Determination of cotton and corn plant fields by employing deep transformer encoder technique and different time-series satellite images

Turkey is a rich country in agriculture because it is located in the middle belt. It is very important to determine the products in the agricultural field in a short time and accurately. Thanks to the satellite images obtained from remote sensing, information can be obtained on many subjects such as the development of agricultural products detection and annual product forecasting. In this study, it is aimed to detect corn and cotton from agricultural products by using Sentinel-1 and Landsat-8 satellite image indexes and deep architecture of agricultural products together. In the first stage, a pilot area was determined to obtain Sentinel-1 and Landsat-8 satellite images of the agricultural products to be determined. While choosing agricultural products, an agricultural land with high corn and cotton products, whose development and harvest times are close, was chosen. Afterwards, the coordinates of 100 sample points from this pilot area were taken with the help of GPS and these coordinates were transferred to Sentinel-1 and Landsat-8 satellite images and reflectance values were obtained. In order to calculate the reflection values of the images, the months of June, July, August and September of the 2016-2021 period, when the development and harvest times of the agricultural products to be determined are close to each other, were preferred. The data set used in the study was obtained with the help of the Google Earth Engine Code Editor (GEE-CE), and a total of 434 images for Sentinel-1 satellite images and 693 images for Landsat-8 for the months of June, July, August, September between 2016-2021. consists of images. At the last stage, the reflection values obtained were classified into three different categories. These are: 1-) Classification with only Sentinel-1 bands, 2-) Classification with B1-B7 bands of Landsat-8 only, 3-) Classification with B1-B7 bands of both Sentinel-1 and Landsat-8. These three different reflection values were given to the Transformer Deep Learning network input and agricultural products (Corn and Cotton) were determined. In the first classification, 85% classification accuracy was obtained when only the reflectance values of Sentilel-1 satellite images were used. In the second classification, 95% classification accuracy was found for the reflectance values of the satellite images of the B1-B7 bands of Landsat-8. In the third classification, when the reflectance values of the B1-B7 bands of Sentinel-1 and Landsat-8 were used together, an average accuracy value of 87,5% was observed.