Betta fish species classification using light weight deep learning algorithm

Abstract

Betta fish sellers and breeders often face challenges in accurately identifying Betta fish species due to variations in colors, patterns, and shapes, leading to potential financial losses and deceptive transactions. To address this issue, we developed a mobile application that employs MobileNet, a deep learning (DL) technique, to classify Betta fish species. The dataset, acquired from online stores, comprises 400 images, with 100 images representing each of the four studied Betta fish species: comb tail, delta tail, spade tail, and veil tail. Prior to model implementation, the dataset undergoes pre-processing with data augmentation techniques, including rotation, shear, zoom-in, horizontal flip, and brightness adjustments, enhancing the model performance. Training utilizes 80% of the data, with the remaining 20% allocated for testing. Three distinct MobileNet models are developed for males, females, and both genders combined, achieving accuracies of 70, 83.75, and 65%, respectively. These trained models are the foundation for a mobile application developed for the Android platform that enables users, particularly Betta fish sellers, and breeders, to efficiently classify Betta fish species, empowering them to set accurate prices based on the identified species.