Paper Title
T-Sne Color and Sfta Texture Features For Aerial Images Palm Oil Plantations Area Classification

Abstract
Indonesia has massive palm oil plantation area. Manual evaluation of the palm oil area would take a lot of effort and time. For that reason, an automated palm oil evaluation is proposed. This paper presents an early study and its results to automated palm oil evaluation. First, an aerial image of the palm oil plantation area is obtained using unmanned aerial vehicle (UAV). The images are then stitched altogether to obtain a single image with much greater area of coverage. Next, the image is divided into several patches, and features are extracted for each patch. This paper used newly joined features to represent a color and textures of the image. The texture features are extracted using Segmentation-based Fractal Texture Analysis (SFTA). The color features are extracted using T-SNE dimensional reduction in the RGB planes. The AdaBoost classifier is adopted to determine the class of these patches and classify it into three categories, i.e. palm oil plantation, non-palm oil plantation, and non-plantation area. Individual and join features performance is tested. The result shows that texture feature alone gives 94.23% classification accuracy, whereas color feature alone yields 45.06%. The join features improves the classification accuracy by 0.85% to give 95.08%. Index Terms- palm oil, plantation, UAV, image processing, SFTA, T-SNE, AdaBoost