METHOD FOR CATEGORIZING A ROCK ON THE BASIS OF AT LEAST ONE IMAGE

Abstract

The present invention relates to a rock classification method wherein at least one image (IMA) of the rock to be classified is acquired, and wherein a decision tree (ARB) classifying the rocks according to several descriptors is used, as well as a machine learning method (APP) from a rock image database (BIR). Machine learning is applied for each descriptor considered.

Claims

1-11. (canceled)

12. A method of classifying a rock by use of a rock image database storing rock images comprising: a) constructing a decision tree that classifies rock images according to descriptors characterizing the rocks; b) associating with each rock image of the rock image database descriptors characterizing the rock; c) acquiring at least one image of the rock to be classified; d) determining at least one descriptor for the at least one acquired image of the rock to be analysed by machine learning from the rock image database with machine learning being applied to each descriptor which was considered; and e) classifying the rock to be classified by use of the at least one determined descriptor and of the constructed decision tree.

13. A rock classification method as claimed in claim 12, wherein the constructed decision tree is a decision tree of naturalistic description of the rock.

14. A rock classification method as claimed in claim 12, wherein the at least one image of the rock to be classified is acquired by one photography, by microscopy or using a scanner.

15. A rock classification method as claimed in claim 13, wherein the at least one image of the rock to be classified is acquired by one photography, by microscopy or using a scanner.

16. A rock classification method as claimed in claim 14, wherein the image of the rock to be classified is acquired by photography by use of a smartphone.

17. A rock classification method as claimed in claim 15, wherein the image of the rock to be classified is acquired by photography by use of a smartphone.

18. A rock classification method as claimed in claim 12, wherein the constructed decision tree classifies lithologic facies of the rock.

19. A rock classification method as claimed in claim 13, wherein the constructed decision tree classifies lithologic facies of the rock.

20. A rock classification method as claimed in claim 14, wherein the constructed decision tree classifies lithologic facies of the rock.

21. A rock classification method as claimed in claim 15, wherein the constructed decision tree classifies lithologic facies of the rock.

22. A rock classification method as claimed in claim 16, wherein the constructed decision tree classifies lithologic facies of the rock.

23. A rock classification method as claimed in claim 17, wherein the constructed decision tree classifies lithologic facies of the rock.

24. A rock classification method as claimed in claim 18, wherein at least one descriptor characterizing the rock is selected from among origin of the rock, chemistry of the rock, presence of foliation, presence of olivine, presence of bedding, and the presence of mica.

25. A rock classification method as claimed in claim 12, wherein the machine learning uses a neural network.

26. A rock classification method as claimed in claim 25, wherein the neural network is a convolutional database.

27. A rock classification method as claimed in claim 12, wherein the method comprises pre-processing the at least one acquired image prior to determining at least one descriptor of the at least one acquired image.

28. A rock classification method as claimed in claim 12, wherein the method comprising inserts into the rock image database the at least one acquired image and the at least one associated determined descriptor.

29. A soil or subsoil exploitation method in accordance with claim 12 comprising exploiting the soil or the subsoil according to the classification of the at least one rock.

30. A soil or subsoil exploitation method as claimed in claim 29, wherein the soil or subsoil exploitation concerns construction of structure on the soil or in the subsoil, storage of a gas in the subsoil, or exploitation of raw materials of the soil or the subsoil.

31. A soil or subsoil exploitation method as claimed in claim 12, wherein the raw materials are the rock itself, or a material or a fluid contained in the soil or in the subsoil.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0030] Other features and advantages of the method according to the invention will be clear from reading the description hereafter of embodiments given by way of non-limitative example, with reference to the accompanying figures wherein:

[0031] FIG. 1 illustrates the steps of the method according to one embodiment of the invention;

[0032] FIG. 2 illustrates the steps of the method according to a second embodiment of the invention;

[0033] FIG. 3 illustrates the steps of the method according to a third embodiment of the invention;

[0034] FIG. 4 illustrates a decision tree according to an embodiment of the invention; and

[0035] FIG. 5 illustrates a decision tree for an example embodiment of the invention for classification of the lithologic facies of a rock.

DETAILED DESCRIPTION OF THE INVENTION

[0036] The present invention relates to a rock classification method. Rock classification is understood to be the determination of a target characteristic of a rock. It may be identification of a rock, for example notably of the lithologic facies of the rock. It may also concern petrophysical property ranges such as grain size and porosity, or geologic age and formation condition ranges, the history of the rock mechanics, etc. The method according to the invention uses a rock image database, it may be a database with images of different rocks.

[0037] The method according to the invention comprises the following steps:

[0038] 1. Decision tree construction;

[0039] 2. Associating descriptors with the rock image database;

[0040] 3. Rock image acquisition;

[0041] 5. Descriptors determination; and

[0042] 6. Rock classification.

[0043] The method according to the invention can be implemented by use of a computer system, notably a computer, a smartphone, etc. Steps 1) and 2) can be carried out in this order or simultaneously. Steps 1) and 2) can be carried beforehand and they can be carried out offline. Steps 3) to 6) can be carried out online. These steps are described in detail in the rest of the description.

[0044] FIG. 1 schematically illustrates, by way of non-limitative example, the steps of the method according to one embodiment of the invention. A decision tree ARB classifying the rocks according to descriptors (characteristics) is constructed beforehand, and descriptors (among those used in the decision tree) are associated with each image of rock image database BIR. Machine learning APP is performed from the rock image database. An image IMA of the rock to be classified is acquired online Only then, by use of machine learning APP, descriptors desc (only three are shown) of the rock to be classified are determined by inference INF, and these descriptors desc are applied to decision tree ARB in order to determine DET the classification car of the rock to be classified.

[0045] According to one embodiment, the method can further comprise a step of image pre-processing. In this case, the method according to the invention comprises the following steps:

[0046] 1. Decision tree construction;

[0047] 2. Associating descriptors with the rock image database;

[0048] 3. Rock image acquisition;

[0049] 4. Image pre-processing;

[0050] 5. Descriptors determination; and

[0051] 6. Rock classification.

[0052] The method according to the invention can be implemented by use of a computer system, notably a computer, a smartphone, etc. Steps 1) and 2) can be carried out in this order or simultaneously. Steps 1) and 2) can be carried beforehand and they can be carried out offline. Steps 3) to 6) can be carried out online. These steps are described in detail in the rest of the description.

[0053] FIG. 2 schematically illustrates, by way of non-limitative example, the steps of the method according to this embodiment of the invention. A decision tree ARB classifying the rocks according to descriptors (characteristics) is constructed beforehand, and descriptors (among those used in the decision tree) are associated with each image of rock image database BIR. Machine learning APP is performed from the rock image database. An image IMA of the rock to be classified is acquired online Onl, then pre-processing PRT of the acquired image is performed, then, by use of machine learning APP, descriptors desc (only three are shown) of the rock to be classified are determined by inference INF, and these descriptors desc are applied to decision tree ARB in order to determine DET the classification car of the rock to be classified.

[0054] According to an implementation of the invention, the method can further comprise a step of updating the image database with the acquired image. In this case, the method according to the invention comprises the following steps:

[0055] 1. Decision tree construction;

[0056] 2. Associating descriptors with the rock image database;

[0057] 3. Rock image acquisition;

[0058] 5. Descriptors determination;

[0059] 6. Rock classification; and

[0060] 7. Updating the image database.

[0061] The method according to the invention can be implemented by use of a computer system, notably a computer, a smartphone, etc. Steps 1) and 2) can be carried out in this order or simultaneously. Steps 1) and 2) can be carried beforehand and they can be carried out offline. Steps 3) to 7) can be carried out online. These steps are described in detail in the rest of the description.

[0062] FIG. 3 schematically illustrates, by way of non-limitative example, the steps of the method according to this embodiment of the invention. A decision tree ARB classifying the rocks according to descriptors (characteristics) is constructed beforehand, and descriptors (among those used in the decision tree) are associated with each image of rock image database BIR. Machine learning APP is performed from the rock image database. An image IMA of the rock to be classified is acquired online Only then, by use of machine learning APP, descriptors desc (only three are shown) of the rock to be classified are determined by inference INF, and these descriptors desc are applied to decision tree ARB in order to determine DET the classification car of the rock to be classified. Finally, the new acquired image and the associated descriptors are added to machine learning APP. This step is illustrated by the arrows relative to descriptors desc pointing towards the machine learning step.

[0063] The embodiments described above can be combined to combine the effects thereof. The method can then comprise a step of pre-processing the image and a step of updating the rock image database.

[0064] 1. Decision Tree Construction

[0065] This step constructs a decision tree that classifies the rocks according to several descriptors (referred to as auxiliary characteristics, different from the target characteristic) classifying the rock. A decision tree is a tool representing a set of options using a tree-like model. The various possible decisions are located at the ends of the branches (the “leaves” of the tree), and they are reached depending on the decisions made at each step. In the present case, each step corresponds to a value of a descriptor, and each branch end corresponds to a rock classification. It is thus possible to precisely determine the classification of a rock by use of several descriptors. Indeed, when the descriptors are known, all the steps allowing to reach the end of a single branch of the decision tree are obtained. It is possible that, to reach the end of a branch of the decision tree, all the descriptors are not used.

[0066] FIG. 4 schematically illustrates, by way of non-limitative example, a decision tree that can be used for the method according to the invention. In this figure, the descriptors are denoted by desc1, desc2, desc3 and desc4, and they are identified by dotted vertical lines, the values of the descriptors are denoted by letters a to m, and the rock classifications are denoted from car1 to car8. For this example, the first decision corresponds to descriptor desc1 that can take the values a or b. When first descriptor desc1 is a, second descriptor desc2 can take the values c or d, whereas when first descriptor desc1 is b, second descriptor desc2 is e or f, and so on. By combining three levels of descriptors, the classification of the rock is determined. For example, if first descriptor desc1 is a, second descriptor desc2 is c and third descriptor desc3 is g, then the associated classification is classification car1.

[0067] According to an embodiment of the invention, the decision tree can be constructed from a naturalistic description of the rocks. In other words, the descriptors and the classifications can be selected to correspond to the rock families. This embodiment allows a naturalistic classification of the rocks to be selected.

[0068] For example, for the embodiment wherein the classification corresponds to the lithologic facies of the rock, the classifications may notably be: dunite, gabbro, orthogneiss, granite, basalt, mica schist, schist, etc. For this embodiment, at least one naturalistic rock descriptor can be determined. In particular, the naturalistic rock descriptor can notably be selected from among:

[0069] origin of the rock: it may notably be plutonic, effusive or basic;

[0070] chemistry of the rock: it may notably be acidic or basic;

[0071] presence of foliation: it may be present or absent;

[0072] presence of olivine: it may be present or absent;

[0073] presence of bedding: it may be present or absent; or

[0074] presence of mica: it may be present or absent.

[0075] These (non-limitative) descriptors provide good discrimination of the most frequently encountered rocks. Indeed, these descriptors allow accounting for the texture, the origin, the structure (bedding, folia), the color related to the chemistry of the rock, and the presence of specific minerals (olivine, mica).

[0076] FIG. 5 schematically illustrates, by way of non-limitative example, a decision tree for this embodiment. In this figure, the symbols are identical to those used in FIG. 4. The first descriptor concerns the origin, which may be plutonic (a), effusive (b) or sedimentary (c). When the origin is plutonic (a) or effusive (b), the second descriptor concerns the chemistry, which may be basic (d, f) or acidic (e, g). When the origin is sedimentary (c), the second descriptor is the foliation that may be present (h) or absent (i). When the origin is plutonic (a) and the chemistry basic (d), the third descriptor is the olivine that may be present (j) or absent (k). In this case, if olivine is present (j), then the rock is a dunite, and if it is absent (k), the rock is a gabbro. When the origin is plutonic (a) and the chemistry acidic (e), the third descriptor is the bedding that may be present (l) or absent (m). In this case, if the bedding is present (l), the rock is an orthogneiss, and if it is absent (m), the rock is a granite. When the origin is effusive (b) and the chemistry basic (f), the rock is a basalt. When the origin is effusive (b) and the chemistry acidic (g), the rock is of another type N/C. When the origin is sedimentary (c) and foliation is present (d), the third descriptor is the presence of mica, which may be present (n) or absent (o). In this case, if mica is present (n), the rock is a mica schist, and if it is absent the rock is a schist. When the rock is sedimentary (a) and foliation is absent (i), the rock is of another type N/C.

[0077] Within the decision tree, the number of steps and values of each branch is not limited to the examples of FIGS. 4 and 5.

[0078] The number of classes (number of branch endpoints of the decision tree) depends on the desired precision for the rock classification. According to one aspect of the invention, for the embodiment wherein the method classifies the lithologic facies of the rock for a broad public, the number of classes can range between 5 and 30, preferably between 8 and 20. According to another aspect of the invention, for the embodiment wherein the method classifies the lithologic facies of the rock for a specialized public, the number of classes can range between 5 and 300, or even more.

[0079] 2. Associating Descriptors with the Rock Image Database

[0080] This step associates with each image of the rock image database of multiple rock descriptors. The descriptors are those used for constructing the decision tree. One learning database is thus formed per descriptor being considered.

[0081] This association of descriptors can be performed by image processing, which can notably analyse the colors, shapes, dimensions of the rocks of the rock image database, or any other information contained in the image. According to a feature of the invention, the descriptors can also comprise geolocation of the image. This geolocation can be related to the image, notably if it originates from a photograph taken by a smartphone. Indeed, this descriptor allows comparing rocks having close geographical areas.

[0082] Alternatively, associating the images with the descriptors can be done by a user.

[0083] The number of rock images of the rock image database can depend on the number of descriptors considered in the decision tree. A large number of rock images allows the rock classification precision to be improved. However, a large number of rock images requires a large memory capacity and it may involve a significant execution time. The rock image database can comprise at least one image per rock classification. According to one embodiment, the rock image database can comprise about a hundred rock images, preferably the number of rock images can be several hundred rock images, preferably the number of rock images can be several thousand rock images, or even several ten thousand rock images.

[0084] 3. Rock Image Acquisition

[0085] This step acquires at least one image of the rock to be classified. Preferably, it can be a digital image.

[0086] According to one implementation of the invention, a microscopic or macroscopic image can be acquired. It may be an image of a thin section of a rock, an image of a rock sample or an image of an outcrop (cliff for example). It is thus possible to determine the classification of the rock whatever the information available.

[0087] According to one embodiment of the invention, the image of the rock to be classified can be acquired by photography, microscopy, a scanner or any other similar way. These ways enable simple acquisition only requiring conventional equipment.

[0088] According to an example embodiment, the rock image can be acquired by photography using a smartphone. Thus, acquisition of the rock image requires no significant equipment and it can be performed anywhere, without any technical prerequisite.

[0089] In order to improve the rock classification precision, several images of the rock can be acquired.

[0090] 4. Image Pre-Processing

[0091] This is an optional step of the method. It pre-processes the image acquired in step 3) prior to the step of determining the descriptors of the acquired image. This step notably allows improving the descriptor determination precision.

[0092] According to one embodiment of the invention, pre-processing the acquired image can select a zone of interest in the acquired image. For example, pre-processing can allow to removing the image background that does not correspond to the rock to be classified.

[0093] In a variant, pre-processing can modify the image, notably in terms of contrast, brightness, enlargement, etc.

[0094] According to one implementation of the invention, pre-processing can be related to the image acquisition method, for example polarized light, infrared wavelength, etc.

[0095] 5. Descriptors Determination

[0096] This step determines at least one descriptor of the acquired image of the rock to be classified by use of machine learning from the rock image database and the descriptors associated in step 2), and by application to the image acquired in step 3) (and possibly pre-processed in step 4). One machine learning process is then applied per descriptor being considered, and a plurality of descriptors of the acquired image are obtained.

[0097] This step allows, for each descriptor being considered, to compare the acquired image with the rock images of the rock image database, and thus to classify the acquired image among the image families having the same properties. Several characteristics (several descriptors) of the rock to be classified are thus determined independently, which allows the rock classification to be precisely determined.

[0098] According to one implementation, this step can use image processing (colors, shapes, dimensions, etc.) to perform machine learning.

[0099] According to one embodiment of the invention, machine learning uses a neural network, preferably a convolutional neural network. The convolutional neural network is particularly suitable for image processing.

[0100] Alternatively, the machine learning technique used can be of any type, notably a non-convolutional neural network, a random forest method, a support vector machine (SVM) or a Gaussian process method.

[0101] 6. Rock Classification

[0102] This step classifies the rock by applying the decision tree constructed in step 1) with the descriptors determined in step 5). In other words, knowing the descriptors, the decisions of the decision tree are known, and the branch end of the decision tree corresponding to the rock to be classified can therefore be known. This branch then indicates the classification of the rock.

[0103] For an image of a rock to be classified, the decision tree of the example corresponding to the embodiment of FIG. 5 and the following descriptors determined in step 5) are applied:

[0104] origin: plutonic

[0105] chemistry: acidic

[0106] bedding: presence.

[0107] Therefore, according to the decision tree, the rock to be classified is an orthogneiss, which corresponds to the classification made by a geologist.

[0108] It is noted that, for this rock type, a method of the prior art using the same rock image database and the same predictor (convolutional neural network), and as described in the preamble of the description, has classified this rock as granite. Therefore, the method according to the invention allows a rock to be classified with more precision than the methods according to the prior art.

[0109] According to one embodiment of the invention, the rock classification can be displayed on a computer system, a computer or a smartphone for example.

[0110] 7. Updating the Image Database

[0111] It is noted that this step is optional in the method. It updates the rock image database with the acquired images. Thus, the learning bases of each descriptor can be expanded each time the method is implemented, which promotes the reliability and robustness of the method.

[0112] According to one implementation of the invention, the rock image database can be updated when the rock classification has been validated by another method (a method according to the prior art or by a geologist).

[0113] Furthermore, the invention relates to a soil or subsoil exploitation method wherein the following steps are carried out:

[0114] a) classifying at least one rock of at least one of the soil and subsoil by use of the rock classification method according to any one of the variants or variant combinations described above; and

[0115] b) exploiting at least one of the soil and subsoil according to the classification of the rock in the previous step.

[0116] Exploitation can notably concern the field of building or engineering structure construction, or the exploitation of raw materials, or gas storage, or the field of risk assessment, etc.

[0117] In the field of construction, the composition of at least one of rocky outcrops and of the subsoil is determined by classification of the rock, and construction is achieved by adapting notably the foundations, the structure of the building according to the rock classification. For these applications, a sample of the rock to be classified can be taken from the soil or the subsoil at shallow burial depth.

[0118] In the field of raw material exploitation (for example for quarries, mines, hydrocarbon recovery, etc.), the composition of at least one of the rocky outcrops and of the subsoil is determined by rock classification, and exploitation of the raw materials is performed (the raw materials can be the rock itself, a material, for example a metal or a fluid, for example hydrocarbons present in the subsoil), by allowing notably to determine suitable zones (i.e. drill areas, dig areas for mines or quarries, etc., for raw materials recovery), and to determine the methods and the tools to be used (for example enhanced oil recovery, drilling tools, the nature of the explosives for mines or quarries, etc.). For these applications, a sample of the rock to be classified can be taken at depth in the subsoil, it may originate from drill cuttings, or from an outcrop, etc.

[0119] In the field of gas storage, CO.sub.2 for example, the composition of the subsoil is determined by rock classification, and storage of the gas in the subsoil is performed in a suitable zone, i.e. an underground zone suitable for gas storage without leakage risk.

[0120] In the field of risk assessment, the composition of a rocky outcrop (cliff) is determined by rock classification, and consolidation operations are carried out if there is a risk of collapse or landslide of the rocky outcrop.

[0121] Thus, this method enables exploitation of at least one of the soil and subsoil in a simple and fast manner, without the need to interact with a geology expert. The method also allows very large amounts of rocks to be treated more rapidly.

[0122] Another possible application of the method according to the invention is information to the public about the rock classification (geotourism).

[0123] The invention is not restricted to the embodiments described above by way of example and that it encompasses all variant embodiments.