Hyperspectral image Mars surface montmorillonite identification method
By constructing spectral data sets and establishing a two-way training model, the problem of low recognition accuracy of Mars minerals is solved, and efficient and automated processing of hyperspectral image data is achieved, and the recognition accuracy is improved.
Patent Information
- Application Number
- CN202411703275.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-26
- Publication Date
- 2025-02-28
AI Technical Summary
In the prior art, Mars mineral recognition accuracy is low and is sensitive to noise, making it difficult to automatically process hyperspectral image data.
By constructing the spectral data set, performing data preprocessing, establishing a two-way training model, using a combination of forward and reverse training, iterative training based on loss function, obtaining an identification model, and achieving automated recognition of Mars surface montmorillonite.
The accuracy of Mars mineral recognition is improved, and the efficient and automated processing of hyperspectral image data is achieved, taking into account the bidirectional dependence between spectral bands.
Smart Images
Figure CN119540649A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a method for identifying montmorillonite on the surface of Mars using hyperspectral images. Background Art
[0002] The Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) is a hyperspectral imaging spectrometer installed on the Mars Reconnaissance Orbiter (MRO) to collect data on Mars. The data obtained based on CRISM have been widely used to identify the distribution of minerals on the surface of Mars, mainly including carbonates, silicates, sulfates, etc. Among them, iron montmorillonite or magnesium montmorillonite, as hydrated minerals, are not only of great significance for preserving the characteristics of microorganisms and indicating the evolution of the geological environment of Mars, but also can provide guidance for the future utilization of in-situ resources.
[0003] Currently, the identification of Martian surface minerals is done through the spectral summary parameter method based on manual calculations.
[0004] However, the spectral summary parameter method is based on manual calculation, has a low degree of automation, and is very sensitive to noise, so the accuracy of mineral identification needs to be improved. Summary of the invention
[0005] The embodiment of the present application provides a method for identifying montmorillonite on the surface of Mars using hyperspectral images, which can solve the technical problem of low accuracy in identifying Martian minerals.
[0006] To achieve the above objectives, the embodiments of the present application adopt the following technical solutions:
[0007] In the first aspect, an embodiment of the present application provides a method for identifying montmorillonite on the surface of Mars using hyperspectral images. The method includes: constructing a spectral data set; preprocessing the data in the spectral data set to obtain model training data; based on the model training data, constructing a bidirectional training model through forward training and reverse training; based on the loss function, iteratively training the bidirectional training model to obtain a recognition model.
[0008] Based on the above description of the method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in the embodiment of the present application, it can be seen that the method for identifying montmorillonite on the surface of Mars using hyperspectral images includes constructing a spectral data set, in which all valid data of the hyperspectral image are included. A bidirectional training model is constructed to fully consider the bidirectional dependencies between the deep-level mineral spectral features hidden in the hyperspectral image and the spectral bands. The optimal recognition model is obtained by iterative training based on the loss function. In this way, the image to be predicted is input into the recognition model to perform mineral identification, and the hyperspectral image data is efficiently processed in an automated manner. At the same time, the bidirectional dependencies between the spectral bands after dimensionality reduction are taken into account to improve the accuracy of Mars mineral identification.
[0009] In a feasible implementation of the first aspect, when executing the step of constructing a bidirectional training model based on model training data through forward training and reverse training, the bidirectional training model includes a bidirectional long short-term memory network model, and the hyperspectral image Mars surface montmorillonite identification method also includes: based on the model training data, through forward training, learning the first spectral feature sequence relationship; based on the model training data, through reverse training, learning the second spectral feature sequence relationship; integrating the first spectral feature sequence relationship and the second spectral feature sequence relationship.
[0010] In a feasible implementation of the first aspect, the bidirectional long short-term memory network model includes a first long short-term memory layer, a second long short-term memory layer and a fully connected layer.
[0011] In a feasible implementation of the first aspect, the first long short-term memory layer includes a plurality of long short-term memory units; and the calculation formula of the long short-term memory unit includes:
[0012]
[0013] Among them, y (t) Represented as the tth dimension of y, which is the input at the current moment; W f Represented as the forget gate weight; b f Expressed as bias; C (t-1) Represents the unit state at the previous moment; h (t-1) Represents the output value of the previous moment; y (t) represents the input value of the network at the previous moment; h represents the output of the first long short-term memory layer; h (t) Represented as the output value of the first long short-term memory layer at the current moment; C (t) Represents the unit state at the current moment; f (t) Represented as the output of the forget gate; Represented as input gate output; O (t) Represented as the output gate output.
[0014] In a feasible implementation of the first aspect, the calculation formula of the fully connected layer includes:
[0015] a=σ(f(h,w)+b);
[0016] Among them, h represents the output of the second long short-term memory layer; f represents the fully connected layer; w represents the weight of the fully connected layer; b represents the bias of the fully connected layer; σ represents the activation function; a represents the output of the fully connected layer, that is, the current prediction result.
[0017] In a feasible implementation of the first aspect, the calculation formula of the loss function includes:
[0018]
[0019] Among them, Loss is represented by the cross entropy loss function; c is represented by the category label; and n is represented by the number of categories.
[0020] In a feasible implementation of the first aspect, when executing the step of preprocessing the data in the spectral dataset to obtain model training data, the hyperspectral image Mars surface montmorillonite identification method also includes: wavelength consistency processing, spectral smoothing, continuum removal and spectral standardization of the data in the spectral dataset.
[0021] In a feasible implementation of the first aspect, the method for identifying montmorillonite on the surface of Mars using hyperspectral images also includes: preprocessing the image to be predicted to obtain data to be input; and inputting the data to be input into the recognition model.
[0022] In a second aspect, an embodiment of the present application provides a hyperspectral image Mars surface montmorillonite identification system, the hyperspectral image Mars surface montmorillonite identification system comprising: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method provided in the first aspect.
[0023] The hyperspectral image Mars surface montmorillonite recognition system constructs a spectral data set by executing the method provided in the first aspect, and the valid data of the hyperspectral image are all in the spectral data set. A bidirectional training model is constructed to fully consider the deep-level mineral spectral features hidden in the hyperspectral image and the bidirectional dependency between the spectral bands. Iterative training based on the loss function is used to obtain the best recognition model. In this way, the image to be predicted is input into the recognition model for mineral recognition, and the hyperspectral image data is efficiently processed in an automated manner. At the same time, the bidirectional dependency between the spectral bands after dimensionality reduction is considered to improve the accuracy of Mars mineral recognition.
[0024] In a third aspect, an embodiment of the present application provides a computer-readable medium having computer program instructions stored thereon, and the computer program instructions can be executed by a processor to implement the method provided in the first aspect.
[0025] The computer program instructions in the computer-readable medium construct a spectral data set by implementing the method provided in the first aspect, and the valid data of the hyperspectral image are all in the spectral data set. A bidirectional training model is constructed to fully consider the bidirectional dependencies between the deep-level mineral spectral features hidden in the hyperspectral image and the spectral bands. The optimal recognition model is obtained by iterative training based on the loss function. In this way, the image to be predicted is input into the recognition model for mineral recognition, and the hyperspectral image data is efficiently processed in an automated manner. At the same time, the bidirectional dependencies between the spectral bands after dimensionality reduction are considered to improve the accuracy of Mars mineral recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 A schematic diagram of the structure of a hyperspectral imagery system for identifying montmorillonite on the surface of Mars provided in an embodiment of the present application;
[0027] Figure 2 A schematic diagram of a process for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application;
[0028] Figure 3 A schematic diagram of a process for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application;
[0029] Figure 4 A schematic diagram of the structure of a bidirectional long short-term memory network model in a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application;
[0030] Figure 5a A schematic diagram of the structure of a bidirectional long short-term memory network model in a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application;
[0031] Figure 5b A schematic diagram of the structure of a multiple bidirectional long short-term memory network model in a method for identifying montmorillonite on the surface of Mars in a hyperspectral image provided in an embodiment of the present application;
[0032] Figure 5c A schematic diagram of the structure of a long short-term memory network based on a convolutional neural network combined with an attention mechanism in a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application;
[0033] Figure 6 A spatial distribution map of iron montmorillonite or magnesium montmorillonite in a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application;
[0034] Figure 7 A spectral curve with representative spectral characteristics corresponding to iron montmorillonite or magnesium montmorillonite in a method for identifying montmorillonite on the surface of Mars in a hyperspectral image provided in an embodiment of the present application. DETAILED DESCRIPTION
[0035] The technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention. In the description of the embodiments of the present invention, unless otherwise specified, "multiple" refers to two or more than two. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.
[0036] In addition, in order to facilitate the clear description of the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, the words "first", "second" and the like are used to distinguish the same items or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first", "second" and the like do not limit the quantity and execution order, and the words "first", "second" and the like do not necessarily limit the difference. At the same time, in the embodiments of the present invention, the words "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or design. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way for easy understanding.
[0037] The principles and features of the present application are described below. The examples given are only used to explain the present application and are not used to limit the scope of the present application.
[0038] Montmorillonite minerals are a class of clay minerals that are widely used in industrial and environmental fields. Montmorillonite can be divided into different types according to the dominant metal cations in its chemical composition, such as sodium montmorillonite, calcium montmorillonite, and iron or magnesium montmorillonite. The basic structure of montmorillonite minerals consists of two tetrahedral silicon oxygen layers sandwiched by an octahedral aluminum oxygen layer (2:1 type layered structure). In iron montmorillonite or magnesium montmorillonite, part or all of the aluminum in the octahedral layer is replaced by iron or magnesium, and this substitution affects its physical and chemical properties.
[0039] Ferruginous Montmorillonite contains iron ions (Fe 3+ or Fe 2 + ), which gives it unique adsorption properties and chemical reactivity.
[0040] Magnesium Montmorillonite, which contains magnesium ions (Mg2+ ), which gives it a higher cation exchange capacity and better swelling properties.
[0041] By taking hyperspectral images of Mars, we can identify iron montmorillonite or magnesium montmorillonite on Mars, thereby achieving the goals of preserving microbial characteristics, indicating the evolution of the Martian geological environment, and guiding future in situ resource utilization.
[0042] At present, the acquisition of important information from hyperspectral images is often carried out by feature extraction, that is, using dimensionality reduction methods to retain as much effective information or features as possible. General dimensionality reduction methods cannot fully consider the nonlinear relationship in hyperspectral data, and conventional machine learning does not well consider the bidirectional dependency between spectral bands after dimensionality reduction, which makes it difficult to fully extract the deep spectral characteristics of Martian surface minerals from CRISM hyperspectral images.
[0043] The embodiment of the present application provides a method for identifying montmorillonite on the surface of Mars using hyperspectral images, which fully considers the nonlinear relationship in hyperspectral data and the bidirectional dependency that actually exists between spectral bands, thereby improving the accuracy of identifying Martian minerals.
[0044] The embodiment of the present application provides a hyperspectral image Mars surface montmorillonite identification system, which can execute the hyperspectral image Mars surface montmorillonite identification method provided by the embodiment of the present application. Figure 1 A schematic diagram of the structure of a hyperspectral imagery system for identifying montmorillonite on the surface of Mars provided in an embodiment of the present application.
[0045] like Figure 1 As shown, the hyperspectral image Mars surface montmorillonite identification system 001 includes at least one processor 011 and a memory 012 communicatively connected to the at least one processor; wherein the memory 012 stores instructions executable by the at least one processor 011, and the instructions are executed by the at least one processor 011 so that the at least one processor 011 can execute the hyperspectral image Mars surface montmorillonite identification method provided in an embodiment of the present application.
[0046] Figure 2 A schematic diagram of a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application. Figure 2 As shown, in some embodiments, the hyperspectral image Mars surface montmorillonite identification method includes the following steps:
[0047] S1, construct spectral dataset.
[0048] In some embodiments, CRISM is used to collect and capture data. In one implementation, CRISM can capture spectral data at different wavelengths, providing detailed spectral information for each pixel. In one implementation, CRISM can perform hyperspectral imaging in the visible and near-infrared bands (0.36-3.92 microns). Based on the collected images, hyperspectral data are obtained. These hyperspectral data can be used as elements of spectral data sets. In one implementation, CRISM can provide a detailed mineral distribution map.
[0049] Figure 3 A schematic diagram of a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application. Figure 3 As shown, in some embodiments, the CRISM hyperspectral image dataset is divided into a training set and a test set, and the validation set is divided from the training set. In one implementation, the constructed dataset is divided into a training set and a test set at a ratio of 10:1, wherein 25% of the total number of training sets is randomly selected as the validation set.
[0050] S2, preprocessing the data in the spectral dataset to obtain model training data.
[0051] In some embodiments, when executing step S2, the hyperspectral image Mars surface montmorillonite identification method further includes: performing wavelength consistency processing, spectrum smoothing, continuum removal and spectrum standardization on the data in the spectral data set. In one implementation, the preprocessing operations of wavelength consistency processing, spectrum smoothing, continuum removal and spectrum standardization are performed on both the training set and the test set.
[0052] S3, based on the model training data, builds a bidirectional training model through forward training and backward training.
[0053] There are many possible implementations of the bidirectional training model.
[0054] For example, in some embodiments, the bidirectional training model can be a bidirectional long short-term memory network model (Bidirectional Long Short-Term Networks, Bi-LSTM), which significantly enhances the model's ability to understand sequence data by combining forward and backward information flows, enabling it to perform well in various complex time series modeling tasks.
[0055] In some embodiments, when executing step S3, the method for identifying montmorillonite on the surface of Mars using hyperspectral images further includes:
[0056] S31, learning the first spectral feature sequence relationship through forward training based on the model training data.
[0057] The preprocessed data is input into the model for forward training to learn the forward spectral feature sequence relationship as the first spectral feature sequence relationship.
[0058] S32, learning the second spectral feature sequence relationship through reverse training based on the model training data.
[0059] The preprocessed data is input into the model for reverse training to learn the reverse spectral feature sequence relationship as the second spectral feature sequence relationship.
[0060] S33, integrating the first spectral feature sequence relationship and the second spectral feature sequence relationship.
[0061] In some embodiments, the bidirectional long short-term memory network model includes a first long short-term memory layer, a second long short-term memory layer, and a fully connected layer.
[0062] In some embodiments, the first LSTM layer includes a plurality of LSTM units. Figure 4 A schematic diagram of the structure of a bidirectional long short-term memory network model in a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application. Figure 5a A schematic diagram of the structure of a bidirectional long short-term memory network model in a method for identifying montmorillonite on the surface of Mars using hyperspectral images provided in an embodiment of the present application. Figure 4 and Figure 5a As shown, in one implementation, when constructing a bidirectional long short-term memory network model, first construct a long short-term memory (LSTM) unit, and then construct a bidirectional LSTM layer on its basis, that is, a first long short-term memory layer and a second long short-term memory layer. Exemplarily, the number of long short-term memory units in the first long short-term memory layer is set to 16, and the number of long short-term memory units in the second long short-term memory layer is set to 8.
[0063] The calculation formula for long short-term memory units includes:
[0064]
[0065] Among them, y (t) Represented as the tth dimension of y, which is the input at the current moment; W f Represented as the forget gate weight; b f Expressed as bias; C (t-1) Represents the unit state at the previous moment; h (t-1) Represents the output value of the previous moment; y (t) represents the input value of the network at the previous moment; h represents the output of the first long short-term memory layer; h (t) Represented as the output value of the first long short-term memory layer at the current moment; C(t) Represents the unit state at the current moment; f (t) Represented as the output of the forget gate; Represented as input gate output; O (t) Represented as the output gate output.
[0066] In one implementation, when constructing a bidirectional long short-term memory network model, other parameter settings also include but are not limited to selecting a cross-loss entropy function as the direction of model optimization, setting an initial learning rate of 0.0001, a batch size of 240, an iteration number of 5000, and an iteration round of 50.
[0067] In some embodiments, the calculation formula of the fully connected layer includes:
[0068] a=σ(f(h,w)+b);
[0069] Among them, h represents the output of the second long short-term memory layer; f represents the fully connected layer; w represents the weight of the fully connected layer; b represents the bias of the fully connected layer; σ represents the activation function; a represents the output of the fully connected layer, that is, the current prediction result.
[0070] In one implementation, the fully connected layer selects a normalized exponential function (softmax) activation function to implement classification, and the dimension is set to a "1" layer. Softmax is a common activation function in deep learning that can be used for multi-classification tasks. Through the Softmax function, the output values of multiple classifications can be converted into a probability distribution in the range of [0, 1] and the sum is 1. The activation function can be a rectified linear unit (Relu), a common nonlinear activation function in deep learning. The Relu function helps alleviate the gradient vanishing problem because the gradient of this function is 1 in the positive interval, which helps to better propagate the gradient.
[0071] Exemplarily, the parameter settings for constructing the Bi-LSTM model include: the number of long short-term memory units in the first long short-term memory layer is set to L1; the number of long short-term memory units in the second long short-term memory layer is set to L2; the maximum number of iterations is M; the learning rate is α; and the batch size is p.
[0072] For example,
[0073] In some other embodiments, the bidirectional training model can be a multiple bidirectional long short-term memory attention mechanism network (BiLSTM-Attention-BiLSTM). BiLSTM-Attention-BiLSTM includes a bidirectional long short-term memory network (Bidirectional Long Short-Term Networks, BiLSTM) layer, a self-attention layer, and a fully connected layer. Figure 5b As shown in the figure, in one implementation, the LSTM unit in the encoding layer is first constructed, and then the BiLSTM layer is constructed on its basis, and a total of one BiLSTM layer is constructed. Then a multi-head self-attention layer is constructed, and a total of eight self-attention heads are constructed. Finally, a BiLSTM layer is also constructed in the decoding layer. Among them, the number of LSTM units in the encoding layer is set to "16", and the number of LSTM units in the decoding layer is set to "8". The cross loss entropy function is selected as the direction of model optimization, and the initial learning rate is set to "0.0001", the batch size is set to "240", the number of iterations is set to "5000", the iteration round is set to "50", and the optimizer is RAdam. The fully connected layer selects the softmax activation function to implement classification, and the dimension is set to one layer. The preprocessed training set is imported into the BiLSTM-Attention-BiLSTM model for iterative training. The model training process is as follows: the preprocessed data is input into the encoding layer, and forward training and reverse training are performed in sequence to learn the bidirectional spectral feature sequence relationship. The learned bidirectional sequence features are then input into the multi-head self-attention layer, and the key information of the spectral data is focused on through the self-attention mechanism. Finally, the result is input into the decoding layer. 25% of the preprocessed training set is randomly selected as the validation set.
[0074] For example,
[0075] In some other embodiments, the bidirectional training model can be a long short-term memory network (CNN-BiLSTM-Attention) based on a convolutional neural network combined with an attention mechanism. CNN-BiLSTM-Attention includes a convolutional neural network (CNN) layer, a BiLSTM layer, a self-attention layer, and a fully connected layer. The local features of the spectral sequence data are extracted through the convolutional layer, and then the two BiLSTM layers are used as encoders and decoders respectively, and the multi-head self-attention mechanism (Multi-head Self-Attention Machanism) is integrated and used as the intermediate layer. Figure 5cAs shown in the figure, in one implementation, first, a one-dimensional convolution layer, a one-dimensional maximum pooling layer, and a fully connected layer are constructed. Secondly, the LSTM unit in the encoding layer is constructed, and then a BiLSTM layer is constructed on its basis, and a total of BiLSTM layers are constructed. Then, a multi-head self-attention layer is constructed, and a total of 8 self-attention heads are constructed. Finally, a BiLSTM layer is also constructed in the decoding layer. Among them, the number of convolution kernels in the convolution layer is "128", the size of the pooling window in the maximum pooling layer is "2", and the number of neurons in the fully connected layer is "64". The number of LSTM units in the encoding layer is set to "16", and the number of LSTM units in the decoding layer is set to "8". The cross-loss entropy function is selected as the direction of model optimization, and the initial learning rate is set to "0.0001", the batch size is "240", the number of iterations is "5000", the iteration round is "50", and the optimizer is RAdam. The fully connected layer selects the softmax activation function to implement classification, and the dimension is set to "1" layer. The preprocessed training set is imported into the CNN-BiLSTM-Attention model for iterative training. The model training process is as follows: the preprocessed data is first input into the convolution layer to extract the local features of the spectral sequence data, and the captured local features are sequentially forward trained and reverse trained to learn the bidirectional spectral feature sequence relationship. The learned bidirectional sequence features are then input into the multi-head self-attention layer to focus on the key information of the spectral data through the self-attention mechanism, and finally the result is input into the decoding layer. 25% of the preprocessed training set is randomly selected as the validation set.
[0076] S4, based on the loss function, iteratively train the bidirectional training model to obtain the recognition model.
[0077] In some embodiments, the loss function calculation formula includes:
[0078]
[0079] Among them, Loss is represented by the cross entropy loss function; c is represented by the category label; and n is represented by the number of categories.
[0080] like Figure 3 As shown, in some embodiments, when executing step S4, the hyperspectral image Mars surface montmorillonite identification method further includes:
[0081] S41, calculate the loss function.
[0082] S42, use the optimizer to update the bidirectional long short-term memory network model through back propagation.
[0083] In some embodiments, the optimizer may be a rectified adaptive moment estimation (RAdam), which is an optimizer in deep learning that solves the problem that adaptive moment estimation (Adam) easily converges to a local optimal solution.
[0084] S43, determining whether the minimum loss function is reached, and if so, obtaining a recognition model.
[0085] In some embodiments, after executing step S43, the hyperspectral image Mars surface montmorillonite identification method further includes:
[0086] S44, test the performance of the model using the test set.
[0087] Exemplarily, 25% of the training set after dimensionality reduction is randomly selected as the validation set.
[0088] In some embodiments, after executing step S4, the method for identifying montmorillonite on the surface of Mars using hyperspectral images further includes:
[0089] S5, preprocessing the image to be predicted to obtain data to be input.
[0090] In some embodiments, pre-processing includes wavelength consistency processing, spectral smoothing, continuum removal, and spectral normalization.
[0091] S6, inputting the data to be input into the recognition model.
[0092] In some embodiments, the standard spectral library is combined to realize the identification of iron montmorillonite or magnesium montmorillonite on the surface of Mars, and the iron montmorillonite or magnesium montmorillonite is obtained. Figure 6 The spatial distribution diagram shown in Figure 7 The spectral curve diagram with representative spectral features is shown in FIG. Figure 6 The red color shown is iron montmorillonite or magnesium montmorillonite.
[0093] In one implementation, the standard spectral library may be the Mars CRISM Analysis Minerals Identified in CRISM Analysis (MICA) spectral library, which is used for the identification and recognition of Martian surface minerals.
[0094] Figure 7 The spectral curve with representative spectral characteristics corresponding to iron montmorillonite or magnesium montmorillonite in the method for identifying montmorillonite on the surface of Mars in a hyperspectral image provided in the embodiment of the present application. Figure 7As shown, in some embodiments, an image from the Earth Science Node of the Planetary Data System is selected, the image is the full-resolution FRT000093BE (93BE), the image band range is 1.05-2.6 μm, and a total of 235 bands are included. The selected representative spectral characteristic curves of iron montmorillonite or magnesium montmorillonite all show typical diagnostic absorption at 1.9 μm, 2.31 μm, and 2.39 μm, which is in good agreement with the average spectrum of the 93BE image and the spectral characteristics of iron montmorillonite or magnesium montmorillonite in the MICA spectral library.
[0095] The hyperspectral imagery method for identifying montmorillonite on the surface of Mars provided in the embodiment of the present application can solve the disadvantage of information redundancy and reduce interference in the accurate identification of minerals.
[0096] Based on the same application concept, a hyperspectral image Mars surface montmorillonite identification system is also provided in an embodiment of the present application. The method corresponding to the hyperspectral image Mars surface montmorillonite identification system may be the hyperspectral image Mars surface montmorillonite identification method in the aforementioned embodiment, and its principle of solving the problem is similar to that of the method. The hyperspectral image Mars surface montmorillonite identification system provided in an embodiment of the present application includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the methods and / or technical solutions of the aforementioned multiple embodiments of the present application.
[0097] Another embodiment of the present application further provides a computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of the present application described above.
[0098] Specifically, the present embodiment may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The 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 thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: 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 this document, 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.
[0099] Computer readable signal media may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer readable program code. 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. Computer readable signal media 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.
[0100] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
[0101] Computer program code for performing the operations of the present application 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 separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of 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., via the Internet using an Internet service provider).
[0102] The secondary flow chart or block diagram in the accompanying drawings shows the possible architecture, function and operation of the equipment, method and computer program product according to various embodiments of the present application. In this regard, each square box in the secondary flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square 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 square box in the block diagram and / or the secondary flow chart, and the combination of the square boxes in the block diagram and / or the secondary flow chart can be implemented with a dedicated system for hardware that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0103] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0104] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or page components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0105] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0106] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.
[0107] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.
[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
[0109] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in a device claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.
Claims
1. A method for identifying montmorillonite on the surface of Mars using hyperspectral images, characterized in that: include: Construct spectral datasets; Preprocessing the data in the spectral data set to obtain model training data; Based on the model training data, a bidirectional training model is constructed through forward training and reverse training; Based on the loss function, the bidirectional training model is iteratively trained to obtain a recognition model.
2. The method for identifying montmorillonite on the surface of Mars using hyperspectral images according to claim 1, characterized in that: When executing the step of constructing a bidirectional training model based on the model training data through forward training and reverse training, the bidirectional training model includes a bidirectional long short-term memory network model, and the hyperspectral image Mars surface montmorillonite identification method also includes: Based on the model training data, learning the first spectral feature sequence relationship through forward training; Based on the model training data, learning the second spectral feature sequence relationship through reverse training; The first spectral feature sequence relationship and the second spectral feature sequence relationship are integrated.
3. The method for identifying montmorillonite on the surface of Mars using hyperspectral images according to claim 2, characterized in that: The bidirectional long short-term memory network model includes a first long short-term memory layer, a second long short-term memory layer and a fully connected layer.
4. The method for identifying montmorillonite on the surface of Mars using hyperspectral images according to claim 3, characterized in that: The first long short-term memory layer includes a plurality of long short-term memory units; the calculation formula of the long short-term memory unit includes: Among them, y (t) Represented as the tth dimension of y, which is the input at the current moment; W f Represented as the forget gate weight; b f Expressed as bias; C (t-1) Represents the unit state at the previous moment; h (t-1) Represents the output value of the previous moment; y (t) represents the input value of the network at the previous moment; h represents the output of the first long short-term memory layer; h (t) Represented as the output value of the first long short-term memory layer at the current moment; C (t) Represents the unit state at the current moment; f (t) Represented as the output of the forget gate; Represented as input gate output; O (t) Represented as the output gate output.
5. The method for identifying montmorillonite on the surface of Mars using hyperspectral images according to claim 4, characterized in that: The calculation formula of the fully connected layer includes: a=σ(f(h,w)+b); Among them, h represents the output of the second long short-term memory layer; f represents the fully connected layer; w represents the weight of the fully connected layer; b represents the bias of the fully connected layer; σ represents the activation function; a represents the output of the fully connected layer, that is, the current prediction result.
6. The method for identifying montmorillonite on the surface of Mars using hyperspectral images according to claim 1 or 2, characterized in that: The calculation formula of the loss function includes: Among them, Loss is represented by the cross entropy loss function; c is represented by the category label; and n is represented by the number of categories.
7. The method for identifying montmorillonite on the surface of Mars using hyperspectral images according to claim 1 or 2, characterized in that: When executing the step of preprocessing the data in the spectral data set to obtain model training data, the hyperspectral image Mars surface montmorillonite identification method also includes: performing wavelength consistency processing, spectral smoothing, continuum removal and spectral standardization on the data in the spectral data set.
8. The method for identifying montmorillonite on the surface of Mars using hyperspectral images according to claim 1 or 2, characterized in that: The hyperspectral image Mars surface montmorillonite identification method further includes: Preprocess the image to be predicted to obtain the data to be input; The data to be input is input into the recognition model.
9. A hyperspectral imagery system for identifying montmorillonite on the surface of Mars, characterized in that: include: at least one processor; a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 8.
10. A computer readable medium having computer program instructions stored thereon, wherein the computer program instructions can be executed by a processor to implement the method according to any one of claims 1 to 8.