Automated method and system for categorising and describing thin sections of rock samples obtained from carbonate rocks
An automated method for categorising and describing rock samples includes: (i) capturing images of samples to generate image data; (ii) classifying the samples by applying convolutional neural network processing to the image data to identify attributes of the samples; (iii) extracting, using a module based on a trained convolution deconvolution neural network, salient features from the attributes to generate salient feature data; (iv) applying, using a module based on a trained recurrent neural network generated from supervised learning, semantic text processing to the salient feature data to generate sample text data indicative of the attributes; (v) storing the sample text data; (vi) searching an image database for categorised images that are similar to the captured images; and (vii) generating output data dependent upon a degree of similarity between the sample text data and the database text data to indicate petrophysical properties of the rocks.
1 . An automated method for categorizing and describing thin sections of reservoir rock samples obtained from carbonate rocks, the method comprising:
capturing, using an input image data module, optical or microscopic images of one or more thin sections of one or more input rock samples obtained from reservoir carbonate rocks so as to generate input rock image data;
classifying, using a classifying module based on a trained convolutional neural network generated from supervised learning, the one or more input rock samples into lithological and quality categories of carbonate thin sections by applying convolutional neural network processing to the input rock image data to identify one or more attributes of the input rock samples;
extracting, using a trained salient feature extraction module based on a trained convolution deconvolution neural network, salient features from the one or more attributes of the input rock samples so as to generate salient feature data;
applying, using a semantic text processing module based on a trained recurrent neural network generated from supervised learning, semantic text processing to the salient feature data using learned associations between salient-feature data and geological descriptive terminology so as to generate input rock sample text data indicative of the one or more attributes of the input rock samples;
storing, using a storage module, the input rock sample text data;
searching, using a search module, a rock image database for categorized images that are similar to the captured images based on the stored rock sample text data and database text data associated with the categorized images stored in the rock image database, in which the rock image database comprises categorized images of categorized rock samples and the database text data, and the database text data is indicative of one or more attributes of the categorized rock samples and is associated with petrophysical measurements of the categorized rock samples; and
generating output data dependent upon a degree of similarity between the rock sample text data and the database text data so that the output data indicates petrophysical properties of the carbonate rocks associated with the thin sections of input rock samples that are identified as being within a selected degree of similarity to the categorized rock samples in the rock image database based on a comparison of the stored rock sample text data with the database text data, and in which the output data is based on the petrophysical measurements of the categorized rock samples identified as being within a selected degree of similarity to the input rock samples.
2 . The method according to claim 1 , in which the output data comprises a textual description of the petrophysical properties of the input rock sample.
3 . The method according to claim 1 , in which extracting the salient features comprises performing image semantic segmentation using deconvolution neural network processing.
4 . The method according to claim 1 , in which the semantic text processing comprises applying recurrent neural network processing to the salient feature data.
5 . The method according to claim 4 , in which the recurrent neural network processing comprises applying a long short term memory (LTSM) stacked network to the salient feature data to generate a textual summary of image content of the image data.
6 . The method according to claim 1 , in which the degree of similarity is user selectable.
7 . The method according to claim 1 , in which the convolutional neural network comprises using an intermediate layer between two convolutional layers.
8 . The method according to claim 7 , in which the intermediate layer comprises using a rectified linear units (ReLU) nonlinear activation function.
9 . The method according to claim 7 , in which the convolutional neural network processing comprising using two or more of: a convolution layer; a max pooling layer; and a fully connected layer.
10 . The method according to claim 7 , in which the intermediate layer comprises a max pooling layer.
11 . The method according to claim 1 , comprising:
generating a labelled image training database based on observing and labelling a plurality of carbonate reservoir rock samples including providing qualitative assessment of salient feature abundance;
generating a semantic text training database comprising text data comprising text data relating to descriptions of carbonate reservoir rock samples;
generating a measured rock properties database comprising rock property data associated with labelled and unlabeled rock training images;
generating the trained convolutional neural network for the classifying module by using supervised learning based on a first subset of the labelled image training database so as to classify the rock training images;
generating the trained convolution deconvolution neural network for the trained salient feature extraction module using supervised learning based on a second subset of the labelled image training database so as to recognize carbonate reservoir rock features and their abundance; and
generating the trained recurrent neural network for the semantic text processing module by using supervised learning based on seed words from the trained convolutional neural network and the trained convolution deconvolution neural network.
12 . A non-transitory computer-readable medium storing a computer program comprising computer readable instructions which, when executed by one or more processors, cause the one or more processors to:
capture, using an input image data module, images of one or more thin sections of one or more input rock samples obtained from reservoir carbonate rocks so as to generate input rock image data;
classify, using a classifying module based on a trained convolutional neural network generated from supervised learning, the one or more input rock samples by applying convolutional neural network processing to the input rock image data to identify one or more attributes of the input rock samples;
extract, using a trained salient feature extraction module based on a trained convolution deconvolution neural network, salient features from the one or more attributes of the input rock samples so as to generate salient feature data;
apply, using a semantic text processing module based on a trained recurrent neural network generated from supervised learning, semantic text processing to the salient feature data so as to generate input rock sample text data indicative of the one or more attributes of the input rock samples;
store, using a storage module, the input rock sample text data;
search, using a search module, a rock image database for categorized images that are similar to the captured images based on the stored rock sample text data and database text data associated with the categorized images stored in the rock image database, in which the rock image database comprises categorized images of categorized rock samples and the database text data, and the database text data is indicative of one or more attributes of the categorized rock samples and is associated with petrophysical measurements of the categorized rock samples; and
generate output data dependent upon a degree of similarity between the rock sample text data and the database text data so that the output data indicates petrophysical properties of the carbonate rocks associated with the thin sections of input rock samples that are identified as being within a selected degree of similarity to the categorized rock samples in the rock image database based on a comparison of the stored rock sample text data with the database text data, and in which the output data is based on the petrophysical measurements of the categorized rock samples identified as being within a selected degree of similarity to the input rock samples.
13 . The non-transitory computer-readable medium of claim 12 , wherein the computer-executable instructions further cause the one or more processors to:
apply recurrent neural network processing to the salient feature data,
wherein the recurrent neural network processing comprises applying a long short term memory (LTSM) stacked network to the salient feature data to generate a textual summary of image content of the image data.
14 . An automated system for categorizing and describing thin sections of reservoir rock samples obtained from carbonate rocks, the system comprising a processor in communication with memory storing instructions executable by the processor to:
capture optical or microscopic images of one or more thin sections of one or more input rock samples obtained from reservoir carbonate rocks so as to generate input rock image data;
classify, based on a trained convolution neural network generated from supervised learning, the one or more rock samples into lithological and quality categories of carbonate thin sections by applying convolutional neural network processing to the input rock image data to identify one or more attributes of the rock samples;
extract, based on a trained convolution deconvolution neural network, salient features from the one or more attributes of the rock samples so as to generate salient feature data;
apply semantic text processing, based on a trained recurrent neural network generated from supervised learning, to the salient feature data using learned associations between salient-feature data and geological descriptive terminology so as to generate input rock sample text data indicative of the one or more attributes of the rock samples;
store the input rock sample text data in a storage module;
search a rock image database for categorized images that are similar to the captured images based on the stored rock sample text data and database text data associated with the categorized images stored in the rock image database, in which the rock image database comprises categorized images of categorized rock samples and the database text data, and the database text data is indicative of one or more attributes of the categorized rock samples and is associated with petrophysical measurements of the categorized rock samples; and
generate output data dependent upon a degree of similarity between the rock sample text data and the database text data so that the output data indicates petrophysical properties of the carbonate rocks associated with the thin sections of input rock samples that are identified as being within a selected degree of similarity to the categorized rock samples in the rock image database based on a comparison of the stored rock sample text data with the database text data, and in which the output data is based on the petrophysical measurements of the categorized rock samples identified as being within a selected degree of similarity to the input rock samples.