Library-based rock thin section interpretation using image encoders
Example methods and systems for library-based rock thin section interpretation using image encoders are disclosed. One example method includes obtaining multiple images of rock thin sections, where each of the multiple images is associated with a respective rock thin section interpretation and a respective image encoding vector. An image of a rock thin section with unknown properties is obtained and encoded using an image encoder to generate an image encoding vector, where the image encoder includes a machine learning based image encoding model or a statistical image feature-based encoder. One or more images from the multiple images are selected based on the image encoding vector of the image and the respective image encoding vector for each of the multiple images of rock thin sections. The respective rock thin section interpretation of each of the one or more images is provided for interpreting the rock thin section with unknown properties.
1 . A computer-implemented method comprising:
obtaining a first plurality of images of rock thin sections, wherein each of the first plurality of images is associated with a respective rock thin section interpretation and a respective image encoding vector;
obtaining an image of a rock thin section with unknown properties;
encoding, using an image encoder, the image of the rock thin section with unknown properties to generate an image encoding vector of the image, wherein the image encoder comprises a machine learning based image encoding model or a statistical image feature-based encoder;
selecting, based on (1) the image encoding vector of the image of the rock thin section with unknown properties and (2) the respective image encoding vector for each of the first plurality of images of rock thin sections, one or more images from the first plurality of images; and
providing the respective rock thin section interpretation of each of the one or more images for interpreting the rock thin section with unknown properties.
2 . The computer-implemented method of claim 1 , wherein the image encoder comprises the machine learning based image encoding model, and the computer-implemented method further comprises:
before obtaining the first plurality of images of rock thin sections:
training, in a self-supervised learning process and based on a second plurality of images of rock thin sections, the machine learning based image encoding model; and
generating, using the trained machine learning based image encoding model, the respective image encoding vector of each of the first plurality of images.
3 . The computer-implemented method of claim 2 , wherein the machine learning based image encoding model comprises a convolutional neural network (CNN) backbone and a multilayer perceptron (MLP) head.
4 . The computer-implemented method of claim 1 , wherein the image encoder comprises the statistical image feature-based encoder, and the image encoding vector of the image of the rock thin section with unknown properties comprises Haralick features of the image of the rock thin section with unknown properties.
5 . The computer-implemented method of claim 1 , wherein selecting the one or more images from the first plurality of images comprises:
determining a respective distance between the image encoding vector of the image of the rock thin section with unknown properties and the respective image encoding vector for each of the first plurality of images of rock thin sections; and
selecting, based on the determined respective distance between the image encoding vector of the image of the rock thin section with unknown properties and the respective image encoding vector for each of the first plurality of images of rock thin sections, the one or more images from the first plurality of images.
6 . The computer-implemented method of claim 1 , further comprising:
after providing the respective rock thin section interpretation of each of the one or more images for interpreting the rock thin section with unknown properties, storing, in a database that stores the first plurality of images of rock thin sections, the image of the rock thin section with unknown properties and the interpretation of the image.
7 . The computer-implemented method of claim 1 , wherein obtaining the first plurality of images of rock thin sections comprises obtaining, from a database, each of the first plurality of images of rock thin sections, the respective rock thin section interpretation, and the respective image encoding vector.
8 . The computer-implemented method of claim 1 , wherein obtaining the image of the rock thin section with unknown properties comprises scanning the rock thin section with unknown properties using a thin section scanner to obtain the image of the rock thin section with unknown properties.
9 . The computer-implemented method of claim 1 , wherein selecting the one or more images from the first plurality of images is based on an encoding space comprising the image encoding vector of the image of the rock thin section with unknown properties and the respective image encoding vector for each of the first plurality of images of rock thin sections.
10 . A non-transitory computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:
obtaining a first plurality of images of rock thin sections, wherein each of the first plurality of images is associated with a respective rock thin section interpretation and a respective image encoding vector;
obtaining an image of a rock thin section with unknown properties;
encoding, using an image encoder, the image of the rock thin section with unknown properties to generate an image encoding vector of the image, wherein the image encoder comprises a machine learning based image encoding model or a statistical image feature-based encoder;
selecting, based on (1) the image encoding vector of the image of the rock thin section with unknown properties and (2) the respective image encoding vector for each of the first plurality of images of rock thin sections, one or more images from the first plurality of images; and
providing the respective rock thin section interpretation of each of the one or more images for interpreting the rock thin section with unknown properties.
11 . The non-transitory computer-readable medium of claim 10 , wherein the image encoder comprises the machine learning based image encoding model, and the operations further comprise:
before obtaining the first plurality of images of rock thin sections:
training, in a self-supervised learning process and based on a second plurality of images of rock thin sections, the machine learning based image encoding model; and
generating, using the trained machine learning based image encoding model, the respective image encoding vector of each of the first plurality of images.
12 . The non-transitory computer-readable medium of claim 10 , wherein the image encoder comprises the statistical image feature-based encoder, and the image encoding vector of the image of the rock thin section with unknown properties comprises Haralick features of the image of the rock thin section with unknown properties.
13 . The non-transitory computer-readable medium of claim 10 , wherein selecting the one or more images from the first plurality of images comprises:
determining a respective distance between the image encoding vector of the image of the rock thin section with unknown properties and the respective image encoding vector for each of the first plurality of images of rock thin sections; and
selecting, based on the determined respective distance between the image encoding vector of the image of the rock thin section with unknown properties and the respective image encoding vector for each of the first plurality of images of rock thin sections, the one or more images from the first plurality of images.
14 . The non-transitory computer-readable medium of claim 10 , wherein the operations further comprise:
after providing the respective rock thin section interpretation of each of the one or more images for interpreting the rock thin section with unknown properties, storing, in a database that stores the first plurality of images of rock thin sections, the image of the rock thin section with unknown properties and the interpretation of the image.
15 . The non-transitory computer-readable medium of claim 10 , wherein obtaining the image of the rock thin section with unknown properties comprises scanning the rock thin section with unknown properties using a thin section scanner to obtain the image of the rock thin section with unknown properties.
16 . A computer-implemented system, comprising:
one or more computers; and
one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform one or more operations comprising:
obtaining a first plurality of images of rock thin sections, wherein each of the first plurality of images is associated with a respective rock thin section interpretation and a respective image encoding vector;
obtaining an image of a rock thin section with unknown properties;
encoding, using an image encoder, the image of the rock thin section with unknown properties to generate an image encoding vector of the image, wherein the image encoder comprises a machine learning based image encoding model or a statistical image feature-based encoder;
selecting, based on (1) the image encoding vector of the image of the rock thin section with unknown properties and (2) the respective image encoding vector for each of the first plurality of images of rock thin sections, one or more images from the first plurality of images; and
providing the respective rock thin section interpretation of each of the one or more images for interpreting the rock thin section with unknown properties.
17 . The computer-implemented system of claim 16 , wherein the image encoder comprises the machine learning based image encoding model, and the one or more operations further comprise:
before obtaining the first plurality of images of rock thin sections:
training, in a self-supervised learning process and based on a second plurality of images of rock thin sections, the machine learning based image encoding model; and
generating, using the trained machine learning based image encoding model, the respective image encoding vector of each of the first plurality of images.
18 . The computer-implemented system of claim 16 , wherein the image encoder comprises the statistical image feature-based encoder, and the image encoding vector of the image of the rock thin section with unknown properties comprises Haralick features of the image of the rock thin section with unknown properties.
19 . The computer-implemented system of claim 16 , wherein selecting the one or more images from the first plurality of images comprises:
determining a respective distance between the image encoding vector of the image of the rock thin section with unknown properties and the respective image encoding vector for each of the first plurality of images of rock thin sections; and
selecting, based on the determined respective distance between the image encoding vector of the image of the rock thin section with unknown properties and the respective image encoding vector for each of the first plurality of images of rock thin sections, the one or more images from the first plurality of images.
20 . The computer-implemented system of claim 16 , wherein the one or more operations further comprise:
after providing the respective rock thin section interpretation of each of the one or more images for interpreting the rock thin section with unknown properties, storing, in a database that stores the first plurality of images of rock thin sections, the image of the rock thin section with unknown properties and the interpretation of the image.