AI and smart glasses in prawn farm management
Obtaining frequent prawn growth data using smart glass, computer vision and machine learning while adding minimal effort to a prawn farmer’s daily workflow
The challenge of monitoring prawn weight and size
In prawn farming, continual monitoring of the average weight and size distribution of prawns in a pond is important to optimise husbandry and harvest strategies.
Current best practice involves casting a net that catches a sample of prawns that are then bulk weighed and individually counted to estimate the average weight. It’s an extremely labour-intensive task and usually only carried out once a week or once a fortnight. The infrequent measurement practices may result in growth issues being unnoticed for weeks that in turn can have a significant economic impact.
High tech hard hat with Google Glass, Raspberry Pi and camera
However, another practice that takes place on prawn farms is where technicians pull up feed trays to check feed consumption and adjust feed rates. This is part of their daily workflow. The trays typically hold a good number of prawns as they’re lifted. In this project, we aimed to take advantage of this practice as it is more frequent – twice daily – than the fortnightly net casting. This is where smart glass, computer vision (CV) and machine learning (ML) can contribute.
Our prototype headset integrates the Google Glass smart glass system, a Raspberry Pi computer, and an Intel Realsense depth camera all housed on a hard hat.
A ‘smart glass’ is an augmented reality technology that uses eyewear to merge what one sees in the real world with virtual information, usually overlaid on one of the glasses’ lenses. The smart glass interface is used to capture images from a depth camera mounted on the headset. We developed computer vision and machine learning algorithms to then convert the images of the prawns in the trays to size estimates of the prawns.
The aim is that farmers can automatically and frequently gather prawn growth measurements hands-free, just by looking at them while they’re going about their existing daily operations.
High accuracy detection of prawns
We deployed the hard hat at our aquaculture research facility on Bribie Island, Queensland, collecting images of prawns for a period of seven weeks. It successfully captured the field quality data and the computer vision part detected prawns with high accuracy and revealed variability of prawn size over time better than cast net-based methods.
We still need to address a number of problems to make the prototype market ready, such as better utilise field quality images with ‘noisy’ depth measurements.
We see a number of other areas that could be explored with this system – better sampling and disease detection in cropping, for example. or getting real time feedback on animal conditions in livestock production.
Meet the team
Dr Ashfaqur Rahman
- Ashfaqur's sensor data analytics team develop machine learning algorithms for transforming sensor data into decisions. He led Digiscape’s work on smart glass, computer vision and machine learning for efficient prawn farm management.
Dr Mingze Xi
- Mingze led the augmented reality component of the aquaculture project. He is a senior experimental scientist with CSIRO's Data61.
Dr Chuong Nguyen
- Chuong's research interest include mixed reality, computer vision, and machine learning for applications in medical, agriculture, ecosystems, supply chain, smart vehicles, smart cities, smart energy, and smart manufacturing.
- John has a robotics and sensing background which has been applied loosely in the marine space for a decade. His current focus is on bringing IoT and machine learning to the aquaculture and, more broadly, agriculture sectors.