Generative Adversarial Networks based Data Augmentation for Paddy Disease Detection using Support Vector Machine

One of the greatest obstacles in applying machine learning to image classification problems is the lack of a standard, online dataset in terms of the number of images of diseases. A lack of training data reduces the accuracy of the supervised models generated for classification. The traditional ways of augmenting datasets using NumPy library functions do not facilitate the generation of new images from existing ones. Generative Adversarial Networks (GANs) unlock that option by generating images which were never existed in the real world. This paper demonstrates the feasibility of augmenting the dataset available to create new images and hence generating a large, authentic, and reliable corpus for paddy diseases. It also indicates that the performance of the classifier trained on the GAN augmented corpus is better than that of augmenting the dataset using traditional methods and not augmenting the dataset at all. The Support Vector Machine (SVM) technique is used to identify diseased plants. The images of the diseased plant were obtained from online archives: Kaggle, UCI, Mendeley, and GitHub. The accuracy achieved by the proposed model is 96.23%.