Real-time Fish Detection and Counting with YOLOv11

Abstract

As the population grows, so does the demand for protein and fish resources. The fishing industry faces the challenge of striking a balance between environmental protection and meeting the demand for food through fish products. Industrial aquaculture technologies, leveraging artificial intelligence (AI) and robotics, address this issue. While traditional methods rely primarily on manual labor, the integration of real-time fish detection and counting technologies demonstrates promising practical results. This article analyzes a fish detection and counting system based on the YOLOv11 model, distinguished by its ability to recognize and localize fish in complex aquatic environments. On the Deepfish dataset with a single fish class, the model achieved 96.4% mAP50, 93.2% precision, and 90.3% recall. The system not only delivers high accuracy in real-world conditions but also processes data in real time, effectively supporting the monitoring and management of fish resources. These findings highlight the potential of applying AI technologies in the fishing industry, laying the groundwork for future intelligent solutions. Due to its scalability, the YOLOv11 model can be further optimized to meet the growing needs of the industry and scientific research.