Comparing the Segment Anything Model with Region Growing Algorithms in the detection of irrigated croplands

Abstract

The advance of remote sensing and geotechnologies has helped to solve agricultural-related problems, especially those connected to management practices such as irrigation. Image segmentation techniques, for example, bring the possibility of identifying areas and borders of irrigated croplands,a factor that can enhance monitoring and yield estimates. In this research field, a recent innovation is the Segment Anything Model (SAM) algorithm. Thus, this study aimed to compare SAM with two well-known remote sensing image segmentation algorithms, Region Growing and Baatz-Schape, in order to delineate irrigated agricultural lands in the Brazilian semiarid region. The findings indicate that SAM has the potential to generate homogeneous segments when examining such lands. However, it requires refinements in order to distinguish fields with varying crops and to improve the high computational cost of SAM, especially for big data. Additionally, the choice and testing of parameters are crucial for the optimal performance of segmentation algorithms.