Investigation of the utility of wheat plant multi-band phenological spectral signatures in crop monitoring and precise yield predictions.
Project no. 21594 · Status: Kapanmış (Yayın bekleniyor) · 2020
Summary
<p>The use of remote sensing techniques in the agricultural field has developed primarily in crop recognition, monitoring of their growth, and crop mapping. Researches continue on crop yield estimation from satellite images. However, soil color, topography, plant variety, etc. Significant deviations in the yield estimation due to the impacts have not been eliminated. In order to determine product pattern and yield information at the international level and to manage agricultural policies, there is a strong need to determine crop reflection values, product development, and yield relations, which are of strategic importance. In order to error rate in yield determination studies, reference parcels and soil maps are needed, in which field yield information is accurately recorded. This project aims to monitor the development of many parcels for known products and determine the yield with deep learning or artificial intelligence. It is planned to evaluate the last 2 years of parcel yield information of TIGEM farms as a reference. Product development and yield may vary based on field and region. The reflected energy level, recorded in different wavelengths from satellite images, is closely related to product development and thus yield. Phenology-based multi-band spectral signature will be created in each satellite image to be taken throughout their phenology. This process will be completed by the formation of spectral signatures explaining the product development on the basis of regions and establishing a yield relationship with the deep learning technique. Within the framework of the project, the plan of the parcels where wheat cultivation is carried out in 5 enterprises located in different climatic zones, which will be cooperated with TİGEM, has been digitized and coordinated. After this stage, the product development from planting to harvest was monitored with Sentinel-2 satellite images to be taken in approximately 6-10 day periods. The statistical relationship between the field yield for each plot and the spectral signatures of the multi-band reflectance values obtained from each satellite was examined. At this stage, bands of images and parcel yield information of previous years for similar climate and crop patterns were added to the system through the software to be developed for deep learning. Thus, a system model has been created to automatically make the same product pattern in the coming years, depending on the location of the enterprises, in the development of different climatic conditions, reflection relations and ultimately yield estimation by machine learning method.</p>