Home // SIGNAL 2023, The Eighth International Conference on Advances in Signal, Image and Video Processing // View article

Supervised Spatial Divide-and-Conquer Applied to Fish Counting

Authors:
Gianna Arencibia-Castellanos
Alejandro González-Fernández
María Castillo-Moral
Rubén Fraile
Juana M. Gutiérrez-Arriola
Fernando Pescador

Keywords: Image processing, Object detection, SS-DCNet, biomass estimation

Abstract:
The estimation of fish biomass plays a crucial role in aquaculture. Performing this task automatically using machine learning algorithms has attracted the attention of the scientific community. This work describes the application of Supervised Spatial Divide-and-Conquer net to counting the number of larvae present in an image of an aquaculture tank. SS-DCNet is among the most robust object counters in the state of the art when applied to different datasets. It is trained with labeled images of turbots in breeding tanks, taking into account that the sizes can be variable and that they can be grouped and overlapped. Data augmentation is applied to obtain a greater number of training instances. The application of this model to counting turbots in images provides a mean relative error lower than 3.5$%$, which is an acceptable accuracy for this task. The main advantage of the model studied is its generalization ability, confirmed by its performance in counting objects in images where the density and the total number of objects are much higher than for the training images. Adapting the model for counting other types of fish, or turbot in other stages of growth, is straightforward since it is not necessary to build large training datasets.

Pages: 43 to 47

Copyright: Copyright (c) IARIA, 2023

Publication date: March 13, 2023

Published in: conference

ISSN: 2519-8432

ISBN: 978-1-68558-057-5

Location: Barcelona, Spain

Dates: from March 13, 2023 to March 17, 2023