Classification of Tilapia Freshness Based on Visual Eye Images Using a Convolutional Neural Network Method

Abstrak

This study aims to classify the freshness levels of tilapia based on visual images using a Convolutional Neural Network (CNN) with a VGG16 architecture. Manual assessment of fish freshness tends to be subjective and inefficient, necessitating a more accurate and faster automated system. The research methods used include data collection, data preprocessing (resizing, augmentation), model design, training, and model evaluation. This study compares two models: a conventional CNN and a transfer learning-based CNN using the VGG16 architecture. The results show that the conventional CNN achieved an accuracy of 99.62% with a loss value of 0.0107, while the CNN with the VGG16 architecture achieved 100% accuracy with a loss value of 0.0002. Based on these results, it can be concluded that the use of the VGG16 architecture is capable of improving the model’s performance in classifying the freshness of tilapia more accurately compared to a CNN without the architecture. The developed system can serve as a solution to assist in the rapid and objective identification of fish freshness.