Determination of product pattern from satellite images using different classification algorithms – a case study of ilgin

Determination of agricultural crop patterns by utilizing spectral data and machine learning algorithms is one of the primary areas of use of remote sensing technology in digital agriculture. In this study, the determination of different crop patterns was investigated by classifying the Sentinel-2 satellite images of 1128.80 hectares of agricultural land located in Ağalar neighborhood of Ilgın district of Konya province with learning algorithms belonging to the year 2024. Common crop types in the study area are distributed as cereals (wheat - barley), sunflower, corn, sugar beet and alfalfa plants. Maximum likelihood and Random forest techniques were used as image classification algorithms. In order to determine the training and test areas to be used in the algorithms, a field study was carried out between May and September and 98 (%70) of the ground truth data taken in 140 parcels were randomly selected as training and 42 (%30) as test data. Accuracy tests and error matrices of machine learning algorithms were created and comparisons were made. Sentinel-2 satellite images representing the vegetation periods of satellite image plant varieties were obtained on 8 different dates. In the classification, B2, B3, B4 and B8 bands with 10 m spatial resolution from these images and the NDVI index created with these bands were used. In the classification made by including the multi-temporal and multi-band image stack in the classification at once, the general accuracy rate was determined as 93.48% in the RO algorithm and 91.31% in the EB algorithm. At the same time, after the classification, it was determined with 90% accuracy which products were planted in 280 parcels whose product declaration was not given in the study area and 179 (18%) parcels with incorrect ÇKS declaration were detected. It has been evaluated that the thematic maps produced as a result of the research can be announced through posters, brochures, social media etc., resources and various communication channels, which can be effective in preventing erroneous and/or misleading statements on a farmer basis, and it has been suggested that the product classification made using satellite images should not be limited to the application of a scientific method, but can be turned into an effective technical tool in monitoring and managing agricultural production processes in practice, supported by local infrastructure facilities. Keywords: Classification, Konya, Maximum likelihood, Product pattern, Random forest, Remote sensing Sentinel-2