Deep Learning-Based Buoyancy Prediction for Intelligent Mussel Farm Monitoring
Abstract
Greenshell mussel farming is the largest contributor to export earnings in Aotearoa New Zealand's (NZ) aquaculture industry. Floats are essential to mussel farm structures, keeping tons of mussels suspended in the water. Monitoring float buoyancy is essential for minimising the loss of mussels due to sinking or excessive energy transfer to the mussels. We propose a novel transformer-based neural network architecture for float buoyancy prediction in NZ's greenshell mussel farms. The model extends a vision transformer to incorporate relationships between adjacent floats, enhancing prediction performance. Results show improved accuracy over baseline methods, suggesting potential for integration into a cost-effective, computer vision-based monitoring system for mussel farm buoyancy at scale. This approach could address the challenges of monitoring buoyancy in expanding coastal and open ocean aquaculture operations.
BibTeX
@inproceedings{mcmillan2024deep,
title = {Deep learning-based buoyancy prediction for intelligent mussel farm monitoring},
author = {McMillan, Carl and Zhao, Junhong and Xue, Bing and Vennell, Ross and Zhang, Mengjie},
booktitle = {2024 39th International Conference on Image and Vision Computing New Zealand (IVCNZ)},
pages = {1–6},
year = {2024},
organization = {IEEE},
abstract = {Greenshell mussel farming is the largest contributor to export earnings in Aotearoa New Zealand's (NZ) aquaculture industry. Floats are essential to mussel farm structures, keeping tons of mussels suspended in the water. Monitoring float buoyancy is essential for minimising the loss of mussels due to sinking or excessive energy transfer to the mussels. We propose a novel transformer-based neural network architecture for float buoyancy prediction in NZ's greenshell mussel farms. The model extends a vision transformer to incorporate relationships between adjacent floats, enhancing prediction performance. Results show improved accuracy over baseline methods, suggesting potential for integration into a cost-effective, computer vision-based monitoring system for mussel farm buoyancy at scale. This approach could address the challenges of monitoring buoyancy in expanding coastal and open ocean aquaculture operations.},
}