Sooty Mold Detection on Citrus Tree Canopy Using Deep Learning Algorithms

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Sooty mold is a common citrus disease hindering photosynthesis.
• Early detection of sooty mold, especially on small leaves, is challenging.
• Existing methods often rely on expensive imaging equipment.

Purpose of the Study:

• To develop a cost-effective method for detecting citrus sooty mold at the canopy level.
• To evaluate the performance of deep learning models using affordable surveillance cameras.
• To investigate nighttime image capture for overcoming lighting challenges.

Main Methods:

• Utilized a low-cost home surveillance camera with night vision for image acquisition.
• Employed deep learning models (YOLOv5m, YOLOv7, CenterNet) for sooty mold detection.
• Trained and tested models on a dataset of 4200 sliced night-captured images.

Main Results:

• YOLOv7 demonstrated the highest detection accuracy (74.4% mAP) for sooty mold at night.
• Performance decreased significantly with day-captured images.
• Nighttime detection using affordable cameras proved effective.

Conclusions:

• A cost-effective surveillance camera combined with deep learning (YOLOv7) can accurately detect citrus sooty mold at night.
• This approach offers a promising solution for real-time orchard disease monitoring.
• Enables growers to effectively identify and manage sooty mold at the canopy level.