Fish oocyte morphology detection using neural networks: a comparison of YOLO architectures

Fish oocyte morphology detection using neural networks: a comparison of YOLO architectures

ABSTRACT The recognition of oocytes, in their maturational stages, allow estimate the ovarian development and the type of spawning of a species. Although, distinguishing oocytes on histological images requires a visual and subjective interpretation by the specialist. With the development of deep lea...

Full description

Saved in:
Bibliographic Details
Main Authors: Yanna Leidy Ketley Fernandes Cruz, Ewaldo Eder Carvalho Santana, Isa Rosete Araujo Nascimento, Antonio Fhillipi Maciel Silva, Raimunda Nonata Fortes Carvalho Neta, José Ribamar de Souza Torres-Junior
Format: Article
Language:English
Published: Universidade Federal do Ceará 2025-01-01
Series:Revista Ciência Agronômica
Subjects:
Maturational phases
Histological images
Artificial intelligence
Convolutional Neural Networks
Centropomus undecimalis
Online Access:http://www.scielo.br/scielo.php?script=sci_arttext&pid=S1806-66902025000100638&lng=en&tlng=en
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:ABSTRACT The recognition of oocytes, in their maturational stages, allow estimate the ovarian development and the type of spawning of a species. Although, distinguishing oocytes on histological images requires a visual and subjective interpretation by the specialist. With the development of deep learning techniques, automatic object detection has become an important mechanism for this task. However, studies that use deep learning techniques have not been widely explored for the analysis of fish oocyte samples so far. In this paper, we propose the use of YOLO, a family of convolutional neural networks, for oocyte morphology detection of Centropomus undecimalis fish. The research uses an image database with 5,680 oocytes with different maturation stadiums (PV - pre-vitellogenesis, VI - early vitellogenesis and VF - late vitellogenesis), in histological images, divided into training, testing and validation, and detection performed by YOLOv3, YOLOv4, and YOLOv5 architectures. The results obtained were promising, highlighting that the YOLOv5l model, in the detection of oocytes of the VF class, reached the best values in the metrics precision, recall, mAP@.5 and mAP@.95 with 85.4%, 95.3%, 95.7%, and 75.9%, respectively. When considering all classes, YOLOv5l was the model that obtained the best results in the analyzed metrics.
ISSN:1806-6690

Similar Items