Identification of Stages of Ripening of Dragon Fruit Using Neural Networks for Smart Agriculture

Abstract

Dragon fruit is a popular fruit with a unique appearance and taste. It is an important fruit in export and domestic markets. However, its maturity detection is still a challenging task due to the complexity of its physical properties. This research study introduces a new approach by utilizing the VGG16 model and SVM to detect the maturity of dragon fruit. For the purpose of increasing the datasets, the data augmentation techniques were applied that was followed by preprocessing, thresholding, edge detection and contour detection, and extracting the ROI. The segmented images were then sent to the VGG-16 model that provided accuracy of 95.93%, 95.31% and 96.54 % for unripe, partially ripe and ripe stages. The features extracted for the fruit region are mean, standard deviation, entropy, contrast, correlation, Inverse difference moments. These are fed to the SVM classifier that generated accuracy of 91.93%, 91.93 % and 92.54% accuracy for unripe, partially ripe and ripe stage −16 performed better than SVM classifier.