Identification of rock weathering by conventional methods and image analysis techniques

Abstract

Observations on weathering patterns are a crucial aspect of geotechnical analysis, as they allow the determination of site quality for various civil and mining engineering applications. There are traditional methods available to assess weathering and the effect of weathering on rock properties. In this study, selected set of laboratory testing were performed to identify the key properties of rock using representative samples from ongoing three quarry sites. The study reveals that fresh rock samples from all three quarry locations maintained a durability of over 98% through four cycles of the slake durability test. Nonetheless, these same samples exhibited decreased strength, which can be attributed to their mineral composition and internal structural arrangements of rock samples tested. Moreover, the overall findings indicate deteriorating values for the tested rock properties which could possibly be caused by rock weathering. Therefore, an attempt was made to look at using the modern technology how accurately the weathered surfaces can be identified and classified. For this purpose, machine learning (ML) techniques with remotely sensed Unpiloted Aerial Vehicles (UAVs) images were utilized. The analysis yielded an impressive F1 score of 0.88 to classify weathering in general. However, the attempts to classify different weathering grades yielded marginal results. These limitations are primarily due to factors such as the number of bands, the spatial resolution of the UAV sensors, and the availability of training data for the ML algorithm. Nonetheless, this study serves as a promising first step in demonstrating the potential of UAVs and appropriate ML models for the classification of weathering patterns, which can be further optimised and deployed for real-time observations. It is highly recommended that laboratory sample testing be carried out in conjunction with image analysis to ensure a detailed and comprehensive understanding of the test results. Keywords: Rock properties, Microstructure, Machine learning