Hyperspectral image Mars surface mineral identification method, device, equipment and medium

Abstract

The invention relates to a hyperspectral image Mars surface mineral identification method and device, equipment and a medium. The method comprises the following steps: acquiring a hyperspectral image for a mineral on the Mars surface; and inputting the hyperspectral image into a pre-trained recognition model, performing layer-by-layer extraction of pixel point features on the hyperspectral image through each layer of classification module in the recognition model, and obtaining a recognition result of the minerals on the surface of the Mars based on the pixel point features obtained through layer-by-layer extraction. Through the method of the invention, high-resolution and high-precision rapid detection can be carried out on Mars minerals through a machine learning technology.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a method, device, equipment and medium for identifying minerals on the surface of Mars using hyperspectral images. Background Art

[0002] Using data collected by CRISM, scientists can analyze the distribution of minerals such as carbonates, silicates, and sulfates on the Martian surface. This analysis is crucial for understanding Martian geological activity and climate change. Furthermore, by detecting and identifying water-bearing minerals, researchers can trace the history of water on Mars, providing a foundation for exploring the possibility of life on Mars and possibly other planets.

[0003] Traditional methods for identifying minerals on the Martian surface primarily rely on spectral summary parameters. While this method has achieved some research results, it relies on manual calculations, resulting in a low degree of automation and high sensitivity to noise, leaving room for improvement in the accuracy of mineral identification. Other methods include spectral matching and identification based on spectral characteristic parameters. While the spectral matching method is simple in principle, it is highly sensitive to noise and requires a high spectral signal-to-noise ratio. While methods based on spectral characteristic parameters have achieved some success in identifying minerals on the Martian surface, their degree of automation and accuracy are limited. Furthermore, CRISM imagery, as a type of hyperspectral data, despite its high characteristic dimensionality, also suffers from information redundancy, which can interfere with accurate mineral identification.

[0004] With technological advances, deep learning networks have been introduced to the identification of Martian surface minerals, primarily including spectral feature classification networks and spectral-spatial feature classification networks. Spectral feature classification networks use the spectral information of hyperspectral data for classification, but they make insufficient use of spatial information. While spectral-spatial feature classification networks attempt to integrate multi-scale information and apply attention mechanisms, these efforts are still relatively rudimentary. Most methods employ pixel-by-pixel analysis and rely on the classification of image pixel blocks, resulting in low recognition efficiency.

[0005] Furthermore, existing technologies often use feature extraction to extract important information from hyperspectral imagery, using dimensionality reduction techniques to retain as much valid information or features as possible. However, these general dimensionality reduction methods fail to fully account for the nonlinear relationships in hyperspectral data and remain insufficient in extracting key information. This can make it difficult to fully extract the deep spectral characteristics of Martian surface minerals from CRISM hyperspectral imagery, leading to inaccurate identification results for Martian surface minerals. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method, device, equipment and medium for identifying minerals on the surface of Mars using hyperspectral images, aiming to solve the technical problem of inaccurate identification results mentioned above.

[0007] In a first aspect, the present invention provides a technical solution to the above-mentioned technical problem as follows: a method for identifying minerals on the surface of Mars using hyperspectral images, the method comprising:

[0008] Acquire hyperspectral images of minerals on the Martian surface;

[0009] The hyperspectral image is input into a pre-trained recognition model, and the pixel features of the hyperspectral image are extracted layer by layer through the classification modules of each layer in the recognition model, and the identification results of the minerals on the surface of Mars are obtained based on the pixel features extracted layer by layer.

[0010] The beneficial effects of the present invention are: by extracting pixel point features of the hyperspectral image layer by layer through each layer classification module, the nonlinear relationship in the hyperspectral data can be fully considered, and the deep-level spectral characteristics of Martian surface minerals can be fully extracted from the hyperspectral image, thereby making the identification results of minerals on the Martian surface more accurate.

[0011] On the basis of the above technical solution, the present invention can also be improved as follows.

[0012] Furthermore, before inputting the hyperspectral image into the pre-trained recognition model, the method further includes:

[0013] performing spectral continuum removal on the hyperspectral image to obtain an image spectrum;

[0014] Performing spectral standardization processing on the image spectrum to obtain a standardized image spectrum.

[0015] The beneficial effect of adopting the above further scheme is that, by removing the spectral continuum, the background effect in the hyperspectral image can be removed, and the absorption characteristics of certain specific substances can be separated. By spectral standardization processing, the influence of differences in test conditions on the spectral similarity calculation can be suppressed, making the final recognition result more accurate.

[0016] Furthermore, the step of removing the spectral continuum from the hyperspectral image to obtain an image spectrum includes:

[0017] The spectral continuum is removed from the hyperspectral image using a first formula to obtain an image spectrum, wherein the first formula is:

[0018]

[0019] Among them, R cris the image spectrum, R is the absorption spectrum of the hyperspectral image, and R c is the continuum of the absorption spectrum.

[0020] The beneficial effect of adopting the above further solution is that, by removing the spectral continuum, the background effect in the hyperspectral image can be removed and the absorption characteristics of certain specific substances can be separated.

[0021] Furthermore, performing spectral standardization on the image spectrum to obtain a standardized image spectrum includes:

[0022] According to the reference spectrum, the image spectrum is spectrally normalized using a second formula to obtain a standardized image spectrum, wherein the second formula is:

[0023]

[0024] Where k is the normalization coefficient determined based on the reference spectrum, is the normalized image spectrum, R cr is the image spectrum before normalization.

[0025] The beneficial effect of adopting the above further solution is that, through spectral standardization processing, the influence of differences in test conditions on the calculation of spectral similarity can be suppressed.

[0026] Furthermore, the recognition model is trained based on the following method:

[0027] Acquire a CRISM hyperspectral image dataset, wherein the CRISM hyperspectral image dataset includes multiple frames of hyperspectral sample images;

[0028] Performing spectral continuum removal and spectral normalization processing on each frame of the hyperspectral sample image in the CRISM hyperspectral image dataset to obtain a training sample;

[0029] Constructing a densely connected random forest classification model, wherein the densely connected random forest classification model includes multiple layers of classification modules, and each layer of classification modules includes random forest classifiers equal to the total number of hyperspectral sample images in the training sample;

[0030] According to the training samples, the densely connected random forest classification model is trained layer by layer to obtain the overall classification accuracy average of multiple random forest classifiers in each layer of classification modules. When the overall classification accuracy average of the classification modules in any layer is less than or equal to the overall classification accuracy average of the classification modules in the previous layer, the training is stopped and the any layer is used as the last layer of the recognition model.

[0031] The beneficial effect of adopting the above further scheme is that, during the training process of the recognition model, a layer-by-layer training method is adopted, and the average value of the overall classification accuracy of the classification module in any layer is used as the judgment factor for the end of model training, thereby improving the model training accuracy.

[0032] Furthermore, in the process of training the densely connected random forest classification model layer by layer, the following steps are also included:

[0033] Obtaining the average classification probability of all hyperspectral sample images corresponding to multiple random forest classifiers in the classification module of each layer;

[0034] For each layer of the classification module, the initial weight corresponding to the classification module of the layer is updated according to the average classification probability corresponding to the classification module of the layer, so as to obtain the target weight corresponding to the classification module of the layer;

[0035] The identification result of the minerals on the surface of Mars is obtained based on the pixel features extracted layer by layer, including:

[0036] Determining target features based on pixel features extracted layer by layer and target weights corresponding to the classification module of the last layer in the recognition model;

[0037] Based on the target features, identification results of the minerals on the Martian surface are obtained.

[0038] The beneficial effect of adopting the above further solution is that the accuracy of the recognition result can be improved based on the determination of the weights of the classification modules at each layer.

[0039] Furthermore, the initial weight corresponding to the classification module of the first layer is 1, and the initial weight of the classification module of each layer after the classification module of the first layer is the target weight of the classification module of the previous layer.

[0040] The beneficial effect of adopting the above further solution is that the target weight of the previous layer of the current layer is used as the initial weight of the current layer, which can fully consider the feature correlation between the classification modules of each layer and improve the accuracy of weight determination.

[0041] In a second aspect, in order to solve the above technical problems, the present invention further provides a hyperspectral imaging device for identifying minerals on the surface of Mars, the device comprising:

[0042] An acquisition module, used to obtain hyperspectral images of minerals on the Martian surface;

[0043] The recognition module is used to input the hyperspectral image into a pre-trained recognition model, extract the pixel features of the hyperspectral image layer by layer through the classification modules of each layer in the recognition model, and obtain the recognition results of the minerals on the surface of Mars based on the pixel features extracted layer by layer.

[0044] In the third aspect, in order to solve the above-mentioned technical problems, the present invention also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, the hyperspectral image Martian surface mineral identification method of the present application is implemented.

[0045] In a fourth aspect, in order to solve the above-mentioned technical problems, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the hyperspectral image Martian surface mineral identification method of the present application is implemented.

[0046] Additional aspects and advantages of the present application will be given in part in the following description, which will become apparent from the following description, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments of the present invention.

[0048] Figure 1 A schematic flow chart of a method for identifying minerals on the surface of Mars using hyperspectral images provided by one embodiment of the present invention;

[0049] Figure 2 A schematic flow chart of another method for identifying minerals on the surface of Mars using hyperspectral images provided by one embodiment of the present invention;

[0050] Figure 3 A mineral mapping of magnesium carbonate, high-calcium pyroxene and fayalite provided in one embodiment of the present invention;

[0051] Figure 4 A comparison chart of the spectral curves of magnesium carbonate and high-calcium pyroxene identified by a model provided in one embodiment of the present invention and the spectral curves of the MICA spectral library;

[0052] Figure 5 A schematic diagram of the structure of a hyperspectral imaging device for identifying minerals on the surface of Mars, provided by one embodiment of the present invention;

[0053] Figure 6 The present invention provides a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The principles and features of the present invention are described below. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0055] The following describes in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems using specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following embodiments of the present invention are described in conjunction with the accompanying drawings.

[0056] The solution provided by the embodiment of the present invention can be applied to any application scenario that requires the identification of Martian surface minerals based on hyperspectral imagery. The solution provided by the embodiment of the present invention can be executed by any electronic device, for example, an electronic device in space or an electronic device on land.

[0057] The embodiment of the present invention provides a possible implementation method, such as Figure 1 As shown in FIG, a flowchart of a method for identifying minerals on the surface of Mars using hyperspectral images is provided. This solution can be executed by any electronic device. For the convenience of description, the method provided by the embodiment of the present invention will be described below using a terrestrial electronic device as an example of the execution subject. Figure 1 As shown in the flowchart, the method may include the following steps:

[0058] Step S10, obtaining a hyperspectral image of minerals on the surface of Mars;

[0059] In step S20, the hyperspectral image is input into a pre-trained recognition model, pixel features of the hyperspectral image are extracted layer by layer through the classification modules of each layer in the recognition model, and the identification results of the minerals on the surface of Mars are obtained based on the pixel features extracted layer by layer.

[0060] Through the method of the present invention, the pixel point features of the hyperspectral image are extracted layer by layer through each layer classification module, which can fully consider the nonlinear relationship in the hyperspectral data and fully extract the deep spectral characteristics of the Martian surface minerals from the hyperspectral image, thereby making the identification results of the minerals on the Martian surface more accurate.

[0061] The solution of the present invention is further described below with reference to the following specific embodiments. In this embodiment, the method for identifying minerals on the surface of Mars using hyperspectral images may include the following steps:

[0062] Step S10, obtaining a hyperspectral image of minerals on the surface of Mars;

[0063] The hyperspectral image may be a CRISM image, which can be obtained based on existing technology.

[0064] After acquiring the hyperspectral image, and before inputting the hyperspectral image into the pre-trained recognition model, the method further includes:

[0065] S01, performing spectral continuum removal on the hyperspectral image to obtain an image spectrum;

[0066] S02, performing spectral standardization processing on the image spectrum to obtain a standardized image spectrum.

[0067] Optionally, one possible implementation of S01 is:

[0068] The spectral continuum is removed from the hyperspectral image using a first formula to obtain an image spectrum, wherein the first formula is:

[0069]

[0070] Among them, R cr is the image spectrum, R is the absorption spectrum of the hyperspectral image, and R c is the continuum of the absorption spectrum.

[0071] Continuum Removal (also known as Continuum Removal) is a spectral analysis method used to separate spectral absorption features. It is widely used in hyperspectral remote sensing to remove background effects and isolate the absorption characteristics of specific substances. The continuum is defined as the linear connection between reflection peaks in a reflectance spectrum curve, where the outer angle of the connecting line at the peak of the reflectance spectrum is greater than 180 degrees. Spectral continuum removal is the division of the reflectance spectrum by the continuum spectrum. After continuum removal, the reflectance at the endpoints is 1, and the reflectance between the endpoints is less than 1.

[0072] Optionally, before S01, the method may further include:

[0073] Use Gaussian filter to denoise the hyperspectral image:

[0074]

[0075] Among them, I' xy is the pixel value after denoising the coordinate (x, y), I (x+u)(y+v) is the value of the corresponding pixel in the original image (hyperspectral image), G uv are the weights of the Gaussian filter, W is the sum of the weights, and a and b define the size of the filter.

[0076] Optionally, one possible implementation of S02 is:

[0077] According to the reference spectrum, the image spectrum is spectrally normalized using a second formula to obtain a standardized image spectrum, wherein the second formula is:

[0078]

[0079] Where k is the normalization coefficient determined based on the reference spectrum, is the normalized image spectrum, R cr is the image spectrum before normalization, that is, R in formula (1) cr After standardization, the standardized image spectrum can be input into a pre-trained recognition model to extract pixel features of the hyperspectral image layer by layer through the classification modules of each layer in the recognition model, and obtain the identification results of the minerals on the surface of Mars based on the pixel features extracted layer by layer.

[0080] Optionally, the reference spectrum may also be processed in S02 after the spectral continuum is removed.

[0081] Wherein, k can be determined based on the following third formula:

[0082]

[0083] in, is the minimum value of the reference spectrum after the spectral continuum is removed, R cr,min It is the value of the image spectrum after the spectral continuum is removed in the corresponding band of the image spectrum.

[0084] In step S20, the hyperspectral image is input into a pre-trained recognition model, pixel features of the hyperspectral image are extracted layer by layer through the classification modules of each layer in the recognition model, and the identification results of the minerals on the surface of Mars are obtained based on the pixel features extracted layer by layer.

[0085] Among them, the hyperspectral image is first subjected to pixel-by-pixel feature extraction through the first-layer classification module. The output of the first-layer classification module is used as the input of the second-layer classification module, which then performs pixel-by-pixel feature extraction. And so on. The feature extraction method performed in this way can be called layer-by-layer extraction.

[0086] The pixel features extracted layer by layer refer to the features extracted by the last layer of classification module. Pixel features refer to the features extracted according to the pixels.

[0087] In actual application, in order to make the recognition result more accurate, multiple frames of hyperspectral images can be continuously acquired and input into the recognition model. The recognition result can be the recognition result after the recognition results corresponding to the multiple frames of hyperspectral images are fused.

[0088] Optionally, the identification result may be the category of the mineral.

[0089] Optionally, the identified minerals can be annotated in the hyperspectral image.

[0090] Optionally, mineral anomalies can also be extracted based on the recognition model.

[0091] Optionally, the above recognition model is trained based on the following method:

[0092] S1, obtaining a CRISM hyperspectral image dataset, wherein the CRISM hyperspectral image dataset includes multiple frames of hyperspectral sample images;

[0093] S2, performing spectral continuum removal and spectral normalization processing on each frame of the hyperspectral sample image in the CRISM hyperspectral image dataset to obtain a training sample; wherein the processing process of performing spectral continuum removal and spectral normalization processing has been described above and will not be repeated here;

[0094] S3, constructing a densely connected random forest classification model, wherein the densely connected random forest classification model includes multiple layers of classification modules, and each layer of classification modules includes random forest classifiers equal to the total number of hyperspectral sample images in the training sample;

[0095] S4: Training the densely connected random forest classification model layer by layer based on the training samples to obtain an average overall classification accuracy of multiple random forest classifiers in each layer of classification modules. When the average overall classification accuracy of the classification modules in any layer is less than or equal to the average overall classification accuracy of the classification modules in the previous layer, training is stopped and the layer is used as the last layer of the recognition model. The layer has a value greater than 1.

[0096] Among them, one layer of classification module corresponds to multiple random forest classifiers, and one random forest classifier corresponds to a classification accuracy.

[0097] In this application, layer-by-layer training can be expressed by an iterative formula. Assume that L i represents the i-th layer classification module of the model, D i represents the training sample corresponding to the i-th layer, and the training process can be expressed as:

[0098] L i+1 =Train(L i ,D i ) (5)

[0099] Model update can be achieved by introducing learning rate α and error feedback E to adjust parameters:

[0100]

[0101] Among them, P new and P old Respectively represent the values ​​before and after the parameter update.

[0102] For a layer of classification modules, the average overall classification accuracy corresponding to the classification modules of this layer can be determined based on the following fourth formula:

[0103]

[0104] in, Represents the average overall classification accuracy corresponding to a layer of classification modules, T represents the number of random forest classifiers corresponding to the classification modules of this layer, OA t Indicates the overall classification accuracy obtained by the tth random forest classifier.

[0105] Among them, for multiple hyperspectral sample images in the training sample, the overall classification accuracy OA is obtained by a random forest classifier. t ,but The overall classification accuracy OA corresponding to T random forest classifiers can be t The average value of .

[0106] Optionally, in the process of training the densely connected random forest classification model layer by layer, the following steps are further included:

[0107] S5, obtaining the average classification probability of all hyperspectral sample images corresponding to multiple random forest classifiers in the classification module of each layer;

[0108] S6, for each layer of the classification modules, updating the initial weight corresponding to the classification modules of the layer according to the average classification probability corresponding to the classification modules of the layer, to obtain the target weight corresponding to the classification modules of the layer;

[0109] In S20, the identification result of the minerals on the surface of Mars is obtained based on the pixel features extracted layer by layer, including:

[0110] Determining target features based on pixel features extracted layer by layer and target weights corresponding to the classification module of the last layer in the recognition model;

[0111] Based on the target features, identification results of the minerals on the Martian surface are obtained.

[0112] Optionally, S5, a possible implementation method for obtaining the average classification probability of all hyperspectral sample images corresponding to multiple random forest classifiers in the classification module of each layer is:

[0113] For a layer of classification modules, the classification probability of a hyperspectral sample image corresponding to multiple random forest classifiers corresponding to the classification module of this layer is calculated. According to the classification probability of each hyperspectral sample image corresponding to multiple random forest classifiers corresponding to the classification module of this layer, the average classification probability of all hyperspectral sample images corresponding to multiple random forest classifiers corresponding to the classification module of this layer is calculated.

[0114] Among them, for a layer of classification modules, calculating the classification probability of a hyperspectral sample image corresponding to multiple random forest classifiers corresponding to the classification modules of this layer can be achieved based on the following fifth formula:

[0115]

[0116] in, represents the average classification probability of all hyperspectral sample images corresponding to multiple random forest classifiers corresponding to the classification module of a layer, T represents the number of random forest classifiers corresponding to the classification module of this layer, G q,t Represents the classification probability of each hyperspectral sample image corresponding to the tth random forest classifier.

[0117] Optionally, one possible implementation of S6 is:

[0118] For each layer of the classification module, the initial weight corresponding to the classification module of the layer is updated according to the average classification probability corresponding to the classification module of the layer by the sixth formula to obtain the target weight corresponding to the classification module of the layer, wherein the sixth formula is:

[0119] When W q(i-1) =1(9)

[0120] Among them, W qi Represents the target weight corresponding to the i-th layer classification module, W q(i-1) represents the initial weight corresponding to the i-1th layer classification module, represents the average classification probability corresponding to the i-th layer classification module, where the initial weight corresponding to the first layer classification module is 1, and the initial weight of each layer classification module after the first layer classification module is the target weight of the previous layer classification module. That is, when i = 1, W q(i-1) =1.

[0121] It can be understood that for a layer of classification modules, the target weight corresponding to the classification module can be used as the weight corresponding to each hyperspectral sample image in the classification module of the layer. Then, in the same classification module, the weight corresponding to each hyperspectral sample image is the same.

[0122] In order to better illustrate and understand the principle of the method provided by the present invention, the solution of the present invention is described below in conjunction with an optional specific embodiment. It should be noted that the specific implementation of each step in this specific embodiment should not be understood as limiting the solution of the present invention. On the basis of the principle of the solution provided by the present invention, other implementations that can be thought of by those skilled in the art should also be considered as within the scope of protection of the present invention.

[0123] In this embodiment, see Figure 2 As shown, this scheme is described in detail:

[0124] See attached Figure 3 ,In order to better illustrate the method and steps of the ,presentation, a mineral mapping experiment was conducted using ,CRISM hyperspectral data as an example.

[0125] (1) The equipment used in the experiment is a graphics workstation with Windows 10 (64-bit) operating system, a CPU of 2.66 GHz, a memory of 32 GB, and a hard disk of 1 TB.

[0126] (2) The specific steps are as follows:

[0127] Step 1: Import the US CRISM hyperspectral remote sensing data and reference spectral data. CRISM can measure visible and infrared wavelengths between 370 and 3920 nanometers, with each band separated by 6.55 nanometers. The reference spectral data is taken from the MICA mineral spectral library.

[0128] Step 2: Remove the spectrum continuum from the spectrum data obtained in step 1

[0129] Continuum Removal is a spectral analysis method used to separate spectral absorption features. It is widely used in hyperspectral remote sensing to remove background effects and separate the absorption characteristics of certain specific substances. The continuum is defined as the linear connection between reflection peaks in the reflectance spectrum curve. The outer angle of the connecting line at the peak of the reflectance spectrum is greater than 180 degrees. Spectral continuum removal is to divide the reflectance spectrum by the continuum spectrum, as shown in the following formula:

[0130]

[0131] Among them, R cr is the absorption spectrum (image spectrum) after the spectral continuum is removed, R is the absorption spectrum of the hyperspectral image, and R c is the continuum of the absorption spectrum; after continuum removal, the reflectivity at the endpoint is 1, and the reflectivity between the endpoints is less than 1. Both the image spectrum and the reference spectrum need to be continuum removed;

[0132] Step three, spectral standardization, i.e. performing spectral standardization processing on the image spectrum to obtain a standardized image spectrum;

[0133] The image spectrum is standardized using the reference spectral standard to suppress the impact of differences in test conditions on the spectral similarity calculation. The formula for calculating the standardization coefficient is as follows:

[0134]

[0135] Where k is the normalization coefficient, is the minimum value of the reference spectrum after the spectral continuum is removed, R cr,min is the value of the image spectrum after the spectral continuum is removed at the corresponding band of the image spectrum; the image spectrum standardization calculation formula is as follows:

[0136]

[0137] Where k is the normalization coefficient, is the normalized image spectrum, R cr is the image spectrum before normalization;

[0138] Step 4: Build a densely connected random forest classification model

[0139] 4.1, the initial weight parameter of each sample in the balanced hyperspectral dataset S′ (training sample) is set to 1, that is, the initial weight corresponding to the first layer classification module is 1;

[0140] 4.2, the set S′ is input into the T random forest classifiers of the first-layer classification module for training, and the overall classification accuracy OA of the set S′ output by each random forest classifier is obtained t , it can be understood that if there are multiple samples in the set S′, then OA t It represents the overall classification accuracy of all samples output by a random forest classifier, which is different from the meaning expressed in the previous formula (7), as well as the classification probability G of each sample (hyperspectral sample image) q,t , and calculate the overall classification accuracy OA output by T random forest classifiers t Average value

[0141]

[0142] And the classification probability G of each sample output by T random forest classifiers q Average value

[0143]

[0144] Bundle Merge into set S′ to obtain set S”;

[0145]

[0146] 4.3, calculate the sample weight W of each sample in the set S' q , that is, the target weight of the first-layer classification module, is calculated as follows:

[0147]

[0148] 4.4. Then, the set S'' is input into the T random forest classifiers of the second-layer classification module for training, and the average overall classification accuracy of the second-layer classification module and the average classification probability of all samples are obtained. The average classification probability of all samples of the second-layer classification module is merged into the set S'' to obtain the set S''. The target weight of each sample in the set S'' is used as the initial weight parameter of each sample in the second-layer classification module. This process is repeated until the average overall classification accuracy of the current layer classification module is less than or equal to the previous layer. The training is then stopped, and the current layer is used as the last layer. The category of each sample output by the last layer classification module is used as the classification result of the weighted deep random forest classification model for the original unbalanced hyperspectral dataset. That is, the category of each sample output by the last layer classification module is used as the classification result of the recognition model for various different samples.

[0149] This paper proposes a method for identifying minerals on the surface of Mars based on a densely connected random forest model. In the face of massive Mars CRISM hyperspectral images, this method is an automated mineral identification method that can quickly and efficiently identify magnesium carbonate and pyroxene on the surface of Mars (see Figure 3 The recognition results, Figure 3 Different substances can be displayed by different grayscale); the model fully considers the hidden nonlinear characteristics in CRISM high-dimensional data that are not taken into account by the traditional spectral summary parameter method. By learning the dependencies of spectral sequences and focusing on the key information of spectral data, it solves the shortcomings of low efficiency and inaccuracy in mineral identification based on the traditional spectral summary parameter method.

[0150] This example uses the Earth Science Node from the Planetary Data System. The image is full-resolution FRT000093BE (93BE), and the image band range is 1.05 to 2.6 μm ( Figure 4 Horizontal axis), contains a total of 235 bands. Figure 4 As can be seen from the left figure, the spectral characteristic curves of the selected representative magnesium carbonate and pyroxene all show typical diagnostic absorption at 1.9μm, 2.3μm, and 2.5μm, which are consistent with the average spectrum corresponding to the 93BE image and the MICA spectral library ( Figure 4 The spectral characteristics of magnesium carbonate and pyroxene in the right side are consistent well.

[0151] Based on Figure 1 Based on the same principle as the method shown in , the embodiment of the present invention also provides a hyperspectral imaging Mars surface mineral identification device 20, such as Figure 5 As shown in , the hyperspectral image Mars surface mineral identification device 20 may include an acquisition module 210 and an identification module 220, wherein:

[0152] An acquisition module 210 is used to acquire hyperspectral images of minerals on the surface of Mars;

[0153] The recognition module 220 is used to input the hyperspectral image into a pre-trained recognition model, extract the pixel features of the hyperspectral image layer by layer through the classification modules of each layer in the recognition model, and obtain the recognition results of the minerals on the surface of Mars based on the pixel features extracted layer by layer.

[0154] Optionally, before inputting the hyperspectral image into the pre-trained recognition model, the device further includes:

[0155] The preprocessing module is used to remove the spectral continuum of the hyperspectral image to obtain an image spectrum; and perform spectral standardization on the image spectrum to obtain a standardized image spectrum.

[0156] Optionally, when the preprocessing module removes the spectral continuum of the hyperspectral image to obtain the image spectrum, it is specifically used to:

[0157] The spectral continuum is removed from the hyperspectral image using a first formula to obtain an image spectrum, wherein the first formula is:

[0158]

[0159] Among them, R cr is the image spectrum, R is the absorption spectrum of the hyperspectral image, and R c is the continuum of the absorption spectrum.

[0160] Optionally, when the preprocessing module performs spectral standardization on the image spectrum to obtain a standardized image spectrum, it is specifically used to:

[0161] According to the reference spectrum, the image spectrum is spectrally normalized using a second formula to obtain a standardized image spectrum, wherein the second formula is:

[0162]

[0163] Where k is the normalization coefficient determined based on the reference spectrum, is the normalized image spectrum, R cr is the image spectrum before normalization.

[0164] Optionally, the recognition model is trained based on the following method:

[0165] Acquire a CRISM hyperspectral image dataset, wherein the CRISM hyperspectral image dataset includes multiple frames of hyperspectral sample images;

[0166] Performing spectral continuum removal and spectral normalization processing on each frame of the hyperspectral sample image in the CRISM hyperspectral image dataset to obtain a training sample;

[0167] Constructing a densely connected random forest classification model, wherein the densely connected random forest classification model includes multiple layers of classification modules, and each layer of classification modules includes random forest classifiers equal to the total number of hyperspectral sample images in the training sample;

[0168] According to the training samples, the densely connected random forest classification model is trained layer by layer to obtain the overall classification accuracy average of multiple random forest classifiers in each layer of classification modules. When the overall classification accuracy average of the classification modules in any layer is less than or equal to the overall classification accuracy average of the classification modules in the previous layer, the training is stopped and the any layer is used as the last layer of the recognition model.

[0169] Optionally, in the process of training the densely connected random forest classification model layer by layer, the following steps are further included:

[0170] A target weight determination module is used to obtain the average classification probability of all hyperspectral sample images corresponding to multiple random forest classifiers in each layer of the classification module; for each layer of the classification module, the initial weight corresponding to the classification module of the layer is updated according to the average classification probability corresponding to the classification module of the layer, and the target weight corresponding to the classification module of the layer is obtained;

[0171] When the identification module 220 obtains the identification result of the minerals on the surface of Mars based on the pixel features extracted layer by layer, it is specifically used to:

[0172] Determining target features based on pixel features extracted layer by layer and target weights corresponding to the classification module of the last layer in the recognition model;

[0173] Based on the target features, identification results of the minerals on the Martian surface are obtained.

[0174] Optionally, the initial weight corresponding to the classification module of the first layer is 1, and the initial weight of the classification module of each layer after the classification module of the first layer is the target weight of the classification module of the previous layer.

[0175] The hyperspectral imaging Mars surface mineral identification device of an embodiment of the present invention can execute the hyperspectral imaging Mars surface mineral identification method provided by the embodiment of the present invention, and its implementation principle is similar. The actions performed by each module and unit in the hyperspectral imaging Mars surface mineral identification device in each embodiment of the present invention correspond to the steps in the hyperspectral imaging Mars surface mineral identification method in each embodiment of the present invention. For the detailed functional description of each module of the hyperspectral imaging Mars surface mineral identification device, please refer to the description of the corresponding hyperspectral imaging Mars surface mineral identification method shown in the previous text, which will not be repeated here.

[0176] Among them, the above-mentioned hyperspectral imaging Mars surface mineral identification device can be a computer program (including program code) running in a computer device, for example, the hyperspectral imaging Mars surface mineral identification device is an application software; the device can be used to execute the corresponding steps in the method provided in the embodiment of the present invention.

[0177] In some embodiments, the hyperspectral image Martian surface mineral identification device provided by an embodiment of the present invention can be implemented by a combination of software and hardware. As an example, the hyperspectral image Martian surface mineral identification device provided by an embodiment of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the hyperspectral image Martian surface mineral identification method provided by an embodiment of the present invention. For example, the processor in the form of a hardware decoding processor can adopt one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs) or other electronic components.

[0178] In other embodiments, the hyperspectral image Martian surface mineral identification device provided by the embodiments of the present invention can be implemented in a software manner. Figure 5 A hyperspectral image Martian surface mineral identification device stored in a memory is shown, which can be software in the form of a program and a plug-in, and includes a series of modules, including an acquisition module 210 and an identification module 220, for implementing the hyperspectral image Martian surface mineral identification method provided by an embodiment of the present invention.

[0179] The modules involved in the embodiments of the present invention may be implemented in software or hardware, wherein the name of a module does not necessarily limit the module itself.

[0180] Based on the same principle as the method shown in the embodiments of the present invention, an electronic device is also provided in the embodiments of the present invention, which may include but is not limited to: a processor and a memory; the memory is used to store computer programs; the processor is used to execute the method shown in any embodiment of the present invention by calling the computer program.

[0181] In an alternative embodiment, an electronic device is provided, such as Figure 6 As shown, Figure 6 The electronic device 4000 shown includes: a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data exchange between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, the number of transceivers 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0182] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the present disclosure. Processor 4001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0183] Bus 4002 may include a path for transmitting information between the above components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. Bus 4002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0184] The memory 4003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0185] The memory 4003 is used to store application code (computer program) for executing the solution of the present invention, and is controlled by the processor 4001. The processor 4001 is used to execute the application code stored in the memory 4003 to implement the content shown in the above method embodiment.

[0186] Among them, the electronic device can also be a terminal device, Figure 6 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0187] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer-readable storage medium is run on a computer, the computer can execute the corresponding contents of the aforementioned method embodiment.

[0188] According to another aspect of the present invention, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various implementations described above.

[0189] Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0190] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0191] The computer-readable storage medium provided by the embodiments of the present invention may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. 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 thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.

[0192] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device executes the method shown in the above embodiment.

[0193] The above description is merely a preferred embodiment of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present invention is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in the present invention.

Claims

1. A method for identifying minerals on the surface of Mars using hyperspectral images, characterized in that: include: Obtain hyperspectral images of minerals on the Martian surface; The hyperspectral image is input into a pre-trained recognition model, pixel features of the hyperspectral image are extracted layer by layer through each layer classification module in the recognition model, and recognition results of the minerals on the surface of Mars are obtained based on the pixel features extracted layer by layer.

2. The method according to claim 1, characterized in that Before inputting the hyperspectral image into the pre-trained recognition model, the method further includes: Performing spectral continuum removal on the hyperspectral image to obtain an image spectrum; The image spectrum is subjected to spectral standardization processing to obtain a standardized image spectrum.

3. The method according to claim 2, characterized in that The removing of the spectral continuum from the hyperspectral image to obtain the image spectrum comprises: The spectral continuum is removed from the hyperspectral image by a first formula to obtain an image spectrum, wherein the first formula is: Among them, R cr is the image spectrum, R is the absorption spectrum of the hyperspectral image, and R c is the continuum of the absorption spectrum.

4. The method according to claim 2, characterized in that: The performing spectral standardization processing on the image spectrum to obtain a standardized image spectrum includes: According to the reference spectrum, the image spectrum is spectrally standardized by a second formula to obtain a standardized image spectrum, wherein the second formula is: Where k is the normalization coefficient determined based on the reference spectrum, is the standardized image spectrum, R cr is the image spectrum before normalization.

5. The method according to any one of claims 1 to 4, characterized in that The recognition model is trained based on the following method: Acquire a CRISM hyperspectral image dataset, wherein the CRISM hyperspectral image dataset includes multiple frames of hyperspectral sample images; Performing spectral continuum removal and spectral standardization processing on each frame of the hyperspectral sample image in the CRISM hyperspectral image dataset to obtain a training sample; Constructing a densely connected random forest classification model, wherein the densely connected random forest classification model comprises multiple layers of classification modules, and each layer of classification modules comprises random forest classifiers equal to the total number of hyperspectral sample images in the training samples; According to the training samples, the densely connected random forest classification model is trained layer by layer to obtain the overall classification accuracy average of multiple random forest classifiers in each layer of classification modules. When the overall classification accuracy average of the classification modules in any layer is less than or equal to the overall classification accuracy average of the classification modules in the previous layer, the training is stopped and the any layer is used as the last layer of the recognition model.

6. The method according to claim 5, characterized in that The process of training the densely connected random forest classification model layer by layer also includes: Obtaining the average classification probability of all hyperspectral sample images corresponding to multiple random forest classifiers in the classification module of each layer; For each layer of the classification modules, the initial weight corresponding to the classification modules of the layer is updated according to the average classification probability corresponding to the classification modules of the layer, so as to obtain the target weight corresponding to the classification modules of the layer; The identification result of the minerals on the surface of Mars is obtained based on the pixel features extracted layer by layer, including: Determine the target feature based on the pixel features extracted layer by layer and the target weight corresponding to the classification module of the last layer in the recognition model; Based on the target features, identification results of the minerals on the surface of Mars are obtained.

7. The method according to claim 6, characterized in that The initial weight corresponding to the classification module of the first layer is 1, and the initial weight of each layer of classification modules after the first layer of classification modules is the target weight of the classification module of the previous layer of classification modules of the classification module of this layer.

8. A hyperspectral imaging device for identifying minerals on the surface of Mars, characterized in that: include: An acquisition module, used to acquire hyperspectral images of minerals on the surface of Mars; The recognition module is used to input the hyperspectral image into a pre-trained recognition model, extract pixel features of the hyperspectral image layer by layer through the classification modules of each layer in the recognition model, and obtain the recognition result of the minerals on the surface of Mars based on the pixel features extracted layer by layer.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images