Development of a Convolutional Neural Network Method for Classifying Ripeness Levels of Servo Variety Tomatoes

The distribution of tomatoes in Indonesia is huge, making it an important commodity in the agricultural sector. However, manual classification of tomato ripeness can lead to human error and decrease supply chain efficiency. Therefore, an automated system capable of classifying tomatoes quickly and accurately is needed, in order to reduce the potential for human error and improve supply chain efficiency. This research aims to develop the Convolutional Neural Network (CNN) method to improve the accuracy of tomato ripeness detection through modifications to the architecture, such as reducing several layers, adding batch normalization, and adding dropouts. The dataset used in this study consists of 500 images taken by the researcher himself which are divided into 5 classes, namely unriped, half-riped, riped, half-rotten, and rotten, with each class containing 100 images. There are 3 proposed CNN models, namely the standard model, as well as the addition of batch normalization and dropout in the architecture. The results showed that the proposed model 3 with the addition of dropout on several layers of its architecture is the optimal model with a parameter of 2.4 million and using a batch size of 16 resulting in an accuracy of 98%, as well as precision, recall, and F1-score values of 98%. With these results, the proposed CNN model is effective in identifying the ripeness level of tomato fruit. This research is expected to be applied in the agricultural industry to improve the efficiency of sorting and distributing tomato fruits according to the desired quality standards.