To develop a federated learning framework for precision aquaculture

Ling, John Tze Jun (2024) To develop a federated learning framework for precision aquaculture. Final Year Project, UTAR.

Preview

PDF Download (6Mb) | Preview

Abstract

In the past few years, the growing population of the world has resulted in the increase in demand for aquaculture products, this has become one of the factors for depletion of wild fish population. In order to address the issue of depleting fish stocks from overfishing and environment pollution, precision aquaculture has been chosen as an alternative with potential to supply fish stocks to fulfil the current demands. Precision aquaculture is the cultivation of years of innovation and evolution in the aquaculture industry together with the use of different technologies to improve its efficiency and automation. However, this innovation is still vulnerable towards factors such as labour shortage, diversity of data and data privacy concerns. The water quality of aquafarms needs to be monitored periodically since it plays such an important role for determining the health and growth of fishes. The accuracy and experience of farmers to measure the water parameters in their farms determine the accuracy of the results but they are prone towards human error. Data also requires to be diverse in order to train machine learning models but cannot be shared across different aquafarms due to privacy concerns at the same time. Hence, this project aims to design a federate learning framework with an IoT Cloud solution for Precision Aquaculture for predicting water quality to improve the overall efficiency and effectiveness of aquaculture operations for production. Inside this paper, aquaculture farms will be referred as edge computing environments which will consist of IoT sensors and devices to obtain the real time water parameter readings from each farm and sending them into the cloud for further processing purposes using MQTT protocol. The data will also be processed locally using federated learning frameworks to train machine learning models which are aggregated from a main model located inside the cloud. After models have successfully been trained locally, they will then be sent back into the cloud for further storage, training and analysis to build the main model for predicting water quality and raising alerts to fish farmers along with recommended solutions. Finally, the collected data and predictions will be displayed in an interactive dashboard to assist farmers in making better data-driven decisions.

Actions (login required)