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.

What problem does this paper attempt to address?

The paper primarily aims to address the issue of fry counting in aquaculture. Traditionally, the identification and counting of fry require manual work, which is not only time-consuming but also prone to errors. To improve efficiency and accuracy, the research team employed machine learning techniques to automatically identify and count fry. Specifically, they evaluated the performance of four neural network architectures (including convolutional neural network-based and transformer-based architectures) at different scales and proposed an image slicing technique to enhance detection accuracy. ### Main Contributions and Objectives: 1. **Creation of a New Dataset**: The researchers constructed a new annotated image dataset containing images of fry from Spotted Sorubim and Dourado. These images were captured using smartphones, which require lower image acquisition standards compared to previous works, making the dataset more applicable to real-world scenarios. 2. **Model Evaluation**: The paper evaluated four neural network architectures, including different versions of YOLOv8 (nano, small, medium, large, extra large), Real-Time Detection Transformer (RT-DETR), Detection Transformer based on ResNet-50 (DETR), and Deformable Detection Transformer (Deformable DETR). These models were assessed on the task of fry counting in high-resolution images. 3. **Application of Image Slicing Technique**: To address hardware limitations, the researchers also used an image slicing technique, which involves dividing the original image into multiple smaller segments for processing. This helps improve the model's performance under limited hardware resources. 4. **Performance Metrics**: The research results showed that using the extra-large Real-Time Detection Transformer (RT-DETR) achieved the smallest Mean Absolute Percentage Error (MAPE) of 4.46% (standard deviation ±4.70%), while the medium-sized YOLOv8 model had a MAPE of 4.71% (standard deviation ±4.98%). In summary, the goal of this study is to develop a method that can automatically and efficiently identify and count fry, reducing the need for manual operations and enhancing the efficiency and sustainability of the aquaculture industry. By using advanced machine learning techniques, particularly deep learning-based methods, the researchers aim to achieve this goal.

Livestock Fish Larvae Counting using DETR and YOLO based Deep Networks

Estimation of Injection Site in Turbot Based on Image Segmentation

Multi-class Fish Stock Statistics Technology Based on Object Classification and Tracking Algorithm

Semi-supervised learning-based live fish identification in aquaculture using modified deep convolutional generative adversarial networks

Deep learning with self-supervision and uncertainty regularization to count fish in underwater images

Shrimp Larvae Counting Based on Improved YOLOv5 Model with Regional Segmentation

Automatic Fish Population Counting by Machine Vision and a Hybrid Deep Neural Network Model

FishNet: Deep Neural Networks for Low-Cost Fish Stock Estimation

Real-time nondestructive fish behavior detecting in mixed polyculture system using deep-learning and low-cost devices

Counting Fish and Dolphins in Sonar Images Using Deep Learning

Fish Detection in Underwater Environments Using Deep Learning

Fish Trajectory Extraction Based on Object Detection

Multi-classification deep neural networks for identification of fish species using camera captured images

Deep but Lightweight Neural Networks for Fish Detection

Computer vision in aquaculture: a case study of juvenile fish counting

Research on Optimization Method of Multi-scale Fish Target Fast Detection Network

Detection Method of Fry Feeding Status Based on YOLO Lightweight Network by Shallow Underwater Images

Deep Learning-Based Fish Detection Using Above-Water Infrared Camera for Deep-Sea Aquaculture: A Comparison Study

A deep neural network for multi-species fish detection using multiple acoustic cameras

Identification and Counting of Pirapitinga Piaractus brachypomus Fingerlings Fish Using Machine Learning

Object counting from aerial remote sensing images: application to wildlife and marine mammals