Urban land identification method based on remote sensing image

Abstract

The application discloses a city land use recognition method based on remote sensing images, and relates to the technical field of city land use planning.The application solves the problem of insufficient fine features of the original recognition mode, locks core boundary pixels from a complex gradient pixel point set by pixel interval quantization, feature circle construction and refined point screening, and significantly improves the contour accuracy.Compared with the traditional single line contour, the refined contour can more completely represent the feature form, is especially suitable for irregular terrains, and reduces classification errors.The pixel mean value and variance features are used as the measurement, the number of combined areas and the internal feature similarity are taken into account, the optimal combination is screened through weighted calculation, scattered division or excessive merging is avoided, and the generated optimal combined area is highly consistent with the requirements of function aggregation and land use cooperation in city planning, thereby significantly improving the planning efficiency and scientificity.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban land planning, in particular to a city land recognition method based on remote sensing images. BACKGROUND

[0002] With the acceleration of urbanization, urban land planning has increasingly urgent needs for high-precision and high-efficiency ground feature recognition and regional division. Traditional urban land recognition relies on manual field investigation or low-resolution remote sensing images, which has limitations such as low efficiency, large error, and difficulty in capturing complex terrain details.

[0003] In the prior art, although there are automatic recognition methods based on remote sensing images, there are generally the following problems: first, the contour extraction accuracy is insufficient, the edge feature capture ability of irregular boundary areas such as lakes and forests is limited, and the contour is easily offset by spectral interference; second, there is a lack of fine processing, only a single line contour is used to represent the ground object, which cannot reflect the spatial distribution characteristics of the ground object edge, affecting the subsequent classification accuracy; third, the regional division strategy is single, and the spatial correlation and functional similarity between ground objects are not considered, making it difficult to meet the needs of land coordination and functional aggregation in urban planning.

[0004] In view of the above problems, a technical solution combining high-precision image processing and intelligent analysis is needed to realize accurate extraction of urban land contour, fine classification of terrain features, and scientific and reasonable regional combination, providing reliable support for optimization of urban land resources. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a city land recognition method based on remote sensing images, which solves the problem of insufficient fine features in the original recognition method.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a city land recognition method based on remote sensing images, comprising the following steps:

[0007] Step 1: Obtain the remote sensing image of the region to be planned, and confirm the gradient pixel points. From the confirmed gradient pixel points, the contour regions existing in the remote sensing image are sequentially calibrated, and the specific method is as follows:

[0008] Confirm the RGB values associated with different image points in the remote sensing image, and confirm the gray values of the confirmed RGB values by using: HD=R×0.114+G×0.587+B×0.299 to confirm the gray values associated with the corresponding image points, and based on the different gray values associated with different image points, the remote sensing image is grayed to confirm the gray image;

[0009] The vertical gradient and the vertical gradient associated with different pixel points in the gray image are confirmed by using the Sobel algorithm, and the confirmed vertical gradient is calibrated as C i The confirmed vertical gradient is calibrated as S i , wherein i represents different pixel points, and the vertical gradient associated with the corresponding pixel point is confirmed by using The comprehensive gradient ZH i associated with the corresponding pixel point is confirmed

[0010] The pixel points satisfying ZH i > Y1 are calibrated as gradient pixel points, Y1 is a preset value, otherwise, no calibration is performed;

[0011] A plurality of gradient pixel points associated with the gray image are sequentially confirmed, and a middle region included by a plurality of continuous gradient pixel points is recorded as a contour region, and a plurality of contour regions associated with the gray image are sequentially calibrated;

[0012] Step two, based on the contour features of the contour region, the contour features are quantified, and a refinement point is selected from a plurality of quantified feature values, a plurality of refinement points are connected, and a refinement contour belonging to the corresponding contour region is confirmed, and the specific method is:

[0013] A plurality of contour regions calibrated in the remote sensing image are determined, and an edge contour of the corresponding contour region is locked from a single contour region determined, a plurality of pixel values associated with the gradient pixel points associated with the edge contour are sequentially confirmed, and the minimum value and the maximum value are determined therefrom to lock the contour pixel interval;

[0014] The contour pixel interval locked by the corresponding edge contour is quantified, the minimum value is quantized as 0, the maximum value is quantized as 2, and the quantization ratio is confirmed: (maximum value-minimum value) ÷ 2=quantization ratio, and a plurality of pixel values existing in the contour pixel interval are quantized by using: X i ÷quantization ratio=L i The quantized value L i corresponding to the pixel point is confirmed

[0015] The corresponding contour region is combined with the two-dimensional coordinate system, and different two-dimensional coordinates associated with different pixel points are confirmed, the two-dimensional coordinates of a plurality of groups of pixel points are processed by using the mean value, the mean value coordinates are confirmed, and are simultaneously calibrated in this contour region as the center point of this contour region, and the pixel point farthest from the center point on the edge contour of the contour region is locked as the feature edge point, and a group of feature circles are confirmed with the center point as the center and the straight line distance between the center point and the feature edge point as the radius

[0016] The edge circle point of the feature circle is taken as the associated end point, the associated connection line between the center point and the associated end point is confirmed, the part of the connection line located in the edge contour is recorded as the contour connection line, a number of pixel points intersected by each contour connection line are recorded as a pixel point set, and a thinning point is selected from each different pixel point set: the different quantization values L associated with the different pixel points in the pixel point set are confirmed i , and the following is adopted: Cz i = |L i -1| The feature difference Cz associated with the corresponding pixel point is confirmed i , and the pixel point associated with the Cz i min is selected as the thinning point and selected from the different feature differences Cz associated with the different pixel points i .

[0017] Based on the thinning point selected from each pixel point set, the adjacent thinning points are connected, the thinning contour belonging to the contour region is confirmed, and the thinning contour is reselected as the edge contour of the contour region;

[0018] Step three, based on the specific pixel features of the corresponding contour region, the belonging classification of the contour region is confirmed, and the belonging classification is labeled, and the specific method is:

[0019] Based on the edge contour after the contour region thinning processing, the pixel values of a number of pixel points located in the edge contour are confirmed, and based on the confirmed minimum value and maximum value, the pixel value interval belonging to the contour region is confirmed;

[0020] The pixel value interval is checked and compared with a number of groups of preset intervals: the intersection range of the pixel value interval and different preset intervals is confirmed, and the proportion value of the intersection range located in the corresponding preset interval is determined, which is the proportion value = the total value length of the intersection range ÷ the total value length of the preset interval. The preset interval that meets the proportion value ≥ 85% is taken as the adaptive interval of the pixel value interval, and the belonging classification associated with the adaptive interval is taken as the belonging classification of the contour region, and is labeled in the remote sensing image at the same time;

[0021] Step four, a number of contour regions associated in the remote sensing image are randomly combined, a combined area is generated, and the optimal process is selected based on different random combination processes. The combined area associated with the optimal process is taken as the optimal combined area and is displayed, and the specific method is:

[0022] The mean value processing is performed on a number of pixel values associated with each contour region, the pixel mean value of each contour region is confirmed, and is taken as the contour feature of the corresponding contour region;

[0023] Adjacent contour regions within the remote sensing image are randomly combined to identify combined regions. Each combined region must include at least two sets of contour regions. The total number of combined regions identified in this random combination process is denoted as G. k Then, variance processing is performed on the contour features associated with several contour regions within the corresponding combined area to confirm the variance eigenvalue F associated with the corresponding combined area. k k represents different combinations of processes;

[0024] G is adopted. k ×A1+F k ×A2=BZ k Determine the calibration feature value BZ associated with the corresponding combination process. k A1 and A2 are both preset fixed coefficient factors, representing different calibration feature values ​​BZ associated with different combination processes. k In the middle, select BZ k The process associated with min is recorded as the optimal process. The combined region associated with the optimal process is recorded as the optimal combined region and displayed.

[0025] This invention provides a method for urban land use identification based on remote sensing imagery. Compared with existing technologies, it has the following advantages:

[0026] This invention combines high-resolution UAV imagery with the Sobel algorithm to accurately capture edge differences in different areas such as lakes and forests. Grayscale processing and gradient thresholding effectively filter out interference information, ensuring that the outline region division fits the true boundary of the ground features, laying a reliable foundation for subsequent analysis.

[0027] By quantizing pixel intervals, constructing feature circles, and filtering refined points, core boundary pixels are locked from complex gradient pixel sets, significantly improving contour accuracy. Compared with traditional single-line contours, refined contours can more completely represent the morphology of ground features, especially suitable for irregular terrain, reducing classification errors.

[0028] Using pixel mean and variance features as measures, and taking into account the number of combined areas and the similarity of internal features, the optimal combination is selected through weighted calculation to avoid scattered division or excessive merging. The generated optimal combination area is highly consistent with the needs of functional agglomeration and land use coordination in urban planning, which significantly improves planning efficiency and scientificity. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0030] Figure 2 This is a schematic diagram illustrating the determination of the outline connection lines of the present invention. Detailed Implementation

[0031] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described, obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application. Embodiments

[0032] Please refer to Figure 1 The present application provides a city land identification method based on remote sensing images, comprising the following steps:

[0033] Step one, the remote sensing image of the planning area is acquired, and the gradient pixel points of the acquired remote sensing image are confirmed, and the outline area existing in the remote sensing image is sequentially calibrated from the confirmed gradient pixel points. Specifically, in the remote sensing image, corresponding to different regional blocks in the image, there is a large difference on the edge with other areas, for example, the lake area has obvious outline characteristics, and the forest area also has obvious outline characteristics. The specific calibration of the outline area can be carried out by confirming the corresponding gradient pixel points, which is convenient for subsequent type identification of different blocks. The remote sensing image is collected by high-definition unmanned aerial vehicle, and the collected image has high resolution. The specific processing method for calibration is:

[0034] The RGB value associated with different image points in the remote sensing image is confirmed, and the confirmed RGB value is confirmed by gray scale, that is, HD=R×0.114+G×0.587+B×0.299. The gray value associated with the corresponding image point is confirmed, and the gray image is processed based on the different gray values associated with different image points.

[0035] The vertical gradient and vertical gradient associated with different pixel points in the gray image are confirmed by using Sobel algorithm, and the confirmed vertical gradient is calibrated as C i The confirmed vertical gradient is calibrated as S i Where i represents different pixel points (the Sobel algorithm is based on the arrangement change mode of the corresponding pixel points, taking the to-be-determined pixel point as the center point, combining eight groups of pixel points around it, and giving different weights to determine the related gradient data associated with the corresponding pixel point. The content is common in the prior art, so it is not described in detail here), and the comprehensive gradient ZH associated with the corresponding pixel point is confirmed i ;

[0036] ZH iThe pixel point of Y1 is calibrated as a gradient pixel point, Y1 is a preset value, and a specific value thereof is determined by an operator according to experience, and otherwise, no calibration is performed;

[0037] The gradient pixel points associated with the gray image are sequentially confirmed, and the intermediate region included in the continuous gradient pixel points is recorded as a contour region (that is, the intermediate region surrounded by the gradient pixel points), and the contour regions associated with the gray image are sequentially calibrated. Specifically, a surrounding ring is generated between different gradient pixel points, and the intermediate region generated by the surrounding ring is the corresponding contour region;

[0038] In step two, based on the different contour regions calibrated in the remote sensing image, the contour features of the contour regions are quantified, and the refined points are selected from the quantified feature values, and the refined points are connected to confirm the refined contour of the corresponding contour region. Specifically, different edge contours have different contour features, and the corresponding contour features are not only a single line, but also a region. In order to achieve better contour refinement effect, the optimal pixel point is locked from the corresponding gradient pixel point, and the contour is refined again. The specific processing method for confirmation is as follows:

[0039] The contour regions calibrated in the remote sensing image are determined, and the edge contour of the corresponding contour region is locked from the determined single contour region. The pixel values associated with the gradient pixel points associated with the edge contour are sequentially confirmed, and the minimum value and the maximum value are determined to lock the contour pixel interval.

[0040] The contour pixel interval locked by the corresponding edge contour is quantified, the minimum value is quantized as 0, the maximum value is quantized as 2, and the quantization ratio is confirmed: (maximum value-minimum value) ÷ 2=quantization ratio. The pixel values existing in the contour pixel interval are quantized by using: X i ÷quantization ratio=L i Confirm the quantized value L of the pixel point i For example, assuming that the associated contour pixel interval is [20, 40], 20 is quantized as 0, 40 is quantized as 2, and the associated quantized value is 20. There are three pixel points in the interval, and the associated pixel values are 25, 30 and 35. 25 is quantized as 1.25, 30 is quantized as 1.5, and 35 is quantized as 1.75.

[0041] The corresponding contour region is combined with a two-dimensional coordinate system, and the different two-dimensional coordinates associated with different pixels are identified. The two-dimensional coordinates of several sets of pixels are averaged to determine the average coordinates, which are then simultaneously marked within this contour region as the center point. The pixel furthest from this center point is then identified on the edge contour of the contour region and denoted as the feature edge point. A set of feature circles is identified with the center point as the center and the straight-line distance between the center point and the feature edge point as the radius. Figure 2 Based on the corresponding contour area, the corresponding center point is locked, and then the corresponding feature circle is constructed by combining the farthest edge point. Then the connection line is confirmed. After the connection line is confirmed, the contour line belonging to the corresponding edge contour can be locked, and the points associated with the contour line are selected for feature selection. The corresponding refinement point is locked from it. Based on the refinement point determined in sequence, the refinement contour belonging to this area can be locked.

[0042] Using the edge circle points of the feature circle as the association endpoints, confirm the association lines between the center point and the association endpoints. Record the portion of the association line within the edge contour as the contour line. Record the several pixels where each contour line intersects as a pixel set, and select refinement points from each different pixel set.

[0043] Identify the different quantization values ​​L associated with different pixels in the pixel set. i Using: Cz i =|L i -1| Confirm the feature difference Cz associated with the corresponding pixel. i Then, from the different feature differences Cz associated with different pixels i In the middle, select Cz i The pixels associated with min are selected as the thinning points;

[0044] Based on the selected thinning points of each pixel set, adjacent thinning points are connected to confirm the thinning contour belonging to this contour region, and the thinning contour is re-used as the edge contour of this contour region. Specifically, the gradient pixel associated with the thinning contour is the most central pixel in the corresponding gradient region. This processing method can effectively complete the thinning of the edge contour of each different contour region, thereby achieving a better region division processing effect.

[0045] Step 3: For the relevant contour areas in the remote sensing image that have been refined, based on the specific pixel features of the corresponding contour areas, confirm the classification of the contour areas and label the classification. Specifically, the classification includes relevant areas with different terrain features such as lakes, tree bark, and flat land.

[0046] The specific method for calibration is as follows:

[0047] Based on the edge contour after the contour region refinement processing, the pixel values of a plurality of pixel points located in the edge contour are confirmed, and based on the minimum value and the maximum value confirmed, the pixel value interval belonging to the contour region is confirmed;

[0048] The pixel value interval is checked and compared with a plurality of groups of preset intervals: the intersection range of the pixel value interval and different preset intervals is confirmed, and the proportion value of the intersection range located in the corresponding preset interval is determined, which is the proportion value = the total value length of the intersection range ÷ the total value length of the preset interval. The preset interval that satisfies the proportion value ≥ 85% is taken as the adaptive interval of the pixel value interval, and the classification to which the adaptive interval is associated is taken as the classification to which the contour region belongs, and is simultaneously marked in the remote sensing image. If it does not satisfy, it will not be marked.

[0049] Specifically, each contour region has an associated pixel value interval, and the corresponding pixel value interval and the corresponding preset interval have an intersection range. The preset interval is determined according to the characteristics of the terrain by the operator, and is a preset value. Each different contour region has a classification to which it belongs, so that the specific classification to which the corresponding contour region belongs can be determined based on the corresponding interval check and verification, facilitating subsequent feature integration.

[0050] Step four, randomly combining a plurality of contour regions associated in the remote sensing image, and generating a combined area, and selecting an optimal process based on different random combination processes, taking the combined area associated with the optimal process as the optimal combined area and displaying it. The specific processing method for random combination is:

[0051] The mean value of a plurality of pixel values associated with each contour region is processed to confirm the pixel mean value of each contour region, and the pixel mean value is taken as the contour feature of the corresponding contour region;

[0052] Randomly combine adjacent contour regions in the remote sensing image to confirm the combined area. Each combined area includes at least two contour regions. The total number of combined areas confirmed in this random combination process is denoted as G k The contour features associated with a plurality of contour regions in the corresponding combined area are subjected to variance processing to confirm the variance feature value F k associated with the corresponding combined area, k represents different combination processes;

[0053] G k ×A1+F k ×A2=BZ k determines the marking feature value BZ associated with the corresponding combination process k , wherein A1 and A2 are preset fixed coefficient factors, and their specific values are determined by the operator according to experience. Different marking feature values BZ associated with different combination processesk In the embodiment, the selected BZ k The combination process associated with the minimum is recorded as the optimal process, and the combination area associated with the optimal process is recorded as the optimal combination area and displayed. Each combination process is associated with a different eigenvalue of variance and a corresponding total number of combination areas. When the determined total number of combination areas in the corresponding combination process is the least and the associated eigenvalue of variance is the smallest, the determined combination areas are more similar, the features of different contour regions in the corresponding combination area are more similar, and the relevant regions are basically the same category. After the combination processing, the optimal combination effect can be achieved, and subsequent relevant personnel can plan the urban area of the specified area.

[0054] Some data in the above formula are dimensionless for numerical calculation, and the contents not described in detail in the specification belong to the prior art known to those skilled in the art.

[0055] The above embodiments are only used to illustrate the technical method of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A method for identifying urban land based on remote sensing images, characterized in that, The method comprises the following steps: Step one, remote sensing image of the region to be planned is acquired, and gradient pixel points are confirmed, and the outline region existing in the remote sensing image is labeled in sequence from the confirmed gradient pixel points; Step two, based on the outline feature of the outline region, the outline feature is quantified, and the thinning points are selected from the quantified feature values, the thinning points are connected, and the thinning outline belonging to the corresponding outline region is confirmed, and the specific method is: The outline region labeled in the remote sensing image is determined, and the edge outline of the corresponding outline region is locked from the determined single outline region, the pixel values associated with the gradient pixel points associated with the edge outline are confirmed in sequence, and the minimum value and the maximum value are determined to lock the outline pixel interval; The profile pixel interval corresponding to the edge profile lock is numerically quantized, the minimum value is quantized as 0, the maximum value is quantized as 2, and the quantization ratio is confirmed: (maximum value-minimum value) ÷ 2=quantization ratio, and the several pixel values existing in the profile pixel interval are quantized, using: X i ÷quantization ratio=L i The quantized value L corresponding to the pixel point is confirmed i ; The corresponding outline region is combined with the two-dimensional coordinate system, and different two-dimensional coordinates associated with different pixel points are confirmed, the two-dimensional coordinates of the pixel points are processed by averaging, the average coordinates are confirmed, and the average coordinates are labeled in the outline region as the center point of the outline region, and the pixel point farthest from the center point is locked from the edge outline of the outline region, which is recorded as the feature edge point, and the center point is taken as the center, and the straight line distance between the center point and the feature edge point is taken as the radius to confirm a group of feature circles; The edge circle point of the feature circle is taken as the associated terminal point, the associated connecting line between the center point and the associated terminal point is confirmed, the part of the connecting line located in the edge outline is recorded as the outline connecting line, and the pixel points intersected by each outline connecting line are recorded as a pixel point set, and the thinning points are selected from each different pixel point set; Based on the thinning points selected from each pixel point set, the adjacent thinning points are connected, the thinning outline belonging to the outline region is confirmed, and the thinning outline is taken as the edge outline of the outline region again; Step three, based on the specific pixel feature of the corresponding outline region, the classification to which the outline region belongs is confirmed, and the classification is labeled; Step four, the outline regions associated with the remote sensing image are randomly combined, and the combined area is generated, and the optimal process is selected based on different random combination processes, and the combined area associated with the optimal process is taken as the optimal combined area and displayed. 2.The urban land recognition method based on remote sensing images according to claim 1, characterized in that, In the step one, the specific method for labeling the outline region in sequence is: The RGB values associated with different image points in the remote sensing image are confirmed, and the confirmed RGB values are confirmed by gray scale, that is, HD=R×0.114+G×0.587+B×0.299, the gray value associated with the corresponding image point is confirmed, and the remote sensing image is processed by gray scale based on the different gray values associated with different image points to confirm the gray image; The vertical gradient associated with different pixel points in the gray image is confirmed by using Sobel algorithm, and the confirmed vertical gradient is calibrated as C i The vertical gradient associated with different pixel points in the gray image is confirmed by using Sobel algorithm, and the confirmed vertical gradient is calibrated as C i Wherein i represents different pixel points, and the horizontal gradient associated with the pixel points is confirmed by using Sobel algorithm The comprehensive gradient ZH associated with the pixel points is confirmed i ; The following will be satisfied: ZH i The pixel point of Y1 is calibrated as a gradient pixel point, and Y1 is a preset value. The gradient pixel points associated with the gray image are confirmed in sequence, and the intermediate region including the continuous gradient pixel points is recorded as the outline region, and the outline regions associated with the gray image are labeled in sequence. 3.The urban land recognition method based on remote sensing images according to claim 2, characterized in that, For the pixels that do not satisfy ZH i No calibration is performed for the pixels that do not satisfy Y1. 4.The urban land recognition method based on remote sensing images according to claim 1, characterized in that, The selection method for selecting the thinning points from the pixel point set is: Confirming different quantization values L associated with different pixels in the pixel set i , adopt: Cz i = |L i -1| Confirm the feature difference Cz associated with the corresponding pixel point i , and select the pixel point associated with Cz i min from different feature differences Cz i associated with different pixels as the refinement point and select. 5.The urban land recognition method based on remote sensing images according to claim 1, characterized in that, In the step three, the specific method for labeling the classification is: Based on the edge contour after the contour region refinement processing, the pixel values of a plurality of pixel points located in the edge contour are confirmed, and based on the confirmed minimum value and maximum value, the pixel value interval belonging to the contour region is confirmed; The pixel value interval is checked and compared with a plurality of groups of preset intervals: the intersection range of the pixel value interval and different preset intervals is confirmed, and the proportion value of the intersection range located in the corresponding preset interval is determined, which is the proportion value = the total value length of the intersection range ÷ the total value length of the preset interval. The preset interval that meets the proportion value ≥ 85% is taken as the adaptive interval of the pixel value interval, and the classification to which the adaptive interval is associated is taken as the classification to which the contour region belongs, and is simultaneously marked in the remote sensing image. 6.The urban land recognition method based on remote sensing images according to claim 1, characterized in that, The preset interval that does not meet the proportion value ≥ 85% is not marked. 7.The urban land recognition method based on remote sensing images according to claim 1, characterized in that, In the step four, the specific way of randomly combining a plurality of contour regions is: The mean value of a plurality of pixel values associated with each contour region is processed, the pixel mean value of each contour region is confirmed, and is taken as the contour feature of the corresponding contour region; combining the adjacent contour regions in the remote sensing image randomly to confirm a combined area, each combined area including at least two groups of contour regions, and the total number of the combined areas confirmed in the random combining process being recorded as G k further performing variance processing on the contour features associated with the contour regions in the corresponding combined area to confirm a variance feature value F associated with the corresponding combined area k k representing different combining processes Adopt: G k xA1 + F k xA2 = BZ k Determine the calibration characteristic value BZ associated with the corresponding combination process k , wherein A1 and A2 are both preset fixed coefficient factors, and BZ k , select the combination process associated with BZ k min, and mark it as the optimal process. The combination area associated with the optimal process is marked as the optimal combination area and displayed.

Images