Fish Tracking, Counting, and Behaviour Analysis in Digital Aquaculture: A Comprehensive Review (2406.17800v2)

Abstract: Digital aquaculture leverages advanced technologies and data-driven methods, providing substantial benefits over traditional aquaculture practices. This paper presents a comprehensive review of three interconnected digital aquaculture tasks, namely, fish tracking, counting, and behaviour analysis, using a novel and unified approach. Unlike previous reviews which focused on single modalities or individual tasks, we analyse vision-based (i.e. image- and video-based), acoustic-based, and biosensor-based methods across all three tasks. We examine their advantages, limitations, and applications, highlighting recent advancements and identifying critical cross-cutting research gaps. The review also includes emerging ideas such as applying multi-task learning and LLMs to address various aspects of fish monitoring, an approach not previously explored in aquaculture literature. We identify the major obstacles hindering research progress in this field, including the scarcity of comprehensive fish datasets and the lack of unified evaluation standards. To overcome the current limitations, we explore the potential of using emerging technologies such as multimodal data fusion and deep learning to improve the accuracy, robustness, and efficiency of integrated fish monitoring systems. In addition, we provide a summary of existing datasets available for fish tracking, counting, and behaviour analysis. This holistic perspective offers a roadmap for future research, emphasizing the need for comprehensive datasets and evaluation standards to facilitate meaningful comparisons between technologies and to promote their practical implementations in real-world settings.

• The paper presents a comprehensive evaluation of digital aquaculture technologies for tracking, counting, and monitoring fish behavior in various underwater conditions.
• It compares vision-based, acoustic, and biosensor methods, addressing challenges such as occlusion, environmental variability, and high implementation costs.
• The study recommends future research on multi-modal data fusion, standardized evaluation metrics, and real-time on-device analysis to enhance aquaculture management.

Overview of Fish Tracking, Counting, and Behaviour Analysis in Digital Aquaculture

The reviewed paper, titled "Fish Tracking, Counting, and Behaviour Analysis in Digital Aquaculture: A Comprehensive Review," presents an extensive analysis of digital technologies applied to aquaculture, focusing specifically on the tasks of fish tracking, counting, and behaviour analysis. This paper addresses the critical need for optimizing production efficiency, enhancing fish welfare, and improving resource management within the aquaculture industry via advanced technological solutions.

Digital Aquaculture Technologies

The paper categorizes current digital aquaculture technologies into three primary methods:

1. Vision-based Methods: These leverage computer vision algorithms and camera systems to monitor fish. Vision-based methods face challenges such as lighting conditions, water clarity, and background complexity, but advancements in deep learning and computer vision (e.g., convolutional neural networks) have improved their robustness and accuracy.
2. Acoustic-based Methods: Employing sonar and hydroacoustic technology, these methods are less affected by water turbidity and poor lighting conditions. Acoustic-based methods use devices such as DIDSON and ARIS to produce high-resolution underwater sonar images, facilitating the tracking and counting of fish. Their primary disadvantages include high costs and technical expertise requirements.
3. Biosensor-based Methods: Biosensors, including accelerometers and heart rate monitors, attached to individual fish enable monitoring of physiological and behavioural data. While they provide detailed individual-level insights, their invasive nature and the practical challenges of deploying them in large-scale operations limit their applicability.

Fish Tracking

2D and 3D Tracking

Fish tracking methods are broadly categorized into 2D and 3D approaches.

• 2D Tracking: Utilized for shallow water environments where fish movement is constrained to two-dimensional planes, these methods have evolved from classical algorithms and kernel correlation filters to deep learning-based tracking algorithms. Despite improvements, challenges such as occlusion and variation in individual fish appearance persist.
• 3D Tracking: Combining data from multiple cameras or using stereoscopic methods, 3D tracking allows comprehensive behavior analysis in natural environments. Complex setups and the need for high precision equipment remain barriers to deployment in practical settings.

Evaluation Metrics

The paper details several evaluation metrics for fish tracking, including Multiple Object Tracking Accuracy (MOTA), Multiple Object Tracking Precision (MOTP), and Identification-Score (IDF1). These metrics, combined with tracking speed measures like frames per second (FPS), enable a standardized assessment of tracking performance.

Fish Counting

Sensor-based and Computer Vision-based Methods

The paper outlines the prominent methods for fish counting in aquaculture:

• Sensor-based Methods: Infrared and resistivity counters are prevalent, though their efficacy is impeded by environmental factors like water turbidity and depth, often underestimating fish counts.
• Vision-based Methods: Divided into detecting-based and density-based approaches, computer vision methods benefit from advances in deep learning which enhance accuracy and robustness. Nonetheless, issues related to varying environmental conditions and fish densities require further research.

Fish Behaviour Analysis

Understanding fish behaviour is essential for effective aquaculture management. The paper categorizes behavior analysis into three primary areas:

1. Feeding Behaviour: Accurate detection of feeding intensity helps optimize feeding strategies and reduce waste. The application of deep learning models to video data has shown promising results but requires further adaptation to real-world aquaculture conditions.
2. Hypoxia Behaviour: Detecting signs of hypoxia through behavioural analysis provides early warnings, though current methods predominantly developed in laboratory settings need validation in practical environments.
3. Abnormal Behaviour: Identifying signs of stress, disease, or aggression through computer vision and acoustic monitoring systems can preemptively address welfare issues.

Dataset Availability

The lack of comprehensive public datasets is highlighted as a significant hindrance. The authors advocate for the development of extensive, high-quality datasets and standardized evaluation metrics to facilitate meaningful comparisons and advancements in technology.

Future Directions and Challenges

The paper identifies key areas for future research and development, including the integration of multimodal data fusion, the creation of more diverse and representative datasets, and the exploration of on-device machine learning for real-time monitoring. Also, the potential of incorporating LLMs and AGI into aquaculture is discussed, which could further advance the field by enabling more intelligent and adaptable systems.

Conclusion

This comprehensive review underscores the importance of leveraging advanced digital technologies to enhance fish tracking, counting, and behaviour analysis in aquaculture. While significant progress has been made, ongoing challenges relating to environmental variability, dataset scarcity, and technological costs must be addressed to fully realize the potential of digital aquaculture. The paper calls for a concerted effort towards creating robust, adaptable, and integrated solutions that can be deployed effectively in diverse aquaculture environments, paving the way for more efficient, sustainable, and welfare-oriented fish farming practices.