Remote sensing image spatial-spectral fusion method, system and medium based on principal component analysis

Abstract

The present invention discloses a remote sensing image space-spectrum fusion method based on principal component analysis, wherein the mean filter images of the panchromatic band and the multispectral band are selected to form a vector sequence, the eigenvalues ​​and eigenvectors are calculated, the vector sequence is subjected to PCA transformation, a component with the largest correlation coefficient with the mean filter image of the panchromatic band is obtained as the first principal component component, the first principal component component is subjected to PCA inverse transformation, and the multispectral data replaces the transformed first principal component component, and at the same time, it is assumed that the feature extraction operator is not related to the multispectral, and the extreme value of the multispectral feature coefficient is solved, thereby obtaining the space-spectrum fusion expression of the mean filter image at high resolution. The present invention also discloses a space-spectrum fusion system, which organically integrates the spectral information of the multispectral band with the spatial information of the panchromatic band, so that the spectral information of the ground objects in the image is improved, and the geometric texture and spatial details of the panchromatic image are highly retained, effectively improving the visual resolution and computer analysis ability of the panchromatic image.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image space-spectrum information transformation and fusion, and relates to a remote sensing image space-spectrum fusion method based on principal component analysis; and also relates to a remote sensing image space-spectrum fusion system and medium based on principal component analysis. Background Art

[0002] With the development of science and technology, the research on multi-platform, multi-sensor, multi-weather, multi-temporal and multi-resolution remote sensing technology has been continuously deepened, and the types of remote sensing images that can be applied to all walks of life are becoming increasingly rich. Different remote sensing image data have different basic image characteristics such as spatial resolution, temporal resolution, and spectral resolution, and they have different application performances and application potentials in different application fields. Even for remote sensing images of the same sensor, when its spectral resolution is high, its spatial resolution is low, and when its spatial resolution is high, its spectral information is relatively simple. Traditional remote sensing image processing focuses on enhancing the general image features of a single image type, such as color, texture, and hierarchy; remote sensing image fusion processing focuses on integrating the basic features of different images into a new remote sensing image map, fully tapping its comprehensive application potential and improving its application performance.

[0003] The existing panchromatic image and multispectral image fusion methods are mainly divided into four categories: component replacement method, multi-resolution analysis method, variational model method, and depth-based image space-spectral fusion. There are two basic indicators to reflect the geometric clarity and spectral fidelity - the correlation coefficient between the fusion result and the original multispectral data, and the average gradient ratio between the fusion result and the original panchromatic data. It is generally believed that the higher the correlation coefficient with the multispectral, the better the spectral characteristic fidelity and the greater the information entropy; the higher the correlation coefficient between the fusion result and the panchromatic band, the greater the average gradient ratio, the higher the geometric clarity, and the better the image details are preserved. However, the spectral information fidelity of the fused multispectral image and the spatial detail information fidelity of the panchromatic image need to be further improved based on mathematical analysis. Summary of the invention

[0004] The first purpose of the present invention is to provide a remote sensing image spatial-spectral fusion method based on principal component analysis, which obtains intermediate bands based on the principle of principal component analysis and then performs image fusion, significantly improving the spectral information of multispectral images and the spatial detail information of high-resolution images; aiming to improve the stability and adaptability of spatial-spectral fusion.

[0005] The second object of the present invention is to provide a remote sensing image spatial-spectral fusion system based on principal component analysis.

[0006] The third object of the present invention is to provide a computer-readable storage medium for implementing a remote sensing image spatial-spectral fusion method based on principal component analysis.

[0007] The first technical solution adopted by the present invention is a remote sensing image spatial-spectral fusion method based on principal component analysis, and the specific operation steps are as follows:

[0008] Step 1: Obtain panchromatic band and multispectral band of remote sensing image; spatially register the panchromatic band and multispectral band images to make the geometric spatial position of the same object in the two images consistent;

[0009] Step 2: The panchromatic band and the multispectral band are combined into a vector sequence, and the mean μ, mean square error σ, eigenvalue of each band, as well as the covariance Cov, correlation coefficient r, and eigenvector between the bands are calculated;

[0010] Step 3: Calculate the mean filter images of the panchromatic band and the multispectral band, construct the default intermediate band of PCA and perform normalization processing to obtain a component with the largest correlation coefficient with the panchromatic band; and construct the intermediate band I according to the weight coefficients of different bands in the component with the largest correlation coefficient with the panchromatic band;

[0011] Step 4: Perform inverse PCA transformation on the maximum component of the correlation coefficient, obtain the band coefficients separately based on the scale invariance assumption, and perform spatial-spectral fusion with the panchromatic band and intermediate band I according to the multispectral band to be fused; calculate the average value and mean square error after fusion, perform histogram matching according to the histogram of the multispectral band, and obtain the spatial-spectral fusion result.

[0012] The present invention is also characterized in that:

[0013] The specific calculation method of mean μ, mean square error σ, covariance Cov, and correlation coefficient r in step 2 is as follows:

[0014]

[0015]

[0016] Where i = 1, 2, ..., n, j = 1, 2, ..., n, n is the number of multispectral bands; R and C are the number of rows and columns involved in the fusion image; p and q are the row and column numbers of the band image pixels; μ i , μ P are the means of the multispectral band and the panchromatic band respectively; σ i , σ P are the mean square error of the multispectral band and the panchromatic band respectively; Cov(M i , M j )、Cov(M i , P) are the covariance between multispectral bands and the covariance between multispectral bands and panchromatic band respectively; r(M i , M j )、r(M i, P) are the correlation coefficients between multispectral bands and the correlation coefficients between multispectral bands and panchromatic bands, respectively.

[0017] The covariance Cov matrix C and eigenvector matrix D described in step 2 are as follows:

[0018] The covariance Cov matrix C is:

[0019]

[0020] In the formula, B n represents the nth multispectral band;

[0021] The eigenvector matrix D is:

[0022]

[0023] a 00 The eigenvector representing the panchromatic band, a 0n Represents the feature vector of the panchromatic band and the nth multispectral band, a n1 Represents the feature vector of the nth multispectral band and the first multispectral band; a nn The feature vector representing the nth multispectral band.

[0024] Step 3 is as follows:

[0025] Calculate the panchromatic band P and multispectral band B i Mean filtered image of :

[0026] Calculate the panchromatic band P and multispectral band B i Mean filtered image of :

[0027] B0′=P-μ P (11)

[0028] B i ′=B i -μi (12)

[0029] Component calculation in image vector:

[0030]

[0031] According to the principle of principal component analysis, the first principal component carries most of the information of the image sequence, and the first principal component is:

[0032]

[0033] Normalize the weight coefficient of the first principal component PC0 to get:

[0034]

[0035] Among them, I is the intermediate band, I′ is the mean filtered image of the intermediate band, μ I is the mean value of I;

[0036] Calculate the mean filtered image correlation coefficient between the intermediate band and the panchromatic band:

[0037]

[0038] Among them, Cov(I, P) is the covariance between the intermediate band and the panchromatic band, σ I is the mean square error of the intermediate band, σ P is the mean square error of the panchromatic band.

[0039] From the inverse PCA transformation, we can get:

[0040]

[0041] In the formula, a ji is a ij The inverse transform of PCj is the inverse transform of PCi;

[0042] For i = 0, the inverse transform P′ of the panchromatic band P is:

[0043]

[0044] Depend on:

[0045]

[0046] We can get:

[0047]

[0048] Will:

[0049]

[0050] Using multispectral data j ′ replaces the intermediate band I, and P′ and B j ′The fused image Pf is:

[0051]

[0052] Right now:

[0053]

[0054] Among them, σ P , σ j , σ I P and B respectively j , the mean square error of I;

[0055] like Independent of the panchromatic band, the redundancy between multispectral features and panchromatic is minimal, the correlation coefficient between Pf and multispectral is the largest, and the multispectral preservation effect of the fusion results is better;

[0056] make Make it linearly independent of the panchromatic band P, and from r(P, E) = 0, we can get the new feature extraction coefficient:

[0057]

[0058] make Keep it unchanged, so that the correlation coefficient between the fusion result and the multi-spectral band is maximized, and the new fusion coefficient k can be obtained. Ep :

[0059]

[0060] The spatial-spectral fusion result is:

[0061]

[0062] The second technical solution adopted by the present invention is a remote sensing image space-spectrum fusion system based on principal component analysis, comprising:

[0063] Data acquisition module, used to collect panchromatic band and multispectral band data of remote sensing images;

[0064] Data processing module: used to form a vector sequence of panchromatic bands and multispectral bands, count the mean, mean square error, eigenvalue of each band, as well as the covariance, correlation coefficient, and eigenvector between bands, and calculate the mean filtered images of panchromatic bands and multispectral bands;

[0065] Principal component analysis module: used to construct the default intermediate band of principal component analysis PCA and perform normalization processing to obtain a component with the largest correlation coefficient with the panchromatic band; and construct the intermediate band according to the weight coefficients of different bands in the component with the largest correlation coefficient with the panchromatic band, and perform PCA inverse transformation on the component with the largest correlation coefficient;

[0066] Fusion module: Based on the multispectral band to be fused, spatial-spectral fusion is performed with the panchromatic band and the intermediate band constructed by the principal component analysis module; the average value and mean square error after fusion are statistically calculated, and histogram matching is performed according to the histogram of the multispectral band to obtain the spatial-spectral fusion result.

[0067] The third technical solution adopted by the present invention is a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a remote sensing image spatial-spectral fusion method based on principal component analysis is implemented.

[0068] The beneficial effects of the present invention are:

[0069] 1. The present invention uses a mean-filtered image composition vector sequence to construct a principal component analysis PCA transformation formula of multi-spectral and panchromatic bands under multi-resolution conditions, obtains an intermediate band, and realizes the fusion of low-resolution multi-spectral bands and high-resolution panchromatic bands, which not only enhances the multi-spectral information before and after the fusion of the panchromatic remote sensing image, but also maintains the details of the spatial information before and after the fusion. Compared with the original image, the spectral information of the objects in the fused image is greatly enriched. And compared with the original panchromatic image, in the fused panchromatic image, the geometric details, textures, edges, layers and other spatial information of objects such as water bodies, vegetation, exposed surfaces, buildings, etc. remain stable. This fusion algorithm belongs to a linear fusion model with clear physical meaning, concise and clear expressions, and fast and efficient calculations.

[0070] 2. The present invention has no limit on the number of bands of the fused target image, which can be greater than or equal to 1, that is, a multispectral image of one band can be fused, or fusion target images of 2 bands, 3 bands, 4 bands or even more bands can be processed. According to actual needs, the fusion target images of a certain number of bands can be fused to obtain the image fusion results of the corresponding number of bands. The algorithm itself has no limit on the number of bands of the fused target image and is open.

[0071] 3. Based on PCA transformation, the present invention uses multispectral images and panchromatic band images to construct intermediate bands. By evaluating the correlation coefficient between the intermediate bands and the panchromatic bands, the best intermediate bands are selected to achieve scheme selection driven by the fusion target. PCA multispectral injection panchromatic fusion is performed according to the selected intermediate bands to obtain the relatively optimal fusion result.

[0072] The effectiveness of the optimal solution was verified by visual judgment. Compared with the original image, the spatial information of the geometric details, texture, edge and layer of the objects in the fused image has been greatly enriched. Compared with the original multispectral image, the spectral characteristics and color display of objects such as water bodies, vegetation, exposed surfaces and buildings in the fused multispectral image remain basically stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a technical flow chart of the present invention;

[0074] Figure 2 It is a full-color band image at a resolution of 2 meters in Example 3 of the present invention;

[0075] FIG3( a ) is a blue band (B) image at 8-meter resolution in Example 3 of the present invention;

[0076] FIG3( b ) is a green band (G) image at a resolution of 8 meters in Example 3 of the present invention;

[0077] FIG3( c ) is a red band (R) image at a resolution of 8 meters in Example 3 of the present invention;

[0078] FIG3( d ) is an image of the near infrared band (N) at a resolution of 8 meters in Example 3 of the present invention;

[0079] Figure 4 It is the intermediate band I image at a resolution of 2 meters in Example 3 of the present invention;

[0080] FIG5(a) is a near-infrared (N') image after fusion at a resolution of 2 meters in Example 3 of the present invention;

[0081] FIG5( b ) is a red band (R′) image after fusion at a resolution of 2 meters in Example 3 of the present invention;

[0082] FIG5( c ) is a green band (G′) image after fusion at a resolution of 2 meters in Example 3 of the present invention;

[0083] FIG5( d ) is a blue band (B′) image after fusion at a resolution of 2 meters in Example 3 of the present invention;

[0084] FIG6( a ) is a true color composite image of multi-spectral R'G'B' bands after fusion at a resolution of 2 meters in Example 3 of the present invention;

[0085] FIG6( b ) is a standard false color composite image of the multi-spectral N'R'G' bands after fusion at a resolution of 2 meters in Example 3 of the present invention; DETAILED DESCRIPTION

[0086] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0087] Example 1

[0088] Step 1: Obtain panchromatic band and multispectral band of remote sensing image; spatially register the panchromatic band and multispectral band images to make the geometric spatial position of the same object in the two images consistent;

[0089] Step 2: The panchromatic band and the multispectral band are combined into a vector sequence, and the mean μ, mean square error σ, eigenvalue of each band, as well as the covariance Cov, correlation coefficient r, and eigenvector between the bands are calculated;

[0090] The specific calculation method of mean μ, mean square error σ, covariance Cov, and correlation coefficient r in step 2 is as follows:

[0091]

[0092]

[0093] Where i = 1, 2, ..., n, j = 1, 2, ..., n, n is the number of multispectral bands; R and C are the number of rows and columns involved in the fusion image; p and q are the row and column numbers of the band image pixels; μ i , μ P are the means of the multispectral band and the panchromatic band respectively; σ i , σ P are the mean square error of the multispectral band and the panchromatic band respectively; Cov(M i , M j )、Cov(M i , P) are the covariance between multispectral bands and the covariance between multispectral bands and panchromatic band respectively; r(M i , M j )、r(M i , P) are the correlation coefficients between multispectral bands and the correlation coefficients between multispectral bands and panchromatic bands, respectively.

[0094] The covariance Cov matrix C is:

[0095]

[0096] In the formula, B n represents the nth multispectral band;

[0097] The eigenvector matrix D is:

[0098]

[0099] a 00 The eigenvector representing the panchromatic band, a 0n Represents the feature vector of the panchromatic band and the nth multispectral band, a n1 Represents the feature vector of the nth multispectral band and the first multispectral band; a nn Represents the feature vector of the nth multispectral band.

[0100] Step 3: Calculate the mean filtered images of the panchromatic band and the multispectral band, construct the default intermediate band of PCA and perform normalization processing to obtain a component with the largest correlation coefficient with the panchromatic band; and construct the intermediate band I according to the weight coefficients of different bands in the component with the largest correlation coefficient with the panchromatic band; the details are as follows:

[0101] Calculate the panchromatic band P and multispectral band B i Mean filtered image of :

[0102] B0′=P-μ P (11)

[0103] B i ′=B i -μi (12)

[0104] Component calculation in image vector:

[0105]

[0106] According to the principle of principal component analysis, the first principal component carries most of the information of the image sequence, and the first principal component is:

[0107]

[0108] Normalize the weight coefficient of the first principal component PC0 to get:

[0109]

[0110] Among them, I is the intermediate band, I′ is the mean filtered image of the intermediate band, μ I is the mean value of I, B j is the default intermediate band; B j ′ is the mean-filtered image of the default intermediate band;

[0111] Calculate the correlation coefficient with the panchromatic band:

[0112]

[0113] Among them, Cov(I, P) is the covariance between the intermediate band and the panchromatic band, σ I is the mean square error of the intermediate band, σ P is the mean square error of the panchromatic band.

[0114] Step 4: Perform inverse PCA transformation on the maximum component of the correlation coefficient, obtain the band coefficients separately based on the scale invariance assumption, and perform spatial-spectral fusion with the panchromatic band and intermediate band I according to the multispectral band to be fused; calculate the average value and mean square error after fusion, perform histogram matching according to the histogram of the multispectral band, and obtain the spatial-spectral fusion result.

[0115] From the inverse PCA transformation, we can get:

[0116]

[0117] In the formula, a ji is a ij The inverse transform of PCj is the inverse transform of PCi, B j 'yes;

[0118] For i = 0, the inverse transform P′ of the panchromatic band P is:

[0119]

[0120] Depend on:

[0121]

[0122] We can get:

[0123]

[0124] Will:

[0125]

[0126] Using multispectral data j ′ replaces the intermediate band I, and P′ and B j ′The fused image Pf is:

[0127]

[0128] Right now:

[0129]

[0130] Among them, σ P , σ j , σ I P and B respectively j , the mean square error of I;

[0131] like Independent of the panchromatic band, the redundancy between multispectral features and panchromatic is minimal, the correlation coefficient between Pf and multispectral is the largest, and the multispectral preservation effect of the fusion results is better;

[0132] make Make it linearly independent of the panchromatic band P, and from r(P, E) = 0, we can get the new feature extraction coefficient:

[0133]

[0134] make Keep it unchanged, so that the correlation coefficient between the fusion result and the multi-spectral band is maximized, and the new fusion coefficient k can be obtained. Ep :

[0135]

[0136] The spatial-spectral fusion result is:

[0137]

[0138] After fusion processing, the spectral resolution of the panchromatic image is significantly improved. The richness of spectral information has comprehensively improved the color, edge clarity and layering of the objects in the image. At the same time, the fused image can not only maintain the richness of remote sensing image information and the overall quality of the image, but also enrich the color of the objects and improve the spectral characteristics of the objects in the remote sensing image.

[0139] Panchromatic remote sensing images have high spatial resolution and can clearly display the texture and spatial details of objects such as water bodies, vegetation, buildings, exposed rocks, and exposed soil. Panchromatic remote sensing images have low spectral resolution and lack spectral information of objects, which makes it impossible to accurately distinguish objects during remote sensing analysis, affecting the application effect of panchromatic images.

[0140] This method organically integrates the spectral information of multispectral bands with the spatial information of high spatial resolution panchromatic bands, thereby improving the spectral information of objects in the image, while highly retaining the geometric texture and spatial details of the panchromatic image, without changing the image spatial characteristics and display effects of typical objects such as water bodies, vegetation, soil, and buildings, effectively improving the visual resolution and computer analysis capabilities of panchromatic images, and improving the application effects and capabilities of remote sensing image mapping, classification, etc.

[0141] Example 2

[0142] The remote sensing image space-spectrum fusion system based on principal component analysis of the present invention comprises:

[0143] Data acquisition module, used to collect panchromatic band and multispectral band data of remote sensing images;

[0144] Data processing module: used to form a vector sequence of panchromatic bands and multispectral bands, count the mean, mean square error, eigenvalue of each band, as well as the covariance, correlation coefficient, and eigenvector between bands, and calculate the mean filtered images of panchromatic bands and multispectral bands;

[0145] Principal component analysis module: used to construct the default intermediate band of principal component analysis PCA and perform normalization processing to obtain a component with the largest correlation coefficient with the panchromatic band; and construct the intermediate band according to the weight coefficients of different bands in the component with the largest correlation coefficient with the panchromatic band, and perform PCA inverse transformation on the component with the largest correlation coefficient;

[0146] Fusion module: Based on the multispectral band to be fused, spatial-spectral fusion is performed with the panchromatic band and the intermediate band constructed by the principal component analysis module; the average value and mean square error after fusion are statistically calculated, and histogram matching is performed according to the histogram of the multispectral band to obtain the spatial-spectral fusion result.

[0147] Example 3

[0148] The present invention discloses a remote sensing image space-spectrum fusion method based on principal component analysis, and the specific operation steps are as follows:

[0149] Step 1. Input remote sensing image

[0150] Open a Gaofen-6 (GF6) remote sensing image of Foshan City, Guangdong Province on January 18, 2021 with panchromatic band (P), blue band (B), green band (G), red band (R), and near infrared (N) (the multispectral image is taken from GF6_PMS_E112_8_N23_2_20210118_L1A1120072629_ORTHO_MS, and the panchromatic image is taken from GF6_PMS_E112_8_N23_2_20210118_L1A1120072629_ORTHO_PAN). Figure 2 is a full-color band map (2-meter resolution), Figure 3(a)-Figure 3(d) This is a multispectral band map (8-meter resolution).

[0151] Step 2. Use the remote sensing image processing software ENVI to form a vector sequence of the panchromatic band P and the multispectral bands B, G, R, and N, and then calculate the image characteristic statistical parameters such as the mean μ and standard deviation σ of each band image. The basic characteristic statistical parameters of each band image are shown in Tables 1 to 4.

[0152] Table 1 Basic characteristic statistical parameters of each band image

[0153] Image bands Mean / μ Standard deviation / σ Panchromatic Band P 797.381014 325.640736 Near infrared 1181.74205 596.326554 Red Band R 756.027566 380.18231 Green Band G 801.835808 272.455739 Blue Band B 740.774641 179.282284

[0154] Table 2 Covariance table of each band image

[0155] Covariance Panchromatic Band P Near infrared Red Band R Green Band G Blue Band B Panchromatic Band P 106041.889 167949.771 109902.375 73866.275 47134.349 Near infrared 167949.771 355605.359 154029.965 93786.627 59174.840 Red Band R 109902.375 154029.965 144538.589 101460.222 65046.199 Green Band G 73866.275 93786.627 101460.222 74232.130 48107.706 Blue Band B 47134.349 59174.840 65046.199 48107.706 32142.137

[0156] Table 3 Correlation coefficients of images in each band

[0157] Correlation coefficient Panchromatic Band P Near infrared Red Band R Green Band G Blue Band B Panchromatic Band P 1 0.864881 0.887721 0.832552 0.807349 Near infrared 0.864881 1 0.679406 0.577246 0.553497 Red Band R 0.887721 0.679406 1 0.979508 0.954317 Green Band G 0.832552 0.832552 0.979508 1 0.984875 Blue Band B 0.807349 0.553497 0.954317 0.984875 1

[0158] Table 4 Table of image feature vectors for each band

[0159]

[0160]

[0161] Step 3. Calculate the mean filter images of the panchromatic band and the multispectral band, construct the default intermediate band of the principal component analysis PCA and perform normalization processing to obtain a component with the largest correlation coefficient with the panchromatic band; and construct the intermediate band I according to the weight coefficients of different bands in the component with the largest correlation coefficient with the panchromatic band;

[0162] Table 5 Intermediate band optimization table

[0163] PCA default mediation r(P,I) PCA default mediation r(P, I) PCA default mediation r(P, I) NRGB 0.95680418 PNRB 0.95699353 NRB 0.95695653 PNRGB 0.95597913 PNGB 0.94289561 NGB 0.95416985 PNRG 0.95646891 NRG 0.95713853 RGB 0.93660392

[0164] Step 4. Use the panchromatic band and the four multispectral band images to construct the intermediate band I. The band operation expression is I = (0*(b1-797.381014)+0.334163961332373*(b2-1181.742054)+0.281587540270588*(b3-756.027566)+0.213670432444854*(b4-801.835808)+0.170578065952185*(b5-740.774641)), where b1, b2, b3, b4, and b5 are the panchromatic band (P), near infrared (N), red band (R), green band (G), and blue band (B), respectively. Calculate the intermediate band I involved in the fusion target image as follows Figure 4 .

[0165] Step 5. Reconstruct the fused multispectral image

[0166] ① Reconstruct the fused near-infrared band (N') image. The calculation expression is (596.3266 / 363.399361)*((b1-797.381014)+1.17340463770544*(0.546077872628157*(b2-1181.742054)-0.915267459712135*0.903611482924039*b3))+1181.742, where b1 is the panchromatic band P, b2 is the near-infrared (N), and b3 is the intermediate band I. The fused near-infrared (N') image (2-meter resolution) is calculated and shown in Figure 5(a).

[0167] ② Reconstruct the fused red band (R') image. The calculation expression is (b1-797.381014)+1.09974919390883*(0.856538369710048*(b2-756.027566)-0.915267459712135*0.92747379348402*b3))+756.0276, where b1 is the panchromatic band P, b2 is the red band (R), and b3 is the intermediate band I. The fused red band (R') image (2-meter resolution) is calculated and shown in Figure 5(b).

[0168] ③ Reconstruct the fused green band (G') image. The calculation expression is (272.4557 / 378.527121)*((b1-797.381014)+1.1863862031307*(1.1952060073875*(b2-801.835808)-0.915267459712135*0.869834662587184*b3))+801.8358, where b1 is the panchromatic band P, b2 is the green band (G), and b3 is the intermediate band I. The fused green band (G') image (2-meter resolution) is calculated and shown in Figure 5(c).

[0169] ④ Reconstruct the fused blue band (B') image. The calculation expression is (179.2823 / 391.123980)*((b1-797.381014)+1.24137120866249*(1.81635758277154*(b2-740.774641)-0.915267459712135*0.843502962545513*b3))+740.7746, where b1 is the panchromatic band P, b2 is the blue band (B), and b3 is the intermediate band I. The fused blue band (B') image (2-meter resolution) is calculated and shown in Figure 5(d).

[0170] Step 6. Synthesize the reconstructed color image

[0171] The true color image synthesized by the fused red band (R'), green band (G') and blue band (B') according to the red, green and blue channels is shown in Figure 6(a); the standard false color image synthesized by the fused near-infrared band (N'), red band (R') and green band (G') according to the red, green and blue channels is shown in Figure 6(b).

[0172] In this case, band data statistical analysis was performed on the original GF6 multispectral image, the multispectral image fused by the method of the present invention, and the multispectral image fused by the ENVIPCA method (traditional PCA method fusion), and the comparison of the image band statistical characteristic parameters is shown in Table 6. It can be seen from the data in the table that compared with the original multispectral image, the spatial resolution of the multispectral image fused by the method of the present invention is increased from the original 8 meters to 2 meters, the spatial accuracy is greatly improved, and the spatial texture and geometric detail information of the ground object are significantly enhanced; the fused image has both the original multispectral and panchromatic image information, and the information consistency index is 0.922; the information entropy of the multispectral band is slightly reduced, but the gradient information of the band is greatly enhanced, the average gradient of the true color image is enhanced from 81.53 to 212.107, and the average gradient of the standard false color image is enhanced from 141.921 to 327.233. The spatial information of the ground object in the fused multispectral image is richer, which is conducive to the development of various image applications such as ground object recognition, interpretation, and analysis. It can be seen from the data in the table that, compared with the ENVIPCA method, the correlation coefficient between the multispectral images fused by the method of the present invention is lower, the information redundancy is smaller, the information entropy index data and gradient index data of the true color and standard false color combined images are higher, and the details and texture features of the objects are better expressed.

[0173] The method, system, medium and device for panchromatic spatial-spectral fusion based on Principal Component Analysis transformation and multispectral injection described in the present invention are mainly aimed at remote sensing images with panchromatic bands and multispectral bands such as near infrared, red, green and blue. First, the panchromatic band and the multispectral band image are spatially registered to make the geometric spatial position of the same object in the above two images consistent; then, the multispectral band image and the panchromatic band image are used to construct a simulated low-resolution panchromatic image, and as the intermediate band I of the fusion, the panchromatic band P, the intermediate band I and the multispectral band M are combined. iAccording to the high spatial resolution image resampling, the above three images are synthesized into one file, and the mean μ, mean square error σ and correlation coefficient r matrix of each band involved in the fusion image are calculated; finally, the fusion target band and fusion source band are selected according to the fusion scheme, and the intermediate band I is spatially-spectrally fused one by one to realize the injection of spectral characteristics and colors of multispectral bands into panchromatic bands, and at the same time improve the spatial geometric information of multispectral bands. This method is suitable for fusing panchromatic bands with multispectral bands. Specifically, the multispectral band is used as the fusion source, and each multispectral band is injected into the panchromatic band in turn, so as to realize the multispectral injection panchromatic spatial-spectral fusion based on PCA transformation, so as to enhance the spatial geometry, texture, edge, layer and other information of the objects in the multispectral image, improve its image clarity and spatial resolution, and at the same time, the multispectral image fusion results can highly maintain the stability of the spectral characteristics and color display of various objects in the original multispectral image. This technical method has a solid theoretical basis, clear physical meaning, wide application objects, simple operation and high computational efficiency. The fused images have bright colors, rich information, stable spectral information, and are easy to visually inspect and automatically classify. Especially in the context of the rapid development of high-resolution satellite remote sensing, they play a huge role in promoting the popularization and application of domestic high-resolution images in various industries at home and abroad.

[0174] Table 6 Comparison of band statistical characteristic parameters of original multispectral images, fused images of the present invention, and ENVIPCA fused images

[0175]

Claims

1. A remote sensing image spatial-spectral fusion method based on principal component analysis, characterized in that: The details are as follows: Step 1: Obtain panchromatic band and multispectral band of remote sensing image; spatially register the panchromatic band and multispectral band images to make the geometric spatial position of the same object in the two images consistent; Step 2: The panchromatic band and the multispectral band are combined into a vector sequence, and the mean μ, mean square error σ, eigenvalue of each band, as well as the covariance Cov, correlation coefficient r, and eigenvector between the bands are counted; Step 3: Calculate the mean filter images of the panchromatic band and the multispectral band, construct the default intermediate band of PCA and perform normalization processing to obtain a component with the largest correlation coefficient with the panchromatic band; and construct the intermediate band I according to the weight coefficients of different bands in the component with the largest correlation coefficient with the panchromatic band; Step 4: Perform inverse PCA transformation on the maximum component of the correlation coefficient, obtain the band coefficients separately based on the scale invariance assumption, and perform spatial-spectral fusion with the panchromatic band and intermediate band I according to the multispectral band to be fused; calculate the average value and mean square error after fusion, perform histogram matching according to the histogram of the multispectral band, and obtain the spatial-spectral fusion result.

2. The remote sensing image spatial-spectral fusion method based on principal component analysis according to claim 1, characterized in that: The specific calculation method of mean μ, mean square error σ, covariance Cov, and correlation coefficient r in step 2 is as follows: Where i = 1, 2, ..., n, j = 1, 2, ..., n, n is the number of multispectral bands; R and C are the number of rows and columns involved in the fusion image; p and q are the row and column numbers of the band image pixels; μ i , μ P are the means of the multispectral band and the panchromatic band respectively; σ i , σ P are the mean square error of the multispectral band and the panchromatic band respectively; Cov(M i , M j )、Cov(M i , P) are the covariance between multispectral bands and the covariance between multispectral bands and panchromatic band respectively; r(M i , M j )、r(M i , P) are the correlation coefficients between multispectral bands and the correlation coefficients between multispectral bands and panchromatic bands, respectively.

3. The remote sensing image space-spectrum fusion method based on principal component analysis according to claim 2 is characterized in that: The covariance Cov matrix C and eigenvector matrix D described in step 2 are as follows: The covariance Cov matrix C is: In the formula, B n represents the nth multispectral band; The eigenvector matrix D is: a 00 The eigenvector representing the panchromatic band, a 0n Represents the feature vector of the panchromatic band and the nth multispectral band, a n1 Represents the feature vector of the nth multispectral band and the first multispectral band; a nn Represents the feature vector of the nth multispectral band.

4. The remote sensing image space-spectrum fusion method based on principal component analysis according to claim 3 is characterized in that: Step 3 is as follows: Calculate the panchromatic band P and multispectral band B i Mean filtered image of : B0′=P-μ P (11) B i ′=B i -m i (12) Component calculation in image vector: B0′ represents the mean filtered image of the full-color band, B j ′ represents the mean filtered image of the jth multispectral band, j = 1, 2, ..., n; According to the principle of principal component analysis, the first principal component carries most of the information of the image sequence, and the first principal component is: Normalize the weight coefficient of the first principal component PC0 to get: Among them, I is the intermediate band, I′ is the mean filtered image of the intermediate band, μ I is the mean value of I, B j is the default intermediate band; B j ′ is the mean-filtered image of the default intermediate band; Calculate the mean filtered image correlation coefficient between the intermediate band and the panchromatic band: Among them, Cov(I, P) is the covariance between the intermediate band and the panchromatic band, σ I is the mean square error of the intermediate band, σ P is the mean square error of the panchromatic band.

5. The remote sensing image space-spectrum fusion method based on principal component analysis according to claim 4 is characterized in that: Step 4 is as follows: From the inverse PCA transformation, we can get: In the formula, a ji is a ij The inverse transform of PCj is the inverse transform of PCi; For i = 0, the inverse transform P′ of the panchromatic band P is: Depend on: We can get: Will: Substituting into the above formula, we get: Using multispectral data j ′ replaces the intermediate band I, and P′ and B j ′The fused image Pf is: Right now: Among them, σ P , σ j , σ I P and B respectively j , the mean square error of I; like Independent of the panchromatic band, the redundancy between multispectral features and panchromatic is minimal, the correlation coefficient between Pf and multispectral is the largest, and the multispectral preservation effect of the fusion results is better; make Make it linearly independent of the panchromatic band P, and from r(P,E)=0, we can get the new feature extraction coefficient: make Keep it unchanged, so that the correlation coefficient between the fusion result and the multi-spectral band is maximized, and the new fusion coefficient k can be obtained. Ep : The spatial-spectral fusion result is:

6. Remote sensing image spatial-spectral fusion system based on principal component analysis, characterized in that: include: Data acquisition module, used to collect panchromatic band and multispectral band data of remote sensing images; Data processing module: used to form a vector sequence of panchromatic bands and multispectral bands, count the mean, mean square error, eigenvalue of each band, as well as the covariance, correlation coefficient, and eigenvector between bands, and calculate the mean filtered images of panchromatic bands and multispectral bands; Principal component analysis module: used to construct the default intermediate band of principal component analysis PCA and perform normalization processing to obtain a component with the largest correlation coefficient with the panchromatic band; and construct the intermediate band according to the weight coefficients of different bands in the component with the largest correlation coefficient with the panchromatic band, and perform PCA inverse transformation on the component with the largest correlation coefficient; Fusion module: Based on the multispectral band to be fused, spatial-spectral fusion is performed with the panchromatic band and the intermediate band constructed by the principal component analysis module; the average value and mean square error after fusion are statistically calculated, and histogram matching is performed according to the histogram of the multispectral band to obtain the spatial-spectral fusion result.

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 remote sensing image spatial-spectral fusion method according to any one of claims 1 to 5 is implemented.

Images