Metallization prediction method and system based on fine sample segmentation and small sample learning strategy

Abstract

The invention belongs to the technical field of metallogenic prediction, and discloses a metallogenic prediction method based on fine sample segmentation and a small sample learning strategy. According to the method, a robust small sample learning framework is provided, the capacity of labeled data is expanded through data enhancement, and the model is pre-trained and finely adjusted by adopting transfer learning, so that the model has excellent prediction efficiency, the problem of overfitting of the model is inhibited to a certain extent, and the uncertainty of model prediction is reduced. The small sample learning framework balances the data set through SMOTE, and the classification precision of the data set in the training process is improved. The prediction points of the small sample learning framework are arched, which is beneficial for focusing potential targets with high probability and low uncertainty. The small sample learning framework has the ability to control overfitting and produces a high confidence exploration target with a low level of uncertainty.

Description

Technical Field

[0001] The present invention belongs to the technical field of mineralization prediction, and in particular relates to a mineralization prediction method and system of fine sample segmentation and small sample learning strategy. Background Art

[0002] Data-driven mineralization prediction based on supervised learning is a typical small-sample task, which is affected by the lack of labeled data, overfitting of the model and uncertainty of prediction. Small-sample learning technology can expand the amount of labeled data through the data augmentation technology of synthetic minority oversampling technique (SMOTE), and can transfer the knowledge learned from a large amount of data to the scarce data of the target. Therefore, a robust small-sample learning framework is proposed to effectively solve the problems existing in the data-driven mineralization prediction task based on supervised learning.

[0003] Data-driven mineralization prediction based on supervised learning faces three major challenges. First, the training of the model relies on a labeled dataset consisting of a large number of positive and negative samples. Second, sophisticated machine learning models will show overfitting in mineralization prediction tasks with scarce data and unbalanced datasets. In any case study of mineralization prediction, the number of non-ore points (negative samples) greatly exceeds that of mineral points (positive samples), which leads to serious dataset imbalance. In addition, due to incomplete understanding of the mineral system, deviations in exploration and measurement, and inherent and random errors in training and applying machine learning models, the prediction results are uncertain.

[0004] Through the above analysis, the problems and defects of the prior art are as follows:

[0005] Data-driven mineralization prediction with supervised learning is a typical small-sample task, which is affected by the lack of labeled data, model overfitting and prediction uncertainty. Summary of the invention

[0006] In view of the problems existing in the prior art, the present invention provides a mineralization prediction method with fine sample segmentation and small sample learning strategy.

[0007] The present invention is implemented in this way: a mineralization prediction method of fine sample segmentation and small sample learning strategy includes:

[0008] S1, obtain mineralization information dataset;

[0009] For the mineralization prediction task, the dataset consists of two parts: the evidence layer and the target variable;

[0010] S2, building a small sample learning framework;

[0011] The small sample learning framework consists of two parts: the first is the data augmentation module to solve the imbalanced data set, and the second is the transfer learning module to train the prediction model using limited labeled data;

[0012] S3, evaluation of the model;

[0013] S4, output prediction results;

[0014] S5, result analysis.

[0015] Further, the mineralization information data set is obtained:

[0016] For the mineralization prediction task, the dataset consists of two parts: the evidence layer and the target variable;

[0017] (a) Evidence layers: An analysis using a mineral systems approach maps the understanding of the mineral system into a set of evidence layers that reflect spatial relationships. The layers can present the key mineralization processes related to mineral source, transport, physical entrapment and chemical deposition;

[0018] (b) Target variable: The explored mineral deposits and known mineral points are obtained as positive samples; negative samples are all non-mineral points. The following principles should be considered when selecting:

[0019] (i) Non-mining sites should be randomly distributed in space;

[0020] (ii) When selecting non-mining point locations, the locations of non-mining points should be located outside a certain range of known mining points as much as possible; therefore, a buffer distance with a high probability of finding at least one other mining point is selected to establish a buffer zone; and the area outside the buffer zone within the study area is used as the area for selecting non-mining point locations.

[0021] Further, the small sample learning framework is constructed:

[0022] The small sample learning framework consists of two parts: one is the data enhancement module to solve the imbalanced data set, and the other is the transfer learning module that uses limited labeled data to train the prediction model. The specific steps of implementation are as follows:

[0023] (a) Data augmentation module, operates as follows:

[0024] (i) Prepare a data set; divide the labeled data set into two categories according to the attribute value of the label (0 and 1), where 0 represents non-mining points and 1 represents mining points;

[0025] (ii) Count the number of samples in the non-mining data set and the number of samples in the mining data set;

[0026] (iii) The original non-mining point data and mining point data are randomly split into three parts, which are used as the test data set, the target domain, and the data set to be processed by SMOTE. 20% and 15% of the mining point samples are randomly selected from the original data set and combined with the same number of randomly selected non-mining point samples to form the test data set and the target domain respectively. Then, SOMTE is performed on the remaining mining point samples and non-mining point samples to generate the source domain to balance the number of positive and negative samples and solve the class imbalance problem.

[0027] (iv) Perform the random division in (iii) 10 times;

[0028] (b) The transfer learning module operates as follows:

[0029] (i) Random sampling with replacement is used to randomly sample the data in the source domain to generate N subsets, namely the support set;

[0030] (ii) N support sets are trained using a pre-training model consisting of N neural networks. The purpose of pre-training is to allow the model to learn some common features and knowledge in the source domain, and then adapt the model to the target domain data;

[0031] (iii) Use samples from the target domain to fine-tune the parameters of the pre-trained model so that the model reaches the optimal state to adapt to the target domain.

[0032] Further, the model is evaluated:

[0033] (a) Evaluation indicators: Calculate the accuracy, recall, F1-score and other indicators of the model to quickly evaluate the effect of parameter optimization; however, this indicator cannot evaluate the predictive performance of the model;

[0034] (b) Plotting of receiver operating characteristic (ROC) curve: The ROC curve can be used to evaluate the predictive performance of the model;

[0035] (i) Calculate the true positive rate TPR and false positive rate FPR;

[0036] (ii) Set multiple thresholds in the output probability of the model and calculate the corresponding TPR and FPR;

[0037] (iii) Draw ROC curve using programming language;

[0038] (c) Plotting the success rate curve

[0039] (i) Calculate the success rate; the success rate refers to the accuracy of positive sample prediction under a certain prediction threshold;

[0040] (ii) setting multiple thresholds in the model output probability and calculating the success rate under each threshold;

[0041] (iii) Plot the success rate curve using programming language.

[0042] Further, the output prediction result is:

[0043] (a) Prediction of mineralization probability: Apply the trained model to the characteristic data of the area to be evaluated to generate a mineralization potential score for each pixel unit;

[0044] (b) Spatial interpolation: The mineralization potential scores of pixel units are interpolated using the Kriging method to convert the discrete potential scores into a continuous potential score map;

[0045] (c) Potential classification: Thresholds can be set based on the scoring results to classify the regions into different potential levels, and then the results can be visualized.

[0046] Further, the result analysis:

[0047] (a) Analysis and discussion: Combined with the geological background and other relevant data, the results of the potential map are analyzed to discuss the possible mineralization mechanism and its guiding significance for resource exploration;

[0048] (b) Result verification: Compare the mineralization prediction map with known mineral sites or historical data to verify the accuracy of the model prediction; if possible, conduct field surveys and collect new data for further verification.

[0049] Another object of the present invention is to provide a mineralization prediction system with fine sample segmentation and small sample learning strategy, including:

[0050] The data set acquisition module is used to obtain the mineralization information data set; for the mineralization prediction task, the data set consists of two parts: the evidence layer and the target variable;

[0051] The construction module is used to build a small sample learning framework; the small sample learning framework consists of two parts: one is the data enhancement module to solve the imbalanced data set, and the other is the transfer learning module that uses limited labeled data to train the prediction model;

[0052] Evaluation module, used for model evaluation;

[0053] Output module, used to output prediction results;

[0054] Analysis module, used for result analysis.

[0055] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the mineralization prediction method of the fine sample segmentation and small sample learning strategy.

[0056] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the mineralization prediction method of the fine sample segmentation and small sample learning strategy.

[0057] Another object of the present invention is to provide an information data processing terminal, which is used to implement the mineralization prediction system of the fine sample segmentation and small sample learning strategy.

[0058] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0059] First, the present invention proposes a robust small sample learning framework, which expands the capacity of labeled data through data enhancement, and adopts transfer learning to pre-train and fine-tune the model, so that the model has excellent prediction efficiency, suppresses the overfitting problem of the model to a certain extent, and reduces the uncertainty of model prediction. The small sample learning framework balances the data set through SMOTE, which improves the classification accuracy of the data set during the training process. The prediction points of the small sample learning framework are arched, which is conducive to focusing on potential targets with high probability and low uncertainty. The small sample learning framework has the ability to control overfitting and produces high-confidence exploration targets with low uncertainty levels.

[0060] The present invention provides a mineralization prediction method with fine sample segmentation and small sample learning strategy. Through a series of technical steps, the present invention can solve the problems of data scarcity and sample imbalance in mineralization prediction tasks, and at the same time can suppress the overfitting problem of the model to a certain extent and reduce the uncertainty of model prediction.

[0061] Second, whether the technical solution of the present invention solves the technical problems that people have been eager to solve but have never been able to solve successfully:

[0062] Most artificial intelligence mineralization predictions are a kind of supervised learning. However, due to the high scarcity of mineral points (often only dozens in one area), compared with the tens of thousands of samples in general field machine learning, the training samples for mineralization prediction are seriously insufficient, which is the biggest bottleneck restricting the improvement of artificial intelligence mineralization prediction accuracy. The present invention integrates the fine sample segmentation scheme for mineralization prediction and the highly targeted small sample learning strategy, and carries out supervised learning mineralization prediction on the basis of a reasonable and effective data enhancement framework, which solves the problem of insufficient sample size and improves the accuracy of mineralization prediction.

[0063] Third, the mineralization prediction method based on fine sample segmentation and small sample learning strategy proposed in the present invention, through innovative data enhancement and transfer learning module design, effectively solves the problems of sample imbalance and insufficient small sample data in the existing technology in industrial applications, and significantly improves the accuracy and application value of mineralization prediction.

[0064] 1. Solved the problems of sample imbalance and insufficient sample size

[0065] In geological exploration, mineralization samples are usually few and unevenly distributed, and traditional models find it difficult to extract effective mineralization features from rare samples. The present invention generates more samples through a data enhancement module, balancing the data ratio of mineralization and non-mineralization areas. This strategy ensures that the model can learn the features under a variety of mineralization conditions, greatly improving the model's adaptability to small sample data, thereby achieving efficient prediction under limited sample conditions.

[0066] 2. Realized the innovative application of transfer learning in mineralization prediction

[0067] The present invention uses a transfer learning module to transfer existing knowledge from other fields to the mineralization prediction task, thus solving the modeling problem caused by insufficient number of geological samples. The introduction of transfer learning not only improves the prediction accuracy of the model, but also significantly reduces the training time of the mineralization prediction model, enhances the adaptability of the model in different mining environments, and provides an efficient solution for geological prospecting.

[0068] 3. Improved the generalization ability and stability of the mineralization prediction model

[0069] Traditional models are prone to instability due to small sample size or uneven data distribution. The present invention uses dual optimization of data enhancement and transfer learning to enable the model to maintain high prediction accuracy in different mining areas and under different mineralization conditions. Such high generalization ability and stability can effectively cope with the complexity of geological conditions in mining areas and improve the reliability and scalability of mineralization prediction.

[0070] 4. Reduced prospecting costs and exploration risks

[0071] The present invention provides a scientific basis for the delineation of mining areas through accurate mineralization prediction, helping users to identify mineralization target areas more efficiently. Compared with traditional exploration methods, this technology greatly reduces the blindness of prospecting and effectively reduces unnecessary exploration costs. At the same time, accurate prediction improves the success rate of mining area exploration, significantly reduces exploration risks, and brings direct economic benefits to resource development companies.

[0072] 5. Significantly enhanced the application value of mineralization prediction

[0073] The present invention provides an intuitive mineralization probability distribution map and mineralization target area positioning, allowing geological experts and decision makers to interpret the prediction results more easily. Visualized mineralization information improves the applicability of the results and provides a reliable scientific basis for decision-makers. This mineralization prediction method can be widely used in mining area exploration, mining rights investment and other fields, providing important support for the rational development and utilization of mineral resources.

[0074] In summary, the technical solution of the present invention has achieved significant technical progress in sample imbalance processing, model optimization, and improvement of prediction accuracy, solving the problem of insufficient prediction accuracy caused by limited and unbalanced samples in the prior art, and providing efficient and practical technical support for geological prospecting and mineral resource development. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is a flow chart of the mineralization prediction method of fine sample segmentation and small sample learning strategy provided by an embodiment of the present invention.

[0076] Figure 2 It is a flow chart of a result analysis method provided by an embodiment of the present invention.

[0077] Figure 3 It is a structural block diagram of the mineralization prediction system of the fine sample segmentation and small sample learning strategy provided by the embodiment of the present invention.

[0078] Figure 4 It is an example of the data enhancement and transfer learning method framework provided by an embodiment of the present invention.

[0079] Figure 5 This is a comparison chart of the model performance of the small sample learning method model provided by an embodiment of the present invention and the support vector machine model of traditional machine learning when performing mineralization prediction using the same data set.

[0080] Figure 6 The output results of mineralization prediction provided by the embodiments of the present invention are respectively outputted using (a) a small sample learning method; (b) a support vector machine; and (c) an artificial neural network. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0082] like Figure 1 As shown, a mineralization prediction method of a fine sample segmentation and small sample learning strategy provided by an embodiment of the present invention comprises the following steps:

[0083] S1, obtain mineralization information dataset;

[0084] For the mineralization prediction task, the dataset consists of two parts: the evidence layer and the target variable;

[0085] S2, building a small sample learning framework;

[0086] The small sample learning framework consists of two parts: the first is the data augmentation module to solve the imbalanced data set, and the second is the transfer learning module to train the prediction model using limited labeled data;

[0087] S3, evaluation of the model;

[0088] S4, output prediction results;

[0089] S5, result analysis.

[0090] The mineralization prediction method of the present invention starts with a mineralization information data set, which consists of two parts: an evidence layer and a target variable. The evidence layer contains data on key geological features in the mineralization process, and the target variable marks the known mineralization locations. The data set is collected through geological surveys and other means, covering a variety of information required for mineralization prediction tasks. The data will be preprocessed at this stage to ensure its integrity and consistency, providing high-quality input data for the subsequent small sample learning framework.

[0091] In order to effectively deal with the problem of small number of samples, the present invention proposes a small sample learning framework. The framework consists of two modules: data enhancement module and transfer learning module. The data enhancement module mainly targets the sample imbalance problem in the data set, and generates more training samples to improve the robustness of the model through data expansion and transformation. The transfer learning module loads limited labeled data on a model pre-trained on a similar task, so that the model still has good prediction ability in the case of limited samples.

[0092] The data enhancement module plays a key role in the small sample learning framework. In order to solve the problem of small and unbalanced samples of mineralization locations in the data set, it uses a variety of methods such as image transformation and noise addition to generate more training samples. Through these operations, the data enhancement module expands the scale of the data set without increasing the number of real samples, thereby effectively solving the problem of data imbalance, enabling the model to learn the characteristics under different mineralization conditions more comprehensively and improve prediction accuracy.

[0093] The transfer learning module is another important component of the small sample learning framework. Since there are few labeled samples of mineralization data, the transfer learning module transfers the existing knowledge to the current mineralization prediction task by using the model that has been trained on related tasks. This module fine-tunes the parameters and adjusts the adaptability of the model to the current task, so that under the condition of limited samples, the model can still show a high prediction accuracy, thereby improving the efficiency of small sample learning.

[0094] After the model training is completed, the performance of the model is evaluated. The prediction ability of the model is verified through multiple indicators (such as accuracy, recall rate, etc.) to ensure that the model has good generalization ability. The evaluated model predicts the unlabeled mineralization area data, outputs the mineralization probability distribution map and mineralization target area, and forms a complete prediction result. These prediction results provide a scientific basis for actual prospecting and help to delineate possible mineralization areas.

[0095] Finally, the mineralization prediction results are analyzed and compared with known mineralization areas to evaluate the reliability and accuracy of the model prediction. Through comparative analysis and feedback adjustment, the data enhancement strategy and model migration parameters can be continuously optimized to improve the practicality of the mineralization prediction model. The results provide reference information for geological exploration, making mineralization prediction more instructive and improving the scientificity and efficiency of resource development.

[0096] The mineralization prediction method proposed in the embodiment of the present invention effectively improves the mineralization prediction accuracy under unbalanced data sets through fine sample segmentation and small sample learning strategy. The specific working principle is divided into the following six key steps:

[0097] 1. Obtaining the mineralization information data set: The system first collects the data set related to the mineralization prediction task, which consists of two parts: the evidence layer and the target variable. The evidence layer includes multi-dimensional geological data such as geology, geophysics, and geochemistry as the input features of the prediction task, while the target variable identifies the mineralization area or the non-ore-forming area. The integrity and quality of the data set directly affect the effect of mineralization prediction.

[0098] 2. Fine sample segmentation strategy: In order to improve the prediction accuracy, the system finely segments the mineralization data. According to different geological characteristics and geographical locations, the data set is divided into multiple small sample groups by region or specific characteristics, and the data stratification method is used to ensure the consistency of data characteristics between each sample group. This segmentation method not only improves the balance of the data, but also provides basic support for subsequent small sample learning.

[0099] 3. Construction of small sample learning framework: The constructed small sample learning framework includes data enhancement module and transfer learning module. The data enhancement module performs data expansion for a small number of mineralization samples in the data set, generates more samples, and balances the data distribution; the transfer learning module uses labeled data with similar geological characteristics to train the model, transfers knowledge to new data, and realizes the mineralization prediction of small samples. Transfer learning greatly reduces the demand for labeled samples and adapts to the situation where there are few and unbalanced samples for mineralization prediction.

[0100] 4. Model training and evaluation: The system inputs the processed data into the prediction model for training. By using deep learning models (such as convolutional neural networks or long short-term memory networks), the model can learn the complex characteristics of the mineralization area. After training, the system uses cross-validation and indicators such as accuracy and recall to evaluate the prediction effect of the model to ensure the reliability and stability of the model.

[0101] 5. Output of prediction results: After model evaluation, the system conducts mineralization prediction in the target area and outputs a probability distribution map of the mineralization area. The map clearly shows the areas with high mineralization potential, helping geological prospectors to determine the priority exploration areas. At the same time, the system supports the output of detailed attribute data for each predicted area, which is convenient for subsequent analysis and verification.

[0102] 6. Result analysis and feedback: Finally, the system analyzes and verifies the prediction results to obtain the correlation information between geological characteristics and mineralization potential. Combined with the predicted mineralization area, geological experts can further adjust the model parameters or introduce new geological data to form a closed-loop feedback. This analysis step ensures that the prediction model can be dynamically optimized according to geological changes, and continuously improve the accuracy and adaptability of mineralization prediction.

[0103] This method achieves efficient mineralization prediction as a whole, while effectively addressing the problems of small samples and data imbalance, and improving the accuracy of geological exploration.

[0104] The embodiment of the present invention provides a method for obtaining a mineralization information data set:

[0105] For the mineralization prediction task, the dataset consists of two parts: the evidence layer and the target variable;

[0106] (a) Evidence layers: The analysis using the mineral system approach maps the understanding of the mineral system into a set of evidence layers that reflect spatial relationships. The layers can present the key mineralization processes related to the source, transport, physical closure and chemical deposition of minerals. The content can be fault zone density, fault zone intersection density, magnetic anomalies, gravity anomalies, element geochemical anomalies, etc. The resolution of the evidence layer can be determined based on the resolution of the evidence map and the distance between the two nearest mineral deposits.

[0107] (b) Target variable: The explored mineral deposits and known mineral points are obtained as positive samples; negative samples are all non-mineral points. The following principles should be considered when selecting:

[0108] (i) Non-mining sites should be randomly distributed in space;

[0109] (ii) When selecting non-mining point locations, the locations of non-mining points should be located outside a certain range of known mining points as much as possible; therefore, a buffer distance with a relatively high probability (such as 85%) of finding at least one other mining point is selected to establish a buffer zone; and the area outside the buffer zone in the study area is used as the area for selecting non-mining point locations.

[0110] The small sample learning framework provided by the embodiment of the present invention is as follows:

[0111] The proposed small sample learning framework consists of two parts: one is the data augmentation module to solve the imbalanced data set, and the other is the transfer learning module to train the prediction model with limited labeled data. The specific steps of implementation are as follows:

[0112] (a) Data augmentation module, operates as follows:

[0113] (i) Prepare a data set; divide the labeled data set into two categories according to the attribute value of the label (0 and 1), where 0 represents non-mining points and 1 represents mining points;

[0114] (ii) Count the number of samples in the non-mining data set and the number of samples in the mining data set;

[0115] (iii) The original non-mining point data and mining point data are randomly split into three parts, which are used as the test data set, the target domain, and the data set to be processed by SMOTE. 20% and 15% of the mining point samples are randomly selected from the original data set and combined with the same number of randomly selected non-mining point samples to form the test data set and the target domain respectively. Then, SOMTE is performed on the remaining mining point samples and non-mining point samples to generate the source domain to balance the number of positive and negative samples and solve the class imbalance problem.

[0116] (iv) Perform the random division in (iii) 10 times;

[0117] (b) The transfer learning module operates as follows:

[0118] (i) Random sampling with replacement is used to randomly sample the data in the source domain to generate N subsets, namely the support set;

[0119] (ii) N support sets are trained using a pre-training model consisting of N neural networks. The purpose of pre-training is to allow the model to learn some common features and knowledge in the source domain, and then adapt the model to the target domain data;

[0120] (iii) Use samples from the target domain to fine-tune the parameters of the pre-trained model so that the model reaches the optimal state to adapt to the target domain.

[0121] Evaluation of the model provided by the embodiment of the present invention:

[0122] (a) Evaluation indicators: Calculate the accuracy, recall, F1-score and other indicators of the model to quickly evaluate the effect of parameter optimization; however, this indicator cannot evaluate the predictive performance of the model;

[0123] (b) Plotting of receiver operating characteristic (ROC) curve: The ROC curve can be used to evaluate the predictive performance of the model;

[0124] (i) Calculate the true positive rate TPR and false positive rate FPR;

[0125] (ii) Set multiple thresholds (from 0 to 1) in the output probability of the model and calculate the corresponding TPR and FPR;

[0126] (iii) Draw ROC curve using programming language;

[0127] (c) Plotting the success rate curve

[0128] (i) Calculate the success rate; the success rate refers to the accuracy of positive sample prediction under a certain prediction threshold;

[0129] (ii) setting multiple thresholds in the model output probability and calculating the success rate under each threshold;

[0130] (iii) Plot the success rate curve using programming language.

[0131] The output prediction results provided by the embodiment of the present invention are:

[0132] (a) Mineralization probability prediction: Apply the trained model to the characteristic data of the area to be evaluated to generate a mineralization potential score (probability value) for each pixel unit;

[0133] (b) Spatial interpolation: The mineralization potential scores of pixel units are interpolated using the Kriging method to convert the discrete potential scores into a continuous potential score map;

[0134] (c) Potential classification: Thresholds can be set based on the scoring results to classify regions into different potential levels (high, medium, and low), and the results can be visualized;

[0135] like Figure 2 As shown, the results analysis provided by the embodiments of the present invention are as follows:

[0136] S501, Analysis and Discussion: Combined with geological background and other relevant data, analyze the results of the potential map, discuss possible mineralization mechanisms and their guiding significance for resource exploration;

[0137] S502, result verification: compare the mineralization prediction map with known mineral points or historical data to verify the accuracy of the model prediction; if possible, conduct field surveys and collect new data for further verification;

[0138] like Figure 3 As shown, a mineralization prediction system with fine sample segmentation and small sample learning strategy provided by an embodiment of the present invention includes:

[0139] The data set acquisition module is used to obtain the mineralization information data set; for the mineralization prediction task, the data set consists of two parts: the evidence layer and the target variable;

[0140] The construction module is used to build a small sample learning framework; the small sample learning framework consists of two parts: one is the data enhancement module to solve the imbalanced data set, and the other is the transfer learning module that uses limited labeled data to train the prediction model;

[0141] Evaluation module, used for model evaluation;

[0142] Output module, used to output prediction results;

[0143] Analysis module, used for result analysis.

[0144] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the mineralization prediction method of the fine sample segmentation and small sample learning strategy.

[0145] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the mineralization prediction method of the fine sample segmentation and small sample learning strategy.

[0146] Another object of the present invention is to provide an information data processing terminal, which is used to implement the mineralization prediction system of the fine sample segmentation and small sample learning strategy.

[0147] The present invention is widely applicable to many fields such as mineral resource exploration and ore deposit evaluation, and provides an efficient and intelligent sample segmentation solution and a small sample learning framework, providing important data support for geological science research.

[0148] The present invention can be applied to the fields of mineral resource exploration, mineral prediction, etc. The present invention provides a computer software system model for intelligent mineral prediction, which improves the accuracy of mineralization prediction by realizing fine sample segmentation and small sample learning.

[0149] like Figure 4 As shown, the technical framework in the implementation process of the present invention can be divided into two parts: "data enhancement" and "transfer learning". The "data enhancement" part provides an algorithm for fine sample segmentation, which divides the 118 mining point samples (positive samples) data into three parts, and uses the 24 positive samples divided as test sets, 18 positive samples as target domain data sets, and the remaining 76 positive samples are used to expand the number of samples to 304. In order to achieve sample balance, the non-mining point samples (negative samples) use the same fine sample segmentation algorithm, dividing 24 negative samples as test sets, 18 samples as target domain data sets, and the remaining 304 negative samples and the 304 positive samples after data enhancement are used together as source domain data sets.

[0150] In the embodiment of the present invention, the fine sample segmentation and small sample learning method provided and the support vector machine (SVM) algorithm of traditional machine learning are used to perform mineralization prediction on the test sets of 10 different data sets and make performance comparisons. The comparison results are as follows: Figure 5 The comparison is based on the AUC value of the model to evaluate the performance. Figure 5 It can be seen that the small sample learning method proposed in the present invention shows more stable performance than the support vector machine on different data sets, and the AUC value of the small sample learning method model is generally higher than the AUC value of the support vector machine model, indicating that the small sample learning method model provided by the present invention has stronger prediction accuracy and generalization ability than the support vector machine model.

[0151] Figure 6 The following is a comparison chart of the mineralization prediction effect of the small sample learning method proposed in this invention and the support vector machine model and artificial neural network model of traditional machine learning. Figure 6 (a) shows the mineralization prediction results of the support vector machine model. Figure 6 (b) shows the mineralization prediction results of the artificial neural network. Figure 6(c) As shown. Small sample learning successfully predicted 61.86% of the mineral points with a high potential area prediction area of ​​7%, and the prediction efficiency was 8.837; support vector machine successfully predicted 72.03% of the mineral points with a high potential area prediction area of ​​12%, and the prediction efficiency was 6.003; artificial neural network successfully predicted 72.88% of the mineral points with a high potential area prediction area of ​​14%, and the prediction efficiency was 5.206. It shows that the method proposed in the present invention has a significant improvement in the efficiency of mineralization prediction, and a high prediction success rate is guaranteed with a very small area, so that the prospecting prediction area can be reduced, a large amount of exploration costs can be saved, and it has important practicality in prospecting practice.

[0152] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0153] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A mineralization prediction method based on fine sample segmentation and small sample learning strategy, characterized in that: The following steps are involved: Obtaining mineralization information data set: Obtaining mineralization information data set, which consists of two parts: evidence layer and target variable. The evidence layer includes geological, geophysical, and geochemical multi-dimensional geological data. The target variable identifies the mineralization area or the non-ore-forming area. Fine sample segmentation strategy: According to the requirements of mineralization prediction accuracy, the mineralization data is finely segmented, and the data set is divided into multiple small sample groups according to geological characteristics and geographical locations. The consistency of data characteristics between sample groups is ensured through data stratification method to improve data balance; Construction of small sample learning framework: Construction of a small sample learning framework, including a data enhancement module and a transfer learning module. The data enhancement module performs data expansion on a small number of mineralization samples in the data set, and the transfer learning module uses annotated data with similar geological features to train the prediction model to achieve mineralization prediction for small sample data; Model training and evaluation: Input the processed data into the prediction model for training, and use the deep learning model to complete the learning of the characteristics of the mineralization area; evaluate the prediction effect of the model through cross-validation, accuracy, and recall rate; Output and analysis of prediction results: Carry out mineralization prediction in the target area, output the probability distribution map of the mineralization area, identify the areas with high mineralization potential and output detailed attribute data to support the subsequent analysis and verification of the results.

2. The method for mineralization prediction based on fine sample segmentation and small sample learning strategy as claimed in claim 1, characterized in that: The obtaining of mineralization information data set: For the mineralization prediction task, the dataset consists of two parts: the evidence layer and the target variable; (a) Evidence layers: An analysis using a mineral systems approach maps the understanding of the mineral system into a set of evidence layers that reflect spatial relationships. The layers can present the key mineralization processes related to mineral source, transport, physical entrapment and chemical deposition; (b) Target variable: The explored mineral deposits and known mineral points are obtained as positive samples; negative samples are all non-mineral points. The following principles should be considered when selecting: (i) Non-mining sites should be randomly distributed in space; (ii) When selecting non-mining point locations, the locations of non-mining points should be located outside a certain range of known mining points as much as possible; therefore, a buffer distance with a high probability of finding at least one other mining point is selected to establish a buffer zone; and the area outside the buffer zone within the study area is used as the area for selecting non-mining point locations.

3. The method for mineralization prediction based on fine sample segmentation and small sample learning strategy as claimed in claim 1, characterized in that: The small sample learning framework is constructed as follows: The small sample learning framework consists of two parts: the first is the data augmentation module to solve the imbalanced data set, and the second is the transfer learning module to train the prediction model using limited labeled data; The specific steps to achieve this are as follows: (a) Data augmentation module, operates as follows: (i) Prepare a data set; divide the labeled data set into two categories according to the attribute value of the label (0 and 1), where 0 represents non-mining points and 1 represents mining points; (ii) Count the number of samples in the non-mining data set and the number of samples in the mining data set; (iii) The original non-mining point data and mining point data are randomly split into three parts, which are used as the test data set, the target domain, and the data set to be processed by SMOTE; 20% and 15% of the mining samples are randomly selected from the original dataset and combined with the same number of randomly selected non-mining samples to form the test dataset and target domain respectively. Then, SOMTE is performed on the remaining mining samples and non-mining samples to generate the source domain to balance the number of positive and negative samples and solve the problem of class imbalance. (iv) Perform the random division in (iii) 10 times; (b) The transfer learning module operates as follows: (i) Random sampling with replacement is used to randomly sample the data in the source domain to generate N subsets, namely the support set; (ii) N support sets are trained using a pre-training model consisting of N neural networks. The purpose of pre-training is to allow the model to learn some common features and knowledge in the source domain, and then adapt the model to the target domain data; (iii) Use samples from the target domain to fine-tune the parameters of the pre-trained model so that the model reaches the optimal state to adapt to the target domain.

4. The method for mineralization prediction based on fine sample segmentation and small sample learning strategy as claimed in claim 1, characterized in that: Evaluation of the model: (a) Evaluation indicators: Calculate the model's accuracy, recall, F1-score and other indicators to quickly evaluate the effect of parameter optimization; however, this indicator cannot evaluate the model's predictive performance; (b) Plotting of receiver operating characteristic (ROC) curve: The ROC curve can be used to evaluate the predictive performance of the model; (i) Calculate the true positive rate TPR and false positive rate FPR; (ii) Set multiple thresholds in the output probability of the model and calculate the corresponding TPR and FPR; (iii) Draw ROC curve using programming language; (c) Plotting the success rate curve (i) Calculate the success rate; the success rate refers to the accuracy of positive sample prediction under a certain prediction threshold; (ii) setting multiple thresholds in the model output probability and calculating the success rate under each threshold; (iii) Plot the success rate curve using programming language.

5. The method for mineralization prediction based on fine sample segmentation and small sample learning strategy as claimed in claim 1, characterized in that: The output prediction result is: (a) Prediction of mineralization probability: Apply the trained model to the characteristic data of the area to be evaluated to generate a mineralization potential score for each pixel unit; (b) Spatial interpolation: The mineralization potential scores of pixel units are interpolated using the Kriging method to convert the discrete potential scores into a continuous potential score map; (c) Potential classification: Thresholds can be set based on the scoring results to classify the regions into different potential levels, and then the results can be visualized.

6. The method for mineralization prediction based on fine sample segmentation and small sample learning strategy as claimed in claim 1, characterized in that: Analysis of the results: (a) Analysis and discussion: Combined with the geological background and other relevant data, the results of the potential map are analyzed to discuss the possible mineralization mechanism and its guiding significance for resource exploration; (b) Result verification: Compare the mineralization prediction map with known mineral sites or historical data to verify the accuracy of the model prediction; if possible, conduct field surveys and collect new data for further verification.

7. A mineralization prediction system using fine sample segmentation and small sample learning strategy to implement the mineralization prediction method using fine sample segmentation and small sample learning strategy as described in any one of claims 1 to 6, characterized in that: The mineralization prediction system of the fine sample segmentation and small sample learning strategy includes: The data set acquisition module is used to obtain the mineralization information data set; for the mineralization prediction task, the data set consists of two parts: the evidence layer and the target variable; The construction module is used to build a small sample learning framework. The small sample learning framework consists of two parts: one is the data enhancement module to solve the imbalanced data set, and the other is the transfer learning module that uses limited labeled data to train the prediction model. Evaluation module, used for model evaluation; Output module, used to output prediction results; Analysis module, used for result analysis.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the mineralization prediction method of fine sample segmentation and small sample learning strategy as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the mineralization prediction method of fine sample segmentation and small sample learning strategy as described in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the mineralization prediction system of fine sample segmentation and small sample learning strategy as described in claim 7.

Images