Automated lesion detection in cotton leaf visuals using deep learning.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Cotton leaf diseases significantly impact global crop yields.
• Accurate and efficient detection of cotton diseases is crucial for agricultural economies.
• Challenges in automated detection include limited datasets, class imbalance, and varied lesion sizes.

Purpose of the Study:

• To develop a precise and scalable automated method for detecting cotton leaf diseases.
• To address challenges in automated disease detection, including class imbalance and limited data.
• To improve the accuracy and efficiency of identifying various cotton crop diseases.

Main Methods:

• Utilized a novel deep learning approach combining data augmentation with Generative Adversarial Networks (GANs).
• Employed an ensemble-based method integrating feature vectors from VGG16, Inception V3, and ResNet50 architectures.
• Implemented and evaluated the method on a public dataset encompassing seven disease and one healthy class.

Main Results:

• Achieved a highest accuracy of 95% in cotton leaf disease detection.
• Obtained an F1-score of 98%, demonstrating high precision and recall.
• The proposed method outperformed existing state-of-the-art techniques in automated disease identification.

Conclusions:

• The novel deep learning method effectively addresses class imbalance and enhances detection accuracy for cotton leaf diseases.
• The ensemble approach combining multiple deep learning architectures provides a robust solution for automated crop disease monitoring.
• This research offers a significant advancement in precision agriculture for improved cotton crop management and yield protection.