Handling of Class Imbalance for Plant Disease Classification with Variants of GANs

Abstract

Plant leaf diseases are one of the major threats to the agriculture sector that significantly contribute to yield losses. Swift and accurate detection of plant leaf diseases is essential to reduce the intensity of the disease thereby minimising economic losses. Here, a deep learning-based Inception-v3 methodology has been proposed to identify and classify various plant leaf diseases using plant leaf image datasets. The approach also employs Generative Adversarial Networks (GANs) to augment the limited datasets. Different classes of GANs are adopted for experimental analysis to evaluate the performance of the proposed model. From the experiment’s results, it is observed that the DCGAN model achieves the highest accuracy and performs better than CGAN as a data augmentation technique in terms of Class Accuracy, Precision, Recall, F1 Score and Accuracy. The DCGAN model also outperforms in terms of evaluation parameters when compared with other techniques in literature.