Details

Autor(en) / Beteiligte

• Bagis, S.
• Ustundag, B. Berk

Titel

Image based automated phenological stage detection of cereal plants

Ist Teil von

• 2012 First International Conference on Agro- Geoinformatics (Agro-Geoinformatics), 2012, p.1-6

Ort / Verlag

IEEE

Erscheinungsjahr

2012

Link zum Volltext

• IEEE_Xplore

Quelle

IEEE/IET Electronic Library

Beschreibungen/Notizen

• Most of the agricultural loss and yield evaluation methods require separate plant parameters depending on phenological stages. In this study, an automated image processing based phenological stages detection system for cereal plants is explained. TARIT Project (www.tarit.org) in Turkey aims to predict crop yield and loss rates in crop lands over a vast region in South-Eastern Turkey. The image data is acquired in RGB format at 30 minutes period while the accompanied meteorological data is sampled at 10 minutes period. We construct a state machine for each field, and states represent the phonological stages. Each state has different transition function for the next stage related logical state. 7 major stages for wheat and barley are considered to determine transition dates although it could be increased in general. The sampled image is enhanced and normalized by using light rate and surface wetness related sensor data. Four different processes are then applied and their outcomes are used to determine the state transition conditions. The first is vegetation area green pixel ratio after RGB to HIS color space conversion. The second is change detection with respect to a virtual reference or recognized pattern for growth rate measurement. The fourth is probabilistic supporting stage transition factors based on agro-meteorological parameters. We set state transition functions depending on combination of these parameters with state dependent thresholds. It has shown that phenological stage transition dates of cereal plants can automatically be assigned less than a week tolerance with respect to expert diagnosis, by using the proposed image based state transition model.