Citrus pests classification using an ensemble of deep learning models

Early diagnosis of plant pests is essential for reducing the consumption of agricultural pesticides as well as saving costs and reducing environmental pollutions. In this paper, an intelligent method based on deep learning is presented to recognize three common citrus pests including citrus Leafminer, Sooty Mold, and Pulvinaria. To this end, an ensemble classifier of deep convolutional neural networks is presented to recognize citrus pests. For constructing this ensemble, three levels of diversity including classifier level, feature level, and data level diversity are considered. In the training phase, data augmentation is used to increase the number of training samples and improve the generalizability of classifiers. The proposed method has been evaluated on a dataset of 1774 citrus leaf images. All images were taken in field conditions by various cameras in distinct time intervals, angles, scales, and light conditions. For experimental analysis, 10-fold cross validation is used to measure the accuracy of CNNs. Based on the experimental results, the proposed ensemble achieved an accuracy of 99.04% which outperformed other competing CNN methods.

Automated summary

This study presents an intelligent method based on deep learning to recognize citrus pests. Data augmentation is used in the training phase to increase the number of training samples. Experimental results show that the proposed method achieved an accuracy of 99.04%, outperforming other competing CNN methods.