A remote sensing image feature identification method and system based on artificial intelligence

Abstract

The present application relates to the field of remote sensing technology, and in particular to a remote sensing image feature discrimination method and system based on artificial intelligence, comprising: collecting optical remote sensing images and preprocessing the optical remote sensing images; performing an initial judgment on the preprocessed optical remote sensing images to determine whether the optical remote sensing images are affected by weather factors; enhancing the optical remote sensing images affected by weather factors; performing preliminary feature region division on the optical remote sensing images to obtain multiple sub-regions, where each sub-region contains the same type of ground objects; using a pre-trained remote sensing image feature discrimination model, performing feature discrimination on each sub-region to obtain the probability that each sub-region belongs to each ground object category; and determining the ground object feature category of each sub-region based on the probability that each sub-region belongs to each ground object category. The present application performs feature discrimination on a single optical remote sensing image, thereby avoiding misjudgment or missed judgment and improving the accuracy of feature discrimination.

Description

Technical Field

[0001] The present application relates to the field of remote sensing technology, and in particular to a remote sensing image feature identification method and system based on artificial intelligence. Background Art

[0002] Remote sensing imagery is data captured through remote sensing technology, obtained from the Earth's surface or atmosphere and displayed as digital images on a computer screen. These images are primarily captured or recorded by satellites, aircraft, drones, or other remote sensing platforms, capturing various features of the Earth's surface and their reflection or radiation in different spectral bands. With the continuous advancement of remote sensing technology and the widespread use of remote sensing platforms such as satellites and drones, the acquisition of remote sensing imagery will become more convenient and efficient.

[0003] Currently, remote sensing images include optical, radar, and infrared images, each with its own advantages and disadvantages. For example, visible light remote sensing images offer high spatial resolution and rich spectral information, making them suitable for urban land use monitoring and capable of clearly distinguishing buildings, roads, and green spaces. However, they are susceptible to factors such as weather and noise, which can reduce the image quality and lead to misjudgments or omissions when identifying remote sensing image features. Synthetic aperture radar (SAR) images offer all-day, all-weather observation capabilities, enabling the acquisition of high-precision terrain data and the creation of topographic maps in topographic surveying. However, some ground features exhibit similar microwave scattering characteristics, making them difficult to distinguish in radar remote sensing images.

[0004] In order to overcome the fact that optical remote sensing images are easily affected by factors such as weather and noise, optical remote sensing images are usually fused with radar remote sensing images. This combines the rich texture information of optical remote sensing images with the all-weather and penetrating advantages of radar remote sensing images, and can avoid the influence of factors such as weather and noise on the feature discrimination of remote sensing images. However, using a single optical remote sensing image for remote sensing image feature discrimination is still difficult to overcome the phenomenon of misjudgment or missed judgment due to the influence of factors such as weather and noise.

[0005] Therefore, how to avoid misjudgment or missed judgment when performing feature discrimination through a single optical remote sensing image and improve the accuracy of remote sensing image feature discrimination is a technical problem that technicians in this field urgently need to solve. Summary of the Invention

[0006] The present application provides a remote sensing image feature identification method and system based on artificial intelligence, which can perform feature identification through a single optical remote sensing image, and can avoid misjudgment or missed judgment, thereby improving the accuracy of remote sensing image feature identification.

[0007] To solve the above technical problems, this application provides the following technical solutions:

[0008] A remote sensing image feature discrimination method based on artificial intelligence includes the following steps: step T110, collecting optical remote sensing images and preprocessing the optical remote sensing images; step T120, performing an initial judgment on the preprocessed optical remote sensing images to determine whether the optical remote sensing images are affected by weather factors; step T130, enhancing the optical remote sensing images affected by weather factors; step T140, performing preliminary feature region division on the optical remote sensing images to obtain multiple sub-regions, where each sub-region contains the same type of ground objects; step T150, using a pre-trained remote sensing image feature discrimination model, performing feature discrimination on each sub-region to obtain the probability that each sub-region belongs to each ground object category; step T160, determining the ground object feature category of each sub-region based on the probability that each sub-region belongs to each ground object category.

[0009] As described above, in the artificial intelligence-based remote sensing image feature identification method, preferably, the optical remote sensing image is input into an image preprocessing model, and the image preprocessing model performs filtering and noise reduction processing on the optical remote sensing image, thereby preprocessing the optical remote sensing image.

[0010] As described above, in the artificial intelligence-based remote sensing image feature discrimination method, preferably, step T120 includes: calculating a grayscale image of the preprocessed optical remote sensing image; calculating an entropy value of the preprocessed optical remote sensing image based on the grayscale image; comparing the entropy value of the optical remote sensing image with a preset influence threshold value; if the entropy value of the optical remote sensing image is less than the influence threshold value, the optical remote sensing image is affected by weather factors; otherwise, the optical remote sensing image is not affected by weather factors.

[0011] As described above, in the artificial intelligence-based remote sensing image feature identification method, preferably, the optical remote sensing image not affected by weather factors and the enhanced optical remote sensing image are preliminarily divided into feature areas based on spectral features, texture features and shape features.

[0012] As described above, in the artificial intelligence-based remote sensing image feature discrimination method, preferably, the pixel values ​​of a sub-region are flattened into a vector; the vector is input into a pre-trained remote sensing image discrimination model for feature discrimination to output the probability that the vector corresponds to each type of land object; and the probability is used as the probability that the corresponding sub-region belongs to each land object category.

[0013] A remote sensing image feature discrimination system based on artificial intelligence includes: an acquisition device and a ground receiving station; the acquisition device includes: an acquisition unit and a preprocessing unit; the ground receiving station includes: a judgment unit, an enhancement unit, a region division unit, a feature discrimination unit and a determination unit; the acquisition unit acquires optical remote sensing images, and the preprocessing unit preprocesses the optical remote sensing images; the judgment unit performs an initial judgment on the preprocessed optical remote sensing images to determine whether the optical remote sensing images are affected by weather factors; the enhancement unit enhances the optical remote sensing images affected by weather factors; the region division unit performs preliminary feature region division on the optical remote sensing images to obtain multiple sub-regions, and each sub-region contains the same type of ground objects; the feature discrimination unit uses a pre-trained remote sensing image feature discrimination model to perform feature discrimination on each sub-region to obtain the probability that each sub-region belongs to each ground object category; the determination unit determines the ground object feature class of each sub-region based on the probability that each sub-region belongs to each ground object category.

[0014] As described above, in the artificial intelligence-based remote sensing image feature identification system, preferably, the preprocessing unit inputs the optical remote sensing image into the image preprocessing model, and the image preprocessing model performs filtering and noise reduction processing on the optical remote sensing image, thereby preprocessing the optical remote sensing image.

[0015] As described above, in the artificial intelligence-based remote sensing image feature discrimination system, preferably, the judgment unit calculates a grayscale image of the preprocessed optical remote sensing image, calculates an entropy value of the preprocessed optical remote sensing image based on the grayscale image, and compares the entropy value of the optical remote sensing image with a preset influence threshold. If the entropy value of the optical remote sensing image is less than the influence threshold, the optical remote sensing image is affected by weather factors; otherwise, the optical remote sensing image is not affected by weather factors.

[0016] As described above, in the artificial intelligence-based remote sensing image feature identification system, preferably, the region division unit performs preliminary feature region division on the optical remote sensing image not affected by weather factors and the enhanced optical remote sensing image based on spectral features, texture features and shape features.

[0017] As described above, in the artificial intelligence-based remote sensing image feature discrimination system, preferably, the feature discrimination unit flattens the pixel values ​​of a sub-region into a vector, inputs the vector into a pre-trained remote sensing image discrimination model for feature discrimination, and outputs the probability that the vector corresponds to each type of land object, and uses the probability as the probability that the corresponding sub-region belongs to each land object category.

[0018] Compared with the above background technology, since this application processes a single optical remote sensing image and also applies a pre-trained remote sensing image feature discrimination model for feature discrimination, this application can perform feature discrimination through a single optical remote sensing image and avoid misjudgment or missed judgment, thereby improving the accuracy of remote sensing image feature discrimination. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of the artificial intelligence-based remote sensing image feature identification method provided by this application;

[0021] Figure 2 This is a schematic diagram of the artificial intelligence-based remote sensing image feature identification method provided in this application. DETAILED DESCRIPTION

[0022] The following describes embodiments of the present invention in detail, with examples of the embodiments illustrated in the accompanying drawings. Throughout, identical or similar reference numerals denote identical or similar elements or elements having identical or similar functions. Spatially related terms such as "upper," "lower," "front," "rear," "left," and "right" are used to facilitate description and explain the positional relationship between two components. The embodiments described below with reference to the accompanying drawings are illustrative and intended only to explain the present invention, and are not to be construed as limiting the present invention.

[0023] Example 1

[0024] like Figure 1 As shown, this application provides a remote sensing image feature identification method based on artificial intelligence, comprising the following steps:

[0025] Step T110: collecting optical remote sensing images and preprocessing the optical remote sensing images;

[0026] Optical remote sensing images of multiple spectral bands are acquired through optical satellites, or optical remote sensing images of multiple spectral bands with higher resolution are acquired by flying aircraft / UAVs equipped with sensors at medium and low altitudes. The acquired optical remote sensing images are pre-processed by the data processing system on the optical satellite or aircraft / UAV, and the pre-processed optical remote sensing images are transmitted to the ground receiving station.

[0027] Among them, the setting of these spectral bands is based on the differences in the reflection or emission characteristics of ground objects in different spectral bands. For example, vegetation has high reflectivity in the near-infrared band and relatively low reflectivity in the red light band. The combination of these bands can facilitate the identification of vegetation.

[0028] An image preprocessing model is run in the data processing system of the optical satellite or aircraft / UAV. The optical remote sensing image is input into the image preprocessing model, and the image preprocessing model performs filtering and noise reduction on the optical remote sensing image, thereby preprocessing the optical remote sensing image, reducing the random noise in the optical remote sensing image, enhancing the overall consistency of the optical remote sensing image, and making the optical remote sensing image smoother.

[0029] Specifically, the preprocessing model is based on the following formula:

[0030]

[0031] The preprocessed optical remote sensing image is calculated; where (x, y) is the pixel with coordinates x and y in the optical remote sensing image; g(x, y) is the grayscale value of the pixel (x, y) in the optical remote sensing image after filtering and denoising; f(x, y) is the original grayscale value of the pixel (x, y); and M×N is the size of the filter window.

[0032] Step T120: performing an initial judgment on the pre-processed optical remote sensing image to determine whether the optical remote sensing image is affected by weather factors;

[0033] After receiving the pre-processed optical remote sensing image, the ground receiving station's processor also performs an initial evaluation of the pre-processed optical remote sensing image to determine whether the pre-processed optical remote sensing image is affected by weather factors (e.g., heavy rain, heavy snow, dense fog, etc.). If the optical remote sensing image is not affected by weather factors, the process proceeds to step T140; if the optical remote sensing image is affected by weather factors, the process proceeds to step T130.

[0034] The processor of the ground receiving station calculates the grayscale image of the preprocessed optical remote sensing image, and then calculates the entropy value of the preprocessed optical remote sensing image based on the grayscale image. An influence threshold is pre-stored in the memory of the ground receiving station. The processor of the ground receiving station compares the entropy value of the optical remote sensing image with the preset influence threshold. The influence of weather factors will reduce the texture complexity of the optical remote sensing image, and the entropy value will also decrease accordingly. If the entropy value of the optical remote sensing image is less than the influence threshold, it is considered that the optical remote sensing image is affected by weather factors. Otherwise, it is considered that the optical remote sensing image is not affected by weather factors.

[0035] Specifically, the processor uses the following formula:

[0036]

[0037] The entropy value ENT of the preprocessed optical remote sensing image is calculated; where I(x,y) is the grayscale image corresponding to g(x,y); and I(x,y)≠0. If I(x,y) is 0, I(x,y) in this case is discarded; K is the number of pixels in the preprocessed optical remote sensing image.

[0038] Step T130: enhancing the optical remote sensing image affected by weather factors;

[0039] If the processor of the ground receiving station determines that the optical remote sensing image is affected by weather factors such as heavy rain, heavy snow, and dense fog, the processor of the ground receiving station calculates the enhancement value of the preprocessed optical remote sensing image, thereby enhancing the preprocessed optical remote sensing image, and enters step T140 after the optical remote sensing image is enhanced.

[0040] Specifically, the processor uses the following formula:

[0041]

[0042] Enhance the preprocessed optical remote sensing image; where G(x,y) is the optical remote sensing image enhancement value after g(x,y) is enhanced; I(x,y) is the grayscale image corresponding to g(x,y); T is the grayscale threshold, which is used to distinguish the grayscale of multiple uniformly distributed regional images in the optical remote sensing image from the grayscale of other parts of the optical remote sensing image; a is the enhancement coefficient, with a>0; b is the enhancement compensation value, with b>0; A is the atmospheric background light intensity, which is the intensity of the light that comes directly from the light source and reaches the observation point after multiple scattering when the atmosphere has a uniform scattering medium; t(x,y) is a grayscale image of the same size as g(x,y), where the value of each pixel (x,y) represents the ability of light to penetrate the uniform scattering medium at that pixel (x,y), and the value range is between 0 and 1; t0 is the minimum transmittance threshold, usually 0.1, to avoid the denominator being too small.

[0043] Step T140: performing preliminary feature region division on the optical remote sensing image to obtain a plurality of sub-regions, where each sub-region contains the same type of ground objects;

[0044] The processor of the ground receiving station divides the optical remote sensing images that are not affected by weather factors and the enhanced optical remote sensing images into preliminary feature areas based on spectral characteristics, texture characteristics and shape characteristics, so that each optical remote sensing image can obtain multiple sub-areas, and a sub-area contains the same type of ground objects.

[0045] Among them, the objects in optical remote sensing images have different reflection or emission characteristics in different bands. For example, vegetation has a higher reflectivity in the near-infrared band, and water bodies have a higher reflectivity in the green light band. The spectral characteristics are extracted by calculating the vegetation index and water body index. The gray-level co-occurrence matrix (GLCM) is used to calculate the texture features. The texture features reflect the spatial distribution pattern of the objects. For example, rough surface texture corresponds to mountains, and smooth texture corresponds to water bodies or large areas of farmland. For objects with obvious geometric shapes, such as buildings and farmland, shape characteristics are extracted, such as perimeter, area, shape factor, etc.

[0046] Step T150: Use a pre-trained remote sensing image feature discrimination model to perform feature discrimination on each sub-region to obtain the probability that each sub-region belongs to each ground feature category;

[0047] The processor of the ground receiving station flattens the pixel values ​​of a sub-region into a vector z = (z1, z2, ..., z c …、z C ); where z1 is the first element of vector z, z2 is the second element of vector z, and z c is the cth element of vector z, z C is the Cth element of vector z, where C is the length of the vector.

[0048] The processor of the ground receiving station inputs the vector z into a pre-trained remote sensing image discrimination model for feature discrimination, thereby outputting the probability that the vector z corresponds to each type of land object, and using this probability as the probability that the sub-area belongs to each land object category.

[0049] Specifically, the processor uses the following formula:

[0050]

[0051] The calculated vector z=(z1, z2, ..., z c …、z C ) belongs to each feature category; where l e is the probability that vector z belongs to the e-th type of land feature, e ranges from 1 to E, and E is the total type of land feature, so the probability that vector z belongs to the 1st to the Eth type of land feature is (l1, l2, ..., l e …、l E ); The weights from the input layer to the hidden layer of the remote sensing image feature discrimination model; is the bias of the hidden layer of the remote sensing image feature discrimination model; C is the number of neurons in the input layer; μ(x) = max(0,x); The weights from the hidden layer to the output layer of the remote sensing image feature discrimination model; is the bias of the output layer of the remote sensing image feature discrimination model; D is the number of neurons in the hidden layer; is the softmax function.

[0052] Step T160: Determine the feature category of each sub-region based on the probability that each sub-region belongs to each feature category;

[0053] After obtaining the probability that the vector z belongs to the 1st to the Eth type of features (l1, l2, ..., l e …、l E ), the processor of the ground receiving station will convert the probability (l1, l2, ..., l e …、l E ) in the maximum value l max The corresponding land feature category is determined as the land feature category of the sub-region corresponding to the vector z.

[0054] Example 2

[0055] like Figure 2 As shown, the present application provides an artificial intelligence-based remote sensing image feature discrimination system 200, including: an acquisition device 210 and a ground receiving station 220; the acquisition device 210 includes: an acquisition unit 211 and a preprocessing unit 212; the ground receiving station 220 includes: a judgment unit 221, an enhancement unit 222, a region division unit 223, a feature discrimination unit 224 and a determination unit 225, and the discrimination unit 221, the enhancement unit 222, the region division unit 223, the feature discrimination unit 224 and the determination unit 225 all run in the processor of the ground receiving station 220.

[0056] The acquisition unit 211 acquires optical remote sensing images, and the pre-processing unit 212 pre-processes the optical remote sensing images.

[0057] Optical remote sensing images of multiple spectral bands are acquired by the acquisition unit 211 of the acquisition device 210 of the optical satellite, or optical remote sensing images of multiple spectral bands with relatively high resolution are acquired by the acquisition unit 211 of the acquisition device carried by the aircraft / UAV flying at medium or low altitude, and the acquired optical remote sensing images are pre-processed by the pre-processing unit 212 of the acquisition device 210 on the optical satellite or the aircraft / UAV, and the pre-processed optical remote sensing images are transmitted to the ground receiving station 220.

[0058] Among them, the setting of these spectral bands is based on the differences in the reflection or emission characteristics of ground objects in different spectral bands. For example, vegetation has high reflectivity in the near-infrared band and relatively low reflectivity in the red light band. The combination of these bands can facilitate the identification of vegetation.

[0059] An image preprocessing model runs in the preprocessing unit 212 of the acquisition device 210 on the optical satellite or aircraft / UAV, and the optical remote sensing image is input into the image preprocessing model. The image preprocessing model performs filtering and noise reduction processing on the optical remote sensing image, thereby preprocessing the optical remote sensing image, reducing random noise in the optical remote sensing image, enhancing the overall consistency of the optical remote sensing image, and making the optical remote sensing image smoother.

[0060] Specifically, the preprocessing model is based on the following formula:

[0061]

[0062] The preprocessed optical remote sensing image is calculated; where (x, y) is the pixel with coordinates x and y in the optical remote sensing image; g(x, y) is the grayscale value of the pixel (x, y) in the optical remote sensing image after filtering and denoising; f(x, y) is the original grayscale value of the pixel (x, y); and M×N is the size of the filter window.

[0063] The judgment unit 221 performs an initial judgment on the pre-processed optical remote sensing image to determine whether the optical remote sensing image is affected by weather factors.

[0064] After receiving the pre-processed optical remote sensing image, the ground receiving station's processor also performs an initial assessment of the image to determine whether it has been affected by weather factors (e.g., heavy rain, snow, dense fog, etc.). If the image is not affected by weather factors, the image is initially segmented into feature regions. If the image is affected by weather factors, the image is enhanced.

[0065] The processor of the ground receiving station calculates the grayscale image of the preprocessed optical remote sensing image, and then calculates the entropy value of the preprocessed optical remote sensing image based on the grayscale image. An influence threshold is pre-stored in the memory of the ground receiving station. The processor of the ground receiving station compares the entropy value of the optical remote sensing image with the preset influence threshold. The influence of weather factors will reduce the texture complexity of the optical remote sensing image, and the entropy value will also decrease accordingly. If the entropy value of the optical remote sensing image is less than the influence threshold, it is considered that the optical remote sensing image is affected by weather factors. Otherwise, it is considered that the optical remote sensing image is not affected by weather factors.

[0066] Specifically, the processor uses the following formula:

[0067]

[0068] The entropy value ENT of the preprocessed optical remote sensing image is calculated; where I(x,y) is the grayscale image corresponding to g(x,y); and I(x,y)≠0. If I(x,y) is 0, I(x,y) in this case is discarded; K is the number of pixels in the preprocessed optical remote sensing image.

[0069] The enhancement unit 222 enhances the optical remote sensing image affected by weather factors.

[0070] If the processor of the ground receiving station determines that the optical remote sensing image is affected by weather factors such as heavy rain, heavy snow, and dense fog, the processor of the ground receiving station calculates the enhancement value of the preprocessed optical remote sensing image, thereby enhancing the preprocessed optical remote sensing image, and after the optical remote sensing image is enhanced, the optical remote sensing image is preliminarily divided into feature areas.

[0071] Specifically, the processor uses the following formula:

[0072]

[0073] Enhance the preprocessed optical remote sensing image; where G(x,y) is the optical remote sensing image enhancement value after g(x,y) is enhanced; I(x,y) is the grayscale image corresponding to g(x,y); T is the grayscale threshold, which is used to distinguish the grayscale of multiple uniformly distributed regional images in the optical remote sensing image from the grayscale of other parts of the optical remote sensing image; a is the enhancement coefficient, with a>0; b is the enhancement compensation value, with b>0; A is the atmospheric background light intensity, which is the intensity of the light that comes directly from the light source and reaches the observation point after multiple scattering when the atmosphere has a uniform scattering medium; t(x,y) is a grayscale image of the same size as g(x,y), where the value of each pixel (x,y) represents the ability of light to penetrate the uniform scattering medium at that pixel (x,y), and the value range is between 0 and 1; t0 is the minimum transmittance threshold, usually 0.1, to avoid the denominator being too small.

[0074] The region division unit 223 performs preliminary feature region division on the optical remote sensing image to obtain a plurality of sub-regions, where each sub-region contains the same type of ground objects.

[0075] The processor of the ground receiving station divides the optical remote sensing images that are not affected by weather factors and the enhanced optical remote sensing images into preliminary feature areas based on spectral characteristics, texture characteristics and shape characteristics, so that each optical remote sensing image can obtain multiple sub-areas, and a sub-area contains the same type of ground objects.

[0076] Among them, the objects in optical remote sensing images have different reflection or emission characteristics in different bands. For example, vegetation has a higher reflectivity in the near-infrared band, and water bodies have a higher reflectivity in the green light band. The spectral characteristics are extracted by calculating the vegetation index and water body index. The gray-level co-occurrence matrix (GLCM) is used to calculate the texture features. The texture features reflect the spatial distribution pattern of the objects. For example, rough surface texture corresponds to mountains, and smooth texture corresponds to water bodies or large areas of farmland. For objects with obvious geometric shapes, such as buildings and farmland, shape characteristics are extracted, such as perimeter, area, shape factor, etc.

[0077] The feature discrimination unit 224 uses a pre-trained remote sensing image feature discrimination model to perform feature discrimination on each sub-region to obtain the probability that each sub-region belongs to each ground feature category.

[0078] The processor of the ground receiving station flattens the pixel values ​​of a sub-region into a vector z = (z1, z2, ..., z c …、z C ); where z1 is the first element of vector z, z2 is the second element of vector z, and z c is the cth element of vector z, z C is the Cth element of vector z, where C is the length of the vector.

[0079] The processor of the ground receiving station inputs the vector z into a pre-trained remote sensing image discrimination model for feature discrimination, thereby outputting the probability that the vector z corresponds to each type of land object, and using this probability as the probability that the sub-area belongs to each land object category.

[0080] Specifically, the processor uses the following formula:

[0081]

[0082] The calculated vector z=(z1, z2, ..., z c …、z C ) belongs to each feature category; where l e is the probability that vector z belongs to the e-th type of land feature, e ranges from 1 to E, and E is the total type of land feature, so the probability that vector z belongs to the 1st to the Eth type of land feature is (l1, l2, ..., l e …、l E ); The weights from the input layer to the hidden layer of the remote sensing image feature discrimination model; is the bias of the hidden layer of the remote sensing image feature discrimination model; C is the number of neurons in the input layer; μ(x) = max(0,x); The weights from the hidden layer to the output layer of the remote sensing image feature discrimination model; is the bias of the output layer of the remote sensing image feature discrimination model; D is the number of neurons in the hidden layer; is the softmax function.

[0083] The determination unit 225 determines the ground feature class of each sub-region based on the probability that each sub-region belongs to each ground feature class.

[0084] After obtaining the probability that the vector z belongs to the 1st to the Eth type of features (l1, l2, ..., l e …、l E ), the processor of the ground receiving station will convert the probability (l1, l2, ..., l e …、l E ) in the maximum value l max The corresponding land feature category is determined as the land feature category of the sub-region corresponding to the vector z.

[0085] Since the present application processes a single optical remote sensing image and also applies a pre-trained remote sensing image feature discrimination model for feature discrimination, the present application can perform feature discrimination through a single optical remote sensing image and avoid the phenomenon of misjudgment or missed judgment, thereby improving the accuracy of remote sensing image feature discrimination.

[0086] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

[0087] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A remote sensing image feature identification method based on artificial intelligence, characterized in that: The steps include: Step T110: collecting optical remote sensing images and preprocessing the optical remote sensing images; Step T120: performing an initial judgment on the pre-processed optical remote sensing image to determine whether the optical remote sensing image is affected by weather factors; Step T120 includes: Calculate the grayscale image of the preprocessed optical remote sensing image; Calculate the entropy value of the preprocessed optical remote sensing image based on the grayscale image; Comparing the entropy value of the optical remote sensing image with a preset influence threshold, if the entropy value of the optical remote sensing image is less than the influence threshold, the optical remote sensing image is affected by weather factors, otherwise the optical remote sensing image is not affected by weather factors; Step T130: enhancing the optical remote sensing image affected by weather factors; Step T140: performing preliminary feature region division on the optical remote sensing image to obtain a plurality of sub-regions, where each sub-region contains the same type of ground objects; Step T150: Use a pre-trained remote sensing image feature discrimination model to perform feature discrimination on each sub-region to obtain the probability that each sub-region belongs to each ground feature category; Step T160: Determine the ground feature category of each sub-region based on the probability that each sub-region belongs to each ground feature category.

2. The method for distinguishing remote sensing image features based on artificial intelligence according to claim 1, characterized in that: The optical remote sensing image is input into the image preprocessing model, and the image preprocessing model performs filtering and noise reduction on the optical remote sensing image, thereby preprocessing the optical remote sensing image.

3. The method for distinguishing remote sensing image features based on artificial intelligence according to claim 1 or 2, characterized in that: Based on spectral characteristics, texture characteristics and shape characteristics, the optical remote sensing images not affected by weather factors and the enhanced optical remote sensing images are preliminarily divided into feature areas.

4. The method for distinguishing remote sensing image features based on artificial intelligence according to claim 1 or 2, characterized in that: Flatten the pixel values ​​of a sub-region into a vector; Input the vector into a pre-trained remote sensing image discrimination model for feature discrimination to output the probability that the vector corresponds to each ground object type; The probability is used as the probability that the corresponding sub-region belongs to each ground feature category.

5. A remote sensing image feature recognition system based on artificial intelligence, characterized in that: include: Collection equipment and ground receiving stations; The acquisition equipment includes: an acquisition unit and a pre-processing unit; the ground receiving station includes: a judgment unit, an enhancement unit, a region division unit, a feature discrimination unit and a determination unit; The acquisition unit acquires the optical remote sensing image, and the preprocessing unit preprocesses the optical remote sensing image; The judgment unit performs an initial judgment on the pre-processed optical remote sensing image to determine whether the optical remote sensing image is affected by weather factors; The judgment unit calculates a grayscale image of the preprocessed optical remote sensing image, calculates an entropy value of the preprocessed optical remote sensing image based on the grayscale image, and compares the entropy value of the optical remote sensing image with a preset influence threshold value. If the entropy value of the optical remote sensing image is less than the influence threshold value, the optical remote sensing image is affected by weather factors; otherwise, the optical remote sensing image is not affected by weather factors. The enhancement unit enhances the optical remote sensing images affected by weather factors; The region division unit performs preliminary feature region division on the optical remote sensing image to obtain multiple sub-regions, where each sub-region contains the same type of ground objects; The feature discrimination unit uses a pre-trained remote sensing image feature discrimination model to perform feature discrimination on each sub-region and obtain the probability that each sub-region belongs to each ground feature category; The determination unit determines the ground feature class of each sub-region according to the probability that each sub-region belongs to each ground feature class.

6. The artificial intelligence-based remote sensing image feature identification system according to claim 5, characterized in that: The preprocessing unit inputs the optical remote sensing image into the image preprocessing model, and the image preprocessing model performs filtering and noise reduction processing on the optical remote sensing image, thereby preprocessing the optical remote sensing image.

7. The artificial intelligence-based remote sensing image feature identification system according to claim 5 or 6, characterized in that: The regional division unit divides the optical remote sensing images that are not affected by weather factors and the enhanced optical remote sensing images into preliminary characteristic regions based on spectral characteristics, texture characteristics and shape characteristics.

8. The artificial intelligence-based remote sensing image feature identification system according to claim 5 or 6, characterized in that: The feature discrimination unit flattens the pixel values ​​of a sub-region into a vector, inputs the vector into a pre-trained remote sensing image discrimination model for feature discrimination, and outputs the probability that the vector corresponds to each type of land object, and uses the probability as the probability that the corresponding sub-region belongs to each land object category.

Images