Mineral resource identification model training method based on data collected by unmanned aerial vehicle

Abstract

The invention relates to the field of artificial intelligence, in particular to a mineral resource recognition model training method based on data collected by an unmanned aerial vehicle. The training method comprises the steps of collecting real geological data, training a data generation model, generating simulated geological data by adopting the trained data generation model, and training a feature extraction model and a classification model by taking the simulated geological data and the real geological data as training samples together. An existing mineral resource identification model training method based on data acquired by an unmanned aerial vehicle has the problems of low model precision and low calculation speed of a mineral resource identification model. According to the mineral resource identification model training method based on the data acquired by the unmanned aerial vehicle provided by the invention, the model precision of the mineral resource identification model is relatively high, and the calculation speed is relatively high.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method for training a mineral resource identification model based on data collected by an unmanned aerial vehicle. Background Art

[0002] Mineral resources, also known as mineral resources, refer to the aggregate of minerals or useful elements that are formed through geological mineralization, naturally occur in the earth's crust or are buried underground or exposed on the surface, in solid, liquid or gaseous form, and have development and utilization value. Before mining mineral resources, it is necessary to conduct a regional survey and detailed exploration.

[0003] Regional survey is to conduct comprehensive surveys of a large area using various methods such as geology, geophysics and geochemistry to obtain a preliminary understanding of the stratigraphic structure, landforms, soil, rock types, structural characteristics and possible mineral signs in the area. The common methods of regional survey are manual survey and ground instrument survey. At present, with the gradual maturity of drone technology, the use of drones equipped with multiple sensors for survey has become a new method of regional survey. The various sensors carried by drones collect geological data when the drones fly in the mineral resource area. The geological data covers various geological information of the mining area, including multi-dimensional attributes such as spectrum, density, temperature, and humidity. The data collected by drones can provide more comprehensive geological data for the exploration area, greatly improving the exploration efficiency. However, because regional surveys usually need to cover a wide area, explorers need to process huge and complex geological data, which will consume a lot of time and manpower. Therefore, people have developed a mineral resource identification model based on drone collected data. The model can analyze geological data and obtain the possible mineral resource categories in the mineral resource area. The mineral resource identification model based on UAV collected data mainly includes a feature extraction model for extracting features that are more critical to classification and a classification model for analyzing the geological data after feature extraction to derive its mineral resource category.

[0004] When training a mineral resource identification model based on drone-collected data, traditional neural network algorithms are usually used to train feature extraction models. Traditional feature extraction methods are not effective when processing complex, nonlinear, high-dimensional data. It is difficult to maintain the content integrity and spatial structure consistency of the data, which affects the prediction accuracy of the model. In addition, traditional neural network algorithms are prone to gradient vanishing or explosion problems, especially in high-dimensional space, which is difficult to train stably, thus affecting the feature extraction effect. When using deep neural network algorithms to train classification models, it is easy to overfit when identifying complex mineral resource categories, and it is difficult to control the computational complexity of the model, resulting in slow model reasoning and poor practicality.

[0005] Therefore, the existing mineral resource identification model training method based on UAV collected data has the problems of low model accuracy and slow calculation speed of the mineral resource identification model. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a mineral resource identification model training method based on drone collected data, which can make the mineral resource identification model have higher model accuracy and faster calculation speed.

[0007] In order to solve the above technical problems, the present invention provides a method for training a mineral resource identification model based on data collected by an unmanned aerial vehicle, comprising:

[0008] S1. Using a plurality of different types of sensors equipped on a drone, real geological data of the mineral resource area is collected in real time when the drone flies over the mineral resource area, the real geological data is preliminarily processed and stored, and the mineral resource category of the preliminarily processed real geological data is marked;

[0009] The marking method is that professional geologists analyze the real geological data in combination with field samples of the mineral resource area for accurate marking;

[0010] S2, using the annotated real geological data to train a data generation model constructed by a generative adversarial network algorithm based on dynamic gradient adjustment, using a generative adversarial network algorithm based on dynamic gradient adjustment when training the data generation model constructed by the generative adversarial network algorithm based on dynamic gradient adjustment to obtain a trained data generation model, using the trained data generation model to generate simulated geological data, and using the simulated geological data and the annotated real geological data together as training samples;

[0011] S3. Determine the data feature dimension of the training sample. When the data feature dimension is greater than or equal to 20, use the training sample to train the feature extraction model constructed by the autoencoder algorithm based on the dual constraints, and use the autoencoder algorithm based on the dual constraints when training the feature extraction model constructed by the autoencoder algorithm based on the dual constraints; when the data feature dimension is less than 20, use the feature extraction model constructed by the neural network algorithm based on dynamic population evolution optimization, and use the neural network algorithm based on dynamic population evolution optimization when training the feature extraction model constructed by the neural network algorithm based on dynamic population evolution optimization; obtain the trained feature extraction model and the training sample after feature extraction;

[0012] S4. Use the training samples after feature extraction to train the classification model constructed by the deep neural decision tree classification algorithm based on quantum transformation pruning. When training the classification model constructed by the deep neural decision tree classification algorithm based on quantum transformation pruning, use the deep neural decision tree classification algorithm based on quantum transformation pruning to obtain a trained classification model.

[0013] As a further improvement of the present invention: the data generation model constructed by training the generative adversarial network algorithm based on dynamic gradient adjustment in S2 includes:

[0014] S201, initializing the network parameters of the generator and discriminator of the generative adversarial network;

[0015] S202. During the adversarial training process, the generator and the discriminator are trained alternately. The generator generates fake geological data, and the discriminator determines whether the input comes from a real geological data set or the generator. In this way, the generator learns to generate geological data that is increasingly difficult for the discriminator to distinguish. The objective functions of the generator and the discriminator are as follows:

[0016]

[0017] In the formula, To obtain the corresponding generator parameters that minimize the expectation, is the weight of the generator, To obtain the discriminator parameters corresponding to the maximum expectation, is the weight of the discriminator, is the expected operation, x is the real geological data sample, p data (x) is the distribution of real geological data, D c (x) is the judgment result of the discriminator on the real sample, z is the input noise of the generator, which comes from the predefined noise distribution p z (z), D c () is the discriminator function, G c () is the generator function, G c (z) is the fake geological data generated by the generator;

[0018] In the adversarial training stage, the discriminator output in the loss function is the judgment result D of the real sample c (x) and the judgment result D of the discriminator on the fake geological data generated by the generator c (G c (z)) is implemented by a neural network layer with a Sigmoid activation function to ensure that the output value is between 0 and 1, expressed as:

[0019]

[0020] In the formula, fc () is the linear output of the discriminator network, is the weight of the discriminator;

[0021] S203. In each round of training, the learning rates of the generator and the discriminator are dynamically adjusted according to the feedback of the discriminator. When the accuracy of the discriminator is too high, the learning rate of the generator is increased and the learning rate of the discriminator is reduced, and vice versa. This dynamic adjustment helps to avoid mode collapse during training and ensure the diversity of geological data generation. Dynamic gradient adjustment is achieved by adjusting the learning rate. The adjustment strategy is expressed as:

[0022]

[0023] In the formula, is the learning rate of the generator, η0 is the basic learning rate, is the adjustment coefficient of the generator learning rate, ∥∥ is the L2 norm, is the gradient of the generator’s loss function with respect to its parameters, is the loss function of the generator, is the learning rate of the discriminator, is the adjustment coefficient of the discriminator learning rate, is the gradient of the discriminator’s loss function with respect to its parameters, is the loss function of the discriminator;

[0024] S204, using the self-similarity principle to guide the generator to focus on the local similarity of geological data, improve the quality and practicality of the fake geological data, and the self-similarity adjustment guides the model to pay attention to the self-similarity characteristics of local geological data by adding a regularization term in the loss function of the generator. The calculation formula of the regularization term is:

[0025]

[0026] In the formula, R c () is the self-similarity regularization term calculation function, G c (z) is the fake geological data generated by the generator, β c is the regularization coefficient of the generative adversarial network, G c (z) i is the i-th characteristic region in the false geological data, G c (z) j is the jth characteristic area in the false geological data, s c () is the self-similarity descriptor calculation function;

[0027] S205, repeating steps S202 to S204 until a preset condition for stopping iteration is met.

[0028] Preferably, in step S202 Said The calculation formulas are:

[0029]

[0030] In the formula, is the weight of the discriminator network, is the bias of the discriminator network, x i is the i-th characteristic attribute of the real geological data sample, G c,i (z) is the i-th output of the generator.

[0031] As a further improvement of the present invention: the feature extraction model constructed by training the autoencoder algorithm based on the dual constraints in S3 includes:

[0032] S311, geological data passes through multiple hidden layers from the input layer through the autoencoder, each layer undergoes linear transformation and nonlinear activation, and finally outputs the extracted features. The transfer function of each layer is expressed as:

[0033] x p,i+1 =Sig(W p,i x p,i +b p,i ),

[0034] In the formula, x p,i+1 is the output vector of the i+1th layer of the autoencoder, Sig() is the Sigmoid activation function, and W p,i is the weight matrix of the i-th layer of the autoencoder, x p,i is the input vector of the i-th layer of the autoencoder, b p,i is the bias vector of the autoencoder;

[0035] S312. The loss function of the autoencoder is designed as a dual constraint loss. One part comes from the reconstruction loss of the geological data content, ensuring that the extracted features can effectively reconstruct the original geological data; the other part is the structure preservation loss, ensuring that the original structure of the geological data is maintained in the feature space. The loss function calculation formula of the autoencoder is:

[0036] L p =λ p1 L p,recon +λ p2 L p,struct ,

[0037] Where, L p is the loss function of the autoencoder, λ p1 To adjust the weight coefficient of the relative importance of reconstruction loss, L p,recon is the reconstruction loss, λ p2The weight coefficient to adjust the structure to maintain the relative importance of losses,

[0038] L p,struct To maintain losses for the structure;

[0039] S313. Calculate the gradient according to the loss function and use the least squares gradient descent method to update the network parameters, considering the stability of the gradient and the adaptability of the update rate to avoid the gradient explosion or disappearance problem in high-dimensional space. The parameter update rule is:

[0040]

[0041] In the formula, θ p,i is the parameter of the i-th layer of the autoencoder, ← is the parameter update operation, η p is the learning rate of the autoencoder, is the loss function L of the autoencoder p The gradient of the parameters of the i-th layer of the autoencoder;

[0042] S314, repeating steps S311 to S313 until a preset stop iteration condition is met.

[0043] Preferably, the reconstruction loss L in S312 is p,recon The calculation formula is:

[0044]

[0045] In the formula, x p is the original input geological data, is the output of the autoencoder;

[0046] The structure retention loss L p,struct The calculation formula is:

[0047] L p,struct =∑ j,k W p,jk ∥z p,j -z p,k ∥ 2 ,

[0048] Where W p,jk is the similarity matrix based on the input geological data, z p,j is the jth geological data point mapped to the feature space, z p,k is the kth geological data point mapped to the feature space.

[0049] As a further improvement of the present invention: it is characterized in that the feature extraction model constructed by training the neural network algorithm based on dynamic population evolution optimization in S3 includes:

[0050] S321. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, each individual represents a configuration of network weights, and the population size is N. p , initialize the weights and biases for the i-th individual, expressed as:

[0051]

[0052] In the formula, is the weight matrix of the i-th individual in the initial state, ~ is subject to a specific distribution, is a normal distribution; The mean is 0 and the variance is σ 2 Normal distribution of 2 is the initialization variance, is the bias of the i-th individual in the initial state; W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual;

[0053] S322. For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. For the i-th individual, use its weight and bias to calculate the loss on the training data set, expressed as:

[0054]

[0055] Where, L pi is the loss of the neural network corresponding to the i-th individual, mps is the number of samples input in the current batch, is the composite loss function, fsig() is the neural network model function, is the characteristic of the jth geological data sample, W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, is the label of the jth geological data sample;

[0056] S323. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. The selection is based on the fitness of the individual. The excellent individual has a higher probability of being selected. The calculation formula for the probability of an individual being selected is:

[0057]

[0058] Where P select (i) is the probability of the i-th individual being selected, γ pse To control the selected pressure parameter, Lpi is the loss of the neural network corresponding to the i-th individual, N p is the population size, L pk is the loss of the neural network corresponding to the kth individual;

[0059] S324. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. The crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0060] '

[0061] W pi =α pcs W p1 +(1-α pcs )W p2 ,

[0062] '

[0063] b pi =α pcs b p1 +(1-α pcs )b p2 ,

[0064] '

[0065] Where W pi is the weight of the neural network corresponding to the individual after the crossover operation, α pcs is the crossover rate,

[0066] W p1 is the weight of the neural network corresponding to the first individual selected, W p2 For the second individual selected

[0067] ′ is the weight of the neural network, b pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first individual selected, b p2 is the bias of the neural network corresponding to the selected second individual;

[0068] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0069]

[0070] ″′

[0071] Where W pi is the weight of the neural network corresponding to the individual after the mutation operation, W piis the weight of the neural network corresponding to the individual after the crossover operation, The mean is 0 and the variance is τ 2 The normal distribution of 2

[0072] ″′

[0073] is the variance of the variation, b pi is the bias of the neural network corresponding to the individual after the mutation operation, b pi is the bias of the neural network corresponding to the individual after the crossover operation;

[0074] S325, repeating steps S322 to S324 until a preset stop iteration condition is met.

[0075] Preferably, in S322, The calculation formula is:

[0076]

[0077] In the formula, is the composite loss function, fsig() is the neural network model function, is the characteristic of the jth geological data sample, W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, is the label of the jth geological data sample, MSE() is the mean square error function, λ ps is the regularization parameter, Reg(W pi ) is the regularization term;

[0078] The regularization term Reg(W pi ) is calculated as:

[0079]

[0080] Where W pi,k is the kth weight of the neural network corresponding to the ith individual.

[0081] As a further improvement of the present invention: the classification model constructed by the training of the deep neural decision tree classification algorithm based on quantum transformation pruning in S4 includes:

[0082] S401, initialize the parameters of the deep neural decision tree, including the weight of the decision node and the depth of the tree, and set the weight of the i-th node in the deep neural decision tree to W u,i , the bias of the i-th node in the deep neural decision tree is b u,i , the initialization method is expressed as:

[0083]

[0084] In the formula, is a normal distribution with a mean of 0 and a standard deviation of 1, and ~ is subject to a specific distribution;

[0085] S402. During the training process, the geological data after input feature extraction passes through the multi-layer decision nodes of the decision tree. Each node makes a decision based on the weight and input feature, guiding the geological data to flow to the left or right subtree until it reaches the leaf node. Let the feature vector input to the deep neural decision tree be x u , expressed as:

[0086] y u,n =Sig(W u,n x u +b u,n ),

[0087] In the formula, y u,n is the output of the nth node of the deep neural decision tree; Sig() is the Sigmoid activation function; W u,n is the weight of the nth node in the deep neural decision tree, b u,n is the bias of the nth node in the deep neural decision tree;

[0088] S403. Considering the classification accuracy and tree complexity, a loss function is used to evaluate the performance of the current decision tree. The loss function of the deep neural decision tree includes classification error and regularization terms to prevent overfitting. The calculation formula is:

[0089]

[0090] Where, L u is the loss function of the deep neural decision tree, N u is the number of samples input to the deep neural decision tree in the current batch, t i is the target output of the i-th sample, y u,i is the classification result of the deep neural decision tree for the i-th sample, λ u is the regularization coefficient of the deep neural decision tree, W u is the weight of the deep neural decision tree; ∥∥ is the L2 norm;

[0091] S404. In each training cycle, the utility and efficiency of each decision node are evaluated according to the quantum transformation principle. Inefficient nodes will be pruned to reduce the complexity of the model while maintaining or improving the performance of the model. The present invention uses the quantum transformation principle to evaluate and adjust the structure of the network. Suppose P u,i Indicates the importance of the i-th node. If P u,i Less than the preset threshold value θ u, then the node will be removed from the model, P u,i The calculation formula is:

[0092]

[0093] Where, L u is the loss function of the deep neural decision tree, W u,i is the weight of the i-th node in the deep neural decision tree, ψ r () is the quantum transformation function; Z r is the input feature vector of the current node of the deep neural decision tree;

[0094] S405, repeating steps S402 to S404 until a preset stop iteration condition is met, indicating that the model training is completed.

[0095] Preferably, the ψ in S404 r (z r ) is calculated as:

[0096]

[0097] In the formula, Z r is the input feature vector of the current node of the deep neural decision tree, λ r is the parameter that controls the sensitivity of quantum transformation, z rj For z r The jth element of corresponds to z rj The quantum noise level of

[0098] The quantum noise level has different compression levels in different parts of the feature representation, thereby more finely controlling the information loss, and the calculation formula is:

[0099]

[0100] In the formula, α r To adjust the coefficient of quantum noise response, z rj For z r The j-th element of .

[0101] The beneficial effects of the present invention are as follows: the mineral resource identification model training method based on drone collected data provided by the present invention can make the model accuracy of the mineral resource identification model higher and the calculation speed faster.

[0102] First, the present invention first uses a generative adversarial network algorithm based on dynamic gradient adjustment to train a data generation model constructed by a generative adversarial network algorithm based on dynamic gradient adjustment before training a feature extraction model and a classification model. The traditional generative adversarial network may face a pattern collapse problem when simulating the distribution of complex geological data, resulting in insufficient diversity of generated geological data, but the generative adversarial network algorithm based on dynamic gradient adjustment adopts a dynamic gradient adjustment strategy, and dynamically adjusts the learning rate of the generator and the discriminator during the confrontation process to achieve more sophisticated geological data distribution learning. Therefore, a large amount of simulated geological data can be obtained by the trained data generation model, and it can be used together with the real geological data as training samples to train feature extraction models and classification models, which can effectively expand the training samples.

[0103] Secondly, the present invention selects feature extraction models based on different algorithms according to the number of data feature dimensions. When the data feature dimension is greater than or equal to 20, the feature extraction model based on the autoencoder algorithm with dual constraints is used to process the geological data collected by the drone. The geological data is highly nonlinear and complex due to terrain and environmental factors. The dual constraint mechanism is adopted. The integrity of the mineral features and the spatial topological structure of the geological data are guaranteed in the feature extraction process through the dual mechanism of content constraint and structure constraint, and the representation of the feature space is optimized to meet the needs of high-dimensional complex geological data. For low-dimensional data with data feature dimensions less than 20, a feature extraction model based on a neural network algorithm optimized by dynamic population evolution is adopted. In the traditional population evolution algorithm, the evolution of all individuals is based on fixed rules. The present invention uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current training data, and adjusts the probability of crossover and mutation according to the changing trend of the loss function, so that the algorithm can adapt to different geological data distributions more flexibly, and improve the generalization ability and training efficiency of the model. The dynamic population evolution optimization algorithm is adopted, and the evolution rules and parameters are adjusted through the self-correction mechanism to improve the adaptability of the neural network to variable geological data, and enhance the effectiveness of feature extraction and the robustness of the model.

[0104] Finally, the present invention adopts a deep neural decision tree classification algorithm based on quantum transformation pruning to improve the classification accuracy and optimize the complexity of the model, effectively controlling the complexity and computational complexity of the classifier, so that it can accurately classify different mineral categories while improving computational efficiency.

[0105] Therefore, after adding the data augmentation model and optimizing the training method of the feature extraction model, the model accuracy of the mineral resource identification model based on UAV data collection is higher. After optimizing the training method of the classification model, the calculation speed of the mineral resource identification model based on UAV data collection is faster. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1It is a principle block diagram of the present invention. DETAILED DESCRIPTION

[0107] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0108] like Figure 1 As shown, the mineral resource identification model training method based on drone collected data provided by the present invention includes:

[0109] S1. Using multiple different types of sensors equipped on the drone, the drone collects real geological data of the mineral resource area in real time when it flies over the mineral resource area, preliminarily processes and stores the real geological data, and marks the mineral resource category of the preliminarily processed real geological data;

[0110] Multiple different types of sensors include high-resolution cameras, multispectral sensors, infrared scanners, and altimeters.

[0111] The drone is designed specifically according to usage requirements and has a preset flight path before flight. At the same time, the drone is equipped with optical cameras, infrared thermal imagers, radars, and combines image recognition algorithms with target detection algorithms to achieve automatic target recognition functions, ensuring that the designated mineral resource area can be fully and accurately covered during flight.

[0112] In practical applications, the attributes of real geological data are usually more than ten, and can reach dozens or even hundreds. Common attributes of real geological data include:

[0113] Ra is reflectivity, which describes the ability of surface materials to reflect light in a specific wavelength band;

[0114] Da is the density, which reflects the density information of soil or rock in the mining area;

[0115] Ea is the height difference, obtained by the altimeter carried by the drone;

[0116] Ta is the temperature index, the surface temperature measured by infrared scanning;

[0117] Pa is the pressure value, which is related to the compaction of the surface material;

[0118] Fa is the fractal dimension, which describes the complexity of the surface texture;

[0119] Ga is the spectral signature, the feature vector extracted from multispectral data;

[0120] Ba is a magnetic index, reflecting the magnetic properties of surface minerals;

[0121] Ha is the humidity index, which represents the humidity level in the collection area.

[0122] The preliminary processing includes image correction, noise filtering and format conversion. The format conversion refers to converting the real geological data into a unified data format, namely HDF5 (Hierarchical Data Format).

[0123] Real geological data are stored in the form of discrete vectors.

[0124] The marking method is that professional geologists analyze the real geological data in combination with field samples in the mineral resource area for accurate marking;

[0125] Mineral resource categories include iron ore, copper ore, gold ore, etc.;

[0126] S2, using the annotated real geological data to train a data generation model constructed by a generative adversarial network algorithm based on dynamic gradient adjustment, after the training is completed, a trained data generation model is obtained, the trained data generation model is used to generate simulated geological data, and the simulated geological data and the annotated real geological data are used together as training samples;

[0127] Generative adversarial network is a deep learning algorithm architecture, which consists of two neural networks, generator and discriminator, which compete with each other to achieve the purpose of generating new data. Dynamic gradient adjustment is to dynamically adjust the learning rate of the generator and discriminator according to the feedback of the discriminator in each round of training, using the gradient of the loss function of the generator and discriminator with respect to their parameters. When the accuracy of the discriminator is too high, the learning rate of the generator is increased and the learning rate of the discriminator is reduced, and vice versa, to help avoid mode collapse during training and ensure the diversity of geological data generation. Training based on the generative adversarial network algorithm for dynamic gradient adjustment The data generation model constructed adopts the generative adversarial network algorithm based on dynamic gradient adjustment. The training includes:

[0128] S201, initialize the network parameters of the generator and discriminator of the generative adversarial network; the generator aims to simulate the distribution of geological data, and the discriminator aims to distinguish between real geological data and fake geological data generated by the generator. Let the generator be G c , the discriminator is D c , the initialization method is expressed as:

[0129]

[0130] In the formula, is the weight of the generator, represents a normal distribution with a mean of 0 and a variance of 0.01. is the weight of the discriminator;

[0131] S202. During the adversarial training process, the generator and the discriminator are trained alternately. The generator generates fake geological data, and the discriminator determines whether the input comes from a real geological data set or the generator. In this way, the generator learns to generate geological data that is increasingly difficult for the discriminator to distinguish. The objective functions of the generator and the discriminator are as follows:

[0132]

[0133] In the formula, To obtain the corresponding generator parameters that minimize the expectation, is the weight of the generator, To obtain the discriminator parameters corresponding to the maximum expectation, is the weight of the discriminator, is the expected operation, x is the real geological data sample, p data (x) is the distribution of real geological data, D c (x) is the judgment result of the discriminator on the real geological data sample, and z is the input noise of the generator, which comes from the predefined noise distribution p z (z), D c () is the discriminator function, G c () is the generator function, G c (z) is the fake geological data generated by the generator;

[0134] In the adversarial training stage, the discriminator output in the loss function is the judgment result D of the real geological data sample. c (x) and the judgment result D of the discriminator on the fake geological data generated by the generator c (G c (z)) is implemented by a neural network layer with a Sigmoid activation function to ensure that the output value is between 0 and 1, expressed as:

[0135]

[0136] In the formula, f c () is the linear output of the discriminator network, which depends on the input x or G c (z) and its network parameters is the weight of the discriminator;

[0137] The linear output of the discriminator network is the weighted sum of the multilayer perceptrons, The calculation formulas are:

[0138]

[0139] In the formula, is the weight of the discriminator network, is the bias of the discriminator network, xi is the i-th characteristic attribute of the real geological data sample, G c,i (z) is the i-th output of the generator.

[0140] S203. In each round of training, the learning rates of the generator and the discriminator are dynamically adjusted according to the feedback of the discriminator. When the accuracy of the discriminator is too high, the learning rate of the generator is increased and the learning rate of the discriminator is reduced, and vice versa. This dynamic adjustment helps to avoid mode collapse during training and ensure the diversity of geological data generation. Dynamic gradient adjustment is achieved by adjusting the learning rate. The adjustment strategy is expressed as:

[0141]

[0142] In the formula, is the learning rate of the generator, η0 is the basic learning rate, is the adjustment coefficient of the generator learning rate, ∥∥ is the L2 norm, is the gradient of the generator’s loss function with respect to its parameters, is the loss function of the generator, is the learning rate of the discriminator, is the adjustment coefficient of the discriminator learning rate, is the gradient of the discriminator’s loss function with respect to its parameters, is the loss function of the discriminator;

[0143] The adjustment coefficients of the learning rates of the generator and discriminator are adjusted according to the second-order moment of the loss gradient to achieve adaptive optimization of the learning rate of the generative adversarial network. The calculation method is expressed as:

[0144]

[0145] In the formula, is the adjustment coefficient of the generator learning rate, σ() is the standard deviation calculation function of the gradient, G c () is the generator function, is the weight of the generator, μ() represents the mean calculation function of the gradient; ∈ is a preset small constant, ∈ is set to 0.001; is the adjustment coefficient of the discriminator learning rate, D c () is the discriminator function, is the weight of the discriminator;

[0146] The standard deviation and mean of the gradient are calculated as:

[0147]

[0148] Where N cg Indicates the number of samples input into the generative adversarial network in the current batch, is the gradient of the generator's loss function with respect to its parameters for the nth sample, represents the gradient of the loss function of the discriminator with respect to its parameters for the nth sample.

[0149] S204. Use the self-similarity principle to guide the generator to focus on the local similarity of geological data, improve the quality and practicality of fake geological data, and combine the self-similar process so that the generated geological data can better simulate the local characteristics of geological data while maintaining the consistency of the overall distribution, thereby enhancing the model's learning ability for small sample geological data. By dynamically adjusting the learning rate of the generator and the discriminator and introducing the self-similar regularization term, the problem of mode collapse in the traditional generative adversarial network is overcome, the generated data diversity is higher, and the generalization ability of the model is enhanced. Self-similarity adjustment guides the model to pay attention to the self-similarity characteristics of local geological data by adding a regularization term to the loss function of the generator. The calculation formula of the regularization term is:

[0150] R c (G c (z)) = β c ∑ i,j ∥G c (z) i -G c (z) j ∥+s c (G c (z)),

[0151] In the formula, R c () is the self-similarity regularization term calculation function, G c (z) is the fake geological data generated by the generator, β c is the regularization coefficient of the generative adversarial network, G c (z) i is the i-th characteristic region in the false geological data, G c (z) j is the jth characteristic region in the false geological data, s c () is the self-similarity descriptor calculation function;

[0152] The self-similarity descriptor measures the similarity within the generated sample, and the calculation method is expressed as:

[0153]

[0154] Where |P| is the total number of geological data feature pairs involved in the comparison, P is all possible geological data feature pairs, It's G c The mean value of (z), G c (z) pis the pth feature pair of fake geological data generated by the generator, τ c is the scale parameter of the regularization, which controls the influence of self-similarity.

[0155] S205, repeating the iteration steps S202 to S204 until the preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is reaching the preset maximum number of iterations, and the preset maximum number of iterations is set to 1000 times.

[0156] To distinguish them, the geological data generated by the data generation model during the training process is called pseudo geological data, and the address data generated by the trained data generation model is called simulated geological data.

[0157] S3. Determine the data feature dimension of the training sample. When the data feature dimension is greater than or equal to 20, use the training sample to train the feature extraction model constructed by the autoencoder algorithm based on the dual constraints; when the data feature dimension is less than 20, use the feature extraction model constructed by the neural network algorithm based on dynamic population evolution optimization; after the training is completed, obtain the trained feature extraction model and the training sample after feature extraction;

[0158] The encoder of the autoencoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder of the autoencoder is used to reconstruct the reduced-dimensional features back to the original space. The result of feature extraction is the output feature of the encoder. The dual constraint mechanism is the content constraint of geological data to ensure the integrity and discrimination of mineral features during feature extraction; the second is structural constraint to maintain the geometric and topological structure of geological data in the original space, thereby optimizing the geological data representation in the feature space. The feature extraction model constructed by the autoencoder algorithm based on dual constraints adopts the autoencoder algorithm based on dual constraints. The training includes:

[0159] S311, geological data passes through multiple hidden layers from the input layer through the autoencoder, each layer undergoes linear transformation and nonlinear activation, and finally outputs the extracted features. The transfer function of each layer is expressed as:

[0160] x p,i+1 =Sig(W p,i x p,i +b p,i ),

[0161] In the formula, x p,i+1 is the output vector of the i+1th layer of the autoencoder, Sig() is the Sigmoid activation function, and W p,i is the weight matrix of the i-th layer of the autoencoder, x p,i is the input vector of the i-th layer of the autoencoder, b p,i is the bias vector of the autoencoder;

[0162] S312. The loss function of the autoencoder is designed as a dual constraint loss. One part comes from the reconstruction loss of the geological data content, ensuring that the extracted features can effectively reconstruct the original geological data; the other part is the structure preservation loss, ensuring that the original structure of the geological data is maintained in the feature space. The loss function calculation formula of the autoencoder is:

[0163] L p =λ p1 L p,recon +λ p2 L p,struct ,

[0164] Where, L p is the loss function of the autoencoder, λ p1 To adjust the weight coefficient of the relative importance of reconstruction loss, L p,recon is the reconstruction loss, λ p2 The weight coefficient to adjust the structure to maintain the relative importance of losses,

[0165] L p,struct To maintain losses for the structure;

[0166] Reconstruction loss L p,recon The calculation formula is:

[0167]

[0168] In the formula, x p is the original input geological data, is the output of the autoencoder;

[0169] The structure preservation loss aims to maintain the original topological structure of geological data. It is calculated using the Laplace eigenmapping method. The structure preservation loss L p,struct The calculation formula is:

[0170] L p,struct =∑ j,k W p,jk ∥z p,j -z p,k ∥ 2 ,

[0171] Where W p,jk is the similarity matrix based on the input geological data, z p,j is the jth geological data point mapped to the feature space, z p,k is the kth geological data point mapped to the feature space.

[0172] The output of the autoencoder Decomposed into the nonlinear mapping relationship between the output and input of the network, the calculation method is expressed as:

[0173]

[0174] In the formula, g p () is the decoding function of the autoencoder, which is responsible for transforming the representation h of the latent space p (x p ,W p,enc ) is mapped back to the original geological data space; h p () is the encoding function, which maps the original input to a latent space; W p,enc is the weight parameter of the encoder in the autoencoder, W p,dec is the weight parameter of the decoder in the autoencoder.

[0175] The encoding function is a superposition of multiple layers of networks, and the output of each layer is a nonlinear transformation of the output of the previous layer, expressed as:

[0176] h p,k+1 =A p ⊙Ser(W p,k h p,k +b p,k ),

[0177] In the formula, h p,k+1 is the input of the k+1th layer of the encoding function, A p represents the attention score of the feature, ⊙ represents element multiplication, Ser() is a dynamic adaptive activation function, and W p,k is the weight of the kth layer of the encoder, h p,k is the input of the kth layer of the encoding function, b p,k is the bias of the kth layer of the encoder;

[0178] Based on the input geological data x p The similarity matrix W p,jk The calculation method is expressed as:

[0179]

[0180] In the formula, x p,j is the jth sample in the original geological data, x p,k is the kth sample in the original geological data; σ p is a scale parameter used to control the similarity sensitivity, σ p Set to 0.1.

[0181] The feature selection module based on the attention mechanism is used to further enhance the sensitivity of the autoencoder to the most informative part of the input geological data, thereby improving the ability of the overall model to identify key features in geological data. The feature attention score A p The calculation method is expressed as:

[0182] A p =sof(W p,att ·h p,k-1 +b p,att ),

[0183] In the formula, sof() is the Softmax function, W p,att is the weight of the learned attention layer, h p,k-1 is the input of the k-1th layer of the encoding function, b p,att is the bias of the learned attention layer;

[0184] A dynamic adaptive activation strategy is used in each layer of the autoencoder to automatically adjust the shape and parameters of the activation function. This can significantly improve the learning ability and generalization performance of the model, especially when processing highly nonlinear mineral geological data. Suppose the input of the dynamic adaptive activation function is x puts , the calculation method of the dynamic adaptive activation function is expressed as:

[0185]

[0186] Where Ser() is a dynamic adaptive activation function; α p is the first dynamic adaptive activation function adjustment parameter, used to adjust the slope of the curve; β p The parameters of the second dynamic adaptive activation function are adjusted to adjust the center position of the function.

[0187] In each iteration, α p and β p The update not only depends on the partial derivatives of the loss function with respect to them, but also includes adjustment terms based on historical gradients to ensure the stability and adaptability of the update. The update method is expressed as:

[0188]

[0189] Where η p is the learning rate of the autoencoder, L p is the loss function of the autoencoder, λ p is a hyperparameter that adjusts the influence of historical gradients, Δα p is the change in the adjustment parameters of the first dynamic adaptive activation function between this iteration and the previous iteration, Δβ p The change in the adjustment parameters of the second dynamic adaptive activation function between this iteration and the previous iteration;

[0190] S313. Calculate the gradient according to the loss function and use the least squares gradient descent method to update the network parameters, considering the stability of the gradient and the adaptability of the update rate to avoid the gradient explosion or disappearance problem in high-dimensional space. The parameter update rule is:

[0191]

[0192] In the formula, θ p,i is the parameter of the i-th layer of the autoencoder, ← is the parameter update operation; η p is the learning rate of the autoencoder, η p Set to 0.01; is the loss function L of the autoencoder p The gradient of the parameters of the i-th layer of the autoencoder;

[0193] S314, repeating the iteration steps S311 to S313 until the preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is reaching the preset maximum number of iterations, and the preset maximum number of iterations is set to 1000 times.

[0194] Dynamic population evolution optimization uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current training data, and adjusts the probability of crossover and mutation according to the changing trend of the loss function. The feature extraction model constructed by the neural network algorithm based on dynamic population evolution optimization adopts the neural network algorithm based on dynamic population evolution optimization. The training includes:

[0195] S321. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, each individual represents a configuration of network weights, and the population size is N. p , initialize the weights and biases for the i-th individual, expressed as:

[0196]

[0197] In the formula, is the weight matrix of the i-th individual in the initial state, ~ is subject to a specific distribution, is a normal distribution; The mean is 0 and the variance is σ 2 Normal distribution of 2 is the initialization variance, is the bias of the i-th individual in the initial state; W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual;

[0198] S322. For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. For the i-th individual, use its weight and bias to calculate the loss on the training data set, expressed as:

[0199]

[0200] Where, L pi is the loss of the neural network corresponding to the i-th individual, mps is the number of samples input in the current batch, is the composite loss function, fsig() is the neural network model function, is the characteristic of the jth geological data sample, W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, is the label of the jth geological data sample;

[0201] The composite loss function contains a regularization term, which can increase the generalization ability of the model.

[0202] The calculation formula is:

[0203]

[0204] In the formula, is the composite loss function, fsig() is the neural network model function, is the characteristic of the jth geological data sample, W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, is the label of the jth geological data sample, MSE() is the mean square error function; ps is the regularization parameter, γ pse Set to 2; Reg(W pi ) is the regularization term;

[0205] Regularization term Reg(W pi ) is calculated as:

[0206]

[0207] Where W pi,k is the kth weight of the neural network corresponding to the ith individual.

[0208] S323. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. The selection is based on the fitness of the individual. The excellent individual has a higher probability of being selected. The calculation formula for the probability of an individual being selected is:

[0209]

[0210] Where P select (i) is the probability of the i-th individual being selected, γ pseTo control the selected pressure parameter, L pi is the loss of the neural network corresponding to the i-th individual, N p is the population size, L pk is the loss of the neural network corresponding to the kth individual;

[0211] S324. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. The crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0212] '

[0213] W pi =α pcs W p1 +(1-α pcs )W p2 ,

[0214] '

[0215] b pi =α pcs b p1 +(1-α pcs )b p2 ,

[0216] '

[0217] Where W pi is the weight of the neural network corresponding to the individual after the crossover operation; α pcs is the crossover rate, α pcs Set to 0.3; W p1 is the weight of the neural network corresponding to the first individual selected, W p2 For the selected

[0218] The weight of the neural network corresponding to the second individual, b pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first individual selected, b p2 is the bias of the neural network corresponding to the selected second individual;

[0219] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0220]

[0221] ″′

[0222] Where W pi is the weight of the neural network corresponding to the individual after the mutation operation, W piis the weight of the neural network corresponding to the individual after the crossover operation, The mean is 0 and the variance is τ 2 The normal distribution of 2 is the variance of the variation, τ 2 Set to 0.04; b ″ pi is the bias of the neural network corresponding to the individual after the mutation operation, ′

[0223] b pi is the bias of the neural network corresponding to the individual after the crossover operation;

[0224] S325, repeating the iteration steps S322 to S324 until the preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is reaching the preset maximum number of iterations, and the preset maximum number of iterations is set to 1000 times.

[0225] S4, using the training samples after feature extraction to train the classification model constructed by the deep neural decision tree classification algorithm based on quantum transformation pruning, and after the training is completed, a trained classification model is obtained;

[0226] The dynamic pruning strategy of quantum transformation (referred to as quantum transformation pruning) evaluates node utility through quantum transformation and dynamically prunes inefficient nodes, thereby optimizing the structure of the decision tree, improving the classification accuracy and computational efficiency of the model, reducing the complexity of the model, and improving the generalization ability of the model, avoiding overfitting, and improving overall performance. The classification model constructed by the deep neural decision tree classification algorithm based on quantum transformation pruning adopts the deep neural decision tree classification algorithm based on quantum transformation pruning. The training includes:

[0227] S401, initialize the parameters of the deep neural decision tree, including the weight of the decision node and the depth of the tree, and set the weight of the i-th node in the deep neural decision tree to W u,i , the bias of the i-th node in the deep neural decision tree is b u,i , the initialization method is expressed as:

[0228]

[0229] In the formula, is a normal distribution with a mean of 0 and a standard deviation of 1, and ~ is subject to a specific distribution;

[0230] S402. During the training process, the geological data after input feature extraction passes through the multi-layer decision nodes of the decision tree. Each node makes a decision based on the weight and input feature, guiding the geological data to flow to the left or right subtree until it reaches the leaf node. Let the feature vector input to the deep neural decision tree be x u , expressed as:

[0231] y u,n =Sig(W u,n x u +b u,n ),

[0232] In the formula, y u,n is the output of the nth node of the deep neural decision tree; Sig() is the Sigmoid activation function, which is used to map the result of the linear combination to the (0, 1) interval to simulate the probability output of the decision; W u,n is the weight of the nth node in the deep neural decision tree, b u,n is the bias of the nth node in the deep neural decision tree;

[0233] S403. Considering the classification accuracy and tree complexity, a loss function is used to evaluate the performance of the current decision tree. The loss function of the deep neural decision tree includes classification error and regularization terms to prevent overfitting. The calculation formula is:

[0234]

[0235] Where, L u is the loss function of the deep neural decision tree, N u is the number of samples input to the deep neural decision tree in the current batch, t i is the target output of the i-th sample, y u,i is the classification result of the deep neural decision tree for the i-th sample, λ u is the regularization coefficient of the deep neural decision tree, W u is the weight of the deep neural decision tree; ∥∥ is the L2 norm, which is used to penalize excessive weight values ​​to improve the generalization ability of the model;

[0236] S404. In each training cycle, the utility and efficiency of each decision node are evaluated according to the quantum transformation principle. Inefficient nodes will be pruned to reduce the complexity of the model while maintaining or improving the performance of the model. The present invention uses the quantum transformation principle to evaluate and adjust the structure of the network. Suppose P u,i Indicates the importance of the i-th node. If P u,i Less than the preset threshold value θ u , then the node will be removed from the model, P u,i The calculation formula is:

[0237]

[0238] Where, L u is the loss function of the deep neural decision tree, W u,i is the weight of the i-th node in the deep neural decision tree, ψ r () is the quantum transformation function, which is used to simulate the quantum state; Zr is the input feature vector of the current node of the deep neural decision tree;

[0239] ψ r (z r ) is calculated as:

[0240]

[0241] In the formula, Z r is the input feature vector of the current node of the deep neural decision tree, λ r is the parameter that controls the sensitivity of quantum transformation, z rj For z r The jth element of corresponds to z rj The quantum noise level of

[0242] The quantum noise level has different compression levels in different parts of the feature representation, thereby more finely controlling the information loss, and the calculation formula is:

[0243]

[0244] In the formula, α r To adjust the coefficient of quantum noise response, α r Set to 0.2; z rj For z r The j-th element of .

[0245] S405, repeating the iteration steps S402 to S404 until the preset stop iteration condition is met, which means that the model training is completed. The preset stop iteration condition is reaching the preset maximum number of iterations, and the preset maximum number of iterations is set to 1000 times.

[0246] At this point, the training of the mineral resource identification model based on drone collected data has been completed.

[0247] Steps S5 to S7 are a brief description of the application process of the mineral resource identification model.

[0248] S5, collect new real geological data of mineral resource areas in real time through multiple sensors;

[0249] S6. Determine the data feature dimension of the training sample. When the data feature dimension is greater than or equal to 20, use the trained feature extraction model constructed based on the dual-constrained autoencoder algorithm to extract features from the new real geological data. When the data feature dimension is less than 20, use the trained feature extraction model constructed based on the dynamic population evolution optimization neural network algorithm to extract features from the new real geological data, and obtain new real geological data after feature extraction.

[0250] S7. Use the trained classification model to label the mineral resource categories of new real geological data.

Claims

1. A mineral resource identification model training method based on drone data collection, characterized in that: include: S1. Using a plurality of different types of sensors equipped on a drone, real geological data of the mineral resource area is collected in real time when the drone flies over the mineral resource area, the real geological data is preliminarily processed and stored, and the mineral resource category of the preliminarily processed real geological data is marked; The marking method is that professional geologists analyze the real geological data in combination with field samples of the mineral resource area for accurate marking; S2, using the annotated real geological data to train a data generation model constructed by a generative adversarial network algorithm based on dynamic gradient adjustment, using a generative adversarial network algorithm based on dynamic gradient adjustment when training the data generation model constructed by the generative adversarial network algorithm based on dynamic gradient adjustment to obtain a trained data generation model, using the trained data generation model to generate simulated geological data, and using the simulated geological data and the annotated real geological data together as training samples; S3. Determine the data feature dimension of the training sample. When the data feature dimension is greater than or equal to 20, use the training sample to train the feature extraction model constructed by the autoencoder algorithm based on the dual constraints, and use the autoencoder algorithm based on the dual constraints when training the feature extraction model constructed by the autoencoder algorithm based on the dual constraints; when the data feature dimension is less than 20, use the feature extraction model constructed by the neural network algorithm based on dynamic population evolution optimization, and use the neural network algorithm based on dynamic population evolution optimization when training the feature extraction model constructed by the neural network algorithm based on dynamic population evolution optimization; obtain the trained feature extraction model and the training sample after feature extraction; S4. Use the training samples after feature extraction to train the classification model constructed by the deep neural decision tree classification algorithm based on quantum transformation pruning. When training the classification model constructed by the deep neural decision tree classification algorithm based on quantum transformation pruning, use the deep neural decision tree classification algorithm based on quantum transformation pruning to obtain a trained classification model.

2. The method for training a mineral resource identification model based on unmanned aerial vehicle data collection according to claim 1, characterized in that: The data generation model constructed by training the generative adversarial network algorithm based on dynamic gradient adjustment in S2 includes: S201, initializing the network parameters of the generator and discriminator of the generative adversarial network; S202. During the adversarial training process, the generator and the discriminator are trained alternately. The generator generates fake geological data, and the discriminator determines whether the input comes from a real geological data set or the generator. In this way, the generator learns to generate geological data that is increasingly difficult for the discriminator to distinguish. The objective functions of the generator and the discriminator are as follows: In the formula, To obtain the corresponding generator parameters that minimize the expectation, is the weight of the generator, To obtain the discriminator parameters corresponding to the maximum expectation, is the weight of the discriminator, is the expected operation, x is the real geological data sample, p data (x) is the distribution of real geological data, D c (x) is the judgment result of the discriminator on the real sample, z is the input noise of the generator, which comes from the predefined noise distribution p z (z), D c () is the discriminator function, G c () is the generator function, G c (z) is the fake geological data generated by the generator; In the adversarial training stage, the discriminator output in the loss function is the judgment result D of the real sample c (x) and the judgment result D of the discriminator on the fake geological data generated by the generator c (G c (z)) is implemented by a neural network layer with a Sigmoid activation function to ensure that the output value is between 0 and 1, expressed as: In the formula, f c () is the linear output of the discriminator network, is the weight of the discriminator; S203. In each round of training, the learning rates of the generator and the discriminator are dynamically adjusted according to the feedback of the discriminator. When the accuracy of the discriminator is too high, the learning rate of the generator is increased and the learning rate of the discriminator is reduced, and vice versa. This dynamic adjustment helps to avoid mode collapse during training and ensure the diversity of geological data generation. Dynamic gradient adjustment is achieved by adjusting the learning rate. The adjustment strategy is expressed as: In the formula, is the learning rate of the generator, η0 is the basic learning rate, is the adjustment coefficient of the generator learning rate, ∥∥ is the L2 norm, is the gradient of the generator’s loss function with respect to its parameters, is the loss function of the generator, is the learning rate of the discriminator, is the adjustment coefficient of the discriminator learning rate, is the gradient of the discriminator’s loss function with respect to its parameters, is the loss function of the discriminator; S204, using the self-similarity principle to guide the generator to focus on the local similarity of geological data, improve the quality and practicality of the fake geological data, and the self-similarity adjustment guides the model to pay attention to the self-similarity characteristics of local geological data by adding a regularization term in the loss function of the generator. The calculation formula of the regularization term is: R c (G c (z))=β c ∑ i,j ∥G c (With) i -G c (With) j ∥+s c (G c (With)), In the formula, R c () is the self-similarity regularization term calculation function, G c (z) is the fake geological data generated by the generator, β c is the regularization coefficient of the generated adversarial network, G c (z) i is the i-th characteristic region in the false geological data, G c (z) j is the jth characteristic area in the false geological data, s c () is the self-similarity descriptor calculation function; S205, repeating steps S202 to S204 until a preset condition for stopping iteration is met.

3. The method for training a mineral resource identification model based on unmanned aerial vehicle data collection according to claim 2 is characterized in that: In step S202 Said The calculation formulas are: In the formula, is the weight of the discriminator network, is the bias of the discriminator network, x i is the i-th characteristic attribute of the real geological data sample, G c,i (z) is the i-th output of the generator.

4. The method for training a mineral resource identification model based on data collected by an unmanned aerial vehicle according to any one of claims 1 to 3, characterized in that: The feature extraction model constructed by training the autoencoder algorithm based on the dual constraints described in S3 includes: S311, geological data passes through multiple hidden layers from the input layer through the autoencoder, each layer undergoes linear transformation and nonlinear activation, and finally outputs the extracted features. The transfer function of each layer is expressed as: x p,i+1 =Sig(W p,i x p,i +b p,i ), In the formula, x p,i+1 is the output vector of the i+1th layer of the autoencoder, Sig() is the Sigmoid activation function, and W p,i is the weight matrix of the i-th layer of the autoencoder, x p,i is the input vector of the i-th layer of the autoencoder, b p,i is the bias vector of the autoencoder; S312. The loss function of the autoencoder is designed as a dual constraint loss. One part comes from the reconstruction loss of the geological data content, ensuring that the extracted features can effectively reconstruct the original geological data; the other part is the structure preservation loss, ensuring that the original structure of the geological data is maintained in the feature space. The loss function calculation formula of the autoencoder is: L p =λ p1 L p,recon +λ p2 L p,struct , Where, L p is the loss function of the autoencoder, λ p1 To adjust the weight coefficient of the relative importance of reconstruction loss, L p,recon is the reconstruction loss, λ p2 The weight coefficient to adjust the structure to maintain the relative importance of losses, L p,struct To maintain losses for the structure; S313. Calculate the gradient according to the loss function and use the least squares gradient descent method to update the network parameters, considering the stability of the gradient and the adaptability of the update rate to avoid the gradient explosion or disappearance problem in high-dimensional space. The parameter update rule is: In the formula, θ p,i is the parameter of the i-th layer of the autoencoder, ← is the parameter update operation, η p is the learning rate of the autoencoder, is the loss function L of the autoencoder p The gradient of the parameters of the i-th layer of the autoencoder; S314, repeating steps S311 to S313 until a preset stop iteration condition is met.

5. The method for training a mineral resource identification model based on unmanned aerial vehicle data collection according to claim 4 is characterized in that: 4 The reconstruction loss L in S312 p,recon The calculation formula is: In the formula, x p is the original input geological data, is the output of the autoencoder; The structure retention loss L p,struct The calculation formula is: L p,struct =∑ j,k IN p,jk ∥z p,j -With p,k ∥ 2 , Where W p,jk is the similarity matrix based on the input geological data, z p,j is the jth geological data point mapped to the feature space, z p,k is the kth geological data point mapped to the feature space.

6. The method for training a mineral resource identification model based on data collected by an unmanned aerial vehicle according to any one of claims 1 to 3, characterized in that: The feature extraction model constructed by training a neural network algorithm based on dynamic population evolution optimization in S3 includes: S321. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, each individual represents a configuration of network weights, and the population size is N. p , initialize the weights and biases for the i-th individual, expressed as: In the formula, is the weight matrix of the i-th individual in the initial state, ~ is subject to a specific distribution, is a normal distribution; The mean is 0 and the variance is σ 2 The normal distribution of 2 is the initialization variance, is the bias of the i-th individual in the initial state; W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual; S322. For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. For the i-th individual, use its weight and bias to calculate the loss on the training data set, expressed as: Where, L pi is the loss of the neural network corresponding to the i-th individual, mps is the number of samples input in the current batch, is the composite loss function, fsig() is the neural network model function, is the characteristic of the jth geological data sample, W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, is the label of the jth geological data sample; S323. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. The selection is based on the fitness of the individual. The excellent individual has a higher probability of being selected. The calculation formula for the probability of an individual being selected is: Where P select (i) is the probability of the i-th individual being selected, γ pse To control the selected pressure parameter, L pi is the loss of the neural network corresponding to the i-th individual, N p is the population size, L pk is the loss of the neural network corresponding to the kth individual; S324. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. The crossover operation randomly selects two individuals for gene exchange, which is expressed as: W′ pi =a pcs W p1 +(1-a pcs )W p2 , b′ pi =a pcs b p1 +(1-a pcs )b p2 , Where W′ pi is the weight of the neural network corresponding to the individual after the crossover operation, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first individual selected, W p2 is the weight of the neural network corresponding to the second individual selected, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation, b p1 is the bias of the neural network corresponding to the first individual selected, b p2 is the bias of the neural network corresponding to the selected second individual; The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as: ″′ Where W pi is the weight of the neural network corresponding to the individual after the mutation operation, W pi is the weight of the neural network corresponding to the individual after the crossover operation, The mean is 0 and the variance is τ 2 The normal distribution of 2 ″′ is the variance of the variation, b pi is the bias of the neural network corresponding to the individual after the mutation operation, b pi is the bias of the neural network corresponding to the individual after the crossover operation; S325, repeating steps S322 to S324 until a preset stop iteration condition is met.

7. The method for training a mineral resource identification model based on unmanned aerial vehicle data collection according to claim 6 is characterized in that: In the S322, the The calculation formula is: In the formula, is the composite loss function, fsig() is the neural network model function, is the characteristic of the jth geological data sample, W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, is the label of the jth geological data sample, MSE() is the mean square error function, λ ps is the regularization parameter, Reg(W pi ) is the regularization term; The regularization term Reg(W pi ) is calculated as: Where W pi,k is the kth weight of the neural network corresponding to the ith individual.

8. The method for training a mineral resource identification model based on data collected by an unmanned aerial vehicle according to any one of claims 1 to 3, characterized in that: The classification model constructed by the deep neural decision tree classification algorithm based on quantum transformation pruning described in S4 includes: S401, initialize the parameters of the deep neural decision tree, including the weight of the decision node and the depth of the tree, and set the weight of the i-th node in the deep neural decision tree to W u,i , the bias of the i-th node in the deep neural decision tree is b u,i , the initialization method is expressed as: In the formula, is a normal distribution with a mean of 0 and a standard deviation of 1, and ~ is subject to a specific distribution; S402. During the training process, the geological data after input feature extraction passes through the multi-layer decision nodes of the decision tree. Each node makes a decision based on the weight and input feature, guiding the geological data to flow to the left or right subtree until it reaches the leaf node. Let the feature vector input to the deep neural decision tree be x u , expressed as: y u,n =Sig(W u,n x u +b u,n ), In the formula, y u,n is the output of the nth node of the deep neural decision tree; Sig() is the Sigmoid activation function; W u,n is the weight of the nth node in the deep neural decision tree, b u,n is the bias of the nth node in the deep neural decision tree; S403. Considering the classification accuracy and tree complexity, a loss function is used to evaluate the performance of the current decision tree. The loss function of the deep neural decision tree includes classification error and regularization terms to prevent overfitting. The calculation formula is: Where, L u is the loss function of the deep neural decision tree, N u is the number of samples input to the deep neural decision tree in the current batch, t i is the target output of the i-th sample, y u,i is the classification result of the deep neural decision tree for the i-th sample, λ u is the regularization coefficient of the deep neural decision tree, W u is the weight of the deep neural decision tree; ∥∥ is the L2 norm; S404. In each training cycle, the utility and efficiency of each decision node are evaluated according to the quantum transformation principle. Inefficient nodes will be pruned to reduce the complexity of the model while maintaining or improving the performance of the model. The present invention uses the quantum transformation principle to evaluate and adjust the structure of the network. Suppose P u,i Indicates the importance of the i-th node. If P u,i Less than the preset threshold value θ u , then the node will be removed from the model, P u,i The calculation formula is: Where, L u is the loss function of the deep neural decision tree, W u,i is the weight of the i-th node in the deep neural decision tree, ψ r () is the quantum transformation function; Z r is the input feature vector of the current node of the deep neural decision tree; S405, repeating steps S402 to S404 until a preset stop iteration condition is met, indicating that the model training is completed.

9. The method for training a mineral resource identification model based on data collected by an unmanned aerial vehicle according to claim 8, characterized in that: The ψ r (z r ) is calculated as: In the formula, Z r is the input feature vector of the current node of the deep neural decision tree, λ r is the parameter that controls the sensitivity of quantum transformation, z rj For z r The jth element of corresponds to z rj The quantum noise level of The quantum noise level has different compression levels in different parts of the feature representation, thereby more finely controlling the information loss, and the calculation formula is: In the formula, α r To adjust the coefficient of quantum noise response, z rj For z r The j-th element of .

Images