Enhancing Fish Freshness Assessment for Sustainable Fisheries: A Deep Learning Approach with MobileNetV1

The rapid expansion of aquaculture intensifies the issues concerning global fisheries and highlights the critical concerns for ocean health due to climate change, pollution, and unsustainable traditional practices. These challenges led to other relevant issues, including fish freshness assessments that, by conventional practices, are inefficient, subjective, and prone to errors. It is, therefore, urgent to introduce advanced technologies in this field to protect consumers' health and ensure economic sustainability. This study proposed using a deep learning approach to automate the classification of fish freshness. For this purpose, the study utilized the Freshness of the Fish Eyes dataset, comprised of 7,809 images across eight different fish species. Each image in the dataset is categorized into either of the three freshness levels: highly fresh, fresh, and not fresh. The researchers used various augmentation techniques to increase the number of datasets, including random vertical flip, random horizontal flip, random rotations, zoom range, and shear range. The dataset was divided into training (75%), testing (10%), and validation (15%) sets, with both augmented and non-augmented versions to evaluate the model's performance under varied conditions. Employing MobileNetV1 architecture, the researchers conducted experiments to assess the model's accuracy, precision, recall, and F1 score, achieving an average accuracy of 88.54%. This result demonstrates the model’s potential as a reliable tool for fish freshness classification. Furthermore, the study explored the differences in accuracy between augmented and non-augmented datasets, revealing insights into the model's adaptability. The findings suggest that data augmentation techniques can significantly enhance model performance, highlighting the importance of robust dataset preparation in machine learning applications. The implications of this research extend beyond the seafood industry, suggesting potential applications in other food quality assessments where visual indicators play a crucial role. By integrating advanced machine learning models like MobileNetV1 into quality control processes, the seafood industry can achieve greater efficiency, accuracy, and consumer confidence.