<![CDATA[Banana ripeness is one of the main factors affecting fruit quality before distribution and consumption. Conventional ripeness assessment is generally performed through visual observation, which may lead to subjective and inconsistent results. Previous studies have reported promising results in banana ripeness classification. However, distinguishing adjacent ripeness stages remains challenging because of their similar visual characteristics. This study proposes a banana ripeness classification model using a Convolutional Neural Network (CNN) with the ResNet50 architecture. The dataset consisted of four ripeness categories, namely unripe, ripe, overripe, and rotten. Data balancing, preprocessing, and augmentation were applied before model training. A total of 1,120 images were used to train and evaluate the model. Transfer learning with full fine-tuning was employed to adapt pretrained visual features to different banana ripeness levels. The experimental results showed that the proposed model achieved an accuracy of 92.86%, while precision, recall, and F1-score reached 93%. Several misclassifications were observed between adjacent ripeness categories due to similarities in visual characteristics. These results indicate that the proposed ResNet50 model can effectively classify banana ripeness levels on the testing dataset based on color and texture information learned automatically from digital images.]]>
Copyrights © 2026
