DBA-DeepLab: Dual-Backbone Attention-Enhanced DeepLab V3+ Model for Plant Disease Segmentation

Area of Science:

• Computer Vision
• Agricultural Science
• Machine Learning

Background:

• Accurate plant disease segmentation is vital for early detection and effective management in agriculture.
• Existing segmentation models often struggle with complex disease patterns and background noise.

Purpose of the Study:

• To develop an advanced deep learning model for precise plant disease segmentation.
• To enhance the accuracy and efficiency of automated disease identification in crops.

Main Methods:

• Proposed a Dual-Backbone Attention-Enhanced DeepLab (DBA-DeepLab) model integrating ResNet-50 and EfficientNet-B3 backbones with a Convolutional Block Attention Module (CBAM).
• Incorporated multi-scale feature extraction, attention mechanisms, and Sobel filtering for edge preservation.
• Trained and validated the model on the PlantDoc dataset using the Adam optimizer over 50 epochs.

Main Results:

• DBA-DeepLab achieved superior segmentation performance compared to standard DeepLabV3+ variants.
• The model demonstrated high accuracy (99.35%), Dice coefficient (91.48%), IoU coefficient (85.85%), precision (96.78%), and recall (100%).
• Grad-CAM visualization confirmed the model's ability to focus on disease-affected regions and reduce background noise.

Conclusions:

• The DBA-DeepLab model offers a highly accurate, efficient, and scalable solution for plant disease segmentation.
• The model shows significant potential for applications in smart farming, automatic disease detection, and precision agriculture.
• The attention mechanism and dual-backbone architecture contribute to improved segmentation accuracy and robustness.