Deep Learning-Based Segmentation of Peach Diseases Using Convolutional Neural Network.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Peach diseases impact crop yield and human health.
• Accurate disease identification and lesion segmentation are crucial for effective management.
• Complex backgrounds and imbalanced data pose challenges for current segmentation methods.

Purpose of the Study:

• To apply and evaluate deep learning instance segmentation models for peach disease identification and segmentation.
• To address challenges of complex backgrounds and imbalanced samples in peach disease segmentation.
• To improve the accuracy and detail of peach disease recognition and segmentation.

Main Methods:

• Instance segmentation models Mask R-CNN and Mask Scoring R-CNN were utilized.
• The Focal Loss function was incorporated to handle difficult and imbalanced samples.
• Experiments were conducted using ResNet50 and ResNet101 as backbone networks.

Main Results:

• Mask Scoring R-CNN with Focal Loss outperformed standard Mask R-CNN and Mask R-CNN with CE loss.
• Segmentation accuracy (segm_mAP_50) improved from 0.236 to 0.254 with ResNet50.
• Segmentation accuracy (segm_mAP_50) increased from 0.452 to 0.463 with ResNet101.

Conclusions:

• The integration of Focal Loss with Mask R-CNN and Mask Scoring R-CNN enhances peach disease segmentation accuracy.
• Instance segmentation models provide detailed information including disease names, locations, and segmented areas.
• This approach offers a robust solution for precise peach disease analysis and management.