ABSTRACT: Determining the mechanical properties of rocks is vital in civil engineering applications such as construction and paving. Traditional geotechnical surveys rely on in-situ and laboratory tests, which are often hindered by limited outcrop access, sampling biases, and high costs. This study explores the potential of hyperspectral imaging and machine learning to predict water absorption and uniaxial compressive strength (UCS) of carbonate rocks as a faster, cost-effective alternative. A total of 123 carbonate rock samples, with UCS values ranging from 7 to 280 MPa, were scanned using a hyperspectral camera operating in the 400-1000 nm range, capturing 448 spectral bands. The data underwent preprocessing, clustering via k-means, dimensionality reduction with Wavelet transforms, and analysis using an Artificial Neural Network to create a predictive water absorption and UCS models. The models achieved strong performance, with correlation coefficient R2 = 0.88 for water absorption and R2 = 0.90 for UCS, and relatively low error: RMSE = 2.93% for water absorption and 19 MPa for UCS. These findings highlight the feasibility of hyperspectral imaging for efficient mechanical characterization of rocks, offering promising applications in remote sensing-based geoengineering.
Bakun-Mazor et al. (Sun,) studied this question.