Enhancing Wheat Disease Diagnosis in a Greenhouse Using Image Deep Features and Parallel Feature Fusion.

Area of Science:

• Agricultural Science
• Plant Pathology
• Computer Vision

Background:

• Visual assessment of wheat diseases is subjective and inefficient.
• Automated, objective, and efficient diagnosis methods are needed for crop management.

Purpose of the Study:

• To develop an automated approach for diagnosing wheat diseases using image processing and deep learning.
• To compare the performance of handcrafted features (HFs) and deep features (DFs) for disease detection.
• To evaluate different deep learning models for wheat disease diagnosis.

Main Methods:

• Collected paired color and color-infrared (CIR) images of wheat plants.
• Developed a semiautomatic webtool for efficient dataset creation and image annotation.
• Extracted HFs and DFs (using models like ResNet101) from segmented images.
• Implemented parallel feature fusion for improved diagnostic accuracy.

Main Results:

• Parallel feature fusion of color and CIR images improved accuracy over single-source features.
• ResNet101 demonstrated the highest accuracy in extracting DFs for disease detection.
• Deep features coupled with parallel fusion achieved accuracies of 75% (leaf rust), 84% (tan spot), and 71% (combined).

Conclusions:

• The developed methodology offers an objective and efficient approach for wheat disease diagnosis.
• Deep learning models, especially ResNet101 with feature fusion, show significant potential for automated plant disease detection.
• The system is suitable for greenhouse applications and can be extended to field settings.