Detecting Irrigated Croplands: A Comparative Study with Segment Anything Model and Region-Growing Algorithms

Abstract

The advance of remote sensing and geotechnologies has helped to solve agricultural-related problems, especially those connected to management practices as irrigation. Segmentation techniques, for example, bring the possibility of identifying areas and borders of irrigated croplands, a factor that can enhance area and yield estimates. In this area, a recent innovation is the Segment Anything Model (SAM) algorithm. Thus, this study aimed to compare SAM with two segmentation algorithms, Region Growing and Baatz-Schape, for identifying irrigated croplands in the Brazilian semiarid region. Results show that SAM has potential to generate homogeneous segments when analyzing irrigated croplands but needs adjustments to separate crop fields with different crops.