A novel method for maize leaf disease classification using the RGB-D post-segmentation image data.

Area of Science:

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

Background:

• Maize (Zea mays L.) is a critical global crop facing significant disease challenges.
• Traditional disease identification methods struggle with efficiency and accuracy in field conditions.
• Accurate disease phenotyping is crucial for maize germplasm resource management.

Purpose of the Study:

• To evaluate the potential of multi-sensor synchronized RGB-D cameras for maize leaf disease classification.
• To develop and compare deep learning models for classifying key maize leaf diseases.
• To assess the effectiveness of pre-segmentation versus post-segmentation image analysis for disease detection.

Main Methods:

• Utilized RGB-D camera depth information to segment maize leaves from complex backgrounds.
• Employed four deep learning models (Resnet50, MobilenetV2, Vgg16, Efficientnet-B3) for classification.
• Classified three major maize diseases: curvularia leaf spot, small spot, and mixed spot diseases.

Main Results:

• Post-segmentation models provided more practical disease classification results with shorter prediction times compared to pre-segmentation.
• Resnet50 and MobilenetV2 demonstrated superior accuracy among post-segmentation models.
• MobilenetV2 exhibited the best performance in terms of model size and single-image prediction time.

Conclusions:

• Post-segmentation analysis using RGB-D camera data offers a robust method for maize leaf disease classification.
• Deep learning models, particularly MobilenetV2, can efficiently identify maize leaf diseases in field settings.
• This approach facilitates the development of portable devices for practical agricultural disease management.