Fragmented croplands with small and irregular fields pose formidable challenges to represent agricultural field boundaries. Existing freeware do not consider novel segmentation algorithms, thus restrict their application to relatively homogeneous, large landholdings. We developed an open source QGIS plug-in ‘HS-FRAG’ to delineate agricultural fields of a fragmented landscape from a multi-spectral imagery using an object-based, hybrid segmentation algorithm. ‘HS-FRAG’ uses Sobel operator to extract the edges from individual spectral bands, followed by watershed algorithm to close the polygons. Three metrics, namely geometric evaluation (GE), quantitative completeness (QC), and spatial correctness (SC) were considered to evaluate segmentation performance. Robustness of the tool was demonstrated on a highly fragmented, heterogeneous cropland in south India using Sentinel-2 (10 m spatial) and Cartosat-2 (1.6 m spatial) imagery. Both Sentinel-2 and Cartosat-2 have resulted in over-segmentation (OS) with a generalization of 1.30, 1.37 and an overall accuracy (OA) of 38.06 %, 75.61 % respectively. Segmentation with Cartosat-2 has outperformed in representing the geometry, number, and extent of the reference polygons. A user guide with working examples is provided for quick acquaintance of ‘HS-FRAG’ tool and its application to other fragmented regions. © 2022 Elsevier B.V.