Tree Crowns Segmentation and Classification in Overlapping Orchards Based on Satellite Images and Unsupervised...

Area of Science:

• Agricultural Engineering
• Remote Sensing
• Computer Science

Background:

• Smart agriculture integrates technology to enhance crop yield and management.
• Automated orchard monitoring relies on accurate tree crown segmentation.
• Overlapping tree canopies pose a significant challenge for precise delineation.

Purpose of the Study:

• To develop and validate a machine learning approach for segmenting and classifying trees in overlapping orchards using satellite imagery.
• To improve automated monitoring and health assessment of individual trees within commercial orchards.
• To provide farmers with tools for understanding tree distribution and health status.

Main Methods:

• Utilized satellite images from the Mohammed VI satellite for the OUARGHA citrus orchard in Morocco.
• Employed machine learning algorithms for row segmentation, tree detection, and crown segmentation.
• Classified trees into 'missing/weak', 'normal', or 'big' based on canopy size and field measurements.

Main Results:

• Achieved a 0.93 F-measure score for row segmentation accuracy.
• Validated tree classification through field comparisons, demonstrating high accuracy.
• Successfully visualized tree distribution and health status on a map.

Conclusions:

• The developed method effectively segments and classifies trees in challenging overlapping orchard environments.
• This approach facilitates automated, rapid tree health monitoring and distribution analysis for farmers.
• Enhances smart agriculture capabilities by providing detailed orchard insights.