Optimized encoder-decoder cascaded deep convolutional network for leaf disease image segmentation.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Deep Learning (DL) automates plant disease identification, crucial for food security.
• Deep Encoder-Decoder Cascaded Network (DEDCNet) precisely segments diseased leaf areas.
• DEDCNet model training requires careful hyperparameter selection for robustness.

Purpose of the Study:

• To propose an Optimized DEDCNet (ODEDCNet) model for enhanced leaf disease image segmentation.
• To introduce the Dingo Optimization Algorithm (DOA) for optimal DEDCNet hyperparameter selection.
• To improve the accuracy and efficiency of automated plant disease diagnosis.

Main Methods:

• Developed ODEDCNet integrating Dingo Optimization Algorithm (DOA) for hyperparameter tuning.
• DOA mimics dingo foraging behavior for efficient exploration and exploitation in search space.
• Trained DEDCNet with optimized hyperparameters for segmentation, followed by CNN and SVM for classification.

Main Results:

• ODEDCNet achieved high accuracy: 97.33% on PlantVillage and 97.42% on Betel Leaf datasets.
• Excellent performance metrics including recall, F-score, Dice coefficient, and precision were obtained.
• Achieved rapid processing times of 0.07 and 0.06 seconds for the respective datasets.

Conclusions:

• The proposed ODEDCNet model with DOA demonstrates superior performance in leaf disease segmentation and classification.
• DOA effectively optimizes DEDCNet hyperparameters, leading to improved accuracy and robustness.
• This automated approach significantly contributes to early plant disease detection and global food security.