A fast classification method for characteristic objects in remote sensing images

Abstract

The present invention discloses a method for rapid classification of remote sensing image feature objects, comprising: step S1: collecting remote sensing hyperspectral images of a region to be classified; step S2: outputting a classification model according to the remote sensing hyperspectral images; step S3: performing band operations on the remote sensing hyperspectral images according to the classification model to generate classified images; step S4: performing threshold segmentation on the classified images to output a classification result map. The beneficial effects of the present invention are: recognizing remote sensing images based on land cover classification indexes, with faster recognition speed, avoiding the defects of low physical and geometric significance and non-universality of some classification methods in the prior art; recognizing multiple types of objects at the same time through land cover classification indexes, with fast recognition speed, avoiding the problem of low efficiency in extracting single-category objects in the prior art, and directly generating a classification result map for dividing multiple types of objects, making the process more convenient.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing imaging, and in particular to a method for quickly classifying characteristic objects in remote sensing images. Background Art

[0002] As a hub connecting people and the natural environment, the development and expansion of cities, as well as changes in their internal structure and layout, are of great significance to regional or global sustainable development. In order to timely grasp the utilization of urban land, the existing technology often uses aerial photographs or satellite photographs to collect remote sensing images of cities, and then identifies, classifies and counts the features according to the spectrum in the remote sensing images. The remote sensing images are then processed to generate the area covered by the features, which is convenient for statistics on the coverage of a certain type of features in the area, and thus timely grasps the local land use situation. Usually, the classification of features is completed through the feature classification index. By selecting the band corresponding to the specific feature for band operation, the corresponding feature can be effectively extracted from the remote sensing image.

[0003] However, the existing technologies often have the following technical problems: 1. Existing land object classification indices are mostly used for specific extraction of one type of land object. For example, the NDVI index is mainly used for extraction and analysis of vegetation. For example, classification indices such as BCI, NDBI, NDISI, IBI, and NDISI are mostly used for extraction of impermeable layers, while NDWI is mostly used for extraction and analysis of water bodies, and EBBI, NDBaI, and NDSI are mostly used for extraction and analysis of bare soil. When it is necessary to extract multiple land objects in remote sensing images, it is often necessary to use multiple classification indices to extract each type of land object separately, which is inefficient; 2. Some indices do not have strict physical and geometric meanings and have low universality. Summary of the invention

[0004] In view of the above problems existing in the prior art, a method for quickly classifying characteristic objects in remote sensing images is now provided.

[0005] The specific technical solutions are as follows:

[0006] A method for quickly classifying characteristic objects in remote sensing images, comprising:

[0007] Step S1: collecting remote sensing hyperspectral images of an area to be classified;

[0008] Step S2: outputting a classification model according to the remote sensing hyperspectral image;

[0009] Step S3: performing band operation on the remote sensing hyperspectral image according to the classification model to generate a classified image;

[0010] Step S4: performing threshold segmentation on the classified image to output a classification result map.

[0011] Preferably, step S2 comprises:

[0012] Step S21: preprocessing the remote sensing hyperspectral image to generate the surface reflectance of the remote sensing hyperspectral image;

[0013] Step S22: generating model parameters of the land cover classification index according to the surface reflectivity;

[0014] Step S23: inputting the model parameters and the surface reflectance into the land cover classification index to generate the classification model.

[0015] Preferably, the step S21 includes:

[0016] Step S211: obtaining pixel brightness values ​​of the remote sensing hyperspectral image from the remote sensing hyperspectral image;

[0017] Step S212: performing radiometric calibration on the remote sensing hyperspectral image according to the pixel brightness value to generate the absolute radiance of the remote sensing hyperspectral image;

[0018] Step S213: performing radiation correction on the remote sensing hyperspectral image according to the absolute radiance to generate an apparent reflectance;

[0019] Step S214: performing atmospheric correction on the apparent reflectivity to generate the surface reflectivity.

[0020] Preferably, the calculation formula of the apparent reflectivity is:

[0021] Where: p is the apparent reflectivity, L λ is the absolute radiance value, d is the distance between the sun and the earth in Julian day generated according to the acquisition time of the remote sensing hyperspectral image, ESUN λ is the average solar radiation brightness outside the atmosphere, θ s It is the solar zenith angle of the area to be classified when the remote sensing hyperspectral image is collected.

[0022] Preferably, the bands of the remote sensing hyperspectral image collected in step S1 include a first band, a second band and a third band;

[0023] The band widths of the first band, the second band and the third band are between 1 nm and 10 nm;

[0024] The first waveband, the second waveband and the third waveband have the same waveband width;

[0025] The wavelength of the first band is in the range of 1169nm to 1300nm, the wavelength of the second band is in the range of 1512nm to 1573nm, and the wavelength of the third band is in the range of 1628nm to 1694nm.

[0026] Preferably, in step S22, the land cover classification index is:

[0027] Where: LCCI is the land cover classification index, A is the first model parameter, is the first surface reflectance of the first band in the wavelength range from a to b, B is the second model parameter, is the second surface reflectance of the second band in the wavelength range from c to d, C is the third model parameter, is the third surface reflectivity of the third band in the wavelength range from e to f.

[0028] Preferably, in step S22, the method for generating the model parameters is:

[0029] Step S221: performing correlation analysis according to the first surface reflectivity, the second surface reflectivity and the third surface reflectivity to establish an o-xyz coordinate system;

[0030] Wherein, the second surface reflectivity corresponds to the x-axis in the o-xyz coordinate system, the third surface reflectivity corresponds to the y-axis in the o-xyz coordinate system, and the second surface reflectivity corresponds to the z-axis in the o-xyz coordinate system;

[0031] Step S222: Projecting and fitting the scattered points on the xoz plane to generate a first straight line, and projecting and fitting the scattered points on the yoz plane to generate a second straight line;

[0032] Step S223: generating the first model parameter and the second model parameter according to the intersection of the first straight line and the second straight line on the z-axis;

[0033] Step S224: selecting a plurality of characteristic pixels corresponding to a preset classification category in the remote sensing hyperspectral image;

[0034] The classification categories include a first classification category and a second classification category;

[0035] Step S225: generating a third straight line according to the first coordinate, wherein the third straight line is perpendicular to the projected fitting straight line of the xoy plane;

[0036] The first coordinate is the feature pixel farthest from the origin o in the second classification category;

[0037] Step S226: generating a second coordinate according to the third straight line, and generating a fourth straight line and a fifth straight line according to the first coordinate and the second coordinate;

[0038] The second coordinate is the intersection of the third straight line and the x-axis;

[0039] Step S227: performing an angle bisector operation based on the fourth straight line and the fifth straight line to generate the third model parameters.

[0040] Preferably, the first classification category is water bodies;

[0041] The second classification categories include: bare soil, dark buildings, light buildings, and vegetation.

[0042] The above technical solution has the following advantages or beneficial effects: remote sensing images are recognized based on the land cover classification index, with a faster recognition speed, avoiding the defects of some classification methods in the prior art that have no physical geometric meaning and are not universal; multiple types of land objects are recognized at the same time through the land cover classification index, with a fast recognition speed, avoiding the problem of only being able to extract land objects of a single category in the prior art, and can directly generate a classification result map for dividing multiple types of land objects, and the process is simpler. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The embodiments of the present invention will be described more fully with reference to the attached drawings, which are provided for illustration and description only and are not intended to limit the scope of the present invention.

[0044] Figure 1 It is an overall schematic diagram of an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of sub-steps of step S2 of an embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of sub-steps of step S21 in an embodiment of the present invention;

[0047] Figure 4A is a schematic diagram of a plane segmentation model in the prior art;

[0048] Figure 4B Schematic diagram of a three-dimensional segmentation model in an embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of sub-steps of step S22 in an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of a remote sensing image of an area to be classified in an embodiment of the present invention;

[0051] Figure 7 is a schematic diagram of a classified image in an embodiment of the present invention;

[0052] Figure 8 This is a classification result diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0055] In view of the many problems existing in the prior art, the present invention provides a land cover classification index LCCI (Land Cover Classification Index), which is suitable for quickly extracting vegetation, water bodies, bright buildings, and dark buildings from remote sensing images. This technical solution saves a lot of time and energy when classifying land use, and realizes efficient extraction of vegetation, water bodies, bright buildings, and dark buildings. It has the advantages of "high accuracy, fast speed, and simplicity", and has strong practicality and popularization.

[0056] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0057] The present invention comprises:

[0058] A fast classification method for remote sensing image feature objects, such as Figure 1 As shown, including:

[0059] Step S1: collecting remote sensing hyperspectral images of an area to be classified;

[0060] Step S2: Outputting a classification model based on remote sensing hyperspectral images;

[0061] Step S3: performing band operations on the remote sensing hyperspectral image according to the classification model to generate a classified image;

[0062] Step S4: Perform threshold segmentation on the classified image to output a classification result map.

[0063] Specifically, in view of the problem that some classification indexes in the prior art are not universal, the present invention introduces a land cover classification index: It is suitable for the rapid classification of vegetation, water bodies, bright buildings and dark buildings in remote sensing hyperspectral images. Among them: LCCI is the land cover classification index, A is the first model parameter, is the first surface reflectance in the first band between wavelengths a and b, B is the second model parameter, is the second surface reflectance in the second band with a wavelength range from c to d, C is the third model parameter, is the third surface reflectance of the third band in the wavelength range from e to f. Based on the three bands in the remote sensing hyperspectral image and the above land cover classification index, the classification model is constructed to quickly separate the four types of land objects, namely vegetation, water bodies, bright buildings and dark buildings, from the remote sensing hyperspectral image.

[0064] In a preferred embodiment, Figure 2 As shown, step S2 includes:

[0065] Step S21: preprocessing the remote sensing hyperspectral image to generate the surface reflectance of the remote sensing hyperspectral image;

[0066] Step S22: generating model parameters of the land cover classification index according to the surface reflectivity;

[0067] Step S23: inputting the model parameters and the surface reflectance into the land cover classification index to generate a classification model.

[0068] In a preferred embodiment, Figure 3 As shown, step S21 includes:

[0069] Step S211: obtaining pixel brightness values ​​of remote sensing hyperspectral images from remote sensing hyperspectral images;

[0070] Step S212: performing radiometric calibration on the remote sensing hyperspectral image according to the pixel brightness value to generate the absolute radiance of the remote sensing hyperspectral image;

[0071] Step S213: performing radiation correction on the remote sensing hyperspectral image according to the absolute radiance to generate apparent reflectance;

[0072] Step S214: Perform atmospheric correction on the apparent reflectance to generate the surface reflectance.

[0073] Specifically, the process of radiation calibration and radiation correction is prior art and will not be described in detail here.

[0074] As an optional implementation, in step S24, the FLAASH method is used to perform atmospheric correction on the apparent reflectance to generate the surface reflectance.

[0075] In a preferred embodiment, the calculation formula of the apparent reflectivity is:

[0076] Where: p is the apparent reflectance, L λ is the absolute radiance value, d is the distance between the sun and the earth in Julian day generated according to the acquisition time of remote sensing hyperspectral image, ESUN λ is the average solar radiation brightness outside the atmosphere, λ s It is the solar zenith angle of the area to be classified when collecting remote sensing hyperspectral images.

[0077] As an optional implementation, the solar zenith angle is calculated as follows: θ s =90°-θ e , where θ e It is the solar altitude angle of the area to be classified when collecting remote sensing hyperspectral images.

[0078] In a preferred embodiment, the bands of the remote sensing hyperspectral image collected in step S1 include a first band, a second band and a third band;

[0079] The band widths of the first band, the second band and the third band are between 1 nm and 10 nm;

[0080] The first band, the second band and the third band have the same band width;

[0081] The wavelength of the first band ranges from 1169 nm to 1300 nm, the wavelength of the second band ranges from 1512 nm to 1573 nm, and the wavelength of the third band ranges from 1628 nm to 1694 nm.

[0082] Specifically, in a preferred embodiment, the remote sensing hyperspectral image collected has three bands with a band width of 5nm, wherein the selection range of the first band should be between 1169nm and 1300nm, the selection range of the second band should be between 1512nm and 1573nm, and the selection range of the third band should be between 1628nm and 1694nm. By respectively setting the three bands located in the above band range, a better segmentation effect can be achieved for a variety of ground objects.

[0083] In a preferred embodiment, in step S22, the land cover classification index is:

[0084] Among them: LCCI is the land cover classification index, A is the first model parameter, is the first surface reflectance in the first band between wavelengths a and b, B is the second model parameter, is the second surface reflectance in the second band between wavelengths c and d, C is the third model parameter, It is the third surface reflectivity in the third band with a wavelength range from e to f.

[0085] Specifically, Figure 4A As shown in FIG. 1 , in the prior art, a plane isosurface segmentation map is usually used, that is, different objects are extracted by segmenting a plane through a number of parallel straight lines. Figure 4B The three-dimensional isosurface segmentation map shown in Figure 4A The plane shown by the dotted line in the middle expands into three-dimensional space to form multiple Figure 4B The plane shown by the midplane MNPQ can effectively separate different landforms to achieve the purpose of distinguishing different landforms.

[0086] In a preferred embodiment, in step S22, the method for generating the model parameters is:

[0087] Step S221: performing correlation analysis according to the first surface reflectivity, the second surface reflectivity and the third surface reflectivity to establish an o-xyz coordinate system;

[0088] Among them, the second surface reflectivity corresponds to the x-axis in the o-xyz coordinate system, the third surface reflectivity corresponds to the y-axis in the o-xyz coordinate system, the second surface reflectivity corresponds to the z-axis in the o-xyz coordinate system, and the focus is the origin o;

[0089] Specifically, in Figure 4B In the stereo coordinate system shown in Perform correlation analysis and select the band with the weakest correlation As the z-axis, the band Corresponding to the x-axis, band Corresponding to the y-axis.

[0090] Step S222: Project and fit the scattered points on the xoz plane to generate a first straight line, and project and fit the scattered points on the yoz plane to generate a second straight line;

[0091] Specifically, the band on the y-axis The coefficient of the first straight line is obtained by fitting and translating the scattered point projection on the xoz plane; the band on the x-axis The coefficient of , that is, the second straight line is obtained by fitting and translating the scattered point projection on the yoz plane, as shown in equation 1) and equation 2):

[0092] Formula 1):

[0093] Formula 2):

[0094] Step S223: generating a first model parameter and a second model parameter according to the intersection of the first straight line and the second straight line on the z-axis;

[0095] Therefore, the straight line and straight line The mean of the intersection points on the z-axis is taken as the value of the straight line and straight line The intersection of the straight line and the z-axis after translation. and straight line The method for finding the mean at the intersection point (0, 0, b) on the z-axis specifically includes:

[0096] Then we have:

[0097] Formula 1):

[0098] Formula 2):

[0099] Thus, the first model parameter A=a2′ and the second model parameter B=a1′ can be obtained.

[0100] Step S224: selecting a plurality of characteristic pixels corresponding to a preset classification category in the remote sensing hyperspectral image;

[0101] The classification categories include the first classification category and the second classification category;

[0102] Step S225: Generate a third straight line according to the first coordinate, the third straight line is perpendicular to the projection fitting straight line of the xoy plane, and the projection fitting straight line of the xoy plane is shown in Formula 3:

[0103] Formula 3):

[0104] The first coordinate is the feature pixel closest to the origin o in the second classification category;

[0105] Specifically, feature pixels are selected for the remote sensing hyperspectral image, and the point farthest from the origin o in the water body class, that is, the feature pixel closest to the origin o in the first classification category, is recorded as (i1, j1, 0); the four types of ground objects, namely, bare soil, dark buildings, bright buildings, and vegetation, that is, the points closest to the origin o in the second classification category, are recorded as the first coordinate, (i2, j2, 0). At this time, the third straight line is obtained according to the first coordinate (i2, j2, 0), so that the third straight line passes through the first coordinate (i2, j2, 0) and is perpendicular to the projection fitting straight line on the xoy plane, that is, the straight line in formula 3).

[0106] Step S226: generating a second coordinate according to the third straight line, and generating a fourth straight line and a fifth straight line according to the first coordinate and the second coordinate;

[0107] The second coordinate is the intersection of the third straight line and the x-axis;

[0108] Specifically, the intersection of the third straight line and the x-axis is recorded as (i3,0,0), that is, the second coordinate. After the corresponding deformation, the fourth straight line is as shown in formula 4):

[0109] Formula 4): in

[0110] Further, the first coordinate (i2, j2, 0) and the second coordinate (i3, 0, 0) are connected to generate a fifth straight line, and the fifth straight line is shown in Formula 5):

[0111] Formula 5):

[0112] Step S227: Perform an angle bisector operation based on the fourth straight line and the fifth straight line to generate third model parameters.

[0113] Specifically, taking the second coordinate (i3,0,0) as the starting point, the angle bisectors of the fourth and fifth straight lines are calculated, as shown in Formula 6):

[0114] Formula 6):

[0115] Thus, the third model parameter C=a can be obtained.

[0116] Furthermore, the first model parameter A, the second model parameter B and the third model parameter C are substituted into the land cover classification index: The classification model can be generated, where A = a2′, B = a1′, C = a, is the first surface reflectance in the first band between wavelengths a and b, is the second surface reflectance in the second band between wavelengths c and d, It is the third surface reflectance of the third band in the wavelength range from e to f. This classification index can effectively distinguish four types of ground objects: vegetation, water, dark buildings, and bright buildings.

[0117] In a preferred embodiment, the first classification category is water bodies;

[0118] The second classification categories include: bare soil, dark buildings, light buildings, and vegetation.

[0119] Specifically, the urban landform to be classified has multiple types of landforms such as vegetation, water bodies, bright buildings and dark buildings. The existing methods for classifying and identifying a certain type of landform cannot cope with this type of landform well, or it takes a long time. This method collects hyperspectral images containing three bands between 1169nm-1300nm, 1512nm-1573nm, and 1628nm-1694nm for the area to be classified, and then establishes a stereo isosurface segmentation model based on the hyperspectral image: And calculate the first model parameter A, the second model parameter B, and the third model parameter C. Then perform band calculations based on the classification model, and simultaneously classify vegetation, water bodies, bright buildings, and dark buildings to generate Figure 6 The classified image is then segmented by threshold to form the following Figure 7 The classification result diagram shown makes the image boundary clearer and more intuitively reflects the areas covered by vegetation, water bodies, bright buildings and dark buildings respectively. In the specific implementation process, threshold segmentation is an existing technology, and its detailed principle is not repeated here.

[0120] The beneficial effects of the present invention are as follows: remote sensing images are identified based on the land cover classification index, with a faster recognition speed, avoiding the defects of some classification methods in the prior art that have low physical and geometric significance and are not universal; multiple types of land objects are identified simultaneously through the land cover classification index, with a fast recognition speed, avoiding the problem of low efficiency in extracting single-category land objects in the prior art, and a classification result map for dividing multiple types of land objects can be directly generated, and the process is simpler.

[0121] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for rapid classification of characteristic objects in remote sensing images, characterized in that: include: Step S1: collecting remote sensing hyperspectral images of an area to be classified; The bands of the remote sensing hyperspectral image collected in step S1 include a first band, a second band and a third band; Step S2: outputting a classification model according to the remote sensing hyperspectral image; Step S3: performing band operation on the remote sensing hyperspectral image according to the classification model to generate a classified image; Step S4: performing threshold segmentation on the classified image to output a classification result map; The step S2 comprises: Step S21: preprocessing the remote sensing hyperspectral image to generate the surface reflectance of the remote sensing hyperspectral image; Step S22: generating model parameters of the land cover classification index according to the surface reflectivity; Step S23: inputting the model parameters and the surface reflectance into the land cover classification index to generate the classification model; In step S22, the land cover classification index is: ； in: LCCI is the land cover classification index, A is the first model parameter, The wavelength range is a to b The first surface reflectance of the first band between B is the second model parameter, The wavelength range is c to d The second surface reflectance of the second band between C is the third model parameter, The wavelength range is e to f a third surface reflectivity of the third waveband between ; The method for generating the model parameters is: Step S221: performing correlation analysis according to the first surface reflectivity, the second surface reflectivity and the third surface reflectivity to establish an o-xyz coordinate system; Wherein, the second surface reflectivity corresponds to the x-axis in the o-xyz coordinate system, the third surface reflectivity corresponds to the y-axis in the o-xyz coordinate system, and the second surface reflectivity corresponds to the z-axis in the o-xyz coordinate system; Step S222: Projecting and fitting the scattered points on the xoz plane to generate a first straight line, and projecting and fitting the scattered points on the yoz plane to generate a second straight line; Step S223: generating the first model parameter and the second model parameter according to the intersection of the first straight line and the second straight line on the z-axis; Step S224: selecting a plurality of characteristic pixels corresponding to a preset classification category in the remote sensing hyperspectral image; The classification categories include a first classification category and a second classification category; Step S225: generating a third straight line according to the first coordinate, wherein the third straight line is perpendicular to the projected fitting straight line of the xoy plane; The first coordinate is the feature pixel farthest from the origin o in the second classification category; Step S226: generating a second coordinate according to the third straight line, and generating a fourth straight line and a fifth straight line according to the first coordinate and the second coordinate; The second coordinate is the intersection of the third straight line and the x-axis; Step S227: performing an angle bisector operation based on the fourth straight line and the fifth straight line to generate the third model parameters.

2. The rapid classification method according to claim 1, characterized in that: The step S21 comprises: Step S211: obtaining pixel brightness values ​​of the remote sensing hyperspectral image from the remote sensing hyperspectral image; Step S212: performing radiometric calibration on the remote sensing hyperspectral image according to the pixel brightness value to generate the absolute radiance of the remote sensing hyperspectral image; Step S213: performing radiation correction on the remote sensing hyperspectral image according to the absolute radiance to generate an apparent reflectance; Step S214: performing atmospheric correction on the apparent reflectivity to generate the surface reflectivity.

3. The rapid classification method according to claim 2, characterized in that: The calculation formula of the apparent reflectance is: in: is the apparent reflectance, is the absolute radiance value, is the sun-earth distance in Julian day generated according to the acquisition time of the remote sensing hyperspectral image, is the average solar radiation brightness outside the atmosphere, It is the solar zenith angle of the area to be classified when the remote sensing hyperspectral image is collected.

4. The rapid classification method according to claim 2, characterized in that: The band widths of the first band, the second band and the third band are between 1 nm and 10 nm; The first waveband, the second waveband and the third waveband have the same waveband width; The wavelength of the first band is in the range of 1169nm~1300nm, the wavelength of the second band is in the range of 1512nm~1573nm, and the wavelength of the third band is in the range of 1628nm~1694nm.

5. The rapid classification method according to claim 1, characterized in that: The first classification category is water bodies; The second classification categories include: bare soil, dark buildings, light buildings, and vegetation.

Images