Olive Tree Classification and Inventory with Medium Resolution Multi-spectral Satellite Imagery

Abstract

Remote sensing is the most widely useful tool for land use planning and decision support system. It gives the accurate information of agricultural activities such as crop identification and classification, crop area, crop growth condition monitoring and yield estimation in a concise and recurring manner for large areas and over long periods of time. This chapter presents a two-step approach for the inventory of olive tree plantations using medium-resolution optical satellite image. First, a multi-scale Conditional Random Fields (CRF) is applied for olive plantations labeling. Second, an algorithm based on a combination of three procedures, mean-sift segmentation, spectral thresholding of Red Band and NDVI and spatial thresholding is carried out to detect and count olive trees. The results of the approach are evaluated on four test sites situated in the Sbikha region, Kairouan-Tunisia and the accuracies are analyzed. Experimental results show that the proposed classification and detection schemes achieve good performances with a mean total accuracy about 87%.