Towards high throughput in-field detection and quantification of wheat foliar diseases with deep learning
Reliable, quantitative information on the presence and severity of crop diseases is critical for site-specific crop management and resistance breeding. Successful analysis of leaves under naturally variable lighting, presenting multiple disorders, and across phenological stages is a critical step towards high-throughput disease assessments directly in the field.Here, we demonstrate the capability of deep learning based keypoint detection for STB pycndia and rust pustules combined with semantic segmentation for leaves, leaf necrosis and insect damage to reliably detect and quantify the presence of Septoria Tritici Blotch, leaf rusts, and insect damage under such conditions. In addition we present the Eschikon Wheat Foliar Disease (EFD) dataset with 418 high resolution images with polygon annotations of leaves, leaf necrosis and insect damage and point annotations of STB pycnidia and rust pustules. With our dataset a solid benchmark is established, but the potential of the proposed method remains under-exploited as significantly more data containing more diverse symptoms can be utilized to further improve the performance and introduce new disorders. Finally, we demonstrated the robustness of the approach by evaluating images of an unstructured canopy. This underlines the potential for extending this work to make more efficient in-field analysis without the need to detach leaves and thus moving towards automated in-field assessment of foliar diseases.