Computer vision applied to fish farming for biometrics, biomass estimation and detection of vision-related pathologies.

• Grant number: 25/02355-8
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: September 01, 2025
• End date: May 31, 2026
• Field of knowledge: Agronomical Sciences - Fishery Resources and Fishery Engineering - Aquaculture
• Principal Investigator: Aristóteles Gomes de Melo Junior
• Grantee: Aristóteles Gomes de Melo Junior

Abstract

The project "Computer vision applied to aquaculture for biometry, biomass estimation, and disease detection" aims to develop an integrated hardware and software system tailored for aquaculture production environments, including excavated ponds and net tanks. The main objective is to provide an innovative solution that combines artificial intelligence and image processing to estimate fish weight, calculate total biomass, and identify early signs of diseases without stressing the animals.The current approach to fish biometry, based on manual sampling every 15 days, is imprecise, time-consuming, and negatively affects fish welfare. This project proposes replacing this method with an automated solution using cameras installed in the tanks to capture images of fish in motion. These images will be analyzed by advanced computer vision and machine learning algorithms, designed to operate under varying lighting conditions, water turbidity, and specific dynamics of each environment. Various types of cameras and configurations will be tested to ensure the quality and accuracy of the data collected.The hardware infrastructure will utilize edge computing, with local servers communicating via Wi-Fi with modules installed directly in the tanks. This architecture ensures fast and efficient processing, reducing reliance on remote servers and increasing system reliability. Initially, the system will be adapted for Tilapia, a species representing 52% of national production, but it will be designed for future expansion to other species through computational model updates.With an innovative approach and a multidisciplinary team, the project aims to achieve high levels of technological readiness during Phase 1, culminating in a functional proof of concept (TRL 6) that will demonstrate the technical and economic feasibility of the proposed solution. The greatest technical challenge will be collecting images in different production environments, ensuring they meet the required parameters for applying the developed models. The success of this stage will pave the way for Phase 2, enabling the technology's commercialization and introducing a new generation of intelligent systems for aquaculture.The implementation of this project will bring direct benefits to the aquaculture sector, enabling more efficient and sustainable management. Reducing resource waste, improving feeding planning, and continuously monitoring fish health will contribute to increased productivity and the sector's competitiveness in the global market. (AU)