Spectral index and waveband combination optimization method and system for hyperspectral image classification

Abstract

The invention particularly relates to a spectral index and wave band combination optimization method and system for hyperspectral image classification. The method comprises the following steps: acquiring surface feature hyperspectral data; creating a wave band combination list according to the ground object hyperspectral data; generating a plurality of wave band arrangement modes for each wave band combination, and calculating a spectral index corresponding to the wave band combination based on each wave band arrangement mode corresponding to the wave band combination; performing pairing t test on the spectral indexes corresponding to the wave band combination, and if obtained p values are smaller than a preset value, taking the spectral indexes corresponding to the wave band combination as alternative spectral indexes; traversing all wave band combinations; and selecting the spectral index with the minimum p value as the optimal spectral index of ground feature classification, and selecting the wave band combination corresponding to the spectral index with the minimum p value as the optimal wave band combination of ground feature classification. According to the method, the precision and the efficiency of hyperspectral image ground feature classification are remarkably improved by optimizing the waveband combination and the spectral index and combining the pairing t test.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral image classification, and particularly relates to a method and system for optimizing spectral indices and band combinations for hyperspectral image classification. Background Art

[0002] Hyperspectral remote sensing images cover multiple bands including ultraviolet, visible light, near-infrared, and short-wave infrared, contain rich spectral information of ground objects, and have the potential for high-precision target classification. However, hyperspectral data has a high dimension, and there is often data correlation between different bands, resulting in high data redundancy, which seriously affects the operation efficiency of hyperspectral image classification.

[0003] To solve the problem of hyperspectral data redundancy, related image classification algorithms usually use feature extraction or data dimensionality reduction methods to reduce the computational complexity. Multi-band spectral indices are a robust spectral analysis method. By using several bands with strong ground object discrimination ability, band operation indices are constructed to replace the high-dimensional hyperspectral raw data, significantly reducing the computational complexity.

[0004] As the key to the implementation of the multi-band spectral index method, the selection of effective spectral indices and characteristic bands depends on the corresponding band optimization method. However, since current multi-spectral indices are often applied to quantitative analysis of specific physical quantities (such as chlorophyll concentration inversion, soil organic matter inversion, etc.), related spectral index and band combination optimization methods are usually constructed based on linear regression. The hyperspectral image classification task is essentially a classification problem. Therefore, the spectral indices for hyperspectral image classification cannot be determined using the existing band optimization methods based on linear regression, resulting in a lack of theoretical basis for the selection of related spectral indices and their characteristic band combinations.

[0005] In order to identify different ground object types through a limited number of detection bands, the present invention proposes an optimal band combination algorithm to preferably select the best detection wavelengths for ground object classification. Using the optimized multi-band spectral indices is expected to significantly reduce the dimension of hyperspectral data and identify ground object types with a lower computational complexity. The core logic of band optimization is to perform statistical analysis on the multi-spectral index values of each ground object. If there are significant differences between the multi-spectral index values of all ground objects, then this index and the corresponding band combination for calculating this index are suitable for ground object discrimination. By traversing all possible band combinations and performing statistical analysis in turn, the multi-spectral index value with the most obvious difference is found, and the band combination for calculating this value is the optimal band combination. Summary of the Invention

[0006] In view of the technical problems that there is a lack of theoretical basis for the selection of multi - band spectral indices and their band combinations in hyperspectral image classification, and the existing methods cannot effectively solve the problems of hyperspectral data redundancy and classification accuracy, a method for optimizing spectral indices and band combinations for hyperspectral image classification is provided. The present invention mainly uses the combined paired t - test to perform statistical analysis on the band combinations, screen out the optimal spectral indices and their band combinations, so as to significantly reduce the dimension of hyperspectral data, improve the classification accuracy and operation efficiency.

[0007] The technical means adopted by the present invention are as follows:

[0008] A method for optimizing spectral indices and band combinations for hyperspectral image classification, the steps include:

[0009] Step 1: Obtain the hyperspectral data of the ground object, and the hyperspectral data of the ground object includes s effective band numbers;

[0010] Step 2: Generate calculation formulas for candidate spectral indices including m bands;

[0011] Step 3: According to the hyperspectral data of the ground object, create a band combination list, and the band combination list includes C s m band combinations;

[0012] Step 4: Generate A m m kinds of band permutations for each of the band combinations. Based on each band permutation corresponding to the band combination, calculate the spectral indices corresponding to the band combination through the calculation formulas of the candidate spectral indices;

[0013] Step 5: Perform paired t - tests on the spectral indices corresponding to the band combinations. If all the obtained p - values are less than the preset value, take the spectral indices corresponding to the band combinations as alternative spectral indices; if one of the obtained p - values is greater than the preset value, skip the current band combination;

[0014] Step 6: Traverse all the band combinations, sequentially execute Step 4 to Step 5, and stop the loop when all the band combinations complete the paired t - tests;

[0015] Step 7: Output all the alternative spectral indices, select the spectral index with the smallest p - value as the optimal spectral index for ground object classification, and select the band combination corresponding to the spectral index with the smallest p - value as the optimal band combination for ground object classification.

[0016] Further, when m = 3, the calculation formula for the candidate spectral index is:

[0017] Index1=(R a -R b ) / (Ra +R c )

[0018] Index2 = R a / (R b +R c )

[0019] Index3 = (R a -R b ) / (R b -R c )

[0020] Index4 = (R a -R b ) / R c

[0021] Among them, Index1 is the first candidate spectral index, Index2 is the second candidate spectral index, Index3 is the third candidate spectral index, Index4 is the fourth candidate spectral index, R a is the spectral reflectance in band a, R b is the spectral reflectance in band b, R c is the spectral reflectance in band c.

[0022] Furthermore, the bands a, b, and c do not overlap with each other.

[0023] Furthermore, the preset p-value of the paired t-test is 0.05.

[0024] Furthermore, step 5 specifically includes:

[0025] Performing a paired t-test on the spectral index corresponding to the band combination. If the p-value of the band combination is greater than the preset value for once, skip the current band combination and perform a paired t-test on the next band combination;

[0026] If all p-values in the band combination are less than the preset value, use the spectral index set corresponding to the band combination as the alternative spectral index.

[0027] A spectral index and band combination optimization system for hyperspectral image classification, used to implement the above spectral index and band combination optimization method for hyperspectral image classification, includes:

[0028] A data acquisition module, used to acquire ground object hyperspectral data with s bands;

[0029] A band combination module, used to generate a band combination list based on the hyperspectral data, and the band combination list includes C s m band combinations;

[0030] A band arrangement module for generating A m m kinds of band arrangements for each band combination;

[0031] A spectral index calculation module for calculating the spectral index corresponding to each band arrangement;

[0032] A paired t - test module for performing a paired t - test on the spectral indices corresponding to each band combination, screening the spectral indices corresponding to the cases where the p - value obtained from the paired t - test is less than a preset value, and taking the affiliated spectral indices as alternative spectral indices;

[0033] A traversal control module for controlling the traversal process of band combinations, sequentially performing band arrangement generation, spectral index calculation, and paired t - test until all band combinations are completed for testing;

[0034] An output module for outputting all alternative spectral indices, selecting the spectral index with the smallest p - value as the optimal spectral index for ground object classification, and selecting the band combination corresponding to the spectral index with the smallest p - value as the optimal band combination for ground object classification.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] 1. The multi - band spectral index optimization method provided by the present invention, combined with the paired t - test, realizes a significant improvement in the accuracy of ground object classification in hyperspectral images, can effectively distinguish ground object types with similar spectral features, reduce classification errors at the same time, and improve the reliability of classification results.

[0037] 2. The present invention provides a multi - band spectral index optimization method and a band combination screening process, realizing providing a theoretical basis for the selection of multi - band spectral indices and their band combinations for hyperspectral image classification, filling the defect of the current lack of multi - band spectral index optimization methods for hyperspectral ground object classification, and solving the problem of the lack of systematicness and scientificity in spectral index selection in the prior art.

[0038] For the above reasons, the present invention can be widely promoted in the fields of hyperspectral image classification, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following - described drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0040] Figure 1This is the flowchart of the method for optimizing spectral indices and band combinations in hyperspectral image classification of the present invention.

[0041] Figure 2 This is the true color image obtained by the AVIRIS airborne hyperspectral sensor in the embodiment of the present invention.

[0042] Figure 3 This is the significance gap map of ground object targets under different spectral indices and band combinations in the embodiment of the present invention.

[0043] Figure 4 This is the hyperspectral image classification result map in the embodiment of the present invention. Detailed implementation manners

[0044] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0045] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order different from those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0046] As Figure 1 shown, the present invention provides a method for optimizing spectral indices and band combinations in hyperspectral image classification, and the specific steps are as follows:

[0047] Step 1: Obtain ground object hyperspectral data, where the ground object hyperspectral data includes s effective band numbers.

[0048] Taking the AVIRIS airborne hyperspectral sensor as an example, ground object hyperspectral data is obtained through the sensor, and a true color image as Figure 2 shown is obtained. This sensor includes 224 bands covering 370 nm to 2150 nm. After removing the bands with relatively low signal-to-noise ratio affected by the atmosphere, it includes 183 effective bands, that is, s = 183.

[0049] Step 2: Generate calculation formulas for candidate spectral indices including m bands.

[0050] Based on existing research, initially determine multiple candidate spectral indices including m (m = 2, 3, 4) bands. Taking the three-band spectral index, i.e., m = 3 as an example, select the following four spectral indices as candidate indices based on the three-band spectral indices used in existing research:

[0051] Index1 = (R a - R b ) / (R a + R c )

[0052] Index2 = R a / (R b + R c )

[0053] Index3 = (R a - R b ) / (R b - R c )

[0054] Index4 = (R a - R b ) / R c

[0055] Among them, Index1 is the first candidate spectral index, Index2 is the second candidate spectral index, Index3 is the third candidate spectral index, Index4 is the fourth candidate spectral index, R a is the spectral reflectance in band a, R b is the spectral reflectance in band b, R c is the spectral reflectance in band c, and bands a, b, and c do not repeat each other.

[0056] Step 3: Create a band combination list based on the ground object hyperspectral data, and the band combination list includes C s m band combinations.

[0057] Since the bands involved in the multi-spectral index should not repeat (a ≠ b ≠ c), it is necessary to first select m bands from all bands without repetition to form the spectral index. For hyperspectral data with s bands, according to the combination principle, there are C s m different band combinations in this list. Similarly, taking the AVIRIS airborne hyperspectral sensor as an example, the band combination list includes C 183 3 = 106 Band combination.

[0058] Step 4: Generate A m m kinds of band arrangements for each band combination. Based on each band arrangement corresponding to the band combination, calculate the spectral index corresponding to the band combination through the calculation formula of the candidate spectral index.

[0059] For the three-band spectral index, that is, when m = 3, each band combination has A3 3 = 6 permutation possibilities. In order to make the paired t-test used in Step 5 valid, 30 to 40 different spectral index values need to be calculated for each ground object to be classified.

[0060] Step 5: Conduct a paired t-test on the spectral index corresponding to the band combination. If all the obtained p-values are less than the preset value, take the spectral index corresponding to the band combination as the alternative spectral index; if one of the obtained p-values is greater than the preset value, skip the current band combination.

[0061] Conduct a paired t-test on the spectral index values of n ground objects to be classified. Since the multi-band spectral index values obtained from the reflection spectra of the ground objects to be classified collected under different conditions are independent of each other, and their values theoretically follow a normal distribution, a paired t-test can be used to evaluate whether there is a significant difference between the multi-band spectral index values under different band combinations.

[0062] As the main output parameter of the t-test, the p-value represents the probability that the two groups of observed values follow a similar distribution. If the p-value is very small, it indicates that the probability of the hypothesis occurring is very small. Even if it occurs, according to the small probability principle, we have reason to reject this hypothesis. Therefore, the smaller the p-value, the more sufficient the reason to reject the original hypothesis.

[0063] In the present invention, for the paired t-test of the multi-spectral indices of n ground objects to be classified, a total of C n 2 tests are required, record the p-values, and set the preset value of the p-value to 0.05.

[0064] If the p-value is greater than the preset value, it indicates that the multi-spectral indices of these two ground objects under the current band combination do not have a significant difference, that is, the multi-spectral index under the current band combination cannot be used to distinguish these two ground objects.

[0065] Conduct a paired t-test on the spectral index corresponding to the band combination. If one of the p-values of the band combination is greater than the preset value, it indicates that at least two ground objects cannot be distinguished by the spectral index under the current band combination. Skip the current loop and conduct a paired t-test on the next band combination;

[0066] If all p-values in the band combination are less than the preset value, it indicates that the multi-band spectral index values of each type of ground object are significantly different from those of any other ground object. Therefore, the spectral index under this band combination can be used to distinguish these ground objects, and the set of spectral indices corresponding to the band combination is used as the alternative spectral indices, and then the paired t-test is performed on the next band combination.

[0067] Step 6: Traverse all band combinations, and steps 4 to 5 are sequentially executed for all band combinations. When the paired t-tests for all band combinations are completed, the loop stops.

[0068] Step 7: Output all the alternative spectral indices, select the spectral index with the smallest p-value as the optimal spectral index for ground object classification, and select the band combination corresponding to the spectral index with the smallest p-value as the optimal band combination for ground object classification.

[0069] After the loop, record the p-value distribution, which can be plotted as Figure 3 the sliced graph shown. The dark area in the graph is the area where the optimal band combination is distributed.

[0070] Use the selected band combination to classify the hyperspectral image, and the obtained result is as Figure 4 shown. The result shows that the selected optimal multi-spectral index and band combination can achieve hyperspectral image classification with a relatively low computational complexity.

[0071] The present invention also includes a system for optimizing spectral indices and band combinations for hyperspectral image classification, which is used to implement the method for optimizing spectral indices and band combinations for hyperspectral image classification. The system includes:

[0072] A data acquisition module, which is used to acquire the hyperspectral data of ground objects with s bands;

[0073] A band combination module, which is used to generate a list of band combinations based on the hyperspectral data. The list of band combinations includes C s m band combinations;

[0074] A band arrangement module, which is used to generate A m m band arrangement methods for a band combination;

[0075] A spectral index calculation module, which is used to calculate the spectral indices corresponding to each band arrangement;

[0076] A paired t-test module, which is used to perform paired t-tests on the spectral indices corresponding to each band combination, and screen the spectral indices corresponding to the p-values obtained from the paired t-tests that are less than the preset value, and use the belonging spectral indices as the alternative spectral indices;

[0077] The traversal control module is used to control the traversal process of band combinations, and sequentially execute band arrangement generation, spectral index calculation, and paired t-test until all band combinations are completed for inspection;

[0078] The output module is used to output all alternative spectral indices, select the spectral index with the smallest p-value as the optimal spectral index for ground object classification, and select the band combination corresponding to the spectral index with the smallest p-value as the optimal band combination for ground object classification.

[0079] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing spectral index and band combination for hyperspectral image classification, characterized in that the steps include: Step 1: Acquire hyperspectral data of ground objects, wherein the hyperspectral data of ground objects includes s effective bands; Step 2, generating a calculation formula for candidate spectral indices including m bands; Step 3: Create a band combination list based on the ground feature hyperspectral data, the band combination list includes C s m Band combinations; Step 4: Generate A for each band combination m m a band arrangement method, and based on each band arrangement method corresponding to the band combination, calculate the spectral index corresponding to the band combination by using the calculation formula of the candidate spectral index; Step 5: Perform a paired t-test on the spectral index corresponding to the band combination. If the obtained p-values ​​are all less than the preset values, the spectral index corresponding to the band combination is used as an alternative spectral index; if the obtained p-value is greater than the preset value, skip the current band combination; Step 6: traverse all band combinations and execute steps 4 to 5 in sequence. When all band combinations complete the paired t test, stop the loop; Step 7: Output all candidate spectral indexes, select the spectral index with the smallest p value as the optimal spectral index for land feature classification, and select the band combination corresponding to the spectral index with the smallest p value as the optimal band combination for land feature classification.

2. The method for optimizing spectral index and band combination for hyperspectral image classification according to claim 1, characterized in that: When m is 3, the candidate spectrum index calculation formula is: Index1=(R a -R b ) / (R a +R c ) Index2=R a / (R b +R c ) Index3=(R a -R b ) / (R b -R c ) Index4=(R a -R b ) / R c Among them, Index1 is the first candidate spectral index, Index2 is the second candidate spectral index, Index3 is the third candidate spectral index, Index4 is the fourth candidate spectral index, R a is the spectral reflectance in band a, R b is the spectral reflectance in band b, R c is the spectral reflectance in band c.

3. The method for optimizing spectral index and band combination for hyperspectral image classification according to claim 2, characterized in that: The band a, band b and band c do not overlap with each other.

4. The method for optimizing spectral index and band combination for hyperspectral image classification according to claim 1, characterized in that: The default p value for the paired t-test was 0.

05.

5. The method for optimizing spectral index and band combination for hyperspectral image classification according to claim 1, characterized in that: Step 5 specifically includes: Perform a paired t test on the spectral index corresponding to the band combination. If the p value of the band combination is greater than a preset value once, skip the current band combination and perform a paired t test on the next band combination. If all p values ​​in the band combination are less than a preset value, the spectral index set corresponding to the band combination is used as a candidate spectral index.

6. A spectral index and band combination optimization system for hyperspectral image classification, used to implement any spectral index and band combination optimization method for hyperspectral image classification according to claims 1-5, characterized in that: include: A data acquisition module is used to acquire ground object hyperspectral data with s bands; A band combination module is used to generate a band combination list based on the hyperspectral data, wherein the band combination list includes C s m Band combinations; Band arrangement module, used to generate A for each band combination m m Band arrangement method; Spectral index calculation module, used to calculate the spectral index corresponding to each band arrangement; A paired t-test module is used to perform a paired t-test on the spectral index corresponding to each band combination, and select the spectral index corresponding to the p value obtained by the paired t-test that is less than a preset value, and use the corresponding spectral index as a candidate spectral index; The traversal control module is used to control the traversal process of the band combination, and sequentially executes the band arrangement generation, spectral index calculation and paired t test until all band combinations are tested; The output module is used to output all candidate spectral indexes, select the spectral index with the smallest p value as the optimal spectral index for land feature classification, and select the band combination corresponding to the spectral index with the smallest p value as the optimal band combination for land feature classification.

Images