Cannabis sativa L. spectral discrimination using satellite imagery machine learning

Monitoring the earth with remote sensing technologies has shown significant improvement, and thanks to these developments, its use for agricultural purposes has increased. Today, satellite remote sensing method is used to detect various plants that are the raw material source of many narcotic substances that are illegally produced. This thesis uses high-resolution satellite images to determine the differences between cannabis and other plants in the study area. For that purpose, spectral signatures and NDVI values of the cultivated cannabis plant in Kastamonu province have been extracted and analyized. In this thesis, which was carried out with a comprehensive literature search, plants' morphological characteristics and phenological development processes were examined. In this context, satellite images of the locations of plant species in Kastamonu and Eskişehir provinces were obtained from the PlanetScope satellite platform. ArcGIS software and QGIS software were preferred for spectral value calculations and NDVI analysis of the satellite images obtained during the study. In addition, the obtained data was used for training a different machine learning algorithms for cannabis classification. Thus, five different machine learning algorithms included in the WEKA software were used as classifiers. The thesis results can be used for understanding the phenology of cannabis plants. The sampling data needs to be increased for future studies, and image classification needs to be performed using higher spatial resolution data.