Yellow rust disease detection from wheat images

Abstract

Yellow rust, a devastating fungal disease caused by Puccinia striiformis have been identified as a major hindrance to wheat production to millions of metric tons being lost annually, in turn exacerbating food insecurity. To minimize its negative effects on wheat yield, early detection and accurate control measures are necessary. However, the current identification methods are labour intensive and error-prone due to their reliance on manual visual observations. As a result, there have been an upsurge in demand for automated and high precision spatial detection of yellow rust in the wheat plant. In this study, we propose a novel approach based on advanced image processing algorithms. Our workflow, which utilizes convolutional neural networks, residual networks, and MobileNet to assess wheat plant multimedia and then determine the presence of yellow rust, is described.

Highly accurate models with reliably high detection rates have been the critical limitation in the field. Our submission can be positioned as a contribution to overcoming this limitation as we developed and evaluated several models with different architectures and detection rates. Our CNN-based model shows unrivalled accuracy of 0.99, which means that the accuracy level is close to perfection. ResNet boasts an accuracy level of 0.84 with the architecture depth, this suggests a relatively reliable detection. MobileNet architecture pursues efficiency and scalability rather than depth; due to maximum accuracy of 1.0, we believe this model’s deployment can be effective in low-resource settings. Overall, automated image processing models with convolutional architectures appear to be a viable solution for detecting yellow rust in wheat images. Creating models that achieve a high level of accuracy can revolutionize disease monitoring in agriculture, detecting the problem when it arises and eliminating the need for regular manual inspection.