Generative Adversarial Networks Based Approach for Data Augmentation in Mango Leaf Disease Detection System

Mango is a precious commodity in India as it is also known as the ‘‘king of Fruits However much of the crop yield is lost due to diseases, even if the government intends to use some machine learning algorithms to reduce the human workforce required to identify and classify the plants there is not an extensive dataset available to train models and get good accuracy. The solution to this issue can be found in the field of Reinforcement learning algorithms to generate new images of infected and healthy Mango leaves. This is a form of data augmentation. In this study, three algorithms were trained to create an entirely new dataset based on a manually collected Mango leaves dataset as inputs. The three algorithms Wasserstein Generative Adversarial Networks (WGAN), Deep Convolutional Generative Adversarial Networks (DCGAN) and Generative Adversarial Networks (GAN) are then compared with each other to find the best algorithm to produce the new dataset