Classification Of Banana Maturity Level Using CNN Architectural Method VGG19

Abstract

Bananas are abundantly available in Indonesia, rich in nutrients, and hold high economic value. However, the post-harvest sorting process still relies on inconsistent human judgment, resulting in losses for farmers. Therefore, this research proposes the use of Convolutional Neural Network (CNN) to classify the ripeness of bananas based on color. The dataset consists of 450 banana images with three ripeness classes: raw, ripe, and overripe, sourced from Kaggle. Data augmentation is performed using Image Data Generator. CNN is designed using the VGG-19 architecture and trained using both Adam and SGD optimizers. The research results show the highest accuracy of 100% with the lowest loss of 0.02 when using the Adam optimizer with 20 epochs. The SGD optimizer also yields 100% accuracy with a loss of 0.04 at epoch 20. The research conclusion indicates that CNN with the VGG-19 architecture can be used for banana ripeness classification with high accuracy rates. For future developments, the model will be enhanced with layer adjustments and preprocessing to improve accuracy and reduce data loss.