Analysing the vegetation of energy plantsby processing UAV images

Abstract

Bioenergy plants are widely used as a form of renewable energy. It is important to monitor the vegetation and accurately estimate the yield before harvest in order to maximize the profit and reduce the costs of production. The automatic tracking of plant development by traditional methods is quite difficult and labor intensive. Nowadays, the application of Unmanned Aerial Vehicles (UAV) became more and more popular in precision agriculture. Detailed, precise, three-dimensional (3D) representations of energy forestry are required as a prior condition for an accurate assessment of crop growth. Using a small UAV equipped with a multispectral camera, we collected imagery of 1051 pictures of a study area in Kompolt, Hungary, then the Pix4D software was used to create a 3D model of the forest canopy. Remotely sensed data was processed with the aid of Pix4Dmapper to create the orthophotos and the digital surface model. The calculated Normalized Difference Vegetation Index (NDVI) values were also calculated. The aim of this case study was to do the first step towards yield estimation, and segment the created orthophoto, based on tree species. This is required, since different type of trees have different characteristics, thus, their yield calculations may differ. How-ever, the trees in the study area are versatile, there are also hybrids of the same species present. This paper presents the results of several segmentation algorithms, such as those that the widely used eCognition provides and other Matlab implementations of segmentation algorithms.