Recent years have seen a significant development in plant stress phenotyping and classification systems based on deep learning, multimodal analysis, and computer vision techniques. To get the most out of machine learning approaches, a plethora of data is required. For the aim of developing models to recognise and diagnose plant stress, it requires a considerable time and effort to gather a sizable amount of training data in the agricultural industry. Data augmentation enriches the variety of training data for machine learning systems without the need to collect new data. Basic image manipulation and deep learning-based image augmentation methods were used to create augmented plant stress disease datasets in this article. The extended simulation shows that the improved dataset made using DCGAN, WacGAN with NST methods performs better than the original dataset produced with traditional picture editing techniques. However, the detection of stress at the time are key variables in determining how much information is available. We'll need to apply data augmentation methods to study performance of different transfer learning techniques. Multi-crop dataset expansion is handled in this study, which is subsequently used to train a bespoke multi-crop stress phenotyping classification.