Data Augmentation Method for Plant Leaf Disease Recognition

Recently, several plant pathogens have become more active due to temperature increases arising from climate change, which has caused damage various crops. If change continues, it will likely be very difficult maintain current crop production, and the problem of a shortage expert manpower is also deepening. Fortunately, research on early diagnosis systems based deep learning actively underway solve these problems, but lack diversity in some hard-to-collect disease samples remains. This imbalanced data bias machine models, causing overfitting problems. In this paper, we propose augmentation method an image-to-image translation model by supplementing insufficient diseased leaf images. The proposed performs between healthy images utilizes attention mechanisms create that reflect evident textures. Through improvements, generated plausible image compared existing methods conducted experiment verify whether could further improve performance classification for plants. experiment, PlantVillage dataset was used, extended built using original images, models evaluated through test set.