The author considers the possibility of areas overgrowing automated determination in agricultural territory to reduce the time required to perform tasks of identifying disturbed lands and decrease the human factor influencing the result. The methodology is based on a system of converting the initial remote sensing data into a segmented image. The purpose is to achieve the highest reliability of percentage at further uncontrolled classification. The initial data is that of space survey with high and medium spatial resolution, geometric and atmospheric correction and vector boundaries of agricultural fields. The author applies the watershed method to the original images to increase the reliability of the final result. The Sobel operator is used as preprocessing method to create a gradient image. Further joining of adjacent homogeneous segments together is carried out using the Full Lambda Schedule method. The segmented image is classified through the k-means clustering technique. Materials from WorldView-3 and Sentinel-2 satellites agricultural territories in the Republic of Kalmykia and Kaluga oblast were used. The classification result is analyzed and on its basis an overgrowth vector mask of agricultural fields is created.