DETEKSI KESEGARAN IKAN BANDENG DENGAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK (CNN)

Abstract

Fish freshness is a key indicator in ensuring food quality and safety, especially in milkfish (Chanos chanos) which is widely consumed in Indonesia. Manual freshness assessment is subjective and requires special skills, so an accurate automated approach is needed. This study aims to develop a digital image-based milkfish freshness classification application using the Convolutional Neural Network (CNN) method with a transfer learning approach. The dataset used consists of 445 milkfish images in two classes: fresh and not fresh, with an augmentation process to enrich the visual variety. Two models were compared: Model A (baseline) and Model B (enhancement with Dropout and fine-tuning). The evaluation results show that Model A has 33% accuracy, 50% precision, and 50% recall, In contrast, Model B has 67% accuracy, 50% precision, and 100% recall, showing more stable prediction in Streamlit-based applications. These findings suggest that the integration of CNN and transfer learning can be effectively applied to support the digitization of fish-based food product quality. Further development is suggested through the addition of training data, multi-class classification, and integration to mobile or IoT devices.