Method for identifying spodumene based on spectral index and double-branch neural network from multi-source remote sensing

Abstract

The present application belongs to the technical field of remote sensing image recognition, and discloses a multi-source remote sensing spodumene recognition method based on spectral index and double-branch neural network. The method comprises the following steps: field sample collection, remote sensing image acquisition and pretreatment; calculating the spodumene index based on the spectral characteristics of spodumene in the remote sensing image, and making a remote sensing index cube; constructing a double-branch neural network model based on a graph convolution network and a one-dimensional convolution neural network, recognizing the spectral characteristics and spatial distribution characteristics of spodumene, and then recognizing spodumene, and evaluating the accuracy of the recognition result. The present application solves the problem that the spectral characteristics of spodumene in visible light-near infrared and short wave infrared band remote sensing images are not obvious, which leads to difficulty in recognition. The mixed deep learning model, i.e. the double-branch neural network model, fully utilizes the graph and spectral information in the remote sensing image, and improves the recognition accuracy and efficiency of spodumene.

Description

Technical Field

[0001] This invention belongs to the field of image remote sensing technology, and particularly relates to a multi-source remote sensing spodumene identification method based on spectral index and dual-branch neural network. Background Technology

[0002] Lithium (Li) is the lightest metal in nature, possessing high specific heat, electrical conductivity, and electrochemical activity. As a crucial component of green energy storage technology, it holds significant economic importance. Lithium has been designated a strategic or critical mineral by many countries and recognized as a "green rare metal" by the United Nations Environment Programme. Global lithium resources are primarily found in salt lake brines (72.3%), with relatively less in ores (20.3%), but lithium-bearing minerals contribute over 50% of the global lithium supply. Among the key lithium-bearing minerals primarily found in granite pegmatite-type lithium deposits, spodumene is the most important. Rapid identification of spodumene and enhanced lithium resource development and utilization are crucial for ensuring stable lithium supply and sustainable global lithium resource development. Remote sensing technology, with its accurate identification and classification capabilities and potential for discovering mineral targets in complex geological environments, has been widely applied in rock and mineral identification and classification in the geological field, achieving significant results. Currently, remote sensing technology using the visible-shortwave infrared band has made significant progress in the detection of various minerals. However, minerals such as spodumene are anhydrous silicate minerals, and their spectral characteristics are mainly concentrated in the thermal infrared band. The application of thermal infrared remote sensing in lithium ore detection is still in its early stages. How to accurately identify spodumene by combining thermal infrared remote sensing, and thus distinguish between ore-bearing and ore-free pegmatites, remains a challenge.

[0003] Based on the above analysis, the problems and defects of the existing technology are as follows: the spectral characteristics of spodumene are not obvious in the visible-near infrared and short-wave infrared remote sensing images, which makes it difficult for the existing technology to identify it, resulting in low accuracy and efficiency of spodumene identification. Summary of the Invention

[0004] To overcome the problems existing in related technologies, the embodiments disclosed in this invention provide a multi-source remote sensing method for identifying spodumene that integrates visible light, shortwave infrared, and thermal infrared, specifically involving a multi-source remote sensing method for identifying spodumene based on spectral indices and a dual-branch neural network. The purpose of this invention is to establish a series of spodumene remote sensing indices based on multi-source satellite imagery data, and to achieve accurate identification of spodumene through a hybrid deep learning model, providing a new method for improving the efficiency of lithium resource exploration.

[0005] The technical solution is as follows: a multi-source remote sensing spodumene identification method based on spectral index and dual-branch neural network, including:

[0006] S1, Field sample collection, remote sensing image acquisition and preprocessing;

[0007] S2. Based on the spectral characteristics of remote sensing images, calculate the spodumene spectral index and create a spodumene spectral remote sensing index cube.

[0008] S3. Construct a dual-branch neural network model based on graph convolutional network and one-dimensional convolutional neural network. Based on this dual-branch neural network model, identify the spatial distribution characteristics and spectral characteristics of spodumene and obtain the identification result image.

[0009] S4, Evaluation of recognition results.

[0010] In step S1, field sample collection, remote sensing image acquisition and preprocessing include: for Landsat 8 multispectral and GF-5 hyperspectral images, radiometric calibration and atmospheric correction are performed to convert the original values ​​into reflectance; for SDGSAT-1 thermal infrared images, radiometric calibration is performed to convert the original values ​​into radiance; after completion, the three images are registered and resampled respectively.

[0011] In step S2, the spodumene spectral index is calculated, and a spodumene spectral remote sensing index cube is constructed, including:

[0012] S201. Select the green band and near-infrared band of the preprocessed Landsat 8 image in the visible-near-infrared range and calculate the band ratio to obtain the spodumene visible-near-infrared remote sensing index BR.

[0013] S202. Select the shortwave infrared band of the preprocessed GF-5 image, extract the spectrum of the pixel where spodumene is located in the collected sample and take the average as the reference spectrum of the shortwave infrared band of spodumene. Use the matched filtering method to calculate the matched filter value MF between all pixels and the reference spectrum, and use it as the shortwave infrared remote sensing index of spodumene.

[0014] S203. For the preprocessed SDGSAT-1 image, extract the three band radiance values ​​of all spodumene samples, combine them in pairs, and use the least squares method to fit the linear relationship of thermal infrared radiance of spodumene in each band combination.

[0015] S204 combines the visible-near infrared remote sensing index, short-wave infrared remote sensing index and multiple thermal infrared remote sensing indices corresponding to each pixel to obtain a three-dimensional spodumene remote sensing index cube, which is arranged in pixel order and stored in two-dimensional CSV format.

[0016] In step S201, the formula for calculating the visible-near-infrared remote sensing index BR of spodumene is as follows:

[0017]

[0018] In the formula, Band3 is the green band and Band5 is the near-infrared band.

[0019] In step S202, the formula for calculating the matched filter value MF is:

[0020]

[0021] In the formula, S i R is the actual spectral value of the pixel in the i-th band. i is the reference spectral value of spodumene in the i-th band, and n is the total number of bands.

[0022] In step S202, the spodumene shortwave infrared remote sensing index ST i The calculation formula is:

[0023] ST i =|Band b -Band a ×m i -c i |

[0024] In the formula, Band a Band b These are two bands in the i-th band combination, m i ,c i These represent the slope and intercept of the linear relationship obtained by linear fitting of spodumene in the t band combination, respectively.

[0025] In step S3, a dual-branch neural network model based on graph convolutional networks and one-dimensional convolutional neural networks is constructed. This dual-branch neural network model is used to identify the spatial distribution characteristics and spectral characteristics of spodumene, including:

[0026] S301, use Matlab software to create the data required for training;

[0027] S302 uses the Python language to build deep learning models.

[0028] In step S301, the training data is prepared using Matlab software, including:

[0029] (1) Read the image data and divide it into training and test sets according to the label of each pixel;

[0030] (2) Perform data conversion, converting the three-dimensional remote sensing image data and label data into two-dimensional matrices respectively;

[0031] (3) Calculate the Laplacian matrix required for the graph convolutional network;

[0032] (4) Save the training and test data as .mat files respectively.

[0033] In step S302, a deep learning model is constructed using the Python language, including:

[0034] (a) Import the Python libraries NumPy, Tensorflow, and SciPy;

[0035] (b) Define auxiliary functions, including:

[0036] `sample_mask` creates a sample mask for data segmentation.

[0037] `create_placeholders` creates TensorFlow placeholders for input data and labels;

[0038] initialize_parameters initializes the network parameters, including weights and biases;

[0039] GCN_layer defines the operation of the graph convolutional layer, with the input being the feature matrix and the Laplacian matrix;

[0040] Mynetwork defines the forward propagation process of a hybrid model, which includes GCN layers and 1D-CNN.

[0041] (c) Define the network structure, which consists of graph convolutional layers and 1D-CNN. The graph convolutional layers perform convolution operations on the features through the Laplacian matrix. The 1D-CNN is used to process the temporal information of the feature vectors. Finally, the network is classified through fully connected layers to output the probability distribution of spodumene categories.

[0042] (d) Use cross-entropy as the loss function, add L2 regularization, and use the Adam optimizer to update the parameters;

[0043] (e) Use leave-one-out cross-validation to train the deep learning model. Leave one sample as the validation set each time and the rest as the training set. Then train the deep learning model on the entire training set and make predictions on the test set to complete the identification of spodumene.

[0044] (f) Save the predicted probabilities of the test set as MAT and CSV files for subsequent analysis.

[0045] Another objective of this invention is to provide a multi-source remote sensing spodumene identification system based on spectral index and dual-branch neural network. This system implements the aforementioned multi-source remote sensing spodumene identification method based on spectral index and dual-branch neural network. The system includes:

[0046] The remote sensing image processing module is used for field sample collection, remote sensing image acquisition and preprocessing.

[0047] The Remote Sensing Index Cube Creation Module is used to calculate the spodumene spectral index based on the spectral characteristics of remote sensing images and to create a spodumene spectral remote sensing index cube.

[0048] The dual-branch neural network model construction module is used to construct a dual-branch neural network model based on graph convolutional networks and one-dimensional convolutional neural networks. Based on this dual-branch neural network model, the spatial distribution characteristics and spectral characteristics of spodumene are identified, and the identification result image is obtained.

[0049] The evaluation module is used to identify and evaluate the results.

[0050] Combining all the above technical solutions, the beneficial effects of this invention are as follows: This invention solves the problem that spodumene is difficult to identify in visible-shortwave infrared remote sensing images due to its unclear spectral features. It uses a hybrid deep learning model to make full use of the image and spectral information in remote sensing images, thereby improving the accuracy and efficiency of spodumene identification.

[0051] Because the identification of spodumene is crucial for the discovery of pegmatite-type lithium deposits, serving as a direct indicator for prospecting, the spodumene identification technology and methods developed in this invention, once applied and transformed, can directly identify the location and extent of spodumene deposits. This is particularly beneficial in high-altitude areas where human access is difficult, significantly narrowing the prospecting area and clarifying prospecting methods. For prospecting companies, this not only saves substantial manpower and material resources in prospecting costs but also clarifies prospecting directions, promotes the discovery of new deposits, and generates economic benefits.

[0052] This invention innovatively proposes spodumene spectral indices in the near-infrared, short-wave infrared, and thermal infrared spectral bands, establishes a spodumene spectral index cube, and introduces a dual-branch neural network for the first time to identify spodumene, filling the technological gap in using thermal infrared remote sensing combined with other remote sensing data and deep learning algorithms to mine spodumene features and then identify spodumene.

[0053] The shortwave infrared band primarily reflects the vibrations between hydroxyl groups and their associated cations. Spodumene, as a chain-like anhydrous silicate mineral, lacks structural water in its crystal structure and therefore exhibits no absorption characteristics in the shortwave infrared band. Since sericite is secondary distributed within the joints and fissures of spodumene, its spectral characteristics in the shortwave infrared range are caused by mica-like hydrous minerals. Therefore, most researchers indirectly identify spodumene by recognizing the spectral characteristics of sericite. However, the diagnostic characteristics of spodumene are actually in the thermal infrared band, but research on this is limited, with only a few scholars studying its ground-based thermal infrared spectral characteristics. Therefore, this invention solves the technical challenge of identifying the spectral characteristics of spodumene in the visible-near-infrared and shortwave infrared bands. By combining thermal infrared, visible-near-infrared, and shortwave infrared spectral bands, it not only constructs spectral indices reflecting the characteristics of spodumene but also enables direct identification of spodumene in remote sensing images. Attached Figure Description

[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0055] Figure 1 This is a diagram of the multi-source remote sensing spodumene identification method and system provided in the embodiments of the present invention;

[0056] Figure 2 The graph shows the linear relationship between thermal infrared radiation brightness in band 1 and band combination.

[0057] Figure 3 Linear relationship of thermal infrared radiance in the combination of Band 1 and Band 3 (green band);

[0058] Figure 4 A linear relationship graph of thermal infrared radiance in the combination of Band 2 and Band 3 (green band);

[0059] Figure 5 Landsat 8BR map in the spodumene remote sensing index raster image;

[0060] Figure 6 This is the GF-5MF map in the spodumene remote sensing index raster image;

[0061] Figure 7 This is the SDGSAT-1ST1 image in the spodumene remote sensing index raster image;

[0062] Figure 8 This is the SDGSAT-1ST2 image in the spodumene remote sensing index raster image;

[0063] Figure 9This is the SDGSAT-1ST3 image in the spodumene remote sensing index raster image;

[0064] Figure 10 This is a diagram illustrating the computation process of a two-branch neural network.

[0065] Figure 11 This is a diagram showing the results of remote sensing identification of spodumene. Detailed Implementation

[0066] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0067] The innovation of this invention lies in the fact that it is the first to create a thermal infrared spectral index for spodumene, and the first to establish a dual-branch neural network recognition model using a cubic approach of visible-near infrared, short-wave infrared, and thermal infrared spodumene optical indexes. This innovatively identifies both the spectral characteristics and spatial distribution characteristics of spodumene, and significantly improves the accuracy and efficiency of spodumene recognition.

[0068] Example 1, as Figure 1 As shown, the multi-source remote sensing spodumene identification method provided in this embodiment of the invention includes:

[0069] S1, Field sample collection, remote sensing image acquisition and preprocessing;

[0070] For example, spodumene and non-spodumene samples were uniformly collected from vertical pegmatite dikes in the study area. Landsat 8 multispectral, GF-5 hyperspectral, and SDGSAT-1 thermal infrared remote sensing images of the study area were acquired, prioritizing images from the same season, with low cloud cover, and good image quality. Image preprocessing was performed using ENVI 5.6 software. For the Landsat 8 multispectral and GF-5 hyperspectral images, radiometric calibration and atmospheric correction were performed to convert the raw values ​​to reflectance. For the SDGSAT-1 thermal infrared images, radiometric calibration was performed to convert the raw values ​​to radiance. After completion, the three images were registered and resampled to 30m × 30m.

[0071] S2. Based on the spectral characteristics of remote sensing images, calculate the spodumene spectral index and create a spodumene spectral remote sensing index cube.

[0072] S201, the green band (Band 3) and near-infrared band (Band 5) in the visible-near-infrared range of the preprocessed Landsat 8 image were selected and their band ratios were calculated to obtain the spodumene visible-near-infrared remote sensing index BR. (See...) Figure 5 Landsat 8BR map in spodumene remote sensing index raster image.

[0073]

[0074] In the formula, Band3 is the green band and Band5 is the near-infrared band.

[0075] S202, the shortwave infrared band (2032-2496nm) of the preprocessed GF-5 image was selected. The average of the spectra of the pixels containing spodumene in the acquired samples was taken as the reference spectrum for the shortwave infrared band of spodumene. The matched filter value MF between all pixels and the reference spectrum was calculated using the matched filtering method, and used as the shortwave infrared remote sensing index of spodumene. See [link to relevant documentation]. Figure 6 GF-5MF map in spodumene remote sensing index raster image.

[0076]

[0077] In the formula, S i R is the actual spectral value of the pixel in the i-th band. i is the reference spectral value of spodumene in the i-th band, and n is the total number of bands.

[0078] S203. For the preprocessed SDGSAT-1 image, extract the radiance values ​​of the three bands for all pixels containing spodumene samples, combine them pairwise, and use the least squares method to fit the linear relationship of thermal infrared radiance of spodumene in each band combination (see...). Figure 2 The linear relationship between the thermal infrared radiance in Band 1 and the band combination in the scatter plot of the SDGSAT-1 satellite thermal infrared bands and the fitted line of spodumene is shown.

[0079] Figure 3 The linear relationship between thermal infrared radiance in the combination of Band 1 and the green band Band 3 is given.

[0080] Figure 4 (A linear relationship diagram of thermal infrared radiance in the combination of Band 2 and Band 3 (green band) is shown), and then the spodumene thermal infrared remote sensing index ST is calculated. i ,See Figure 7 SDGSAT-1ST1 image in spodumene remote sensing index raster image. Figure 8 SDGSAT-1ST2 image in spodumene remote sensing index raster image. Figure 9 SDGSAT-1ST3 image in the spodumene remote sensing index raster image.

[0081] ST i =|Band b -Band a ×m i -c i |

[0082] In the formula, Band a Band b These are two bands in the i-th band combination, m i ,c i These represent the slope and intercept of the linear relationship obtained by linear fitting of spodumene in the t band combination, respectively.

[0083] S204 combines one visible-near-infrared remote sensing index, one short-wave infrared remote sensing index, and three thermal infrared remote sensing indices corresponding to each pixel to obtain a three-dimensional spodumene remote sensing index cube, which is then arranged in pixel order and stored in a two-dimensional CSV format.

[0084] S3. Construct a dual-branch neural network model based on graph convolutional network and one-dimensional convolutional neural network. Based on this dual-branch neural network model, identify the spatial distribution characteristics and spectral characteristics of spodumene and obtain the identification result image.

[0085] For example, a dual-branch neural network model based on graph convolutional networks and one-dimensional convolutional neural networks is constructed. The spatial information of the spodumene remote sensing spectral index cube is extracted using the GCN model, and the spectral information of the spodumene remote sensing spectral index cube is extracted using 1D-CNN. Finally, the spodumene identification based on the dual-branch neural network model is completed. The specific steps are as follows:

[0086] S301, use Matlab software to create the training data required, including:

[0087] (1) Read the image data and divide it into training and test sets according to the label of each pixel;

[0088] (2) Data conversion is performed, converting the three-dimensional remote sensing image data and label data into two-dimensional matrices respectively. All band information of each pixel is expanded into a vector, so that each pixel occupies a row in the matrix. This conversion combines the spatial structure information and spectral information of the original image, which facilitates subsequent feature processing and analysis;

[0089] (3) Calculate the Laplacian matrix required for the graph convolutional network. The Laplacian matrix is ​​constructed based on the similarity between samples and reflects the connectivity of data points in the graph structure. By calculating the Laplacian matrix, the local geometric structure in the data can be captured, which can be used to smooth and aggregate node features in the graph convolutional network, thereby improving the learning ability of the model;

[0090] (4) Save the training and test data as .mat files respectively.

[0091] S302, a recognition model using Python to construct a two-branch neural network, including:

[0092] (a) Import the necessary Python libraries such as NumPy, Tensorflow, and SciPy;

[0093] (b) Define auxiliary functions, including:

[0094] sample_mask: Creates a sample mask for data segmentation;

[0095] `create_placeholders`: Creates TensorFlow placeholders for input data and labels.

[0096] initialize_parameters: Initializes network parameters, including weights and biases;

[0097] GCN_layer: Defines the operation of the graph convolutional layer. The input is the feature matrix and the Laplacian matrix. The feature matrix is ​​first multiplied by the weight matrix and then multiplied by the Laplacian matrix to achieve feature smoothing and aggregation, thereby extracting local features in the graph structure.

[0098] mynetwork: Defines the forward propagation process of the hybrid model. The input features are first convolved and smoothed through graph convolutional layers (GCN layers), and then extracted together with another part of the data through a one-dimensional convolutional neural network (1D-CNN).

[0099] (c) Define the network structure, which consists of graph convolutional layers and a 1D-CNN. The graph convolutional layers perform convolution and smoothing operations on the features using the Laplacian matrix to extract local features from the graph structure. Subsequently, the 1D-CNN further convolves these features to extract higher-level spatial features, and then processes them through pooling and normalization. Finally, classification is performed through fully connected layers, outputting a probability distribution of spodumene categories. That is, each grid cell corresponds to a probability value for that grid cell being spodumene. Through this probability distribution, users can directly obtain the classification confidence of each input grid cell from the model output (see details in [link to documentation]). Figure 10 (Diagram of the computation process of a two-branch neural network);

[0100] (d) Cross-entropy is used as the loss function to measure the difference between the model's predictions and the true labels. L2 regularization is introduced to prevent overfitting, and the complexity of the model is limited by incorporating the sum of squared weight parameters into the loss function. Finally, the Adam optimizer is used to optimize the total loss (the sum of cross-entropy loss and L2 regularization loss). By adjusting the model parameters, the loss is gradually reduced, thereby improving the model's accuracy and generalization ability.

[0101] (e) Leave-one-out cross-validation is used for deep learning model training, where one sample is used as the validation set and the remaining samples as the training set in each round to optimize the model parameters. In each round, the model updates the weights through forward and back propagation, minimizes the loss function including cross-entropy loss and L2 regularization, and evaluates its performance on the validation set. After all cross-validation rounds are completed, the model is finally trained on the entire training set to fully utilize all data to optimize the parameters. Subsequently, the model makes predictions on the test set, calculates the class probability distribution for each sample through forward propagation, and classifies the samples according to these probabilities, ultimately achieving accurate identification of spodumene. (See...) Figure 11 (Remote sensing identification results of spodumene); the overall parameters of the model are shown in Table 1.

[0102] Table 1. Core parameters of the dual-branch neural network model

[0103]

[0104]

[0105] (f) Save the predicted probabilities of the test set as MAT and CSV files for subsequent analysis.

[0106] S4, Evaluation of Recognition Results

[0107] For example, to evaluate model performance, calculate metrics such as accuracy, precision, recall, and F1 score, and plot the AUC-ROC curve.

[0108] Accuracy measures the proportion of samples that are correctly classified out of all samples. It takes into account the correctness of all predictions.

[0109]

[0110] Precision is the ratio of true positives to all predicted positives, representing the model's ability to avoid false positives. It reflects the model's ability to avoid false alarms.

[0111]

[0112] Recall is the ratio of true positives to all actual positives, reflecting the model's ability to identify relevant instances.

[0113] It reflects the model's ability to identify relevant instances.

[0114]

[0115] The F1 score is the harmonic mean of precision and recall, providing a balance between the two.

[0116]

[0117] AUC-ROC measures a model’s ability to distinguish between classes by evaluating the trade-off between the true positive rate (TPR) and the false positive rate (FPR) at different threshold settings.

[0118]

[0119] Among them, a true positive (TP) is an instance that is correctly identified as positive, a false positive (FP) is an instance that is incorrectly identified as positive, a true negative (TN) is an instance that is correctly identified as negative, and a false negative (FN) is an instance that is incorrectly identified as negative.

[0120] Compared with the use of visible-near infrared index, shortwave infrared index and combination of the two alone, the results of multi-source remote sensing index identification with spodumene thermal infrared remote sensing index have achieved better results in all evaluation indicators, as shown in Table 2.

[0121] Table 2 Quantitative evaluation results of different combinations of remote sensing indices

[0122]

[0123] Example 2: This embodiment of the invention provides a multi-source remote sensing spodumene identification system, which includes:

[0124] The remote sensing image processing module is used for field sample collection, remote sensing image acquisition and preprocessing.

[0125] The Remote Sensing Index Cube Creation Module is used to calculate the spodumene spectral index based on the spectral characteristics of remote sensing images and to create a spodumene spectral remote sensing index cube.

[0126] The dual-branch neural network model construction module is used to construct a dual-branch neural network model based on graph convolutional networks and one-dimensional convolutional neural networks. Based on this dual-branch neural network model, the spatial distribution characteristics and spectral characteristics of spodumene are identified, and the identification result image is obtained.

[0127] The evaluation module is used to identify and evaluate the results.

[0128] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A multi-source remote sensing method for identifying spodumene based on spectral index and a dual-branch neural network, characterized in that, The method includes: S1, Field sample collection, remote sensing image acquisition and preprocessing; S2. Based on the spectral characteristics of remote sensing images, calculate the spodumene spectral index and create a spodumene spectral remote sensing index cube. S3. Construct a dual-branch neural network model based on graph convolutional network and one-dimensional convolutional neural network. Based on this dual-branch neural network model, identify the spatial distribution characteristics and spectral characteristics of spodumene and obtain the identification result image. S4, Evaluation of recognition results; In step S2, the spodumene spectral index is calculated, and a spodumene spectral remote sensing index cube is constructed, including: S201. Select the green band and near-infrared band of the preprocessed Landsat 8 image in the visible-near-infrared range and calculate the band ratio to obtain the spodumene visible-near-infrared remote sensing index BR. S202. Select the shortwave infrared band of the preprocessed GF-5 image, extract the spectrum of the pixel where spodumene is located in the collected sample and take the average as the reference spectrum of the shortwave infrared band of spodumene. Use the matched filtering method to calculate the matched filter value MF between all pixels and the reference spectrum, and use it as the shortwave infrared remote sensing index of spodumene. S203. For the preprocessed SDGSAT-1 image, extract the three band radiance values ​​of all spodumene samples in the pixel, combine them in pairs, and use the least squares method to fit the linear relationship of thermal infrared radiance of spodumene in each band combination to obtain three thermal infrared remote sensing indices. S204 combines the visible-near infrared remote sensing index, short-wave infrared remote sensing index and three thermal infrared remote sensing indices corresponding to each pixel to obtain a three-dimensional spodumene remote sensing index cube, which is arranged in pixel order and stored in two-dimensional CSV format. In step S203, the thermal infrared remote sensing index The calculation formula is: ； In the formula, The first Two bands in a band combination, Lithium spodumene in the 1st The slope and intercept of the linear relationship obtained by linear fitting of a combination of bands.

2. The method according to claim 1, wherein, In step S1, field sample collection, remote sensing image acquisition and preprocessing include: for Landsat 8 multispectral and GF-5 hyperspectral images, radiometric calibration and atmospheric correction are performed to convert the original values ​​into reflectance; for SDGSAT-1 thermal infrared images, radiometric calibration is performed to convert the original values ​​into radiance; after completion, the three images are registered and resampled respectively.

3. The method according to claim 1, wherein, In step S201, the formula for calculating the visible-near-infrared remote sensing index BR of spodumene is as follows: ； In the formula, is a green waveband, is a near infrared waveband.

4. The method according to claim 1, wherein, In step S202, the formula for calculating the matched filter value MF is: ； In the formula, For the pixel in the first The actual spectral value of the band, For spodumene in the first Reference spectral values ​​for the band, This represents the total number of bands.

5. The method according to claim 1, wherein, In step S3, a dual-branch neural network model based on graph convolutional networks and one-dimensional convolutional neural networks is constructed. This dual-branch neural network model is used to identify the spatial distribution characteristics and spectral characteristics of spodumene, including: S301, use Matlab software to create the data required for training; S302 uses the Python language to build deep learning models.

6. The method according to claim 5, wherein, In step S301, the training data is prepared using Matlab software, including: (1) Read the image data and divide it into training and test sets according to the label of each pixel; (2) Perform data conversion, converting the three-dimensional remote sensing image data and label data into two-dimensional matrices respectively; (3) Calculate the Laplacian matrix required for the graph convolutional network; (4) Save the training and test data as .mat files respectively.

7. The method according to claim 5, wherein, In step S302, a deep learning model is constructed using the Python language, including: (a) Import the Python libraries NumPy, Tensorflow, and SciPy; (b) Define auxiliary functions, including: `sample_mask` creates a sample mask for data segmentation. `create_placeholders` creates TensorFlow placeholders for input data and labels; initialize_parameters initializes the network parameters, including weights and biases; GCN_layer defines the operation of the graph convolutional layer, with the input being the feature matrix and the Laplacian matrix; Mynetwork defines the forward propagation process of a hybrid model, which includes GCN layers and 1D-CNN. (c) Define the network structure, which consists of graph convolutional layers and 1D-CNN. The graph convolutional layers perform convolution operations on the features through the Laplacian matrix. The 1D-CNN is used to process the temporal information of the feature vectors. Finally, the network is classified through fully connected layers to output the probability distribution of spodumene categories. (d) Use cross-entropy as the loss function, add L2 regularization, and use the Adam optimizer to update the parameters; (e) Use leave-one-out cross-validation to train the deep learning model. Leave one sample as the validation set each time and the rest as the training set. Then train the deep learning model on the entire training set and make predictions on the test set to complete the identification of spodumene. (f) Save the predicted probabilities of the test set as MAT and CSV files for subsequent analysis.

Images