The Detection Method of Potato Foliage Diseases in Complex Background Based on Instance Segmentation and Semantic...

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Potato early and late blights significantly impact crop yield.
• Accurate disease diagnosis is crucial for effective potato farm management.
• Traditional methods face limitations in detecting crop diseases.

Purpose of the Study:

• To develop an integrated deep learning framework for segmenting and detecting potato foliage diseases.
• To address limitations of traditional computer vision in complex backgrounds.
• To provide a robust solution for rapid and accurate crop disease identification.

Main Methods:

• An integrated framework combining instance segmentation (Mask R-CNN), classification (VGG16, ResNet50, InceptionV3), and semantic segmentation (UNet, PSPNet, DeepLabV3+) was devised.
• Mask R-CNN segmented potato leaves in complex backgrounds.
• Classification models identified leaf diseases, while semantic segmentation models delineated affected areas.

Main Results:

• The Mask R-CNN instance segmentation achieved 81.87% average precision and 97.13% precision.
• The classification models demonstrated 95.33% accuracy.
• Semantic segmentation models achieved 89.91% mean intersection over union (MIoU) and 94.24% mean pixel accuracy (MPA).

Conclusions:

• The developed three-stage deep learning model effectively segments and detects potato foliage diseases in natural settings.
• This framework offers a new approach for potato disease assessment and classification.
• It lays a theoretical foundation for advanced agricultural disease management systems.