TS Analyser

TS Analyser software presents an intelligent identification of rock thin sections using image processing. It provides a semi-automatic approach for quantitative and qualitative petrographic thin section analysis that combines with image processing. The method can greatly improve the identification efficiency and reduce human error in the identification results. The traditional identification of thin sections depends on visual observation, while the intelligent identification of sections using image processing is based on the size of the samples and uses deep learning technology to realize the quantitative and qualitative extraction of information.

The developed application is capable of detecting more than 15 properties of thin sections. Eight of them are already used in industries as general properties of thin sections such as cleavages, area, roundness, extinction angle, colours, percentage, perimeter, and Average Diameter. The rest of the features introduced by the software itself include six colour images (see the analysis section), mixed colour images, and average extinction colours. Features like Fracture, inclusions, twinning, and pore type can also be identified by developing the cleavage detection mythology that was introduced in this software. Crystal shape and Relief can be identified by developing the S Segment process.

Segmentation of particles can be done by various image processing. The proposed method allowed us to identify, cleavages, and other optical properties with reasonable accuracy. However further development of mythology and algorithms may be needed for more accurate results. By evaluating research data, it also found that the speed of colour change may vary from mineral to mineral concerning rotation. Therefore, nearly 20 properties of thin sections can be detected by image processing of thin section videos.

These detected properties with more sample video footage are very much enough to train the AI that can used for intelligent recognition of mineral components in thin sections. Well-trained CNN or RNN can used to train the AI with the properties that are detected by software. Training of AI may consume time and the experience of experts but later on, it will produce very accurate and quick analysis.