Combining a Standardized Growth Class Assessment, UAV Sensor Data, GIS Processing, and Machine Learning...

Area of Science:

• Agricultural Engineering
• Remote Sensing
• Machine Learning

Background:

• Accurate assessment of grapevine vigor is crucial for efficient vineyard management and the automation of viticulture machinery.
• Existing methods may lack the precision required for automated systems, necessitating advanced data-driven approaches.

Purpose of the Study:

• To develop and evaluate Machine Learning (ML) classifiers for predicting grapevine growth classes using multispectral Unmanned Aerial Vehicle (UAV) sensor data.
• To integrate spectral, structural, and texture features for enhanced classification accuracy in precision viticulture.

Main Methods:

• Utilized multispectral UAV data to generate spectral, structural (from Digital Terrain/Surface Models), and texture (GLCM) features.
• Employed a pixel- and object-based image segmentation technique to create vine row masks for extracting canopy-exclusive features.
• Trained Random Forest Classifier (RFC) and Support Vector Machines (SVM) models, optimizing hyperparameters with grid search and cross-validation.

Main Results:

• Machine learning models successfully predicted grapevine growth classes with high accuracy.
• Feature importance analysis indicated the significant contribution of spectral, structural, and texture data for vigor assessment.
• Optimized models demonstrated robust performance in classifying different growth stages.

Conclusions:

• ML-based analysis of multispectral UAV data provides a reliable method for assessing grapevine vigor.
• This approach supports precision viticulture by enabling automated and accurate vineyard management decisions.
• The methodology offers a scalable solution for monitoring vine health and optimizing agricultural practices.