Grape Leaf Spot Identification Under Limited Samples by Fine Grained-GAN

In practice, early detection of disease is high importance to practical value, corresponding measures can be taken at the stage plant disease. However, in or when a rare occurs, there are limited training samples which makes machine learning especially deep models hardly work well while stronger representative ability needs large-scale data. To solve this problem, fine grained-GAN based grape leaf spot identification method was proposed for local area image data augmentation generated images were added and fed them into further strengthen generalization classification models, effectively improve accuracy robustness prediction. Including 500 early-stage every category augmentation, 1000 sub study. The improved faster R-CNN integrated as detector. After that, segmented sub-images mixed input original used testing. Experimental results showed that had achieved higher on five state-of-art models; ResNet-50 got 96.27% accuracy, obtained significant improvements than other methods verified its satisfactory performance. This great significance diseases samples.