Hyperspectral remote sensing mineral end member extraction method based on staged comprehensive learning strategy

Abstract

The invention relates to application of a meta-heuristic algorithm and multi-source hyperspectral remote sensing data in the field of mineral spectrum unmixing, in particular to a hyperspectral remote sensing mineral end member extraction method based on a staged comprehensive learning strategy. The method comprises the following steps: updating mineral end members in a ground spectrum based on a development stage to guide exploration of mineral end members in an airborne image, globally adjusting mineral end member indexes in the ground spectrum based on an exploration stage to accurately position local positions of the mineral end members in the airborne image, and establishing a relation between the ground spectrum and the airborne image; and extracting candidate mineral end members with strong relevance. According to the method, the feasibility of mineral spectrum unmixing in the multi-source hyperspectral remote sensing data is provided, and the defect that mineral end members in ground spectrums and airborne images are difficult to extract at the same time in a traditional method is overcome.

Description

Technical Field

[0001] The present invention relates to the application of meta-heuristic algorithm and multi-source hyperspectral remote sensing data in the field of mineral spectrum unmixing, in particular to a hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy. Background Art

[0002] Minerals, as an important natural resource, record the formation of geological conditions and geochemical environmental elements, and have extremely important scientific research and application value in the green economy. Different minerals usually contain a series of characteristic absorption bands, and these characteristic absorption bands have stable wavelength positions and waveforms in minerals. Extracting pure mineral spectra and estimating their proportion in mixed pixels is of great significance for mine environmental protection and mineral spectroscopy research.

[0003] Hyperspectral remote sensing uses continuous spectral features to distinguish the weak information differences between different components belonging to the same type of land objects, providing a good theoretical and technical basis for its application in mineral end-member extraction. Due to the characteristics of different minerals, they also show significant characteristics in terms of spectral absorption characteristics and reflectance values. On the other hand, the spectral characteristics in the visible-near infrared-short-wave infrared wavelength range depend on the chemical composition and crystal structure of the mineral, which are mainly caused by the electron transition and molecular vibration of the crystal field. In the description of the spectral characteristics of minerals, the absorption peak as an important feature reflects the single diagnostic indicator of the mineral, providing prior knowledge for the accurate analysis of mineral information. However, a single data source is difficult to meet the application requirements of high spatial and spectral resolution at the same time. Multiple models need to be constructed separately according to the characteristics of the data, which is difficult to meet the actual application requirements of high-precision mineral end-member extraction.

[0004] In view of this, the present invention is proposed. Summary of the invention

[0005] The present invention proposes a hyperspectral remote sensing mineral endmember extraction method, system, computer and computer-readable storage medium based on a phased comprehensive learning strategy, aiming to solve the problems of large differences in mineral endmembers and low computational efficiency in the prior art.

[0006] The present invention uses the high resolution of ground spectra to analyze regional mineral characterizations; uses phased comprehensive learning to assign appropriate evolutionary stages to ground spectra and airborne images, fully utilizing the advantages of the data itself to establish connections between mineral endmembers, and uses a model to simultaneously extract mineral endmembers from ground spectra and airborne images. The present invention provides the feasibility of applying multi-source hyperspectral remote sensing data to mineral endmember extraction, improving the accuracy of mineral endmembers and the portability of related technologies.

[0007] First, a hyperspectral remote sensing mineral endmember extraction method based on a phased comprehensive learning strategy is proposed, including: obtaining ground spectra of mineral samples, extracting potential mineral endmembers from the ground spectra, and using sequence index coding for the ground spectra; obtaining airborne images marked with the locations of mineral sample sampling points, and using coordinate position coding for pixels in the airborne images; performing a local search for the sequence index coding based on the development phase for the ground spectra, and covering the coordinate position coding range through the exploration phase, expanding from local to global; locating the locations of potential mineral endmembers based on the development phase for the airborne images, and globally updating the potential mineral endmembers in the ground spectra through the exploration phase, the information obtained in the development phase and the exploration phase complement each other, and a connection between the ground spectra and the airborne images is established, and the sequence index coding and the coordinate position coding are iteratively updated respectively to extract the actual mineral endmember combinations in the ground spectra and the airborne images.

[0008] Secondly, a hyperspectral remote sensing mineral endmember extraction device based on a phased comprehensive learning strategy is proposed, including: a ground spectrum preprocessing module, which is configured to: obtain the ground spectrum of the mineral sample, extract potential mineral endmembers from the ground spectrum, and the ground spectrum uses sequence index coding; an airborne image preprocessing module, which is configured to: obtain airborne images marked with the locations of mineral sample sampling points, and the pixels in the airborne images use coordinate position coding; a model, which is configured to: perform local search on the sequence index coding based on the development stage for the ground spectrum, cover the coordinate position coding range through the exploration stage, and expand from local to global; locate the location of potential mineral endmembers based on the development stage for the airborne image, and globally update the potential mineral endmembers in the ground spectrum through the exploration stage. The potential mineral endmember information obtained in the development stage and the exploration stage complement each other, establish a connection between the ground spectrum and the airborne image, iteratively update the sequence index coding and the coordinate position coding, and extract the actual mineral endmember combination in the ground spectrum and the airborne image.

[0009] In some examples, the coordinate position encoding is the encoding of the absolute position of the pixel in the airborne image, and the potential mineral end members of each mineral are determined based on its horizontal and vertical coordinate positions in the airborne image, and its value range is the maximum number of rows and columns of the airborne image.

[0010] In some examples, a Sigmond function is used to map the decimal ground spectrum sequence index to binary, where 0 and 1 represent selection as or not as a potential mineral end member, respectively.

[0011] In some examples, the development phase includes searching the spatial neighborhood, the exploration phase includes searching the global space, and the coordinate position encoding promotes both the sequence index encoding and the coordinate position encoding update at the next iteration. The formulas for the development phase and the exploration phase are as follows:

[0012]

[0013] in, represents the optimal sequence index code and optimal coordinate position code of the current potential mineral end member, x gbest represents the globally optimal sequence index encoding and the optimal coordinate position encoding, x i It represents the sequence index code and coordinate position code of the current potential mineral end member, and a and F represent random numbers between (0,1).

[0014] In some examples, the development phase optimizes sequence index encoding in one-dimensional space, which affects the update of coordinate position encoding, enhances the diversity presentation of ground spectra in local space, and searches for mineral end members with better global performance in airborne imagery.

[0015] In some examples, the exploration phase optimizes the coordinate position encoding in two-dimensional space, which promotes the update of the sequence index encoding in the next iteration, thereby achieving local update of potential mineral end members in airborne imagery.

[0016] In some examples, ground spectra perform local searches for sequence index codes in the development phase, and airborne images perform global searches in the exploration phase. Then, airborne images determine the local space of their optimal potential mineral endmembers, thereby performing global spatial adjustments to the potential mineral endmembers of the ground spectra, completing information interaction of potential mineral endmembers between multi-source hyperspectral remote sensing data, and establishing a connection between ground spectra and airborne images.

[0017] In the third aspect, a computer is proposed, comprising: a processor; a memory, comprising one or more computer program modules; wherein the one or more computer program modules are stored in the memory and configured to be executed by the processor, and the one or more computer program modules include instructions for implementing the hyperspectral remote sensing mineral endmember extraction method based on a staged comprehensive learning strategy.

[0018] In a fourth aspect, a computer-readable storage medium is proposed for storing non-temporary computer-readable instructions, which, when executed by a computer, can implement the hyperspectral remote sensing mineral endmember extraction method based on a phased comprehensive learning strategy.

[0019] The present invention has the following beneficial effects: (1) The present invention uses multi-source hyperspectral remote sensing data, giving full play to the complementary advantages of ground spectra and airborne images, making it highly portable in the field of mineral end-member extraction. (2) The present invention adopts a phased comprehensive learning strategy, and improves the accuracy of mineral end-member extraction by assigning appropriate evolutionary stages to ground spectra and airborne images respectively. (3) The present invention establishes a mapping relationship between ground spectra and airborne images through information interaction at different stages in the algorithm, thereby enhancing the interpretability of the extracted mineral end-members. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a method for extracting end-members of minerals from hyperspectral remote sensing based on a phased comprehensive learning strategy provided in one embodiment of the present invention.

[0021] Figure 2 This is an application example of the present invention.

[0022] Figure 3 The present invention provides an embodiment of the present invention providing a process of spectral information divergence variation between four mineral end members and standard mineral spectra in ground spectra and airborne images. DETAILED DESCRIPTION

[0023] Figure 1 The present invention is a flowchart of a method for extracting mineral endmembers from hyperspectral remote sensing based on a phased comprehensive learning strategy. Based on the development phase, the potential mineral endmembers in the ground spectrum are updated to guide the exploration of potential mineral endmembers in airborne images. Then, based on the exploration phase, the potential mineral endmember index in the ground spectrum is globally adjusted to accurately locate the local position of the potential mineral endmember in the airborne image, establish a connection between the ground spectrum and the airborne image, and extract potential mineral endmembers with strong correlation. The present invention provides the feasibility of unmixing mineral spectra in multi-source hyperspectral remote sensing data, making up for the deficiency that traditional methods are difficult to simultaneously extract mineral endmembers from ground spectra and airborne images. The present invention mainly proposes a large framework to solve the endmember extraction of multi-source data. Therefore, the present invention does not limit the specific algorithm of the phased comprehensive learning strategy, and basic meta-heuristic algorithms are applicable.

[0024] (1) Use a handheld spectrometer to test ground mineral samples to obtain ground spectra and extract potential mineral end members of major altered minerals. The SVC HR1024 ground spectrometer can be used to collect ground spectra. ENVI 5.3 software can be used to extract altered mineral end members.

[0025] (2) Use airborne sensors to acquire images, mark the sampling point locations of mineral samples in the images, and obtain pixel spectra. Airborne sample images can be acquired using the CASI / SASI platform, and atmospheric correction can be performed using the fast line-of-sight atmospheric analysis model in ENVI 5.3 software. Hybrid pixel decomposition technology is used to decompose the pixel spectra containing multiple altered minerals. The sampling point locations are measured on the ground by GPS. The reason for marking the sampling point locations of mineral samples in airborne images is to correspond to the mineral category, that is, to know which mineral the end member to be found belongs to. The ground spectrum and the pixel spectrum are acquired at the same time to avoid local time differences.

[0026] (3) The encoding in the randomly initialized phased optimization process is as follows: the ground spectrum uses sequence index encoding, and the airborne image uses coordinate position encoding.

[0027] The coordinate position is encoded as the absolute position of the pixel in the airborne image. The potential mineral end member of each mineral is determined based on its horizontal and vertical coordinate position in the airborne image, and its value range is the maximum number of rows and columns in the image.

[0028] Initially, the ground spectrum sequence index code is decimal, and the Sigmond function is used to map the decimal ground spectrum sequence index code to binary, where 0 and 1 represent selection or non-selection as a candidate end member, respectively.

[0029] (4) Comprehensive learning in stages - early stage of algorithm operation

[0030] In the development phase, the sequence index encoding is optimized in one-dimensional space, which affects the update of the coordinate position encoding, enhances the diversity presentation of ground spectra in local space, and searches for potential mineral end members with better global performance in airborne images. The formula for the development phase is as follows:

[0031]

[0032] Among them, x i Indicates the current individual sequence index and coordinate position encoding, Indicates the optimal encoding of the current iteration sequence index and coordinate position. gbest represents the optimal encoding of the global iteration sequence index and coordinate position, a represents a random number between (0,1). The "individual" mentioned above is the encoding of the potential mineral end member in the algorithm.

[0033] (5) Comprehensive learning in stages - late stage of algorithm operation

[0034] The exploration phase optimizes the coordinate position in two-dimensional space, which promotes the update of the sequence index in the next iteration and realizes the local update of the mineral end members in the airborne image. The formula of the exploration phase is as follows:

[0035]

[0036] Among them, x i Indicates the current individual sequence index and coordinate position encoding, Indicates the optimal encoding of the current iteration sequence index and coordinate position. gbest It represents the optimal encoding of the global iteration sequence index and coordinate position, and F represents a random number between (0,1).

[0037] (6) Establishing the connection between ground-based spectra and airborne images

[0038] The ground spectra were searched locally for sequence indexes in the development phase, and the airborne images were searched globally in the exploration phase. Then, the local space of the best potential mineral endmembers was determined in the airborne images, and the potential mineral endmembers in the ground spectra were adjusted globally to establish the connection between the ground spectra and the airborne images.

[0039] (7) Iteratively update the sequence index and coordinate position

[0040] Through a phased comprehensive learning strategy, global and local searches are performed on the sequence index encoding and coordinate position encoding, 30 iterations are used to update the individual position, output the optimal encoding of the sequence index and coordinate position, calculate the spectral information divergence and evaluate the individual position.

[0041] (8) Extraction of optimal mineral end-member combinations

[0042] Through a phased comprehensive learning strategy, the output is the optimal encoding of sequence index and coordinate position, and the optimal mineral endmember combination in ground spectrum and airborne imagery is extracted to achieve mineral spectral unmixing of multi-source hyperspectral remote sensing data.

[0043] The present invention also proposes an embodiment of a hyperspectral remote sensing mineral end member extraction system based on a phased comprehensive learning strategy. The system includes: a ground spectrum preprocessing module, an airborne image preprocessing module, and a model.

[0044] The ground spectrum preprocessing module is configured to obtain the ground spectrum of the mineral sample, extract potential mineral end members from the ground spectrum, and encode the ground spectrum using sequence index. The airborne image preprocessing module is configured to obtain the airborne image marked with the location of the mineral sample sampling point, and encode the pixels in the airborne image using coordinate position. For more detailed preprocessing of the ground spectrum and airborne image, refer to the above steps (1)-(3).

[0045] A model is configured as follows: for ground spectra, local search of sequence index is performed based on the development phase, and the range of coordinate position coding is covered through the exploration phase, expanding from local to global; for airborne images, the location of potential mineral end members is located based on the development phase, and the potential mineral end members in the ground spectra are globally updated through the exploration phase. The potential mineral end member information obtained in the development phase and the exploration phase complement each other, establish the connection between the ground spectra and the airborne images, iteratively update the sequence index code and the coordinate position code, and extract the actual mineral end member combination in the ground spectra and the airborne images. For a more detailed description of the model, refer to the above steps (4)-(8).

[0046] The present invention also provides an embodiment of a computer. The computer includes a processor and a memory. The memory is used to store non-temporary computer-readable instructions (such as one or more computer program modules). The processor is used to run the non-temporary computer-readable instructions, and when the non-temporary computer-readable instructions are run by the processor, one or more steps in the hyperspectral remote sensing mineral end member extraction method based on the staged comprehensive learning strategy described above can be executed. The memory and the processor can be interconnected through a bus system and / or other forms of connection mechanisms.

[0047] For example, the processor may be a central processing unit (CPU), a graphics processing unit (GPU), or other forms of processing units with data processing capabilities and / or program execution capabilities. For example, the central processing unit (CPU) may be an X86 or ARM architecture, etc. The processor may be a general-purpose processor or a special-purpose processor, and may control other components in the computer to perform desired functions.

[0048] For example, the memory may include any combination of one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), compact disc read-only memory (CD-ROM), USB memory, flash memory, etc. One or more computer program modules may be stored on the computer-readable storage medium, and the processor may run one or more computer program modules to implement various functions of the computer.

[0049] The present invention also provides an embodiment of a computer-readable storage medium, which is used to store non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, one or more steps in the hyperspectral remote sensing mineral end member extraction method based on the staged comprehensive learning strategy can be implemented. The hyperspectral remote sensing mineral end member extraction method and system based on the staged comprehensive learning strategy provided by the embodiment of the present invention can be stored in a computer-readable storage medium when it is implemented in the form of software and sold or used as an independent product. For the relevant description of the storage medium, please refer to the corresponding description of the memory in the computer above, which will not be repeated here.

[0050] like Figure 2 As shown, a specific application case is listed below. The Longjiaoshan area is located in Daye City, Hubei Province, in the southeastern Hubei ore cluster area of ​​the middle and lower reaches of the Yangtze River in my country, with an area of ​​about 163 square kilometers. The study area is rich in metal mineral resources, such as gold, copper, zinc, etc. Magmatic activity is the basis for the formation of ore deposits. The Longjiaoshan area has experienced multiple periods of magmatic intrusions to form various igneous rocks such as granite and basalt. These magmatic activities provide the source of minerals. In addition, hydrothermal activity plays a key role in the formation of ore deposits. High-temperature and high-pressure hydrothermal fluids deep underground flow through the crust, dissolve and transport minerals, and finally precipitate under suitable conditions to form ore deposits. The main mineral types include calcite, alunite, illite, kaolinite, prehnite and calcium hydroxide.

[0051] The ground spectrum and airborne imagery are updated through the development and exploration phases of the phased comprehensive learning strategy, and the connection between the ground spectrum and airborne imagery is established so that the information can be complementary. Figure 3 As shown in the figure, through experimental verification, the variation process of spectral information divergence between mineral endmembers and standard mineral spectra in ground spectra and airborne images shows good multi-source hyperspectral remote sensing data fusion capability, and the average spectral information divergence of mineral endmembers reaches 0.104. Compared with the traditional mineral endmember extraction method, the present invention maintains the spectral consistency of similar minerals in multi-source hyperspectral remote sensing data, and has high interpretability for mineral absorption characteristics. In addition, the phased comprehensive learning strategy provides appropriate stages for ground spectra and airborne images respectively, establishes the connection between multi-source hyperspectral remote sensing data according to the advantages of the data itself, and improves the accuracy of mineral endmember extraction.

Claims

1. A hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy, characterized in that: include: Obtaining the ground spectrum of the mineral sample, extracting potential mineral end members from the ground spectrum, and encoding the ground spectrum using a sequence index; Obtain airborne images marked with the locations of mineral sample sampling points, where pixels in the airborne images are encoded using coordinate positions; Based on the development stage, a local search is performed on the sequence index code for the ground spectrum, and the coordinate position code range is covered through the exploration stage, expanding from local to global. Based on the development stage, the potential mineral endmembers are located for the airborne imagery, and the potential mineral endmembers in the ground spectrum are globally updated through the exploration stage. The potential mineral endmember information obtained in the development stage and the exploration stage complement each other, and the connection between the ground spectrum and the airborne imagery is established. The sequence index code and the coordinate position code are iteratively updated respectively, and the actual mineral endmember combination in the ground spectrum and the airborne imagery is extracted.

2. The hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy according to claim 1 is characterized in that: Coordinate position coding is the coding of the absolute position of the pixel in the airborne image. The potential mineral end member of each mineral is determined based on its horizontal and vertical coordinate position in the airborne image, and its value range is the maximum number of rows and columns in the airborne image.

3. The hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy according to claim 1 is characterized in that: The decimal ground spectrum sequence index was mapped to binary using the Sigmond function, where 0 and 1 respectively indicated that it was selected as or not as a potential mineral end member.

4. The hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy according to claim 1 is characterized in that: The development phase includes searching the spatial neighborhood, and the exploration phase includes searching the global space. The coordinate position encoding promotes the update of the sequence index encoding and the coordinate position encoding at the next iteration. The formulas for the development phase and the exploration phase are as follows: in, represents the optimal sequence index code and optimal coordinate position code of the current potential mineral end member, x gbest represents the globally optimal sequence index encoding and the optimal coordinate position encoding, x i It represents the sequence index code and coordinate position code of the current potential mineral end member, and a and F represent random numbers between (0,1).

5. The hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy according to claim 4 is characterized in that: In the development phase, the sequence index encoding is optimized in one-dimensional space, which affects the update of the coordinate position encoding.

6. The hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy according to claim 4 is characterized in that: In the exploration phase, the coordinate position coding is optimized in two-dimensional space, which promotes the update of sequence index coding in the next iteration and realizes the local update of potential mineral end members in airborne images.

7. The hyperspectral remote sensing mineral end member extraction method based on a phased comprehensive learning strategy according to claim 1 is characterized in that: The ground spectrum performs a local search for sequence index codes in the development phase, and the airborne imagery performs a global search in the exploration phase. Then, the airborne imagery determines the local space of its optimal potential mineral endmembers, thereby making global spatial adjustments to the potential mineral endmembers of the ground spectrum, completing the information exchange of potential mineral endmembers between multi-source hyperspectral remote sensing data, and establishing a connection between the ground spectrum and airborne imagery.

8. A hyperspectral remote sensing mineral end member extraction system based on a phased comprehensive learning strategy, characterized in that: include: A ground spectrum preprocessing module is configured to: obtain a ground spectrum of a mineral sample, extract potential mineral end members from the ground spectrum, and encode the ground spectrum using a sequence index; The airborne image preprocessing module is configured to: obtain airborne images marked with the locations of mineral sample sampling points, wherein the pixels in the airborne images are encoded using coordinate positions; A model is configured as follows: for ground spectra, local search is performed on sequence index codes based on the development phase, and the range of coordinate position codes is covered through the exploration phase, expanding from local to global; for airborne images, the locations of potential mineral endmembers are located based on the development phase, and potential mineral endmembers in ground spectra are globally updated through the exploration phase. The potential mineral endmember information obtained in the development phase and the exploration phase complement each other, a connection is established between ground spectra and airborne images, and the sequence index codes and coordinate position codes are iteratively updated respectively to extract actual mineral endmember combinations in ground spectra and airborne images.

9. A computer, characterized in that: include: processor; a memory including one or more computer program modules; Wherein, the one or more computer program modules are stored in the memory and configured to be executed by the processor, and the one or more computer program modules include instructions for implementing the hyperspectral remote sensing mineral end member extraction method based on a staged comprehensive learning strategy as described in any one of claims 1-7.

10. A computer-readable storage medium for storing non-transitory computer-readable instructions, characterized in that: When the non-transitory computer-readable instructions are executed by a computer, the hyperspectral remote sensing mineral end-member extraction method based on a phased comprehensive learning strategy described in claims 1-7 can be implemented.

Images