Wheat Rust Disease Detection and Classification in Complex Natural Backgrounds

Abstract

Wheat rust disease poses a significant threat to global wheat yield and grain quality. Early detection of this disease will help to minimize the loss caused by its effects.Existing models work well on images taken in controlled environment, where a uniform background is placed behind the leaf, but these models fail to produce good results in natural settings. In this paper a complete pipeline is proposed that uses yolov8 with an unsupervised leaf rotation algorithm to localize the leaf in such a way that maximum background is being removed at the earliest. Then the performance of multiple segmentation models have been compared to show how good localisation will give good segmentation results even on small dataset.Segmentation models ie UNet, Segment-Anything (SAM),Segnet, LinkNet, PSPNet, FPN, DeepLabv3+(Xception), DeepLabv3+(MobileNet) are used. As Unet has outperformed all the other segmentation model with an IOU score of 0.9563, dataset for classification is prepared by segmenting the leaves using this trained Unet. Lastly for classification, performance of multiple convolution neural network ie. VGG16, Resnet101v2, Xception, MobileNetV2 and Transformer based models ie. Swin transfomer and MobileVit has been compared. Swin transformer have outperformed the state of the art CNN models with an accuracy of 95.8%. This research also highlights that effective preprocessing techniques can enable deep models to achieve very good performance, even when dealing with very small datasets.This research can be extended by identifying the disease on all the leaves in the frame.