Authors - Riyazahemed A Jamadar, Anoop Sharma, Kishor Wagh Abstract - Recent developments in deep learning have provided a broad spectrum of algorithms and tools for precision agriculture. It is evident from the research work hitherto carried out in this domain, that the deep learning based system out-performs when trained with large size datasets. As the public datasets available for this domain are relatively smaller, the deep learning models cannot be leveraged to its full capacity. To beat this issue, we are proposing a data augmentation technique that combines Generative Adversarial Networks(GANs) with Transfer Learning that substantially increases the size of datasets. The prosed technique uses GAN for generating synthetic images of pomegranate leaves, whereas through Transfer Learning fine tuning and reduced convergence time of the model are achieved. The dataset used for GAN is real-time images of pomegranate plant leaves and DenseNet is trained with combination of real pomegranate images and GAN generated synthetic images for Transfer Learning. The experimental results show the efficacy of the proposed work to higher levels when compared with standalone approaches.
Open Zoom