Enhancing texture feature for mineral classification in tight sandstone rock thin-section images using super-resolution techniques

Abstract

In petrological studies, accurately classifying minerals and pore objects in rock thin-section images is essential for understanding geological conditions of oil and gas reservoirs. A key challenge in current methodologies is the limited image resolution, which restricts the accuracy of texture feature recognition crucial for effective classification. This study introduces super-resolution (SR) techniques to overcome this limitation, aiming to enhance image resolution and thereby improve the recognition and classification of texture features in thin-section imagery. To evaluate the effectiveness of various SR methods, we conducted a series of experiments evaluating their impact on mineral type classification, employing advanced techniques such as Gabor filtering and SVM classifiers for texture feature extraction and classification.Our experiments demonstrated that all evaluated SR techniques can significantly enhance classification accuracy, with each method improving accuracy by at least 19.3% for mixed color spaces. Notably, the Enhanced Deep Super-Resolution (EDSR) model, while less intuitive for human visual assessment, emerged as highly effective in extracting intricate texture features, resulting in an approximate 24.2% increase in accuracy within multi color spaces. These results suggest that the most effective SR models for enhancing texture details are not necessarily aligned with human visual evaluation. The novelty of this work is grounded in its comprehensive assessment of SR techniques in petrological imaging, marking a substantial advancement over previous methods. This study not only enhances the discernibility of critical texture features in rock thin-section imagery but also provides a clear pathway for integrating advanced image processing techniques into petrological studies.