Advanced deep transfer learning techniques for efficient detection of cotton plant diseases

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Cotton is a vital global cash crop, threatened by diseases impacting yield and quality.
• Accurate disease identification is crucial for effective management and sustainable agriculture.
• Image recognition and deep learning offer powerful tools for early crop disease detection.

Purpose of the Study:

• To investigate deep transfer learning models for cotton plant disease detection.
• To evaluate the performance of various models including EfficientNet, Xception, ResNet, Inception, VGG, DenseNet, MobileNet, and InceptionResNet.
• To identify the most effective model for accurate and efficient cotton disease diagnosis.

Main Methods:

• Utilized a dataset of healthy and diseased cotton plant images (Bacterial Blight, Target Spot, Powdery Mildew, Aphids, Army Worm).
• Applied image pre-processing and feature extraction techniques (contour generation, point identification, cropping, adaptive thresholding).
• Evaluated multiple deep transfer learning models for disease classification.

Main Results:

• EfficientNetB3 demonstrated superior performance with 99.96% accuracy, 0.149 loss, and 0.386 RMSE.
• Other models (excluding VGG19) showed high precision, recall, and F1 scores (around 0.98-1.00).
• Deep transfer learning proved effective for cotton disease diagnosis.

Conclusions:

• Deep transfer learning offers a cost-effective and efficient solution for cotton crop disease monitoring.
• This approach supports sustainable farming practices and enhances crop output.
• The study highlights the potential of AI in improving agricultural management and disease control.