Advancing arsenic contamination detection in boring cores using hyperspectral imaging and convolutional neural networks

Detection of hazardous arsenic contamination in geological formations is a critical challenge in construction and environmental monitoring. Traditional methods for identifying arsenic-containing zones in boring cores rely on time-consuming leaching tests and expert analyses, which lead to delays and increased tunnel construction costs. This study proposes an advanced approach that combines hyperspectral imaging (HSI) and convolutional neural networks (CNNs) to identify arsenic-containing areas in boring cores rapidly and accurately. Boring cores consisting of mudstone and tuff were analyzed, and arsenic concentrations were measured using a handheld X-ray fluorescence (XRF) analyzer. These XRF measurements served as viable proxies for leaching potential. Hyperspectral images were captured under controlled conditions and preprocessed for CNN training. Initial models using spectral data alone achieved low prediction accuracy (40%–70%) owing to the subtle spectral variations caused by arsenic concentrations. By integrating geological classification with spectral data, the model demonstrated significant improvement, achieving a prediction accuracy exceeding 80%. This study highlights the potential of combining HSI with CNNs for the efficient detection of arsenic-containing areas, thereby offering a scalable, cost-effective, and rapid solution for identifying hazardous zones in boring cores. These findings pave the way for improved environmental safety and reduced delays in construction and resource extraction projects.