Fish classification is a mix of animal sciences and artificial intelligence. With the advent of machine learning in artificial intelligence, classification has been done using computer vision algorithms and now deep learning is gaining prominence. Betta fish classification is not much explored. The wild species of Betta Splendens which are native to the Kingdom of Thailand are taken in the research reported in this paper. BettaNet architecture, a modified version of ResNet 152 is used to classify 6 species of wild species of betta. The experimental results show that the proposed BettaNet architecture holds better in performance in terms of accuracy and F1-scores. Two different datasets were used and the performance obtained by the proposed architecture reduced the cross-entropy loss over different experimental configurations.

鱼类分类是动物科学和人工智能的结合。随着人工智能中机器学习的出现，分类已经使用计算机视觉算法完成，现在深度学习正变得越来越重要。斗鱼分类没有太多探索。本文报道的研究采用了原产于泰国的野生斗鱼芨芨草。BettaNet 架构，ResNet 152 的修改版本，用于对 6 种野生斗鱼进行分类。实验结果表明，所提出的 BettaNet 架构在准确性和 F1 分数方面具有更好的性能。使用了两个不同的数据集，并且通过所提出的架构获得的性能减少了不同实验配置的交叉熵损失。