Milkfish Freshness Detection Based On Eye Images Using Convolutional Neural Network (CNN) With Mobilenetv3 Architecture On A Mobile Application

Abstract

Indonesia has abundant fishery resources, making it one of the world's largest producers and consumers of fish. One of the most commonly consumed types is milkfish (Chanos chanos). Before consumption, it is important to determine the freshness level of the fish. This freshness can be identified using a Convolutional Neural Network (CNN) model with the MobileNetV3 architecture, which is efficient and suitable for mobile application implementation. This study aims to detect the freshness level of milkfish based on eye images using the MobileNetV3 CNN architecture implemented in a mobile application. The dataset used consists of 500 images, divided into training, validation, and testing sets with proportions of 70%, 20%, and 10%, respectively. The data underwent preprocessing, including resizing and image augmentation, to increase data variation. The model was developed using hyperparameter tuning with both random search and grid search methods. The results show that random search achieved better performance with a training accuracy of 92.88%, validation accuracy of 89.90%, and an overall test accuracy of 91%. The trained model was successfully implemented into a mobile application named ScanBang, which can classify the freshness level of milkfish and display its confidence score in a practical and user-friendly manner.