Image-Based Wheat Fungi Diseases Identification by Deep Learning

Area of Science:

• Agricultural Science
• Plant Pathology
• Computer Vision

Background:

• Fungal diseases in cereals significantly impact crop yields globally.
• Early detection and management of plant diseases are crucial for agricultural productivity.
• Mobile-based image analysis offers a practical solution for field disease monitoring.

Purpose of the Study:

• To develop and validate a method for recognizing five key fungal diseases in wheat shoots.
• To enable identification of single and multiple disease infections, including plant developmental stages.
• To create a user-friendly tool for field-based plant disease assessment.

Main Methods:

• Generation of a labeled dataset (WFD2020) comprising 2414 wheat fungal disease images.
• Application of an image hashing algorithm to reduce training data degeneracy.
• Utilizing a convolutional neural network with the EfficientNet architecture for disease recognition.
• Implementation of a Telegram bot for real-time plant disease assessment.

Main Results:

• The developed algorithm achieved a high accuracy of 0.942 in disease recognition.
• The best performance was obtained using a training strategy combining data augmentation and style transfer.
• The system can distinguish between healthy plants, single, and multiple disease infections.

Conclusions:

• The proposed convolutional neural network-based method effectively identifies fungal diseases in wheat.
• The Telegram bot provides a practical tool for farmers to monitor crop health in field conditions.
• This approach supports timely interventions to mitigate yield losses caused by fungal pathogens.