Quality prediction of seabream Sparus aurata by deep learning algorithms and explainable artificial intelligence

Area of Science:

• Food Science
• Computer Science
• Artificial Intelligence

Background:

• Assessing fish freshness is crucial for food safety and quality control.
• Traditional methods for determining fish freshness can be time-consuming and destructive.
• Developing objective, non-destructive methods for fish quality assessment is an ongoing research area.

Purpose of the Study:

• To evaluate the effectiveness of various machine learning algorithms in determining sea bream freshness.
• To assess the quality changes in refrigerated sea bream using image analysis.
• To validate the performance of Convolutional Neural Network (CNN), DenseNet121, Inception V3, and ResNet50 models.

Main Methods:

• Utilized Convolutional Neural Network (CNN), DenseNet121, Inception V3, and ResNet50 algorithms.
• Analyzed eye and gill images of sea bream categorized into fresh, moderate, and spoiled classes.
• Employed Explainable Artificial Intelligence (XAI) algorithms, including Grad-CAM and LIME, for result interpretation.

Main Results:

• Machine learning models achieved high prediction accuracy, with the lowest performance at 98.42% for the spoiled class using eye parameters.
• The models demonstrated excellent capability in distinguishing between different freshness levels of sea bream.
• Explainable AI methods provided insights into the decision-making process of the models.

Conclusions:

• Convolutional Neural Network (CNN) and DenseNet121 models, combined with Grad-CAM and LIME, offer a reliable non-destructive method for assessing sea bream freshness.
• Image-based machine learning is a promising approach for real-time quality control in the fish industry.
• This technology can help ensure the quality and safety of seafood products.