A citizen science approach for the collection of data to train deep learning models

Abstract

Machine learning techniques that give good predictions require a considerable amount of data, which sometimes can be a challenge to collect. In this study, citizen science is being introduced as an auxiliary technique to provide more data for the training of a deep learning network for the classification of Maltese Flora; a field which to this date lacks sufficient training data. In the first part of this study, we investigate the training of a deep learning model that makes use of a limited training dataset utilising techniques such as data augmentation, data scraping and transfer learning. This improved experimented off-the-shelf model generated a low accuracy highlighting the relevance of the initial hypothesis that citizen science is needed for the improvement of deep-learning models. In the second phase, citizen science was used as a data crowdsourcing technique through a mobile communication system. A study was conducted to determine the opinion of the public with only a small percentage showing a lack of interest in participating. Therefore, a dynamic educational application was implemented for the public exploiting Artificial Intelligence advancements to identify Maltese Flora in real-time whilst gathering images used to enhance the dataset. The deep learning model was re-trained on this dataset showing a significant increase in performance. Visualizations of the current Maltese flora distribution were also generated utilizing this data. This study demonstrated that the use of citizen science is essential for the improvement of deep learning models so that they can be employed in more widespread applications.