Incorporating deep features extracted from convolutional neural networks to utilize machine learning classifiers for improved identification of Maize Leaf Disease

Abstract

Maize is one of the most produced crops in the world and a significant contributor to the economy of various countries. Maize leaf diseases can lead to hamper of crop production and eventually reduce profit of agricultural farms. Through accurately identifying maize leaf disease earlier, farmers can take the necessary steps to minimize damages. In this paper, we propose to incorporate features extracted from deep convolutional neural networks and train them using machine learning classifiers for the identification of maize leaf diseases with high accuracy. For feature extraction, we trained 5 CNN models, which are InceptionResNetV2, DenseNet121, EfficientNetV2S, Xception and InceptionV3, reaching accuracy of 99.172%, 98.965%, 98.654%, 98.344% and 98.965%. Furthermore, the features extracted using these models were used to train K-Nearest Neighbors and Support Vector Classifier. The K-Nearest Neighbors classifier reach an accuracy of 99.586%, while the Support Vector Classifier reached an accuracy of 99.379%.