Comparative Study on Crop Type Classification Using Support Vector Machine on UAV Imagery

Abstract

In Indian agricultural practices, single crop cultivation is rare and uncommon. This poses a real challenge for crop type classification using single date imagery. Site-specific information of crop type is required for agricultural management which includes technologies aiming at productivity and profit while practicing eco-friendly environment. Unmanned Air Vehicles (UAV) are effective image acquisition platforms for many agricultural applications. UAV’s can acquire high levels of spatial details compared to standard remote sensing platforms. Single date RGB imagery of 5 cm spatial resolution obtained from processing the raw data was used for the classification of different types of crop. Traditional pixel-based analysis of remote sensing data results in inaccurate classification due to low spatial resolution, mixed pixels, and crop pattern variability. This can be overcome by using high-resolution UAV data and machine learning methods like Support Vector Machine (SVM). In the present study SVM kernel functions namely linear, sigmoid, radial basis and polynomial function are adopted and compared for mapping the crop types. The classification shows that the radial and sigmoid kernel functions give high accuracy when compared with the rest by performing the accuracy assessment for all four classifiers. These crop classifications are important for greenhouse gas modeling, agrarian policy, and agro-environmental studies.