A modern deep learning framework in robot vision for automated bean leaves diseases detection

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Bean crops are susceptible to diseases like angular leaf spot and bean rust, significantly reducing yield.
• Early disease detection and treatment are crucial for maintaining crop quality and productivity.
• Automated systems using AI and image processing show promise but face challenges with accurate diagnosis, leading to ineffective treatments.

Purpose of the Study:

• To propose a novel deep learning framework for early and accurate detection of bean leaf diseases using robot vision.
• To develop a two-stage system for detecting bean leaves and diagnosing diseases under uncontrolled environmental conditions.
• To evaluate the performance of various deep learning models for classifying bean leaf health.

Main Methods:

• A U-Net architecture with a pre-trained ResNet34 encoder was used for bean leaf detection in images.
• Five deep learning models (Densenet121, ResNet34, ResNet50, VGG-16, VGG-19) were assessed for disease classification.
• A dataset of 1295 images featuring healthy, angular leaf spot, and bean rust classes was utilized for evaluation.

Main Results:

• The Densenet121 model achieved high performance in binary classification, with a Classification Accuracy Rate (CAR) of 98.31% and AUC of 100%.
• In multi-classification tasks, Densenet121 achieved a CAR of 91.01%, processing each image in under 2 seconds.
• The framework demonstrated robust performance in identifying bean leaf health and diseases.

Conclusions:

• The proposed deep learning framework offers an effective solution for early and accurate detection of bean leaf diseases.
• The system aids in precise disease management, potentially reducing unnecessary chemical applications and improving crop yields.
• This AI-driven approach enhances the efficiency and reliability of automated crop monitoring systems.