Construction of deep learning-based disease detection model in plants

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Early crop disease detection is crucial for maintaining crop quality and yield.
• Traditional disease identification requires specialized expertise and extensive experience.
• Automated systems can significantly enhance early disease detection capabilities in agriculture.

Purpose of the Study:

• To develop an automated, stepwise disease detection model for crops using image analysis.
• To improve the accuracy and generalizability of crop disease identification.
• To create a foundation for smart farming applications through enhanced disease detection.

Main Methods:

• A convolutional neural network (CNN) algorithm was employed, integrating five pre-trained models.
• A stepwise classification model was constructed, involving crop classification, disease detection, and disease classification.
• The category 'unknown' was incorporated to enhance model generalizability for diverse applications.

Main Results:

• The developed disease detection model achieved high accuracy (97.09%) in classifying crops and disease types.
• The model demonstrated expendability, with accuracy improving for previously unmodeled crops upon their inclusion in the training dataset.
• The system shows significant potential for application in smart farming, particularly for Solanaceae crops.

Conclusions:

• The automated crop disease detection system offers a viable solution for early-stage disease identification.
• The model's accuracy and generalizability can be further improved by expanding the training dataset with more crop varieties.
• This technology has the potential to revolutionize crop management practices in smart farming environments.