Investigating the effect of radar images in classifying land use classes in machine learning based algorithms

Acquiring knowledge about the types of land uses and the stages of their change provides basic and very important information to researchers and decision makers. One of the most common and useful methods in remote sensing is to access the maximum information contained in satellite data by combining radar and optical satellite images. In general, the main purpose of this study was to investigate the effect of the presence of SAR images in the classification of optical multi-temporal satellite images in machine learning-based classification algorithms, including random forest, Cart Decision Tree and Support Vector Machine. In the above paper, the Normalized Difference Vegetation Index (NDVI) dataset, along with slope layers, a digital elevation model and a corrected Sentinel-2 satellite image was supervised by the three methods mentioned. Once again, this was done with the presence of the Sentinel-1 satellite SAR image database. Finally, in the post-processing stage, the individual pixels were connected to neighboring classes. This was done by majority filtering. The final results were validated with ground data. The results showed that in the study of all classes, the overall accuracy and kappa coefficient in the presence of SAR dataset and for all three classification methods improved by only 3%, but in the one-to-one study of the classes, the producer accuracy of the random forest method in the dual agriculture class improved. It has been significant and its value has increased from 0.74 to 0.84. In the support vector machine method, dry farming and orchard classes have had a more significant improvement, which have increased from 0.75 and 0.78 to 0.84 and 0.92, respectively. Finally, it can be said that the addition of radar images to the classification has a positive and significant effect only in the mentioned classes, and also the obvious advantage of the random forest method compared to other methods is quite obvious.