Livestock Fish Larvae Counting using DETR and YOLO based Deep Networks

TL;DR

This paper evaluates various neural network architectures, including transformers and CNNs, for counting fish larvae in aquaculture, introducing a new dataset and achieving high accuracy with real-time detection models.

Contribution

It presents a new annotated fish larvae dataset and compares multiple neural network architectures, including DETR and YOLO, for effective and real-time larvae counting.

Findings

Achieved a MAPE of 4.46% with DETR

Achieved a MAPE of 4.71% with YOLOv8

Introduced a new dataset with less data collection requirements

Abstract

Counting fish larvae is an important, yet demanding and time consuming, task in aquaculture. In order to address this problem, in this work, we evaluate four neural network architectures, including convolutional neural networks and transformers, in different sizes, in the task of fish larvae counting. For the evaluation, we present a new annotated image dataset with less data collection requirements than preceding works, with images of spotted sorubim and dourado larvae. By using image tiling techniques, we achieve a MAPE of 4.46% ($\pm 4.70$) with an extra large real time detection transformer, and 4.71% ($\pm 4.98$) with a medium-sized YOLOv8.