Lithology identification method and device based on dual-channel convolutional neural network

Abstract

The invention provides a lithology identification method and device based on a two-channel convolutional neural network, and relates to the technical field of information processing, and the method comprises the steps: obtaining a hyperspectral image sample and a multi-temporal multispectral image sample; preprocessing the hyperspectral image sample and the multi-temporal multi-spectral image sample, respectively inputting the preprocessed hyperspectral image sample and the preprocessed multi-temporal multi-spectral image sample into a one-dimensional convolution model and a two-dimensional convolution model, and outputting to obtain a target spectral feature and a target spatial feature; fusing the target spectral feature and the target spatial feature to obtain a fused feature, and training a preset fusion model based on the fused feature to obtain a target fusion model; and outputting a to-be-identified image to the target fusion model, and determining the lithology type corresponding to the to-be-identified image. According to the method, different features are learned by using the one-dimensional convolution model and the two-dimensional convolution model respectively, and the output features of two different channels are fused, so that the lithology of the remote sensing image is identified, and the accuracy and reliability of an identification result are improved.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a lithology identification method and device based on a dual-channel convolutional neural network. Background Art

[0002] Lithology identification can provide an important research basis for various geological work, such as mineral exploration and geological mapping. Traditional lithology identification mainly relies on manual field investigation methods, which is not only time-consuming and labor-intensive, but also difficult to carry out in areas with harsh natural conditions. In recent years, with the development of remote sensing technology, it has been possible to carry out large-scale detection of complex terrain, providing a reliable means for field lithology identification, and to a certain extent promoting the development of geological work. Hyperspectral imaging is a collection of spectral image data that captures each pixel band in hundreds of continuous and narrow bands through remote sensing systems. It has the characteristics of many bands, rich information, and high spectral resolution. The information contained in hyperspectral images can reflect the external characteristics of the sample, such as size, shape, and volume. These characteristics determine the unique advantages of hyperspectral images in some tasks that require internal and external feature detection. The continuous band data of hyperspectral images provides detailed information, which can be used to identify the types of ground objects. Multi-temporal multispectral imaging is a technology that obtains multispectral image data of the target area at different time nodes to reveal the changing characteristics and spectral characteristics of the target in the time series. By acquiring image data continuously or periodically, the dynamic changes of the target in the time series can be monitored. Different spectral bands correspond to different material properties and environmental information, and multispectral data can provide richer information dimensions. Therefore, hyperspectral images and multi-phase multispectral images can be used for lithology identification.

[0003] However, most of the remote sensing images currently used are multispectral images. The spectral characteristics of similar rock types are difficult to distinguish, and the influence of external environmental factors is rarely considered. There is relatively little research on remote sensing rock type identification by integrating the spectral characteristics and spatial characteristics of remote sensing images. How to use various remote sensing detection methods to improve the accuracy of rock type identification has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that the spectral characteristics of similar lithologies are difficult to distinguish and the influence of external environmental factors is rarely considered, resulting in low lithology identification accuracy. The present invention provides a lithology identification method and device based on a dual-channel convolutional neural network.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] A first aspect of an embodiment of the present application provides a lithology identification method based on a dual-channel convolutional neural network, comprising:

[0007] Obtain hyperspectral image samples and multi-temporal multispectral image samples;

[0008] The hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are output;

[0009] The target spectral feature and the target spatial feature are fused to obtain a fusion feature, and a preset fusion model is trained based on the fusion feature to obtain a target fusion model;

[0010] The image to be identified is output to the target fusion model to determine the lithology type corresponding to the image to be identified.

[0011] Optionally, after obtaining the hyperspectral image samples and the multi-temporal multispectral image samples, the method further includes:

[0012] The hyperspectral image samples and the multi-temporal multispectral image samples are preprocessed, including band removal, bad line and stripe processing, radiation calibration, atmospheric correction and image fusion.

[0013] Optionally, the hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are outputted, including:

[0014] The pixels in the hyperspectral image sample are divided into a training set and a validation set to train the one-dimensional convolution model, learn the spectral characteristics of each pixel, and output the target spectral characteristics.

[0015] Optionally, the hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are outputted, including:

[0016] The pixel blocks in the multi-phase multispectral image samples are divided into a training set and a validation set to train the two-dimensional convolution model, learn the spatial features of each pixel block, and output the target spatial features; each pixel block includes at least nine pixels.

[0017] Optionally, fusing the target spectral feature and the target spatial feature to obtain a fused feature includes:

[0018] The target spectral feature and the target spatial feature are fused by splicing to obtain a fused feature.

[0019] Optionally, the training of a preset fusion model based on the fusion feature to obtain a target fusion model includes:

[0020] The preset fusion model is trained based on the fusion features, and the parameters of the preset fusion model are adjusted in combination with the cross entropy loss function to obtain the target fusion model.

[0021] Optionally, the step of outputting the image to be identified to the target fusion model to determine the lithology type corresponding to the image to be identified includes:

[0022] The SHAP method is used to determine the contribution size and direction of each type of spatial feature to the identification of lithology types.

[0023] The second aspect of the embodiment of the present application provides a lithology identification device based on a dual-channel convolutional neural network, including: a sample acquisition module, an input and output module, a fusion training module and a determination module, wherein:

[0024] The sample acquisition module is configured to acquire hyperspectral image samples and multi-temporal multispectral image samples;

[0025] The input-output module is configured to input the hyperspectral image samples and the multi-temporal multispectral image samples into a one-dimensional convolution model and a two-dimensional convolution model respectively after preprocessing, and output the target spectral features and the target spatial features;

[0026] The fusion training module is configured to fuse the target spectral feature and the target spatial feature to obtain a fusion feature, and train a preset fusion model based on the fusion feature to obtain a target fusion model;

[0027] The determination module is configured to output the image to be identified to the target fusion model to determine the lithology type corresponding to the image to be identified.

[0028] A third aspect of an embodiment of the present application provides an electronic device, comprising a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the lithology identification method based on a dual-channel convolutional neural network described in the first aspect.

[0029] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0030] Compared with the prior art, the technical solution provided by this application has the following beneficial effects:

[0031] The present invention provides a lithology identification method and device based on a dual-channel convolutional neural network, which obtains hyperspectral image samples and multi-phase multispectral image samples, pre-processes the hyperspectral image samples and the multi-phase multispectral image samples, and respectively inputs them into a one-dimensional convolution model and a two-dimensional convolution model, outputs target spectral features and target spatial features, fuses the target spectral features and the target spatial features to obtain fusion features, and trains a preset fusion model based on the fusion features to obtain a target fusion model, outputs the image to be identified to the target fusion model, and determines the lithology type corresponding to the image to be identified. By using a one-dimensional convolution model and a two-dimensional convolution model to learn different features respectively, and fusing the output features of two different channels, the lithology of the remote sensing image can be identified, and the accuracy and reliability of the identification result are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic diagram of a process flow of a lithology identification method based on a dual-channel convolutional neural network provided in an embodiment of the present application;

[0033] Figure 2 A schematic diagram of the structure of a lithology identification device based on a dual-channel convolutional neural network provided in an embodiment of the present application;

[0034] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] Below, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application.

[0036] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "include", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0037] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0038] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flow charts may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions, when executed by the processor, may create a device for implementing the functions / operations described in these block diagrams and / or flow charts.

[0039] In some embodiments, see Figure 1 , Figure 1 A schematic diagram of a process flow of a lithology identification method based on a dual-channel convolutional neural network provided in an embodiment of the present application; The lithology identification method based on a dual-channel convolutional neural network provided in an embodiment of the present application includes:

[0040] S110, obtaining hyperspectral image samples and multi-temporal multispectral image samples.

[0041] Hyperspectral images can capture dozens to hundreds of narrow, continuous spectral bands, with band widths typically less than 10nm, providing ultra-high spectral resolution and a wide spectral range. Multi-temporal multispectral images usually record a limited number of spectral bands, with wider bands and relatively low spectral resolution, and the spectral range is limited to the visible and near-infrared regions. Multi-temporal multispectral images can include image data of the same region at different times. Here, hyperspectral images and multi-temporal multispectral images need to have the same resolution and range to ensure that the subsequent two channels have the same number of samples.

[0042] In one example, the hyperspectral image sample is a GF-5B hyperspectral image, and the multi-temporal multispectral image sample is a Landsat-8 multi-temporal multispectral image.

[0043] In some embodiments, after acquiring the hyperspectral image sample and the multi-temporal multispectral image sample at S110, the method further includes:

[0044] Preprocessing of hyperspectral image samples and multi-temporal multispectral image samples includes band removal, bad line and stripe processing, radiation calibration, atmospheric correction and image fusion.

[0045] Band rejection is to remove invalid or unreliable bands caused by equipment failure, noise interference or poor data quality to improve the accuracy of subsequent processing and analysis; bad line and streak processing is to eliminate or weaken bad line and streak noise in the image caused by sensor failure, data transmission error or non-uniformity in the imaging process; radiation calibration is to convert the original DN (digital number) value recorded by the sensor into a radiation brightness or reflectivity value with physical meaning; atmospheric correction is to eliminate the influence of the atmosphere on the image, such as atmospheric absorption and scattering, to restore the true reflectivity or radiation brightness of the surface; image fusion is to fuse image data of different resolutions or different sources to improve the resolution and information content of the image.

[0046] S120, pre-processing the hyperspectral image samples and the multi-temporal multispectral image samples, respectively inputting them into a one-dimensional convolution model and a two-dimensional convolution model, and outputting target spectral features and target spatial features.

[0047] The input and output of the one-dimensional convolution model are both one-dimensional arrays. In one-dimensional convolution, the convolution kernel slides along the one-dimensional direction of the input data. Each slide performs a dot product operation with a local area of ​​the input data and adds a bias term. During the sliding process, the weights of the convolution kernel are shared. The input and output of the two-dimensional convolution model are both two-dimensional matrices. The convolution operation is performed on the input image by sliding the two-dimensional convolution kernel to extract the local features of the image. In two-dimensional convolution, the convolution kernel slides along the two-dimensional direction of the input image (usually width and height) and performs a cross-correlation operation with a local area of ​​the input image. During the operation, the weight sharing strategy is also adopted.

[0048] In one example, for the GF-5B hyperspectral image, the spectral curve composed of the remaining 293 bands after band removal is the target spectral feature; for the multi-temporal Landsat-8 image, the TCT transform is used to extract the multi-year average humidity, greenness and brightness characteristics, the GLCM transform is used to extract 8 texture features, and the TIRS band of the Landsat8 image is used with the help of a single-channel algorithm to extract the surface temperature feature (LST), thus obtaining a spectral feature image containing 293 bands and a spatial feature image containing 12 target spatial features.

[0049] In some embodiments, S120, the hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are output, including:

[0050] The pixels in the hyperspectral image samples are divided into a training set and a validation set to train the one-dimensional convolution model, learn the spectral characteristics of each pixel, and output the target spectral characteristics.

[0051] In this embodiment, a sample set and a label set of a spectral channel are established with a single pixel as a sample, which are recorded as X_spectral and Y_spectral respectively. Then, the training set and the validation set are divided according to the ratio and input into the one-dimensional convolution model to learn the spectral characteristics of each pixel. After the input and output errors of the one-dimensional convolution model meet the expectations, at least one frame of the hyperspectral image sample is used as input to obtain the target spectral characteristics.

[0052] In some embodiments, S120, the hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are output, including:

[0053] The pixel blocks in the multi-temporal and multispectral image samples are divided into a training set and a validation set to train the two-dimensional convolutional model, learn the spatial features of each pixel block, and output the target spatial features; each pixel block includes at least nine pixels.

[0054] In this embodiment, each pixel on the image is traversed and taken as the center to construct a pixel block. The label is based on the central pixel. The sample set and label set of the spatial channel are established with the pixel block as the sample, which are recorded as X_spatial and Y_spatial respectively. Then the training set and the validation set are divided according to the proportion and input into the two-dimensional convolution model to learn the spatial features of each pixel block. After the input and output errors of the two-dimensional convolution model meet the expectations, at least one frame of the multi-phase multispectral image sample is used as input to obtain the target spatial features. Exemplarily, the pixel block can be a 3×3 pixel block, a 5×5 pixel block.

[0055] S130, fusing the target spectral feature and the target spatial feature to obtain a fusion feature, and training a preset fusion model based on the fusion feature to obtain a target fusion model.

[0056] Fusion is performed at the feature level, that is, the target spectral features and target spatial features are combined into a unified feature vector. During the model training phase, the parameters of the preset fusion model need to be continuously corrected through the error function so that the error function of the preset fusion model is reduced to the expected threshold.

[0057] In some embodiments, S130, the target spectral feature and the target spatial feature are fused to obtain a fused feature, including:

[0058] The target spectral features and target spatial features are fused by splicing to obtain fused features.

[0059] Here, it is necessary to pre-vectorize, normalize and / or standardize the target spectral features and target spatial features to ensure that the two features are in a similar range in value and in the same dimension. On this basis, the target spectral features and target spatial features are merged into a new feature vector by splicing. Exemplarily, assuming that the target spectral feature vector s is a one-dimensional array of length n, and the target spatial feature vector p is a one-dimensional array of length m (if the target spatial feature is two-dimensional, it needs to be flattened first), then the spliced ​​fusion feature vector f will be a one-dimensional array of length n+m.

[0060] In an optional embodiment, the pixels used for training in the two data sets containing the target spectral features and the target spatial features are respectively input into the two channels of the preset fusion model. Channel one uses one-dimensional convolution to accept spectral samples, and channel two uses two-dimensional convolution to accept spatial samples. The outputs of the two channels are respectively defined as:

[0061] X_spe = ReLU(W1X1+b1);

[0062] X_spe = ReLU(W2X2+b2);

[0063] Among them, W1 and b1 represent the weight matrix and bias vector of the spectral channel respectively, and X1 represents the output of the previous layer; W2 and b2 represent the weight matrix and bias vector of the spatial channel respectively, and X2 represents the output of the previous layer. After the two channels are trained, the features are fused by splicing.

[0064] In some embodiments, S130, training a preset fusion model based on the fusion feature to obtain a target fusion model includes:

[0065] The preset fusion model is trained based on the fusion features, and the parameters of the preset fusion model are adjusted in combination with the cross entropy loss function to obtain the target fusion model.

[0066] The cross entropy loss function L can be expressed as:

[0067]

[0068] Among them, M is the total number of samples, N is the total number of categories, and y n (i) is the indicator function of the true label of the i-th sample in category n, that is, when the sample belongs to category n, y n (i) =1, otherwise y n (i) =0;y n (i) is the probability that the model predicts that the i-th sample belongs to category c.

[0069] S140: Output the image to be identified to the target fusion model to determine the lithology type corresponding to the image to be identified.

[0070] Lithology mainly includes two categories: rock and soil. Among them, rock can be further divided into sedimentary rock, igneous rock and metamorphic rock according to its origin.

[0071] In some embodiments, S140, outputting the image to be identified to the target fusion model, and determining the lithology type corresponding to the image to be identified, includes:

[0072] The SHAP method is used to determine the contribution size and direction of each type of spatial feature to the identification of lithology types.

[0073] The SHAP method is based on the Shapley value theory, which aims to explain the contribution of each feature in the prediction model to the model output. The Shapley value is a concept derived from cooperative game theory and is used to fairly distribute the benefits of cooperation. In machine learning models, the SHAP value can be interpreted as the average marginal contribution of each feature to the model prediction results. Here, the confusion matrix can be used to calculate the overall accuracy (OA), average accuracy (AA) and Kappa coefficient to evaluate the accuracy of the model:

[0074]

[0075]

[0076]

[0077] Where N is the number of categories, M is the total number of samples, and TP i is the true positive class of the i-th class, FNi is the false negative class of the i-th class, and FPi is the false positive class of the i-th class.

[0078] By selecting samples of different lithology categories, the SHAP value can be calculated. For the SHAP value of feature Xi The calculation formula is as follows:

[0079]

[0080] Among them, N is the set of all features, expressed as {X1, X2, ..., XM}; S is a subset of the feature set N, excluding feature i; |S| is the size of the subset S; f(S) is the predicted output of the model when it only contains the feature subset S; f(S∪{i}) is the predicted output of the model when it contains the feature subset S and feature i. By extracting the SHAP value of the spatial feature Xi, drawing its SHAP feature importance bar chart and SHAP force diagram, the contribution size and contribution direction of these features to the identification of specific categories of lithology can be determined, and the interpretability of the model can be achieved to a certain extent. It should be noted that when performing SHAP interpretation on the model, it is necessary to select a suitable interpreter, and the number and order of features need to be consistent with the band order of the remote sensing image used.

[0081] The embodiment of the present application obtains hyperspectral image samples and multi-phase multispectral image samples, pre-processes the hyperspectral image samples and the multi-phase multispectral image samples, and respectively inputs them into a one-dimensional convolution model and a two-dimensional convolution model, outputs target spectral features and target spatial features, fuses the target spectral features and the target spatial features to obtain fusion features, and trains a preset fusion model based on the fusion features to obtain a target fusion model, outputs the image to be identified to the target fusion model, and determines the lithology type corresponding to the image to be identified. By using a one-dimensional convolution model and a two-dimensional convolution model to learn different features respectively, and fusing the output features of two different channels, the lithology of the remote sensing image can be identified, and the accuracy and reliability of the identification results are improved.

[0082] In some embodiments, see Figure 2 , Figure 2 The structural diagram of a lithology identification device based on a dual-channel convolutional neural network provided in an embodiment of the present application. The lithology identification device 200 based on a dual-channel convolutional neural network provided in an embodiment of the present application includes: a sample acquisition module 210, an input and output module 220, a fusion training module 230 and a determination module 240, wherein:

[0083] The sample acquisition module 210 is configured to acquire hyperspectral image samples and multi-temporal multispectral image samples.

[0084] The input-output module 220 is configured to input the hyperspectral image samples and the multi-temporal multispectral image samples into the one-dimensional convolution model and the two-dimensional convolution model respectively after preprocessing, and output the target spectral features and the target spatial features.

[0085] The fusion training module 230 is configured to fuse the target spectral feature and the target spatial feature to obtain a fusion feature, and train a preset fusion model based on the fusion feature to obtain a target fusion model.

[0086] The determination module 240 is configured to output the image to be identified to the target fusion model to determine the lithology type corresponding to the image to be identified.

[0087] In some embodiments, the lithology identification device 200 based on a dual-channel convolutional neural network further includes a preprocessing module; the preprocessing module is specifically configured as follows:

[0088] Preprocessing of hyperspectral image samples and multi-temporal multispectral image samples includes band removal, bad line and stripe processing, radiation calibration, atmospheric correction and image fusion.

[0089] In some embodiments, the input-output module 220 is specifically configured as follows:

[0090] The pixels in the hyperspectral image samples are divided into a training set and a validation set to train the one-dimensional convolution model, learn the spectral characteristics of each pixel, and output the target spectral characteristics.

[0091] In some embodiments, the input-output module 220 is specifically configured as follows:

[0092] The pixel blocks in the multi-temporal and multispectral image samples are divided into a training set and a validation set to train the two-dimensional convolutional model, learn the spatial features of each pixel block, and output the target spatial features; each pixel block includes at least nine pixels.

[0093] In some embodiments, the fusion training module 230 is specifically configured as follows:

[0094] The target spectral features and target spatial features are fused by splicing to obtain fused features.

[0095] In some embodiments, the fusion training module 230 is specifically configured as follows:

[0096] The preset fusion model is trained based on the fusion features, and the parameters of the preset fusion model are adjusted in combination with the cross entropy loss function to obtain the target fusion model.

[0097] In some embodiments, the determination module 240 is specifically configured as follows:

[0098] The SHAP method is used to determine the contribution size and direction of each type of spatial feature to the identification of lithology types.

[0099] The lithology identification device based on a dual-channel convolutional neural network provided in the embodiment of the present application can implement each process in the embodiment corresponding to the above-mentioned lithology identification method based on a dual-channel convolutional neural network. To avoid repetition, they will not be described here.

[0100] It should be noted that the lithology identification device based on a dual-channel convolutional neural network provided in the embodiment of the present application and the lithology identification method based on a dual-channel convolutional neural network provided in the embodiment of the present application are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned lithology identification method based on a dual-channel convolutional neural network, and the repeated parts will not be repeated.

[0101] In some embodiments, see Figure 3 , Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. An electronic device 300 provided in an embodiment of the present application includes a processor 310 and a memory 320; the memory 320 stores a computer program, wherein the computer program implements the above-mentioned lithology identification method based on a dual-channel convolutional neural network when executed by the processor.

[0102] Specifically, the processor 310 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 310 may also include an onboard memory for cache purposes. The processor 310 may be a single processing unit or multiple processing units for executing different actions of the method flow according to the embodiment of the present application.

[0103] The memory 320 may be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, the memory 320 may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device, component or propagation medium. Specific examples of the memory 320 include: a magnetic storage device, such as a magnetic tape or a hard disk (HDD); an optical storage device, such as a compact disk (CD-ROM); a random access memory (RAM) or flash memory; and / or a wired / wireless communication link.

[0104] The present application also provides a computer-readable medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned lithology identification method based on a dual-channel convolutional neural network is implemented. The computer-readable medium may be included in the device / apparatus / system described in the above-mentioned embodiment; or it may exist independently without being assembled into the device / apparatus / system. The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed, the method according to the embodiment of the present application is implemented.

[0105] According to an embodiment of the present application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, optical cable, radio frequency signal, etc., or any suitable combination of the above.

[0106] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present application. Therefore, the scope of the present application should not be limited to the above-described embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

Claims

1. A lithology identification method based on a dual-channel convolutional neural network, characterized in that: include: Obtain hyperspectral image samples and multi-temporal multispectral image samples; The hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are output; The target spectral feature and the target spatial feature are fused to obtain a fusion feature, and a preset fusion model is trained based on the fusion feature to obtain a target fusion model; The image to be identified is output to the target fusion model to determine the lithology type corresponding to the image to be identified.

2. The lithology identification method based on a dual-channel convolutional neural network according to claim 1 is characterized in that: After obtaining the hyperspectral image samples and the multi-temporal multispectral image samples, the method further includes: The hyperspectral image samples and the multi-temporal multispectral image samples are preprocessed, including band removal, bad line and stripe processing, radiation calibration, atmospheric correction and image fusion.

3. The lithology identification method based on a dual-channel convolutional neural network according to claim 1 is characterized in that: The hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are outputted, including: The pixels in the hyperspectral image sample are divided into a training set and a validation set to train the one-dimensional convolution model, learn the spectral characteristics of each pixel, and output the target spectral characteristics.

4. The lithology identification method based on a dual-channel convolutional neural network according to claim 1 is characterized in that: The hyperspectral image samples and the multi-temporal multispectral image samples are pre-processed and then respectively input into a one-dimensional convolution model and a two-dimensional convolution model, and the target spectral features and the target spatial features are outputted, including: The pixel blocks in the multi-phase multispectral image samples are divided into a training set and a validation set to train the two-dimensional convolution model, learn the spatial features of each pixel block, and output the target spatial features; each pixel block includes at least nine pixels.

5. The lithology identification method based on dual-channel convolutional neural network according to claim 1 is characterized in that: The step of fusing the target spectral feature and the target spatial feature to obtain a fused feature includes: The target spectral feature and the target spatial feature are fused by splicing to obtain a fused feature.

6. The lithology identification method based on dual-channel convolutional neural network according to claim 1 is characterized in that: The step of training a preset fusion model based on the fusion features to obtain a target fusion model includes: The preset fusion model is trained based on the fusion features, and the parameters of the preset fusion model are adjusted in combination with the cross entropy loss function to obtain the target fusion model.

7. The lithology identification method based on dual-channel convolutional neural network according to claim 1 is characterized in that: Outputting the image to be identified to the target fusion model to determine the lithology type corresponding to the image to be identified includes: The SHAP method is used to determine the contribution size and direction of each type of spatial feature to the identification of lithology types.

8. A lithology identification device based on a dual-channel convolutional neural network, characterized in that: include: Sample acquisition module, input and output module, fusion training module and determination module, among which, The sample acquisition module is configured to acquire hyperspectral image samples and multi-temporal multispectral image samples; The input-output module is configured to input the hyperspectral image samples and the multi-temporal multispectral image samples into a one-dimensional convolution model and a two-dimensional convolution model respectively after preprocessing, and output the target spectral features and the target spatial features; The fusion training module is configured to fuse the target spectral feature and the target spatial feature to obtain a fusion feature, and train a preset fusion model based on the fusion feature to obtain a target fusion model; The determination module is configured to output the image to be identified to the target fusion model to determine the lithology type corresponding to the image to be identified.

9. An electronic device comprising a processor and a memory; the memory stores a computer program, wherein: When the computer program is executed by the processor, the computer program implements the lithology identification method based on a dual-channel convolutional neural network described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images