Traditional methods for monitoring aquaculture environments are often labor-intensive and inefficient. To address these challenges, researchers have turned to computer vision technologies, which offer real-time monitoring and management capabilities.
However, fish detection remains a critical yet challenging task due to factors such as variable lighting conditions, water turbidity, and pond floor dynamics.
Researchers from the Vellore Institute of Technology have developed a new model called AquaYOLO, a cutting-edge model designed to optimize fish detection in small-scale aquaculture environments.
The need for advanced fish detection
Fish farming, particularly in regions like southern India, plays a vital role in supporting local economies and ensuring food security. However, as aquaculture expands, traditional monitoring methods are becoming increasingly impractical.
Computer vision, powered by deep learning, offers a promising solution. Fish detection is a crucial step in this process but faces significant challenges, including fluctuating lighting conditions, water clarity, and the presence of aquatic vegetation.
Introducing AquaYOLO
Developed by researchers from the Vellore Institute of Technology, AquaYOLO is a state-of-the-art model specifically designed for aquaculture applications. Based on the YOLO (You Only Look Once) framework, AquaYOLO incorporates several advanced features to enhance fish detection accuracy and efficiency:
- CSP Layers and Enhanced Convolutional Operations: AquaYOLO’s backbone utilizes Cross-Stage Partial (CSP) layers and optimized convolutional operations to extract hierarchical features, ensuring robust feature representation.
- Multi-Scale Feature Fusion: The model employs oversampling, concatenation, and multi-scale fusion in its head to improve feature representation across different scales.
- Precise Localization: AquaYOLO uses a 40×40 scale for box regression and omits the final C2f layer to ensure accurate fish localization, even in complex environments.
Performance and validation
Researchers tested AquaYOLO using the DePondFi dataset, which contains over 50,000 bounding box annotations across 8,150 images collected from aquaculture ponds in southern India.
According to study results, the model achieved impressive performance metrics:
- Precision: 0.889
- Recall: 0.848
- Mean Average Precision (mAP@50): 0.909
These metrics demonstrate AquaYOLO’s ability to accurately detect fish in real time, even under challenging conditions.
The model was also evaluated on other benchmark datasets, including DeepFish and OzFish, to assess its generalization capabilities. AquaYOLO consistently outperformed existing models, proving its robustness and adaptability across different aquatic environments.
Key contributions of AquaYOLO
- New Application: AquaYOLO addresses a unique research problem by focusing on real-time fish detection in southern India’s pond environments, an area with limited prior research.
- Model Optimization: The model incorporates advanced architectural improvements, such as CSP layers and multi-scale feature fusion, to enhance detection accuracy and computational efficiency.
- Generalization: AquaYOLO has been validated on multiple datasets, demonstrating its ability to generalize across different aquatic environments and fish species.
Implications for aquaculture
AquaYOLO’s ability to provide accurate, real-time fish detection has significant implications for the aquaculture industry. By automating monitoring processes, the model can help fish farmers optimize feeding, reduce waste, and improve overall fish health. Additionally, AquaYOLO can be integrated with other technologies, such as water quality sensors, to create comprehensive smart aquaculture systems.
Future directions
While AquaYOLO has demonstrated remarkable performance, there are still areas for improvement. Detecting small or distant fish, particularly in highly turbid waters, remains a challenge.
Future research will focus on enhancing the model’s ability to handle occlusions and improve detection in low-visibility conditions. Additionally, integrating AquaYOLO with automated feeding systems and other IoT devices could further enhance its utility in aquaculture.
Conclusion
AquaYOLO represents a significant advancement in aquaculture monitoring, offering an efficient and accurate solution for real-time fish detection. Its ability to address the specific challenges of aquaculture pond environments makes it a valuable tool for fish farmers looking to optimize their operations.
By leveraging deep learning and computer vision, AquaYOLO provides a powerful tool for real-time fish detection, contributing to the sustainable management of aquatic resources. As global demand for seafood continues to rise, technologies like AquaYOLO will play a crucial role in ensuring the efficiency and sustainability of aquaculture practices.
Open access funding was provided by Vellore Institute of Technology.
Contact
A. Sasithradevi
Center for Advanced Data Science, Vellore Institute of Technology
Chennai, 600127, India
Email: sasithradevi.a@vit.ac.in
Reference (open access)
Vijayalakshmi, M., Sasithradevi, A. AquaYOLO: Advanced YOLO-based fish detection for optimized aquaculture pond monitoring. Sci Rep 15, 6151 (2025). https://doi.org/10.1038/s41598-025-89611-y
