The inability to collect actionable data on-site, in a time-efficient manner, directly hinders how farm management strategies can react to problems as they arise. Many sectors within the aquaculture industry have therefore begun to adopt a 'precision farming' approach, applying various technologies to facilitate automated, real-time monitoring and analysis of various production, health, and welfare related parameters of their farmed animals. Owing to the fact that behavioural assessments are often carried out through some form of visual observations, typically aided with underwater feed cameras, a number of emerging technologies integrating with such video surveillance & recording systems have the potential to radically transform the practicality and effectiveness of not just behavioural WIs, but welfare assessments overall. This section not only outlines how such innovations might address the logistical hurdles outlined in the previous section, but how they may further enhance the unique advantages associated with behavioural assessment.

 Improving the operational feasibility of quantitative, behavioural WIs:

Regardless of the farming system or staff training involved, technological advances will be required to turn the majority of measurable behavioural analyses into OWIs. Emerging solutions that offer real-time, objective, automated, and continuous tracking & monitoring of fish behaviour will need to be developed and adapted to the farm environment and its specific demands for welfare monitoring. 

Automating and facilitating the detection & tracking of behaviours

Computer vision and machine learning have already found real world applications in facilitating non-intrusive, automated methods for in-situ monitoring of fish behaviours.

What is computer vision?
This refers to a technology that enables machines/computers to process, interpret, and construct explicit information and meaningful descriptions of physical objects via image analysis. Innovations in this area have grown rapidly, becoming more powerful and accessible alongside the developments in digital cameras and computer processing power.

What is machine learning?
This refers to a data-driven process in which models automatically generate predictive insights by identifying patterns within the data they are given. Put simply, it is similar to teaching a computer to recognise patterns and make sense of information it is provided with.

Through these technologies, video tracking (i.e., tracking of moving objects and monitoring their activities through processing the sequence of images captured in a video recording) can be applied to help automatically quantify behavioural WIs. The benefits of automated systems that can detect behavioural changes revolve around their efficiency after their implementation, and they include:

1) No labour required to obtain behavioural information; subtle changes are also far less likely to go unnoticed.

2) Abnormal deviations from normal levels of activity can be automatically detected with no requirement of staff involvement.

 Sonar, Near-Infrared (NIR), and optic video imaging, in combination with sophisticated machine learning models, have already been used to facilitate regular behavioural monitoring under sub-optimal conditions. Such technologies will prove essential in helping to overcome the aforementioned hurdles of assessing behaviour in an aquatic environment (sub-optimal conditions and poor water quality, complex individual behaviours within large groups, etc.).

Automating and facilitating analysis of behaviours

The description of precision farming technologies outlined so far have referred to the initial detection and tracking of behaviours. However, to automate in-depth analyses of these behaviours (to the point where actionable information on farmed animal welfare can be obtained in the same way as if it was carried out by an expert), further advancements are still needed. 

Deep learning & Convolutional Neural Networks (CNN) are powerful machine learning technologies that have already been successfully applied to various computer vision applications including behavioural recognition and classification. This integrated system could eventually play a central role in on-farm welfare assessments by providing fast, objective, practical, and reliable ways to analyse massive amounts of monitoring data, all without any influence from observer bias.

To learn more about precision farming you may be interested in this OpenLearn Create course - Precision Aquaculture