Deep transfer learning for comprehensive diagnosis of cotton leaf pathologies

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Traditional methods for cotton leaf disease identification are often slow and inaccurate.
• The cotton sector faces significant challenges due to diseases like cotton leaf blight.
• Automated disease detection is crucial for enhancing crop yields and farm productivity.

Purpose of the Study:

• To develop an automated system for diagnosing cotton leaf blast disease using deep learning and image processing.
• To evaluate the performance of various deep learning models for cotton disease identification.
• To enhance precision agriculture through reliable and accurate disease prediction tools.

Main Methods:

• Utilized a dataset of over 4200 cotton leaf images, including healthy and diseased samples.
• Implemented and compared deep learning architectures: Convolutional Neural Network (CNN), InceptionV3, ResNet50, VGG16, VGG19, and Xception.
• Assessed model performance based on validation accuracy, complexity, and inference speed.

Main Results:

• The VGG19 model achieved the second-highest validation accuracy at 95.97%.
• ResNet50 demonstrated high accuracy, outperforming previous models.
• The developed deep learning models significantly surpassed existing methodologies in accuracy and speed.

Conclusions:

• Deep learning offers a precise and efficient automated approach for diagnosing cotton leaf diseases.
• This technology can equip farmers with tools for early disease detection, mitigating crop damage.
• Further research incorporating advanced models like Vision Transformers aims to improve diagnostic accuracy and efficacy.