LeafGenEx-A Novel Method for Generating Healthy and Diseased Wheat Leaf Images using CycleGAN

Abstract LeafGenEx, an innovative approach to wheat leaf disease detection, uses data augmentation, deep learning, and explainable AI to improve disease detection accuracy. The procedure starts with preprocessing methods, such as image flipping, rotation, and cropping, to extend the original dataset. Segmentation is eventually carried out utilising a pre-trained ResNet network and GradCAM, allowing for exact identification of disease-affected regions. An upgraded CycleGAN model is used to produce synthetic images of healthy and diseased leaves, yielding a higher Fréchet Inception Distance (FID) score than previous DCGAN and CycleGAN models. The generated images are combined with the original dataset to form a comprehensive dataset for training a deep learning detection model. To categorise leaf images according to their different diseases, the authors use a method based on transfer learning with InceptionV3. GradCAM, an explainable AI approach, is used to evaluate the deep learning model's results and determine the most important leaf sections for disease detection. LeafGenEx's effectiveness is proved by its capacity to generate high-quality synthetic images, improve disease detection accuracy, and produce interpretable data, making it an important tool for precision agriculture and early disease intervention.