CLASSIFICATION OF VIETNAMESE WHITE DRAGON FRUIT SIZE USING IMAGE PROCESSING AND CONVOLUTIONAL NEURAL NETWORK

Abstract

The objectives of this research were twofold: 1) to create a model for classifying the size of Vietnamese white dragon fruit from photographs using a convolutional neural network, and 2) to evaluate the effectiveness of the developed model. The method comprises four main components: 1) image acquisition, 2) image data preparation, which involves image processing, cropping, and enhancement, 3) dragon fruit size classification modeling, and 4) measuring model performance. The dataset for modeling, obtained from mirrorless digital cameras capturing Vietnamese white dragon fruit, is divided into five groups: 000 (Class 1), 00 (Class 2), 0 (Class 3), 1 (Class 4), and 2 (Class 5). Each of these groups contains 400 images, resulting in a total of 2000 images. Among these, 80 percent, or 1600 images were used for model training while the remaining 20 percent, or 400 images were allocated for model testing. The images have a width and height of 256 pixels, and a convolutional neural network is employed as the classification model. This research was compared with mango size classification models and apple size classification models, evaluating model performance in terms of precision at 96 percent, recall at 99 percent, accuracy at 96 percent, and an F1-score of 97 percent, respectively.