Grape Leaf Black Rot Detection Based on Super-Resolution Image Enhancement and Deep Learning.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Accurate detection of grape leaf diseases is crucial for crop management.
• Small disease spots (<32x32 pixels) pose significant challenges for current deep learning detection methods due to fuzzy convolutional substrate information.
• Existing methods struggle with efficiency and accuracy, particularly for small disease lesions.

Purpose of the Study:

• To develop a robust algorithm for enhancing super-resolution images of grape leaves to improve the detection of black rot disease spots.
• To improve the accuracy and efficiency of detecting small black rot lesions on grape leaves using deep learning.
• To adapt and enhance the YOLOv3 network for better performance in identifying grape leaf diseases.

Main Methods:

• Employed bilinear interpolation for super-resolution image enhancement, increasing pixel count and local details.
• Introduced a modified YOLOv3 network incorporating the Spatial Pyramid Pooling (SPP) module and replacing Intersection over Union (IOU) with Generalized Intersection over Union (GIOU).
• Utilized pre-trained YOLOv3 weights for rapid model convergence and evaluated performance on Plant Village and orchard field datasets.

Main Results:

• The YOLOv3-SPP model achieved 95.79% detection accuracy and 94.52% recall on the Plant Village test set, outperforming original YOLOv3 by 5.94% and 10.67% respectively.
• In orchard field tests, the method demonstrated 86.69% precision and 82.27% recall, with performance improving to 94.05% accuracy and 93.26% recall for images with simple backgrounds.
• The enhanced detection method effectively addresses the challenge of detecting small disease targets, significantly improving detection effectiveness for grape leaf black rot.

Conclusions:

• The proposed super-resolution enhancement and YOLOv3-SPP algorithm provide an effective solution for detecting small black rot lesions on grape leaves.
• The integration of SPP and GIOU modules enhances the network's ability to handle variations in object scale and improves detection robustness.
• This approach shows significant potential for practical application in agricultural environments for disease monitoring and management.