TAF PUBLISHING

Fruit quality, including color, texture, size, and defects, significantly influences their market value and customer preference. Conventional manual fruit quality assessment methods are both time-consuming and laborintensive. Image processing methods, particularly convolutional neural networks (CNNs), have proven to be efficient in automating fruit quality recognition to address this issue. This study compares various CNN models' accuracy in classifying images of fresh and rotten apples, bananas, and oranges. We assessed three pre-trained CNN models—MobileNet V2, ResNet50, and VGG19—alongside the K-Nearest Neighbors (KNN) algorithm. The findings suggest that VGG19 achieved the highest accuracy at 99.56%, followed by MobileNet V2 at 98.37%, and ResNet50 at 94.21%. The accuracy of the KNN algorithm, however, was notably lower at 68.18%. This study sheds light on the effectiveness of different CNN models for assessing fruit quality and offers direction for future research in fruit image classification.