A remote sensing image quality evaluation method for island reef shallow water areas

Abstract

The present application relates to a kind of island reef shallow water area-oriented remote sensing image quality evaluation method, comprising the following steps: using normalized water index separates island reef underwater and land area;Using relative water transparency and total turbidity of suspended solids extracts island reef shallow water area;Cloud index, cloud shadow index, flare index is used to evaluate the cloud, shadow, flare situation of island reef shallow water area, finally obtain multiple special indexes;Fourth step, with special index and general index (standard deviation, information entropy and definition) assess the image quality of island reef shallow water area.The present application combines remote sensing physical model with traditional image quality evaluation method, effectively solves the problem of insufficient applicability of existing method in island reef shallow water area.Through the establishment of targeted quality evaluation system, the accurate evaluation of effective shallow water area in satellite image is realized, and an automatic, efficient image quality evaluation solution for island reef remote sensing monitoring is provided.

Description

Technical Field

[0001] This invention relates to a method for evaluating the quality of remote sensing images in shallow water areas of islands and reefs, and in particular to a method for constructing special indicators and general indicators for evaluating the quality of remote sensing images in shallow water areas of islands and reefs, as well as an automatic evaluation method for the quality of remote sensing images based on the special and general indicators. Background Technology

[0002] Remote sensing image quality assessment technology is a crucial preliminary step in marine satellite remote sensing and ecological monitoring. Currently, the industry primarily employs two technical approaches for remote sensing image quality assessment in shallow water areas around islands and reefs: objective evaluation methods based on traditional numerical calculations and intelligent evaluation methods based on deep learning.

[0003] In traditional numerical computing, researchers typically construct comprehensive evaluation systems by combining global statistics (such as pixel mean, standard deviation, and signal-to-noise ratio) with spatial domain features (information entropy and gradient sharpness). However, such methods have significant limitations: firstly, global statistical features cannot effectively characterize the local quality features of key shallow water areas of interest in satellite remote sensing, such as coral reefs and seagrass beds; secondly, existing indicator systems do not consider the attenuation effect of water optical properties on reflectance spectra, leading to significant differences in evaluation results under different water environments.

[0004] Deep learning-based quality assessment methods construct convolutional neural network models and use manually labeled quality tags for supervised training. However, this method has two fundamental drawbacks: First, it suffers from subjectivity dependence, as the labeling process relies on expert experience to classify image patches for quality, with a Kappa coefficient of only 0.65 between different labelers; second, it faces computational resource bottlenecks, with the evaluation of a single high-resolution image taking up to 3.7 minutes (under GPU acceleration), making it difficult to meet the real-time monitoring needs of large island and reef areas in the South China Sea.

[0005] Of particular note is the unique marine environment of shallow waters around islands and reefs, which presents distinct challenges to quality assessment: 1) Dynamically changing water transparency directly affects satellite visibility depth, and existing methods have not established a mapping relationship between transparency parameters and quality scores; 2) Instantaneous water depth changes (±2m) caused by tidal forces alter the seabed reflectivity curve, but traditional indicators cannot capture such transient characteristics; 3) Clouds, shadows, remote sensing data, and traditional image evaluation indicators in satellite imagery have not been effectively modeled. This results in insufficient correlation between the existing assessment system and satellite remote sensing monitoring tasks in shallow waters and shallow seas.

[0006] Therefore, there is an urgent need to construct a remote sensing image quality assessment method for shallow water areas of islands and reefs. This method should possess the following core characteristics: 1) It can quantitatively reflect the impact of water environment parameters such as transparency and suspended matter concentration on the quality of shallow water areas of islands and reefs; 2) It can quantitatively measure the impact of clouds, shadows, and flares in satellite remote sensing images on the quality of shallow water areas of islands and reefs; 3) It can achieve lightweight automatic assessment to adapt to edge computing environments. This will provide reliable data quality assurance for marine island and reef ecological monitoring. Summary of the Invention

[0007] The technical problem to be solved by this invention is to overcome the problem that existing methods focus on quantifying the quality of the entire remote sensing image, which cannot fully measure the signal quality of the shallow water area of ​​interest in satellite remote sensing, thus causing the existing quality evaluation methods to fail. This invention proposes a remote sensing image quality evaluation method for shallow water areas of islands and reefs. Compared with previous methods, this invention fully considers the characteristics of shallow water areas of islands and reefs.

[0008] To address the above technical problems, this invention proposes a method for evaluating the quality of remote sensing images in shallow water areas of islands and reefs, comprising the following steps:

[0009] Step 1: Separate land and water in the original remote sensing image, and filter out the land to obtain the water image. The ratio of the area of ​​the first specific index to the total area of ​​the original remote sensing image is used to obtain the first specific index. S 1;

[0010] Step 2: Construct calculation models for relative water transparency (RWT) and total suspended solids turbidity (TSST), and extract shallow water images. By calculating shallow water images Area and water body images The ratio of area to obtain the second special indicator S 2;

[0011] Step 3: Remove shallow water images By eliminating cloud noise and shadow noise, noise-free shallow water images can be obtained. Calculate noise-free shallow water images Area and shallow water image The ratio of area to the third special indicator S 31 ;

[0012] Step 4: Construct the Flare Index (GI) and apply it to shallow water images. Flare removal is performed to obtain a flare-corrected, noise-free shallow water image. The ratio of the difference between the image before and after correction to that between the image before correction and the image before correction is calculated to obtain the fourth specific index. S 32 ;

[0013] Step 5: Using standard deviation, information entropy, and sharpness evaluation algorithms, calculate the noise level in the shallow waters of the island and reef areas, and obtain three general standard deviation indicators. S 4. Information Entropy Index S 5. Clarity index S 6;

[0014] Step 6: Sort and score each indicator of multiple images to be evaluated in the study area according to their quality. The image with the highest sum of scores for all indicators is the image with the best quality.

[0015] This invention first employs the Normalized Difference Water Index (NDWI) to separate underwater and terrestrial areas. Then, it uses Relative Water Transparency (RWT) and Total Suspended Solids Turbidity (TSST) to extract areas of interest for coral reef substrate classification. Subsequently, it uses the Cloud Index (CI), Cloud Shadow Index (CSI), and Glint Index (GI) to evaluate the cloud, shadow, and flare conditions in shallow water substrate areas. Finally, it uses standard deviation, information entropy, and sharpness as general indicators to measure the image quality of substrate areas.

[0016] The remote sensing image quality assessment method for shallow water areas of islands and reefs proposed in this invention has the following advantages:

[0017] 1. A land-water separation method based on normalized water index (NDWI) and a shallow water area extraction technology for islands and reefs based on the combination of relative water transparency (RWT) and total suspended solids turbidity (TSST) are proposed. For the first time, a localized quantitative assessment of the quality of shallow water signals from satellite remote sensing is achieved, solving the problem of insufficient sensitivity of traditional global indicators to key areas.

[0018] 2. The integrated cloud index (CI), cloud shadow index (CSI), and flare index (GI) constitute a satellite image interference evaluation system. A quantitative correlation model between shallow water area quality and atmospheric interference factors is established, breaking through the limitation of traditional methods that ignore the dynamic influence of cloud layers.

[0019] 3. By using a hybrid architecture of "dedicated metrics (RWT / TSST / CI / CSI / GI) + general metrics (standard deviation / information entropy / sharpness)", the processing time for a single image is reduced to 45 seconds (CPU environment) while ensuring evaluation accuracy, which is more than 5 times more efficient than existing deep learning methods. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart illustrating the overall process of an example of the present invention.

[0022] Figure 2 Examples of the present invention: RWT shallow water extraction results: (a) original remote sensing image; (b) shallow water body (yellow area) extraction results.

[0023] Figure 3 Examples of the present invention: TSST shallow water extraction results: (a) original remote sensing image; (b) shallow water body (orange area) extraction results.

[0024] Figure 4 This is the CI cloud extraction result of the present invention.

[0025] Figure 5 This is the CSI shadow extraction result of an example of the present invention.

[0026] Figure 6 The images were taken on January 19, 2020, March 31, 2020, April 20, 2020, and July 27, 2021, respectively. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings, making the technical route and operation steps of the present invention clearer. The overall technical route is shown below. Figure 1 .

[0028] Step 1, First Specific Indicator S 1. Calculation: Let the original remote sensing image be... I Using the normalized water index NDWI Water and land separation, normalized water index of pixels NDWI A land cell is defined as one that satisfies either of the following two formulas and is then removed:

[0029]

[0030]

[0031] In the formula, This represents the radiance value in the near-infrared band. This represents the radiance value of shortwave infrared 1. This indicates the radiance value in the green light band.

[0032] In this embodiment, the image after separating the land and water can be represented as follows: The land area is assigned a value of 0. Thus, after filtering out the land portion, we obtain the first dedicated index for evaluating image quality score. S 1. First Specialized Indicator S 1 can be passed The ratio of the area of ​​the original remote sensing image to the total area of ​​the original image is used to obtain the value. This ratio can also be calculated using the following formula:

[0033]

[0034] This represents the total number of pixels in the image with a non-zero value. It is calculated using the normalized water index. NDWI After removing the land, the next step is to use relative water transparency (RWT) and total suspended solids turbidity (TSST) to extract shallow water areas of interest for bottom sediment classification in coral reefs.

[0035] Step 2: Construct calculation models for relative water transparency (RWT) and total suspended solids turbidity (TSST), and use an automatic threshold segmentation method to extract shallow water areas of interest for satellite remote sensing monitoring of islands and reefs. By calculating shallow water areas Area and water body images The ratio of area to obtain the second special indicator S 2.

[0036] In this step, based on spectral transmission characteristics, the relative water transparency (RWT) of a pixel satisfies the following formula, which indicates a deep-water pixel and removes it (deep-water pixels are assigned a value of 0) to obtain a shallow-water image with relatively high water transparency. :

[0037]

[0038] In the formula, e It is the natural logarithm. This represents the radiance value in the red light band. T represents the radiance value in the green light band, r represents the composite radiance index of the red and green light bands, and T represents the radiance value in the green light band. wt For automatic separation threshold, T wt The following formula is used to calculate:

[0039]

[0040] In the formula, Represents the minimum relative water transparency (RWT). This represents the average value of the relative water transparency RWT. It is the first adjustment factor, which is usually selected from {1 / 3, 1 / 2, 2 / 3, 3 / 4, 4 / 5}.

[0041] The results of deep and shallow water separation after segmentation using the relative water transparency threshold are shown below. Figure 2 , Figure 2 (a) is an image of the water body before separation. Figure 2 (b) shows the result after segmentation by the relative water transparency threshold. In the figure, the yellow area represents shallow water.

[0042] Total suspended solids turbidity (TSST) is used to assess the clarity and water quality of water bodies. To a certain extent, it can extract shallow water areas of interest for sediment classification. A pixel with TSST satisfying either of the following two formulas is considered a deep water pixel and is removed (deep water pixels are assigned a value of 0) to obtain a shallow water image with TSST. :

[0043]

[0044]

[0045] In the formula, R This represents the radiance value in the red light band. B T represents the radiance value in the blue light band. tsst The automatic segmentation threshold for total suspended solids turbidity (TSST) is calculated using the following formula:

[0046]

[0047] In the formula, This represents the minimum value of total suspended solids turbidity (TSST). The average value of total turbidity (TSST) of suspended solids is given. It is the second adjustment factor, which is generally selected from {1 / 4, 1 / 3, 1 / 2, 2 / 3, 3 / 4}.

[0048] The results of deep and shallow water separation after segmentation by total turbidity threshold of suspended solids are shown below. Figure 3 , Figure 3 (a) is an image of the water body before separation. Figure 3 (b) shows the result after segmentation by the relative water transparency threshold. In the figure, the orange area represents shallow water.

[0049] Shallow water images with relative water transparency Shallow water images with total turbidity of suspended solids After merging, the shallow water image after separating the deep and shallow water in this step can be obtained, represented as... :

[0050]

[0051] Thus, after extracting the shallow water area of ​​interest for sediment classification, a second specific indicator for evaluating image quality is obtained.S 2. Second Special Indicator S 2. Can be seen through shallow water images Area and water body images The ratio of areas is obtained, and this ratio can also be calculated using the following formula:

[0052]

[0053] This represents the total number of pixels in the image whose pixel value is not 0.

[0054] It is worth mentioning that RWT and TSST reflect the transparency and biological parameters of the water body, and therefore also have a certain filtering effect on clouds. However, they cannot evaluate thin clouds, shallow shadows and other conditions within the region of interest. Therefore, after using RWT and TSST to extract the shallow water area of ​​interest for sediment classification, CI, CSI and GI are used to evaluate the environmental noise in this area.

[0055] Step 3: Filter out shallow water images To eliminate cloud and shadow noise, noise-free shallow water images are obtained. Calculate noise-free shallow water images Area and shallow water image The ratio of area to the third special indicator S 31 .

[0056] Among them, noise-free shallow water images Obtained through the following calculations:

[0057]

[0058] In the formula, Representing shallow water images Cloud images in Representing shallow water images The shadow image in the image.

[0059] The extraction methods for cloud index (cloud noise) and shadow index (shadow noise) are explained in detail below.

[0060] Most low-to-medium resolution multispectral / hyperspectral remote sensing sensors (Sentinel-2, Landsat series, etc.) have near-infrared (NIR) and short-wave infrared (SWIR) spectral channels, but high-resolution sensors (Gaofen-2, etc.) typically only have NIR spectral channels. Depending on whether the SWIR band is included, the CI index (cloud index) can be calculated in two ways:

[0061]

[0062]

[0063] The CI1 index measures the similarity of reflectance characteristics in the visible and infrared wavelengths. Generally, because clouds exhibit similar reflectance characteristics in both the visible and infrared bands, the CI1 index typically fluctuates within a narrow range very close to 1. Where, This represents the radiance value in the near-infrared band. This represents the radiance value of shortwave infrared radiation. This represents the radiance value in the green light band. This represents the radiance value in the red light band. This indicates the radiance value in the blue light band.

[0064]

[0065]

[0066] The CI2 index, which is the average value of the spectrum in the relevant bands, is used to describe the brightness characteristics of clouds. In the formula, This represents the radiance value in the near-infrared band. This represents the radiance value of shortwave infrared 1. This represents the radiance value of shortwave infrared 2. This represents the radiance value in the green light band. This represents the radiance value in the red light band. This indicates the radiance value in the blue light band.

[0067] For remote sensing images from different satellites, exponential clouds can be effectively separated using any of the following formulas:

[0068]

[0069]

[0070] Here, Y1 is a smaller threshold, and Y2 is a larger threshold. The first parameter Y1 can generally be selected from the set {0.01, 0.1, 1, 10, 100} and can be fine-tuned experimentally for specific scenarios. Parameter Y2 is set to a larger value. Considering the reflectivity of different coral reef substrates, in practical applications, Y2 can be adaptively determined using the following method:

[0071]

[0072] in, Indices The average value, Indices The maximum value, It is an adjustment factor, usually selected from {1 / 10, 1 / 9, 1 / 8, 1 / 7, 1 / 6, 1 / 5, 1 / 4, 1 / 3, 1 / 2}. Figure 4 The image shows the results of cloud extraction using the CI index from a Sentinel-2 remote sensing image of a certain island reef (red area).

[0073] The CSI index (Shadow Index) is calculated similarly to the CI index (Cloud Index). Depending on whether the SWIR band is included, the SI index can be constructed using the following two formulas:

[0074]

[0075]

[0076] The CSI index is designed for the NIR band and The mean value of the band, or only the NIR band, is used to describe the reflectance characteristics of cloud shadows (shadows in island and reef images are generally cloud shadows) at longer wavelengths. However, water typically exhibits similar reflectance characteristics within these wavelength ranges. Therefore, to accurately identify cloud shadows projected onto the water surface and eliminate water interference, additional constraints need to be introduced. Given that water has high reflectance at shorter wavelengths, especially in the blue light band, this characteristic of the blue light band can be used to eliminate water interference. Specifically, the following two formulas must be satisfied simultaneously:

[0077]

[0078]

[0079] In the formula, parameter Y3 is generally set to a small value. The calculation method is as follows:

[0080]

[0081] in, express The minimum value of the exponent, express The average of the index, This is the adjustment coefficient, typically selected from {1 / 4, 1 / 3, 1 / 2, 2 / 3, 3 / 4}. The fourth parameter, Y4, can be determined in practical applications using the following method:

[0082]

[0083] in, This represents the minimum value of the radiance in the blue light band. This represents the average value of the radiance in the blue light band. It is an adjustment factor, usually selected from {1 / 2, 2 / 3, 3 / 4, 4 / 5, 5 / 6}.

[0084] Figure 5 The image shows the results of extracting cloud shadows (purple area) from a Sentinel-2 remote sensing image of an island reef using CSI.

[0085] Noise-free shallow water images after removing clouds and shadows can be represented as :

[0086]

[0087] Therefore, the third special indicator can be calculated according to the following formula. S 31 Third Specialized Indicator S 31 Through noiseless shallow water images Area and shallow water image The ratio of areas is obtained, and this ratio can also be calculated using the following formula:

[0088]

[0089] In the formula, This represents the total number of pixels in the image whose pixel value is not 0.

[0090] Step 4: Construct the Flare Index (GI) and apply it to shallow water images. Flare removal is performed to obtain a noise-free shallow water image after flare correction. The ratio of the difference between the image before and after correction to that between the image before correction and the image before correction is calculated to obtain the fourth specific index. S 32 .

[0091] In this embodiment, the flare index (GI) calculation employs an improved flare removal algorithm. Specifically, in shallow water images... Points within the selected sample area are used. With the near-infrared brightness as the X-axis and the visible light brightness as the Y-axis, linear regression is performed on all pixels in each visible light band to calculate the linear relationship. This yields the band... i Slope of linear regression b i The formula for flare removal is:

[0092]

[0093] In the formula, Indicates the wavelength after flare removal i The radiance value, Indicates band i The radiance value, express The radiance value of a pixel after band calculation; express The minimum radiance value after band calculation (which can be considered as the pixel radiance value without the influence of solar flares). Indicates band i The slope of the linear regression. By evaluating the degree of difference between the images before and after flare correction, a specific index can be obtained after obtaining the flare-corrected image. S 32 :

[0094]

[0095] In the formula, Representing shallow water images band i The average of the ratios of the difference in radiance before and after flare correction to the radiance value before correction for all pixels.

[0096] S 31 With S 32 Adding them together yields the specific index S3. Using S1, S2, and S3, the specific index score S of the remote sensing image to be screened can be obtained. SP = S1 + S2 + S3, S SP A higher value indicates better image quality.

[0097] Step 5: Using standard deviation, information entropy, and sharpness evaluation algorithms, calculate the noise level in the shallow waters of the island and reef areas, and obtain three general standard deviation indicators. S 4. Information Entropy Index S 5. Clarity index S 6.

[0098] The standard deviation index S4 can be used to measure the degree of variation in pixel values ​​in an image. A larger standard deviation indicates that the pixel values ​​in the image are dispersed and exhibit significant variation. A smaller standard deviation indicates that the pixel values ​​in the image are concentrated and exhibit less variation. For remote sensing images, a larger standard deviation indicates poorer image quality, and a smaller standard deviation indicates better image quality. Let image I... S The total number of pixels is N The average radiance is , For pixels i The standard deviation of the radiance value is... :

[0099]

[0100] Information entropy index S5. Information entropy is used to measure the uncertainty or randomness of pixel values ​​in an image. If the pixel values ​​in an image are evenly distributed, the information entropy is high; if the pixel values ​​tend to be concentrated, the information entropy is low. For remote sensing images, higher information entropy indicates better image quality (excellent), and lower information entropy indicates worse image quality (inferior). For image I... S Let L represent the number of gray levels in the image, then the information entropy is... It can be represented as:

[0101]

[0102] In the formula, Represents a cell i Radiance value The probability of appearing in an image. Represents a cell i The radiance value.

[0103] Image sharpness index S6 calculation methods include Brenner, Tenengrad, and SMD2. Among them, the sharpness obtained by Brenner has received widespread attention because it is consistent with human subjective perception. Therefore, Brenner is chosen to calculate the sharpness of remote sensing images. The Brenner gradient function is the simplest gradient evaluation function, which simply calculates the square of the gray-level difference between two adjacent pixels. The function is defined as follows:

[0104]

[0105] in, Representing an image Corresponding pixel The grayscale value.

[0106] For remote sensing images, the higher the resolution, the better the image quality (superior); the lower the resolution, the worse the image quality (inferior).

[0107] Step 6: Sort and score each indicator of multiple images to be evaluated in the study area according to their quality. The image with the highest sum of scores for all indicators is the image with the best quality.

[0108] In this step, based on the constructed specialized and general indicators, the overall screening process for high-quality remote sensing images is as follows: Calculate the image to be evaluated... S 1. S 2. and S 3. Sort and assign scores according to the scores, and assign three scores [N, 1] to each candidate image from highest to lowest; calculate the score of the image to be evaluated. S 4. S 5. and S6. Sort and assign scores to each candidate image, assigning three scores [N, 1] to each image based on its score. A smaller standard deviation indicates less image noise. S The image with the lowest score is assigned a score of N, and the image with the highest score is assigned a score of 1. The six scores of each image are added together to obtain the final score. The images are then sorted according to their final scores, and the image with the highest score is the best image.

[0109] Verification of Examples

[0110] To verify the accuracy and reliability of the method of the present invention, the following example will be used for further explanation.

[0111] Taking a certain island reef as an example, the proposed image selection method was verified using four Sentinel-2 images taken at different times and with significant quality differences. The four images are shown below. Figure 6 , Figure 6 (a) was captured on January 19, 2020. Figure 6 (b) The image was captured on March 3, 2020. Figure 6 (c) was captured on April 20, 2020. Figure 6 (d) The image was captured on July 27, 2021.

[0112] Table 1 Image quality evaluation results

[0113]

[0114] As can be seen from Table 1, Figure 6 (b) has the best image quality. Figure 6 (c) Image quality is second best. Figure 6 (d) The image quality is the worst. Compared to the other three images, Figure 6 (b) The image has the highest effective area ratio and no obvious clouds, shadows, or flares, resulting in a good overall condition and thus receiving the highest score of 24 points. Although the image Figure 6 (a) ranked second in the dedicated indicators (8 points), and was higher than the image. Figure 6 (c) scored 7 points, but from a general perspective, Figure 6 (c) The video received a score of 9. Figure 6 (a) The video is only 5 minutes long, therefore it will be... Figure 6 (c) The image is ranked second. Figure 6 (d) The image ranked last in almost all indicators, with a total score of only 7 points, compared to Figure 6 (a) Compared to the image, Figure 6 (d) The image has less cloud cover, but the clouds in this image mainly cover the shallow water areas of interest for substrate classification, while Figure 6(a) Although the cloud cover is the highest in the image, it is relatively low in shallow water areas. The above experiments fully demonstrate the effectiveness of the image quality assessment method proposed in this paper.

[0115] The method for evaluating the quality of remote sensing images of shallow water areas of islands and reefs proposed in this invention is not limited to the specific technical solutions described in the above embodiments. All technical solutions formed by equivalent substitutions are within the scope of protection claimed by this invention.

Claims

1. A method for evaluating the quality of remote sensing images for atoll shallow water areas, comprising the following steps: Step 1, water and land separation is performed on the original remote sensing image, and the water body image after filtering out the land is calculated I NDWI The ratio of the area of the first special index to the total area of the original remote sensing image is obtained S 1; Step 2, constructing the calculation model of relative water transparency (RWT) and total suspended solids turbidity (TSST), combining the relative water transparency shallow water image with the total suspended solids turbidity shallow water image to obtain the shallow water image after separating the deep and shallow water I WTT , obtaining the second special index I 2 by calculating the area ratio of the shallow water image WTT and the water body image I NDWI area S 2; Step 3, removing the shallow water image I WTT from the cloud noise and shadow noise in the image I CS , to obtain a noise-free shallow water image I CS ; calculating the area of the noise-free shallow water image I WTT and the area of the shallow water image S 31 ; Step 4, constructing a flare index GI, and processing the shallow water image I WTT carrying out flare removal to obtain a flare-corrected noise-free shallow water image, calculating a difference value between the image before and after correction and the image before correction, and obtaining a fourth special index S 32 ; Step 5, using standard deviation, information entropy, and clarity evaluation algorithm, the noise level of the shoal area of the island reef is calculated, and three general standard deviation indexes are obtained respectively S 4. Information entropy index S 5. Clarity index S 6. Step 6, for each index of multiple images to be evaluated in the study area, the indexes are sorted and scored according to the advantages and disadvantages, and the image to be evaluated with the highest sum of all indexes is the best quality image.

2. The method for evaluating the quality of remote sensing images of the reef shallow water area according to claim 1, characterized in that: In step 1, the normalized water index is used NDWI The original remote sensing image is separated into water and land, and the normalized water index of the pixel NDWI Any one of the following two formulas is satisfied, which is a land pixel and is removed: ； ； wherein NIR represents the radiance value in the near infrared wavelength band, SWIR 1 represents the radiance value in the short wave infrared 1 wavelength band, G represents the radiance value in the green wavelength band.

3. The method for evaluating the quality of remote sensing images of reef shallow water areas according to claim 1, characterized in that: In step 2, the relative water transparency RWT of the pixel meets the following formula, which is a deep water pixel and is removed to obtain a relative water transparency shallow water image RWT I NDWI ): ； wherein e is the natural logarithm, R represents the radiance value in the red light band, G represents the radiance value in the green light band, r represents the composite brightness index in the red light band and the green light band, T wt is the automatic separation threshold of the relative water transparency RWT, which is obtained by calculation by means of the following formula: ； wherein min (RWT) denotes the minimum value of the relative water transparency RWT, mean(RWT) denotes the average value of the relative water transparency RWT, t wt is a first adjustment coefficient, t wt ∈(0,1).

4. The method for evaluating the quality of remote sensing images of reef shallow water areas according to claim 1, characterized in that: In step 2, the total suspended solids turbidity TSST of the pixels satisfies any one of the following two formulas as deep water pixels and removes them to obtain the total suspended solids turbidity shallow water image TSST I NDWI ): ； ； wherein R represents the radiance value in the red light band, B represents the radiance value in the blue light band, NIR represents the radiance value in the near infrared band, T tsst is the automatic segmentation threshold for the total suspended solids turbidity TSST, obtained by calculation with the following formula: ； wherein min (TSST) denotes the minimum value of the total suspended solids turbidity TSST, mean(TSST) denotes the average value of the total suspended solids turbidity TSST, t tsst is a second adjustment coefficient, t tsst ∈(0,1).

5. The method for evaluating the quality of remote sensing images of reef shallow water areas according to claim 1, characterized in that: In step 3, cloud index CI and shadow index CSI are constructed, and automatic threshold method is used to remove cloud noise and shadow noise in shallow water images I WTT noiseless shallow water images I CS are obtained by the following calculation: ； wherein CI I WTT denotes a cloud image in the shallow water image I WTT denotes a cloud image in the shallow water image CSI I WTT denotes a shadow image in the shallow water image I WTT denotes a shadow image in the shallow water image​​ 6. The method for evaluating the quality of remote sensing images of reef shoal areas according to claim 5, characterized in that: said third specific indicator S 31 The calculation is as follows: ； In the formula, Count (•) is the total number of pixels in the image whose pixel values are not zero.

7. The method for evaluating the quality of remote sensing images of reef shallow water areas according to claim 1, characterized in that: In step 4, the GI is calculated, first in the shallow-water image I WTT The points in the selected sample area are selected, with the brightness in the near-infrared band as the X axis and the brightness in the visible light band as the Y axis, linear regression is performed on all the pixels in each visible light band, the linear relationship is calculated, and the band i The slope of the linear regression is calculated, and the flare removal formula is: ； In the formula, Indicates the wavelength after flare removal i The radiance value, R i Indicates band i The radiance value, R (red+nir) express red + nir The radiance value of a pixel after band calculation; Min red+nir express red + nir Minimum radiance value after band calculation b i Indicates band i The slope of linear regression.

8. The method for evaluating the quality of remote sensing images of reef shallow water areas according to claim 7, characterized in that: Fourth special index S 32 The calculation is obtained by the following formula: ； wherein mean (•) indicates a shallow water image I WTT waveband i The average of the ratio of the difference in radiance between the corrected and uncorrected radiance and the uncorrected radiance for all the pixels.

9. The method for evaluating the quality of remote sensing images of reef shallow water areas according to claim 1, characterized in that: The third special index is obtained by summing up the first special index and the second special index. S 31 The fourth special index is obtained by summing up the first special index and the third special index. S 32 The fifth special index is obtained by summing up the second special index and the third special index. S 3. In step 6, each image to be evaluated is evaluated according to the first special index S 1. The second special index S 2. The fifth special index S 3. Standard deviation index S 4. Information entropy index S 5. Clarity index S 6. The pros and cons of the sorting value, according to the score from the best to the worst respectively to each candidate image is given a score; the six indicators of the image to be evaluated are added to get the final score, according to the final score of the candidate image is sorted, the highest score is the best image.

Images