Automatic High-Throughput Detection of Fluid Inclusions in Thin-Section Images using a Novel Algorithm

Detection and examination of fluid inclusions can lead to insight into diagenesis and history of a hydrocarbon reservoir. Currently, geologists are faced with a large collection of digital images, making manual investigation highly inefficient and time-consuming. A combination of data acquisition by simple and robust light microscopy and an advanced downstream computational solution is ideal with respect to cost efficiency and stable large-scale and high-throughput ability. Moreover, a strict deterministic algorithm for automatic fluid inclusion detection is not subjected to any human biases that would otherwise violate detection consistency. A novel algorithm for fluid inclusion detection has been developed and implemented in the statistical scripting language R and the object-oriented programming language C# under the .NET 4.0 computational framework. The algorithm is a result of thorough understanding of the image representation of fluid inclusions and optimized based on empirical correlations. It is based on sequential image section-specific intensity-centric selection criteria, intensity distribution discrimination and conditional statistics. Moreover, a multi-dimensional scoring-scheme has been developed and implemented. The software was able to successfully identify a series of fluid inclusions on the same thin-section image containing hydrocarbon and aqueous phases, respectively. Due to the modular structure the software is highly flexible and can be tailor-made to specific analytical needs, such as selective identification of solid phases.