Comparison of SVM and RF Algorithms for Crop Mapping Using Bi-Temporal Optical and Radar Data with Limited Training Samples

Comparison of SVM and RF Algorithms for Crop Mapping Using Bi-Temporal Optical and Radar Data with Limited Training Samples

This paper aims to compare two state-of-the-art classification algorithms, namely Support Vector Machine (SVM) and Random Forest (RF) algorithms, in terms of accuracy and running time, in order to crop mapping from multi-temporal optical and radar images with limited training samples. The optical data are RapidEye images and the radar data are UAVSAR images. The case study is an agricultural area near Winnipeg, Manitoba, Canada. From each RapidEye image, 38 optical features, and from each UAVSAR image, 49 radar features were extracted. The results indicated RF was more efficient in the classification of radar features, while SVM was more efficient in the classification of optical and stacked features. Furthermore, regarding running time, RF was much faster than SVM in all scenarios.