Method for detecting and counting at least one geological constituent of a rock sample

Abstract

To this end, the invention relates to a method for detecting and counting a geological constituent (GIC) in an image acquired from a rock sample (IER), using a machine learning (ML) localization detection method. Figure 1 to be published

Description

Title of the invention: Method for detecting and counting at least one geological constituent of a rock sample. Technical field

[0001] The present invention relates to the field of detection of geological constituents of a rock sample.

[0002] Characterizing a rock sample often requires detecting and counting certain specific constituents present within the sample, such as microfossils, nanofossils, plant debris, minerals, pollen grains, or the like. For a specialist, this work can be carried out by directly observing the rock sample with the naked eye, or by cutting a thin slice, called a "thin section," for observation under an optical microscope. This allows for more precise categorization of the rock and estimation of its physical properties, for applications in civil engineering, site remediation, exploration for underground mineral or energy resources, etc.

[0003] Furthermore, when these geological constituents are microfossils (micropaleontology) or plant remains (palynology) in sedimentary rocks, this type of analysis makes it possible to estimate the age, the environment (marine, coastal, continental, etc.), and the climatic context at the time of sediment deposition. Applied to several samples from the same sedimentary sequence, it then becomes a means of reconstructing the evolution of the climate in a geographical area through geological ages (paleoclimatology). Previous technique

[0004] The detection and counting of certain geological constituents is traditionally carried out manually by a human expert, which requires adequate training and is often very tedious.

[0005] Regarding microfossils, one can consult, in particular, the Manual of Micropaleontology published in 2011 by Mathieu, Bellier, and Granier (ISBN 978-2-916733-04-3), as well as the foraminifera database at http: / / www.marinespecies.org / foraminifera / index.php. Such a database contains a very large number of data points. Therefore, this database can only be used by an expert and not in an automated process.

[0006] As regards minerals, a comparable work is the atlas of petrography published in 2001 by McKenzie and Guilford at Dunod (ISBN 978-2-100054-59-6).

[0007] Regarding pollen, see Alfred Traverse's Paleopalynology published in 2007 (ISBN 978-1-4020-5610-9).

[0008] In order to make these analyses both more accessible and faster, the automation of detection and counting using numerical methods has been considered. This can be achieved through expert systems, more or less advanced thresholding techniques, various image analysis and processing methods, and especially through machine learning methods, particularly those based on artificial neural networks.

[0009] For microfossils, see in particular: Beaufort, L., & Dollfus, D. (2004). Automatic recognition of coccoliths by dynamical neural networks. Marine Micropaleontology, 51(1-2), 57-73. This method is also described in patent application WO2015132531A1. However, such a method requires a large number of training images for the automatic learning method. The patent application specifically mentions the use of 10,000 images obtained by microscopy, and 100 images per morphological group. This large number of training images requires a very significant amount of preparation time and a large number of preparatory operations: the thin sections must be prepared (in particular by cutting the rock, impregnating it, and ground) and observations must be made using the microscope for each thin section.Next, for each image, the relevant components must be manually identified, which requires considerable work for an expert. Furthermore, this method requires images taken under various conditions: with different polarizations and in natural light, which implies the need to acquire multiple images of the same sample, thus requiring more time and specialized equipment.

[0010] Regarding a machine learning method applied to minerals, reference can be made to the document: Thompson, S., Fueten, F., & Bockus, D.

[2001] Mineral identification using artificial neural networks and the rotating polarizer stage. Computers & Geosciences, 27(9), 1081-1089. Similarly, this method requires a large number of training images (at least several hundred, or even several thousand), which raises the same drawbacks as those previously mentioned.

[0011] Another method is described in the document: Carvalho, LE, Fauth, G., Fauth, SB, Krahl, G., Moreira, AC, Femandes, CP, & Von Wangenheim, A. (2020). Automated Microfossil Identification and Segmentation Using a Deep Leaming Approach. Marine Micropaleontology, 101890. https: / / doi.org / 10.1016 / j.marmicro.2020.1018900. This is a method for detecting foraminifera in scanned core sections, based on an image segmentation method. This method also requires a large number of training images.

[0012] Furthermore, patent application EP 36737154 describes a system and a method for Determining the type of microfossils can be achieved by applying machine learning. However, this method requires segmenting the acquired images into a plurality of image data, which necessitates user intervention. Summary of the invention

[0013] The present invention aims to automatically detect, classify, and count geological constituents of a rock sample, with limited preparation time and a limited number of operations (by taking a single image per rock sample). To this end, the invention relates to a method for detecting and counting a geological constituent of a rock sample using a machine learning localization detection method.

[0014] The invention relates to a method for detecting and counting at least one geological constituent of a rock sample from training images, each geological constituent belonging to a geological constituent class. This method involves the following steps:

[0015] a. Each geological constituent is surrounded on said training images by means of a predefined geometric shape; b. Each predefined geometric shape on the said training images is associated with the said class of the said geological constituent surrounded; c. A machine learning location-based detection algorithm is trained to detect at least one geological constituent and to associate a class by means of said predefined geometric shapes and said associated classes of said training images; d. An image of said rock sample is acquired; e. A localization detection model constructed from said is applied a machine learning algorithm applied to said image acquired from said rock sample to surround at least one geological constituent with said predefined shape and to associate a class of geological constituent; and f. The number of geological constituents for each class is counted for said image acquired from said rock sample.

[0016] According to one embodiment, said classes of geological constituent are chosen from microfossils, nanofossils, plant debris, minerals, pollen grains and subdivisions of these elements.

[0017] According to one implementation, for each class of geological constituent, a color is associated.

[0018] According to one feature, said machine learning algorithm uses an artificial neural network, preferably a convolutional neural network, a network of a fully convolutional neural network or a region-based convolutional neural network.

[0019] According to one embodiment, said localization-based detection model implements the following steps:

[0020] i. At least one region of predefined geometric shape is generated comprising a geological constituent and a class is associated with each region of interest; ii. The position and / or at least one dimension of said region of interest is adjusted to surround said geological constituent.

[0021] According to one embodiment, said localization detection model generates at least one predefined geometric shape to surround a geological constituent.

[0022] Advantageously, said predefined geometric shape is chosen from a point, a square, a rectangle, an ellipse, a polygon, a circle.

[0023] According to one embodiment, said image of said rock sample is acquired from a thin section of said rock sample.

[0024] According to one embodiment, said image of said rock sample is acquired by means of an optical or electron microscope, with polarized or non-polarized light, a photograph, a scanner, a synchrotron tomography, or X-ray imaging.

[0025] According to one aspect, said process comprises at least one additional step selected from:

[0026] - Portions of said acquired image of said rock sample comprising said at least one geological constituent are extracted and a database of images of said at least one geological constituent is constructed with said portions extracted from said acquired image of said rock sample, or - Morphological and / or textural characteristics of said at least one geological constituent are determined, or - A supervised classification method is applied to categorize said at least one geological constituent, - At least one physical property of said rock is determined, or - A deformation is determined in at least one geological constituent.

[0027] The invention also relates to a method for exploiting soil or subsoil, in which the following steps are carried out:

[0028] a. At least one geological constituent of a rock sample is detected by means of the detection method according to one of the preceding characteristics; and b. The said soil or subsoil is exploited according to the said detection of the said geological constituent of the said at least one rock.

[0029] Advantageously, said exploitation of the land or subsoil relates to construction of a work on said ground or in the subsoil, the storage of gas in the subsoil, or the exploitation of raw materials from said ground or subsoil, preferably said raw materials being the rock itself, or a material, or a fluid contained in said ground or subsoil.

[0030] Furthermore, the invention relates to a method for determining the climate in a geographical area throughout geological ages, in which the following steps are implemented:

[0031] a. At least two rock samples are taken from different depths of an underground formation; b. At least one geological constituent is detected for each rock sample using the detection method according to one of the preceding characteristics; and c. The said climate and the geological age in the said geographical area are determined based on the said at least one detected geological constituent.

[0032] Other features and advantages of the process according to the invention will become apparent from the following description of non-limiting examples of embodiments, with reference to the figures attached and described below. List of figures

[0033] [Fig.1]

[0034] Fig. 1 illustrates the steps of the detection process according to one embodiment of the invention.

[0035] [Fig.2]

[0036] Fig. 2 illustrates a first variant of the predefined geometric shape used by the process according to the invention.

[0037] [Fig.3]

[0038] Fig. 3 illustrates a second variant of the predefined geometric shape used by the process according to the invention.

[0039] [Fig.4]

[0040] Figure 4 illustrates an image acquired from a rock sample and the geological constituents detected by means of the process according to one embodiment of the invention. Description of the implementation methods

[0041] The present invention relates to a method for detecting and counting geological constituents in a rock sample. A rock sample is defined as a piece of rock, for example (but not limited to) obtained by coring. Geological constituents are understood to mean at least one constituent contained within the rock. Such a geological constituent of rock can be selected from among the constituent classes The following geological constituents are identified: microfossils, nanofossils, plant debris, minerals, pollen grains, or any subdivision thereof; that is, different types of microfossils, different types of nanofossils, different types of plant debris, different types of minerals, different types of pollen grains, or any similar element. The detection of geological constituents corresponds to locating the geological constituents within the rock sample. The method according to the invention also makes it possible to identify the class of geological constituents detected and to count the number of geological constituents for each class.

[0042] The detection method according to the invention is based on the use of an image of the rock sample and on the implementation of a machine learning algorithm to automate this detection.

[0043] According to one embodiment of the invention, the image acquired from the rock sample, and the set of training images implemented for training the machine learning algorithm can be of any type.

[0044] According to one embodiment of the invention, the image may be that of an entire rock sample. Alternatively, the image may be that of a core sample taken from an underground formation. Alternatively, the image may be that of a thin section of the rock sample. The samples observed may, in particular, be so-called "thin sections" of three types: very thin sections of material on the order of 30 µm, thicker sections on the order of 100 µm, or polished sections with a thickness of up to 10 mm. To be able to properly observe these different types of samples under the microscope, it is important to have a flat surface with very low roughness (virtually smooth).

[0045] In addition, the image can be obtained in different ways, for example by means of an optical or electron microscope, with polarized or non-polarized light, by means of a scanner, by means of photography, by means of tomography with synchrotron, or by means of X-ray imaging or any similar system.

[0046] According to a first preferred embodiment of the invention, the image can be obtained by a microscope from a thin section of the rock sample. Thus, small geological constituents can be detected and counted more easily.

[0047] According to a second preferred embodiment of the invention, the image can be obtained by scanning a core sample taken from an underground formation.

[0048] According to the invention, the detection method comprises the following steps:

[0049] 1- Training the machine learning algorithm

[0050] 2 - Acquisition of the image of the rock sample

[0051] 3 - Detection and counting of the geological constituents of the rock sample

[0052] In a little more detail, the method according to the invention comprises the steps following:

[0053] 1- Training the machine learning algorithm

[0054] 1.1- Surrounding the geological constituents of the training images

[0055] 1.2- Association of a class

[0056] 1.3- Training the machine learning algorithm

[0057] 2- Acquisition of the image of the rock sample

[0058] 3 - Detection and counting of the geological constituents of the rock sample

[0059] 3.1 - Detection by localization and association of a class

[0060] 3.2 - Counting of geological constituents

[0061] Step 1 and its associated substeps 1.1 to 1.3 can be performed offline, only once, and beforehand. Steps 2 and 3, and their associated substeps 3.1 and 3.2, can be performed online for each rock sample considered. In other words, if it is desired to detect the geological constituents of several rock samples (for example, several thin sections of the same rock), steps 2 and 3 are repeated for each rock sample, as the machine learning algorithm only needs to be trained once. The steps will be detailed later in the description.

[0062] According to one embodiment, in which the machine learning algorithm is already trained (which implies that the localization detection model is already built), the steps of the process can be limited to:

[0063] 2- Acquisition of the image of the rock sample

[0064] 3 - Detection and counting of the geological constituents of the rock sample

[0065] 3.1 - Detection by location and association of a class

[0066] 3.2 - Counting of geological constituents

[0067] The process according to the invention can be implemented using a computer system, in particular a computer, in particular steps 1.3, 2, and 3 (3.1 and 3.2).

[0068] Figure 1 illustrates, schematically and without limitation, the steps of the detection process according to one embodiment of the invention. Initially, the offline steps (Off) are performed using the IAP training images. For each training image, the geological constituents are outlined (ENT) by means of a predefined shape. Then, a class is associated (ASS) with each predefined shape. The ALG machine learning algorithm for detection by location is then trained. Online (Onl), an image of the rock sample (1ER) is acquired. By applying the MOD model resulting from the ALG machine learning algorithm to the acquired image of the rock sample (1ER), the geological constituents (cge) of the rock sample are detected and counted, along with their associated class (cia).

[0069] 1- Training the machine learning algorithm

[0070] This step involves training the machine learning algorithm in order to to automate the detection, classification, and counting of geological constituents from training images, for example, from at least 10 training images. At the end of this step, the machine learning algorithm forms a location-based detection model, which takes as input a rock sample image and outputs the detection and classification of geological constituents.

[0071] Furthermore, the method according to the invention may include a preliminary step of acquiring training images, this step being implemented in the same way as step 2 of acquiring the image of the rock sample. For example, this preliminary step may consist of preparing thin sections and performing microscopy on these thin sections to form the training images.

[0072] According to one embodiment of the invention, the machine learning algorithm may be a supervised classification machine learning algorithm, for example, an artificial neural network, preferably a convolutional neural network (CNN), a fully convolutional neural network (FCNN), or a region-based convolutional neural network (R-CNN). Convolutional neural networks, fully convolutional neural networks, and region-based convolutional neural networks are particularly well-suited for image processing.

[0073] Alternatively, the machine learning algorithm can be of any type, including a non-convolutional neural network, a random forest method, a support vector machine (SVM) method, or a Gaussian process method.

[0074] Alternatively, the machine learning algorithm may be an unsupervised classification algorithm.

[0075] 1.1 - Surrounding geological constituents of training images

[0076] In this step, each geological component is surrounded on the training images by means of a predefined geometric shape. The predefined geometric shape delimits portions of the training images that contain a geological component. Preferably, the same predefined geometric shape can be used for all geological components, and only the dimensions of the predefined geometric shape vary from one geometric shape to another according to the dimensions of the geological components.

[0077] According to one embodiment, the predefined geometric shape can be chosen from a point, a square, a rectangle, an ellipse, a polygon, a circle, or any similar shape.

[0078] Figure 2 illustrates, schematically and without limitation, two portions of a training image. The figure on the left is the portion of the training image. This portion 1 of the training image includes a geological component 2. The figure on the right is the portion of the training image on which the predefined geometric shape 3, here a rectangle, surrounding the geological component 2, has been represented.

[0079] Figure 3 illustrates, schematically and without limitation, two portions of a training image. The figure on the left is the portion of the training image. This portion 1 of the training image includes a geological component 2. The figure on the right is the portion of the training image on which the predefined geometric shape 3, here an ellipse, surrounding the geological component 2, has been represented. 1.2 - Class Association

[0080] In this step, a geological constituent class is associated with each predefined geometric shape in each training image. In other words, the class of the geological constituent enclosed by the predefined geometric shape is identified (for example: microfossil, nanofossil, minerals, or the type of nanofossil, the type of mineral, the type of plant debris, etc.), and the identified geological constituent class is associated with the predefined geometric shape. Thus, the learning algorithm enables the detection and classification of geological constituents.

[0081] According to one embodiment, a color can be associated with each class of geological constituent. In this embodiment, the predefined geometric shape can be the color corresponding to the identified class of geological constituent. In this way, the visualization of the different classes of geological constituent is facilitated.

[0082] Alternatively, a parameter can be associated with each class of geological constituent. For this embodiment, the parameter corresponding to the class of geological constituent can be associated with the predefined geometric shape.

[0083] 1.3 - Training the learning algorithm

[0084] In this step, the machine learning algorithm for location-based detection is trained using training images based on predefined geometric shapes and their associated geological constituent classes. In other words, each detected object is characterized by a geometric shape described by its coordinates and at least one descriptor, such as at least one dimension, as well as by a class. Training the learning algorithm makes it possible to build a location-based detection model for geological constituents.

[0085] The machine learning algorithm can be trained in particular by one of the following methods:

[0086] - The method described in particular in the document: “J. Redmon, A. Farhadi

[2018] YOLOv3: An Incremental Improvement. Eprint arXiv, cs.CV, 1804.02767”. This is a so-called “one-step” method for real-time object detection in images. The method consists of applying a neural network to an entire image. This network can divide the image into a grid of 19x19 regions and can predict five boxes locating the objects of interest per cell of the grid. A classification is also predicted for each of the boxes. In total, 1805 boxes are predicted per image, but many do not contain an object or are redundant. A filter is then applied to all the predictions to retain only the relevant boxes. These boxes represent the objects of interest in the image. This "one-step" method improves the inference performance of detection. - The method described in particular in the document: “S. Ren, K. He, R. Girshick, J. Sun

[2015] Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. Eprint arXiv, cs.CV, 1506.01497”. This is a so-called “two-step” method that relies on a single neural architecture used during two successive tasks: • Application of neural networks and a neural layer The "Region Proposal Layer" (which can be translated as "region proposal layer") generates a set of proposed regions (e.g., 2000) of regions of interest from an image. These regions contain the objects to be detected and are filtered to eliminate irrelevant and redundant areas. • Application of the neural network and a "Spatial Pooling" layer to classify each region of interest and adjust the position and / or at least one dimension of the region of interest from the image used in the previous step.

[0087] This "two-step" method improves detection accuracy.

[0088] 2 - Acquisition of the image of the rock sample

[0089] This step involves acquiring an image of the rock sample to be analyzed. To achieve this, one can implement one of the image acquisition methods adapted to the process according to the invention: optical or electron microscopy, polarized or non-polarized, photography, scanning, synchrotron tomography, X-ray imaging, etc. Furthermore, an image can be acquired directly from the rock sample or from of a thin section of the rock sample.

[0090] According to one embodiment of the invention, when the detection method according to the invention uses thin section images, this step may include a substep of preparing a thin section from the rock sample to be analyzed.

[0091] 3 - Detection and counting of the geological constituents of the rock sample

[0092] In this step, the geological constituents of the rock sample are detected, classified, and counted using the acquired image of the rock sample and a localization-based detection model built from a machine learning algorithm. Thus, a localization-based detection model trained from the machine learning algorithm is applied to the acquired image of the rock sample. A single acquired image of the rock sample is sufficient to detect the geological constituents.

[0093] For the embodiment, in which the process is implemented for thin sections of rock, this step and the previous one can be implemented for several thin sections of the same rock, in order to improve the representativeness of the detection and counting of the geological constituents of the rock.

[0094] 3.1 - Detection by localization and association of a class

[0095] During this step, the localization detection model formed by is applied The machine learning algorithm obtained in step 1.3 is applied to the image acquired from the rock sample in step 2. The detection model automatically surrounds at least one geological component of the acquired image with a predefined geometric shape, and automatically associates a geological component class with this predefined shape. Preferably, the predefined geometric shape is identical to the one used in step 1.1; only the dimensions of the predefined shape vary from one geological component to another to accommodate the component's dimensions. At the end of this step, a new image can be created that combines the acquired image of the rock sample with the predefined geometric shapes superimposed on it.

[0096] For the embodiment in which a color or parameter is associated with a class, at this step, the color or parameter can also be assigned to each predefined geometric shape identified in the acquired image of the rock sample. Associating a color facilitates the visualization and counting of the different geological constituent classes in the acquired image of the rock sample. 3.2 - Counting of geological constituents

[0097] During this step, the number of constituents is counted automatically geological values ​​for each class of geological constituent in the acquired image of the rock sample. In other words, the number of predefined geometric shapes identified in step 3.1 for each class of geological constituent in the image acquired from the rock sample in step 2 is counted. This allows us to obtain a distribution of the geological constituents for a rock sample to be analyzed.

[0098] Figure 4 illustrates, by way of non-limiting illustration, in its left part, an image 4 acquired from a rock sample, and in its right part, image 9 of image 4 with predefined geometric shapes surrounding the geological constituents. This acquired image 4 was obtained by microscopy of a thin section of the rock sample. Image 9 was obtained by the detection and counting method according to the invention. For the detection method according to the invention, the machine learning algorithm was trained with 15 training images. In addition, a convolutional neural network and the two-step method were implemented. In image 9, the geological constituents are identified by rectangles (predefined geometric shapes) 5, 6, 7, and 8. Each type of rectangle corresponds to a class of foraminifera (which is a type of microfossil). The method according to the invention makes it possible to determine the following distribution:

[0099] - Six foraminifera of the class corresponding to the rectangles referenced 5, - Ten foraminifera of the class corresponding to the rectangles referenced 6, - One foraminifera of the class corresponding to the rectangle referenced 7, and - Six foraminifera of the class corresponding to the rectangles referenced 8.

[0100] Thus, the method according to the invention makes it possible to automatically detect and count the geological constituents of an image of a rock sample.

[0101] Furthermore, the method according to the invention may include at least one of the following additional steps:

[0102] - The proportion of the volume of the rock sample occupied by the geological constituents is determined from the detection of the geological constituent, in other words from the acquired image, and / or - Morphological and / or textural characteristics (e.g., shape, dimensions, presence of striations, etc.) of geological constituents are estimated, for example by means of image processing, and / or - Portions of the acquired image of the rock sample containing a geological constituent are extracted, and a database of images of geological constituents is constructed from the extracted portions of the acquired image of the rock sample, and / or - A supervised classification method is applied to categorize geological constituents more precisely, starting from the detection of the constituent. geological. For example, one can determine the subspecies to which each detected microfossil belongs. This step may require having previously trained an artificial neural network to perform this categorization, using a dedicated image database, and independently of the learning algorithm of the detection and counting method according to the invention, and / or - At least one physical property of the rock is determined; indeed, the detection and counting of geological constituents makes it possible to determine physical properties, including porosity, compaction, mechanical resistance, or permeability, depending on the type and shape of the geological constituents detected, and / or - We determine the deformation of the geological constituent that occurred between the time of deposition of the geological constituent and the time of analysis of the rock sample. For example, we can analyze the shape of the geological constituent and compare its shape with other geological constituents belonging to the same class.

[0103] These steps can be carried out automatically, preferably using computer-based means, such as a computer.

[0104] Furthermore, the invention relates to a method for exploiting the soil or subsoil. This method involves the following steps:

[0105] a. At least one geological constituent of a rock sample is detected by means of the detection method according to any one of the variants or combinations of variants described above; and b. The soil and / or subsoil is exploited according to the geological constituents of the rock sample detected in the previous step.

[0106] The first step makes it possible in particular to determine the physical properties of the rock to be analyzed, the exploitation being implemented according to these physical properties.

[0107] For this soil or subsoil exploitation process, the process may include a preliminary step of taking a rock sample from the soil or subsoil.

[0108] The operation may relate in particular to the field of building construction or civil engineering works, or the field of raw material exploitation, or the field of gas storage, or the field of risk determination, or the field of site remediation, etc.

[0109] In the field of construction, the composition of rock outcrops and / or the subsoil is determined by rock categorization, and construction is carried out by adapting, in particular, the foundations and the building structure according to the rock categorization. For these applications, the rock to be categorized may be taken from the ground or subsoil in a shallow manner.

[0110] In the field of raw material exploitation (for example for quarries, mines, hydrocarbon recovery, etc.), the constitution of rock outcrops and / or subsoil is determined by detecting the geological constituents of the rock, and raw material exploitation is carried out (raw materials can be the rock itself, a material, for example a metal, or a fluid, for example hydrocarbons, present in the subsoil), in particular by allowing the determination of suitable areas (i.e. drilling areas, areas to be excavated for mines or quarries, etc. in order to recover raw materials), and the determination of methods and tools to be used (for example enhanced hydrocarbon recovery, drilling tools, the nature of explosive devices for mines or quarries, etc.).For these applications, the rock to be categorized may be taken from deep underground, may result from drilling debris, or may come from a rock outcrop, etc.

[0111] In the field of gas storage, for example CO2, the constitution of the subsoil is determined by the categorization of the rock, and gas storage is carried out in the subsoil in an appropriate area, that is to say in an underground area suitable for storing the gas without leakage.

[0112] In the field of risk determination, the constitution of a rock outcrop (cliff) is determined by the categorization of the rock, and a consolidation operation is carried out if there is a risk of subsidence or collapse of the rock outcrop.

[0113] Thus, this process allows for the exploitation of the soil and / or subsoil in a simple and rapid manner, without requiring the services of an expert geologist. The process also allows for the faster processing of very large quantities of rock.

[0114] The invention also relates to a method for determining the climate in a geographical area throughout geological ages, in which the following steps are implemented:

[0115] a. At least two rock samples are taken from different depths of an underground formation, the rock samples being able to come from the same succession of rock deposits; b. At least one geological constituent is detected for each rock sample using the detection method according to any of the variants or combinations of variants described above; and c. The said climate and the geological age in the geographical area are determined based on said at least one detected geological constituent.

[0116] As will be understood, the invention is not limited to the embodiments described above by way of example; on the contrary, it encompasses all variants of realization.

Claims

Claims

1. Method for detecting and counting at least one geological constituent (cge) of a thin section of a rock sample from training images (IAP), each geological constituent belonging to a geological constituent class (cia), characterized in that the following steps are implemented: a. Each geological constituent (cge) is surrounded (ENT) on said training images (IAP) by means of a predefined geometric shape; b. We associate (ASS) with each predefined geometric shape (cge) on said training images (IAP) said class of said surrounded geological constituent; c. A machine learning algorithm (ALG) for location detection is trained to detect at least one geological constituent (cge) and to associate a class (cia) using said predefined geometric shapes and said associated classes of said training images (IAP); d. An image is acquired from said thin section of said rock sample (1ER); e. A location detection model (MOD) constructed from said machine learning algorithm (ALG) is applied to said acquired image of said thin section of said rock sample (1ER) to surround at least one geological constituent (cge) by said predefined shape and to associate a geological constituent class (cia); and f. The number of geological constituents (cge) for each class (cia) is counted for said image acquired from said thin section of said rock sample (1ER).

2.

3.

4. Method according to claim 1, wherein said classes (cia) of geological constituent are chosen from microfossils, nanofossils, plant debris, minerals, pollen grains and subdivisions of these elements. Method according to one of the preceding claims, in which for each class (cia) of geological constituent, a color is associated. Method according to one of the preceding claims, in which said al- Machine learning algorithm (ALG) uses an artificial neural network, preferably a convolutional neural network, a fully convolutional neural network, or a region-based convolutional neural network.

5. Method according to one of the preceding claims, in which said location detection model implements the following steps: i. At least one region of predefined geometric shape comprising a geological constituent is generated and a class is associated with each region; ii. The position and / or at least one dimension of said region is adjusted to surround said geological constituent.

6. A method according to one of claims 1 to 4, wherein said location detection model generates at least one predefined geometric shape to surround a geological constituent.

7. Method according to one of the preceding claims, in which said predefined geometric shape is chosen from a point, a square, a rectangle, an ellipse, a polygon, a circle.

8. Method according to one of the preceding claims, in which said image of said rock sample (1ER) is acquired by means of an optical or electronic microscope, with polarized or non-polarized light, a photograph, a scanner, a tomography with synchrotron, or X-ray imaging.

9. Method according to one of the preceding claims, wherein said method comprises at least one additional step chosen from: - Extracting portions of said acquired image of said rock sample comprising said at least one geological constituent and constructing a database of images of said at least one geological constituent with said portions extracted from said acquired image of said rock sample, or - Determining morphological and / or textural characteristics of said at least one geological constituent, or - Applying a supervised classification method to categorize said at least one geological constituent, - Determining at least one physical property of said rock, or - Determining a deformation of said at least one constituent geological.

10. Process of exploiting a soil or subsoil, in which one puts in implement the following steps: a. At least one geological constituent (gc) of a rock sample is detected by means of the detection method according to one of the preceding claims; and b. Said soil or subsoil is exploited based on said detection of said geological constituent (GC) of said at least one rock.

11. A method of exploiting a soil or subsoil according to claim 10, wherein said exploitation of the soil or subsoil relates to the construction of a structure on said soil or in the subsoil, the storage of gas in the subsoil, or the exploitation of raw materials from said soil or subsoil, preferably said raw materials being the rock itself, or a material, or a fluid contained in said soil or in said subsoil.

12. A method of determining the climate in a geographical area through geological ages, in which the following steps are implemented: a. At least two rock samples are taken from different depths of an underground formation; b. At least one geological constituent (gc) is detected for each rock sample by means of the detection method according to one of claims 1 to 10; and c. The said climate and the geological age in the said geographical area are determined based on the said at least one geological constituent (GCC) detected.

Images