CNN Models for Identification of Macro-Nutrient Deficiency in Onion Leaves (Allium cepa L.)
Laurence Kobe B. Navarro, Kristin Clarisse H. Mateo, Cyrel O. Manlises
Abstract
Onions are one of the most common vegetables in the agricultural industry in the Philippines. However, nutrient deficiency affects the crop yield. One of the ways to determine the health status or nutrient deficiency of the onions is to check the color and texture of the onion leaves. In this study, different convolutional neural network (CNN) models, namely ResNet-50, VGG-16, and InceptionV3, were used to identify the primary macro-nutrient deficiency (nitrogen, potassium, or phosphorus). The three transfer learning models were compared for their results including the learning rate, training accuracy, validation accuracy, and best epoch time. The ResNet-50, VGG-16, and InceptionV3 models obtained an overall accuracy of 54, 85, and 58%, respectively. The best CNN model for the developed system was the VGG-16 model which achieved an accuracy of 85%.