CN105651263B - Shallow water depth multi-source remote sensing merges inversion method - Google Patents

Abstract

The shallow water depth multi-source remote sensing fusion inversion method includes: the first step: preprocessing the multi-spectral remote sensing image to obtain the sea surface reflectance; the second step: obtaining and processing the measured water depth value on site; Step 4: Multi-source water depth inversion fusion; Step 5: Water depth inversion accuracy verification; n kinds of single-source sounding inversion results and their corresponding water depth segment identification images and fusion parameters are used as input , multi-source water depth inversion fusion is carried out pixel by pixel; after the water depth inversion accuracy verification is completed, the final water depth value is output as the actual water depth value of the remote sensing image. Compared with the existing inversion methods, this method can comprehensively utilize the different responses of various remote sensing data sources to water depth information, mine the water depth data, improve the inversion accuracy, and after decision-making fusion processing, it is especially suitable for shallow water in complex situations. Regional ocean bathymetry.

Description

Multi-source Remote Sensing Fusion Inversion Method for Shallow Sea Depth

technical field

The invention relates to a method for measuring ocean water depth, which belongs to the technical field of space remote sensing, and in particular relates to a shallow water depth multi-source remote sensing fusion inversion method that can use multiple remote sensors to detect ocean water depth.

Background technique

Ocean bathymetry is the necessary basic data to ensure ship navigation, carry out port wharf and marine engineering construction, and formulate relevant plans for coasts and islands. Compared with on-site water depth measurement methods, remote sensing technology has many advantages such as wide coverage, short cycle, low cost, and high spatial resolution. Since the 1970s, research on various passive remote sensing water depth inversion models has been carried out at home and abroad. Commonly used visible light water depth inversion models mainly include analytical models, semi-analytical and semi-empirical models, and statistical models. Using different models, inversion applications have been carried out in the fields of bathymetry around rivers, lakes, reservoirs, islands and coastal zones in recent years.

The inversion of water depth with visible light remote sensing is an effective solution to obtain the water depth of complex shallow sea terrain, especially for inversion to obtain water depth data in areas that are inaccessible and difficult to enter by ships. However, because the model is difficult to take into account the physical mechanism and parameterization, the accuracy of the existing visible light water depth remote sensing inversion model has limited room for improvement.

Water depth multi-source remote sensing inversion can overcome the limitation of environmental conditions in single-source image imaging, and the richer band information and different spectral resolutions provided by multi-source remote sensing images are also conducive to the extraction of water depth information. Currently, there are Multi-sources are applied to the research work of water depth remote sensing inversion, but most of them are used for the interpolation of spatial information, and have not been developed and applied at the level of decision-making fusion. Decision fusion can make full use of existing remote sensing image resources and information, and provides a new way to improve the accuracy of optical remote sensing water depth retrieval.

Chinese patent (Application No. 201310188829.4, application publication date CN 104181515A) discloses "a shallow water depth inversion method based on blue-yellow band hyperspectral data". Most of the models used to solve the inversion of clean water depth by means of optical remote sensing are established for multi-spectral data. This type of algorithm is restricted by the wide band of multi-spectral data and the lack of spectral information. The data proposed a new method for inversion of shallow water depth in clean water by using blue-yellow band (450-610 nm) hyperspectral data. This method can accurately extract shallow water depth distribution information within 30 meters, and for a remote sensor, only Algorithm coefficient calibration is required once, and the universality of the algorithm is significantly improved. However, the party uses a single-source remote sensor to obtain images as the detection data source, and the available remote sensing image spectral data bands and spectral information range are limited, which is not conducive to improving the accuracy of shallow sea depth inversion for bathymetry, especially in complex situations However, the detection effect of water depth in shallow sea areas is insufficient.

Contents of the invention

The present invention provides a shallow water depth multi-source remote sensing fusion inversion method based on decision-making fusion, which is used to solve the problem of using only the image of a single-source remote sensor as a data source in the prior art, and the use of spectral data bands and spectral information of remote sensing images Problems with limited range, poor precision and accuracy of bathymetry.

The multi-source remote sensing fusion inversion method for shallow sea depth includes the following steps:

The first step: preprocessing the multispectral remote sensing images to obtain the sea surface reflectance;

The preprocessing includes radiance conversion, atmospheric correction and solar flare removal;

Step 2: Acquisition and processing of on-site measured water depth values;

Obtain the water depth data and the corresponding latitude and longitude coordinates of the experimental area, confirm the tide height value at the measurement time through the tide table, correct the water depth data to obtain the water depth of the theoretical depth datum, and then calculate the theoretical depth datum according to the acquisition time of the multispectral remote sensing image. Tide correction of the instantaneous water depth is performed on the water depth data to obtain the instantaneous water depth;

Step 3: single-source water depth inversion and water depth section identification;

According to the relationship between the water depth at the water depth control point and the reflectance of the corresponding image pixel, the multi-band model is used for statistical regression, and the parameters of the water depth inversion of the source image are output as an input of the multi-source inversion fusion, and the multi-band model is used to perform statistical regression. The parameters of the band model are calibrated, and the formula of the multi-band model is as follows,

X _i =Ln(ρ _i -ρ _si ) (2)

Among them, Z is the water depth, n is the number of bands participating in the inversion, A ₀ and A _i are undetermined coefficients; ρ _i is the reflectance data of the i-th band, and ρ _si is the reflectance of the deep water in this band;

The sounding control points are divided into multiple sounding segments as input, and the average relative error of each sounding segment is output as another input of multi-source sounding inversion fusion, that is, the fusion parameter; as the fusion parameter of multi-source sounding inversion fusion input, Also includes the Kappa coefficient of the output single-source image and the segmented average accuracy of each depth segment;

Among them, n is the number of water depth control points, k represents the water depth section, in formula 3, δ _k is the average relative error, z _i is the measured value of the i-th water depth control point, z _i ' is its inversion value, formula 4 middle, is the Kappa coefficient, x _ii represents the number of correctly classified control points, x _i+ and x _+i are the row and column boundary values of the error matrix when performing segmental statistics on water depth control points, in formula 5, δ _{ma_k} is the segmental average precision , PA _k is the producer's accuracy of the kth water depth section, UA _k is the user's accuracy of the kth water depth section;

Using the fusion parameters and the whole scene remote sensing image, calculate the single-source water depth inversion result, and correct the theoretical depth datum and then segment the result to obtain the water depth segment identification image;

Step 4: Multi-source water depth inversion fusion;

Taking n types of single-source water depth inversion results and their corresponding water depth section identification images and fusion parameters as input, multi-source water depth inversion fusion is carried out pixel by pixel, specifically including;

a) When the number of votes in a water depth segment is t, and Description has The inversion results of one or more images are in the same water depth segment, where Indicates rounding down. At this time, if there are 2 or more images with the same water depth retrieval value, assign this value directly to the current pixel; otherwise, compare the average values of these images in the water depth range For relative error and average precision, the image with the largest average precision in the water depth segment is used as the final pixel value; only when the image with the largest average accuracy in the water depth segment corresponds to the largest average relative error in the water depth segment, select the image with the second highest average accuracy;

b) When the maximum number of votes t satisfies And there are x (x≥2) with the number of votes as t. At this time, compare the Kappa coefficient and the average accuracy of the n classifiers in their corresponding water depth segments. If the image with the largest Kappa coefficient and the largest average accuracy are judged to be the same water depth segment, and it is the same source image, the water depth value of the image pixel is taken as the result; if it is not the same scene image, it is determined that the average relative error of the two scenes in the water depth segment is small; if the image with the largest Kappa coefficient is determined The water depth segment with the largest average precision is different from the water depth value of the former; if there is only one water depth segment with the number of votes t, then in the water depth segment with the number of votes t, the water depth segment with the largest average accuracy is taken as the final pixel value ; When the average relative error of the image corresponding to the image with the largest average accuracy of the water depth segment is also the largest, select the image with the second average accuracy;

c) When the maximum number of votes t=1, select the water depth value corresponding to the image with the largest Kappa coefficient;

Step 5: Water depth inversion accuracy verification;

The accuracy verification is to compare the single-source inversion results before fusion with the multi-source inversion results after fusion by using the sounding checkpoint. After the accuracy verification of sounding inversion is completed, the final sounding value is output as the actual sounding value of the remote sensing image.

In the shallow water depth multi-source remote sensing fusion inversion method described above, the radiance conversion in the first step is to convert the DN value of the remote sensing image into a radiance value; Wavelet method; the atmospheric correction can use FLAASH, dark pixel or 6S atmospheric correction method.

The reference image selected for decision-making fusion in the present invention depends on the scale and resolution of the required water depth image, and there is no special requirement. If the image with the largest spatial resolution is selected as the benchmark, although the fusion speed is improved to a certain extent, the spatial matching uses the decision fusion value at the coordinates of the pixel center to represent the entire pixel, and the loss of information is relatively large. Therefore, from the perspective of processing efficiency and inversion fusion accuracy, it is preferable to use the water depth inversion result image generated by the image with the highest resolution as the reference. Matching is performed to obtain all single-source inversion water depth values and other information at the coordinates, and decision-making fusion is performed to reduce the amount of information that may be lost and ensure the accuracy of inversion.

Beneficial effects of the present invention:

Compared with the existing inversion methods, this method can comprehensively utilize the different responses of various remote sensing data sources to water depth information, expand the use range of remote sensing image spectral data bands and spectral information, mine water depth data, and improve inversion. Accuracy, after decision fusion processing, especially suitable for ocean bathymetry in shallow water areas under complex conditions.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is the fusion flow chart of water depth multi-source inversion of the present invention;

Figure 3a is the scatter diagram of the multi-source fusion results of water depth;

Figure 3b is a scatter diagram of single-source WorldView-2 water depth inversion results;

Figure 3c is a scatter diagram of the single-source Pleiades water depth inversion results;

Figure 3d is a scatter diagram of single-source QuickBird water depth inversion results;

Figure 3e is a scatter diagram of single-source SPOT-6 water depth inversion results;

Fig. 4 is the fusion result of water depth multi-source remote sensing inversion of the present invention;

detailed description

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention.

In conjunction with accompanying drawing 1, the specific embodiment of the present invention is further described in detail, and the shallow water depth multi-source remote sensing fusion inversion method specifically includes the following steps:

Step 1: Multispectral remote sensing data preprocessing:

Firstly, the multi-spectral remote sensing images obtained through multi-source sensors and involved in the fusion of water depth inversion should be preprocessed, including radiance conversion, atmospheric correction and solar flare removal. Radiance conversion is the process of converting image DN values into radiance. Different remote sensing data products correspond to different radiance conversion formulas. Generally, the following two methods are used:

The corresponding parameters in the formula can be obtained in the metadata file of the image. After obtaining the multi-spectral radiance image, use FLAASH or dark pixel, 6S and other methods to perform atmospheric correction to obtain sea surface albedo data; Methods such as mean value or wavelet are used to remove solar flares.

Second: Acquisition and processing of on-site measured water depth values:

Use the multi-beam bathymeter or other bathymetric means to obtain the water depth data of the experimental area, and obtain the corresponding latitude and longitude coordinates at the same time. Correct the water depth data by confirming the tide height value at the time of measurement with the tide table.

Step 3: Single-source water depth inversion and water depth segment identification:

According to the relationship between the water depth at the water depth control point and the reflectance of the corresponding image pixel, the multi-band model is used for statistical regression, and the parameters of the water depth inversion of the source image are output as an input of the multi-source inversion fusion, and the multi-band model is used to perform statistical regression. The parameters of the band model are calibrated, and the formula of the multi-band model is as follows,

X _i =Ln(ρ _i -ρ _si ) (2)

Among them, Z is the water depth, n is the number of bands participating in the inversion, A ₀ and A _i are undetermined coefficients; ρ _i is the reflectance data of the i-th band, and ρ _si is the reflectance of the deep water in this band.

The sounding control point is divided into multiple sounding segments as input, and the average relative error of each sounding segment is output as another input of multi-source sounding inversion fusion, that is, the fusion parameter;

The input fusion parameters also include the Kappa coefficient of the output single-source image and the segment average precision of each water depth segment;

Among them, n is the number of water depth control points, k represents the water depth section, in formula 3, δ _k is the average relative error, z _i is the measured value of the i-th water depth control point, z _i ' is its inversion value, formula 4 middle, is the Kappa coefficient, x _ii represents the number of correctly classified control points, x _i+ and x _+i are the row and column boundary values of the error matrix when performing segmental statistics on water depth control points, δ _{ma_k} is the segmental average precision, PA _k is Producer accuracy of the kth water depth segment, UA _k is the user accuracy of the kth water depth segment;

Using the obtained parameters and the whole scene remote sensing image, calculate the single-source water depth inversion result, and correct the instantaneous water depth to the theoretical depth datum, and then segment the result to obtain the water depth segment identification image;

Here, the error matrix of the WorldView-2 image used in the multi-source inversion fusion method at the water depth control point is taken as an example, and it is shown in Table 1.

Table 1: Error matrix of WorldView-2 water depth control points

In the error matrix, the columns represent the ground reference verification information, the classification obtained from the behavioral remote sensing data, the main diagonal elements (such as x ₁₁ , expressed as x _ii in the formula) are the correctly classified pixels, and the off-diagonal elements are the remote sensing data classification The number of pixels in error relative to the ground reference. Therefore, in this experiment, 1 to 4 represent four water depth sections divided into 2m, 5m, and 10m intervals, z is the measured water depth, and z' is the inversion water depth. The columns represent the number of control points in the 4 measured sounding intervals, the rows represent the number of control points in the 4 sounding intervals obtained by inversion of remote sensing images, and the main diagonal is the pixel inversion sounding depth assigned to the correct measured sounding interval The number of points, on the contrary, the number of points outside the line is the number of wrong segments. The producer accuracy (PA) in the error matrix is the probability that the pixel corresponding to this point is classified as k when a water depth control point is assumed to be in the k category when inverting the water depth from the remote sensing image. The number is divided by the sum of the kth column (expressed as x _{+ i} in the formula). User Accuracy (UA) is the percentage of the actual measured water depth of the sounding control point belonging to the kth class when the corresponding pixel of a certain control point is classified into the kth class by image inversion water depth, and its calculation is correctly classified as the kth class The number of divided by the sum classified as k (that is, the sum of row k, that is, x _i+ in the formula).

The Kappa coefficient is a measure of agreement or precision between a remote sensing classification map and reference data, expressed by the probability agreement given by the main diagonal and the total number of rows and columns. The Kappa coefficient in the example is 0.7686, which can be explained that the water depth distribution obtained after inverting water depth from this WorldView-2 image is 76.86% better than the randomly divided water depth segments.

The closer the producer precision and user precision are to 1, the better. The ideal situation is that both producer precision and user precision are 1. Therefore, in order not to lose bias in consideration of trade-offs, but also to reduce the number of parameters finally involved in the decision-making fusion, in this embodiment, the mean value of the two is taken as the fusion parameter, that is, the segment average precision.

Using the fusion parameters and the whole scene remote sensing image, calculate the single-source water depth inversion result, and correct the theoretical depth datum and then segment the result to obtain the water depth segment identification image;

Step 4: Multi-source water depth inversion fusion:

The single-source bathymetric inversion results, bathymetric identification images and fusion parameters are used as the input of multi-source bathymetric inversion fusion, and the fusion is carried out pixel by pixel. The final value is determined by the number of votes cast by 4 single sources (i.e. remote sensing images) in the water depth section of the current pixel, as follows:

a) When the number of votes for a certain water depth segment is greater than or equal to 3, it means that there are 3 or more image inversion results in the same water depth segment. At this time, if there are 2 or more images with equal water depth value, assign this value directly to the current pixel, otherwise, compare the average relative error and average precision of these images in the water depth section, and try to choose the one with the largest average accuracy and small average relative error as the water depth value of the current pixel . On the premise that the average precision is relatively large, the average relative error of the image in this water depth section is investigated, and if the average relative error of the image with the highest average precision is also the largest, then the image with the second highest average precision is discarded and selected;

b) When the maximum number of votes is equal to 2, and the votes are 2 and 2, it means that the inversion water depths of two images respectively fall in the same water depth range. At this time, examine the Kappa coefficient and the average precision of the four classifiers in their corresponding water depth segments. If the image with the largest Kappa coefficient and the largest average accuracy are both determined to be in the same water depth segment and are images of the same origin, then select the image pixel As a result, if the water depth value is not the same scene image, the average relative error of the two scenes in this water depth segment is selected. If the water depth segment determined by the image with the largest Kappa coefficient is different from the image with the largest average precision, the former water depth value is selected. When the maximum number of votes is equal to 2, and the number of votes is 2, 1, 1, in the water depth section where the number of votes is 2, try to choose the water depth value of the current pixel with the largest average precision and small average relative error;

c) When the maximum number of votes is 1, that is, the classification results of the four single sources are all different, that is to say, the inversion water depths of the four scene images are distributed in the four water depth segments, and the image with the largest Kappa coefficient is believed at this time.

Step 5: Water depth inversion accuracy verification:

Use checkpoints to verify the accuracy of single-source inversion results and multi-source inversion results, and calculate the average relative error and average absolute error of the whole and different water depth segments, so as to verify the accuracy of multi-source water depth inversion fusion.

(1) Water depth multi-source remote sensing inversion fusion parameters and fusion model implementation

This embodiment carries out water depth multi-source reflection on QuickBird on January 10, 2008, WorldView-2 on February 7, 2010, Pleiades on March 9, 2012, and SPOT-6 on April 5, 2013. The process of evolution and fusion is compared and verified. Table 2 shows the inversion parameters obtained by inverting the water depth through the logarithmic linear model of the blue, green, and red bands, and the segmented average relative error at the control points. Depth retrieval fusion of multi-source remote sensing images is based on WorldView-2 depth retrieval images, and is combined with the depth retrieval results of Pleiades, QuickBird and SPOT-6 images for decision-making. The order of the image types in Table 2 is arranged according to the spatial resolution of the image gradually increasing from left to right, and the segment average precision and segment average relative error of 1-4 represent the points divided by 2m, 5m, and 10m in turn. 4 water depth sections.

Table 2: Water depth single-source remote sensing inversion parameters and multi-source decision-making fusion parameters

After comparative analysis, it can be seen that the SPOT-6 image has the highest overall segmentation accuracy, and the QuickBird image has the worst accuracy. Considering the segment average precision and segment average relative error, the best in the first segment is the Pleiades image, which has the highest segment accuracy and the average relative error in this segment is also smaller, followed by the SPOT-6 image , its average relative error is 8 percentage points smaller than the former, but the segment average precision is worse than the former. Although the segment average accuracy of the Pleiades image is the best in the second segment, its average relative error in this water depth segment is the largest among the four scene images. Therefore, under the premise of ensuring a high segment average accuracy, The average relative error is also better for the SPOT-6 image. In the third and fourth segments, the SPOT-6 image is the best in both segment average precision and segment average relative error.

As shown in Figure 2, in the results of multi-source inversion and fusion of water depth, the generated inversion water depth fusion image has 1002 rows and 1054 columns, that is, a total of 1056108 pixels. According to statistics, the number of pixels whose water depth value is determined according to the second rule is the largest, which is 860835, accounting for 81.51% of all pixels, indicating that when the decision fusion is performed pixel by pixel, the maximum number of votes of most pixels accounts for half of the total Above, that is to say, there are at least three scene images whose retrieved depth values in this pixel are in the same depth range, and the final depth value depends on the depth retrieval image with the highest average accuracy in this depth range. Secondly, the sixth rule is the most executed, 72132, accounting for 6.83%, and the least is the ninth rule, only 128 pixels. Only when the water depth values retrieved by the four images are all in different water depth segments, the ninth rule is used, which means that although the water depth inversion capabilities of the four images are different, the results in the water depth segments will not differ too much , only a small part will have obvious differences, so the images of these four scenes can all play a certain role in decision-making fusion.

(2) Overall accuracy verification analysis of water depth multi-source remote sensing inversion fusion

The accuracy of the fusion results of water depth multi-source inversion is compared with the single-source results before fusion, and the accuracy evaluation indicators obtained are shown in Table 3 below.

Table 3: Comparison of the overall accuracy of water depth multi-source inversion fusion

The three evaluation indicators all show that the results after water depth multi-source decision-making fusion are more significant than the results of the original image inversion. The average relative error from small to large is the decision fusion image, SPOT-6 image, QuickBird image, Pleiades image and WorldView-2 image. Compared with the worst WorldView-2 image, the average relative error of the fused image at the water depth control point The reduction was more than 40 percentage points, and when the result image was initialized before fusion, the inversion result of this image was used as the benchmark, indicating that the decision-making fusion did greatly improve the inversion result of the original image. Even compared with the SPOT-6 image, which has the best inversion accuracy among the 4-scene images, the relative error of the fused image is reduced by 12.7 percent. The difference between the minimum value and the maximum value of the average absolute error is 1.4m, which is obtained by comparing the fusion image or SPOT-6 image with the Pleiades image. The average absolute error of the QuickBird image and WorldView-2 image is relatively large, with values of 1.6m and 1.8m, which differs from the minimum value by as much as 0.8m and 1m. The Kappa coefficient used to evaluate the segmentation accuracy also shows that the pixels of the fused image are more accurate in discriminating the ownership of the water depth segment, followed by the Pleiades image and the SPOT-6 image, and the image accuracy of the water depth segment identification obtained from the QuickBird image inversion is poor . However, it is generally believed that when the Kappa value is greater than 0.80, the consistency between the classification map and the ground reference information is very high or the accuracy is very high. The Kappa values of these four images are all greater than this critical value, indicating that the consistency is relatively good. The worst is the WorldView-2 image, which ranks last with a Kappa value of 0.6139.

As shown in Figures 3a, 3b, 3c, 3d, and 3e, the scatter diagrams of the measured water depth and the retrieved water depth before and after the integration of multi-source inversion of water depth are given. Through the scatter diagram, it can be found that except for the Pleiades image, the inversion effect of the other three scene images is not ideal for water depth points below 2m. In the WorldView-2 image scatter diagram, the distribution of data points is relatively concentrated, and the large average relative error should be affected by the data points in the shallow water area. The maximum bathymetry retrieved from the WorldView-2 image and the Pleiades image is beyond the range of the measured bathymetry checkpoint, which does not appear in the scatter plots of the QuickBird image and the SPOT-6 image.

As shown in Figure 4, this embodiment combines the single-source inversion results of different resolutions through decision fusion. The shallow water area around the island is less than 20 meters in texture and has a small change in water depth. You can clearly see where the North Island is located. In the southwest and northeast of the island, the texture of the deeper deep water area is relatively rough; at a depth of about 20m, the water depth gradient is relatively large, and the transition from shallow to deep is more obvious.

(3) Segmented accuracy verification analysis of water depth multi-source remote sensing inversion fusion

Observing the segmental error distribution of water depth multi-source inversion fusion, as shown in Table 4, as the depth increases, the average relative error and the average absolute error have no regular increase or decrease trend.

Table 4: Segmentation error comparison of water depth multi-source inversion fusion

In the 0-2m water depth section, although the average relative error of the inversion results is generally low, there are still significant gaps. The highest accuracy is the depth multi-source inversion fusion image, with an average relative error of 39.1% and an average absolute error of 0.3m. The second is the Pleiades image, which differs from the former by 3.9% in mean relative error and equal in mean absolute error. Then there are QuickBird images, SPOT-6 images and WorldView-2 images, the average relative error and average absolute error are gradually increasing, especially the WorldView-2 image, its average relative error in this water depth range is SPOT Compared with the best inversion fusion image, the difference is as much as 210.1%, and the average absolute error is almost four times that of the inversion fusion image, which is 1.1m. In the 2-5m water depth section, the inversion fusion image and SPOT-6 image have the highest accuracy, and the average relative error and average absolute error of the two are equal, 5.3% and 0.2m, respectively. The inversion capability of WorldView-2 images in this water depth section has been greatly improved compared with the shallow water section, with an average relative error of 28.8% and an average absolute error of 1.0m. 6 images, the gap is quite obvious. The QuickBird image ranks fourth with an average relative error of 32.8% and an average absolute error of 1.4m, while the Pleiades image, which has the best inversion accuracy in the 0-2m water depth range, has the worst inversion accuracy in this water depth range. In the 5-10m water depth section, the average relative error and average absolute error are arranged in ascending order, including the inversion fusion image, SPOT-6 image, QuickBird image, Pleiades image, and WorldView-2 image. There is a difference of 8 percentages between the minimum and maximum average relative errors, and a maximum difference of 0.6m between the average absolute errors. In the water depth range of 10-20m, the smallest average relative error and average absolute error are from the SPOT-6 image, and the largest is the Pleiades image, with values of 6.3%, 22.5%, and 3.4m, 0.9m, respectively. The inversion accuracy of the fusion image is good, with an average relative error of 6.4% and an average absolute error of 1.0m.

In the fusion inversion of multi-source inversion of water depth, the SPOT-6 image has the highest accuracy among all remote sensing water depth inversion images in other water depths, except that the inversion ability in the shallow water section is not good. The Pleiades image effectively makes up for the deficiency of the SPOT-6 image in shallow water, but its accuracy at 2-5m and 10-20m is the worst among the four scenes. The inversion accuracy of WorldView-2 images is the worst in the two water depth ranges of 0-2m and 5-10m, and the accuracy in the other two water depth ranges is average. However, the inversion accuracy of QuickBird images in each water depth range is at a medium level. Except that the accuracy in the 10-20m water depth section is slightly lower than that of the SPOT-6 image, the inversion accuracy of the multi-source inversion fusion image is the best in other water depth sections.

Compared with the existing inversion method, the present invention can comprehensively utilize the different responses of multiple remote sensing data sources to water depth information, expand the use range of remote sensing image spectral data bands and spectral information, mine water depth data, and improve Inversion accuracy, after decision-making fusion processing, is especially suitable for ocean bathymetry in shallow water areas under complex conditions.

The technical contents not described in detail in the present invention are all known technologies.

Claims (2)

1. shallow water depth multi-source Remote Sensing Images inversion method, it is characterised in that comprise the following steps：

The first step：Multi-spectrum remote sensing image is pre-processed, obtains extra large table reflectivity；

The pretreatment includes radiance conversion, atmospheric correction and solar flare and removed；

Second step：Field measurement water depth value obtains and processing；

The bathymetric data of test block and corresponding latitude and longitude coordinates are obtained, the tidal height value at measurement moment is confirmed by tide table, will Bathymetric data correction obtains the depth of water of theoretical depth reference plane, further according to the acquisition moment of multi-spectrum remote sensing image, to theoretical deep The bathymetric data of degree reference plane carries out the tidal correction of the instantaneous depth of water to obtain the instantaneous depth of water；

3rd step：Single source Depth extraction and depth of water segment identification；

According to the relation between the depth of water at depth of water control point and corresponding image picture element reflectivity, counted using multiband model Return, the input that the parameter exported in the source image Depth extraction merges as multi-source inverting, and multiband model is entered Row parameter calibration, multiband model formation is as follows,

<mrow> <mi>Z</mi> <mo>=</mo> <msub> <mi>A</mi> <mn>0</mn> </msub> <mo>+</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>A</mi> <mi>i</mi> </msub> <msub> <mi>X</mi> <mi>i</mi> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>

X_i=Ln (ρ_i-ρ_si) (2)

Wherein, Z is the depth of water, and n is the wave band number for participating in inverting, A₀And A_iFor undetermined coefficient；ρ_iIt is the i-th wave band reflectivity data, ρ_siIt is the reflectivity at the i-th wave band deep water；

Depth of water control point is divided into multiple depth of water sections as inputting, the average relative error of each depth of water section is exported, as multi-source water Another input, i.e. fusion parameters of deep inverting fusion；The fusion parameters inputted as the fusion of multi-source Depth extraction, in addition to it is defeated The Kappa coefficients of the single source image gone out and the segmental averaging precision of each depth of water section；

<mrow> <msub> <mi>&amp;delta;</mi> <mi>k</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mi>n</mi> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <mfrac> <mrow> <mo>|</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>-</mo> <msup> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>&amp;prime;</mo> </msup> <mo>|</mo> </mrow> <msub> <mi>z</mi> <mi>i</mi> </msub> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow>

<mrow> <mover> <mi>K</mi> <mo>^</mo> </mover> <mo>=</mo> <mfrac> <mrow> <mi>n</mi> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>k</mi> </munderover> <msub> <mi>x</mi> <mrow> <mi>i</mi> <mi>i</mi> </mrow> </msub> <mo>-</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>k</mi> </munderover> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mrow> <mi>i</mi> <mo>+</mo> </mrow> </msub> <mo>&amp;times;</mo> <msub> <mi>x</mi> <mrow> <mo>+</mo> <mi>i</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <msup> <mi>n</mi> <mn>2</mn> </msup> <mo>-</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>k</mi> </munderover> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mrow> <mi>i</mi> <mo>+</mo> </mrow> </msub> <mo>&amp;times;</mo> <msub> <mi>x</mi> <mrow> <mo>+</mo> <mi>i</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>

<mrow> <msub> <mi>&amp;delta;</mi> <mrow> <mi>m</mi> <mi>a</mi> <mo>_</mo> <mi>k</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mrow> <msub> <mi>PA</mi> <mi>k</mi> </msub> <mo>+</mo> <msub> <mi>UA</mi> <mi>k</mi> </msub> </mrow> <mn>2</mn> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>

Wherein, n is depth of water control point number, and k represents depth of water section, in formula 3, δ_kFor average relative error, z_iIt is i-th of depth of water control Make the measured value of point, z_i' it is its inverting value, in formula 4,For Kappa coefficients, x_iiRepresent the control point number correctly classified, x_i+、 x_+iIt is the ranks boundary value of error matrix when segmentation statistics is carried out to depth of water control point, in formula 5, δ_{ma_k}It is segmental averaging precision, PA_kIt is producer's precision of k-th of depth of water section, UA_kIt is user's precision of k-th of depth of water section；

Using fusion parameters and whole scape remote sensing image, single source Depth extraction result is calculated, and is corrected theoretical deep Result is segmented after degree reference plane, obtains depth of water segment identification image；

4th step：Multi-source Depth extraction merges；

Using the Depth extraction result in n kind lists source and its corresponding depth of water segment identification image and fusion parameters as input, by pixel Carry out the fusion of multi-source Depth extraction, specifically include；

A) when the poll of some depth of water section is t, andExplanation hasThe inverting knot of kind or more kind image Fruit be in same depth of water section, whereinExpression rounds downwards, now, if 2 kinds or image of more than two kinds obtain phase Deng Depth extraction value, then directly assign this value for current pixel, otherwise, compare this several image in the average relative of the depth of water section Error and mean accuracy, using depth of water section mean accuracy maximum as final pixel value；Only when the depth of water section mean accuracy is maximum Image corresponding to depth of water section average relative error it is also maximum when, select the image that takes second place of mean accuracy；

B) when maximum number of votes obtained t meetsAnd it is t to have x (x >=2) numbers of votes obtained, now contrasts Kappa coefficients With n grader respective corresponding depth section mean accuracy, if the maximum image of Kappa coefficients and mean accuracy are maximum all It is determined as same depth of water section, and is homologous image, by the water depth value of the image picture element as a result；If not same scape shadow Picture, it is determined that this two scape average relative error in the depth of water section is less；If the water that the maximum image of Kappa coefficients is judged The deep section difference maximum with mean accuracy, then take the former water depth value；It is t's in votes if only 1 number of votes obtained is t In depth of water section, using depth of water section mean accuracy maximum as final pixel value；When the image pair that the depth of water section mean accuracy is maximum When the depth of water section average relative error answered is also maximum, the image that takes second place of mean accuracy is selected；

C) as maximum votes t=1, the water depth value corresponding to the maximum image of Kappa coefficients is selected；

5th step：Depth extraction precision test；

The precision test utilizes multi-source inversion result after single source inversion result before the development fusion of depth of water checkpoint and fusion Compare, after the completion of Depth extraction precision test, using final water depth value as the actual water depth value output data of remote sensing images.

2. shallow water depth multi-source Remote Sensing Images inversion method according to claim 1, it is characterised in that in the first step Radiance conversion be that remote sensing image DN values are converted into spoke brightness value；The solar flare, which removes, can use median method, Value method or wavelet method；The atmospheric correction can use FLAASH, dark pixel or 6S atmospheric correction methods.