Deep Learning Based Ripeness Level and Shelf-Life Time Prediction for Custard Apples

Food wastage stands as a prominent global concern, Many food processing industries have turned to smart automated systems to minimize human error. This study seeks to develop a deep learning-based system focused on evaluating the ripeness of custard apples and predicting their remaining shelf-life. The proposed system aims to be user-friendly, adaptable and will be integrated with mobile phone applications or edge devices which makes it easy to use by food processing industries as well as the general public. The system consists of one mobile application with UI to capture the image and show the result to the user and two deep learning-based models, first is object detection model based on YOLOv5s architecture to detect the ripeness level of custard apple and other is EfficientNet-B0 classification model which classify the custard apple into its expected shelf-life day. The proposed model for ripeness level detection achieves ${9 3 \%}$ accuracy and shelf-life day prediction model achieves ${9 2. 7 \%}$ accuracy. Additionally, the mean squared error and mean absolute error for remaining shelf-life day prediction were calculated to be 0.0854 and 0.0769, respectively. These results suggest the potential effectiveness of the developed system in addressing food quality and wastage concerns, offering practical benefits to both industry stakeholders and consumers.