Automated Animal Population Counting System Using Deep Learning for Sustainable Aquaculture

Current manual sampling methods for monitoring shrimp populations in industrial farms are prone to inaccuracies, can stress or damage the shrimps, and are labor-intensive. This hampers the ability to accurately estimate biomass, growth rates, and health status, affecting quality control and operational efficiency.

A mid-sized aquaculture farm or seafood cultivation company aiming to improve monitoring and management of farmed aquatic species through automation.

• Develop an automated system to estimate and monitor the number of shrimps in farm settings using computer vision and deep learning techniques.
• Achieve a counting accuracy with a relative miscount error below 6% across diverse conditions and densities.
• Enhance biomass estimation accuracy to improve operational decision-making and sales forecasting.
• Create a scalable solution adaptable to different farm environments and aquatic species.
• Reduce manual sampling efforts and minimize shrimp stress and damage during monitoring.

• Automated image capture interface for real-time monitoring.
• Deep neural network models for object detection (e.g., YOLOv5, Faster R-CNN) to identify individual shrimps.
• Density map-based models for population estimation in high-density images.
• Robust data labeling and model training pipelines to improve detection accuracy under varying conditions.
• Model validation and performance assessment, including out-of-distribution sample testing.
• Visualization dashboard for live inference and historical analytics.
• Model augmentation techniques to handle overlapping objects and high-density scenarios.

• Farm surveillance camera systems for live image acquisition.
• Data storage solutions for labeled datasets and model outputs.
• Operational management platforms to incorporate biomass and count data for decision making.

• Model accuracy with less than 6% relative counting error.
• System scalability to handle large datasets and multiple farm locations.
• High system availability to support continuous monitoring.
• Data security and privacy compliance for farm data.

The implementation of this automated shrimp counting system is expected to significantly improve monitoring accuracy, achieving count errors below 6%, thereby enabling more precise biomass estimation and operational planning. It will reduce manual labor, lower shrimp stress, and facilitate scalable deployment across multiple farms, ultimately leading to enhanced product quality, increased efficiency, and better resource management in aquaculture operations.