Betta Fish Classification Using Faster R-CNN Approach with Multi-Augmentation

Betta fish contests have become a popular hobby among fish enthusiasts worldwide. Accurate and consistent assessment of betta fish is crucial in these contests, with judging standards based on physical shape, color patterns, and other attributes. However, the evaluation of fish shape characteristics often requires greater attention. This study proposes an assessment method emphasizing the movement of Halfmoon Betta fish in an aquarium, enhanced by multi-augmentation image techniques. During the testing phase, the approaches consist of comparing the detection outcomes of Faster R-CNN without augmentation and Faster R-CNN with multi-augmentation. The main contribution of this research is the deployment of advanced approaches for identifying objects and using several augmentations to improve the performance of the model. The experimental results show that the model that includes multi-augmentation obtains a mean Average Precision (mAP) of 99%, which is higher than the model without augmentation that achieves a mAP of 97%. This means that the model with multi-augmentation is able to recognise objects with more accuracy.