Hyperspectral end member extraction method based on evolutionary multi-mode

Abstract

The invention discloses a hyperspectral end member extraction method based on evolutionary multimodality. The method comprises the following steps: 1, acquiring hyperspectral remote sensing image data and constructing population individuals; 2, grouping the obtained hyperspectral remote sensing image population by using a population grouping evolution mechanism, and respectively evolving to generate a population with higher quality; and 3, carrying out frontier repair and multi-generation evolution on the high-quality population to obtain a group of non-dominated Pareto solution sets so as to obtain a group of hyperspectral end member extraction results. According to the method, the quality of the end members can be measured from multiple angles, the precision of the end members is improved, meanwhile, more diversified excellent end member combinations in a decision-making space are obtained, and therefore the end member extraction effect can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image endmember extraction, and in particular to a hyperspectral endmember extraction method based on evolutionary multimodality. Background Art

[0002] Hyperspectral images contain rich radiation, spatial and spectral information. These spectral bands can provide diagnostic spectral information to identify different land cover types. However, due to the limitation of resolution, there are mixed pixels in hyperspectral remote sensing images, which affects the recognition of ground objects and the accuracy of hyperspectral remote sensing images. Hyperspectral unmixing can solve the problem of mixed pixels by searching for the basic ground object types that constitute the mixed pixels and calculating the proportion of each basic ground object type in the mixed pixels. Endmember extraction is responsible for finding pure pixels in the unmixing process. However, due to the large search space of endmember extraction, for scenes containing tens of thousands of pixels and several endmembers, the number of combinations has reached a very high level, and existing methods are difficult to balance exploration and utilization. In the actual application of hyperspectral unmixing, there are situations where the accuracy is the same but the endmember combinations are completely different. Mining such individuals can provide solutions with excellent target space and more diverse decision space for different scenarios.

[0003] There are at least the following technical problems in the prior art:

[0004] (1) In sparse large-scale decision spaces, existing evolutionary computation-based methods have difficulty in achieving an effective balance between global exploration and local search, which leads to a decrease in search performance and affects the evolutionary efficiency of the population.

[0005] (2) Most of the existing endmember extraction methods are based on evolutionary computing and its improved frameworks, such as particle swarm optimization and differential evolution, which cannot effectively search all pixel positions.

[0006] (3) Existing methods only consider individual fitness performance and the diversity of population target space, ignoring its diversity in the decision space. Summary of the invention

[0007] The present invention aims to address the deficiencies of the above-mentioned prior art and proposes a hyperspectral endmember extraction method based on evolutionary multimodality, in order to measure the quality of endmembers from multiple angles to explore the entire image, obtain more diverse solutions in practical applications, and further improve the accuracy of endmembers, thereby achieving a more perfect endmember extraction effect.

[0008] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:

[0009] The invention provides a hyperspectral endmember extraction method based on evolutionary multimodality, which comprises the following steps:

[0010] S1, obtain a hyperspectral remote sensing image matrix R with a dimension of L×N, where L represents the number of bands of the hyperspectral remote sensing image and N represents the number of pixels of the hyperspectral remote sensing image; let X represent the pixel vector in the hyperspectral remote sensing image matrix R, and , represents the i-th pixel, if , then it means the i-th pixel For the end element; if , then it means the i-th pixel is a mixed pixel;

[0011] S2, using equations (1) to (3) to construct a multi-objective optimization function :

[0012] (1)

[0013] (2)

[0014] (3)

[0015] In formula (1) to formula (3), is the first optimization objective function, is the second optimization objective function, , , , is the dimension-reduced matrix R of the hyperspectral remote sensing image end member value; Represents the number of end members of the hyperspectral remote sensing image matrix R; is the i-th pixel The reconstruction value of

[0016] S3, define the current number of iterations as t, and initialize t=1; define the maximum number of iterations as t max ;

[0017] The distribution of end members and mixed pixels in N pixels constitutes an individual, thereby randomly initializing NP individuals and forming the t-th generation population. ;

[0018] S4, according to equations (2) and (3), calculate the non-dominated level of each individual in the t-th generation population, and divide the A non-dominated individuals with a non-dominated level of "1" in the t-th generation population into the t-th generation non-dominated sub-population ,in, represents the non-dominated subpopulation of the tth generation The ath individual in the tth generation population is divided into the tth generation dominated sub-population. ,in, represents the dominated subpopulation of the tth generation The bth individual of ;

[0019] S5, using a niche-based local search strategy Process and get the ath offspring individual , thus obtaining the tth generation non-dominated offspring population ; Then through the selection operation of fast non-dominated sorting method and crowding distance calculation from and Select the better A individuals from the population to form the t+1th generation non-dominated subpopulation ;

[0020] S6, from Randomly select the i-th individual As the current b individual The guidance vector is used, and the global search strategy based on SAD guidance is used to and Process and get the bth offspring individual , thus obtaining the tth generation dominated offspring population ; Then through the selection operation of fast non-dominated sorting method and crowding distance calculation from and Select the better B individuals from the population to form the t+1th generation dominated subpopulation ;

[0021] S7, will and Merge into the t+1th generation merged population , and Perform frontier patching to obtain the t+1 generation complement set ;in, represents the zth complement of the t+1th generation;

[0022] S8, and Merge into the t+1th generation repair population , and selects from the fast non-dominated sorting method and the end-member distance calculation Select the best NP individuals to form the t+1 generation population ;

[0023] S9, after assigning t+1 to t, return to step S4 and execute until t>t max So far, we can obtain the high-quality population after evolution. ,from The individual corresponding to the non-dominated level "1" is selected as the optimal endmember extraction result of the spectral remote sensing image.

[0024] The endmember extraction method based on evolutionary multimodality described in the present invention is also characterized in that the local search strategy based on niche in S5 is performed according to the following steps:

[0025] Step a: For the current ath individual The P end members in the formula (4) are used to calculate the End Member and Endmember The spectral angle distance between , thus obtaining an end member similarity matrix with dimension P×P :

[0026] (4)

[0027] Step b: Calculate the End Member With another The sum of the spectral angular distances of the end members , thus obtaining the sum of P spectral angular distances , and The end member corresponding to the minimum value in It is determined as the end member to be optimized;

[0028] Step c: Calculate the current t-th generation population respectively The sum of the serial numbers corresponding to all end members in each individual in Compare the sum of the serial numbers corresponding to all end members in and select The n closest individuals constitute In the microhabitat of the decision space, select the individual with the smallest sum of the first optimization objective function and the second optimization objective function in the microhabitat As learning objectives;

[0029] Step d: Calculate the end member to be optimized using formula (4) Respectively with learning target individuals The spectral angle distance between each end member in , and the minimum spectral angle distance is The corresponding end member As the end member to be optimized The evolutionary goal is A new end member is randomly selected within the image of radius r , and The endmembers to be optimized in Replace the individual to obtain a new offspring .

[0030] Furthermore, the global search strategy based on SAD guidance in S6 is performed as follows:

[0031] Step a: Use formula (4) to calculate the current b-th individual Each end member is associated with the guide vector The spectral angle distance between each end member in is retained and The P / 2 end members with the smallest spectral angle distance in Zhongyu The P / 2 endmembers with larger spectral angle distance are taken as the endmembers to be updated;

[0032] Step b: One of the end elements to be updated is The matched end member is locked, and the dynamic radius is used to lock the locked end member and its surroundings. A new endmember is randomly selected within the image range to replace the corresponding endmember to be updated, thereby completing Replace the P / 2 end members to be updated and obtain the bth offspring individual .

[0033] Furthermore, S7 includes:

[0034] S7.1: and Merge into the t+1th generation merged population , and After the individuals in are sorted in ascending order according to the first objective function, they are evenly divided into Z individual clusters, each of which contains NP / Z individuals;

[0035] According to formula (2) and formula (3), the non-dominated level of each individual in the z-th individual cluster is calculated, and the individual with the smallest first objective function value is selected from the individuals corresponding to the non-dominated level of "1" And calculate The k individuals with the closest Euclidean distance in the target space constitute The neighbor set of ;Depend on All endmembers contained in all individuals in the t+1th generation constitute the zth endmember set ;

[0036] S7.2: Select the individual with the smallest second objective function value from the zth individual cluster , and calculate Mid-end member and end member set The spectral angle distance of the middle endmember is selected, so that the P endmember pairs corresponding to the minimum spectral angle distance are selected, which belongs to The end members in the t+1th generation constitute the zth complement ; Thus, the t+1th generation Z complements are obtained and constitute the t+1th generation complement set .

[0037] Furthermore, S8 includes:

[0038] S8.1: According to equations (2) and (3), calculate The non-dominated level of each individual in is constructed as the neighbor set of all individuals whose non-dominated level is "1", and the non-dominated level of each individual in the neighbor set is set to "1".

[0039] S8.2: Use the endmember distance competition method to Perform environmental selection and retain NP individuals to form the t+1 generation population When individuals are in the same non-dominated level and compete, the average Euclidean distance of the end members within the individual on the spectral image of R is calculated, so as to retain the individuals corresponding to the larger average Euclidean distance.

[0040] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the hyperspectral endmember extraction method, and the processor is configured to execute the program stored in the memory.

[0041] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the hyperspectral endmember extraction method when the computer program is executed by a processor.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. The method of the present invention is different from the traditional one. Starting from the practical application of hyperspectral unmixing, the evolutionary multimodal method is used to mine endmember combinations with excellent target space and more diverse solution space, which is more suitable for endmember extraction of hyperspectral remote sensing images of various complex scenes.

[0044] 2. The present invention proposes a group evolution strategy for the population, focusing on convergence and diversity for the frontier solution and the non-frontier solution respectively, and adopting different update strategies to fully explore the sparse large-scale search space for hyperspectral endmember extraction, so as to be more suitable for the endmember extraction of hyperspectral remote sensing images of real big data.

[0045] 3. The present invention utilizes the frontier repair mechanism and combines the advantages of excellent individuals to obtain a better end member combination, thereby further improving the accuracy of end member extraction.

[0046] 4. The present invention also develops an environmental selection competition mechanism based on endmember distance for the problem of hyperspectral endmember extraction, which retains potential multimodal individuals by widening the Pareto frontier, fully explores the diversity of the decision space, and improves the effect of endmember extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0048] In this embodiment, a hyperspectral endmember extraction method based on evolutionary multimodality is used to extract endmembers of hyperspectral remote sensing images. Specifically, Figure 1 As shown, the method is performed in the following steps:

[0049] Step 1: Obtain a hyperspectral remote sensing image matrix R with a dimension of L×N, where L represents the number of bands of the hyperspectral remote sensing image and N represents the number of pixels of the hyperspectral remote sensing image; let X represent the pixel vector in the hyperspectral remote sensing image matrix R, and , represents the i-th pixel, if , then it means the i-th pixel For the end element; if , then it means the i-th pixel Mixed pixels.

[0050] Step 2: Use equations (1) to (3) to construct a multi-objective optimization function :

[0051] (1)

[0052] (2)

[0053] (3)

[0054] In formula (1) to formula (3), is the first optimization objective function, is the second optimization objective function, , , , is the dimension-reduced matrix R of the hyperspectral remote sensing image end member value; Represents the number of end members of the hyperspectral remote sensing image matrix R; is the i-th pixel The reconstruction value of .

[0055] Step 3: Define the current number of iterations as t and initialize t=1; define the maximum number of iterations as t max ;

[0056] The distribution of end members and mixed pixels in N pixels constitutes an individual, thereby randomly initializing NP individuals and forming the t-th generation population. .

[0057] Step 4: Individuals in the population often exhibit different performances in the target space. For the frontier individuals with a non-dominated level of "1", further searching in their local neighborhood will most likely find more potential excellent solutions; the remaining dominated individuals can be fully explored in the decision space, which can improve the diversity of the population and enhance the overall quality of the population.

[0058] To this end, according to equations (2) and (3), the non-dominated level of each individual in the t-th generation population is calculated, and the A non-dominated individuals with a non-dominated level of "1" in the t-th generation population are divided into the t-th generation non-dominated sub-population ,in, represents the non-dominated subpopulation of the tth generation The ath individual in the tth generation population is divided into the tth generation dominated sub-population. ,in, represents the dominated subpopulation of the tth generation The bth individual of .

[0059] Step 5: The tth generation non-dominated subpopulation Evolution;

[0060] Step 5.1: For the current ath individual The P end members in the formula (4) are used to calculate the End Member and End Member The spectral angle distance between , thus obtaining an end member similarity matrix with dimension P×P :

[0061] (4)

[0062] The smaller the value of, the smaller the corresponding end member and The more similar.

[0063] Step 5.2: Calculate the End Member With another The sum of the spectral angular distances of the end members , thus obtaining the sum of P spectral angular distances , The smaller the value, the higher the similarity between the end member and other end members. The end member corresponding to the minimum value in Only updating some endmembers of excellent solutions can improve the diversity of non-dominated individuals while ensuring the quality of offspring.

[0064] Step 5.3: The position of the endmember to be updated in the individual has been locked. Next, in the decision space, Create a microhabitat and select excellent end members of excellent individuals in the microhabitat as learning targets for the end members to be updated;

[0065] Calculate the current t generation population separately The sum of the serial numbers corresponding to all end members in each individual in Compare the sum of the serial numbers corresponding to all end members in and select The n closest individuals constitute In the microhabitat of the decision space, select the individual with the smallest sum of the first optimization objective function and the second optimization objective function in the microhabitat As learning objectives.

[0066] Step 5.4: Calculate the end member to be optimized using formula (4) Respectively with learning target individuals The spectral angle distance between each end member in , and the minimum spectral angle distance is The corresponding end member As the end member to be optimized The evolution target is that the search range of the hyperspectral image is very large and the spatial correlation between pixels is relatively strong. Therefore, in this embodiment, the perturbation is performed near the target end member. A new endmember is randomly selected within the 5×5 image range around it , and The endmembers to be optimized in Replace and get the ath offspring individual .

[0067] Thus, we get the tth generation non-dominated offspring population ; Then through the selection operation of fast non-dominated sorting method and crowding distance calculation from and Select the better A individuals from the population to form the t+1th generation non-dominated subpopulation .

[0068] Step 6, t-th generation subpopulation Evolution;

[0069] Step 6.1: From Randomly select the i-th individual As the current b individual The guidance vector of the current b-th individual is calculated using formula (4) Each end member is associated with the guide vector The spectral angle distance between each end member in the , whose value reflects the similarity between the end members in two individuals, is retained. and The P / 2 end members with the smallest spectral angle distance in Zhongyu The P / 2 endmembers with larger spectral angle distance are taken as the endmembers to be updated;

[0070] Step 6.2: One of the end elements to be updated is The matched end member is locked, and the dynamic radius is used to lock the locked end member and its surroundings. A new endmember is randomly selected within the image range to replace the corresponding endmember to be updated, thereby completing Replace the P / 2 end members to be updated and obtain the bth offspring individual ; Then we get the tth generation dominated offspring population ; Then through the selection operation of fast non-dominated sorting method and crowding distance calculation from and Select the better B individuals from the population to form the t+1th generation dominated subpopulation .

[0071] Step 7: The population grouping evolution strategy in the previous stage can produce excellent offspring locally. In order to further improve the accuracy of the endmembers and enhance the diversity of the optimal solutions, a repair mechanism for the frontier individuals is adopted. The population individuals are first clustered, and then new supplementary individuals are generated in each cluster by integrating the common excellent features of the excellent solutions of the two objective functions. This strong combination can combine the advantages of excellent individuals and obtain a better combination of endmembers.

[0072] Will and Merge into the t+1th generation merged population , and Perform frontier patching to obtain the t+1 generation complement set ;in, Represents the zth complement of the t+1th generation.

[0073] Step 7.1: and Merge into the t+1th generation merged population , and After the individuals in are sorted in ascending order according to the first objective function, they are evenly divided into Z individual clusters, each of which contains NP / Z individuals;

[0074] According to formula (2) and formula (3), the non-dominated level of each individual in the z-th individual cluster is calculated, and the individual with the smallest first objective function value is selected from the individuals corresponding to the non-dominated level of "1" And calculate The k individuals with the closest Euclidean distance in the target space constitute The neighbor set of ;Depend on All endmembers contained in all individuals in the t+1th generation constitute the zth endmember set , The end member in has the excellent characteristic of making the value of the individual first objective function smaller;

[0075] Step 7.2: Select the individual with the smallest second objective function value from the zth individual cluster , which contains end members with the excellent characteristic of making the value of the second objective function of the individual smaller, respectively calculate Mid-end member and end member set The spectral angle distance of the middle endmembers reflects the matching degree of the two sets of excellent endmembers. The P endmember pairs corresponding to the minimum spectral angle distance are selected. The end members in the t+1th generation constitute the zth complement ; Thus, the t+1th generation Z complements are obtained and constitute the t+1th generation complement set ,The complement generated based on this method can take into account the excellent characteristics that both objective function values ​​are relatively small.

[0076] Step 8: and Merge into the t+1th generation repair population , and selects from the fast non-dominated sorting method and the end-member distance calculation Select the best NP individuals to form the t+1 generation population ;

[0077] Firstly, the first frontier is widened, and the non-frontier individuals in the neighbor set of the individuals on the first frontier are pulled to the first frontier to retain them for the next generation. Meanwhile, a new competition method for individuals on the same frontier is proposed, and the end-member distance competition is used to replace the original crowding distance sorting method in environmental selection to enhance the diversity of non-dominated solutions in the decision space.

[0078] Step 8.1: According to equations (2) and (3), calculate The non-dominated level of each individual in the , constructs the neighbor set for all individuals whose non-dominated level is "1", and sets the non-dominated level of each individual in the neighbor set to "1". Individuals in the neighbor set can be retained to the next generation by being pulled to the first frontier.

[0079] Step 8.2: When individuals are on the same frontier, the traditional environment selection uses crowding distance competition, which is the difference between individuals in the target space. Here, considering their diversity in the decision space, the end member distance competition method is used to Perform environmental selection and retain NP individuals to form the t+1 generation population When individuals are in the same non-dominated level and compete, the average Euclidean distance of the end members within the individual on the spectral image of R is calculated, so as to retain the individuals corresponding to the larger average Euclidean distance.

[0080] Step 9: After assigning t+1 to t, return to step S4 and execute until t>t max So far, we can obtain the high-quality population after evolution. ,from The individual corresponding to the non-dominated level "1" is selected as the optimal endmember extraction result of the spectral remote sensing image.

[0081] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0082] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

[0083] The specific implementation method of the present invention is illustrated and the effect of the method of the present invention is verified by using examples of real data sets.

[0084] 1) Prepare real data sets;

[0085] The present invention uses the Samson dataset as a standard dataset to verify the effectiveness of the hyperspectral endmember extraction method based on the global to local evolutionary algorithm. The Samson dataset is a dataset widely used in endmember extraction problems. In the Samson dataset, the image size is 95 pixels × 95 pixels, with a total of 156 bands. The image contains three types of endmembers: land, trees, and water. The method of this patent is EMO, which uses MODPSO, IMODPSO, (μ+λ)MODE, and GLEA for endmember extraction. Table 1 shows the minimum SAD values ​​of the endmembers obtained by the embodiments of the present invention and the existing endmember extraction method and the true endmember.

[0086] 2) Evaluation indicators;

[0087] The SAD values ​​of the real end members are used to evaluate the performance of different multi-objective algorithms and the end member extraction of the present invention, as shown in Table 1:

[0088]

[0089] It can be seen from Table 1 that the accuracy of the endmembers extracted by the present invention is further evaluated by comparing the SAD values ​​with the real endmembers. Among the three endmembers: land, tree and water, the SAD values ​​of the present invention can reach the minimum, indicating that the endmembers extracted by the present invention are closest to the real endmembers.

Claims

1. A hyperspectral endmember extraction method based on evolutionary multimodality, characterized in that: The following steps are involved: S1, obtain a hyperspectral remote sensing image matrix R with a dimension of L×N, where L represents the number of bands of the hyperspectral remote sensing image and N represents the number of pixels of the hyperspectral remote sensing image; let X represent the pixel vector in the hyperspectral remote sensing image matrix R, and , represents the i-th pixel, if , then it means the i-th pixel For the end element; if , then it means the i-th pixel is a mixed pixel; S2, using equations (1) to (3) to construct a multi-objective optimization function : (1) (2) (3) In formula (1) to formula (3), is the first optimization objective function, is the second optimization objective function, , , , is the dimension-reduced matrix R of the hyperspectral remote sensing image end member value; Represents the number of end members of the hyperspectral remote sensing image matrix R; is the i-th pixel The reconstruction value of S3, define the current number of iterations as t, and initialize t=1; define the maximum number of iterations as t max ; The distribution of end members and mixed pixels in N pixels constitutes an individual, thereby randomly initializing NP individuals and forming the t-th generation population. ; S4, according to equations (2) and (3), calculate the non-dominated level of each individual in the t-th generation population, and divide the A non-dominated individuals with a non-dominated level of "1" in the t-th generation population into the t-th generation non-dominated sub-population ,in, represents the non-dominated subpopulation of the tth generation The ath individual in the tth generation population is divided into the tth generation dominated sub-population. ,in, represents the dominated subpopulation of the tth generation The bth individual of ; S5, using a niche-based local search strategy Process and get the ath offspring individual , thus obtaining the tth generation non-dominated offspring population ; Then through the selection operation of fast non-dominated sorting method and crowding distance calculation from and Select the better A individuals from the population to form the t+1th generation non-dominated subpopulation ; S6, from Randomly select the i-th individual As the current b individual The guidance vector is used, and the global search strategy based on SAD guidance is used to and Process and get the bth offspring individual , thus obtaining the tth generation dominated offspring population ; Then through the selection operation of fast non-dominated sorting method and crowding distance calculation from and Select the better B individuals from the population to form the t+1th generation dominated subpopulation ; S7, will and Merge into the t+1th generation merged population , and Perform frontier patching to obtain the t+1 generation complement set ;in, represents the zth complement of the t+1th generation; S8, and Merge into the t+1th generation repair population , and selects from the fast non-dominated sorting method and the end-member distance calculation Select the best NP individuals to form the t+1 generation population ; S9, after assigning t+1 to t, return to step S4 and execute until t>t max Until then, we can obtain a high-quality population after evolution. ,from The individuals corresponding to the non-dominated level "1" are selected as the optimal endmember extraction results of the spectral remote sensing image.

2. The method for extracting endmembers based on evolutionary multimodality according to claim 1, characterized in that: The niche-based local search strategy in S5 is carried out in the following steps: Step a: For the current ath individual The P end members in the formula (4) are used to calculate the Endmember and Endmember The spectral angle distance between , thus obtaining an end member similarity matrix with dimension P×P : (4) Step b: Calculate the End Member With another The sum of the spectral angular distances of the end members , thus obtaining the sum of P spectral angular distances , and The end member corresponding to the minimum value in It is determined as the end member to be optimized; Step c: Calculate the current t-th generation population respectively The sum of the serial numbers corresponding to all end members in each individual in Compare the sum of the serial numbers corresponding to all end members in and select The n closest individuals constitute In the microhabitat of the decision space, select the individual with the smallest sum of the first optimization objective function and the second optimization objective function in the microhabitat As learning objectives; Step d: Calculate the end member to be optimized using formula (4) Respectively with learning target individuals The spectral angle distance between each end member in , and the minimum spectral angle distance is The corresponding end member As the end member to be optimized The evolutionary goal is A new end member is randomly selected within the image of radius r , and The endmembers to be optimized in Replace the individual to obtain a new offspring .

3. The method for extracting endmembers based on evolutionary multimodality according to claim 1, characterized in that: The SAD-guided global search strategy in S6 is performed as follows: Step a: Use formula (4) to calculate the current b-th individual Each end member is associated with the guide vector The spectral angle distance between each end member in is retained and The P / 2 end members with the smallest spectral angle distance in Zhongyu The P / 2 endmembers with larger spectral angle distance are taken as the endmembers to be updated; Step b: One of the end elements to be updated is The matched end member is locked, and the dynamic radius is used to lock the locked end member and its surroundings. A new endmember is randomly selected within the image range to replace the corresponding endmember to be updated, thereby completing Replace the P / 2 end members to be updated and obtain the bth offspring individual .

4. The method for extracting endmembers based on evolutionary multimodality according to claim 1, characterized in that: S7 includes: S7.1: and Merge into the t+1th generation merged population , and After the individuals in are sorted in ascending order according to the first objective function, they are evenly divided into Z individual clusters, each of which contains NP / Z individuals; According to formula (2) and formula (3), calculate the non-dominated level of each individual in the z-th individual cluster, and select the individual with the smallest first objective function value from the individuals corresponding to the non-dominated level of "1" And calculate The k individuals with the closest Euclidean distance in the target space constitute The neighbor set of ;Depend on All endmembers contained in all individuals in the t+1th generation constitute the zth endmember set ; S7.2: Select the individual with the smallest second objective function value from the zth individual cluster , and calculate Mid-end member and end member set The spectral angle distance of the middle endmember is selected, so that the P endmember pairs corresponding to the minimum spectral angle distance are selected, which belongs to The end members in the t+1th generation constitute the zth complement ; Thus, the t+1th generation Z complements are obtained and constitute the t+1th generation complement set .

5. The method for extracting endmembers based on evolutionary multimodality according to claim 1, characterized in that S8 include: S8.1: According to equations (2) and (3), calculate The non-dominated level of each individual in is constructed as the neighbor set of all individuals whose non-dominated level is "1", and the non-dominated level of each individual in the neighbor set is set to "1". S8.2: Use the endmember distance competition method to Perform environmental selection and retain NP individuals to form the t+1 generation population ; When individuals are in the same non-dominated level and compete, the average Euclidean distance of the end members within the individual on the spectral image of R is calculated, so that the individuals corresponding to the larger average Euclidean distance are retained.

6. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the hyperspectral endmember extraction method according to any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the hyperspectral endmember extraction method according to any one of claims 1 to 5 are executed.

Images