Rock type determination system, rock type determination method, and rock type determination program

Abstract

To effectively improve determination accuracy of a rock kind by machine learning.SOLUTION: A rock kind determination system comprises: acquisition means which acquires image data obtained by imaging a rock in association with position information of a location where the rock is imaged or collected; rock kind determination means which determines a rock kind of a rock imaged in the image data by inputting the image data acquired by the acquisition means to a learning model generated by machine learning; storage means which stores geological information indicating distribution of the rock; and narrowing means which extracts at least one or more rocks distributed within a prescribed distance range from the location where the rock is imaged or collected by referring to the geological information on the basis of the position information acquired by the acquisition means when determining the rock kind, and narrows down a determination result of the rock kind by the rock kind determination means on the basis of the extracted rock.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a rock type determination system, a rock type determination method, and a rock type determination program, and in particular to a technology suitable for determining rock type using machine learning. [Background technology]

[0002] In tunnel and dam construction, safe and economical design and construction methods are determined while understanding the geological conditions of the construction site, such as the type of rocks that make up the tunnel face and outcrop, and the strength of the rock mass. As a technology for understanding such geological conditions, for example, Patent Document 1 discloses a technology for determining the strength of the rock mass from image data of the tunnel face using a neural network. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2018-017570 A Summary of the Invention [Problem to be solved by the invention]

[0004] To determine the type of rock using machine learning such as a neural network, all rocks registered in the training data must be considered as candidates. Since there are a great many types of rocks and the training data is also enormous, there is an issue that if all rocks are considered as candidates without any narrowing down, the accuracy of rock type determination will relatively decrease.

[0005] The technology disclosed herein has been made in consideration of the above circumstances, and aims to effectively improve the accuracy of rock type determination using machine learning. [Means for solving the problem]

[0006] The rock type determination system disclosed herein comprises an acquisition means for acquiring image data of a rock in association with location information of the point where the rock was imaged or collected, a rock type determination means for determining the rock type of the rock imaged in the image data by inputting the image data acquired by the acquisition means into a learning model generated by machine learning, a storage means for storing geological information indicating the distribution of rocks, and a narrowing-down means for determining the rock type by referring to the geological information based on the location information acquired by the acquisition means, extracting at least one rock distributed within a predetermined distance range from the point where the rock was imaged or collected, and narrowing down the rock type determination result by the rock type determination means based on the extracted rocks.

[0007] In addition, it is preferable that the rock type determination means calculates the probability that the rock captured in the image data belongs to each of all the rocks registered in the learning model, and the narrowing down means selects rocks that are likely to be distributed around or coexist with the rock from the rock's formation process extracted from the geological information and sets them as determination candidates, and that the rock with the highest probability among the rocks set as determination candidates among the probabilities calculated by the rock type determination means is set as the rock type determination result.

[0008] In addition, when the rocks extracted from the geological information include any of shale, slate, sandstone, conglomerate, greenstone, chert, limestone, and tuff as rocks constituting an accretionary complex, it is preferable that the narrowing-down means sets at least shale, slate, sandstone, conglomerate, greenstone, chert, limestone, and tuff as the determination candidates.

[0009] In addition, when the rocks extracted from the geological information include sandstone, mudstone, shale, or conglomerate other than an accretionary complex, it is preferable that the narrowing-down means sets at least sandstone, mudstone, shale, conglomerate, and tuff as the judgment candidates.

[0010] In addition, when the rocks extracted from the geological information include either pyroclastic rock or volcanic rock, it is preferable that the narrowing-down means sets at least tuff, tuff breccia, rhyolite, dacite, andesite, and basalt as the judgment candidates.

[0011] In addition, when serpentine is included in the rocks extracted from the geological information, the narrowing-down means preferably sets at least gabbro and serpentine as the determination candidates.

[0012] In addition, when the rocks extracted from the geological information include plutonic rocks, the narrowing-down means preferably adds at least granite, diorite, and gabbro to the candidates, and in the case of granite and diorite, also sets aplite as an intrusive rock and hornfels that has been metamorphosed by the influence of heat at the time of intrusion as candidates.

[0013] In addition, when crystalline schist is included in the rocks extracted from the geological information, the narrowing-down means preferably sets at least pelitic schist, psammitic schist, siliceous schist, and green schist as the determination candidates.

[0014] In addition, when the rocks extracted from the geological information include either gneiss or amphibolite, the narrowing-down means preferably sets at least pelitic gneiss, psammitic gneiss, granite gneiss, and amphibolite as the determination candidates, and also sets granites that are likely to intrude under high temperatures during the formation of gneiss as the determination candidates.

[0015] In addition, when hornfels is included in the rocks extracted from the geological information, the narrowing-down means preferably sets at least hornfels as one of the determination candidates, and also sets granite and diorite, which are heat sources for contact metamorphism, as one of the determination candidates.

[0016] It is also preferable that the narrowing-down means further adds at least andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite as intrusive rocks to the candidates to be determined.

[0017] In addition, it is preferable that the narrowing-down means sets the predetermined distance range to a first distance when the rock captured in the image data is a rock that has originally been in that location, and sets the predetermined distance range to a second distance longer than the first distance when the rock captured in the image data is a rock that may have been moved from another location.

[0018] Moreover, the machine learning is preferably machine learning using a convolutional neural network.

[0019] The rock type determination method disclosed herein includes acquiring image data of a rock in association with location information of the point where the rock was imaged or collected, and inputting the acquired image data into a learning model generated by machine learning to determine the rock type of the rock imaged in the image data.When determining the rock type, geological information indicating the distribution of rocks is referenced based on the acquired location information to extract at least one rock distributed within a predetermined distance range from the point where the rock was imaged or collected, and narrowing down the rock type determination results based on the extracted rocks.

[0020] The rock type determination program disclosed herein is characterized in that it causes a computer to function as: an acquisition means for acquiring image data of a rock in association with location information of the point where the rock was imaged or collected; a rock type determination means for determining the rock type of the rock imaged in the image data by inputting the image data acquired by the acquisition means into a learning model generated by machine learning; a storage means for storing geological information indicating the distribution of rocks; and a narrowing-down means for, when determining the rock type, extracting at least one rock distributed within a predetermined distance range from the point where the rock was imaged or collected by referring to the geological information based on the location information acquired by the acquisition means, and narrowing down the rock type determination result by the rock type determination means based on the extracted rocks. Effect of the Invention

[0021] According to the technology disclosed herein, it is possible to effectively improve the accuracy of rock type determination using machine learning. [Brief description of the drawings]

[0022] [Figure 1] FIG. 1 is a schematic overall configuration diagram showing a rock type determination system according to an embodiment of the present invention. [Diagram 2] This is a schematic functional block diagram explaining the processing procedure for rock type determination using the learning model related to this embodiment. [Diagram 3] This is a flowchart that explains the process of determining rock type and the process of narrowing down determination candidates in this embodiment. [Figure 4] FIG. 1 is a schematic diagram showing an example of geological map data showing the distribution of rocks. [Diagram 5] 13 is a schematic diagram showing an example of rock addition data used when selecting rocks to be added to rock type determination candidates. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] Hereinafter, the rock type determination system, rock type determination method, and rock type determination program according to the present embodiment will be described with reference to the accompanying drawings. The same components are denoted by the same reference numerals, and their names and functions are also the same. Therefore, detailed description thereof will not be repeated.

[0024] [Rock type determination system] FIG. 1 is a schematic overall configuration diagram showing a rock type determination system 1 according to this embodiment.

[0025] As shown in Fig. 1, the rock type determination system 1 includes a communication terminal 10 and a determination device 20. The communication terminal 10 and the determination device 20 are configured to be capable of bidirectional communication via a communication network such as the Internet.

[0026] The communication terminal 10 is, for example, a mobile terminal such as a smartphone, a tablet, or a PDA, and includes an imaging unit 11 such as a camera, a location information acquisition unit 12 such as a GPS (Global Positioning System), an input unit 13, and an output unit 14.

[0027] The imaging unit 11 generates image data of the rock S to be judged (e.g., a rock mass such as excavation debris generated at the site). The position information acquisition unit 12 stores the image data captured by the imaging unit 11 in association with position information (e.g., latitude, longitude, etc. obtained by GPS). The input unit 13 is used by an operator to input various instructions. The output unit 14 is used to output various information. The input unit 13 and the output unit 14 are realized, for example, by a touch panel display of the communication terminal 10. The image data and position information stored in the communication terminal 10 are transmitted to the judgment device 20 in response to the operation of the input unit 13 by the operator.

[0028] The rocks S to be determined are not limited to rock masses generated by excavation of a face or an outcrop, and may be rocks that may have been moved from another location, such as rocks constituting an outcrop or gravel (including unconsolidated gravel layers) on a riverbed, as long as they can be imaged by the imaging unit 11. When collecting and imaging rocks, it is desirable to perform imaging at the point where the rocks were collected (including the vicinity) in order to associate the image data with positional information of the place where the rocks were present.

[0029] The determination device 20 is a computer system that determines the rock type of the rock S based on image data and the like transmitted from the communication terminal 10. The determination device 20 mainly includes a control unit 21 and a storage unit 30.

[0030] The control unit 21 includes a processor such as a CPU (or GPU), and executes a rock type determination program read from the storage unit 30. The control unit 21 also functions as a device including a CNN (Convolutional Neural Network) 22 and a narrowing-down processing unit 27 by executing the rock type determination program.

[0031] The CNN 22 inputs the image data transmitted from the communication terminal 10 into a learning model generated by machine learning, thereby determining the rock type of the rock S captured in the image data. In this embodiment, the CNN 22 determines the rock type using a convolutional neural network. Note that the CNN 22 may determine the rock type using a neural network other than a convolutional neural network. The result of the rock type determination by the CNN 22 is transmitted to the communication terminal 10 and output to the output unit 14 of the communication terminal 10.

[0032] The narrowing-down processing unit 27 performs narrowing-down processing to narrow down rock types that are candidates for determination. Specifically, the narrowing-down processing unit 27 refers to the geological map data 35 stored in the storage unit 30 based on the position information of the image data transmitted from the communication terminal 10, and extracts at least one rock distributed within a range of a predetermined radial distance R from the point where the rock S was imaged (or collected). In addition, the narrowing-down processing unit 27 narrows down the candidates for determination by limiting the rock type determined by the CNN 22 to the extracted rocks. The narrowing-down processing will be described in more detail later.

[0033] Here, the predetermined radial distance R may be either a fixed value or a variable value, but in this embodiment, it is described as a variable value. The radial distance R is automatically set by the narrowing down processing unit 27 in response to the operation of the communication terminal 10 by the operator. Specifically, the radial distance R is set to a relatively short distance (e.g., about 3 km) when the imaged rock S is a rock that originally exists there, such as a rock that constitutes a face or an outcrop. On the other hand, the radial distance R is set to a relatively long distance (e.g., 30 km) when the imaged rock S is a rock that may have been moved from another place, such as gravel in a riverbed.

[0034] The storage unit 30 includes, for example, a RAM, a ROM, etc., and stores a rock type determination program for controlling the determination device 20. In addition to the rock type determination program, the storage unit 30 also stores learning image data 31 for the CNN 22, teacher data 32, learned parameters 33, determination image data 34, geological map data 35, and additional rock data 36, ​​etc.

[0035] The training image data 31 and the teacher data 32 are data used when training the CNN 22. The trained parameters 33 and the judgment image data 34 are data used when the trained CNN 22 judges the rock type of the rock S.

[0036] The learning image data 31 is image data of rocks whose rock type is known in advance, and is input to the feature extraction unit 23 (see FIG. 2) of the CNN 22 during learning of the CNN 22. The teacher data 32 is correct answer data indicating the correct answer of the probability that a rock belongs to each of a plurality of rock types, and is compared with the data output from the classification processing unit 24 (see FIG. 2) of the CNN 22 during learning of the CNN 22. In other words, the learning image data 31 and the teacher data 32 constitute a supervised learning dataset.

[0037] The trained parameters 33 are parameters adjusted by the training of the CNN 22, and include weighting coefficients indicating the weights of connections between artificial neurons. The determination image data 34 is image data transmitted from the communication terminal 10, and is input to the feature extraction unit 23 (see FIG. 2) of the CNN 22 when determining the rock type of the rock S using the trained CNN 22.

[0038] The geological map data 35 is an example of the geological information of the present disclosure, and is a data set in which the type of rock is associated with the location information (e.g., latitude, longitude, etc.) where the rock is located. For example, the Seamless Geological Map (registered trademark) provided by the National Institute of Advanced Industrial Science and Technology can be used as the geological map data 35. The additional rock data 36 is used to supplement the candidates for determination with rocks with small distribution scales that are not included in the geological map data 35. The geological map data 35 and the additional rock data 36 will be described in detail later.

[0039] [Judgment process] FIG. 2 is a schematic functional block diagram illustrating the processing procedure for determining rock type using the learning model according to this embodiment.

[0040] As shown in Figure 2, the CNN 22 includes a feature extraction unit 23 and a classification processing unit 24. When the determination image data 34 is input to the feature extraction unit 23, the classification processing unit 24 is configured to output the probability P that the rock S depicted in the determination image data 34 belongs to each of a plurality of rock types.

[0041] The feature extraction unit 23 performs convolution processing and pooling processing on the input determination image data 34, and outputs a feature map indicating the features of the determination image data 34. Specifically, the feature extraction unit 23 has a plurality of convolution layers 23A and a plurality of pooling layers 23B arranged alternately. Note that the convolution layer 23A and the pooling layer 23B are not limited to the plurality of layers shown in the illustrated example, and one convolution layer 23A and one pooling layer 23B may be provided.

[0042] The convolution layer 23A obtains a feature map by performing a convolution operation using a filter having desired characteristics on the image data. Here, the filter used in the convolution layer 23A is obtained by learning of the CNN 22, and as the learning of the CNN 22 progresses, a filter suitable for extracting particularly important features for determining the rock type is set. The pooling layer 23B obtains a new, reduced feature map by performing a pooling process on the feature map input from the convolution layer 23A. The feature map obtained by the final pooling layer 23B is input to the input layer 24A of the classification processing unit 24.

[0043] The input layer 24A of the classification processing unit 24 outputs data obtained by multiplying the feature map input from the pooling layer 23B by a weighting coefficient to the intermediate layer 24B. Similarly, the intermediate layer 24B outputs data obtained by multiplying the data input from the input layer 24A by a weighting coefficient to the output layer 24C. Note that the weighting coefficient is appropriately changed and set to an optimal value as the learning of the CNN 22 progresses. The output layer 24C calculates the probability P that the rock S shown in the judgment image data 34 belongs to each of a plurality of rock types (all rocks registered in the learning image data 31 and the teacher data 32) based on the data input from the intermediate layer 24B.

[0044] Here, if all rocks (rock types) registered in the learning image data 31 or the teacher data 32 are treated as judgment candidates, the judgment accuracy will be relatively low. In this embodiment, in order to prevent such a decrease in judgment accuracy, a process of narrowing down the judgment candidates is performed. Specifically, the output layer 24C outputs the probability P of all the calculated rock types to the narrowing down processing unit 27. The narrowing down processing unit 27 selects the rock type with the highest probability P among the rock types that are treated as judgment candidates by the narrowing down processing described later, from the probability P of each rock type input from the output layer 24C, and transmits the rock type to the communication terminal 10 as the judgment result. The judgment result (rock type) transmitted to the communication terminal 10 is output to the output unit 14 (see FIG. 1). The narrowing down processing performed by the narrowing down processing unit 27 will be described in more detail below.

[0045] [Refine processing] FIG. 3 is a flowchart illustrating the flow of the rock type determination process and the process of narrowing down the determination candidates according to this embodiment.

[0046] 3, in step S100, the communication terminal 10 is used to capture an image of the rock S, and the image data for determination is stored in association with location information (longitude, latitude, etc.). Next, in step S110, it is selected whether the captured rock S is (A) originally located at the site, such as a face or outcrop, or (B) possibly moved from another location, such as gravel in a riverbed. The selection may be made by the operator operating the input unit 13 of the communication terminal 10.

[0047] In step S120, the determination image data and position information stored in the communication terminal 10, and (A) or (B) selected in step S110 are transmitted to the determination device 20 in response to the operation of the communication terminal 10 by the operator.

[0048] In steps S130 and S140, rock type determination of the rock S is performed using the trained CNN 22 based on the determination image data. In parallel with the rock type determination by the CNN 22, a process of narrowing down determination candidates is performed in steps S200 to S230. The flow of the rock type determination process using the CNN 22 (steps S130 and S140) has already been described based on FIG. 2, so a detailed description will be omitted here.

[0049] [Refine processing] In step S200, a radial distance R for extracting rocks from the geological map data 35 is set. Specifically, in the above-mentioned step S110, (A) if a rock that is originally at the location is selected, the radial distance R is set to a relatively short first radial distance R1 (e.g., about 3 km). On the other hand, in the above-mentioned step S110, (B) if a rock that may have been moved from another location is selected, the radial distance R is set to a relatively long second radial distance R2 (e.g., about 30 km). The first radial distance R1 and the second radial distance R2 are automatically set by the narrowing-down processing unit 27. Note that the setting of the radial distance R is not limited to the two types, the first radial distance R1 and the second radial distance R2, and may be set to any value by the operator operating the input unit 13 of the communication terminal 10 and inputting a specific value (distance).

[0050] In step S210, based on the location information (latitude, longitude, etc.) associated with the determination image data in step S100 and the radial distances R1, R2 set in step S200, at least one rock that exists within the range of radial distances R1, R2 from the point where rock S was imaged (or collected) is extracted by referring to the geological map data 35.

[0051] FIG. 4 is a schematic diagram showing an example of geological map data showing the distribution of rocks. As shown in FIG. 4, when the first radial distance R1 is set as the radial distance R, the narrowing-down processing unit 27 extracts all rocks included within the radial distance R1 from the point T1 where the rock S was imaged (or collected). In the example shown in FIG. 4, shale A and plutonic rock B are extracted. When the second radial distance R2 (>R1) is set as the radial distance R, such as when the rock S was imaged (or collected) near a river, the narrowing-down processing unit 27 extracts all rocks included within the radial distance R2 from the point T2 where the rock S was imaged (or collected). In the example shown in FIG. 4, sandstone C, chert D, pyroclastic rock E, and plutonic rock B are extracted.

[0052] Here, the geological map data 35 includes only rocks that have a large distribution large enough to be represented on a geological map, and does not include rocks that have a small distribution scale. Therefore, even if rocks within the range of the radial distances R1 and R2 are extracted from the geological map data 35, they may not be sufficient as candidates for determining the rock type.

[0053] Therefore, in this embodiment, in consideration of the formation process of rocks, etc., additional correction is performed to supplement the determination candidates with not only the rocks extracted from the geological map data 35, but also all rocks that may be distributed around the points T1 and T2 where the rock S was imaged (or collected). Details of the additional correction are described below.

[0054] [Additional correction] In step S220, the narrowing down processing unit 27 selects rocks to be candidates for determination by referring to the rock addition data 36 shown in FIG. 5 based on the rocks extracted from the geological map data 35 in the above-mentioned step S210. This rock addition data 36 includes a plurality of major categories and minor categories that define additional rocks corresponding to each of the major categories. In the rock addition data 36, ​​the major categories #1, #2, #4, #6, and #7 are rock group names, and the corresponding minor categories define the names of rocks belonging to each group. In addition, in the rock addition data 36, ​​the major categories #3, #5, #8, and #9 are individual rock names, and the corresponding minor categories define the names of rocks that tend to coexist with or be distributed around rocks in the major categories. The selection procedure for rocks in each category will be described below.

[0055] (addition complex) When the rocks extracted from the geological map data 35 include any of the rocks constituting an accretionary wedge (accretionary complex), specifically, when any of shale, slate, sandstone, conglomerate, greenstone, chert, limestone, and tuff classified as sedimentary rocks in #1 of the additional rock data 36 is included, the narrowing down processing unit 27 sets all of these sedimentary rocks (shale, slate, sandstone, conglomerate, greenstone, chert, limestone, and tuff) as candidates for determination. Furthermore, the narrowing down processing unit 27 also adds andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite classified as intrusive rocks in #2 of the additional rock data 36 to the candidates for determination.

[0056] (Rocks other than accretionary complexes) If the rocks extracted from the geological map data 35 include any of sandstone, mudstone, shale, and conglomerate other than the accretionary complex shown in the major category #3 of the additional rock data 36, ​​the narrowing down processing unit 27 sets the sandstone, mudstone, shale, conglomerate, and tuff specified in the minor category #3 as rocks that are often deposited in general sedimentary basins as the determination candidates. Furthermore, the narrowing down processing unit 27 also adds andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite, which are classified as intrusive rocks in #2 of the additional rock data 36, ​​to the determination candidates.

[0057] (pyroclastic rocks, volcanic rocks) If the rocks extracted from the geological map data 35 include either pyroclastic rock or volcanic rock, which are indicated in the major category #4 of the additional rock data 36, ​​the narrowing down processing unit 27 sets tuff and tuff breccia, which are specified in the minor category #4 of pyroclastic rock, as candidates for determination, as rocks that often coexist. The narrowing down processing unit 27 also sets rhyolite, dacite, andesite, and basalt, which are specified in the minor category #4 of volcanic rock, as candidates for determination, as rocks that often coexist. Furthermore, the narrowing down processing unit 27 adds porphyrite, granite porphyry, and dolerite, which are classified as intrusive rocks in #2 of the additional rock data 36, ​​as candidates for determination.

[0058] (Serpentine) If the rocks extracted from the geological map data 35 include serpentine, which is indicated in the major category #5 of the additional rock data 36, ​​the narrowing down processing unit 27 sets, as the determination candidates, gabbro, which both constitute the deep part of the plate, and serpentine, which is a serpentine with a carbonate mineral vein, which are specified in the minor category #5. Furthermore, the narrowing down processing unit 27 also adds andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite, which are classified as intrusive rocks in #2 of the additional rock data 36, ​​to the determination candidates.

[0059] (plutonic rock) When the rocks extracted from the geological map data 35 include plutonic rocks shown in the major category #6 of the additional rock data 36, ​​the narrowing down processing unit 27 sets granites, diorite, and gabbro, which are specified in the minor category #6 as rocks that often coexist, as the determination candidates. The narrowing down processing unit 27 also adds andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite, which are classified as intrusive rocks in #2 of the additional rock data 36, ​​to the determination candidates. In addition, in the case of granites and diorite, the narrowing down processing unit 27 also sets aplite as an intrusive rock in #2. In the case of granites and diorite, the narrowing down processing unit 27 also sets hornfels, which has been metamorphosed by the influence of heat at the time of intrusion, as the determination candidate.

[0060] (Crystalline Schist) When the rocks extracted from the geological map data 35 include crystalline schist, which is specified in the major category #7 of the additional rock data 36, ​​the narrowing down processing unit 27 sets all crystalline schists, which are specified in the minor category #7, such as pelitic schist, psammitic schist, siliceous schist, and green schist, that have different parent rocks but have been subjected to the same metamorphism, as the determination candidates. Furthermore, the narrowing down processing unit 27 also adds andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite, which are classified as intrusive rocks in #2 of the additional rock data 36, ​​to the determination candidates.

[0061] (gneiss, amphibolite) If the rock extracted from the geological map data 35 is either gneiss or amphibolite, which is specified in the major category #8 of the additional rock data 36, ​​the narrowing down processing unit 27 sets all gneisses that have different parent rocks but have undergone the same metamorphism, such as pelitic gneiss, psammitic gneiss, and granite gneiss, and amphibolites that tend to coexist with these, as candidates for determination. The narrowing down processing unit 27 also adds andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite, which are classified as intrusive rocks in #2 of the additional rock data 36, ​​to the candidates for determination. The narrowing down processing unit 27 also sets granites, which tend to intrude under high temperatures when forming gneiss, as candidates for determination.

[0062] (Hornfels) When the rocks extracted from the geological map data 35 include hornfels, which is specified in the major category #9 of the additional rock data 36, ​​the narrowing down processing unit 27 sets hornfels, which is specified in the minor category #9, and also granites and diorite, which are heat sources of contact metamorphism, as candidates for determination. The narrowing down processing unit 27 also adds andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite, which are classified as intrusive rocks in #2 of the additional rock data 36, ​​to the candidates for determination.

[0063] The rocks extracted from the geological map data 35 and the rocks selected from the additional rock data 36 in this manner are set as the final rock type determination candidates. In step S230, of all rock type probabilities P calculated in the classification process in step S140, one rock type with the highest probability P among the rock types narrowed down to the determination candidates in step S220 is selected and set as the rock type determination result. In step S240, the determination result is transmitted to the output unit 14 of the communication terminal 10, and the rock type determination process is terminated.

[0064] According to the embodiment described above in detail, when the CNN 22 determines the rock type of the rock S based on the image data of the captured rock S, the determination candidates are narrowed down based on the position information of the image data and information obtained from the geological map data 35, etc., so that the CNN 22 does not need to determine all rocks registered in the teacher data 33, etc. as candidates. This makes it possible to effectively prevent a relative decrease in the accuracy of rock type determination, and to improve the determination accuracy.

[0065] The narrowing down of the determination candidates is performed by extracting at least one rock distributed within a predetermined radius from the point where the rock was imaged or collected, based on the position information of the image data and referring to the geological map data 35. Furthermore, taking into consideration the formation process of the rock extracted from the geological map data 35, other rocks that may be distributed around the extracted rock or may coexist with the rock are selected from the additional rock data 36, ​​and the selected other rocks are set as determination candidates.

[0066] This series of processes is performed automatically by the narrowing-down processing unit 27 of the control unit 21 by referring to the geological map data 35 and the additional rock data 36. In other words, it is configured to automatically and effectively set rocks with small distribution scales that are not included in the geological map data 35 as judgment candidates. This eliminates the need for narrowing-down and additional work by people with advanced expertise, such as geological engineers, and enables workers at the construction site to easily grasp the rock types at the construction site simply by operating the communication terminal 10.

[0067] [others] The present disclosure is not limited to the above-described embodiments, and can be modified as appropriate without departing from the spirit of the present disclosure.

[0068] For example, in the above embodiment, it has been described that other rocks selected from the rock addition data 36 are added to the rocks extracted from the geological map data 35 as candidates for rock type determination, but if the rock types registered in the geological map data 35 are sufficient, the rock addition process using the rock addition data 36 may be omitted.

[0069] In the above embodiment, the rock S is imaged using a mobile terminal such as a smartphone, tablet, or PDA, but the image may be captured using a digital camera equipped with a position identification function such as RTK-GNSS. Alternatively, the rock type may be determined by importing image data captured using a digital camera without a position identification function into a personal computer. In this case, a function may be added to the determination device 20 that allows the operator to select the position information where the rock was imaged or collected on a map.

[0070] The rock type determination result does not need to be transmitted to a mobile terminal carried by the operator who captured the image of the rock S, but may be transmitted to another user terminal, such as a terminal installed in the management office of the construction site. The rock type determination program does not need to be stored in the memory unit 30 of the determination device 20, but may be stored in a computer-readable storage medium, such as a magneto-optical disk, a CD-ROM, or a DVD-ROM, or may be configured so that the determination device 20 executes a program received via a network, such as the Internet or a communication line. [Explanation of symbols]

[0071] 1...rock type determination system, 10...communication terminal, 11...imaging unit, 12...location information acquisition unit, 13...input unit, 14...output unit, 20...determination device, 21...control unit, 22...CNN, 23...feature extraction unit, 23A...convolution layer, 23B...pooling layer, 24...classification processing unit, 24A...input layer, 24B...intermediate layer, 24C...output layer, 27...narrowing processing unit, 30...memory unit, 31...learning image data, 32...teaching data, 33...learned parameters, 34...determination image data, 35...geological map data, 36...additional rock data

Claims

1. An acquisition means for acquiring image data of a rock in association with position information of a point where the rock was imaged or collected; A rock type determination means for determining the rock type of the rock captured in the image data by inputting the image data acquired by the acquisition means into a learning model generated by machine learning using a convolutional neural network; A storage means for storing geological information indicating the distribution of rocks; A narrowing-down means is provided for, when determining the rock type, extracting at least one rock distributed within a predetermined distance range from the point where the rock was imaged or collected by referring to the geological information based on the position information acquired by the acquisition means, and narrowing down the rock type determination result obtained by the rock type determination means based on the extracted rocks. A rock type determination system.

2. The rock type determination means calculates the probability that the rock captured in the image data belongs to each of all rocks registered in the learning model, The narrowing-down means selects rocks that may be distributed around the rock or may coexist with the rock from the formation process of the rock extracted from the geological information, and sets them as candidates for determination, and determines the rock type as the rock type determination result based on the probability that the rock belongs to each of all the rocks calculated by the rock type determination means, among the rocks set as the candidates for determination. The rock type determination system according to claim 1 .

3. The narrowing down means is When the rocks extracted from the geological information include any of shale, slate, sandstone, conglomerate, greenstone, chert, limestone, and tuff as rocks constituting an accretionary complex, at least shale, slate, sandstone, conglomerate, greenstone, chert, limestone, and tuff are set as the determination candidates. The rock type determination system according to claim 2.

4. The narrowing down means is When the rocks extracted from the geological information include any of sandstone, mudstone, shale, and conglomerate other than the accretionary complex, at least sandstone, mudstone, shale, conglomerate, and tuff are set as the judgment candidates. The rock type determination system according to claim 2 or 3.

5. The narrowing down means is When the rocks extracted from the geological information include either pyroclastic rock or volcanic rock, at least tuff, tuff breccia, rhyolite, dacite, andesite, and basalt are set as the determination candidates. The rock type determination system according to any one of claims 2 to 4.

6. The narrowing down means is When serpentine is included in the rocks extracted from the geological information, at least gabbro and serpentine are set as the judgment candidates. A rock type determination system according to any one of claims 2 to 5.

7. The narrowing down means is If the rocks extracted from the geological information include plutonic rocks, at least granite, diorite, and gabbro are set as the candidates. In addition, in the case of granite and diorite, aplite as an intrusive rock and hornfels that has been metamorphosed by the heat at the time of intrusion are also set as the candidates. A rock type determination system according to any one of claims 2 to 6.

8. The narrowing down means is When the rocks extracted from the geological information include crystalline schist, at least pelitic schist, psammitic schist, siliceous schist, and green schist are set as the candidates for determination. A rock type determination system according to any one of claims 2 to 7.

9. The narrowing down means is When the rocks extracted from the geological information include either gneiss or amphibolite, at least pelitic gneiss, psammitic gneiss, granite gneiss, and amphibolite are set as the candidates, and granites that are likely to intrude under high temperatures when gneiss is formed are also set as the candidates. A rock type determination system according to any one of claims 2 to 8.

10. The narrowing down means is If hornfels is included in the rocks extracted from the geological information, at least hornfels is set as the determination candidate, and granites and diorite, which are heat sources for contact metamorphism, are also set as the determination candidates. The rock type determination system according to any one of claims 2 to 9.

11. The narrowing down means is Furthermore, at least andesite, rhyolite, dacite, porphyrite, basalt, granite porphyry, and dolerite are to be added to the above-mentioned candidates as intrusive rocks. The rock type determination system according to any one of claims 2 to 10.

12. The narrowing down means is When the rock captured in the image data is a rock that has been there since its original location, the predetermined distance range is set to a first distance, and when the rock captured in the image data is a rock that may have been moved from another location, the predetermined distance range is set to a second distance that is longer than the first distance. A rock type determination system according to any one of claims 1 to 11.

13. Acquire image data of the rock in association with location information of the location where the rock was imaged or collected; The acquired image data is input into a learning model generated by machine learning using a convolutional neural network, thereby determining the rock type of the rock captured in the image data; When determining the type of rock, geological information showing the distribution of rocks is referenced based on the acquired position information, and at least one rock distributed within a predetermined distance range from the point where the rock was imaged or collected is extracted, and the rock type determination result is narrowed down based on the extracted rock. A method for determining rock type.

14. Computer, An acquisition means for acquiring image data of a rock in association with position information of a point where the rock was imaged or collected; a rock type determination means for determining the rock type of the rock captured in the image data by inputting the image data acquired by the acquisition means into a learning model generated by machine learning using a convolutional neural network; a storage means for storing geological information indicating the distribution of rocks; When determining the type of rock, the geological information is referred to based on the position information acquired by the acquisition means, and at least one rock distributed within a predetermined distance range from the point where the rock was imaged or collected is extracted, and the rock type determination result obtained by the rock type determination means is narrowed down based on the extracted rock. A rock type determination program comprising:

Images