|||0000-0002-6191-2201
Tipo de recurso: artículo
Fecha de publicación:2024
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:2117/411165
Acceso en línea:https://hdl.handle.net/2117/411165
https://dx.doi.org/10.5821/iwp.2024.23.14158
Access Level:acceso abierto
Palabra clave:Ocean bottom -- Research
Artificial intelligence -- Engineering applications
Artificial intelligence
Object detection
Classification
YOLOv8
Ecosystem monitoring
Fons marins -- Investigació
Intel·ligència artificial -- Aplicacions a l'enginyeria
Intel·ligència artificial
Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial
Àrees temàtiques de la UPC::Enginyeria electrònica::Instrumentació i mesura
Descripción
Sumario:This study explores the transformative impact of artificial intelligence (AI) in ecosystem monitoring, specifically object detection with YOLO (You Only Look Once), emphasising the search for optimal tools and model efficiency. The shift from manual counting to AI-based detection significantly reduces time investment. Methodologically, the YOLO model is employed, and comprehensive training strategies are outlined. The threefold data division ensures unbiased evaluation, and diverse configurations are explored for optimal model performance. Key metrics, including IoU, Precision, Recall, and mAP, along with tools like confusion matrices, contribute to a thorough understanding of the model’s capabilities. Additionally, the model itself serves as a semi-automatic labelling tool.