Abstract:

In this study, crop detection was carried out in an agricultural region in the Artuklu, Kızıltepe and Derik districts of the city of Mardin through classification of multi-temporal Sentinel-2 images from 2018. In the classification, the Random Forest (RF) algorithm was used through a parcel-based approach. The detected crops are corn, wheat, cotton, chickpeas, lentils and the others. The images acquired on six different dates (April 8, May 23, July 12, August 11, September 5 and October 5) were selected as the images. In the classification, the 10m spatial resolution bands Blue (B), Green (G), Red (R) and Near Infrared (NIR) were used. Furthermore, for each image date a Normalized Difference Vegetation Index (NDVI) band was computed and used as additional band in classification. The classification process through the RF algorithm was carried out using the stack of 30 bands that includes five bands (B, G, R, NIR, NDVI) for each image. For training samples and the accuracy assessment of the results, the existing Farmers’ Registry System (FRS) was utilised. As a result of the classification process, the overall accuracy of 96.35% and the Kappa value of 93.13% were achieved.