A CNN-RF Hybrid Approach for Rice Paddy Fields Mapping in Indramayu using Sentinel-1 and Sentinel-2 Data

Rice, cultivated in paddy fields, is one of the staple foods in the world, especially in Asia. In Indonesia, a substantial number of rice paddy fields have been converted into residential or industrial areas, threatening food security. Therefore, it is necessary to monitor the adequacy of rice paddy field areas. Recently, remote sensing has become the most widely used method for mapping rice paddy fields. This research focuses on developing a classification model for rice paddy field mapping using remote sensing with radar and optical data fusion, including input variations in polarization, texture, and optical derivative indices. This study proposes the CNN-RF method, which combines a convolutional neural network (CNN) as a feature extractor and a random forest (RF) as a classifier. The experiment used combinations of input data, including variations of single and multisource data, to achieve optimal results. Research findings in some districts of Indramayu show that the scheme combining Sentinel-1 features with GLCM (gray-level co-occurrence matrix) and Sentinel-2 features with selected bands provides the best results, with an overall accuracy of 98.23% and a Kappa coefficient of 0.96, using the CNN-RF method. CNN-RF outperforms other classifiers owing to the hybrid learning combination, which improves the accuracy through feature extraction by CNN and handles complex relationships between features while reducing overfitting by RF. This study contributes to the development of accurate and efficient rice paddy fields mapping techniques using remote sensing.