Enhanced VGG16 Architecture in Mango Ripeness Classification

Abstract

This study introduces an Enhanced VGG16 Architecture for mango ripeness classification, addressing the critical challenge of accurately determining fruit ripeness in agricultural settings.
Leveraging deep learning and computer vision techniques, the research developed a machine learning model to classify mango ripeness stages with high precision. The methodology employed an experimental design using a dataset of mango images from Kaggle and local sources. The
Enhanced VGG16 model was developed by modifying the standard VGG16 architecture, incorporating a Multi-Layer Perceptron block, Batch Normalization, and Dropout layers to improve feature extraction and classification capabilities. The results demonstrated significant
improvements over the standard VGG16 model. The Enhanced VGG16 achieved an overall accuracy of 89%, with exceptional performance in classifying ripeness stages. Precision and recall rates exceeded 97% for "Ripe" and "Unripe" categories, with notably improved performance in
challenging classifications like "Early Ripe" stages. Comparative analysis revealed the Enhanced VGG16's superior generalization, reducing misclassifications and providing a more reliable approach to ripeness detection. This research contributes to precision agriculture by offering a
data-driven solution for mango ripeness assessment, potentially reducing post-harvest losses and improving quality control in the agricultural sector.