Classification of Banana Ripeness Using Transfer Learning VGG16 and Logistic Regression

Abstract

Banana ripeness classification plays an important role in improving the efficiency of postharvest handling, sorting, and distribution processes. Conventional visual inspection is subjective and often produces inconsistent results due to human judgment. This study proposes an automatic banana ripeness classification model by combining Transfer Learning VGG16 as a feature extractor with Logistic Regression as the classification algorithm. The dataset consists of 1,490 images of Cavendish bananas categorized into four ripeness levels: unripe, semi-ripe, ripe, and overripe. Image feature extraction was performed using the pre-trained VGG16 model, while Logistic Regression was employed to classify the extracted feature vectors. The proposed model was implemented using Orange Data Mining and evaluated under three validation scenarios, namely 10-fold cross-validation, 80:20 random sampling, and 70:30 random sampling. Model performance was assessed using Accuracy, Precision, Recall, F1-Score, Area Under the Curve (AUC), Matthews Correlation Coefficient (MCC), Confusion Matrix, and ROC Curve. Experimental results indicate that the 10-fold cross-validation scenario achieved the highest performance, obtaining 97.8% accuracy, 97.8% precision, 97.8% recall, 97.8% F1-score, 0.999 AUC, and 0.970 MCC. These findings demonstrate that the proposed approach provides accurate and computationally efficient banana ripeness classification, making it suitable for supporting automated fruit sorting applications.