Improved early-stage crop classification using a novel fusion-based machine learning approach with Sentinel-2A and...

Area of Science:

• Remote Sensing
• Agricultural Science
• Machine Learning

Background:

• Early-stage crop classification is difficult due to similar spectral and texture features among crop types.
• Accurate crop identification is crucial for effective agricultural management and monitoring.

Purpose of the Study:

• To develop and evaluate a novel fusion-based deep learning approach for enhanced early-stage crop classification.
• To improve the accuracy and reliability of crop identification using integrated remote sensing data.

Main Methods:

• A fusion-based deep learning approach was proposed, integrating textural and spectral features.
• Landsat 8-9 and Sentinel-2A data were merged using the Gram-Schmidt approach.
• Textural features were extracted using multi-patch Gray Level Co-occurrence Matrix (GLCM), and spectral features (EVI, NDVI) were obtained via spectral indices.
• Five machine learning models, including deep neural network (DNN), were trained and evaluated.

Main Results:

• The proposed fusion-based deep learning approach demonstrated high performance.
• Deep Neural Network (DNN) achieved an accuracy of 0.89, precision of 0.88, recall of 0.91, and F1-score of 0.90.
• The integration of fused spectral and textural data significantly improved classification outcomes.

Conclusions:

• The fusion-based deep learning approach is effective for enhancing early-stage crop classification accuracy.
• Combining diverse data sources and advanced machine learning techniques offers a promising solution for agricultural remote sensing challenges.