Publication

Abstract : Accurate fruit classification requires the identification and classification of fruits with precision according to characteristics such as type, size, color, and ripeness using methods like computer vision and machine learning. Although existing approaches primarily concentrate on tomatoes, papayas, and bananas, this research presents a model using convolutional neural networks like EfficientNetB0 and ResNet50 to overcome their shortcomings. The CNN models classify tomatoes automatically into phases such as unripe, ripe, old/rotten, and damaged; papayas into mature, partially mature, and non-mature phases; and bananas into fresh, ripe, old/rotten, and unripe phases. Experimental results show that EfficientNetB0 performed better than ResNet50, registering overall accuracies of 78.67% for tomatoes, 72.1% for papayas, and 74.18% for bananas, compared to ResNet50’s slightly lower accuracies of 62.8%, 69.85%, and 65.31% respectively. Of particular interest was a 13.35% improvement in the classification of tomatoes compared to earlier research. These results demonstrate EfficientNetB0’s capacity to extract fine image features essential for accurate ripeness determination, possibly optimizing harvest timing and quality control, and minimizing post-harvest losses in agriculture.