Experimental Analysis for Classifying Mixed Classes Using Sentinel 1 and Sentinel 2 Data

In developing countries like India, the cropping pattern is highly heterogeneous, with similar crops grown in adjacent small patches called mixed classes. Classifying these mixed classes using single sensor data is challenging due to their random behavior. To overcome this challenge, a combination of optical and SAR data is used as they provides complementary information. By exploiting the absorption and reflectance characteristics of plant pigments across different optical bands, vegetation indices are derived, that characterize different plant biophysical properties and are more sensitive to canopy parameters than individual bands. Similarly, SAR backscattering coefficients and their derived parameters offer insights into the structural characteristics of the plants. Therefore, this study explored the potential of using multiple vegetation indices from Sentinel-1 and Sentinel-2 data to solve the mixed class classification problem. The Random Forest (RF) classifier, known for modeling complex relationships, is employed. Experimental results showed that combining vegetation indices with Sentinel-1 derived parameters significantly improved the classification accuracy of the mixed classes compared to relying solely on band values. This approach increased the accuracy from 87.2% to 96.5%.