Ensemble learning weighted average meta-classifier for palm diseases identification

Abstract

Crop diseases lead to significant losses for farmers and threaten the global food supply. The date palm, valued for its nutritional benefits and drought resistance in desert climates, is a vital export crop for many countries in the Middle East and North Africa, second only to hydrocarbons. However, various diseases pose a threat to this important plant. Therefore, early disease prediction using deep learning (DL) is essential to prevent the deterioration of date palm crops. The aim of this paper is to apply a robust ensemble method (EL) combining tree transfer learning (TL) models Resnet50, DenseNet201, and InceptionV3, and compares its performance with the CNN-SVM model and the tree TL models mentioned previously. The models were applied to a date palm dataset containing three classes: White scale, brown spot, and healthy leaf. The training and validation sets were applied to a public dataset, while the testing set was applied to a local dataset captured manually to check the model’s performance. As a result, we considered that the ensemble method gave very satisfactory results compared to other methods. Our hybrid model reached a testing accuracy of 98% while achieving an amazing training and validation accuracy of 99.94% and 98.14%, respectively.