A robust deep learning model for fish image classification

Abstract Fish image classification presents significant challenges due to the complexities of underwater environments and the diversity of species. This study introduces a hybrid deep learning approach combining pre-trained ResNeXt variants as feature extractors with Twin Fuzzy Extreme Learning Machines (TFELM), an enhancement of Fuzzy Extreme Learning Machines (FELM) known for handling noise and improving generalization. By leveraging ResNeXt’s cardinality-based transformations and the specialized capabilities of TFELM, our model effectively addresses dataset-specific nuances, enhancing classification accuracy and robustness. Evaluations on two Kaggle datasets yielded remarkable accuracy rates of 99.53% and 95.18%, respectively, outperforming other state-of-the-art methods. These findings demonstrate the effectiveness of our approach in advancing fish species classification, particularly for applications in marine biodiversity monitoring and conservation. The integration of fuzzy logic with deep learning underscores the model’s resilience in managing complex and diverse datasets, highlighting its potential to revolutionize ecological monitoring technologies.