Wheat Spike Blast Image Classification Using Deep Convolutional Neural Networks

Area of Science:

• Plant Pathology
• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Wheat blast poses a significant threat to global wheat production, with a scarcity of resistant cultivars.
• Current methods for assessing wheat spike blast severity rely on human evaluation, which can be subjective and limited.
• Accurate and efficient disease severity assessment is crucial for breeding programs and disease management.

Purpose of the Study:

• To develop and evaluate deep convolutional neural networks (CNNs) for classifying wheat spike blast disease severity (DS).
• To assess the reliability of human assessments for wheat blast DS classification using inter-rater agreement analysis.
• To explore the potential of image-based AI for facilitating future wheat spike blast phenotyping.

Main Methods:

• Collected Red Green Blue (RGB) images of wheat spikes under controlled conditions.
• Conducted inter-rater agreement analysis to ensure reliability of human-collected and classified data.
• Trained CNN models using the image data to classify wheat blast severity into three categories.

Main Results:

• Inter-rater agreement analysis confirmed high accuracy and low bias in human data collection and classification.
• Trained CNN models demonstrated a promising approach for classifying wheat spike blast severity from images.
• Models trained on both non-matured and matured spikes achieved the highest precision, recall, and F1 scores.

Conclusions:

• Deep convolutional neural networks provide an accurate and reliable method for classifying wheat spike blast severity.
• AI-driven image analysis can overcome limitations of manual disease assessment in wheat.
• This approach offers a strong foundation for future advancements in high-throughput wheat blast phenotyping.