An evaluation of calibrated and uncalibrated high-resolution RGB data in time series analysis for coal spoil characterisation: A comparative study

Abstract Minor errors in the spoil deposition process, such as placing stronger materials with higher shear strength over weaker ones, can lead to potential dump failure. Irregular deposition and inadequate compaction complicate coal spoil behaviour, necessitating a robust methodology for temporal monitoring. This study explores using unmanned aerial vehicles (UAV) equipped with red–green–blue (RGB) sensors for efficient data acquisition. Despite their prevalence, raw UAV data exhibit temporal inconsistency, hindering accurate assessments of changes over time which could be attributed to radiometric errors. To this end, the study introduces an empirical line calibration with invariant targets (ELC-IT), for precise calibration across diverse scenes, particularly in the context of UAV imagery used to monitor the evolving nature of spoil dumps. To evaluate the effectiveness of this calibration approach, accuracy assessment of an object-based classification is conducted on both calibrated and uncalibrated data. This classification involves several steps: performing segmentation, carrying out feature extraction, and integrating the extracted features and ground truth labels collected over the time period of UAV image capture into machine learning pipelines. Calibrated RGB data exhibit a substantial performance advantage, achieving a 90.7% overall accuracy for spoil pile classification using ensemble (subspace discriminant), representing a noteworthy 7% improvement compared to classifying uncalibrated data. The study highlights the critical role of data calibration in optimising UAV effectiveness for spatio-temporal mine dump monitoring. These findings play a crucial role in informing and refining sustainable management practices within the domain of mine waste management.