Conception and implementation of an AI-Powered computer vision : application prototype for detecting, counting, and classifying tilapia fish eggs

Abstract

The aquaculture sector is instrumental in global food security and has seen exponential growth, contributing to over half of the global fish production. In Algeria, this sector is growing, backed by government initiatives aimed at economic diversification and food security. One bottleneck in aquaculture efficiency is the manual counting and classification of fish eggs, specifically tilapia, which is labour-intensive and prone to human error. This study aims to develop a prototype system to alleviate this issue, focusing on tilapia egg classification and counting. The system comprises a mobile application integrated with a Mask R-CNN model for object detection and classification. We employ novel techniques such as dynamic dataset cleansing to rectify a skewed dataset and a multi-level transfer learning pipeline for model training. Our model achieved a peak mAP of 86.7%, with robust performance across different developmental stages of tilapia eggs. The results hold significant promise for enhancing the commercial production cycle in aquaculture by automating and optimizing egg counting and classification. This has broader implications for the aquaculture industry, both globally and in Algeria, offering a pathway to achieving the production goals set by the Ministry of Fisheries and Marine Resources.