Identification of agricultural crops using landsat 8 satellite image indices by machine learning techniques

When spectral features obtained from developing satellite-imaging technologies are used together with machine learning techniques, it is possible to detect agricultural crops on a large area with high accuracy in a short time thanks to the system trained with data on a small area. In this study, it is aimed to determine agricultural products by using satellite image indexes and different machine learning techniques. In the first stage, Landsat-8 satellite images of the area where the object to be detected is located were obtained and agricultural products were used as objects. Considering the development and harvest times of agricultural products, an agricultural land where wheat and lentil products are concentrated was selected. In order to compute the reflection indexes in the images, a time period in which agricultural products are close to the development and harvest time was preferred; Landsat-8 satellite images corresponding to May and August 2018 were used. Then, the coordinates corresponding to the sample points determined in the pilot agricultural area were imported to the Landsat-8 satellite images and NDVI values for these points were calculated with the help of the reflection indexes corresponding to the 4th and 5th bands of the satellite images. In the last stage, agricultural crops (Lentil and Wheat) were determined by using the obtained NDVI index values as the inputs of different machine learning techniques (K Nearest Neighbor, Support Vector Machines and Naive Bayes). As a result, the best performance was achieved with the Naive Bayes method with an average accuracy of 86.4%. In addition, it was observed that NDVI values obtained from the satellite image of the development period showed higher performance in the detection phase.