Fish Location Forecasting in the Norwegian Ocean Using Deep Learning Techniques

Róason, Jón (2024) Fish Location Forecasting in the Norwegian Ocean Using Deep Learning Techniques. Bachelor's Thesis, Artificial Intelligence.

Preview	Text bAI2024RoasonJ.pdf Download (6MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (156kB)

Abstract

Successful forecasting can increase fishing efficiency by yielding larger catches, lowering fuel consumption, and reducing environmental pollution. This study aims to develop a neural network-based machine learning model capable of predicting the locations where fish will be present in the following 4 days, given the environmental features and heuristics from the previous 4 days. Datasets in matrix formats consisting of reported catch quantities, sea salinity values, and sea surface temperature are used to train the models. The U-Net model architecture is the focus of the study, as it is specialized for mapping matrices to matrices and has shown good predictive performance for tasks involving geo-spatial data. The results showed that the trained models performed poorly on test data, failing to predict future fishing locations accurately. The shortcomings of the models mainly resulted from the sparsity of the catch dataset and the inability to process temporal data natively. Despite these limitations, the study highlights the potential for future research to explore other advanced neural network architectures, such as recurrent U-Nets. Future research should also explore using more comprehensive ocean data collection methods, such as using echo-sounding data, which could mitigate sparsity, enhance model performance, and contribute to sustainable fishing practices.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Cardenas Cartagena, J.D.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	24 Sep 2024 13:30
Last Modified:	24 Sep 2024 13:30
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/34256

Actions (login required)

View Item