Attention-optimized DeepLab V3 + for automatic estimation of cucumber disease severity

Area of Science:

• Agricultural science
• Computer vision
• Plant pathology

Background:

• Accurate plant disease severity estimation is crucial for crop management and predicting yield loss.
• Traditional methods struggle with complex field conditions like variable sunlight and backgrounds.
• A robust method is needed for disease severity estimation using images taken in real-world agricultural settings.

Purpose of the Study:

• To develop an efficient and accurate image-based method for estimating plant disease severity under field conditions.
• To create an optimized neural network model capable of handling complex image backgrounds and lighting variations.
• To validate the proposed method's performance in segmenting diseased plant leaf areas and quantifying severity.

Main Methods:

• Developed a semantic segmentation model integrating hybrid attention (spatial and channel) and transfer learning.
• Calculated disease severity as the ratio of lesion pixels to total leaf pixels.
• Validated the model on cucumber leaves with downy mildew and powdery mildew under natural conditions.

Main Results:

• The hybrid attention mechanism effectively extracted detailed features of lesions and leaves.
• Transfer learning enhanced the model's segmentation accuracy.
• The model achieved high segmentation performance (MIoU=81.23%, FWIoU=91.89%) and accurate severity estimation (R²=0.9578).
• Outperformed existing models in complex field scenarios.

Conclusions:

• The developed AI tool efficiently estimates plant disease severity in field conditions.
• This research supports the application of artificial intelligence for rapid disease assessment and control in agriculture.