Deep learning for automating fish age prediction from otolith images

Summary

Accurate estimation of fish ageing is crucial for assessing the status of fish stocks and developing management plans for ensuring their sustainability. Prevailed methods of fish ageing are based on readings of otolith images by experts, a process that is often time-consuming and costly. Neural networks are powerful tools for automating processes across various data domains. In this study, we investigated the feasibility of neural networks to provide an automatic estimation of fish age, using a dataset of 5027 otolith images of red mullet (Mullus barbatus). To accomplish this, a pre-trained convolutional neural network was adopted and configured for our case study, considering fish age estimation as a multi-class classification task. In addition, we explored, for first time, the potential benefit of multitask learning for improving network’s predictability, with the auxiliary task being the prediction of fish size. Results showed that, without multitask learning, the ages of the red mullet were predicted correctly by 64.4%, performing better on the younger Age-0 and Age-1 classes (F1 score > 0.8) and moderately on older age classes (F1 score between 0.50-0.54). Multitask learning increased the correct age prediction to 69.1% and was proved a better approach to estimate older age groups, with F1 score being between 0.57-0.64. Potential improvements of the presented approach and benefits for fisheries science are also discussed.