<p>Mine waste dumps consist of individual, blob-like spoil piles, each with unique geological and geotechnical attributes that contribute to the overall stability of the dump. Manually characterising these individual spoil piles presents challenges due to issues of accessibility, safety risks, and time consumption. Analysis of remotely acquired images, through object-based classification, offers a promising solution for the effective identification and characterisation of individual spoil piles. However, object-based classification’s effectiveness hinges on segmentation, an aspect often overlooked in spoil pile analysis. Therefore, this study aims to identify and compare various segmentation approaches to pave the way for image-based spoil characterisation. A comparative analysis is conducted between traditional segmentation approaches and those rooted in deep learning methodologies. Among the diverse segmentation approaches evaluated, the morphology-based deep learning segmentation approach, Segment Anything Model (SAM), exhibited superior performance compared to other approaches. This outcome underscores the effectiveness of incorporating morphological data and deep learning techniques for accurate and efficient segmentation of spoil pile. The findings of this study provide valuable insights into the optimisation of segmentation strategies, thereby contributing to the application of image-based monitoring of spoil piles and promoting the sustainable and hazard free management of mine dump environments.</p>