A shadow recovery method and system for high spatial resolution remote sensing images

Abstract

A high spatial resolution remote sensing image shadow restoration method includes: inputting a remote sensing image containing three visible light bands; converting it to the HIS color space; calculating the normalized shadow index; performing image segmentation on the normalized shadow index, extracting and optimizing the shadow mask; identifying and numbering the shadow blocks in the shadow mask; creating an adaptive rectangular window W that covers all shadow pixels of a certain shadow block; performing linear regression analysis on the spectral values ​​of non-shadow pixels and shadow pixels in the window; and using the linear regression equation to restore shadows to all shadow blocks in all bands of the remote sensing image. This high spatial resolution remote sensing image shadow restoration method overcomes the shortcomings of a single statistical measure in accurately characterizing the spectral statistical characteristics between non-shadow and shadow areas, and fully considers the differences in background features in different shadow areas. The shadow restoration results have high fidelity and are simple, stable, reliable, and computationally efficient.

Description

Technical Field

[0001] This invention relates to the field of digital image processing technology in photogrammetry and remote sensing, and particularly to a method and system for shadow restoration of high spatial resolution remote sensing images. Background Technology

[0002] High spatial resolution remote sensing imagery (HDSA) from sources such as IKONOS, QuickBird, SPOT, Pleiades, WorldView, domestically developed high-resolution satellites, and low-altitude UAVs plays a crucial role in resource and environmental remote sensing due to its refined ground feature resolution capabilities. Utilizing HDSA imagery for urban resource and environmental monitoring is an important tool for urban resource and environmental planning, protection, and management. However, due to the widespread presence of building shadows in urban areas, HDSA imagery often contains a large amount of shadow, severely restricting the efficient use of HDSA remote sensing data. Urban shadows appear as low-brightness areas in remote sensing imagery and mask the radiometric characteristics of ground features within these shadowed areas. When using HDSA imagery for urban land use classification and resource surveys, the presence of shadowed areas significantly reduces the accuracy of ground feature classification, thereby reducing data utilization efficiency. Furthermore, the acquisition cost of urban HDSA imagery is typically high, and lower data utilization efficiency inevitably increases data usage costs. Furthermore, since urban shadows are caused by fixed structures such as buildings blocking direct solar radiation, and the variations in solar azimuth and altitude angles are also fixed, certain shadowed areas will persist in remote sensing images from different periods, thus forming permanent blind spots in urban remote sensing monitoring. Therefore, shadow restoration of high-resolution remote sensing images has significant practical implications.

[0003] Classic methods for shadow restoration in remote sensing images include histogram matching, gamma correction, and mean and variance transformation (MVT). Histogram matching uses histograms to match the same features in the shadowed area with those in the adjacent unshaded area, and then applies the matching parameters to the entire shadowed area to achieve shadow correction. This method is sensitive to the window size used for histogram matching and has a significant impact on the shadow correction results. Gamma correction uses a single gamma factor for the entire shadowed area, ignoring the differences in background features between different shadowed areas. The mean and variance transformation (MVT) assumes a linear relationship between the shadowed and unshaded areas, constructing this linear relationship by extracting statistical variables (mean and variance) from each region to achieve shadow correction. MVT, with its high accuracy and ease of operation, has significant advantages over other shadow correction methods and is widely used. However, the spectral values ​​between shaded and unshaded areas are not strictly linearly correlated, and their correlation is difficult to accurately describe using only the mean and variance. This is a shortcoming of the MVT method. Furthermore, the MVT method uses statistical parameters of both shaded and unshaded areas of the entire image to correct for different shaded objects, thus neglecting the differences in background features across different shaded areas. This aspect also needs improvement. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned shortcomings of the prior art and, based on adaptive window analysis, provide a simpler, more efficient, and more adaptable method and apparatus for shadow restoration of high spatial resolution remote sensing images.

[0005] The technical solution of this invention is as follows: First, a method for shadow restoration of high spatial resolution remote sensing images based on adaptive window analysis is provided, which includes the following steps:

[0006] Step 1: Input Remote Sensing Image

[0007] Input a high-resolution remote sensing image image-A containing three visible light bands: red (R), green (G), and blue (B).

[0008] Step 2: HIS color space transformation

[0009] The conversion formula for RGB color space to HIS color space is as follows:

[0010]

[0011]

[0012]

[0013] In formulas 1, 2, and 3, H, I, and S represent hue, brightness, and saturation in the HIS color space, respectively, and R, G, and B represent the DN values ​​corresponding to the red (R), green (G), and blue (B) bands of the remote sensing image A, respectively.

[0014] Step 3: Calculate the normalized shading index

[0015]

[0016] In Formula 4, NDUI is the normalized shading index, and I and S are the brightness and saturation in the HIS color space, respectively.

[0017] Step 4: Shadow Detection and Extraction

[0018] Based on the Normalized Shadow Index (NDUI), a shadow segmentation threshold T is determined by analyzing its mean and standard deviation, and through manual interactive interpretation. The value of T is typically between 0.31 and 0.36. Image segmentation is then performed using the NDUI threshold to extract shadow region information from the remote sensing image, resulting in the shadow mask Mask1.

[0019] Step 5: Shadow Mask Filtering

[0020] Dilation and erosion operations are performed sequentially on the shadow mask Mask1. Through the morphological filtering described above, small patches in the shadow mask Mask1 are eliminated, resulting in the optimized shadow mask Mask2.

[0021] Step 6: Shaded Block Identification

[0022] Identify and number (i = 1, 2, 3, ..., n) all connected shadow blocks in the shadow mask Mask2 with more than 15 shadow pixels, and ensure that each shadow block is not connected to the others.

[0023] Step 7: Determine the statistical analysis window

[0024] Let the size of the rectangular image R corresponding to the shaded area numbered i be N1×M1. An adaptive rectangular window W (i.e., a statistical analysis window) of size N2×M2 is created centered on the geometric center of this rectangular image. The rectangular window W must satisfy two conditions: first, the rectangular window W must cover all pixels of the shaded area, i.e., N2≥N1 and M2≥M1; second, the number of corresponding non-shaded pixels in the adaptive rectangular window W must be greater than or equal to 50% of all pixels in the entire adaptive rectangular window W. If this condition is not met, the window size of the adaptive rectangular window W can be gradually increased until the condition is met; for example, the values ​​of N2 and M2 can be increased sequentially by 1 each time.

[0025] Step 8: Linear Regression Analysis

[0026] For a specific band in a remote sensing image, the histograms, mean μ, and standard deviation σ of all shaded and non-shaded pixels within an adaptive rectangular window are statistically analyzed. A shaded pixel spectral vector X(1,2,…,7) is constructed by taking different spectral values ​​of shaded pixels from low to high according to a specific rule, such that the values ​​of X1, X2,…,X7 are equal to μ-2σ, μ-1.5σ, μ-σ, μ, μ+σ, μ+1.5σ, and μ+2σ, respectively. Based on the histogram of the shaded pixels, the cumulative percentage vector AP1(1,2,…,7) corresponding to the shaded pixel spectral vector X(1,2,…,7) is obtained. Based on the histogram of the non-shaded pixels, the spectral value closest to the cumulative percentage vector AP1(1,2,…,7) of the shaded pixels is searched to obtain the cumulative percentage vector AP2(1,2,…,7) of the non-shaded pixels, and the corresponding non-shaded pixel spectral vector Y(1,2,…,7) is constructed. Linear regression analysis is performed on the non-shaded pixel spectral vector Y(1,2,…,7) and the shaded pixel spectral vector X(1,2,…,7) of a certain window to obtain the values ​​of the parameters k and b in the linear regression equation Y=k×X+b for that window. Each of the n different shaded blocks corresponds to a unique linear regression equation.

[0027] Step 9: Shadow Restoration

[0028] For a given adaptive rectangular window, the linear regression equation Y = k*X + b is used to perform shadow restoration on a shadow block in the remote sensing image image-A.

[0029] Similarly, row shadow restoration is performed on all shadow blocks in the three bands of red (R), green (G) and blue (B) in the remote sensing image image-A to obtain the shadow-restored remote sensing image-B.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: The shadow restoration method for high spatial resolution remote sensing images based on adaptive window analysis provided by the present invention makes up for the shortcomings of a single statistical quantity in accurately characterizing the spectral statistical characteristics between non-shadow areas and shadow areas, and fully considers the differences in background land features in different shadow areas. The shadow restoration results have high fidelity and have the advantages of simple and easy-to-implement algorithm, stability and reliability, and high computational efficiency.

[0031] The present invention also provides an apparatus for implementing a high spatial resolution remote sensing image shadow restoration method, comprising:

[0032] The image acquisition module takes a high-resolution remote sensing image, image-A, that contains three visible light bands: red (R), green (G), and blue (B).

[0033] The HIS color space conversion module converts the RGB color space to the HIS color space.

[0034] The normalized shading index calculation module calculates the normalized shading index.

[0035] The shadow detection and extraction module uses a threshold T to segment the image based on the normalized shadow index, thereby extracting shadow region information from the remote sensing image and obtaining the shadow mask Mask1.

[0036] The shadow mask filtering module performs dilation and erosion operations on the shadow mask Mask1 sequentially, and obtains the optimized shadow mask Mask2 through morphological filtering.

[0037] The shadow block identification module identifies and numbers all connected shadow blocks in the shadow mask Mask2 with more than 15 pixels (i = 1, 2, 3, ..., n), and ensures that each shadow block is not connected to others.

[0038] Determine the statistical analysis window module, and create an adaptive rectangular window W with the geometric center of the rectangular image corresponding to the shaded area as the center, so that the rectangular window W covers all pixels of the shaded area and the number of non-shaded pixels is greater than or equal to 50% of the total number of pixels in the rectangular window.

[0039] The linear regression analysis module performs linear regression analysis on the non-shaded pixel spectral vector Y(1,2,…,7) and the shaded pixel spectral vector X(1,2,…,7) in the statistical analysis window to obtain the values ​​of the equation parameters k and b of the linear regression equation Y=k×X+b for this window.

[0040] The shadow restoration module uses a linear regression equation to restore shadows in all shadow blocks of all bands in the remote sensing image image-A, resulting in the shadow-restored remote sensing image image-B.

[0041] The aforementioned high spatial resolution remote sensing image shadow restoration system is used to implement the aforementioned high spatial resolution remote sensing image shadow restoration method, and can specifically achieve the beneficial effects corresponding to the method, which will not be elaborated here. Attached Figure Description

[0042] Figure 1 This is a flowchart of the high spatial resolution remote sensing image shadow restoration method in an embodiment of the present invention;

[0043] Figure 2 Image-A is a high spatial resolution remote sensing image in this embodiment of the invention.

[0044] Figure 3The normalized shadow index of the high spatial resolution remote sensing image in this embodiment of the invention;

[0045] Figure 4 This refers to the shadow mask optimized from high spatial resolution remote sensing images in this embodiment of the invention.

[0046] Figure 5 This is a schematic diagram of the shadow block identifier and statistical analysis window in an embodiment of the present invention;

[0047] Figure 6 This is a schematic diagram of linear regression analysis of non-shadow pixels and shadow pixels in an embodiment of the present invention;

[0048] Figure 7 Image-B is the remote sensing image after shadow restoration in this embodiment of the invention.

[0049] Figure 8 This is a schematic diagram of the high spatial resolution remote sensing image shadow restoration system of the present invention. Detailed Implementation

[0050] The present invention will be further described in detail below with reference to specific embodiments. Methods or functional components not specifically described in the embodiments are all prior art.

[0051] Example

[0052] like Figure 1-8 As shown, this invention provides a method for shadow restoration of high spatial resolution remote sensing images based on adaptive window analysis, comprising the following steps:

[0053] Step 1: Input Remote Sensing Image

[0054] Input a high-resolution remote sensing image image-A containing three visible light bands: red (R), green (G), and blue (B).

[0055] This embodiment uses a true-color, high-resolution remote sensing image of a certain urban area acquired by the Pleiades neo satellite, with a spatial resolution of 0.5 meters. A portion of this area is cut out as input data for this embodiment of the invention.

[0056] The Pleiades satellites are the successors to the SPOT satellite family and are high-resolution commercial remote sensing satellites owned by the French company Astrium. Pleiades 1 was successfully launched on December 17, 2011, and began commercial operation, boasting an ultra-high spatial resolution of 0.5 meters and a swath width of 20 km x 20 km. It offers panchromatic (P) imagery and four multispectral MS bands (B, G, R, NIR). Pleiades 2 was successfully launched on December 1, 2012. Pleiades neo 3 was successfully launched on April 28, 2021, and Pleiades neo 4 was successfully launched on August 16, 2021, from the European Ariane Vega launch vehicle. Satellites 3 and 4 are in the same orbit with a 180° phase, forming a new constellation. Satellites 3 and 4 provide 30 cm panchromatic and 1.2 m multispectral imagery.

[0057] Specifically, such as Figure 2 The image shown is a true-color high spatial resolution remote sensing image of a certain urban area in this embodiment.

[0058] Step 2: HIS color space transformation

[0059] The conversion formula for RGB color space to HIS color space is as follows:

[0060]

[0061]

[0062]

[0063] In formulas 1, 2, and 3, H, I, and S represent hue, brightness, and saturation in the HIS color space, respectively, while R, G, and B represent the DN values ​​corresponding to the red (R), green (G), and blue (B) bands of the remote sensing image A.

[0064] Step 3: Calculate the normalized shading index

[0065]

[0066] In Formula 4, NDUI is the normalized shading index, and I and S are the brightness and saturation in the HIS color space, respectively.

[0067] Specifically, such as Figure 3 The image shown is the normalized shadow index of the remote sensing image in an embodiment of the present invention.

[0068] Step 4: Shadow Detection and Extraction

[0069] Based on the Normalized Shadow Index (NDUI), the shadow segmentation threshold T is determined by its mean and standard deviation characteristics and through manual interactive interpretation. The value of T is usually between 0.31 and 0.36.

[0070] In the embodiment of the invention, the threshold T is set to 0.33. All pixels with NDUI values ​​greater than 0.33 are extracted as shadow pixels through image segmentation to obtain the shadow mask Mask1.

[0071] Step 5: Shadow Mask Filtering

[0072] Dilation and erosion operations are performed sequentially on the shadow mask Mask1. Through the morphological filtering described above, small patches in the shadow mask Mask1 are eliminated, resulting in the optimized shadow mask Mask2.

[0073] Specifically, such as Figure 4 The image shown is the shadow mask Mask2 after optimization of remote sensing imagery in an embodiment of the present invention.

[0074] Step 6: Shaded Block Identification

[0075] Identify and number (i = 1, 2, 3, ..., n) all connected shadow blocks in the shadow mask Mask2 with more than 15 shadow pixels, and ensure that each shadow block is not connected to the others.

[0076] Step 7: Determine the statistical analysis window

[0077] Let the size of the rectangular image R corresponding to the shaded area numbered i be N1×M1. Create an adaptive rectangular window W of size N2×M2 centered on the geometric center of the rectangular image. The rectangular window W must satisfy two conditions: First, the rectangular window W must cover all pixels of the shaded area, i.e., N2≥N1 and M2≥M1; Second, the number of corresponding non-shaded pixels in the adaptive rectangular window W must be greater than or equal to 50% of the total number of pixels in the rectangular window. If this condition is not met, the window size of the adaptive rectangular window W can be gradually increased until the condition is met.

[0078] Specifically, such as Figure 5 The diagram shown is a schematic of the shaded block identifier and statistical analysis window in an embodiment of the present invention.

[0079] Step 8: Linear Regression Analysis

[0080] For a specific band in a remote sensing image, the histograms, mean μ, and standard deviation σ of all shaded and unshaded pixels within an adaptive rectangular window are statistically analyzed.

[0081] Specifically, the histograms and statistical characteristics of shadowed and non-shadowed pixels in the red band in the statistical analysis window of this implementation case are as follows: Figure 6As shown in Table 1, the mean μ of shaded pixels in the statistical analysis window was 23.32 and the standard deviation σ was 5.72, while the mean μ of unshaded pixels was 135.52 and the standard deviation σ was 55.32.

[0082] Table 1. Spectral statistical characteristics of shaded and unshaded pixels in a statistical analysis window.

[0083]

[0084] A shadow pixel spectral vector X(1,2,…,7) is established by taking different spectral values ​​of shadow pixels from low to high according to specific rules, such that the values ​​of X1, X2,…, X7 are equal to μ-2σ, μ-1.5σ, μ-σ, μ, μ+σ, μ+1.5σ, and μ+2σ, respectively, with specific values ​​of 11.88, 14.74, 17.6, 23.32, 29.04, 31.9, and 34.76, as shown in Table 1. The cumulative percentage vector AP1(1,2,…,7) corresponding to the shadow pixel spectral vector X(1,2,…,7) can be obtained from the shadow pixel histogram, with specific values ​​of 0.18%, 1.79%, 19.25%, 66.16%, 88.71%, 92.97%, and 95.86%. Based on the histogram statistics of non-shaded pixels, the spectral value closest to the cumulative percentage value vector AP1(1,2,…,7) is used to obtain the cumulative percentage value vector AP2(1,2,…,7) for non-shaded pixels, with specific values ​​of 0.19%, 1.78%, 19.37%, 66.05%, 88.83%, 92.98%, and 95.96%. The corresponding non-shaded pixel spectral vectors for the cumulative percentage value vector AP2(1,2,…,7) are Y(1,2,…,7), with specific values ​​of 2, 29, 86, 153, 210, 230, and 249. Linear regression analysis is performed on the non-shaded pixel spectral vector Y(1,2,…,7) and the shaded pixel spectral vector X(1,2,…,7) of a certain statistical analysis window, yielding the linear regression equation Y = 11.086 × X - 121.53, with a correlation coefficient R. 2 =0.9838. Each of the n different shaded areas corresponds to a unique linear regression equation.

[0085] Specifically, such as Figure 6 The diagram shown is a schematic diagram of linear regression analysis of the non-shaded pixel spectral vector Y and the shaded pixel spectral vector X in an embodiment of the present invention.

[0086] Step 9, Shadow Restoration

[0087] For a certain statistical analysis window, the linear regression equation Y = k × X + b is used to restore the shadow of a certain shadow block in the remote sensing image.

[0088] Similarly, shadow restoration is performed on all shadow blocks in the three bands of red (R), green (G), and blue (B) of remote sensing image A to obtain shadow-restored remote sensing image B.

[0089] Specifically, such as Figure 7 The image shown is image-B after shadow restoration in an embodiment of the present invention.

[0090] To evaluate the shadow correction effect of the method of this invention, an analysis is conducted from both visual and quantitative perspectives. Common image visual features such as image color, texture, hue, and contrast can serve as important indicators for the qualitative evaluation of shadow correction results. If the shadow correction effect is good, these image visual features in the shadow area should have good consistency with those in the adjacent non-shaded area, and vice versa. The same ground feature in the shadow area and the adjacent non-shaded area should have the same reflectance. Therefore, after shadow correction, the pixel value of the same ground feature in the shadow area should be statistically the same as the pixel value of the same ground feature in the adjacent non-shaded area. Based on this characteristic, the relative root-mean-square error (rRMSE) can be used to quantitatively evaluate the shadow correction accuracy for a certain ground feature.

[0091]

[0092] In Formula 5, N represents the number of pixels randomly selected from a certain type of land cover in the shaded area after shading correction. L represents the value of the i-th pixel in the selected shadow pixels. ref This represents the average pixel value of the land cover in the non-shaded area, and is used as a reference value for the land cover in the shaded area after shading correction.

[0093] For comparative studies, the method of this invention and the widely used Mean-Variance Transform (MVT) method were used to perform shadow correction and restoration on the high-resolution remote sensing images in the case study. Reference 1 (Nakajima, T., Guo, T. and Yasuoka, Y., 2002. Simulated recovery of information in shadow areas on IKONOS image by combing ALS data, The 23rd Asian Conference on Remote Sensing, Kathmandu, Nepal.) provides the calculation formula and specific operation method of the MVT method.

[0094] In terms of visual evaluation, the original image contained a large number of urban building shadows, with most of the shadowed areas consisting of urban roads and a small amount of green belts. After shadow correction using the method of this invention and the MVT method, the pixel values ​​in the shadowed areas were well compensated, and the image brightness was basically at the same level as the non-shadowed areas. Through observation of local details in the corrected image, it was found that compared with the MVT shadow correction results, the image features such as color and texture in the shadowed areas corrected by the method of this invention have better consistency with the non-shadowed areas. In addition, compared with the MVT method, the shadow details corrected by the method of this invention are clearer and richer, and the transition with the non-shadowed areas is smoother and more natural.

[0095] In terms of quantitative evaluation, two typical shadow-covered features—asphalt road surface and green belt—were selected from high-resolution remote sensing images. The accuracy of shadow correction was quantitatively evaluated by calculating the rRMSE values ​​after shadow correction using the method of this invention and the MVT method. The statistical results are shown in Table 2.

[0096] Table 2. rRMSE values ​​of typical land features in the shaded area before and after shadow correction using the method of the present invention and the MVT method.

[0097]

[0098] For asphalt pavements, the shading correction error (rRMSE) of the method of this invention ranges from 4% to 10%, with an average error of 6%; while the shading correction error of the MVT method ranges from 12% to 21%, with an average error of 16%. For green belts, the minimum shading correction error of the method of this invention is 12% (blue light band), the maximum error is 26% (red light band), and the average error is 16%; relatively speaking, the MVT method performs slightly worse, with a shading correction error range of 16% to 24% and an average error of 18%. The method of this invention has the following characteristics: it has high correction accuracy for shading areas of asphalt pavements (the most important feature in urban shading areas); however, its correction accuracy is relatively low for green belts, especially in the red light band; in terms of overall accuracy, the method of this invention is superior to the MVT method.

[0099] As can be seen from the above description, the high spatial resolution remote sensing image shadow restoration method of the present invention has technical advantages such as simple and easy-to-implement algorithm, stability and reliability, high computational efficiency, and excellent results. From a market application perspective, it is easy to embed into various remote sensing image processing software as a functional module, and has broad application prospects and market potential.

[0100] The method of this invention can also be applied to shadow restoration processing of high spatial resolution remote sensing images in the visible light, near-infrared and short-wave infrared bands. The operation steps and methods are similar to those for remote sensing images in the visible light, near-infrared and short-wave infrared bands, and will not be repeated here.

[0101] This invention also provides an apparatus for implementing a high spatial resolution remote sensing image shadow restoration method. (See reference) Figure 8 The implementing device includes:

[0102] Image acquisition module 101 takes a high-resolution remote sensing image image-A containing three visible light bands: red (R), green (G), and blue (B).

[0103] HIS color space conversion module 102 converts RGB color space to HIS color space.

[0104] The normalized shading index calculation module 103 calculates the normalized shading index.

[0105] The shadow detection and extraction module 104 uses a threshold T to segment the normalized shadow index, thereby extracting shadow region information from the remote sensing image and obtaining the shadow mask Mask1.

[0106] The shadow mask filtering module 105 performs dilation and erosion operations on the shadow mask Mask1 sequentially, and obtains the optimized shadow mask Mask2 through morphological filtering.

[0107] The shadow block identification module 106 identifies and numbers (i = 1, 2, 3, ..., n) all connected shadow blocks in the shadow mask Mask2 with more than 15 pixels, and ensures that each shadow block is not connected to the others.

[0108] Define the statistical analysis window module 107, and create an adaptive rectangular window W with the geometric center of the rectangular image corresponding to the shadow block as the center, so that the rectangular window W covers all pixels of the shadow block and the number of non-shadow pixels is greater than or equal to 50% of the total number of pixels of the rectangular window.

[0109] The linear regression analysis module 108 performs linear regression analysis on the non-shaded pixel spectral vector Y(1,2,…,7) and the shaded pixel spectral vector X(1,2,…,7) of the statistical analysis window to obtain the values ​​of the equation parameters k and b of the linear regression equation Y=k×X+b for this window.

[0110] The shadow restoration module 109 uses a linear regression equation to restore shadows in all shadow blocks of all bands in the remote sensing image, resulting in the shadow-restored remote sensing image-B.

[0111] The aforementioned high spatial resolution remote sensing image shadow restoration system is used to implement the corresponding high spatial resolution remote sensing image shadow restoration method in the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0112] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples. Within the framework of this invention, technical features of the above embodiments or different embodiments can also be combined, and many other variations of different aspects of the invention as described above exist, which are not provided in detail for the sake of brevity. For those skilled in the art, the present invention can have combinations and modifications of the various technical features described above. Improvements, modifications, equivalent substitutions, or applications of the structure or method of the present invention to other fields to achieve the same effect, without departing from the spirit and scope of the invention, all fall within the scope of protection of this invention.

Claims

1. A method for shadow restoration in high spatial resolution remote sensing images, characterized in that, Includes the following steps: Step 1: Input Remote Sensing Image Input a high-resolution remote sensing image image-A containing three visible light bands: red (R), green (G), and blue (B); Step 2: HIS color space transformation The conversion formula for RGB color space to HIS color space is as follows: In formulas 1, 2, and 3, H, I, and S represent hue, brightness, and saturation in the HIS color space, respectively, and R, G, and B represent the DN values ​​corresponding to the red (R), green (G), and blue (B) bands of the remote sensing image-A. Step 3: Calculate the normalized shading index In Formula 4, NDUI is the normalized shading index, and I and S are the brightness and saturation in the HIS color space, respectively. Step 4: Shadow Detection and Extraction Based on the normalized shadow index NDUI, the shadow segmentation threshold T is determined according to its mean and standard deviation characteristics. The threshold T is used to segment the image using the normalized shadow index NDUI, thereby extracting the shadow area information in the remote sensing image and obtaining the shadow mask Mask1. Step 5: Shadow Mask Filtering Morphological filtering is used to eliminate small patches in shadow mask Mask1, resulting in optimized shadow mask Mask2. Step 6: Shaded Block Identification Identify and number all shadow blocks in the shadow mask Mask2 (i = 1, 2, 3, ..., n), and ensure that each shadow block is not connected to the others; Step 7: Determine the statistical analysis window Let the size of the rectangular image R corresponding to the shaded area numbered i be N1×M1. Create an adaptive rectangular window W of size N2×M2 with the geometric center of the rectangular image R as the center. The adaptive rectangular window W covers all pixels of the shaded area, that is, N2≥N1 and M2≥M1. Step 8: Linear Regression Analysis Linear regression analysis was performed on the adaptive rectangular window W to obtain the linear regression equation of the adaptive rectangular window W; Step 9: Shadow Restoration The linear regression equation of the adaptive rectangular window W is used to restore the shadows in the remote sensing image image-A; the shadows in all shadows of the three bands of the remote sensing image image-A (red (R), green (G) and blue (B)) are restored to obtain the shadow-restored remote sensing image image-B.

2. The high spatial resolution remote sensing image shadow restoration method according to claim 1, characterized in that: In step 4, the shadow segmentation threshold T is between 0.31 and 0.

36.

3. The high spatial resolution remote sensing image shadow restoration method according to claim 1, characterized in that: In step 5, dilation and erosion operations are performed on the shadow mask Mask1 sequentially.

4. The high spatial resolution remote sensing image shadow restoration method according to claim 1, characterized in that: In step 6, the number of shadow pixels in each shadow block is greater than 15.

5. The high spatial resolution remote sensing image shadow restoration method according to claim 1, characterized in that: In step 7, if the number of non-shadow pixels within the adaptive rectangular window W is greater than or equal to 50% of all pixels in the adaptive rectangular window W, then the window size of the adaptive rectangular window W is gradually increased until the condition is met.

6. The high spatial resolution remote sensing image shadow restoration method according to claim 5, characterized in that: In step 7, the method for gradually increasing the window size of the adaptive rectangular window W is to increase the values ​​of N2 and M2 by 1 in sequence.

7. The high spatial resolution remote sensing image shadow restoration method according to claim 1, characterized in that: Step 8 specifically includes the following operations: For a certain band in the remote sensing image image-A, calculate the histogram, mean μ, and standard deviation σ of all shaded and unshaded pixels in each adaptive rectangular window W; A shadow pixel spectral vector X(1,2,…,7) is established by taking different spectral values ​​of shadow pixels from low to high, such that the values ​​of X1,X2,…,X7 are equal to μ-2σ, μ-1.5σ, μ-σ, μ, μ+σ, μ+1.5σ and μ+2σ, respectively. The cumulative percentage vector AP1(1,2,…,7) corresponding to the shadow pixel spectral vector X(1,2,…,7) is obtained from the histogram of the shadow pixels. Based on the histogram of non-shaded pixels, search for the spectral value that is closest to the cumulative percentage vector AP1(1,2,…,7) of shaded pixels to obtain the cumulative percentage vector AP2(1,2,…,7) of non-shaded pixels, and establish the corresponding spectral vector Y(1,2,…,7) of non-shaded pixels; Linear regression analysis is performed on the non-shaded pixel spectral vector Y(1,2,…,7) and the shaded pixel spectral vector X(1,2,…,7) in an adaptive rectangular window W to obtain the values ​​of the parameters k and b of the linear regression equation Y=k×X+b for the adaptive rectangular window W.

8. The high spatial resolution remote sensing image shadow restoration method according to claim 7, characterized in that: Linear regression analysis was performed on each of the n adaptive rectangular windows W to obtain n unique linear regression equations.

9. A high spatial resolution remote sensing image shadow restoration system, used to implement the method as described in claim 1, characterized in that, include: The image acquisition module is used to input a high-resolution remote sensing image image-A containing three visible light bands: red (R), green (G), and blue (B). The HIS color space conversion module is used to convert the RGB color space to the HIS color space. The normalized shading index calculation module is used to calculate the normalized shading index. The shadow detection and extraction module uses a threshold T to segment the normalized shadow index, thereby extracting shadow area information in the remote sensing image and obtaining the shadow mask Mask1. The shadow mask filtering module performs dilation and erosion operations on the shadow mask Mask1 sequentially, and obtains the optimized shadow mask Mask2 through morphological filtering. The shadow block identification module identifies and numbers all connected shadow blocks in the shadow mask Mask2 with more than 15 pixels (i = 1, 2, 3, ..., n), and ensures that each shadow block is not connected to others. Determine the statistical analysis window module, and create an adaptive rectangular window W with the geometric center of the rectangular image corresponding to the shaded area as the center, so that the adaptive rectangular window W covers all pixels of the shaded area and the number of non-shaded pixels is greater than or equal to 50% of all pixels of the adaptive rectangular window W. The linear regression analysis module performs linear regression analysis on the non-shaded pixel spectral vector Y(1,2,…,7) and the shaded pixel spectral vector X(1,2,…,7) of the adaptive rectangular window W, and obtains the values ​​of the equation parameters k and b of the linear regression equation Y=k×X+b for the adaptive rectangular window W. The shadow restoration module uses a linear regression equation to restore shadows in all shadow blocks of all bands in the remote sensing image image-A, resulting in the shadow-restored remote sensing image image-B.

Images