The NWRD Dataset: An Open-Source Annotated Segmentation Dataset of Diseased Wheat Crop

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Wheat stripe rust disease (WRD) poses a significant threat to global food security by reducing crop yield.
• Manual inspection of WRD is inefficient and labor-intensive for large-scale wheat fields.
• Data scarcity for specific crop diseases hinders the development of effective Artificial Intelligence (AI) and Deep Learning (DL) models.

Purpose of the Study:

• To address the challenge of data scarcity for WRD detection.
• To develop a semantic segmentation dataset for precise identification of diseased areas in wheat crops.
• To enable targeted treatment and promote sustainable agriculture by limiting fungicide application.

Main Methods:

• Introduction of the NUST Wheat Rust Disease (NWRD) dataset, featuring manually annotated multileaf images.
• Collection and preprocessing of images under diverse illumination and background conditions.
• Application of the UNet semantic segmentation model combined with an adaptive patching with feedback (APF) technique.

Main Results:

• The NWRD dataset facilitates semantic segmentation for estimating WRD spread.
• The UNet model with APF achieved a precision of 0.506, recall of 0.624, and F1 score of 0.557 for the rust class.
• Demonstrated the feasibility of AI-driven semantic segmentation for targeted disease management.

Conclusions:

• The NWRD dataset is a valuable resource for advancing AI-based crop disease detection.
• The UNet and APF approach shows potential for accurate and efficient wheat rust disease segmentation.
• This research supports environmentally friendly farming by enabling precise application of treatments.