A method for scene classification of aerial remote sensing images based on data processing

Abstract

The present invention discloses a method for classifying aerial remote sensing images based on data processing, which relates to the field of image processing technology and includes the following steps: S1, using sensors to collect real-time aerial remote sensing images; S2, using shadow detection sets to construct remote sensing digital templates, processing the pixels of the real-time aerial remote sensing images, and obtaining the remote sensing fitness of each pixel; S3, using isolation forests to process the remote sensing fitness of all pixels, and obtaining the classification results of the real-time aerial remote sensing images. The present invention uses remote sensing digital templates to calculate the remote sensing fitness of pixels, which can quantify the characteristics of the pixels and provide a reliable basis for subsequent target classification; the isolation forest algorithm can process large-scale data sets, is suitable for real-time aerial remote sensing images, which contain a large number of pixels, and effectively identifies abnormal values, thereby judging whether there is a target and completing scene classification.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method for classifying aerial remote sensing image scenes based on data processing. Background Art

[0002] In recent years, aerial remote sensing and drone technology have developed rapidly. As an effective information carrier, aerial remote sensing images contain rich shape, texture, and scene semantic information and are widely used in applied research such as intelligence reconnaissance, environmental monitoring, natural disaster prevention, and water resource protection. Aerial remote sensing images usually contain complex ground feature information and a variety of target types. How to accurately and quickly identify whether there are target types in aerial remote sensing images and complete the scene classification of aerial remote sensing images has become an urgent problem to be solved. However, traditional aerial remote sensing image classification methods based on target recognition often rely on manual feature extraction, which is not only time-consuming and labor-intensive, but also difficult to adapt to complex and changing image scenes. It has exceeded the scope of human real-time interpretation and understanding, and it is difficult to complete classification in a timely and accurate manner. Summary of the Invention

[0003] In order to solve the above problems, the present invention proposes a scene classification method for aerial remote sensing images based on data processing.

[0004] The technical solution of the present invention is: a method for classifying aerial remote sensing images based on data processing comprises the following steps:

[0005] S1. Use sensors to collect real-time aerial remote sensing images;

[0006] S2. Using the shadow detection set to construct a remote sensing digital template, the pixels of the real-time aerial remote sensing image are processed to obtain the remote sensing fitness of each pixel;

[0007] S3. Use isolation forest to process the remote sensing fitness of all pixels to obtain the classification results of real-time aerial remote sensing images.

[0008] Furthermore, S2 includes the following sub-steps:

[0009] S21, constructing a shadow detection set for the real-time aerial remote sensing image according to the solar horizon angle and the solar azimuth angle when the real-time aerial remote sensing image is collected;

[0010] S22, construct remote sensing digital template using shadow detection set;

[0011] S23. Use remote sensing digital templates to process real-time aerial remote sensing images to obtain remote sensing fitness of each pixel.

[0012] The beneficial effect of the above further scheme is: in the present invention, the solar sky-ground angle and the solar azimuth angle are key factors affecting the formation and distribution of shadows in remote sensing images. Different solar sky-ground angles and solar azimuth angles will cause the shape, size and position of shadows in remote sensing images to change. Constructing a shadow detection set based on the specific conditions when acquiring images in real time can enhance the adaptability of the shadow detection algorithm to different conditions, thereby more accurately identifying shadow areas in remote sensing images. An accurate shadow detection set provides a reliable foundation for subsequent processing such as target recognition using shadow information. In addition, the construction of the remote sensing digital template takes into account a variety of factors, which can enhance the robustness of the algorithm, enabling it to cope with more complex surface environments and more diverse surface objects. The remote sensing fitness generated by the remote sensing digital template for each pixel can be used as important input information for subsequent classification and recognition processes.

[0013] Furthermore, in S21, the shadow detection set of real-time aerial remote sensing images The expression is:

[0014] ;

[0015] Where, Indicates the sun angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth angle when collecting real-time aerial remote sensing images. Represents the radiance of real-time aerial remote sensing images, Indicates the radiant emittance of the sensor.

[0016] The beneficial effect of the above further solution is that, in the present invention, the solar zenith angle refers to the angle between the incident sunlight at the pixel position at that moment and the normal to the ground plane of that pixel. The solar azimuth angle refers to the angle of the ray (with the vertex being the surface pixel point) that starts from the north direction and rotates clockwise to the projection of the incident sunlight at that pixel at a certain moment on the earth's surface. The radiant brightness of a remote sensing image refers to the radiant energy received by the remote sensing sensor from the earth's surface or other target objects, and the brightness value formed on the sensor. Utilizing this angular information, the reflectivity and radiation characteristics of the surface can be more accurately estimated, thereby improving the quality and usability of remote sensing images. The radiant emittance of the sensor is used to describe the radiant flux radiated into the hemispherical space per unit area of ​​the image sensor surface.

[0017] Furthermore, S22 includes the following sub-steps:

[0018] S221, generating a detection matrix according to the shadow detection set;

[0019] S222. Fill the F-norm value of the detection matrix into the center position of the remote sensing digital template to generate a remote sensing digital template.

[0020] The beneficial effect of the above further solution is that in the present invention, the F-norm value of the detection matrix is ​​filled into the center position of the remote sensing digital template, which can effectively represent the shadow characteristics. The detection matrix provides important input information for subsequent processing. By analyzing and processing the detection matrix, shadow characteristics can be further extracted to support tasks such as remote sensing image classification and recognition. Remote sensing digital template The expression is , after filling the F norm value into the center position of the remote sensing digital template, fill the elements of the shadow detection set into the first row in sequence, fill the elements of the shadow detection set into the last row in reverse order, and then fill the elements of the first column of the second row and the third column of the second row, where, represents the F norm of the detection matrix.

[0021] Furthermore, in S221, the detection matrix The expression is:

[0022] ;

[0023] Where, Indicates the sun angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth angle when collecting real-time aerial remote sensing images. Represents a real-time shadow detection set for aerial remote sensing images The standard deviation of all elements in , Represents the maximum function.

[0024] Furthermore, in S23, each pixel of the real-time aerial remote sensing image is input into the remote sensing digital template, and the shadow adaptability of each pixel is output using the maximum eigenvalue of the remote sensing digital template;

[0025] The location in the real-time aerial remote sensing image is The shadow fitness of the pixel The calculation formula is:

[0026] ;

[0027] Where, represents the horizontal coordinate of the pixel, represents the vertical coordinate of the pixel, Indicates the hue of the pixel, Indicates the mean of all tones of the four neighboring pixels corresponding to the pixel. Indicates the saturation of the pixel, Indicates the mean of all saturation values ​​of the four neighboring pixels corresponding to the pixel. represents the brightness of the pixel, Indicates the mean value of all brightness of the four neighboring pixels corresponding to the pixel. Represents the mean of all hues of the pixel's D neighboring pixels. Represents the mean of all saturation values ​​of the pixel’s D neighboring pixels. Represents the mean brightness of all D neighboring pixels corresponding to the pixel, Indicates the maximum eigenvalue of the remote sensing digital template, Indicates the sun angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth when collecting real-time aerial remote sensing images.

[0028] The beneficial effects of the above further solution are as follows: In the present invention, in the HSV model, hue is expressed as an angle, typically ranging from 0° to 360°; saturation is expressed as the ratio between the purity of the selected color and the maximum purity of the color, typically expressed as a percentage or a normalized value; and lightness indicates the brightness of the color, typically expressed as a percentage or a normalized value. The present invention extracts the elements at the corresponding position in the remote sensing digital template based on the neighboring pixels of each pixel and participates in the calculation.

[0029] Furthermore, S3 includes the following sub-steps:

[0030] S31, randomly selecting a number of remote sensing fitnesses from the remote sensing fitnesses of all pixels as the root nodes of each isolated tree to obtain an isolation forest;

[0031] S32, using remote sensing fitness to traverse each isolated tree, and obtain the height of remote sensing fitness in each isolated tree;

[0032] S33, calculating the average height of all isolated trees for each remote sensing fitness;

[0033] S34. Determine whether there is a remote sensing fitness with an average height greater than or equal to a threshold. If so, the detection target exists in the real-time aerial remote sensing image. Otherwise, the detection target does not exist in the real-time aerial remote sensing image, and the scene classification is completed.

[0034] The above further embodiment has the beneficial effect that, in the present invention, in aerial remote sensing image scene classification, anomalous data points may represent potential target areas. By calculating the remote sensing fitness of each isolated tree and determining whether there are remote sensing fitness values ​​with an average height greater than or equal to a threshold, anomalous data points, i.e., potential targets or regions of interest, can be accurately detected. The threshold can be set manually.

[0035] Furthermore, in S32, the remote sensing fitness is at the height of the isolated tree The calculation formula is:

[0036] ;

[0037] Where, represents the standard deviation of all remote sensing fitness in the root node of the isolated tree, represents the split value of the isolated tree, represents the maximum depth of the isolation tree, represents a natural constant, Indicates the number of remote sensing fitness in the root node of the isolated tree.

[0038] The split value is the threshold used at each node to split the data into the left or right subtree.

[0039] The beneficial effects of the present invention are:

[0040] (1) The present invention uses the characteristic parameters of real-time aerial remote sensing images to extract shadow detection sets, providing important clues for subsequent target detection. By constructing remote sensing digital templates, the pixels of real-time aerial remote sensing images can be standardized to improve the efficiency and accuracy of image processing. At the same time, the remote sensing digital templates can also be adjusted and optimized as needed to adapt to different surface features and classification requirements.

[0041] (2) The present invention uses remote sensing digital templates to calculate the remote sensing fitness of pixels, which can quantify the characteristics of pixels and provide a reliable basis for subsequent target classification. The isolation forest algorithm can process large-scale data sets and is suitable for real-time aerial remote sensing images, which contain a large number of pixels. It can also effectively identify outliers, thereby determining whether there is a target and completing scene classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The flowchart of the aerial remote sensing image scene classification method based on data processing. DETAILED DESCRIPTION

[0043] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0044] like Figure 1 As shown, the present invention provides a method for classifying aerial remote sensing images based on data processing, comprising the following steps:

[0045] S1. Use sensors to collect real-time aerial remote sensing images;

[0046] S2. Using the shadow detection set to construct a remote sensing digital template, the pixels of the real-time aerial remote sensing image are processed to obtain the remote sensing fitness of each pixel;

[0047] S3. Use isolation forest to process the remote sensing fitness of all pixels to obtain the classification results of real-time aerial remote sensing images.

[0048] In this embodiment of the present invention, S2 includes the following sub-steps:

[0049] S21, constructing a shadow detection set for the real-time aerial remote sensing image according to the solar horizon angle and the solar azimuth angle when the real-time aerial remote sensing image is collected;

[0050] S22, construct remote sensing digital template using shadow detection set;

[0051] S23. Use remote sensing digital templates to process real-time aerial remote sensing images to obtain remote sensing fitness of each pixel.

[0052] In the present invention, the solar sky-ground angle and solar azimuth are key factors affecting the formation and distribution of shadows in remote sensing images. Different solar sky-ground angles and solar azimuths will cause the shape, size and position of shadows in remote sensing images to change. Constructing a shadow detection set based on the specific conditions when acquiring images in real time can enhance the adaptability of the shadow detection algorithm to different conditions, thereby more accurately identifying shadow areas in remote sensing images. An accurate shadow detection set provides a reliable foundation for subsequent processing such as target recognition using shadow information. In addition, the construction of the remote sensing digital template takes into account a variety of factors, which can enhance the robustness of the algorithm, enabling it to cope with more complex surface environments and more diverse surface objects. The remote sensing fitness generated by the remote sensing digital template for each pixel can be used as important input information for subsequent classification and recognition processes.

[0053] In the embodiment of the present invention, in S21, the shadow detection set of the real-time aerial remote sensing image The expression is:

[0054] ;

[0055] Where, Indicates the sun angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth angle when collecting real-time aerial remote sensing images. Represents the radiance of real-time aerial remote sensing images, Indicates the radiant emittance of the sensor.

[0056] In this paper, the solar zenith angle refers to the angle between the incident sunlight at a pixel's location and the normal to the ground plane at that pixel at that moment. The solar azimuth angle refers to the angle of the ray (with the vertex being the surface pixel point) extending clockwise from true north to the point where the projection of the incident sunlight at that pixel is located on the Earth's surface at a specific moment. The radiance of a remote sensing image refers to the brightness value reflected on the sensor as a result of the radiant energy received by the remote sensing sensor from the Earth's surface or other target objects. Utilizing this angular information allows for more accurate estimation of the Earth's surface reflectivity and radiative properties, thereby improving the quality and usability of remote sensing images.

[0057] In this embodiment of the present invention, S22 includes the following sub-steps:

[0058] S221, generating a detection matrix according to the shadow detection set;

[0059] S222. Fill the F-norm value of the detection matrix into the center position of the remote sensing digital template to generate a remote sensing digital template.

[0060] In this invention, the F-norm value of the detection matrix is ​​filled into the center position of the remote sensing digital template to achieve an effective representation of shadow features. The detection matrix provides important input information for subsequent processing. By analyzing and processing the detection matrix, shadow features can be further extracted to support tasks such as remote sensing image classification and recognition. Remote sensing digital template The expression is , after filling the F norm value into the center position of the remote sensing digital template, fill the elements of the shadow detection set into the first row in sequence, fill the elements of the shadow detection set into the last row in reverse order, and then fill the elements of the first column of the second row and the third column of the second row, where, represents the F-norm of the detection matrix.

[0061] In the embodiment of the present invention, in S221, the detection matrix The expression is:

[0062] ;

[0063] Where, Indicates the sun angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth angle when collecting real-time aerial remote sensing images. Represents a real-time shadow detection set for aerial remote sensing images The standard deviation of all elements in , Represents the maximum function.

[0064] In the embodiment of the present invention, in S23, each pixel of the real-time aerial remote sensing image is input into the remote sensing digital template, and the shadow adaptability of each pixel is output using the maximum eigenvalue of the remote sensing digital template;

[0065] The location in the real-time aerial remote sensing image is The shadow fitness of the pixel The calculation formula is:

[0066] ;

[0067] Where, represents the horizontal coordinate of the pixel, represents the vertical coordinate of the pixel, Indicates the hue of the pixel, Indicates the mean of all tones of the four neighboring pixels corresponding to the pixel. Indicates the saturation of the pixel, Indicates the mean of all saturation values ​​of the four neighboring pixels corresponding to the pixel. represents the brightness of the pixel, Indicates the mean value of all brightness of the four neighboring pixels corresponding to the pixel. Represents the mean of all hues of the pixel's D neighboring pixels. Represents the mean of all saturation values ​​of the pixel’s D neighboring pixels. Represents the mean brightness of all D neighboring pixels corresponding to the pixel, Indicates the maximum eigenvalue of the remote sensing digital template, Indicates the sun angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth when collecting real-time aerial remote sensing images.

[0068] In the present invention, in the HSV model, hue is expressed as an angle, typically ranging from 0° to 360°; saturation is expressed as the ratio between the purity of the selected color and the maximum purity of the color, typically expressed as a percentage or a normalized value; and lightness indicates the brightness of the color, typically expressed as a percentage or a normalized value. The present invention extracts the elements at the corresponding position in the remote sensing digital template based on the neighboring pixels of each pixel for calculation.

[0069] In this embodiment of the present invention, S3 includes the following sub-steps:

[0070] S31, randomly selecting a number of remote sensing fitnesses from the remote sensing fitnesses of all pixels as the root nodes of each isolated tree to obtain an isolation forest;

[0071] S32, using remote sensing fitness to traverse each isolated tree, and obtain the height of remote sensing fitness in each isolated tree;

[0072] S33, calculating the average height of all isolated trees for each remote sensing fitness;

[0073] S34. Determine whether there is a remote sensing fitness with an average height greater than or equal to a threshold. If so, the detection target exists in the real-time aerial remote sensing image. Otherwise, the detection target does not exist in the real-time aerial remote sensing image, and the scene classification is completed.

[0074] In the present invention, in aerial remote sensing image scene classification, anomalous data points may represent potential target areas. By calculating the remote sensing fitness of each isolated tree and determining whether there are remote sensing fitness values ​​with an average height greater than or equal to a threshold, anomalous data points, i.e., potential targets or regions of interest, can be accurately detected. The threshold can be set manually.

[0075] In the embodiment of the present invention, in S32, the remote sensing fitness is at the height of the isolated tree. The calculation formula is:

[0076] ;

[0077] Where, represents the standard deviation of all remote sensing fitness in the root node of the isolated tree, represents the split value of the isolated tree, represents the maximum depth of the isolation tree, represents a natural constant, Indicates the number of remote sensing fitness in the root node of the isolated tree.

[0078] The split value is the threshold used at each node to split the data into the left or right subtree.

[0079] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for scene classification of aerial remote sensing images based on data processing, characterized in that: The following steps are involved: S1. Collect real-time aerial remote sensing images using sensors; S2, using the shadow detection set to construct a remote sensing digital template, process the pixels of the real-time aerial remote sensing image, and obtain the remote sensing fitness of each pixel; S3, using the isolation forest to process the remote sensing fitness of all pixels to obtain the classification results of the real-time aerial remote sensing image; The S2 comprises the following sub-steps: S21, constructing a shadow detection set for the real-time aerial remote sensing image according to the solar zenith angle and the solar azimuth angle when the real-time aerial remote sensing image is collected; S22, construct remote sensing digital template using shadow detection set; S23, using remote sensing digital templates to process real-time aerial remote sensing images to obtain remote sensing fitness of each pixel; In S21, real-time shadow detection of aerial remote sensing images The expression is: ； In the formula, Indicates the solar zenith angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth when collecting real-time aerial remote sensing images. Represents the radiance of real-time aerial remote sensing images, Indicates the radiant emittance of the sensor; The S22 comprises the following sub-steps: S221, generating a detection matrix according to the shadow detection set; S222, filling the F-norm value of the detection matrix into the center position of the remote sensing digital template to generate a remote sensing digital template; In S23, each pixel of the real-time aerial remote sensing image is input into the remote sensing digital template, and the remote sensing fitness of each pixel is output using the maximum eigenvalue of the remote sensing digital template; The position in the real-time aerial remote sensing image is The remote sensing fitness of the pixel The calculation formula is: ； In the formula, represents the horizontal coordinate of the pixel, represents the vertical coordinate of the pixel, Indicates the hue of the pixel, It indicates the mean value of all tones of the four neighboring pixels corresponding to the pixel. Indicates the saturation of the pixel. It indicates the mean of all saturation values ​​of the four neighboring pixels corresponding to the pixel. represents the brightness of the pixel, It indicates the mean value of all brightness of the four neighboring pixels corresponding to the pixel. It represents the mean value of all hues of the pixel’s D neighboring pixels. It represents the mean of all saturation values ​​of the D neighboring pixels corresponding to the pixel. It represents the mean brightness of all pixels in the D neighboring pixels corresponding to the pixel. Represents the maximum eigenvalue of the remote sensing digital template, Indicates the solar zenith angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth when collecting real-time aerial remote sensing images; The S3 comprises the following sub-steps: S31, randomly selecting a number of remote sensing fitnesses from the remote sensing fitnesses of all pixels as the root nodes of each isolated tree to obtain an isolated forest; S32, traversing each isolated tree using remote sensing fitness to obtain the height of the remote sensing fitness in each isolated tree; S33, calculating the average height of all isolated trees for each remote sensing fitness; S34, judging whether there is a remote sensing fitness whose average height is greater than or equal to the threshold, if so, there is a detection target in the real-time aerial remote sensing image, otherwise there is no detection target in the real-time aerial remote sensing image, and the scene classification is completed.

2. The method for scene classification of aerial remote sensing images based on data processing according to claim 1, characterized in that: In S221, the detection matrix The expression is: ； In the formula, Indicates the solar zenith angle when collecting real-time aerial remote sensing images. Indicates the solar azimuth when collecting real-time aerial remote sensing images. Represents a set of shadow detections for real-time aerial remote sensing images The standard deviation of all elements in , Represents the maximum value function.

3. The method for scene classification of aerial remote sensing images based on data processing according to claim 1, characterized in that: In S32, the remote sensing fitness is the height of the isolated tree The calculation formula is: ； In the formula, represents the standard deviation of all remote sensing fitness in the root node of the isolated tree, represents the split value of the isolated tree, represents the maximum depth of an isolated tree, represents a natural constant, Represents the number of remote sensing fitness in the root node of the isolated tree.

Images