Automatic detection and delineation of citrus trees from VHR satellite imagery

Abstract

In this article, we present a novel approach to detecting and delineating individual citrus trees through very-high resolution (VHR) GeoEye-1 satellite images at two different test sites. The approach is based on vegetation extraction, fast radial symmetry (FRS) transform, and simple object-based hierarchical operations. Our basic assumption is that each citrus tree presents a symmetric feature in the image. Multiple parameter combinations were tested to determine the optimum parameter set. The results calculated with the combination of optimum parameters were then evaluated based on both pixel- and object-based approaches. Promising results (up to 90% accuracy) were obtained for both detection and delineation rates, especially in areas with regular planting patterns and minimum tree crown overlap. The results indicate that object-based evaluation improves the accuracy at certain detection and delineation rates.

Acknowledgement

The author would like to thank the anonymous reviewers for their valuable comments and constructive suggestions on this article.

Additional information

Funding

This study was supported by the Scientific Research Council of Turkey [Project Number 114Y671]. The author would like to thank Hürcan ATAY from ATAY Mühendislik for his support during this project.