Intelligent Image Segmentation for Organic-Rich Shales Using Random Forest, Wavelet Transform, and Hessian Matrix

Scanning electron microscope (SEM) image can capture the distribution, topology, and morphology of microstructural constituents of geological materials. Segmentation of SEM image is needed to delineate/locate the various microstructural constituents. To locate locating kerogen/organic matter and pores in shale samples, we test an automated SEM-image segmentation workflow involving feature extraction followed by machine learning, as an alternative to threshold-based and object-based segmentation. For each pixel in the SEM image, 16 features are generated and then fed to a random forest classifier to determine the presence of the four shale components, namely: 1) pore/crack; 2) rock matrix including clay, calcite, and quartz; 3) pyrite; and 4) organic/kerogen components. With the help of feature extraction techniques such as wavelet transform and Hessian affine region detector, the proposed segmentation methodology can segment one 2058 pixel × 2606 pixel SEM image in approximately 30 s. The performance of the trained classifier, quantified in terms of overall F1 score, on the validation data set was higher than 0.9. The newly developed method is significantly robust in comparison to the popular histogram thresholding and object-based segmentation methods.

I. Introduction

Scanning electron microscope (SEM) image analysis facilitates the quantification of the microstructural topology, morphology, and connectivity of distinct components in a material. Image analysis has been well adopted in industries requiring material characterization. Image segmentation is a crucial step in image analysis. Segmentation is the division of an image into spatially continuous, disjoint, and homogeneous regions [1]. Manual segmentation performed by the subject matter expert is the most reliable segmentation approach; however, it requires considerable time, attention, and patience. We propose an automated image-segmentation method that couples classical feature extraction with supervised classification algorithms.