Potential utilisation of Convolutional Neural Network (CNN)-based banana bunch ripeness classification to effectuate banana harvesting process

Abstract

The classification of banana bunch maturity represents a vital preliminary phase for maintaining fruit quality. However, prior studies related to non-destructive maturity classification have predominantly focused on ready-to-sell finger bananas despite the application of industrial-scale banana harvesting, which is done by bunches. This research aimed to categorize banana fruit bunches' ripeness status before the harvesting process. The classification process distinguishes between two maturity levels (unripe and ripe) utilizing the model comparison between Convolutional Neural Network (CNN), Visual Geometry Group (VGG) 16, and EfficientNet methodology. The dataset comprises 500 banana bunch images for labeling purposes. The data was partitioned in a 4:1 ratio for training and testing. The developed model utilizes CNN architecture that includes convolutional (Conv2D), pooling (MaxPooling2D), and fully connected layers. Evaluation outcomes indicate that the model effectively classifies the maturity of banana bunches, demonstrating high accuracy, precision, and recall. The conventional basic CNN resulted in the most optimal model among VGG16 and EfficientNet with precision up to 91.11%. This CNN-based classification system is anticipated to be integrated into the banana industry, aiming to maintain the harvested banana bunches. By employing CNN for classifying the maturity of banana bunches, the harvesting process can be made more efficient with less time needed. Furthermore, the system enhances automation and consistency in product quality while decreasing dependence on manual labor. Additionally, the classification outcomes can be directed towards appropriate processing pathways, thereby facilitating the implementation of smart technology-driven postharvest systems over time.