An Innovative Fusion-Based Scenario for Improving Land Crop Mapping Accuracy

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Agricultural Science

Background:

• Accurate land crop mapping from satellite imagery is crucial for agricultural monitoring.
• The performance of crop classification is influenced by feature selection algorithms and classifiers.
• Combining diverse data sources and methods can potentially improve mapping accuracy.

Purpose of the Study:

• To develop and evaluate a fusion-based framework for enhancing land crop mapping accuracy.
• To investigate the effectiveness of combining features from Sentinel 1, Sentinel 2, and Landsat-8 imagery.
• To compare different feature selection algorithms and classifiers for optimal land crop classification.

Main Methods:

• Feature extraction from Sentinel 1, Sentinel 2, and Landsat-8 satellite imagery.
• Application of feature selection algorithms: Recursive Feature Elimination (RFE), Random Forest (RF), and Boruta.
• Implementation of classifiers: Artificial Neural Network (ANN), Support Vector Machine (SVM), and RF.
• Fusion of classification results using a decision-level voting strategy to create final crop maps and uncertainty maps.

Main Results:

• Random Forest (RF) demonstrated higher accuracy than RFE and Boruta for feature selection.
• RF classifier outperformed SVM and ANN classifiers in land crop classification.
• The fusion-based voting scenario achieved higher overall accuracy compared to individual scenarios.

Conclusions:

• Combining features from multiple sensors and employing diverse feature selection algorithms and classifiers improves land crop classification accuracy.
• A decision-level fusion strategy, particularly voting, is effective for enhancing the reliability of land crop maps.
• The developed framework offers a robust approach for accurate and reliable land crop mapping.