A Comparative Analysis of Machine Learning with WorldView-2 Pan-Sharpened Imagery for Tea Crop Mapping

Area of Science:

• Agricultural remote sensing
• Geospatial analysis
• Environmental monitoring

Background:

• Tea is a vital East Asian crop vulnerable to climate change.
• Accurate land use/land cover (LULC) mapping is crucial for tea cultivation management.
• High-resolution satellite imagery offers potential for detailed crop assessment.

Purpose of the Study:

• To interpret tea land use/land cover (LULC) using WorldView-2 imagery in central Taiwan.
• To compare pixel-based image analysis (PBIA) and object-based image analysis (OBIA) for tea crop classification.
• To develop a framework for accurate, field-survey-free tea crop identification.

Main Methods:

• Utilized very high resolution WorldView-2 imagery.
• Extracted 80 variables including spectral bands, principal components, and GLCM texture indices.
• Employed pixel-based (SVM, RF, ML, LR) and object-based classification approaches.

Main Results:

• Pixel-based image analysis (PBIA) using Support Vector Machine (SVM) achieved 84.70% accuracy.
• Object-based image analysis (OBIA) with Maximum Likelihood (ML) classifier yielded superior accuracy (96.04%) with fewer variables.
• OBIA demonstrated higher efficiency and accuracy for tea crop classification.

Conclusions:

• Object-based image analysis (OBIA) is more effective than PBIA for classifying tea crops using WorldView-2 data.
• This study provides a novel framework for real-time tea crop identification in subtropical regions.
• The findings support improved agricultural land management and sustainable agricultural product supply.