Mineral abundance remote sensing inversion method and device and computer equipment

Abstract

The present application relates to a kind of mineral abundance remote sensing inversion method, device and computer equipment.Wherein, mineral abundance remote sensing inversion method, including based on first position information, second position information, remote sensing data, spectral data and the mineral abundance of spectral data corresponding, establish data set;Mineral remote sensing inversion network model is constructed;Mineral remote sensing inversion network model is trained using data set, and based on the mineral remote sensing inversion network model that has been trained and remote sensing data, the abundance information of target mineral is obtained.Using the multi-head self-attention mechanism in model makes the model can be parallel computing, can accelerate training process, while better handle long distance dependence relationship, applicable to handle longer sequence data, in addition, model can directly access the information of entire input sequence in each layer, can more fully utilize global information and further improve inversion precision.

Description

Technical Field

[0001] This invention relates to the field of remote sensing exploration, and more specifically, to a method, apparatus, and computer equipment for remote sensing inversion of mineral abundance. Background Technology

[0002] Remote sensing technology plays a crucial role in mineral exploration. Utilizing multispectral, hyperspectral, and radar data acquired from satellite, aerial, and other remote sensing platforms, high-precision detection and monitoring of Earth's surface features, such as landforms, geological structures, mineralization and alteration, and vegetation cover, can be achieved. Traditional algorithms for mineral exploration using remote sensing data extract shallow, artificial features, which are highly susceptible to noise and difficult to determine, requiring significant human intervention and expert experience. Deep learning-based methods, however, can effectively extract deeper mineral information from remote sensing hyperspectral data, exhibiting better robustness against noise and uncertainty. Nevertheless, the inventors have found that current methods still suffer from poor inversion accuracy. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, apparatus, and computer equipment for remote sensing inversion of mineral abundance to improve inversion accuracy.

[0004] To achieve the above objectives, this application provides a method for remote sensing inversion of mineral abundance, comprising the following steps:

[0005] Determine the primary location information of the target mineral outcrop area and the secondary location information of the background feature area within the study area;

[0006] Acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data of the study area;

[0007] A dataset is established based on the first location information, the second location information, remote sensing data, spectral data, and the mineral abundance corresponding to the spectral data;

[0008] A mineral remote sensing inversion network model is constructed. This model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module. The convolutional layers of the ResNet network's residual structure are used to transform and extract features from the input data, yielding a first feature representation. The first normalization module normalizes the first feature representation to obtain a second feature representation. The multi-head attention module performs self-attention operations on the second feature representation to obtain a multi-head self-attention calculation result. The second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation. The feedforward network module processes the third feature representation to obtain a fourth feature representation, which is then output to the skip connections of the residual structure.

[0009] A mineral remote sensing inversion network model was trained using a dataset, and the abundance information of the target mineral was obtained based on the trained mineral remote sensing inversion network model and remote sensing data.

[0010] In one embodiment, the step further includes:

[0011] The remote sensing data is preprocessed to obtain preprocessed remote sensing data;

[0012] The steps to create a dataset include:

[0013] A dataset is established based on the first location information, the second location information, and the preprocessed remote sensing data.

[0014] In one embodiment, the step of preprocessing remote sensing data to obtain preprocessed remote sensing data includes:

[0015] Band filtering is performed on the remote sensing data to obtain the remaining band data;

[0016] One or more of the following processing methods are applied to the remaining band data: spectral and radiometric correction, geometric correction, atmospheric correction, and spectral reconstruction.

[0017] In one embodiment, the step further includes:

[0018] The accuracy of the trained mineral remote sensing inversion network model is evaluated, and if the accuracy evaluation meets the standard, the step of obtaining the location distribution map of the target mineral is performed.

[0019] In one embodiment, the step of the multi-head attention module performing a self-attention operation on the second feature representation to obtain the multi-head self-attention calculation result is based on the following formula:

[0020] Q=W q X; K=W k X; V=W v X;

[0021] ;

[0022] ;

[0023] head i =Attention(Q i , K i V i ) ;

[0024] MultiHead(Q, K, V)=Concat(head1, head h WO ;

[0025] Where X represents N embedded feature codes; W q W k W v These represent the weights obtained through training a fully connected network; Q is the initial representation of the corresponding query vector calculated through linear transformation; K is the initial representation of the corresponding key vector K calculated through linear transformation; V is the initial representation of the corresponding value vector calculated through linear transformation; d is the dimension. Let be the query weight matrix for the i-th head; Let be the key weight matrix of the i-th head; Let be the numerical weight matrix for the i-th head; , , head i W O This is the output weight matrix; MultiHead(Q,K,V) represents multi-head attention.

[0026] In one embodiment, the residual structure includes a 1×1 convolution kernel, a 3×3 convolution kernel, and a 1×1 convolution kernel connected in sequence.

[0027] In one embodiment, the residual structure includes 3×3 convolution kernels connected in sequence.

[0028] In one embodiment, in the step of processing the third feature representation to obtain the fourth feature representation, the feedforward network module obtains the fourth feature representation based on the following formula:

[0029] ;

[0030] in, ; This is the weight matrix of the first fully connected layer. It is the bias vector of the first fully connected layer. This is the weight matrix of the second fully connected layer. It is the bias vector of the second fully connected layer.

[0031] On one hand, embodiments of the present invention provide a mineral abundance remote sensing inversion device, comprising:

[0032] The determination module is used to determine the first location information of the target mineral outcrop area and the second location information of the background feature area in the study area;

[0033] The acquisition module is used to acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data of the study area.

[0034] The dataset creation module is used to create a dataset based on the first location information, the second location information, remote sensing data, spectral data, and the mineral abundance corresponding to the spectral data.

[0035] The model building module is used to construct a mineral remote sensing inversion network model. This model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module. Specifically, the convolutional layers of the ResNet network's residual structure are used to transform and extract features from the input data, yielding a first feature representation. The first normalization module normalizes the first feature representation to obtain a second feature representation. The multi-head attention module performs self-attention operations on the second feature representation to obtain the multi-head self-attention calculation result. The second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation. The feedforward network module processes the third feature representation to obtain a fourth feature representation, which is then output to the skip connections of the residual structure.

[0036] The output module is used to train a mineral remote sensing inversion network model using a dataset, and to obtain the abundance information of the target mineral based on the trained mineral remote sensing inversion network model and remote sensing data.

[0037] On the other hand, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and a computer-readable storage medium stores the computer program, wherein the computer program is configured to execute the above-described mineral abundance remote sensing inversion method at runtime.

[0038] One of the above technical solutions has the following advantages and beneficial effects:

[0039] The aforementioned mineral abundance remote sensing inversion method utilizes the multi-head self-attention mechanism in the model, enabling parallel computation without relying on the sequence order as in traditional recurrent neural network algorithms. This improves information transfer between layers, allowing for more efficient use of computational resources, accelerating the training process, and better handling of long-distance dependencies. It is suitable for processing longer sequence data. Furthermore, the model can directly access the information of the entire input sequence at each layer, thus making fuller use of global information, improving the model's expressive power, and further enhancing inversion accuracy. Attached Figure Description

[0040] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic flowchart of a mineral abundance remote sensing inversion method in one embodiment;

[0043] Figure 2 This is a first schematic diagram of the residual structure in one embodiment;

[0044] Figure 3 This is a second schematic diagram of the residual structure in one embodiment;

[0045] Figure 4 This is a schematic structural block diagram of a mineral remote sensing inversion network model in one embodiment;

[0046] Figure 5 This is a schematic flowchart illustrating the steps of preprocessing remote sensing data to obtain preprocessed remote sensing data in one embodiment.

[0047] Figure 6 This is a block diagram of a mineral abundance remote sensing inversion device in one embodiment. Detailed Implementation

[0048] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0050] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustration and has no specific meaning in itself. Therefore, "module" and "part" may be used interchangeably.

[0051] It is understood that the term "connection" in the following embodiments should be understood as "electrical connection," "communication connection," etc., if the connected circuits, modules, units, etc., have electrical signal or data transmission with each other.

[0052] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising,” “including,” or “having,” etc., specify the presence of the stated feature, whole, step, operation, component, part, or combination thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof.

[0053] In one embodiment, such as Figure 1 As shown, a method for remote sensing inversion of mineral abundance is provided, including the following steps:

[0054] S110, determine the first location information of the target mineral outcrop area and the second location information of the background landform area in the study area;

[0055] The background feature area refers to other feature areas that are not related to the target mineral. The study area mentioned above can refer to a portion of the study area or the entire study area.

[0056] Specifically, by conducting field surveys and sampling, and analyzing surface rocks and soil, the primary location information of the target mineral outcrop area and the secondary location information of the background feature area can be determined; the primary and secondary location information can be geographic coordinates.

[0057] S120: Acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data of the study area;

[0058] Among them, remote sensing data is airborne hyperspectral data, spectral data is continuous spectral data of target minerals and other background features measured by shortwave infrared spectrometer; mineral abundance is mineral composition and content, which can be determined by X-ray diffraction experiments or by industry software to determine the mineral composition and content corresponding to each spectral data.

[0059] Specifically, remote sensing data can be acquired using a hyperspectral imager, which can obtain hyperspectral data with a width of 60 km and a spatial resolution >30 m. This data has spectral resolutions of 10 nm in the visible-near infrared band and 20 nm in the shortwave infrared band, totaling 166 bands.

[0060] S130, a dataset is established based on the first location information, the second location information, remote sensing data, spectral data, and the mineral abundance corresponding to the spectral data;

[0061] Specifically, the dataset can include a training dataset and a validation dataset. In a specific example, satellite remote sensing image pixels (remote sensing data) at the first location are labeled with the target mineral area, and satellite remote sensing image pixels (remote sensing data) at the second location are labeled with the background land cover area. The dataset is then created using the correspondence between the target mineral's spectral data and its mineral composition and content labels, and the spectral data of other background land cover and the other background land cover. The dataset is then randomly divided into a training dataset and a validation dataset according to a preset ratio. In one specific example, the ratio could be 7:3. In another specific example, based on the content of the target mineral in the sample corresponding to each spectral data point, the spectral data is divided into three groups, labeled with target mineral content of 40%-60%, 60%-80%, and 80%-100% (data with content below 40% are deleted). Furthermore, before model training, a domain-knowledge-based data augmentation method is used, combining the classic Hapke radiative transfer model with a small number of ground-based measurement points to augment the training label data.

[0062] S140, Construct a mineral remote sensing inversion network model; the mineral remote sensing inversion network model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module; wherein, the convolutional layers of the residual structure of the ResNet network are used to perform feature transformation and extraction on the input data to obtain a first feature representation; the first normalization module is used to normalize the first feature representation to obtain a second feature representation; the multi-head attention module performs self-attention operation on the second feature representation to obtain the multi-head self-attention calculation result; the second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation; the feedforward network module processes the third feature representation to obtain a fourth feature representation and outputs it to the skip connection of the residual structure;

[0063] Specifically, ResNet is a convolutional neural network architecture that, while possessing the characteristics of a conventional network structure, effectively overcomes the "degradation" problem caused by increasing the number and depth of layers by using residual structures. The ResNet network consists of convolutional layers, max-pooling layers, residual modules, average pooling layers, fully connected layers, and Batch Normalization (BN) layers. A batch normalization layer is inserted between each network layer, which helps accelerate the training convergence process and provides a regularization effect, making the model more stable and reliable. Convolutional layers are used for basic feature extraction operations, extracting features from the input through convolution. Max-pooling layers reduce the spatial dimensionality of the feature map while retaining the most important features. Average pooling layers are similar to max-pooling layers, but use the average value instead of the maximum value to reduce the dimensionality of the feature map. Fully connected layers transform the feature map of the last layer of the network into the final output. The residual structure, located between the max-pooling and average pooling layers, preserves the feature information of the input data, and the training results improve with increasing network layers. However, for higher-level ResNet networks, as the number of hidden layers increases to a certain extent, the total number of network parameters also increases dramatically. In this case, varying degrees of network degradation can still occur, leading to accuracy saturation or a decrease during training. Furthermore, a large number of parameters not only affects the computer's GPU memory, reducing the training and prediction speed, but also significantly impacts the training effect. In a specific example, the residual module is as follows: Figure 2 As shown, the weight layer consists of two layers, using two 3×3 convolutional kernels directly, or as follows: Figure 3 As shown, the weight layer in the residual block consists of three layers: first a 1×1 convolutional kernel, then a 3×3 convolutional kernel, and finally a 1×1 convolutional kernel. The first and last 1×1 convolutions are used to reduce the dimensionality of the input and output data, respectively, thereby reducing the parameters of the residual module and the amount of computation while ensuring the accuracy of the model.

[0064] Furthermore, the first normalization module is used to normalize the feature representation to enhance the stability and convergence of the network. The first and second normalization modules can employ layer normalization, which is highly efficient for the hidden states in stable recurrent neural networks. For an image, each segmented piece represents a word; therefore, it solves visual problems using semantic logic. The formula for calculating the parameters of layer normalization is as follows:

[0065] ;

[0066] ;

[0067]

[0068] The two formulas above are for calculating the mean (μ) and standard deviation (σ), respectively. Index of the presentation layer, Indicates the number of heads. Indicate attention head Attention weights in layer l To set a minimum value to prevent the denominator from being 0. and These are trainable parameters that can undergo affine transformations.

[0069] Furthermore, a multi-head attention module connects multiple attention modules, allowing the same attention mechanism to learn different behaviors. This is achieved by combining different subspace representations of queries, keys, and values, enabling the autonomous learning to focus on different things. The computation process is as follows:

[0070] First, provide the input information, using X=[x1, x2, ...]. , x n ] represents the encoding of N embedded features, which is also the second feature representation mentioned above, W q W k W v The weights, obtained through training a fully connected network, are used to calculate the initial representations of the corresponding query vector Q, key vector K, and value vector V through linear transformation. The calculation process is as follows:

[0071] Q=W q X; K=W k X; V=W v X;

[0072] Next, the self-attention score is calculated, and the result is normalized using Softmax. The dot product of each query vector Q and each key vector K is performed to obtain the weight matrix; the Softmax function is used to normalize the weight matrix to obtain the key weights; since the weight matrix is ​​too large, the gradient after Softmax is too small, so it is scaled by dividing the weight matrix by the square root of its dimension d; finally, the weights for each value vector V are weighted and summed with V to obtain the final result. The formula for calculating the self-attention score is as follows:

[0073] ;

[0074] The Transformer network framework actually uses MHSA. Multi-head self-attention (MSA) means combining the SA information learned from different heads. Then, the results from each head are concatenated, and finally, the concatenated result is multiplied by the projection matrix OW, which compresses the output matrix to obtain the MSA calculation result. The specific calculation formula is as follows:

[0075] ;

[0076] head i =Attention(Q i , K i V i ) ;

[0077] MultiHead(Q, K, V)=Concat(head1, head h W O ;

[0078] Where X represents N embedded feature codes; W q W k W v These are the weights obtained through training a fully connected network; Q is the initial representation of the corresponding query vector calculated through linear transformation; K is the initial representation of the corresponding key vector K calculated through linear transformation; and V is the initial representation of the corresponding value vector calculated through linear transformation. Let be the query weight matrix for the i-th head; Let be the key weight matrix of the i-th head; Let be the numerical weight matrix for the i-th head; , , head i W O This is the output weight matrix; MultiHead(Q,K,V) represents multi-head attention.

[0079] The feedforward network module consists of two linear transformations with ReLU activation in between. Furthermore, each of these two sub-layers has residual connections around it, and the sum of computations between the sub-layer inputs and the output produced by the sub-layer itself is normalized by a normalization layer.

[0080] ;

[0081] in, ; This is the weight matrix of the first fully connected layer. It is the bias vector of the first fully connected layer. This is the weight matrix of the second fully connected layer. It is the bias vector of the second fully connected layer.

[0082] The steps in S140 above are as follows: Figure 4 As shown, the first normalization module, the multi-head attention module, the second normalization module, and the feedforward network module are integrated between the last convolutional layer and the skip connection of the residual structure. By utilizing the powerful self-attention mechanism, the relationship between features can be captured more meticulously, resulting in a more powerful feature representation capability and long-distance dependency modeling capability.

[0083] S150 uses a dataset to train a mineral remote sensing inversion network model, and based on the trained mineral remote sensing inversion network model and remote sensing data, obtains the abundance information of the target mineral.

[0084] The abundance information of the target mineral includes its distribution location, composition, and content.

[0085] Specifically, a mineral remote sensing inversion network model is trained using a dataset to obtain a trained mineral remote sensing inversion network model. By inputting remote sensing data into the trained mineral remote sensing inversion network model, the abundance information of the target mineral can be obtained.

[0086] The aforementioned mineral abundance remote sensing inversion method utilizes the multi-head self-attention mechanism in the model, enabling parallel computation without relying on the sequence order as in traditional recurrent neural network algorithms. This improves information transfer between layers, allowing for more efficient use of computational resources, accelerating the training process, and better handling of long-distance dependencies. It is suitable for processing longer sequence data. Furthermore, the model can directly access the information of the entire input sequence at each layer, thus making fuller use of global information, improving the model's expressive power, and further enhancing inversion accuracy.

[0087] In one embodiment, the above-mentioned mineral abundance remote sensing inversion method further includes a preprocessing step, specifically: preprocessing the remote sensing data to obtain preprocessed remote sensing data. During the dataset creation process, the dataset is built based on the first location information, the second location information, the preprocessed remote sensing data, the spectral data, and the mineral abundance corresponding to the spectral data. Specific steps can be found in the above-described description of dataset creation.

[0088] In one embodiment, such as Figure 5 As shown, the steps for preprocessing remote sensing data to obtain preprocessed remote sensing data include:

[0089] S510, perform band filtering on remote sensing data to obtain the remaining band data;

[0090] S520 performs one or more of the following processing methods on the remaining band data: spectral and radiometric correction, geometric correction, atmospheric correction, and spectral reconstruction.

[0091] Specifically, the remote sensing data can be the domestically produced ZY1-02D ​​data. For ZY1-02D ​​data, due to the influence of water vapor absorption, the corresponding bands Band98-Band102 and Band125-Band134 are empty and need to be removed. Due to sensor-specific factors, there are overlapping wavelengths in the visible and near-infrared bands Band72-Band76 and the short-wave infrared bands Band77-Band79, affecting the spectral characteristics and needing to be removed. However, for ZY1-02D ​​data, the signal-to-noise ratio of bands Band77-Band79 is lower than that of bands Band72-Band76, so the data in bands Band77-Band79 are removed here. Finally, after band-by-band review, the lower-quality bands Band161-Band166 are removed. In summary, after band selection, 142 bands are ultimately retained for subsequent processing.

[0092] Furthermore, for bad lines in hyperspectral images, this application uses the average value of adjacent rows or columns of pixels to replace the corresponding pixel value of the bad line to complete the bad line repair. Radiometric calibration and atmospheric correction are performed using ENVI 5.3 software. For hyperspectral data, this application can use a Savitzky-Golay filter to smooth the spectral domain noise of the data. The Savitzky-Golay filter is widely used for data stream smoothing and noise reduction. It is a filtering method based on local polynomial least squares fitting in the time domain. When extracting the original signal from a noisy environment, this method can effectively smooth the signal while preserving the signal shape as much as possible. Different window widths and smoothing polynomial degrees will result in different smoothing effects, thus adapting to different situations. Practical experience shows that when the number of left and right side points of the filter kernel is 5 and the smoothing polynomial degree is 4, the filtering smoothing effect is better and the basic spectral characteristics are better preserved. Radiometric correction aims to remove instrument system errors and convert the original image DN values ​​into radiance values. Geometric correction aims to eliminate spatial geometric distortions in the image, ensuring that ground features in the image match their actual geographical locations. Atmospheric correction aims to remove the influence of the atmospheric environment during radiative transmission, converting radiance values ​​into reflectance values ​​to achieve spectral reconstruction. In one specific embodiment, system software can be used to perform radiometric and geometric corrections on remote sensing data. During geometric correction, DEM (Digital Elevation Model) data of the study area is also incorporated to further improve correction accuracy. Atmospheric correction and spectral reconstruction were completed using the reflectance of the calibration field acquired synchronously during flight, employing an empirical linear method.

[0093] In one embodiment, the trained mineral remote sensing inversion network model can be subjected to accuracy evaluation. If the accuracy evaluation meets the standards, the next step, namely obtaining the location distribution map of the target mineral, can be performed. In the accuracy evaluation stage, this application can use a validation dataset and introduce parameters such as confusion matrix, F1-Score, average F1 score, mIoU (Mean Intersection over Union), FWIoU (Frequency-Weighted Intersection over Union), overall accuracy, and Kappa coefficient to evaluate the accuracy of the inversion results.

[0094] In one embodiment, such as Figure 6 As shown, a mineral abundance remote sensing inversion device is provided, including:

[0095] The determination module is used to determine the first location information of the target mineral outcrop area and the second location information of the background feature area in the study area;

[0096] The acquisition module is used to acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data of the study area.

[0097] The dataset creation module is used to create a dataset based on the first location information, the second location information, remote sensing data, spectral data, and the mineral abundance corresponding to the spectral data.

[0098] The model building module is used to construct a mineral remote sensing inversion network model. This model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module. Specifically, the convolutional layers of the ResNet network's residual structure are used to transform and extract features from the input data, yielding a first feature representation. The first normalization module normalizes the first feature representation to obtain a second feature representation. The multi-head attention module performs self-attention operations on the second feature representation to obtain the multi-head self-attention calculation result. The second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation. The feedforward network module processes the third feature representation to obtain a fourth feature representation, which is then output to the skip connections of the residual structure.

[0099] The output module is used to train a mineral remote sensing inversion network model using a dataset, and to obtain the abundance information of the target mineral based on the trained mineral remote sensing inversion network model and remote sensing data.

[0100] Specific limitations regarding the mineral abundance remote sensing inversion device can be found in the limitations of the mineral abundance remote sensing inversion method described above, and will not be repeated here. Each module in the aforementioned mineral abundance remote sensing inversion device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module. It should be noted that the module division in this embodiment is illustrative and only represents a logical functional division; other division methods may be used in actual implementation.

[0101] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0102] Determine the primary location information of the target mineral outcrop area and the secondary location information of the background feature area within the study area;

[0103] Acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data of the study area;

[0104] A dataset is established based on the first location information, the second location information, remote sensing data, spectral data, and the mineral abundance corresponding to the spectral data;

[0105] A mineral remote sensing inversion network model is constructed. This model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module. The convolutional layers of the ResNet network's residual structure are used to transform and extract features from the input data, yielding a first feature representation. The first normalization module normalizes the first feature representation to obtain a second feature representation. The multi-head attention module performs self-attention operations on the second feature representation to obtain a multi-head self-attention calculation result. The second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation. The feedforward network module processes the third feature representation to obtain a fourth feature representation, which is then output to the skip connections of the residual structure.

[0106] A mineral remote sensing inversion network model was trained using a dataset, and the abundance information of the target mineral was obtained based on the trained mineral remote sensing inversion network model and remote sensing data.

[0107] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, performs the following steps:

[0108] Determine the primary location information of the target mineral outcrop area and the secondary location information of the background feature area within the study area;

[0109] Acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data of the study area;

[0110] A dataset is established based on the first location information, the second location information, remote sensing data, spectral data, and the mineral abundance corresponding to the spectral data;

[0111] A mineral remote sensing inversion network model is constructed. This model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module. The convolutional layers of the ResNet network's residual structure are used to transform and extract features from the input data, yielding a first feature representation. The first normalization module normalizes the first feature representation to obtain a second feature representation. The multi-head attention module performs self-attention operations on the second feature representation to obtain a multi-head self-attention calculation result. The second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation. The feedforward network module processes the third feature representation to obtain a fourth feature representation, which is then output to the skip connections of the residual structure.

[0112] A mineral remote sensing inversion network model was trained using a dataset, and the abundance information of the target mineral was obtained based on the trained mineral remote sensing inversion network model and remote sensing data.

[0113] In specific implementation, the embodiments of this application can be referred to the above embodiments and have corresponding technical effects.

[0114] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0115] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0117] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0118] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0121] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0122] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for remote sensing inversion of mineral abundance, characterized in that, include: Determine the primary location information of the target mineral outcrop area and the secondary location information of the background feature area within the study area; Acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data for the study area; The remote sensing data is preprocessed to obtain preprocessed remote sensing data; A dataset is established based on the first location information, the second location information, the preprocessed remote sensing data, the spectral data, and the mineral abundance corresponding to the spectral data. A mineral remote sensing inversion network model is constructed. This model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module. Specifically, the convolutional layers of the ResNet network's residual structure are used to transform and extract features from the input data, yielding a first feature representation. The first normalization module normalizes the first feature representation to obtain a second feature representation. The multi-head attention module performs self-attention operations on the second feature representation to obtain a multi-head self-attention calculation result. The second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation. The feedforward network module processes the third feature representation to obtain a fourth feature representation, which is then output to the skip connections of the residual structure. The mineral remote sensing inversion network model is trained using the dataset, and the abundance information of the target mineral is obtained based on the trained mineral remote sensing inversion network model and the remote sensing data; the abundance information of the target mineral includes the distribution location information, composition and content of the target mineral.

2. The mineral abundance remote sensing inversion method according to claim 1, characterized in that, The steps of preprocessing the remote sensing data to obtain preprocessed remote sensing data include: The remote sensing data is filtered by band to obtain the remaining band data; The remaining band data is processed using one or more of the following methods: spectral and radiometric correction, geometric correction, atmospheric correction, and spectral reconstruction.

3. The mineral abundance remote sensing inversion method according to claim 1, characterized in that, It also includes the following steps: The accuracy of the trained mineral remote sensing inversion network model is evaluated, and if the accuracy evaluation meets the standard, the step of obtaining the location distribution map of the target mineral is performed.

4. The mineral abundance remote sensing inversion method according to claim 1, characterized in that, In the step of the multi-head attention module performing a self-attention operation on the second feature representation to obtain the multi-head self-attention calculation result, the multi-head self-attention calculation result is obtained based on the following formula: Q=W q X;K=W k X;V=W v X; ； ； head i =Attention(Q i , K i , V i ) ; MultiHead(Q, K, V)=Concat(head1, head h )W O ; Where X represents N embedded feature codes; W q W k W v These represent the weights obtained through training a fully connected network; Q is the initial representation of the corresponding query vector calculated through linear transformation; K is the initial representation of the corresponding key vector K calculated through linear transformation; V is the initial representation of the corresponding value vector calculated through linear transformation; d is the dimension. Let be the query weight matrix for the i-th head; Let be the key weight matrix of the i-th head; Let be the numerical weight matrix for the i-th head; , , head i W O This is the output weight matrix; MultiHead(Q,K,V) represents multi-head attention.

5. The mineral abundance remote sensing inversion method according to claim 1, characterized in that, The residual structure includes a 1×1 convolution kernel, a 3×3 convolution kernel, and a 1×1 convolution kernel connected in sequence.

6. The mineral abundance remote sensing inversion method according to claim 1, characterized in that, The residual structure comprises 3×3 convolution kernels connected in sequence.

7. The mineral abundance remote sensing inversion method according to claim 1, characterized in that, In the step where the feedforward network module processes the third feature representation to obtain the fourth feature representation, the fourth feature representation is obtained based on the following formula: ； in, ; This is the weight matrix of the first fully connected layer. It is the bias vector of the first fully connected layer. This is the weight matrix of the second fully connected layer. It is the bias vector of the second fully connected layer.

8. A mineral abundance remote sensing inversion device, characterized in that, include: The determination module is used to determine the first location information of the target mineral outcrop area and the second location information of the background feature area in the study area; The acquisition module is used to acquire remote sensing data, spectral data, and mineral abundance corresponding to the spectral data of the study area; The dataset creation module is used to preprocess the remote sensing data to obtain preprocessed remote sensing data; and to create a dataset based on the first location information, the second location information, the preprocessed remote sensing data, the spectral data, and the mineral abundance corresponding to the spectral data. A model building module is used to construct a mineral remote sensing inversion network model. This model includes a ResNet network, a first normalization module, a multi-head attention module, a second normalization module, and a feedforward network module. Specifically, the convolutional layers of the ResNet network's residual structure are used to transform and extract features from the input data, obtaining a first feature representation. The first normalization module normalizes the first feature representation to obtain a second feature representation. The multi-head attention module performs self-attention operations on the second feature representation to obtain a multi-head self-attention calculation result. The second normalization module normalizes the multi-head self-attention calculation result to obtain a third feature representation. The feedforward network module processes the third feature representation to obtain a fourth feature representation and outputs it to the skip connections of the residual structure. The output module is used to train the mineral remote sensing inversion network model using the dataset, and to obtain the abundance information of the target mineral based on the trained mineral remote sensing inversion network model and the remote sensing data; the abundance information of the target mineral includes the distribution location information, composition and content of the target mineral.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, The computer program is configured to execute the mineral abundance remote sensing inversion method as described in any one of claims 1 to 7 when it is run.

Images