Fish species classification using otolith images

Abstract

Otoliths are calcium carbonate structures found in the inner ear of fish. They are used for balance and hearing, and their shape and size can be used to identify fish species and age. The primary objective is to facilitate an automatic otolith identification using the power of deep learning method. The proposed method focused on the utilisation of transfer learning to achieve better accuracy over other traditional methods, which is essential in case of availability of limited labeled data. The objective of using transfer learning in fish otolith identification and classification tasks is to improve the performance of machine learning models by transferring knowledge from a pre-trained model that has been trained on a large dataset of labeled images. This approach proved to be effective in this study with a small dataset of otolith images as it was able to achieve better accuracy beating other models and traditional machine learning methods which relied heavily on feature engineering and feature extractions. The classification accuracy of the proposed method of fine tuning a pre trained Xception model on imagenet dataset on the otolith dataset is 93.75% which is better than previously used inception-V3 and VGG16 models