A mine target area prediction method, system, device and storage medium

Abstract

The application discloses a kind of mine target area prediction method, system, equipment and storage medium, including obtaining the element content chart of the region to be detected, and according to element content chart constructs chemical data set and is calculated to obtain element feature map, element feature map is carried out to obtain element scale feature map by inflation convolution operation, constructs a well-trained mine target area prediction source task network and multiple same structures target convolutional neural network, element scale feature map is input in corresponding target convolutional neural network with selected weight, to control target convolutional neural network in calculation by selected weight and carry out voting election to obtain final prediction result, can be combined multiple scale features to carry out mine target area prediction, while selectively migrate weight parameters in mine target area prediction source task network are used for each scale network training, solve the problem of few data, improve the accuracy and reliability of intelligent prospecting prediction.

Description

Technical Field

[0001] This invention relates to the field of mineral target area prediction technology, and in particular to a mineral target area prediction method, system, device and storage medium. Background Technology

[0002] Mineral resources are an important strategic resource for ensuring national stability and economic development. After years of geological exploration, surface minerals have gradually become scarce, leading to increased difficulty in mineral exploration. As a result, mineral exploration has begun to extend towards deeper parts of the earth's surface and overburdened areas.

[0003] In recent years, with the integration and development of artificial intelligence technology and geology, traditional geological prospecting has begun to transform into intelligent prospecting. Intelligent prospecting mainly utilizes machine learning technology to predict prospecting target areas by mining the correlation between geological variables and mineralization characteristics, typically requiring a large amount of data for training. However, in actual research, problems such as limited geological sample data and irregular mineral area characteristics affect the accuracy and reliability of intelligent prospecting predictions. Summary of the Invention

[0004] This invention aims to at least address the technical problems existing in the prior art. To this end, this invention proposes a method, system, device, and storage medium for predicting mineral target areas, which can combine features at multiple scales to predict mineral target areas, solving the problem of insufficient data and improving the accuracy and reliability of intelligent mineral exploration prediction.

[0005] In a first aspect, the present invention provides a method for predicting mineral target areas, comprising the following steps:

[0006] Obtain the elemental content map of the region to be detected, and construct a chemical dataset based on the elemental content map;

[0007] Elemental feature maps were calculated based on the chemical dataset.

[0008] The element feature maps are subjected to dilation and convolution to obtain element scale feature maps of different scales.

[0009] Construct a well-trained source task network for predicting mineral target areas and multiple target convolutional neural networks with the same number of scales as the element scale feature map. The source task network for predicting mineral target areas is used to calculate the selected weights, which are the weight set of all channels of the source task network for predicting mineral target areas. The network structure is the same between any two target convolutional neural networks, and the network structure is the same between the target convolutional neural networks and the source task network for predicting mineral target areas.

[0010] The element scale feature map and the selected weights are input into the corresponding target convolutional neural network, so as to control the calculation in the target convolutional neural network through the selected weights, and obtain the mineral target area prediction probability value of each target convolutional neural network.

[0011] The final prediction result is obtained by voting based on the target area prediction probability value of each target convolutional neural network.

[0012] According to embodiments of the present invention, at least the following technical effects are achieved:

[0013] This method obtains the elemental content map of the region to be detected, constructs a chemical dataset based on the elemental content map, calculates elemental feature maps based on the chemical dataset, and performs dilated convolution operations on the elemental feature maps to obtain elemental scale feature maps at different scales. It then constructs a well-trained source network for predicting the mineral target area and multiple target convolutional neural networks with the same number of scales as the elemental scale feature maps. The source network for predicting the mineral target area is used to calculate the selected weights, which are the set of weights for all channels of the source network. The network structure is identical between any two target convolutional neural networks. With the same network structure as the source task network for mineral target area prediction, the element-scale feature maps and selected weights are input into the corresponding target convolutional neural networks. The selected weights control the calculation in the target convolutional neural networks to obtain the mineral target area prediction probability value of each target convolutional neural network. The final prediction result is obtained by voting based on the mineral target area prediction probability value of each target convolutional neural network. It can combine features of multiple scales for mineral target area prediction, and selectively transfer the weight parameters in the source task network for mineral target area prediction to train each scale network. This solves the problem of insufficient data and improves the accuracy and reliability of intelligent mineral exploration prediction.

[0014] According to some embodiments of the present invention, obtaining an elemental content map of the region to be detected and constructing a chemical dataset based on the elemental content map includes:

[0015] The chemical data of the area to be detected is transformed into an elemental content map using the inverse distance weighted interpolation method.

[0016] The chemical dataset is constructed by cropping the element content map using a 256×256 sliding window with a step size of 128.

[0017] According to some embodiments of the present invention, the calculation formula for obtaining the elemental feature map based on the chemical dataset is as follows:

[0018]

[0019]

[0020] M = [u1, u2, ..., u n ]

[0021]

[0022] Where D is the chemical dataset, Let h be the feature vector of the h-th data sample with the i-th element. Let H be the label of the h-th data sample of the i-th element, H be the number of data samples of the i-th element, n be the number of elements, and u be the label of the h-th data sample of the i-th element. i Let i be the weight corresponding to the i-th element in the mask. Let h be the feature vector of the h-th data sample whose i-th element has undergone the mask operation.

[0023] According to some embodiments of the present invention, performing dilation convolution on the element feature map to obtain element scale feature maps of different scales includes:

[0024] Pre-set the expansion coefficient ρ = {ρ1, ρ2, ..., ρ m The formula for calculating element-scale feature maps at different scales by performing dilation and convolution on the element feature map is as follows:

[0025] r = S-1

[0026]

[0027] in, For the i-th element, the t-th element v The t-th data sample after dilation convolution v The element scale feature map of a data sample, where u,g is the position (p) of the element feature map. row ,p col The nearest neighboring units, For the i-th element, the t-th element v The weights of the data samples are given by o(·,·), which are the positional weights on the dilated convolution kernel, S is the size of the dilated convolution kernel, and r is the extreme value of the nearest neighbor unit.

[0028] According to some embodiments of the present invention, the construction of a well-trained source task network for predicting mineral target areas and multiple target convolutional neural networks having the same number of scales as the element scale feature map, wherein the source task network for predicting mineral target areas is used to calculate selected weights, the selected weights being the set of weights for all channels of the source task network for predicting mineral target areas, the network structure being identical between every two target convolutional neural networks, and the network structure being identical between the target convolutional neural networks and the source task network for predicting mineral target areas, includes:

[0029] Step S11: Pre-set the target area prediction source task network N. S and multiple target convolutional neural networks All networks share the same network structure and have l convolutional layers. The number of input and output channels for the convolutional kernel in the l-th layer are M and M, respectively. l and M l+1 The target area prediction source task network N S The convolution kernel of the l-th layer is The target convolutional neural network N T All convolutional kernels in the l-th layer are defined as And the tth v The convolutional kernels of the target convolutional neural network and the target area prediction source task network N S The convolutional kernels are consistent with those of the target convolutional neural network. sparse matrices with the same convolution kernel size

[0030] Step S12: Based on the sparse matrix and the target area prediction source task network N... S The convolution kernels are used to perform Hadamard product calculations to obtain the selected weights. The weights are then determined based on the sparse matrix and the target area prediction source task network N. S The formula for calculating the Hadamard product using the convolution kernel is:

[0031]

[0032]

[0033] in, Here, represents the weight of the j-th channel in the l-th layer of the target area prediction source task network after selection, and ⊙ represents the Hadamard product. The kernel size is K×K;

[0034] According to some embodiments of the present invention, the step of inputting the element scale feature map and the selected weights into the corresponding target convolutional neural network, so as to control the calculation in the target convolutional neural network through the selected weights, to obtain the target convolutional neural network prediction probability value of each target convolutional neural network, includes:

[0035] Step S21: Calculate the convolutional output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network. The calculation formula for calculating the convolutional output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network is as follows:

[0036]

[0037] in, For the tth v The convolution output value of a target convolutional neural network, where * represents the convolution operation. For the tth v The convolution kernel of the j-th channel of the l-th layer of a target convolutional neural network, where b is a pre-set convolution bias.

[0038] Step S22: Sort the target convolutional neural networks according to the size of the input element-scale feature maps. Guide the target convolutional neural networks with input element-scale feature maps of smaller scales to learn a top-down self-distillation loss. The calculation formula for the top-down self-distillation loss obtained by guiding the target convolutional neural networks with input element-scale feature maps of smaller scales to learn a top-down self-distillation loss is as follows:

[0039]

[0040] in, Represents the target convolutional neural network Self-distillation loss, The input element-scale feature map is larger than the target convolutional neural network. The input element scale feature map is the largest of the t-th elements. j A target convolutional neural network, for weights, for The weights, f() is the softmax operation;

[0041] Step S23: Calculate the classification loss of the target convolutional neural network based on the element-scale feature map of the target convolutional neural network, wherein the calculation formula for the classification loss of the target convolutional neural network based on the element-scale feature map of the target convolutional neural network is:

[0042]

[0043] in, The classification loss is L(), where L() is the cross-entropy calculation.

[0044] Step S24: Calculate the objective function value based on the classification loss and the self-distillation loss. If the preset maximum number of iterations is reached or the objective function value reaches a preset value, output the target area prediction probability value of each target convolutional neural network based on the convolution output value. If the objective function value is less than the preset value, update the convolution kernel size of the target convolutional neural network based on the objective function value, and return to step S12 with the updated kernel size until the preset maximum number of iterations is reached or the objective function value reaches a preset value. Then, output the target area prediction probability value of each target convolutional neural network based on the convolution output value. The calculation formula for the objective function value calculated based on the classification loss and the self-distillation loss is:

[0045]

[0046] Among them, L total Let β be the objective function, and β be the preset self-distillation loss coefficient.

[0047] A second aspect of the present invention provides a mineral target area prediction system, the mineral target area prediction system comprising:

[0048] The data acquisition module is used to acquire the elemental content map of the area to be detected and to construct a chemical dataset based on the elemental content map.

[0049] The element feature calculation module is used to calculate the element feature map based on the chemical dataset.

[0050] The element feature dilation module is used to perform dilation convolution operations on the element feature map to obtain element scale feature maps of different scales.

[0051] The network construction module is used to construct a well-trained source task network for predicting mineral target areas and multiple target convolutional neural networks with the same number of scales as the element scale feature map. The source task network for predicting mineral target areas is used to calculate the selected weights, which are the weight set of all channels of the source task network for predicting mineral target areas. The network structure is the same between any two target convolutional neural networks, and the network structure is the same between the target convolutional neural networks and the source task network for predicting mineral target areas.

[0052] The target area prediction module is used to input the element scale feature map and the selected weights into the corresponding target convolutional neural network, so as to control the target convolutional neural network to perform calculations through the selected weights, and obtain the target area prediction probability value of each target convolutional neural network.

[0053] The prediction result output module is used to obtain the final prediction result by voting based on the prediction probability value of the target area of ​​each target convolutional neural network.

[0054] This system acquires the elemental content map of the region to be detected, constructs a chemical dataset based on the elemental content map, calculates elemental feature maps based on the chemical dataset, and performs dilated convolution operations on the elemental feature maps to obtain elemental scale feature maps at different scales. It then constructs a well-trained mineral target area prediction source task network and multiple target convolutional neural networks with the same number of scales as the elemental scale feature maps. The mineral target area prediction source task network is used to calculate the selected weights, which are the set of weights for all channels of the network. The network structure is identical between any two target convolutional neural networks. With the same network structure as the source task network for mineral target area prediction, the element-scale feature maps and selected weights are input into the corresponding target convolutional neural networks. The selected weights control the calculation in the target convolutional neural networks to obtain the mineral target area prediction probability value of each target convolutional neural network. The final prediction result is obtained by voting based on the mineral target area prediction probability value of each target convolutional neural network. It can combine features of multiple scales for mineral target area prediction, and selectively transfer the weight parameters in the source task network for mineral target area prediction to train each scale network. This solves the problem of insufficient data and improves the accuracy and reliability of intelligent mineral exploration prediction.

[0055] According to some embodiments of the present invention, the target area prediction module further includes:

[0056] The convolution output value calculation module is used to calculate the convolution output value of the target convolutional neural network based on the selected weights of the j-th channel of the l-th layer of the target area prediction source task network. The calculation formula for calculating the convolution output value of the target convolutional neural network based on the selected weights of the j-th channel of the l-th layer of the target area prediction source task network is as follows:

[0057]

[0058] in, For the tth v The convolution output value of a target convolutional neural network, where * represents the convolution operation. For the tth v The convolution kernel of the j-th channel of the l-th layer of a target convolutional neural network, where b is a pre-set convolution bias.

[0059] The self-distillation loss calculation module is used to sort the target convolutional neural network according to the size of the input element-scale feature maps, and guide the target convolutional neural network with small-scale element-scale feature maps to learn a top-down self-distillation loss. The calculation formula for the top-down self-distillation loss obtained by guiding the target convolutional neural network with small-scale element-scale feature maps to learn a top-down self-distillation loss is as follows:

[0060]

[0061] in, Represents the target convolutional neural network Self-distillation loss, The input element-scale feature map is larger than the target convolutional neural network. The input element scale feature map is the largest of the t-th elements. j A target convolutional neural network, for weights, for The weights, f() is the softmax operation;

[0062] The classification loss calculation module is used to calculate the classification loss of the target convolutional neural network based on the element-scale feature map of the target convolutional neural network. The calculation formula for the classification loss of the target convolutional neural network based on the element-scale feature map is as follows:

[0063]

[0064] in, The classification loss is L(), where L() is the cross-entropy calculation.

[0065] The iterative module is used to calculate the objective function value based on the classification loss and the self-distillation loss. If the preset maximum number of iterations is reached or the objective function value reaches a preset value, the predicted probability value of the target area for each target convolutional neural network is output based on the convolution output value. If the objective function value is less than the preset value, the convolution kernel size of the target convolutional neural network is updated based on the objective function value, and the updated kernel size is returned to continue executing the Hadamard product calculation module until the preset maximum number of iterations is reached or the objective function value reaches the preset value. Then, the predicted probability value of the target area for each target convolutional neural network is output based on the convolution output value. The calculation formula for the objective function value calculated based on the classification loss and the self-distillation loss is as follows:

[0066]

[0067] Among them, L total Let β be the objective function, and β be the preset self-distillation loss coefficient.

[0068] A third aspect of the present invention provides an electronic device for predicting mineral target areas, comprising at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, the instructions being executed by the at least one control processor to enable the at least one control processor to perform the above-described method for predicting mineral target areas.

[0069] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the above-described mineral target area prediction method.

[0070] It should be noted that the beneficial effects of the second to fourth aspects of the present invention compared with the prior art are the same as the beneficial effects of the above-described mineral target area prediction system compared with the prior art, and will not be described in detail here.

[0071] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0072] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0073] Figure 1 This is a flowchart of a mineral target area prediction method according to an embodiment of the present invention;

[0074] Figure 2 This is a detailed framework diagram of a mineral target area prediction method according to an embodiment of the present invention;

[0075] Figure 3 This is an experimental result diagram of a mineral target area prediction method according to an embodiment of the present invention;

[0076] Figure 4 This is a flowchart of a mineral target area prediction system according to an embodiment of the present invention. Detailed Implementation

[0077] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0078] In the description of this invention, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of technical features indicated, or implicitly indicating the order of the technical features indicated.

[0079] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the drawings and are only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0080] In the description of this invention, it should be noted that, unless otherwise explicitly defined, terms such as "setting," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0081] In recent years, with the integration and development of artificial intelligence technology and geology, traditional geological prospecting has begun to transform into intelligent prospecting. Intelligent prospecting mainly utilizes machine learning technology to predict prospecting target areas by mining the correlation between geological variables and mineralization characteristics, typically requiring a large amount of data for training. However, in actual research, problems such as limited geological sample data and irregular mineral area characteristics affect the accuracy and reliability of intelligent prospecting predictions.

[0082] To address the aforementioned technical deficiencies, referring to... Figure 1 The present invention also provides a method for predicting mineral target areas, comprising:

[0083] Step S101: Obtain the element content map of the region to be detected, and construct a chemical dataset based on the element content map.

[0084] Step S102: Calculate the elemental characteristic map based on the chemical dataset.

[0085] Step S103: Perform dilation convolution on the element feature maps to obtain element scale feature maps of different scales.

[0086] Step S104: Construct a well-trained source task network for predicting the ore target area and multiple target convolutional neural networks with the same number of scales as the element scale feature maps. The source task network for predicting the ore target area is used to calculate the selected weights, which are the set of weights for all channels of the source task network for predicting the ore target area. The network structure is the same between any two target convolutional neural networks, and the network structure is the same between the target convolutional neural networks and the source task network for predicting the ore target area.

[0087] Step S105: Input the element scale feature map and the selected weights into the corresponding target convolutional neural network, so as to control the calculation in the target convolutional neural network through the selected weights, and obtain the mining target area prediction probability value of each target convolutional neural network.

[0088] Step S106: Based on the predicted probability value of the target area of ​​each target convolutional neural network, a voting election is conducted to obtain the final prediction result.

[0089] This method obtains the elemental content map of the region to be detected, constructs a chemical dataset based on the elemental content map, calculates elemental feature maps based on the chemical dataset, and performs dilated convolution operations on the elemental feature maps to obtain elemental scale feature maps at different scales. It then constructs a well-trained source network for predicting the mineral target area and multiple target convolutional neural networks with the same number of scales as the elemental scale feature maps. The source network for predicting the mineral target area is used to calculate the selected weights, which are the set of weights for all channels of the source network. The network structure is identical between any two target convolutional neural networks, and the target convolutional neural networks are... The source task networks for predicting mineral target areas share the same network structure. The element-scale feature maps and selected weights are input into the corresponding target convolutional neural networks. The selected weights control the calculations within the target convolutional neural networks, resulting in the mineral target area prediction probability value for each target convolutional neural network. Based on the mineral target area prediction probability values ​​of each target convolutional neural network, a voting process is conducted to obtain the final prediction result. This approach can combine features from multiple scales for mineral target area prediction and selectively transfer weight parameters from the source task networks for training each scale network. This solves the problem of insufficient data and improves the accuracy and reliability of intelligent mineral exploration prediction.

[0090] In some embodiments, step S101 may include, but is not limited to, steps S201 to S202:

[0091] Step S201: Convert the chemical data of the area to be detected into an elemental content map using the inverse distance weighted interpolation method.

[0092] Step S202: The element content map is cropped using a 256×256 sliding window with a step size of 128 to construct a chemical dataset.

[0093] In some embodiments, the formula for calculating the elemental feature map based on the chemical dataset is as follows:

[0094]

[0095]

[0096] M = [u1, u2, ..., u n ]

[0097]

[0098] Where D is the chemical dataset, Let h be the feature vector of the h-th data sample with the i-th element. Let H be the label of the h-th data sample of the i-th element, H be the number of data samples of the i-th element, n be the number of elements, and u be the label of the h-th data sample of the i-th element. i Let i be the weight corresponding to the i-th element in the mask. Let h be the feature vector of the h-th data sample whose i-th element has undergone the mask operation.

[0099] In some embodiments, step S103 may include, but is not limited to, step S301:

[0100] Step S301: Pre-set the expansion coefficient ρ={ρ1,ρ2,…,ρ m The formula for calculating element-scale feature maps at different scales by performing dilation and convolution on the element feature maps is as follows:

[0101] r = S-1

[0102]

[0103] in, For the i-th element, the t-th element v The t-th data sample after dilation convolution v The element-scale feature map of each data sample, where u,g is the position of the element-scale feature map (p row ,p col The nearest neighboring units, For the i-th element, the t-th element v The weights of each data sample are given by o(·,·), which are the positional weights on the dilated convolution kernel, S is the size of the dilated convolution kernel, and r is the extreme value of the nearest neighbor unit.

[0104] In some embodiments, step S104 may include, but is not limited to, steps S401 to S402:

[0105] Step S401: Pre-set the target area prediction source task network NS and multiple target convolutional neural networks All networks share the same network structure and have l convolutional layers. The number of input and output channels for the convolutional kernel in the l-th layer are M and M, respectively. l and M l+1 Mineral target area prediction source task network N S The convolution kernel of the l-th layer is Target Convolutional Neural Network N T All convolutional kernels in the l-th layer are defined as And the tth v The convolutional kernels of a target convolutional neural network and the target area prediction source task network N s The convolutional kernels are consistent with those of the target convolutional neural network. sparse matrices with the same convolution kernel size

[0106] Step S402: Based on the sparse matrix and the target area prediction source task network N s The convolution kernels are used to calculate the Hadamard product, and the selected weights are obtained. Based on the sparse matrix and the target area prediction source task network N... s The formula for calculating the Hadamard product using the convolution kernel is:

[0107]

[0108]

[0109] in, Let φ be the weight of the j-th channel in the l-th layer of the target area prediction source task network after selection, and ⊙ be the Hadamard product. The kernel size is K×K.

[0110] In some embodiments, step S105 may include, but is not limited to, steps S501 to S504:

[0111] Step S501: Calculate the convolutional output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network. The formula for calculating the convolutional output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network is as follows:

[0112]

[0113] in, For the tth v The convolution output value of a target convolutional neural network, where * represents the convolution operation. For the tth vThe convolution kernel of the j-th channel of the l-th layer of a target convolutional neural network, where b is a pre-set convolution bias.

[0114] Step S502: Sort the target convolutional neural networks according to the size of the input element-scale feature maps. Use the target convolutional neural network with large-scale input element-scale feature maps to guide the target convolutional neural network with small-scale input element-scale feature maps to learn a top-down self-distillation loss. The formula for calculating the top-down self-distillation loss obtained by using the target convolutional neural network with large-scale input element-scale feature maps to guide the target convolutional neural network with small-scale input element-scale feature maps is as follows:

[0115]

[0116] in, Represents the target convolutional neural network Self-distillation loss, The input element-scale feature map is larger than the target convolutional neural network. The input element scale feature map is the largest of the t-th elements. j A target convolutional neural network, for weights, for The weights are f(), which is the softmax operation.

[0117] Step S503: Calculate the classification loss of the target convolutional neural network based on the element-scale feature map of the target convolutional neural network. The formula for calculating the classification loss of the target convolutional neural network based on the element-scale feature map is as follows:

[0118]

[0119] in, L() is the classification loss, and L is the cross-entropy calculation.

[0120] Step S504: Calculate the objective function value based on the classification loss and self-distillation loss. If the preset maximum number of iterations is reached or the objective function value reaches a preset value, output the target area prediction probability value for each target convolutional neural network based on the convolution output value. If the objective function value is less than the preset value, update the convolution kernel size of the target convolutional neural network based on the objective function value, and return to step S402 with the updated kernel size until the preset maximum number of iterations is reached or the objective function value reaches a preset value. Then, output the target area prediction probability value for each target convolutional neural network based on the convolution output value. The formula for calculating the objective function value based on the classification loss and self-distillation loss is as follows:

[0121]

[0122] Among them, L total Let β be the objective function, and β be the preset self-distillation loss coefficient.

[0123] Reference Figure 2 To facilitate understanding by those skilled in the art, a set of preferred embodiments is provided below:

[0124] I. Data Acquisition and Processing:

[0125] The data in this invention is for intelligent prediction of mineral exploration target areas based on geochemical data from Pangxi Cave. First, the chemical data of the area to be detected is converted into an elemental content map by inverse distance weighted interpolation. Then, the elemental content map is cropped by a 256×256 sliding window with a step size of 128 to construct a chemical dataset.

[0126] The formula for calculating the elemental characteristic map based on the chemical dataset is as follows:

[0127]

[0128]

[0129] M = [u1, u2, ..., u n ]

[0130]

[0131] Where D is the chemical dataset, Let h be the feature vector of the h-th data sample with the i-th element. Let H be the label of the h-th data sample of the i-th element, H be the number of data samples of the i-th element, n be the number of elements, and u be the label of the h-th data sample of the i-th element. i Let i be the weight corresponding to the i-th element in the mask. Let h be the feature vector of the h-th data sample whose i-th element has undergone the mask operation.

[0132] II. Feature map dilation:

[0133] Pre-set the expansion coefficient ρ = {ρ1, ρ2, ..., ρ m The formula for calculating element-scale feature maps at different scales by performing dilation and convolution on the element feature maps is as follows:

[0134] r = S-1

[0135]

[0136] in, For the i-th element, the t-th elementv The t-th data sample after dilation convolution v The element-scale feature map of each data sample, where u,g is the position of the element-scale feature map (p row ,p col The nearest neighboring units, For the i-th element, the t-th element v The weights of each data sample are given by o(·,·), which are the positional weights on the dilated convolution kernel, S is the size of the dilated convolution kernel, and r is the extremum of the nearest neighbor unit. For example, for a dilated convolution kernel of size 3, the nearest neighbor unit of u,g is 1.

[0137] III. Weight Parameter Transfer and Knowledge Learning:

[0138] Step A1: Pre-set the target area prediction source task network N S and multiple target convolutional neural networks All networks share the same network structure and have l convolutional layers. The number of input and output channels for the convolutional kernel in the l-th layer are M and M, respectively. l and M l+1 Mineral target area prediction source task network N S The convolution kernel of the l-th layer is Target Convolutional Neural Network N T All convolutional kernels in the l-th layer are defined as And the tth v The convolutional kernels of a target convolutional neural network and the target area prediction source task network N S The convolutional kernels are consistent with those of the target convolutional neural network. sparse matrices with the same convolution kernel size

[0139] Step A2: Based on the sparse matrix and the target area prediction source task network N S The convolution kernels are used to calculate the Hadamard product, and the selected weights are obtained. Based on the sparse matrix and the target area prediction source task network N... S The formula for calculating the Hadamard product using the convolution kernel is:

[0140]

[0141]

[0142] in, Let φ be the weight of the j-th channel in the l-th layer of the target area prediction source task network after selection, and ⊙ be the Hadamard product. The kernel size is K×K.

[0143] Step A3: Calculate the convolutional output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network. The formula for calculating the convolutional output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network is as follows:

[0144]

[0145] in, For the tth v The convolution output value of a target convolutional neural network, where * represents the convolution operation. For the tth v The convolution kernel of the j-th channel of the l-th layer of a target convolutional neural network, where b is a pre-set convolution bias.

[0146] Step A4: Sort the target convolutional neural networks according to the size of the input element-scale feature maps. Use the target convolutional neural network with large-scale input element-scale feature maps to guide the target convolutional neural network with small-scale input element-scale feature maps to learn a top-down self-distillation loss. The formula for calculating the top-down self-distillation loss obtained by using the target convolutional neural network with large-scale input element-scale feature maps to guide the target convolutional neural network with small-scale input element-scale feature maps is as follows:

[0147]

[0148] in, Represents the target convolutional neural network Self-distillation loss, The input element-scale feature map is larger than the target convolutional neural network. The input element scale feature map is the largest of the t-th elements. j A target convolutional neural network, for weights, for The weights are f(), which is the softmax operation.

[0149] Step A5: Calculate the classification loss of the target convolutional neural network based on its element-scale feature map. The formula for calculating the classification loss of the target convolutional neural network based on its element-scale feature map is as follows:

[0150]

[0151] in, L() is the classification loss, and L is the cross-entropy calculation.

[0152] Step A6: Calculate the objective function value based on the classification loss and self-distillation loss. If the preset maximum number of iterations is reached or the objective function value reaches a preset value, output the target area prediction probability value for each target convolutional neural network based on the convolution output value. If the objective function value is less than the preset value, update the convolution kernel size of the target convolutional neural network based on the objective function value, and return to step A2 to continue executing until the preset maximum number of iterations is reached or the objective function value reaches a preset value. Then, output the target area prediction probability value for each target convolutional neural network based on the convolution output value. The formula for calculating the objective function value based on the classification loss and self-distillation loss is as follows:

[0153]

[0154] Among them, L total Let β be the objective function, and β be the preset self-distillation loss coefficient.

[0155] IV. Experimental Data Output:

[0156] Reference Figure 3 , Figure 3 This is the output image of the experimental results. "Predict no-mining area" represents the predicted no-mining area, "predict mining area" represents the predicted mining area, and "mining area" represents the mining area.

[0157] Additionally, refer to Figure 4 One embodiment of the present invention provides a mineral target area prediction system, including a data acquisition module 1100, an element feature calculation module 1200, an element feature expansion module 1300, a network construction module 1400, a mineral target area prediction module 1500, and a prediction result output module 1600, wherein:

[0158] The data acquisition module 1100 is used to acquire the elemental content map of the area to be detected and to construct a chemical dataset based on the elemental content map.

[0159] The element feature calculation module 1200 is used to calculate element feature maps based on chemical datasets.

[0160] The element feature dilation module 1300 is used to perform dilation convolution operations on the element feature map to obtain element scale feature maps of different scales.

[0161] The network construction module 1400 is used to construct a well-trained source task network for predicting mineral target areas and multiple target convolutional neural networks with the same number of scales as the element-scale feature maps. The source task network for predicting mineral target areas is used to calculate the selected weights, which are the set of weights for all channels of the source task network for predicting mineral target areas. The network structure is the same between any two target convolutional neural networks, and the network structure is the same between the target convolutional neural networks and the source task network for predicting mineral target areas.

[0162] The target area prediction module 1500 is used to input the element scale feature map and the selected weights into the corresponding target convolutional neural network, so as to control the target convolutional neural network to perform calculations and obtain the target area prediction probability value of each target convolutional neural network.

[0163] The prediction result output module 1600 is used to vote and select the final prediction result based on the prediction probability value of the target area of ​​each target convolutional neural network.

[0164] This system acquires the elemental content map of the region to be detected, constructs a chemical dataset based on the elemental content map, calculates elemental feature maps based on the chemical dataset, and performs dilated convolution operations on the elemental feature maps to obtain elemental scale feature maps at different scales. It then constructs a well-trained source network for predicting the mineral target area and multiple target convolutional neural networks with the same number of scales as the elemental scale feature maps. The source network for predicting the mineral target area is used to calculate the selected weights, which are the set of weights for all channels of the source network. The network structure is identical between any two target convolutional neural networks, and the target convolutional neural networks are... The source task networks for predicting mineral target areas share the same network structure. The element-scale feature maps and selected weights are input into the corresponding target convolutional neural networks. The selected weights control the calculations within the target convolutional neural networks, resulting in the mineral target area prediction probability value for each target convolutional neural network. Based on the mineral target area prediction probability values ​​of each target convolutional neural network, a voting process is conducted to obtain the final prediction result. This approach can combine features from multiple scales for mineral target area prediction and selectively transfer weight parameters from the source task networks for training each scale network. This solves the problem of insufficient data and improves the accuracy and reliability of intelligent mineral exploration prediction.

[0165] In some embodiments, the target area prediction module further includes:

[0166] The convolution output value calculation module is used to calculate the convolution output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network. The formula for calculating the convolution output value of the target convolutional neural network based on the weights selected for the j-th channel of the l-th layer of the target area prediction source task network is as follows:

[0167]

[0168] in, For the tth v The convolution output value of a target convolutional neural network, where * represents the convolution operation. For the tth v The convolution kernel of the j-th channel of the l-th layer of a target convolutional neural network, where b is a pre-set convolution bias.

[0169] The self-distillation loss calculation module sorts the target convolutional neural networks according to the size of the input element-scale feature maps. It then guides target convolutional neural networks with smaller input element-scale feature maps to learn a top-down self-distillation loss. The formula for calculating the top-down self-distillation loss is as follows:

[0170]

[0171] in, Represents the target convolutional neural network Self-distillation loss, The input element-scale feature map is larger than the target convolutional neural network. The input element scale feature map is the largest of the t-th elements. j A target convolutional neural network, for weights, for The weights, f() is the softmax operation;

[0172] The classification loss calculation module is used to calculate the classification loss of the target convolutional neural network based on its element-scale feature map. The formula for calculating the classification loss based on the element-scale feature map of the target convolutional neural network is as follows:

[0173]

[0174] in, For classification loss, L() is the cross-entropy calculation;

[0175] The iterative module calculates the objective function value based on the classification loss and self-distillation loss. If the preset maximum number of iterations is reached or the objective function value reaches a preset value, the predicted probability value of the target area for each target convolutional neural network is output based on the convolution output value. If the objective function value is less than the preset value, the convolution kernel size of the target convolutional neural network is updated based on the objective function value, and the updated kernel size is returned to continue executing the Hadamard product calculation module until the preset maximum number of iterations is reached or the objective function value reaches the preset value. Then, the predicted probability value of the target area for each target convolutional neural network is output based on the convolution output value. The formula for calculating the objective function value based on the classification loss and self-distillation loss is as follows:

[0176]

[0177] Among them, L total Let β be the objective function, and β be the preset self-distillation loss coefficient.

[0178] It should be noted that this system embodiment is based on the same inventive concept as the above system embodiment. Therefore, the relevant content of the above method embodiment is also applicable to this system embodiment, and will not be repeated here.

[0179] This application also provides an electronic device for predicting mineral target areas, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the mineral target area prediction method as described above.

[0180] The processor and memory can be connected via a bus or other means.

[0181] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0182] The non-transient software program and instructions required to implement the mineral target area prediction method of the above embodiments are stored in memory. When executed by a processor, the mineral target area prediction method in the above embodiments is executed, for example, the method described above is executed. Figure 1 The method steps S101 to S106.

[0183] This application also provides a computer-readable storage medium storing computer-executable instructions for performing the above-described mineral target area prediction method.

[0184] The computer-readable storage medium stores computer-executable instructions that are executed by a processor or controller, for example, by a processor in the above-described electronic device embodiment, causing the processor to perform the target area prediction method in the above-described embodiment, for example, to perform the above-described... Figure 1 The method steps S101 to S106.

[0185] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program units, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program units, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0186] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for predicting mineral target areas, characterized in that, The mineral target area prediction method includes: Obtain the elemental content map of the region to be detected, and construct a chemical dataset based on the elemental content map; Elemental feature maps were calculated based on the chemical dataset. The element feature maps are subjected to dilation and convolution to obtain element scale feature maps of different scales. Construct a well-trained source task network for predicting mineral target areas and multiple target convolutional neural networks with the same number of scales as the element scale feature map. The source task network for predicting mineral target areas is used to calculate the selected weights, which are the weight set of all channels of the source task network. The network structure is identical between any two target convolutional neural networks, and the network structure is identical between the target convolutional neural networks and the source task network for predicting mineral target areas. Specifically: Step S11: Pre-set the target area prediction source task network. and multiple target convolutional neural networks Among them, all networks have the same network structure and have The nth convolutional layer, the nth The number of input and output channels of the convolution kernel in the layer are respectively and The mineral target area prediction source task network The The convolution kernel of the layer is The target convolutional neural network The All convolutional kernels of a layer are defined as And the first The convolutional kernels of the target convolutional neural network and the target area prediction source task network The convolutional kernels are consistent with those of the target convolutional neural network. sparse matrices with the same convolution kernel size ; Step S12: Based on the sparse matrix and the target area prediction source task network... The convolution kernels are used to perform Hadamard product calculations to obtain the selected weights. The weights are then determined based on the sparse matrix and the target area prediction source task network. The formula for calculating the Hadamard product using the convolution kernel is: in, For the target area prediction source task network, the first The first layer Weighting after selecting each channel For Hadama accumulation, The kernel size is ; The element scale feature map and the selected weights are input into the corresponding target convolutional neural network, so as to control the calculation in the target convolutional neural network through the selected weights, and obtain the mineral target area prediction probability value of each target convolutional neural network. The final prediction result is obtained by voting based on the target area prediction probability value of each target convolutional neural network.

2. The method for predicting a mineral target area according to claim 1, characterized in that, The step of obtaining the elemental content map of the region to be detected and constructing a chemical dataset based on the elemental content map includes: The chemical data of the area to be detected is transformed into an elemental content map using the inverse distance weighted interpolation method. The chemical dataset is constructed by cropping the element content map using a 256×256 sliding window with a step size of 128.

3. The method for predicting a mineral target area according to claim 2, characterized in that, The calculation formula for obtaining the elemental feature map based on the chemical dataset is as follows: in, For the chemical dataset, For the first The element of the first Feature vectors of data samples For the first The element of the first Labels for each data sample For the first The number of data samples for each element The number of elements, To be in the mask and the first The weights corresponding to each element. For the first after mask operation The element of the first The feature vector of each data sample.

4. The method for predicting a mineral target area according to claim 3, characterized in that, The step of performing dilation convolution on the element feature map to obtain element scale feature maps of different scales includes: Preset expansion coefficient The calculation formula for performing dilation and convolution on the element feature map to obtain element scale feature maps of different scales is as follows: in, For the first The element of the first The data sample after dilation convolution is the first... The element-scale feature map of a data sample, where u,g is the position of the element-scale feature map. Neighboring units, For the first The element of the first The weights of each data sample, The position weights on the dilated convolution kernel are... The size of the dilated convolution kernel. is the extreme value of the nearest neighbor unit.

5. The method for predicting a mineral target area according to claim 4, characterized in that, The step of inputting the element scale feature map and the selected weights into the corresponding target convolutional neural network, so as to control the calculation in the target convolutional neural network through the selected weights, to obtain the mineral target area prediction probability value of each target convolutional neural network, includes: Step S21: Based on the predicted source task network of the mining target area... The first layer After selecting the channels, the weights are calculated to obtain the convolutional output value of the target convolutional neural network, wherein the value is calculated based on the weights of the target area prediction source task network. The first layer The formula for calculating the convolution output value of the target convolutional neural network after selecting the weights of each channel is as follows: in, For the first The convolution output value of a target convolutional neural network. For convolution operations, For the first The nth target convolutional neural network The first layer A convolutional kernel with 1 channel, The pre-set convolution bias; Step S22: Sort the target convolutional neural networks according to the size of the input element-scale feature maps. Guide the target convolutional neural networks with input element-scale feature maps of smaller scales to learn a top-down self-distillation loss. The calculation formula for the top-down self-distillation loss obtained by guiding the target convolutional neural networks with input element-scale feature maps of smaller scales to learn a top-down self-distillation loss is as follows: in, Represents the target convolutional neural network Self-distillation loss, The input element-scale feature map is larger than the target convolutional neural network. The input element scale feature map of the largest number A target convolutional neural network, for weights, for weights, For softmax operation; Step S23: Calculate the classification loss of the target convolutional neural network based on the element-scale feature map of the target convolutional neural network, wherein the calculation formula for the classification loss of the target convolutional neural network based on the element-scale feature map of the target convolutional neural network is: in, For the classification loss, Calculate the cross-entropy; Step S24: Calculate the objective function value based on the classification loss and the self-distillation loss. If the preset maximum number of iterations is reached or the objective function value reaches a preset value, output the target area prediction probability value of each target convolutional neural network based on the convolution output value. If the objective function value is less than the preset value, update the convolution kernel size of the target convolutional neural network based on the objective function value, and return to step S12 with the updated kernel size until the preset maximum number of iterations is reached or the objective function value reaches a preset value. Then, output the target area prediction probability value of each target convolutional neural network based on the convolution output value. The calculation formula for the objective function value calculated based on the classification loss and the self-distillation loss is: in, Let the objective function be... This is the preset self-distillation loss coefficient.

6. A mineral target area prediction system, characterized in that, The target area prediction system includes: The data acquisition module is used to acquire the elemental content map of the area to be detected and to construct a chemical dataset based on the elemental content map. The element feature calculation module is used to calculate the element feature map based on the chemical dataset. The element feature dilation module is used to perform dilation convolution operations on the element feature map to obtain element scale feature maps of different scales. The network construction module is used to construct a well-trained source task network for predicting mineral target areas and multiple target convolutional neural networks with the same number of scales as the element scale feature map. The source task network for predicting mineral target areas is used to calculate the selected weights, which are the weight set of all channels of the source task network. The network structure is identical between any two target convolutional neural networks, and the network structure is identical between the target convolutional neural networks and the source task network for predicting mineral target areas. Specifically: Step S11: Pre-set the target area prediction source task network. and multiple target convolutional neural networks Among them, all networks have the same network structure and have The nth convolutional layer, the nth The number of input and output channels of the convolution kernel in the layer are respectively and The mineral target area prediction source task network The The convolution kernel of the layer is The target convolutional neural network The All convolutional kernels of a layer are defined as follows: And the first The convolutional kernels of the target convolutional neural network and the target area prediction source task network The convolutional kernels are consistent with those of the target convolutional neural network. sparse matrices with the same convolution kernel size ; Step S12: Based on the sparse matrix and the target area prediction source task network... The convolution kernels are used to perform Hadamard product calculations to obtain the selected weights. The weights are then determined based on the sparse matrix and the target area prediction source task network. The formula for calculating the Hadamard product using the convolution kernel is: in, For the target area prediction source task network, the first The first layer Weighting after selecting each channel For Hadama accumulation, The kernel size is ; The target area prediction module is used to input the element scale feature map and the selected weights into the corresponding target convolutional neural network, so as to control the target convolutional neural network to perform calculations through the selected weights, and obtain the target area prediction probability value of each target convolutional neural network. The prediction result output module is used to obtain the final prediction result by voting based on the prediction probability value of the target area of ​​each target convolutional neural network.

7. A mineral target area prediction system according to claim 6, characterized in that, The target area prediction module also includes: The convolution output value calculation module is used to calculate the output value based on the predicted source task network of the mining target area. The first layer After selecting the channels, the weights are calculated to obtain the convolutional output value of the target convolutional neural network, wherein the value is calculated based on the weights of the target area prediction source task network. The first layer The formula for calculating the convolution output value of the target convolutional neural network after selecting the weights of each channel is as follows: in, For the first The convolution output value of a target convolutional neural network. For convolution operations, For the first The nth target convolutional neural network The first layer A convolutional kernel with 1 channel, The pre-set convolution bias; The self-distillation loss calculation module is used to sort the target convolutional neural network according to the size of the input element-scale feature maps, and guide the target convolutional neural network with small-scale element-scale feature maps to learn a top-down self-distillation loss. The calculation formula for the top-down self-distillation loss obtained by guiding the target convolutional neural network with small-scale element-scale feature maps to learn a top-down self-distillation loss is as follows: in, Represents the target convolutional neural network Self-distillation loss, The input element-scale feature map is larger than the target convolutional neural network. The input element scale feature map of the largest number A target convolutional neural network, for weights, for weights, For softmax operation; The classification loss calculation module is used to calculate the classification loss of the target convolutional neural network based on the element-scale feature map of the target convolutional neural network. The calculation formula for the classification loss of the target convolutional neural network based on the element-scale feature map is as follows: in, For the classification loss, Calculate the cross-entropy; The iterative module is used to calculate the objective function value based on the classification loss and the self-distillation loss. If the preset maximum number of iterations is reached or the objective function value reaches a preset value, the predicted probability value of the target area for each target convolutional neural network is output based on the convolution output value. If the objective function value is less than the preset value, the convolution kernel size of the target convolutional neural network is updated based on the objective function value, and the updated kernel size is returned to continue executing the Hadamard product calculation module until the preset maximum number of iterations is reached or the objective function value reaches the preset value. Then, the predicted probability value of the target area for each target convolutional neural network is output based on the convolution output value. The calculation formula for the objective function value calculated based on the classification loss and the self-distillation loss is as follows: in, Let the objective function be... This is the preset self-distillation loss coefficient.

8. A mineral target area prediction device, characterized in that, It includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enable the at least one control processor to perform the target area prediction method as described in any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the target area prediction method as described in any one of claims 1 to 5.

Images