The delineation of tea gardens from high resolution digital orthoimages using mean-shift and supervised machine learning methods

Abstract

Rize district is an important tea production site in Turkey, which is known for high quality tea. Determining the temporal changes is very crucial from the viewpoint of agricultural management and protection of tea areas. In addition, delineation of tea gardens using photogrammetric evaluation techniques for a single orthoimage takes approximately 8 h of labour work, which is both costly and time-consuming process. To overcome these issues, a method is proposed for demarcation of tea gardens from high-resolution orthoimages. In this article, a hierarchical object-based segmentation using mean-shift (MS) and supervised machine learning (ML) methods are investigated for delineation of tea gardens. First, the MS algorithm was applied to partition the images into homogeneous segments (objects) and then from each segment, various spectral, spatial and textural features were extracted. Finally, four most widely used supervised ML classifiers, support vector machine (SVM), artificial neural network (ANN), random forest (RF), and decision trees (DTs), were selected for classification of objects into tea gardens and other types of trees. Photogrammetrically evaluated tea garden borders were taken as reference data to evaluate the performance of the proposed methods. The experiments showed that all selected supervised classifiers were effective for delineation of the tea gardens from high-resolution images.

Keywords:

• Tea garden extraction
• mean-shift segmentation
• support vector machine
• artificial neural network
• decision trees
• random forest

Acknowledgments

We are thankful to the EMI Group Turkey for providing the important data for this study which is a part of TEYDEP Project entitled ‘Development of Object Based Neural Network Image Processing System Determination of Vegetation and Forestry Boundaries’ (Project Nr. 7140512). It was supervised by EMI Group-Turkey, and consulted by Prof. Dr. Bulent Bayram.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This study was funded by TEYDEP Project entitled ‘Development of Object Based Neural Network Image Processing System Determination of Vegetation and Forestry Boundaries’ (Project Nr. 7140512).