CN-121997152-A - Shallow low-temperature hydrothermal deposit modeling method based on short-wave infrared

CN121997152ACN 121997152 ACN121997152 ACN 121997152ACN-121997152-A

Abstract

The invention discloses a shallow low-temperature hydrothermal deposit modeling method based on short wave infrared, which relates to the technical field of artificial intelligent mineral prediction and comprises the steps of carrying out weighted normalization processing on multi-source heterogeneous geological data, mapping the multi-source heterogeneous geological data to a unified feature space through a nonlinear fusion coding algorithm with a weight coefficient to generate fusion feature vectors, constructing a geological structure perceived self-attention mechanism, introducing a geological priori matrix reflecting terrain gradient and localization abnormal distribution on the basis of keys, inquiry and value matrix, dynamically adjusting local geological coupling weight parameters in the process of attention calculation, establishing a geological consistency loss mechanism, carrying out space constraint on attention response and alteration labels, forming a composite loss structure through a nonlinear amplification item, a logarithmic penalty item and a smooth boundary constraint item, and jointly participating in optimization on the geological consistency loss mechanism, classification loss and regularization item.

Inventors

LI XIAO
CAI WEN
ZHANG LONG
ZHAO WENCHAO
WANG YIHAI
LIU BINBIN
XU CHUNMING
LIU TAO
CHEN LONG

Assignees

根河市比利亚矿业有限责任公司

Dates

Publication Date: 20260508
Application Date: 20251224

Claims (10)

1. The shallow low-temperature hydrothermal deposit modeling method based on short-wave infrared is characterized by comprising the following steps of: carrying out weighted normalization processing on the multi-source heterogeneous geological data, and mapping the multi-source heterogeneous geological data to a unified feature space through a nonlinear fusion coding algorithm with a weight coefficient to generate a fusion feature vector; A self-attention mechanism of geological structure perception is constructed, a geological priori matrix reflecting the terrain gradient and localization abnormal distribution is introduced on the basis of keys, inquiry and value matrix, and local geological coupling weight parameters are dynamically adjusted in the attention calculation process; And (3) establishing a geological consistency loss mechanism, performing space constraint on attention response and a change label, forming a composite loss structure through a nonlinear amplification term, a logarithmic penalty term and a smooth boundary constraint term, and participating the geological consistency loss mechanism, classification loss and regularization term together into optimization.
2. The method for modeling a shallow cryogenic hydrothermal deposit based on short-wave infrared radiation as defined in claim 1, wherein the weighted normalization of the multi-source heterogeneous geological data comprises, Normalizing the spectral reflectivity data according to the wave bands, and normalizing the gray values of different image sources to be in a unified numerical value interval; Carrying out standardization processing on the geochemical element concentration data according to the abnormal sensitivity of each element, wherein the standardization result keeps consistent proportion on a numerical scale; Smoothing the terrain gradient data according to the gradient change rate, wherein the smoothing result is in convergence distribution in a high gradient area so as to reduce the numerical deviation; After normalization, various data respectively correspond to a spectral feature weight coefficient, a localization feature weight coefficient and a topography feature weight coefficient, and are determined by a preset or self-learning mode, and the proportional relation of different data sources in fusion feature calculation is adjusted; And carrying out spatial registration on remote sensing spectrum data, geochemical data and topographic data before weighted fusion, resampling various data to the same grid resolution by adopting a uniform geographic projection coordinate system, enabling the positions of corresponding pixels to correspond to each other in space one by one, and obtaining a primary fusion feature matrix through matrix superposition after weighting, wherein the obtained matrix is used as input of a nonlinear fusion coding step.
3. The method for modeling shallow low-temperature hydrothermal deposit based on short-wave infrared rays of claim 2, wherein the generating of the fusion feature vector comprises the steps of inputting weighted normalized spectrum data, localization data and topography data into a fusion function, obtaining multi-source input feature data through linear superposition, transforming the multi-source input feature data by using a nonlinear activation unit, wherein the nonlinear activation unit adopts a hyperbolic tangent function, and fusion signals are limited in a limited interval under the action of the activation function; Automatically adjusting dynamic weights of the input features according to standard deviation changes of each type of input features, and automatically reducing balance adjustment weight factors of the corresponding input features when the standard deviation fluctuation amplitude of the input features is larger than the standard deviation fluctuation threshold of the input features; outputting the fusion codes into multidimensional feature vectors, wherein each component corresponds to a response value after the combination of input features, and each parameter in the fusion process is updated through back propagation in a training stage.
4. The method for modeling a shallow cryogenic hydrothermal deposit based on short wave infrared light of claim 3, wherein the self-attention mechanism of the geological structure comprises, Establishing a key matrix, a query matrix and a value matrix in a geological priori attention model, calculating the characteristic correlation among all space units, generating a geological priori matrix through joint calculation of a topography gradient and a geochemistry anomaly degree, and representing the space difference of a geological structure; When the attention calculation is executed, a geological prior matrix and an attention similarity matrix are overlapped according to geological coupling weights, so that a comprehensive attention score is formed; and dynamically adjusting the attention weight based on the relative deviation degree of the local value and the global statistical characteristic in the geological priori matrix.
5. The method for modeling a shallow cryogenic hydrothermal deposit based on short wave infrared light of claim 4, wherein dynamically adjusting local geologic coupling weight parameters during the calculation of attention comprises, When the attention weighting operation is executed, monitoring a local difference value between the input characteristic and the geological priori matrix, and taking the local difference value as a self-adaptive adjustment basis of local geological coupling weight parameters; when the local difference value is larger than a set geological difference sensitive threshold, increasing the value of the local geological coupling weight parameter in the range of the corresponding space unit; Continuously updating the local geological coupling weight parameters in a training stage, and determining that the local geological coupling weight parameters are converged to a stable interval according to an optimization criterion in a plurality of iterative processes; Through a dynamic adjustment mechanism, the attention matrix in the geological priori attention model is adapted in each round of calculation according to the spatial distribution characteristics of geological priors, and a weighting matrix structure updated in real time along with the change of geological conditions is formed.
6. The method for modeling a shallow cryogenic hydrothermal deposit based on short wave infrared light of claim 5, wherein establishing a geological consistency loss mechanism comprises, In the multi-mode geological feature recognition model, calculating the attention response distribution output by the model and the spatial difference value of the alteration tag, and constructing a loss function according to the spatial difference; The loss function consists of three sub-items, including a nonlinear amplification item, a logarithmic penalty item and a smooth boundary constraint item; Combining the three sub-items into a composite loss structure in a weighting mode, wherein a fixed weight or an adaptive weight mode can be adopted for calculation in the training process, the fixed weight is a preset constant, and the adaptive weight is dynamically adjusted according to the convergence rate of each sub-item loss; the multi-mode geological feature recognition model performs parameter updating according to the convergence condition of the composite loss function, and the composite loss structure performs a complete back propagation process in each iteration.
7. The method for modeling a shallow cryogenic hydrothermal deposit based on short wave infrared as defined in claim 6, wherein the co-participating in the optimization of the geological consistency loss mechanism with the classification loss and the regularization term includes, In a multi-mode geological feature recognition model training stage, a geological consistency loss function, a sample classification loss function and a regular smooth function are overlapped in proportion to form a comprehensive optimization target; Simultaneously minimizing a weighted sum of the three sub-items in each round of training iterations; The training control unit dynamically adjusts each weight according to the loss change rate, and when the geological consistency loss descending rate is lower than the classification loss, the geological consistency loss weight balance convergence speed is automatically increased; when the gradient oscillation is larger than a preset gradient oscillation threshold value, automatically increasing the weight of the regularization term; and the joint optimization mechanism executes gradient normalization processing in each round of iteration, and all parameters of the multi-mode geological feature recognition model are synchronously updated in a unified optimization framework.
8. The shallow low-temperature hot liquid deposit modeling system based on the short wave infrared is characterized by comprising a multi-source feature mapping module, a geological priori guiding module and a geological coupling optimizing module, wherein the shallow low-temperature hot liquid deposit modeling method based on the short wave infrared is adopted according to any one of claims 1-7; the multi-source feature mapping module is used for carrying out weighted normalization processing on multi-source heterogeneous geological data, mapping the multi-source heterogeneous geological data to a unified feature space through a nonlinear fusion coding algorithm with a weight coefficient, and generating a fusion feature vector; the geology priori guiding module is used for constructing a self-attention mechanism of the perception of the geological structure, introducing a geology priori matrix reflecting the terrain gradient and localization abnormal distribution on the basis of keys, inquiry and value matrix, and dynamically adjusting local geology coupling weight parameters in the attention calculation process; The geological coupling optimization module is used for establishing a geological consistency loss mechanism, carrying out space constraint on attention response and the alteration tag, forming a composite loss structure through a nonlinear amplification item, a logarithmic penalty item and a smooth boundary constraint item, and participating in optimization together with a classification loss and a regularization item.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the shortwave infrared-based shallow cryogenic hot fluid deposit modeling method of any of claims 1 to 7.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the short wave infrared based shallow low temperature hydrothermal deposit modeling method of any of claims 1 to 7.

Description

Shallow low-temperature hydrothermal deposit modeling method based on short-wave infrared Technical Field The invention relates to the technical field of artificial intelligence mineral prediction, in particular to a shallow low-temperature hydrothermal deposit modeling method based on short-wave infrared. Background Shallow low-temperature hydrothermal deposit is an important type of ore-forming geology research, the formation process of the shallow low-temperature hydrothermal deposit relates to complex multi-field coupling and multi-scale geological response characteristics, with the progress of remote sensing, geochemistry and topography measurement technologies, geological information acquisition means are remarkably improved, a large amount of multi-source heterogeneous geological data can provide support for deposit characteristic analysis, traditional methods based on statistical analysis, spectrum matching and localization anomaly identification can extract part of mineralization clues, but the efficient modeling is difficult when the high-dimensional and multi-modal data are processed, in recent years, artificial intelligence and deep learning technologies are gradually applied to the geological information identification field, geological mode excavation and feature extraction are realized through structures such as convolutional neural networks and transformers, and a new research direction is provided for intelligent deposit prediction. The existing method for analyzing the mineral deposit based on artificial intelligence generally has the problems of limited multi-source data fusion capability, insufficient perception of a geological structure, weak interpretation of a model and the like, and in the aspect of data fusion, the traditional method mostly adopts a linear weighting or simple splicing strategy, nonlinear correlation between remote sensing spectrum, a localization element and a topography feature is not considered, so that deviation exists in feature representation, multi-scale coupling effect in a shallow low-temperature hot liquid mineral deposit is difficult to reflect, in the process of feature recognition, the existing model mostly only depends on global correlation calculation of a self-attention mechanism, geological priori constraint is not introduced, effective perception of spatial information such as topography gradient change, localization abnormal distribution and the like is not available, a change zona and mineralization boundary are difficult to accurately recognize, in the stage of model optimization, and the existing training method usually adopts a single loss function, geological space continuity and structural consistency are not considered, overfitting or boundary blurring of the model under a complex geological background is easy to be caused, and therefore, the existing technology cannot realize accurate recognition and structural analysis of the shallow low-temperature hot liquid mineral deposit etching feature through uniform feature space expression, a geological awareness mechanism and a spatial consistency optimizing strategy. Disclosure of Invention The present invention has been made in view of the above-described problems. The invention solves the technical problems that the existing ore deposit characteristic analysis technology has poor multi-source geological data fusion precision, low geological structure space perception capability, lack of geological consistency constraint in model training and how to realize high-precision identification and structural analysis of shallow low-temperature hydrothermal ore deposit alteration zonation characteristics through an artificial intelligent model. The technical scheme includes that the shallow low-temperature hydrothermal deposit modeling method based on short wave infrared comprises the steps of carrying out weighted normalization processing on multi-source heterogeneous geological data, mapping the multi-source heterogeneous geological data to a unified feature space through a nonlinear fusion coding algorithm with weight coefficients to generate fusion feature vectors, constructing a geological structure perceived self-attention mechanism, introducing a geological priori matrix reflecting terrain gradient and localization abnormal distribution on the basis of keys, inquiry and value matrix, dynamically adjusting local geological coupling weight parameters in the attention computing process, establishing a geological consistency loss mechanism, carrying out space constraint on attention response and alteration labels, forming a composite loss structure through nonlinear amplification items, logarithmic penalty items and smooth boundary constraint items, and enabling the geological consistency loss mechanism, classification loss and regularization items to participate in optimization. The method comprises the steps of carrying out weighted normalization processing on multi-source heterogeneous geological