Training sample enhancement method and device for full convolutional neural network ore prediction

Abstract

The application discloses a kind of training sample enhancement methods and devices for whole convolutional neural network prospecting prediction, belong to mineral resources exploration technical field, solve the deficiency of training sample data enhancement mode in whole convolutional neural network prospecting prediction in prior art.In target prospecting area, model area for generating training sample and the geosciences big data space database of model area are obtained;The size and spatial position of the cumulative distribution function of the random clipping window based on the model area are determined respectively;Random parameter combination is formed based on each cumulative distribution function, and training sample is clipped in model area based on random parameter combination.This application can randomly generate any number, size, and location of training samples, ensuring the number of samples required for training the prospecting prediction model and the randomness and diversity of the samples, which can better optimize the prospecting prediction model based on the whole convolutional neural network and improve the quality of prospecting prediction.

Description

Technical Field

[0001] This invention belongs to the field of mineral resource exploration technology, and specifically relates to a training sample enhancement method and device for mineral exploration prediction using fully convolutional neural networks. Background Technology

[0002] Geoscience big data is a product of the deep integration of earth sciences and information technology. Its core lies in the integration and application of multi-dimensional information about the Earth system through data acquisition, storage, analysis, and visualization technologies. This field not only drives paradigm shifts in traditional geological research but also provides intelligent solutions for resource development, finding wide application in mineral exploration prediction.

[0003] In mineral exploration prediction based on geoscientific big data, fully convolutional neural network (FCN) models are currently a cutting-edge and hot research area, achieving some success in mineral exploration. However, because mineralization is a low-probability event, the number of formed ore bodies is limited, making it difficult to meet the basic requirement of large-capacity training samples for FCN models. To address this issue, data augmentation methods are needed. Currently, in the field of computer vision, although many methods are used for data augmentation, such as flipping, rotating, cropping, and scaling, these methods, for geoscientific big data, either alter the geological properties of the samples (e.g., generating new samples by changing the fracture direction through rotation) or generate samples lacking randomness (e.g., generating new samples by cropping with a fixed window size and step size using a sliding window). These are shortcomings of traditional FCN data augmentation methods for training samples in mineral exploration prediction. Summary of the Invention

[0004] In view of the above analysis, the embodiments of the present invention aim to provide a training sample augmentation method and apparatus for mineral exploration prediction of fully convolutional neural networks, so as to solve one or more of the above-mentioned problems existing in the prior art.

[0005] The objective of this invention is achieved as follows:

[0006] A training sample augmentation method for mineral exploration prediction using fully convolutional neural networks, comprising the following steps:

[0007] Within the target mineral exploration area, acquire the model region used to generate training samples, as well as the geoscientific big data spatial database of the model region;

[0008] Determine the cumulative distribution functions of the size and spatial location of the random cropping window based on the model region, respectively;

[0009] Random parameter combinations are formed based on each cumulative distribution function, and training samples are cropped out within the model region based on these random parameter combinations.

[0010] This application presents a training sample augmentation method for mineral exploration prediction using fully convolutional neural networks. The method acquires a model region for generating training samples and a geoscientific big data spatial database of the model region within the target mineral exploration area. It then determines the cumulative distribution functions for the size and spatial location of random clipping windows based on the model region. Random parameter combinations are formed based on these cumulative distribution functions, and training samples are clipped from the model region based on these random parameter combinations. This method avoids the problems of traditional training sample data augmentation in geoscientific big data, and can randomly generate any number of training samples of varying sizes and locations. This ensures the required number of samples and the randomness and diversity of samples for training large-scale mineral exploration prediction models, thereby better optimizing the mineral exploration prediction model based on fully convolutional neural networks and improving the quality of mineral exploration predictions.

[0011] As one embodiment, the process of establishing a geoscientific big data spatial database includes the following steps:

[0012] Acquire geoscientific big data within the target mineral exploration area; this includes geological data, mineral data, geophysical data, geochemical data, heavy mineral data, and remote sensing data.

[0013] Unify the spatial coordinate system and data type format of geoscientific big data, and generate a geoscientific big data spatial database composed of predicted elements and ore body annotations.

[0014] As one embodiment, the process of acquiring a geoscientific big data spatial database of a model region within a target mineral exploration area for generating training samples includes the following steps:

[0015] Within the target mineral exploration area, select an area where the exploration level meets the set exploration conditions as the initial area;

[0016] By cropping the target mineral exploration area based on the distribution range of the initial region, a model region and its geoscientific big data spatial database are obtained.

[0017] As one embodiment, the process of determining the cumulative distribution function of the size and spatial location of the random cropping window based on the model region is as follows:

[0018] Cumulative distribution function of the size of the random cropping window As shown in the following formula:

[0019] ;

[0020] in, This indicates the size of the randomly cropped window. This indicates the starting point for setting the size of the randomly cropped window. Indicates the height of the model region, Indicates the width of the model region;

[0021] Cumulative distribution function of the height of the spatial location of the random cropping window As shown in the following formula:

[0022] ;

[0023] in, Indicates the height of the spatial position of the random cropping window;

[0024] Cumulative distribution function of the width of the random cropping window at its spatial location As shown in the following formula:

[0025] ;

[0026] in, This indicates the width of the spatial position of the randomly cropped window.

[0027] As one embodiment, the process of forming random parameter combinations based on each cumulative distribution function and cropping training samples within the model region based on these random parameter combinations includes the following steps:

[0028] Based on the cumulative distribution function, a random generator is used to generate random parameter combinations, including the size and spatial location of the random cropping window;

[0029] Use a cropping tool to crop the training samples for each combination of random parameters.

[0030] As one embodiment, the process of generating a random parameter combination including the size and spatial location of a randomized cropping window using a random generator based on a cumulative distribution function includes the following steps:

[0031] The range of random numbers determined by the combination of random parameters is an integer between the interval [a, b].

[0032] Using Mason slew as a random number generator, pseudo-random numbers between [0,1) are generated by initializing a random seed;

[0033] Multiply the pseudo-random number by the interval width (b-a+1) and round down to obtain an integer offset of [0, ba]. Add a lower limit 'a' to the integer offset to ensure that the result includes both endpoints, and obtain random integers between the intervals [a, b]. Combine these random integers to form a random parameter combination.

[0034] As one embodiment, the process of cropping training samples for each random parameter combination using a cropping tool includes the following steps:

[0035] Determine the starting row and column coordinates x1 and y1 of the cropping window;

[0036] The cropping window is directly mapped to the pixel row and column indices of the raster data, and the values ​​are assigned as x1=w, y1=h, width=d, height=d; where d is the size of the random cropping window, width is the number of columns in the cropping window, and height is the number of rows in the cropping window.

[0037] This application also provides a training sample augmentation device for mineral exploration prediction using a fully convolutional neural network, including:

[0038] The region selection module is used to obtain the model region for generating training samples and the geoscientific big data spatial database of the model region within the target mineral exploration area.

[0039] The function determination module is used to determine the cumulative distribution functions of the size and spatial location of the random cropping window based on the model region, respectively.

[0040] The sample cropping module is used to generate random parameter combinations based on each cumulative distribution function, and to crop training samples within the model region based on these random parameter combinations.

[0041] This application's embodiment of the training sample augmentation device for mineral exploration prediction using fully convolutional neural networks acquires a model region for generating training samples and a geoscientific big data spatial database of the model region within a target mineral exploration area. It determines the cumulative distribution functions of the size and spatial location of random clipping windows based on the model region. Based on these cumulative distribution functions, it forms random parameter combinations and clips training samples from the model region using these random parameter combinations. This avoids the problems of traditional training sample data augmentation in geoscientific big data, and can randomly generate any number of training samples of varying sizes and locations. This ensures the required number of samples and the randomness and diversity of samples for training large-scale mineral exploration prediction models, enabling better optimization of the fully convolutional neural network-based mineral exploration prediction model and improving the quality of mineral exploration prediction.

[0042] At least one embodiment of this application also provides a data control device, including:

[0043] One or more memories that store computer-executable instructions non-transitory;

[0044] One or more processors are configured to run computer-executable instructions, wherein the computer-executable instructions are executed by the one or more processors to implement a training sample augmentation method for mineral prospecting prediction based on any embodiment of the present application.

[0045] The aforementioned data control device acquires a model region for generating training samples and a geoscientific big data spatial database of the model region within the target mineral exploration area; determines the cumulative distribution function of the size and spatial location of the random clipping window based on the model region; forms random parameter combinations based on each cumulative distribution function; and clips training samples from the model region based on these random parameter combinations. This avoids the problems of traditional training sample data augmentation in geoscientific big data, and can randomly generate any number of training samples of varying sizes and locations. This ensures the required number of samples and the randomness and diversity of samples for training the large-scale mineral exploration prediction model, enabling better optimization of the fully convolutional neural network-based mineral exploration prediction model and improving the quality of mineral exploration predictions.

[0046] At least one embodiment of this application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement a training sample augmentation method for mineral prospecting prediction based on any embodiment of this application.

[0047] The aforementioned non-transient computer-readable storage medium acquires a model region for generating training samples and a geoscientific big data spatial database of the model region within the target mineral exploration area. It determines the cumulative distribution function of the size and spatial location of the random clipping window based on the model region. Based on these cumulative distribution functions, it forms random parameter combinations and clips training samples from the model region using these random parameter combinations. This avoids the problems of traditional training sample data augmentation in geoscientific big data, and can randomly generate any number of training samples of varying sizes and locations. This ensures the required number of samples and the randomness and diversity of samples for training large-scale mineral exploration prediction models, enabling better optimization of mineral exploration prediction models based on fully convolutional neural networks and improving the quality of mineral exploration predictions.

[0048] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings.

[0050] Figure 1 A flowchart illustrating a training sample augmentation method for mineral exploration prediction using a fully convolutional neural network, according to an embodiment of the application.

[0051] Figure 2 A flowchart of a training sample augmentation method for mineral exploration prediction using a fully convolutional neural network, according to a preferred embodiment;

[0052] Figure 3 This is a schematic diagram of the clipping within the model area;

[0053] Figure 4 This is a structural diagram of a training sample augmentation device module for mineral exploration prediction using a fully convolutional neural network, according to an embodiment of the application.

[0054] Figure 5 A schematic block diagram of a data control device provided by the present invention;

[0055] Figure 6 This is a schematic diagram of a non-transitory computer-readable storage medium provided by the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0057] To facilitate understanding of the embodiments of this application, further explanation and description will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application. In the drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.

[0058] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “(the)” are also intended to include the plural forms. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values ​​that would be recognized by one of ordinary skill in the art.

[0059] This application provides a training sample augmentation method for mineral exploration prediction using fully convolutional neural networks.

[0060] Figure 1 Here is a flowchart of a training sample augmentation method for mineral exploration prediction using a fully convolutional neural network, as described in one embodiment of the application. Figure 1 As shown, a training sample augmentation method for mineral exploration prediction using a fully convolutional neural network according to an embodiment of the application includes steps S100 to S102:

[0061] S100: Obtain the model region and the geoscientific big data spatial database of the model region within the target mineral exploration area for generating training samples.

[0062] S101, determine the cumulative distribution function of the size and spatial location of the random cropping window based on the model region respectively;

[0063] S102, based on each cumulative distribution function, forms a combination of random parameters, and based on the combination of random parameters, cuts out training samples within the model region.

[0064] This application's embodiments are applied to the preliminary stage of a large-scale mineral exploration prediction model, serving as training samples for optimization of the target mineral exploration area. In this application's embodiments, the target mineral exploration area generally refers to areas within the target working area or areas with similar metallogenic geological structures.

[0065] Based on the data requirements of large-scale mineral exploration prediction models, the process of establishing a geoscientific big data spatial database according to embodiments of this application includes the following steps:

[0066] Acquire geoscientific big data within the target mineral exploration area; this includes geological data, mineral data, geophysical data, geochemical data, heavy mineral data, and remote sensing data.

[0067] Unify the spatial coordinate system and data type format of geoscientific big data, and generate a geoscientific big data spatial database composed of predicted elements and ore body annotations.

[0068] Depending on the data type, geoscientific big data exhibits differences in spatial coordinate systems and data format. Coordinate transformation can be used to unify the spatial coordinate systems of different geoscientific big data datasets, and data conversion can be used to unify the data format (e.g., *.tiff data format), thus forming a geoscientific big data spatial database for the target mineral exploration area.

[0069] The prediction elements include geological data, mineral data, geophysical data, geochemical data, heavy mineral data, and remote sensing data. Ore body labels are used to indicate mineral information for the corresponding area or location.

[0070] Based on this, the model region is a sub-region of the target mineral exploration area. In one embodiment, Figure 2 A flowchart of a preferred embodiment of a training sample augmentation method for mineral exploration prediction using a fully convolutional neural network is shown below. Figure 2 As shown, step S100, which involves acquiring a geoscientific big data spatial database of the model region within the target mineral exploration area for generating training samples, includes steps S200 and S201:

[0071] S200: Select an area within the target mineral exploration area that has reached the set exploration conditions as the initial area;

[0072] S201, by cropping the target mineral exploration area through the distribution range of the initial area, obtains the model area and its geoscientific big data spatial database.

[0073] The exploration conditions are set at a preliminary exploration level or higher, and areas with an exploration level of preliminary exploration level or higher are used as the initial area. A cropping method is employed, using the distribution range of the initial area to crop the target working area or similar metallogenic geological structure background areas (within which the same geoscientific big data spatial database must also be established), thereby obtaining the predicted elements of the model area and the geoscientific big data spatial database of ore body annotations. For example... Figure 3 As shown, this represents the distribution range of geoscientific big data in a certain model region, with a height of H (number of pixels) and a width of W (number of pixels).

[0074] Based on the determination of the model region, the cumulative distribution functions of the three parameters—size (d), spatial location (h, w) of the random cropping window for newly generated training samples within the model region—can be further determined. Therefore, the process of determining the cumulative distribution functions of the size and spatial location of the random cropping window based on the model region in step S101 is as follows:

[0075] Cumulative distribution function of the size of the random cropping window As shown in the following formula:

[0076] ;

[0077] in, This indicates the size of the randomly cropped window. This indicates the starting point for setting the size of the randomly cropped window. Indicates the height of the model region, Indicates the width of the model region;

[0078] like Figure 3 The diagram illustrates different training samples generated by different random cropping window sizes d. Theoretically, d can take any integer value from 1 to min(H,W) with equal probability, i.e., d ~ U(1,min(H,W)), meaning d follows a discrete uniform distribution from 1 to min(H,W). However, since the newly generated image must be able to extract target ore body information of a certain scale and shape, setting the starting point of the uniform distribution to 1 is inappropriate. The starting point of d needs to be determined by considering the data accuracy and the size and shape of the target object (ore body), requiring specific analysis for each specific problem. Therefore, the starting point is set to... That is: d~U( ,min(H,W)), In other words, d obeys arrive The discrete uniform distribution.

[0079] Cumulative distribution function of the height of the spatial location of the random cropping window As shown in the following formula:

[0080] ;

[0081] in, Indicates the height of the spatial position of the random cropping window;

[0082] like Figure 3 The diagram illustrates different training samples generated by different heights h at different spatial locations of random cropping windows. Since h can take any integer value from 0 to (Hd) with equal probability, i.e., h ~ U(0, (Hd)), That is, h follows a discrete uniform distribution from 0 to (Hd).

[0083] Cumulative distribution function of the width of the spatial location of the random cropping window As shown in the following formula:

[0084] ;

[0085] in, This indicates the width of the spatial position of the randomly cropped window.

[0086] like Figure 3 The diagram illustrates different training samples generated by different widths w at different spatial locations of the random cropping window. Since w can take any integer value from 0 to (Wd) with equal probability, i.e., w ~ U(0, (Wd)), it means that w follows a discrete uniform distribution from 0 to (Wd). .

[0087] Based on this, a uniform distribution of parameters is achieved in the cumulative distribution function, which facilitates the random parameter combination in step S101. Furthermore, the random parameter combination method in step S101 will not introduce any adverse factors to geoscientific big data.

[0088] As one of the preferred embodiments, such as Figure 2 As shown, step S101, which involves forming random parameter combinations based on each cumulative distribution function and then cropping training samples within the model region based on these random parameter combinations, includes steps S300 and S301:

[0089] S300, based on the cumulative distribution function, uses a random generator to generate random parameter combinations including the size and spatial position of the random cropping window;

[0090] S301 uses a cropping tool to crop training samples for each combination of random parameters.

[0091] Step S300, which involves generating a random parameter combination including the size and spatial location of a random cropping window based on the cumulative distribution function, includes the following steps:

[0092] The range of random numbers determined by the combination of random parameters is an integer between the interval [a, b].

[0093] Using Mason slew as a random number generator, pseudo-random numbers between [0,1) are generated by initializing a random seed;

[0094] Multiply the pseudo-random number by the interval width (b-a+1) and round down to obtain an integer offset of [0, ba]. Add a lower limit 'a' to the integer offset to ensure that the result includes both endpoints, and obtain random integers between the intervals [a, b]. Combine these random integers to form a random parameter combination.

[0095] In practical applications, given that d, h, and w all follow a discrete uniform distribution, and their values ​​are respectively any integer values ​​between (d_start, min(H, W)), (0, (Hd), and (0, (Wd)), we can use the inverse operations Φ-1(d), Φ-1(h), and Φ-1(w) of their cumulative distribution functions F(d), F(h), and F(w) as random generators. We can use a for loop to generate N sets of (d, h, w) combination parameters.

[0096] This is because the randint(a,b) discrete uniform distribution function achieves closed-interval integer randomization through pseudo-random number generation and mathematical mapping. Its implementation process is as follows:

[0097] (1) The range of values ​​for the random number is determined to be integers between the interval [a, b], where a and b are any integers ≥ 0, and b > a;

[0098] (2) The Mersenne Twister algorithm is used as the underlying random number generator. This algorithm generates pseudo-random numbers between [0,1) by initializing a random seed (based on the system time by default);

[0099] (3) Multiply the pseudo-random number by the interval width (b - a + 1) and round down to obtain the integer offset [0, ba]. Finally, add the lower limit a to ensure that the result includes both endpoints. This gives you a random integer between the interval [a, b], with each integer having a probability of 1 / 2. To ensure equal probability.

[0100] As one preferred embodiment, when random sampling is performed on parameter d, the value of a is d. 起 The value of b is min(H,W); when random sampling is performed on parameter h, the value of a is 0 and the value of b is Hd; when random sampling is performed on parameter w, the value of a is 0 and the value of b is Hd.

[0101] The process of cropping training samples for each random parameter combination using a cropping tool in step S301 includes the following steps:

[0102] Determine the starting row and column coordinates x1 and y1 of the cropping window;

[0103] The cropping window is directly mapped to the pixel row and column indices of the raster data, and the values ​​are assigned as x1=w, y1=h, width=d, height=d; where d is the size of the model region, width is the number of columns in the cropping window, and height is the number of rows in the cropping window.

[0104] In practical applications, based on N sets of (d, h, w) combined parameters, the Window(x1, y1, width, height) class from the rasterio.windows module in Python can be used as a cropping tool, where x1=w, y1=h, width=d, height=d. This ultimately generates N training samples for training and optimizing a mineral exploration prediction model based on a fully convolutional neural network. The Window(x1, y1, width, height) class is used to define the raster data operation area and can be used for image retrieval. Its workflow is as follows:

[0105] (1) x1 and y1 represent the starting row and column coordinates of the top left corner of the window (based on pixel index, counting from 0), and width and height specify the number of columns and rows of the window respectively, determining the size of the operation area;

[0106] (2) The window coordinate system directly corresponds to the pixel row and column index of the raster data. For example, Window(0, 0, 128,128) represents a 128×128 pixel area starting from the top left corner;

[0107] (3) Assign values ​​x1=w, y1=h, width=d, height=d. This will allow the sub-region with position (h,w) and size d to be retrieved from the model region of size (H,W), thereby generating a new training sample.

[0108] This application presents a training sample augmentation method for mineral exploration prediction using fully convolutional neural networks. The method acquires a model region for generating training samples and a geoscientific big data spatial database of the model region within the target mineral exploration area. It then determines the cumulative distribution functions for the size and spatial location of random clipping windows based on the model region. Random parameter combinations are formed based on these cumulative distribution functions, and training samples are clipped from the model region based on these random parameter combinations. This method avoids the problems of traditional training sample data augmentation in geoscientific big data, and can randomly generate any number of training samples of varying sizes and locations. This ensures the required number of samples and the randomness and diversity of samples for training large-scale mineral exploration prediction models, thereby better optimizing the mineral exploration prediction model based on fully convolutional neural networks and improving the quality of mineral exploration predictions.

[0109] This application also provides a training sample enhancement device for mineral exploration prediction using a fully convolutional neural network.

[0110] Figure 4 This is a structural diagram of a training sample augmentation device module for mineral exploration prediction using a fully convolutional neural network, as described in one embodiment of the application. Figure 4As shown, a training sample augmentation device for mineral exploration prediction using a fully convolutional neural network according to one embodiment of the application includes:

[0111] The region selection module 100 is used to acquire the model region for generating training samples and the geoscientific big data spatial database of the model region within the target mineral exploration area.

[0112] The function determination module 101 is used to determine the cumulative distribution functions of the size and spatial location of the random cropping window based on the model region, respectively;

[0113] The sample cropping module 102 is used to form a combination of random parameters based on each cumulative distribution function, and to crop training samples within the model region based on the combination of random parameters.

[0114] This application's embodiment of the training sample augmentation device for mineral exploration prediction using fully convolutional neural networks acquires a model region for generating training samples and a geoscientific big data spatial database of the model region within a target mineral exploration area. It determines the cumulative distribution functions of the size and spatial location of random clipping windows based on the model region. Based on these cumulative distribution functions, it forms random parameter combinations and clips training samples from the model region using these random parameter combinations. This avoids the problems of traditional training sample data augmentation in geoscientific big data, and can randomly generate any number of training samples of varying sizes and locations. This ensures the required number of samples and the randomness and diversity of samples for training large-scale mineral exploration prediction models, enabling better optimization of the fully convolutional neural network-based mineral exploration prediction model and improving the quality of mineral exploration prediction.

[0115] At least one embodiment of this application also provides a data control device. Figure 5 This is a schematic block diagram of a data control device provided for at least one embodiment of this application. For example, such as... Figure 5 As shown, the data control device 20 may include one or more memories 200 and one or more processors 201. The memories 200 are used to store computer-executable instructions non-transitory; the processors 201 are used to run the computer-executable instructions, which, when run by the processors 201, can cause the processors 201 to perform one or more steps in the training sample augmentation method for mineral prospecting prediction based on any embodiment of this application.

[0116] The specific implementation and explanation of each step of the training sample augmentation method for mineral exploration prediction using fully convolutional neural networks can be found in the relevant content of the above-mentioned embodiment of the training sample augmentation method for mineral exploration prediction using fully convolutional neural networks, and will not be repeated here. It should be noted that... Figure 5The components of the data control device 20 shown are merely exemplary and not limiting. The data control device 20 may have other components depending on the actual application requirements.

[0117] In one embodiment, the processor 201 and the memory 200 can communicate directly or indirectly with each other. For example, the processor 201 and the memory 200 can communicate via a network connection. The network can include wireless networks, wired networks, and / or any combination of wireless and wired networks; this application does not limit the type and function of the network. Alternatively, the processor 201 and the memory 200 can also communicate via a bus connection. The bus can be a Peripheral Component Interconnect Standard (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. For example, the processor 201 and the memory 200 can be located at a remote data server (cloud) or a distributed energy system (local), or at a client (e.g., a mobile device such as a mobile phone). For example, the processor 201 can be a central processing unit (CPU), a tensor processor (TPU), or a graphics processing unit (GPU), etc., with data processing and / or instruction execution capabilities, and can control other components in the data control device 20 to perform desired functions. The central processing unit (CPU) can be an x86 or ARM architecture, etc.

[0118] In one embodiment, memory 200 may include any combination of one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, flash memory, etc. One or more computer-executable instructions may be stored on the computer-readable storage medium, and processor 201 may execute these computer-executable instructions to implement various functions of data control device 20. Various application programs and various data, as well as various data used and / or generated by the application programs, may also be stored in memory 200.

[0119] It should be noted that the data control device 20 can achieve similar technical effects to the aforementioned training sample augmentation method for mineral exploration prediction using fully convolutional neural networks, and the repetitions will not be repeated here.

[0120] At least one embodiment of this application also provides a non-transitory computer-readable storage medium. Figure 6 This is a schematic diagram of a non-transitory computer-readable storage medium provided for at least one embodiment of this application. For example, such as... Figure 6 As shown, one or more computer-executable instructions 301 may be stored non-transitory on the non-transitory computer-readable storage medium 30. For example, when the computer-executable instructions 301 are executed by a computer, the computer may perform one or more steps in a training sample augmentation method for mineral prospecting prediction based on a fully convolutional neural network according to any embodiment of this application.

[0121] In one embodiment, the non-transitory computer-readable storage medium 30 can be applied to the data control device 20 described above, for example, it can be the memory 200 in the data control device 20.

[0122] In one embodiment, the description of the non-transitory computer-readable storage medium 30 can be found in the description of the memory 200 in the embodiment of the data control device 20, and will not be repeated hereafter.

[0123] It should be noted that the memory 200 stores different non-transient computer-executable instructions, and the data control device 20 corresponds to a firmware upgrade device. When the computer-executable instructions are executed by the processor 201, the processor 201 can perform one or more steps in the training sample augmentation method for mineral prospecting prediction based on any embodiment of this application.

[0124] The following points should be noted regarding this application:

[0125] (1) The accompanying drawings of the embodiments of this application only involve the structures involved in the embodiments of this application. Other structures can be referred to the general design.

[0126] (2) For clarity, the thickness and dimensions of layers or structures are enlarged in the accompanying drawings used to describe embodiments of the invention. It will be understood that when an element such as a layer, film, region, or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element, or there may be intermediate elements present.

[0127] (3) Where there is no conflict, the embodiments and features in the embodiments of this application can be combined with each other to obtain new embodiments. The above are only specific implementations of this application, but the protection scope of this application is not limited thereto, and the protection scope of this application shall be determined by the protection scope of the claims.

[0128] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A training sample augmentation method for mineral exploration prediction using fully convolutional neural networks, characterized in that, Including the following steps: Within the target mineral exploration area, acquire the model region used to generate training samples, as well as the geoscientific big data spatial database of the model region; The cumulative distribution functions for the size and spatial location of the random cropping window based on the model region are determined respectively, as follows: The cumulative distribution function of the size of the random cropping window As shown in the following formula: ； in, This indicates the size of the randomly cropped window. This indicates the starting point for setting the size. Indicates the height of the model region. This represents the width of the model region; The cumulative distribution function of the spatial position of the random cropping window As shown in the following formula: ； in, This indicates the height of the spatial position of the random cropping window; The cumulative distribution function of the width of the spatial location As shown in the following formula: ； in, This indicates the width of the spatial position of the randomly cropped window; The process includes the following steps: First, a combination of random parameters is formed based on each cumulative distribution function. Then, training samples are cropped from the model region based on this combination of random parameters. The range of random numbers determined by the combination of random parameters is an integer between the interval [a, b]. Using Mason slew as a random number generator, pseudo-random numbers between [0,1) are generated by initializing a random seed; where a and b are any integers ≥ 0, and b > a; Multiply the pseudo-random number by the interval width (b-a+1) and round down to obtain an integer offset of [0, ba]. Add a lower limit 'a' to the integer offset to ensure that the result includes both endpoints, and obtain a random integer between the interval [a, b]. Combine these random integers to form the random parameter combination. The training samples for each of the aforementioned random parameter combinations are cropped using a cropping tool.

2. The training sample augmentation method for mineral exploration prediction based on fully convolutional neural networks according to claim 1, characterized in that, The process of establishing the aforementioned geoscientific big data spatial database includes the following steps: Obtain geoscientific big data data within the target mineral exploration area; wherein, the geoscientific big data data includes geological data, mineral data, geophysical data, geochemical data, heavy mineral data, and remote sensing data; The spatial coordinate system and data type format of the aforementioned geoscientific big data are unified to generate a geoscientific big data spatial database composed of predicted elements and ore body annotations.

3. The training sample augmentation method for mineral exploration prediction based on fully convolutional neural networks according to claim 1, characterized in that, The process of acquiring a model region for generating training samples and a geoscientific big data spatial database of the model region within the target mineral exploration area includes the following steps: Within the target mineral exploration area, select an area where the exploration level meets the set exploration conditions as the initial area; The target mineral exploration area is cropped by the distribution range of the initial area to obtain the model area and its geoscientific big data spatial database.

4. The training sample augmentation method for mineral exploration prediction based on fully convolutional neural networks according to claim 1, characterized in that, The process of cropping training samples for each of the random parameter combinations using a cropping tool includes the following steps: Determine the starting row and column coordinates x1 and y1 of the cropping window; The cropping window is directly mapped to the pixel row and column indices of the raster data, and the values ​​are assigned as x1=w, y1=h, width=d, height=d; where d is the size of the random cropping window, width is the number of columns of the cropping window, and height is the number of rows of the cropping window.

5. A training sample augmentation device for mineral exploration prediction using fully convolutional neural networks, characterized in that, include: The region selection module is used to acquire, within the target mineral exploration area, a model region for generating training samples and a geoscientific big data spatial database of the model region; The function determination module is used to determine the cumulative distribution functions of the size and spatial location of the random cropping window based on the model region, as follows: The cumulative distribution function of the size of the random cropping window As shown in the following formula: ； in, This indicates the size of the randomly cropped window. This indicates the starting point for setting the size. Indicates the height of the model region. This represents the width of the model region; The cumulative distribution function of the spatial position of the random cropping window As shown in the following formula: ； in, This indicates the height of the spatial position of the random cropping window; The cumulative distribution function of the width of the spatial location As shown in the following formula: ； in, This indicates the width of the spatial position of the randomly cropped window; The sample cropping module is used to generate random parameter combinations based on each cumulative distribution function, and to crop training samples from the model region based on the random parameter combinations, including the following steps: The range of random numbers determined by the combination of random parameters is an integer between the interval [a, b]. Using Mason slew as a random number generator, pseudo-random numbers between [0,1) are generated by initializing a random seed; where a and b are any integers ≥ 0, and b > a; Multiply the pseudo-random number by the interval width (b-a+1) and round down to obtain an integer offset of [0, ba]. Add a lower limit 'a' to the integer offset to ensure that the result includes both endpoints, and obtain a random integer between the interval [a, b]. Combine these random integers to form the random parameter combination. The training samples for each of the aforementioned random parameter combinations are cropped using a cropping tool.

6. A non-transitory computer-readable storage medium, characterized in that, A non-transitory computer-readable storage medium stores computer-executable instructions that, when executed by a processor, implement the training sample augmentation method for mineral exploration prediction based on any one of claims 1 to 4.

7. A data control device, characterized in that, include: One or more memories that store computer-executable instructions non-transitory; One or more processors configured to run computer-executable instructions, wherein the computer-executable instructions are executed by the one or more processors to implement the training sample augmentation method for mineral prospecting prediction for fully convolutional neural networks as described in any one of claims 1 to 4.

Images