High-precision extraction method and system of vegetation in multispectral remote sensing images based on cluster analysis index

Abstract

The present invention belongs to remote sensing image data processing technology, and is a method and system for high-precision extraction of vegetation from multispectral remote sensing images based on cluster analysis index. The method includes: inputting the near-infrared band and red light band of the multispectral image; matching the corresponding feature points of the near-infrared band image and the red light band image to make the geometric spatial position of the same object consistent; calculating the image feature parameters of the near-infrared band and the red light band; deriving and constructing an equilibrium transformation function, and matching and optimizing the red light band and the near-infrared band, integrating and outputting the vegetation analysis intermediate band; calculating the cluster analysis index pixel by pixel based on the information of the vegetation analysis intermediate band; defining the cluster analysis index threshold, and extracting the vegetation information of the object according to the relationship between the cluster analysis index value of each pixel and the threshold. The present invention integrates and outputs the vegetation information contained in the two bands for complementary optimization, thereby improving the vegetation extraction accuracy of multispectral remote sensing images.

Description

Technical Field

[0001] The invention belongs to remote sensing image data processing technology, and specifically relates to a multi-spectral remote sensing image ground object vegetation high-precision extraction method and system based on Cluster Analysis Index (CAI). Background Art

[0002] In recent years, remote sensing and related technologies have developed rapidly. However, from the perspective of convenience, stability, and economy of data acquisition, medium and high-resolution multispectral remote sensing images such as the Landsat series, Sentinel series, and GF-series are still one of the most commonly used data types in scientific research and production practice. Multispectral remote sensing images contain spectral information in multiple bands. Different bands have different reflection and absorption characteristics for ground objects. Therefore, different band combinations of multispectral images can be used to identify and analyze different types of ground objects. For example, due to the influence of chlorophyll in the leaves, vegetation strongly absorbs the visible blue light band (450nm) and the red light band (650nm), and strongly reflects green light (540nm); due to the complex leaf cavity structure, the reflectivity of vegetation in the near-infrared band (700-800nm) rises sharply, such as Figure 1 The unique spectral characteristics of green vegetation in the red and near-infrared bands make these two bands commonly used in vegetation identification, analysis and other related research.

[0003] According to the spectral characteristics of vegetation, scholars have combined and calculated different bands of multispectral remote sensing images for qualitative and quantitative evaluation of vegetation coverage and vegetation growth. These different combinations and calculation methods are collectively referred to as vegetation indices. Currently, the commonly used vegetation indices (see Table 1) are mostly based on the combination of different bands or the introduction of parameters to construct equations. These indices are either based on experience and lack of theoretical support for band calculation combinations, or introduce parameters without clear physical meanings but it is difficult to obtain specific values ​​in actual application.

[0004] Table 1 List of commonly used vegetation indices

[0005]

[0006]

[0007] In order to better highlight the useful information in the image and improve the interpretation and recognition effect, the pixel value, grayscale distribution, frequency domain information, etc. of the image can be adjusted and processed to achieve the purpose of image enhancement. Image enhancement methods can be divided into two categories: spatial domain image enhancement and transform domain image enhancement. Spatial domain image enhancement is to directly improve the pixels, hue, and contrast of the image itself, such as grayscale transformation enhancement method, histogram enhancement method, local average method, etc.; transform domain image enhancement is to transform and filter the image to optimize edge details, such as low-pass filtering method, wavelet transform method, Fourier transform method, etc.

[0008] After enhancing the red light band and near-infrared band of the multispectral image using image enhancement methods, a cluster analysis index is constructed based on the vegetation index calculation principle. This can further highlight the characteristics of the vegetation spectral curve and help distinguish vegetation from other landforms. However, there are currently no relevant methods proposed and application examples in the field of vegetation classification and extraction from multispectral remote sensing images.

[0009] The currently commonly used method for extracting vegetation from multispectral remote sensing images based on traditional vegetation indices (i.e., traditional methods) has the following problems: (1) the accuracy is generally low and is easily affected by interference factors; (2) the physical meaning of the parameters is unclear and relies on statistical values ​​or empirical values; (3) it is difficult to form a standardized technology and is not suitable for large-scale promotion and application. Summary of the invention

[0010] In order to solve the technical problems existing in the prior art, the present invention provides a method and system for high-precision extraction of vegetation from multispectral remote sensing images based on cluster analysis index, which cleverly combines the vegetation index calculation principle and image enhancement method, integrates and outputs the red light band and near-infrared band that can concentrate on expressing the unique spectral characteristics of vegetation, makes the vegetation information contained in the two bands complementary and optimized, and then calculates and obtains the cluster analysis index results, thereby improving the vegetation extraction accuracy of multispectral remote sensing images.

[0011] A high-precision method for extracting vegetation from multispectral remote sensing images based on cluster analysis index proposed in an embodiment of the present invention includes the following steps:

[0012] S1, near infrared band and red light band of input multispectral image;

[0013] S2, matching the corresponding feature points of the near-infrared band image and the red light band image, so that the geometrical spatial position of the same ground object in the near-infrared band image and the red light band image is consistent;

[0014] S3, calculate and count the image characteristic parameters of near infrared band and red light band;

[0015] S4. Based on the principle of image enhancement and the principle of scale invariance, a balanced transformation function is derived and constructed; the red light band and the near infrared band are matched and optimized by using the balanced transformation function, and the intermediate band for vegetation analysis is integrated and output;

[0016] S5, based on the information of the intermediate bands of vegetation analysis, the cluster analysis index is calculated pixel by pixel;

[0017] S6. Determine a cluster analysis index threshold, and extract ground feature vegetation information based on the relationship between the cluster analysis index value of each pixel and the threshold.

[0018] In addition, a high-precision extraction system of ground objects and vegetation in multispectral remote sensing images based on cluster analysis index proposed in an embodiment of the present invention includes the following modules:

[0019] Input module, used to input near infrared band and red light band of multispectral image;

[0020] The matching module is used to match the corresponding feature points of the near-infrared band image and the red light band image, so that the geometric spatial position of the same object in the near-infrared band image and the red light band image is consistent;

[0021] The characteristic parameter calculation module is used to calculate and count the image characteristic parameters of the near-infrared band and the red light band;

[0022] The intermediate band output module is used to derive and construct a balanced transformation function based on the image enhancement principle and the scale invariance principle; the balanced transformation function is used to match and optimize the red light band and the near infrared band, and integrate and output the intermediate band for vegetation analysis;

[0023] The cluster analysis index calculation module is used to calculate the cluster analysis index pixel by pixel based on the information of the intermediate band of vegetation analysis;

[0024] The extraction module is used to define a cluster analysis index threshold value, and extract ground feature vegetation information according to the magnitude relationship between the cluster analysis index value of each pixel and the threshold value.

[0025] From the above technical scheme, it can be seen that the present invention integrates the principle of image enhancement and the principle of vegetation index calculation, and uses the balanced transformation function to integrate and output the red light band and the near infrared band containing rich vegetation information to obtain the vegetation analysis intermediate band hmβ; then the cluster analysis index is calculated based on the vegetation analysis intermediate band hmβ and the threshold is determined, which optimizes and enhances the vegetation spectral information and realizes the high-precision and high-efficiency extraction of vegetation in multi-spectral remote sensing images; finally, the overall accuracy is used as the basic indicator supplemented by visual interpretation to evaluate the vegetation extraction results. Compared with the prior art, the technical effects achieved by the present invention include:

[0026] 1. The physical meaning of each parameter is clear, the calculation is convenient, and the operability is strong.

[0027] The present invention can directly perform calculations on the red light band and near-infrared band images of multispectral remote sensing images without the need for a preprocessing process. In addition, the parameters in the algorithm have clear meanings, convenient calculations, strong operability, and wide practical application value.

[0028] 2. It only relies on the intrinsic characteristics of image data and has strong applicability.

[0029] The present invention is applicable to multispectral image data formed by different mounting platforms and various types of sensors, and only relies on the intrinsic characteristics of the spectral information of the image band. It has no special requirements on the sensor type and mounting platform, and has strong applicability.

[0030] 3. The optimization highlights the vegetation spectral information, and the results are accurate.

[0031] The present invention integrates the information of red light band and near infrared band, outputs the intermediate band hmβ for vegetation analysis, optimizes and enhances the spectral information of vegetation in multispectral images, is conducive to distinguishing vegetation from other ground objects, and improves the accuracy and efficiency of vegetation extraction in multispectral remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a vegetation spectrum characteristic curve diagram in the prior art;

[0033] Figure 2 is a flow chart of an extraction method in an embodiment of the present invention;

[0034] Figure 3 It is a true color image of 4, 3, and 2 band combination in the embodiment of the present invention;

[0035] Figure 4 It is a near infrared band image in the embodiment of the present invention;

[0036] Figure 5 This is a red light band image in an embodiment of the present invention;

[0037] Figure 6 It is a schematic diagram of spatial registration of band images in an embodiment of the present invention;

[0038] Figure 7 It is the intermediate band hmβ image diagram for vegetation analysis in the embodiment of the present invention;

[0039] Figure 8 This is a rendering of the vegetation extraction effect of the extraction method according to an embodiment of the present invention;

[0040] Fig. 9 To extract the vegetation effect map based on the traditional method. DETAILED DESCRIPTION

[0041] 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.

[0042] Example

[0043] This embodiment provides a high-precision method for extracting vegetation from multispectral remote sensing images based on cluster analysis index; Figure 2 As shown, the specific steps include:

[0044] S1. Input the near infrared band and red light band of the multispectral image. The near infrared band and red light band may be the near infrared band data and red light band data of the multispectral image of any type of carrying platform or sensor.

[0045] In this embodiment, a Landsat8Oli satellite remote sensing image having multiple bands such as a near infrared band (NIR) and a red light band (R) is opened, and the resolution of the near infrared band and the red light band are both 30 meters. Figure 3 It is a true color image composed of 4, 3, and 2 bands. Figure 4 This is a near-infrared image. Figure 5 This is a red light band image, in which the image stretching effect is set to Linear2%.

[0046] S2, matching the corresponding feature points of the near-infrared band image and the red light band image, so that the geometric spatial position of the same object in the near-infrared band image and the red light band image is consistent, so as to achieve the spatial registration of the near-infrared band image and the red light band image, such as Figure 6 shown.

[0047] S3. Calculate and count the image characteristic parameters such as the mean and standard deviation of the near-infrared band and the red light band.

[0048] In this embodiment, the calculation formula used is as follows:

[0049]

[0050] Among them, the remote sensing image is composed of pixels. Let B i is the near infrared band image or red light band image in the multispectral band, n represents the band image B i The nth row, l represents the band image B i The first column of B i (n, l) represents band image B i The pixel in the nth row and lth column of i The total number of rows and columns; μ i , σ i Band image B iThe mean and standard deviation of the near infrared band of the multispectral image are μ and μ respectively. NIR , σ NIR , the mean and standard deviation of the red light band are μ R , σ R .

[0051] The basic characteristic parameters of the images of each band obtained by calculation in this embodiment are statistically shown in Table 2.

[0052] Table 2 Statistics of basic characteristic parameters of images in each band

[0053] Image bands Mean / μ Standard deviation / σ Near Infrared NIR 10360.758728 7435.623113 Red Band R 5092.313953 3674.009464

[0054] S4. Based on the principles of image enhancement and scale invariance, a balanced transformation function is derived and constructed; the balanced transformation function is used to match and optimize the red light band and near infrared band of the multispectral image, and the intermediate band hmβ of vegetation analysis is integrated and output. The image of the intermediate band (30-meter resolution) is shown in Figure 7 .

[0055] In this embodiment, the expression of the equalization transformation function is:

[0056]

[0057] Among them, σ R represents the standard deviation of the red light band, σ NIR represents the standard deviation of the near infrared band, μ NIR represents the mean value of the near-infrared band, μ R Represents the mean value of the red light band; NIR(n,l) represents the near-infrared band reflectance of the pixel in the nth row and the lth column; hmβ(n,l) represents the value of the nth row and the lth column of the output intermediate band hmβ.

[0058] S5. Based on the information of the intermediate band hmβ of vegetation analysis, the cluster analysis index CAI is calculated pixel by pixel, and its expression is:

[0059]

[0060] Among them, R(n,l) represents the red light band reflectance of the pixel in the nth row and lth column, and CAI(n,l) represents the value of the output clustering analysis index CAI in the nth row and lth column.

[0061] S6. Determine the cluster analysis index threshold, and extract the ground vegetation information according to the relationship between the cluster analysis index CAI value of each pixel and the threshold t.

[0062] In this step, the threshold is set according to the distribution of CAI values ​​of vegetation and non-vegetation in the sample points, which is specifically expressed as follows:

[0063]

[0064] In this embodiment, 668 sample points are selected, including 452 vegetation points and 216 non-vegetation points. The distribution of CAI values ​​of vegetation points and non-vegetation points is statistically analyzed. The CAI values ​​of vegetation points are all greater than (equal to) 1, and the CAI values ​​of non-vegetation points are all less than 1, so the threshold t is determined to be 1.

[0065] The results of ground vegetation extraction based on the method of this embodiment are shown in Figure 8 The results of vegetation extraction based on traditional methods are shown in Fig. 9 In the figure, the vegetation area is shown in black and the non-vegetation area is shown in white.

[0066] S7. Select overall accuracy as a verification indicator to verify the vegetation extraction accuracy of this embodiment. The overall accuracy expression is:

[0067]

[0068] During the verification process, the sample point selection method is as follows: a 250×250 grid is delineated on the image with pixels as the unit, and the center point of the grid is taken as the sample point. The total number of sample points is 668. Combining the remote sensing images and visual interpretation of Google Earth images, it is determined that there are 452 vegetation points and 216 non-vegetation points among the 668 sample points.

[0069] The vegetation extraction accuracy of the method of this embodiment is compared with that of the traditional method. The results show that the vegetation extraction accuracy of the method is higher than 95%, which is significantly better than the traditional method. The accuracy statistical analysis of the two methods is shown in Table 3, and the vegetation extraction accuracy comparison results of the two methods are shown in Table 4.

[0070] Table 3 Statistical analysis of the accuracy of the two methods

[0071]

[0072] Table 4 Comparison of vegetation extraction accuracy between the two methods

[0073] Vegetation extraction method Number of correctly classified points Overall accuracy (%) This method 641 / 668 95.96 Traditional methods 588 / 668 88.02

[0074] Based on the same inventive concept, this embodiment also provides a high-precision extraction system for ground objects and vegetation in multispectral remote sensing images based on cluster analysis index, including the following modules:

[0075] Input module, used to input near infrared band and red light band of multispectral image;

[0076] The matching module is used to match the corresponding feature points of the near-infrared band image and the red light band image, so that the geometric spatial position of the same object in the near-infrared band image and the red light band image is consistent;

[0077] The characteristic parameter calculation module is used to calculate and count the image characteristic parameters of the near-infrared band and the red light band;

[0078] The intermediate band output module is used to derive and construct a balanced transformation function based on the image enhancement principle and the scale invariance principle; the balanced transformation function is used to match and optimize the red light band and the near infrared band, and integrate and output the intermediate band for vegetation analysis;

[0079] The cluster analysis index calculation module is used to calculate the cluster analysis index pixel by pixel based on the information of the intermediate band of vegetation analysis;

[0080] The extraction module is used to define a cluster analysis index threshold value, and extract ground feature vegetation information according to the magnitude relationship between the cluster analysis index value of each pixel and the threshold value.

[0081] The modules of the extraction system in this embodiment are respectively used to implement the steps of the aforementioned extraction method. For the detailed implementation process, refer to the corresponding steps described in the extraction method.

[0082] The high-precision extraction method and system of ground objects and vegetation in multispectral remote sensing images based on cluster analysis index proposed in this embodiment are mainly aimed at remote sensing images with multispectral bands such as near-infrared, red, green, and blue bands. The overall accuracy of ground object vegetation extraction in this embodiment is ≥95%, indicating that the method is highly operable, widely applicable, and has high extraction accuracy. The theoretical basis of this embodiment is solid, the physical meaning of each indicator is clear, the operation and processing process is clear and simple, the extraction result is of high quality and good effect, and it has a good promoting effect on the application of remote sensing technology.

[0083] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A high-precision extraction method for vegetation in multispectral remote sensing images based on cluster analysis index, characterized in that: The following steps are involved: S1, near infrared band and red light band of input multispectral image; S2, matching the corresponding feature points of the near-infrared band image and the red light band image, so that the geometrical spatial position of the same ground object in the near-infrared band image and the red light band image is consistent; S3, calculate and count the image characteristic parameters of near infrared band and red light band; S4. Based on the principle of image enhancement and the principle of scale invariance, a balanced transformation function is derived and constructed; the red light band and the near infrared band are matched and optimized by using the balanced transformation function, and the intermediate band for vegetation analysis is integrated and output; S5, based on the information of the intermediate bands of vegetation analysis, the cluster analysis index is calculated pixel by pixel; S6, defining a cluster analysis index threshold, and extracting ground feature vegetation information according to the magnitude relationship between the cluster analysis index value of each pixel and the threshold; The image feature parameters in step S3 include mean and standard deviation, and the calculation formula is: Among them, the remote sensing image is composed of pixels. Let B i is the near infrared band image or red light band image in the multispectral band, n represents the band image B i The nth row, l represents the band image B i The first column, 3 i (n, l) represents band image B i The pixel in the nth row and lth column of i The total number of rows and columns; μ i , σ i Band image B i The mean and standard deviation of the near infrared band of the multispectral image are μ and μ respectively. NIR , σ NIR , the mean and standard deviation of the red light band are μ R , σ R ; The equalization transformation function derived and constructed in step S4 is: Among them, NIR(n, l) represents the near-infrared band reflectance of the pixel in the nth row and the lth column, and hmβ(n, l) represents the value of the nth row and the lth column of the output intermediate band hmβ; The calculation expression of cluster analysis index is: Wherein, R(n, l) represents the red light band reflectance of the pixel in the nth row and the lth column, and CAI(n, l) represents the value of the output clustering analysis index CAI in the nth row and the lth column.

2. The extraction method according to claim 1, characterized in that Step S6 sets a threshold value based on the distribution of cluster analysis index values ​​of vegetation and non-vegetation in the sample points.

3. The extraction method according to claim 2, characterized in that In step S6, the cluster analysis index threshold is set to 1.

4. The extraction method according to claim 1, characterized in that The extraction method further comprises the steps of: S7. Select overall accuracy as the verification indicator to verify the vegetation extraction accuracy.

5. A high-precision extraction system for vegetation in multispectral remote sensing images based on cluster analysis index, characterized in that: Includes the following modules: Input module, used to input near infrared band and red light band of multispectral image; The matching module is used to match the corresponding feature points of the near-infrared band image and the red light band image, so that the geometric spatial position of the same object in the near-infrared band image and the red light band image is consistent; The characteristic parameter calculation module is used to calculate and count the image characteristic parameters of the near-infrared band and the red light band; The intermediate band output module is used to derive and construct a balanced transformation function based on the image enhancement principle and the scale invariance principle; the balanced transformation function is used to match and optimize the red light band and the near infrared band, and integrate and output the intermediate band for vegetation analysis; The cluster analysis index calculation module is used to calculate the cluster analysis index pixel by pixel based on the information of the intermediate band of vegetation analysis; An extraction module is used to define a cluster analysis index threshold value, and extract ground feature vegetation information according to the magnitude relationship between the cluster analysis index value of each pixel and the threshold value; The image feature parameters calculated by the feature parameter calculation module include mean and standard deviation, and the calculation formula is: Among them, the remote sensing image is composed of pixels. Let B i is the near infrared band image or red light band image in the multispectral band, n represents the band image B i The nth row, l represents the band image B i The first column of B i (n, l) represents band image B i The pixel in the nth row and lth column of i The total number of rows and columns; μ i , σ i Band image B i The mean and standard deviation of the near infrared band of the multispectral image are μ and μ respectively. NIR , σ NIR , the mean and standard deviation of the red light band are μ R , σ R ; The equalization transformation function derived and constructed by the intermediate band output module is: Among them, NIR(n, l) represents the near-infrared band reflectance of the pixel in the nth row and the lth column, and hmβ(n, l) represents the value of the nth row and the lth column of the output intermediate band hmβ; The calculation expression of the clustering analysis index in the clustering analysis index calculation module is: Wherein, R(n, l) represents the red light band reflectance of the pixel in the nth row and the lth column, and CAI(n, l) represents the value of the output clustering analysis index CAI in the nth row and the lth column.

6. The extraction system according to claim 5, characterized in that The extraction module sets the threshold value based on the distribution of cluster analysis index values ​​of vegetation and non-vegetation in the sample points.

Images