Island sea area multi-source remote sensing data fusion method based on mixed scale regression optimization

Abstract

The invention relates to the technical field of remote sensing data fusion, and discloses an island sea area multi-source remote sensing data fusion method based on mixed scale regression optimization, which comprises the following steps: step 1, acquiring a multi-source original remote sensing image of an island sea area, and processing the multi-source original remote sensing image; 2, establishing an MRA model based on detail scale regression; 3, improving the MRA model by using an HPM injection method; 4, adding the detail image of the full-scale regression, the spatial scale residual error information and the spectral scale residual error information into scale regression, generating an iterative convergence mixed scale regression model, and completing multi-source remote sensing data fusion; the method has the advantages of better balancing island space detail enhancement, clear edge and island surrounding sea area spectrum reservation, adapting to high-precision water color inversion, having good adaptive ability and the like, and solves the problems of poor space detail enhancement, sea area water body spectrum distortion and the like of an existing panchromatic sharpening method in an island sea area application scene.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing and image processing, and specifically relates to a method for fusion of multi-source remote sensing data of island sea areas based on mixed scale regression optimization. Background Art

[0002] With the development of image acquisition technology, high-precision satellite images play an important role in many important marine remote sensing applications, including environmental monitoring, spectral decomposition, water quality assessment, and water color inversion. Nevertheless, since optical remote sensing systems are limited by imaging indicators such as optical diffraction limit, modulation transfer function, and signal-to-noise ratio, it is often a difficult task to simultaneously obtain satellite-borne remote sensing images with both high spatial and high spectral properties or design multispectral (MS) instruments with high spatial resolution. At the same time, many existing Earth observation programs have also attempted to alleviate these limitations by including multiple dedicated satellites that meet specific spatial spectral requirements. Sentinel-2 (S2) and Sentinel-3 (S3) satellites exemplify this trend. The MSI products of S2 are generally more suitable for land cover characterization tasks, while the OLCI products of S3 focus more on the spectral characteristics of oceans, inland waterways, and coastal areas due to their limited spatial resolution.

[0003] Ocean fusion is different from land image fusion. The water bodies in island waters are changing all the time. There are significant deviations in the spectral and spatial features of the two images to be fused. In addition, the inherent deviation of the sensor and the inconsistency of the imaging time of the remote sensing images will lead to spatial blur and spectral distortion, which cannot be solved by the point-surface information fusion method based on geographically weighted regression. For this reason, we tend to use the covariance-guided feature matching method of spatial-spectral information fusion to alleviate it. At the same time, the unprecedented availability of sentinel data between sensors also brings us the opportunity to deal with this limitation from the perspective of pan-sharpening in the spatial-spectral information fusion framework and injection gain. Pan-sharpening is to fuse high spatial resolution panchromatic (PAN) images and high spectral resolution multispectral (MS) images and their corresponding spatial and spectral information to achieve complementary feature fusion to obtain high spatial and high spectral resolution images (HRMS).

[0004] Classical algorithms for pan-sharpening retain their importance with the key advantage of high computational efficiency, and can be divided into two major categories: component substitution (CS) and multi-resolution analysis (MRA). The former, often referred to as the "spectral class", relies on projection into the transform domain of the original MS image to separate its spatial information and replace it with the PAN image. Many pioneering pan-sharpening techniques belong to the CS class, thanks to their ease of implementation. Notable examples are intensity-hue-saturation (IHS), principal component substitution (PCA), Gram-Schmidt (GS) spectral sharpening, GS adaptive methods, and band-dependent spatial details (BDSD), etc., but the spectral distortion is serious compared to the latter, which is often referred to as the "spatial class", because the MRA class directly applies multi-scale decomposition to the PAN image to extract its spatial components and injects the extracted spatial information into the interpolated MS image. In addition, there are VO methods and deep learning DL methods. The principle of the VO method is derived from image restoration. It uses observation models and sparse representation methods to construct energy functions, and then uses optimization algorithms to solve them to obtain pan-sharpened images. However, since the target image is usually estimated under the assumption of appropriate co-registered PAN and LRMS images, the computational cost is relatively high. Unlike the VO method, the ML method directly learns the mapping from the observed true value to the true value through a neural network, such as PNN and PANnet. However, deep learning usually requires a large amount of labeled training data to achieve the required performance, and there are few remote sensing data sets in the corresponding field. Different satellites have different data types, so it is difficult to train different satellite data sets at the same time, which is particularly disadvantageous in ocean island application scenarios. In addition, training network models also takes a lot of time.

[0005] Based on this, the present invention proposes a multi-source remote sensing data fusion method for island and sea areas based on mixed-scale regression optimization to solve the problems existing in the above-mentioned prior art. Summary of the invention

[0006] 1. Technical issues to be resolved

[0007] In view of the shortcomings of the prior art, the present invention provides a method for fusion of multi-source remote sensing data of island sea areas based on hybrid scale regression optimization, which has the advantages of being able to better balance the enhancement of island spatial details, retaining edge clarity and the spectrum of the sea areas around the islands, adapting to high-precision water color inversion and having good adaptability. It solves the problems of poor spatial detail enhancement and spectral distortion of sea water bodies in the existing panchromatic sharpening method in island sea area application scenarios.

[0008] (II) Technical solution

[0009] In order to achieve the above-mentioned purpose of improving the resolution of remote sensing images, retaining the spectrum of islands and sea areas to a greater extent, and realizing it quickly based on context, the present invention provides the following technical solutions:

[0010] The multi-source remote sensing data fusion method for island and sea areas based on mixed-scale regression optimization includes:

[0011] Step 1: Obtain multi-source original remote sensing images of the island and sea area, and process the multi-source original remote sensing images by reprojection, resampling, and band processing to generate coupled image pairs;

[0012] Step 2: Based on the remote sensing image processed in step 1, an MRA model based on detail scale regression is established;

[0013] Step 3: Use the HPM injection method to improve the MRA model established in step 2, so that the injection coefficient of the original HPM established in step 2 is added with the information of the upscale regression;

[0014] Step 4: Add the full-scale regression detail image obtained in step 3, the spatial scale residual information, and the spectral scale residual information to the scale regression, calculate the mixed weight coefficient through mutual information to generate an iterative convergent mixed-scale regression model, obtain the final panchromatic sharpening result, and complete the multi-source remote sensing data fusion.

[0015] As one of the preferred embodiments of the present invention, the process of establishing a multi-resolution analysis model based on detail scale regression includes:

[0016] Step 2.1: The original multispectral image MS obtained by step 1 is interpolated into an MS image of the original panchromatic image PAN ratio by 23 interpolation method.

[0017] Step 2.2: Based on the generalized Laplacian pyramid of Gaussian filter and the matching modulation transfer function of the multispectral sensor, the low-resolution PAN image P is obtained using the original PAN image P processed in step 1. L ;

[0018] Step 2.3: Inject coefficient g to obtain the pan-sharpening result of the kth spectral band for:

[0019]

[0020]

[0021]

[0022]

[0023] Among them, IPL is the detail image after full-scale regression, μ and std represent the mean and standard deviation respectively, COV is the covariance of the two images, and the injection coefficient g is obtained through multiple iterations. k , P is the original full-color PAN image.

[0024] As one of the preferred embodiments of the present invention, the HPM injection method model is:

[0025]

[0026] in, is the result of pan-sharpening of the kth spectral band, is the MS image of the original panchromatic image PAN ratio, P L is a low-resolution PAN image, and P is the original PAN image.

[0027] As one of the preferred embodiments of the present invention, the process of adding detail scale regression to the HPM injection method includes:

[0028] Step 3.1: Assume that the digital MS or PAN image acquired by sensor s is defined as the convolution of the sensor spatial response and the total energy collected by the sensor in its spectral band and is:

[0029]

[0030] Among them, ε s , k s are the additive constant (or offset) and the multiplicative constant (or gain), respectively; s is the spatial response of the sensor, ω S is the integral of the sensor radiation frequency weighted by the relative spectral response;

[0031] The numbers in formula (5) are rewritten as:

[0032]

[0033]

[0034]

[0035] in, and Represent the spatial responses of LR-MS image, LR-PAN image and HR-PAN image respectively; ω ms and ω P Represent the total energy of MS and PAN images respectively; k ms and k p represents a multiplication constant;

[0036] Step 3.2: The purpose of pan-sharpening is to obtain high-resolution MS images The spatial response of HR-PAN The total energy is ω ms ; The target expression is defined as:

[0037]

[0038] According to the HPM injection method, substituting equations (7)-(9) into equation (5), we get:

[0039]

[0040] set up The multiplication coefficient of , converts equation (11) into the target expression (12):

[0041]

[0042] The spatial response of the HR-MS sensor is the same as that of the PAN sensor, so the first equation is:

[0043]

[0044] MTF-GLP uses filters to match the spatial response of the LR-MS sensor, resulting in the second equation:

[0045]

[0046] According to multiple regression, the third equation is:

[0047]

[0048] get:

[0049]

[0050]

[0051] Based on (16) and (17), a linear affine function is proposed to solve for:

[0052]

[0053] Therefore, the problem is transformed into finding the coefficients m and n to obtain Problems;

[0054] Step 3.3: Mixed-scale regression The coefficients m and n between the HR-PAN image and the HR-MS image are calculated as follows:

[0055]

[0056]

[0057] Where E(X) represents the average value of image X;

[0058] Rewrite equation (12) as:

[0059]

[0060] in,

[0061]

[0062] As one of the preferred embodiments of the present invention, the detail image of the full-scale regression obtained by processing in step 3 and the spatial scale residual information and the spectral scale residual information are added to the scale regression, and an iterative convergent hybrid scale regression model is generated by calculating the hybrid weight coefficient through mutual information to obtain the final panchromatic sharpening result. The process of completing the multi-source remote sensing data fusion includes:

[0063] Step 4.1: The detail image I adjusted by full-scale regression PL The high residual scale information of the covariance with P, and considering I PL and The low residual scale information of the covariance between is added to the scale regression, resulting in:

[0064]

[0065] Among them, MI is based on the calculated detail image I PL With P L The mutual information of

[0066]

[0067] Step 4.2: Introduce the final injection coefficient, calculate the mixing weight coefficient through mutual information to generate an iterative convergent mixed-scale regression model, and obtain a closed solution through the iterative process.

[0068] (III) Beneficial effects

[0069] Compared with the prior art, the present invention provides a method for fusion of multi-source remote sensing data of island and sea areas based on mixed scale regression optimization, which has the following beneficial effects:

[0070] 1. This method adds mixed scale information, i.e., the detailed image after full-scale regression, spatial scale residual information, and spectral scale residual information, to the scale regression, which greatly enriches the scale regression information to better balance the island spatial detail enhancement, edge clarity, and the retention of the spectrum of the sea area around the island, so that it can better adapt to high-precision water color inversion;

[0071] 2. This method utilizes the correlation of mutual information at the level of hybrid weight calculation. Compared with other similar algorithms, it can adaptively calculate the hybrid weight of the correlation of the input image. It has the advantages of simplified process and fast implementation, effectively improving its adaptability. It has excellent performance in the application scenarios of islands and seas, including improving the resolution of remote sensing images, retaining the spectrum of islands and seas to the greatest extent, and high efficiency based on rapid implementation based on context; and it has a certain inhibitory effect on the spectral distortion of panchromatic sharpening in islands and seas; it effectively solves the problems of poor spatial detail enhancement and spectral distortion of sea water bodies in the existing panchromatic sharpening methods in the application scenarios of islands and seas, and achieves the purpose of improving the resolution of remote sensing images, retaining the spectrum of islands and seas to the greatest extent, and rapid implementation based on context. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a flow chart of the multi-source remote sensing data fusion method of island and sea areas based on mixed-scale regression optimization of the present invention (taking the Sentinel series as an example);

[0073] Figure 2 This is an example of the original processed full-color Huangyan Reef remote sensing image of the present invention;

[0074] Figure 3 This is an example of the original processed multi-spectral Huangyan Reef remote sensing image of the present invention;

[0075] Figure 4 The full-resolution result map of the Huangyan Reef data pair using various methods of the present invention;

[0076] Figure 5 It is to frame a sub-area on the Huangyan Reef data and make a water body remote sensing spectral error map.

[0077] In the picture: Figure 4In the figure, Figure (a) is the full-resolution result of the Huangyanjiao data pair based on the BDSD-PC method, Figure (b) is the full-resolution result of the Huangyanjiao data pair based on the C-GSA method, Figure (c) is the full-resolution result of the Huangyanjiao data pair based on the AWLP method, Figure (d) is the full-resolution result of the Huangyanjiao data pair based on the MTF-GLP method, Figure (e) is the full-resolution result of the Huangyanjiao data pair based on the MTF-GLP-HPM method, Figure (f) is the full-resolution result of the Huangyanjiao data pair based on the MTF-GLP-HPM-R method, and Figure (g) is the full-resolution result of the Huangyanjiao data pair based on the MTF-GLP-HPM-CBD method. Figure (h) is the full-resolution result of the Huangyanjiao data pair based on the MTF-GLP-REG-FS method, Figure (i) is the full-resolution result of the Huangyanjiao data pair based on the TV method, Figure (j) is the full-resolution result of the Huangyanjiao data pair based on the RR method, Figure (k) is the full-resolution result of the Huangyanjiao data pair based on the MF method, Figure (l) is the full-resolution result of the Huangyanjiao data pair based on the FE-HPM method, Figure (m) is the full-resolution result of the Huangyanjiao data pair based on the PWMBF method, and Figure (n) is the full-resolution result of the Huangyanjiao data pair based on the MTF-GLP-HPM-HSMI method;

[0078] exist Figure 5In the figure, from left to right are the framed sub-regions, and Figure (a) is a water remote sensing spectral error map of the framed sub-region on the Huangyanjiao data based on the BDSD-PC method, Figure (b) is a water remote sensing spectral error map of the framed sub-region on the Huangyanjiao data based on the C-GSA method, Figure (c) is a water remote sensing spectral error map of the framed sub-region on the Huangyanjiao data based on the AWLP method, Figure (d) is a water remote sensing spectral error map of the framed sub-region on the Huangyanjiao data based on the MTF-GLP method, Figure (e) is a water remote sensing spectral error map of the framed sub-region on the Huangyanjiao data based on the MTF-GLP-HPM method, Figure (f) is a water remote sensing spectral error map of the framed sub-region on the Huangyanjiao data based on the MTF-GLP-HPM-R method, and Figure (g) is a water remote sensing spectral error map of the framed sub-region on the Huangyanjiao data based on the MTF-GLP-HPM-CBD method. Figure (h) is the water remote sensing spectral error map of the sub-region framed on the Huangyanjiao data based on the MTF-GLP-REG-FS method, Figure (i) is the water remote sensing spectral error map of the sub-region framed on the Huangyanjiao data based on the TV method, Figure (j) is the water remote sensing spectral error map of the sub-region framed on the Huangyanjiao data based on the RR method, Figure (k) is the water remote sensing spectral error map of the sub-region framed on the Huangyanjiao data based on the MF method, Figure (l) is the water remote sensing spectral error map of the sub-region framed on the Huangyanjiao data based on the FE-HPM method, Figure (m) is the water remote sensing spectral error map of the sub-region framed on the Huangyanjiao data based on the PWMBF method, and Figure (n) is the water remote sensing spectral error map of the sub-region framed on the Huangyanjiao data based on the MTF-GLP-HPM-HSMI method. DETAILED DESCRIPTION

[0079] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0080] Embodiment 1:

[0081] See also Figure 1-Figure 5 The present invention provides a technical solution: a method for fusion of multi-source remote sensing data of island and sea areas based on hybrid scale regression optimization, taking Sentinel 2 (S2) and Sentinel 3 (S3) in the Sentinel series as panchromatic and multispectral images as examples, combined with Figure 1 ,include:

[0082] Step 1: Obtain multi-source original remote sensing images of the island and sea area, and process the multi-source original remote sensing images by reprojection, resampling, and band processing to generate coupled image pairs;

[0083] Step 1.1: Cloud-free products downloaded from the Copernicus Open Access Center and processed through the Sentinel Application Platform (SNAP). In the case of Sentinel 2 (S2), the Level-2B MSI images were spatially resampled to 20m using the Sen2Cor processor (using its default settings) to generate a uniform data cube as the panchromatic image PAN input. In the case of Sentinel 3 (S3), the OLCI images were corrected using the available Radiance to Reflectance processor, and in addition, the OLCI images were reprojected onto the relevant S2 tiles to collect and construct the overlap area between the two sensors. Through these steps, we generated S2, S3 coupled image pairs representing the same range on the Earth's surface.

[0084] Step 1.2: Define the input image of the multi-source inter-sensor pan-sharpening scheme under consideration, i.e., the low-resolution MS image LRMS (I MS ), the original PAN image (I PAN ). On the one hand, we only consider those MSI and OLCI bands centered on the same wavelength, namely the red (R), green (G), blue (B), and near-infrared (IR) bands, which are located at 665, 560, 490, and 865 nm. In this way, we define I MS ={Oa04,Oa06,Oa08,Oa17}, and then we generate I PAN The MSI is band averaged and its spatial size is adjusted to R×OLCI. In this embodiment, R=4.

[0085] Step 2: Based on the remote sensing image processed in step 1, a multiresolution analysis (MRA) model based on detail scale regression is established on the generalized Laplacian pyramid based on Gaussian filter and the matching modulation transfer function of the multispectral sensor (MTF-GLP);

[0086] Step 2.1: The original multispectral image MS (low-resolution MS image) obtained by step 1 is interpolated into an MS image of the original panchromatic image P ratio by 23 interpolation method. High-resolution MS images

[0087] Step 2.2: Then, the MTF-GLP algorithm is used to obtain a low-resolution PAN image P from the original PAN image P processed in step 1. L ;

[0088] Step 2.3: Finally, the injection coefficient g is used to control the injection information, such as formula (1) to obtain the full color sharpening result The pan-sharpening result of the kth spectral band is:

[0089]

[0090] Among them, this embodiment adopts detailed mixed scale regression, and obtains the appropriate injection coefficient g through multiple iterations. k , which is defined as follows:

[0091]

[0092]

[0093]

[0094] Among them, I PL is the detail image after full-scale regression, μ and std represent the mean and standard deviation respectively, COV is the covariance of the two images, and the injection coefficient g is obtained through multiple iterations. k , P is the original full-color PAN image.

[0095] Step 3: The MRA model can be improved by using a high-pass modulation (HPM) injection method. Therefore, this embodiment uses the HPM injection method to improve the performance of the MRA model based on scale regression established in step 2, so that the original HPM injection coefficient is added with the scale regression information. The HPM injection method model is as follows:

[0096]

[0097] in, is the result of pan-sharpening of the kth spectral band, is the MS image of the original panchromatic image PAN ratio, P L is a low-resolution PAN image, and P is the original PAN image;

[0098] Now, let us explain the process of incorporating detail scale regression into the HPM injection method:

[0099] Step 3.1: Assume that the digital MS or PAN image acquired by sensor s (HR or LR) is defined as the convolution of the sensor spatial response and the total energy collected by the sensor in its spectral band, and its value is

[0100]

[0101] ε s , k s They are the additive constant (or offset) and the multiplicative constant (or gain); these two constants are introduced to obtain the full digital range of the A / D converter, but ε s It can usually be ignored; s represents the spatial response of the sensor, and ω S is the integral of the sensor radiation frequency weighted by the relative spectral response;

[0102] From this we can rewrite

[0103]

[0104]

[0105]

[0106] in, and Represent the spatial responses of LR-MS image, LR-PAN image and HR-PAN image respectively; ω ms and ω P Represent the total energy of MS and PAN images respectively; k ms and k p represents a multiplication constant;

[0107] Step 3.2: Therefore, the purpose of pan-sharpening is to obtain high-resolution MS images. Its spatial response is HR-PAN The total energy is ω ms ; The target expression is defined as

[0108]

[0109] According to the HPM injection method, substituting equations (7)-(9) into equation (5), we can obtain

[0110]

[0111] By properly setting The multiplication coefficient of can be converted into the target expression, namely formula (12).

[0112] This shows that except All variables except are different from those in (11); therefore, additional transition signs are used when expressing these variables to distinguish actual variables from estimated variables;

[0113]

[0114] Since the spatial response of the HR-MS sensor is the same as that of the existing PAN sensor, the first equation is expressed as

[0115]

[0116] In addition, since the LR-PAN image P L is constructed from the MTF-GLP, which utilizes filters to match the spatial response of the LR-MS sensor, so the second equation is defined as follows:

[0117]

[0118] According to multiple regression, the third equation is defined as

[0119]

[0120] Therefore, we get

[0121]

[0122]

[0123] Based on (16) and (17), a linear affine function is proposed to solve Its mathematical expression is:

[0124]

[0125] Therefore, the problem becomes to find the coefficients m and n to obtain problem.

[0126] Step 3.3: The above mixed-scale regression calculates the coefficients m and n between the HR-PAN image and the HR-MS image, and the calculation formula is as follows:

[0127]

[0128]

[0129] Where E(X) represents the average value of image X;

[0130] Therefore, we rewrite equation (12) as follows

[0131]

[0132] in,

[0133]

[0134] Step 4: According to formula (4), the injection coefficient based on scale regression only considers the covariance of the fine-scale and detail image. Therefore, the regression scale information in g is single, which affects the performance of the model in spectral preservation. In order to enrich the scale information required for regression and improve the final multi-source panchromatic sharpening result, the mixed scale information, i.e., the detail image after full-scale regression, the spatial scale residual information and the spectral scale residual information are added to the scale regression. Finally, the mixed weight coefficient is calculated by mutual information to generate an iterative convergent mixed scale regression model to obtain the final panchromatic sharpening result, which is:

[0135] Step 4.1: The detail image I adjusted by full-scale regression PL The high residual scale information of the covariance with P, and considering I PL and The low residual scale information of the covariance between can be obtained:

[0136]

[0137] Among them, MI is based on the calculated detail image I PL With P L In fact, both extremely small and extremely large MI will lead to poor results, as shown below:

[0138]

[0139] Step 4.2: The final injection coefficient is introduced, and mixed-scale regression is achieved through an iterative process; and the iterative process finally obtains a closed solution, and the iterative method can achieve better performance.

[0140] get Figure 2 and Figure 3 Sentinel 3 is a multispectral image, and Sentinel 2 is a panchromatic image. Combined with the experimental results Figure 4 and Figure 5 The fused image enhances the spatial detail information well, which is reflected in the clear edges of the islands, the image detail resolution is significantly improved, and the layering is more distinct than the original image. At the same time, the visual spectrum is well preserved, which is also reflected in the Figure 4 The spectral error is calculated in the selected sub-region. Figure 5 Spectral error diagram proof; Table 1 is the evaluation index of the panchromatic sharpening result, including Q4, Qavg, SAM and ERGAS of reduced resolution: D s and HQNR.

[0141] Table 1: Evaluation indicators of pan-sharpening results

[0142]

[0143] Table 1 shows the index values ​​of 14 pan-sharpening methods at reduced resolution and full resolution in Experiments 1 and 2. Among all the methods, the proposed MTF-GLP-HPM-HSMI achieves the lowest SAM and the highest Q4 at reduced resolution; and the lowest spectral distortion at full resolution. and the highest HQNR; from the overall results, MTF-GLP-HPM-HSMI has good performance and certain stability in both reduced-resolution evaluation and full-resolution evaluation.

[0144] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0145] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-source remote sensing data fusion method for island and sea areas based on mixed-scale regression optimization, characterized by: include: Step 1: Obtain multi-source original remote sensing images of the island and sea area, and process the multi-source original remote sensing images by reprojection, resampling, and band processing to generate coupled image pairs; Step 2: Based on the remote sensing image processed in step 1, an MRA model based on detail scale regression is established; Step 3: Use the HPM injection method to improve the MRA model established in step 2, so that the injection coefficient of the original HPM established in step 2 is added with the information of the upscale regression; Step 4: Add the full-scale regression detail image obtained in step 3, the spatial scale residual information, and the spectral scale residual information to the scale regression, calculate the mixed weight coefficient through mutual information to generate an iterative convergent mixed-scale regression model, obtain the final panchromatic sharpening result, and complete the multi-source remote sensing data fusion.

2. The method for fusion of multi-source remote sensing data of island and sea areas based on mixed-scale regression optimization as claimed in claim 1, characterized in that: The process of establishing a multi-resolution analysis model based on detail scale regression includes Step 2.1: The original multispectral image MS obtained by step 1 is interpolated into an MS image of the original panchromatic image PAN ratio by 23 interpolation method. Step 2.2: Based on the generalized Laplacian pyramid of Gaussian filter and the matching modulation transfer function of the multispectral sensor, the low-resolution PAN image P is obtained using the original PAN image P processed in step 1. L ; Step 2.3: Inject coefficient g to obtain the pan-sharpening result of the kth spectral band for: Among them, I PL is the detail image after full-scale regression, μ and std represent the mean and standard deviation respectively, COV is the covariance of the two images, and the injection coefficient g is obtained through multiple iterations. k , P is the original full-color PAN image.

3. The method for fusion of multi-source remote sensing data of island and sea areas based on mixed-scale regression optimization as claimed in claim 1, characterized in that: The HPM injection method model is: in, is the result of pan-sharpening of the kth spectral band, is the MS image of the original panchromatic image PAN ratio, P L is a low-resolution PAN image, and P is the original PAN image.

4. The method for fusion of multi-source remote sensing data of island and sea areas based on mixed-scale regression optimization as claimed in claim 3 is characterized by: The process of adding detail scale regression to the HPM injection method includes: Step 3.1: Assume that the digital MS or PAN image acquired by sensor s is defined as the convolution of the sensor spatial response and the total energy collected by the sensor in its spectral band and is: Among them, ε s , k s are the additive constant (or offset) and the multiplicative constant (or gain), respectively; s is the spatial response of the sensor, ω S is the integral of the sensor radiation frequency weighted by the relative spectral response; The numbers in formula (5) are rewritten as: in, and Represent the spatial responses of LR-MS image, LR-PAN image and HR-PAN image respectively; ω ms and ω P Represent the total energy of MS and PAN images respectively; k ms and k p represents a multiplication constant; Step 3.2: The purpose of pan-sharpening is to obtain high-resolution MS images The spatial response of HR-PAN The total energy is ω ms ; The target expression is defined as: According to the HPM injection method, substituting equations (7)-(9) into equation (5), we get: set up The multiplication coefficient of , converts equation (11) into the target expression (12): The spatial response of the HR-MS sensor is the same as that of the PAN sensor, so the first equation is: MTF-GLP uses filters to match the spatial response of the LR-MS sensor, resulting in the second equation: According to multiple regression, the third equation is: get: Based on (16) and (17), a linear affine function is proposed to solve for: Therefore, the problem is transformed into finding the coefficients m and n to obtain Problems; Step 3.3: Mixed-scale regression The coefficients m and n between the HR-PAN image and the HR-MS image are calculated as follows: Where E(X) represents the average value of image X; Rewrite equation (12) as: in, 5. The method for fusion of multi-source remote sensing data of island and sea areas based on mixed-scale regression optimization as claimed in claim 1, characterized in that: The detailed image of the full-scale regression obtained in step 3 and the spatial scale residual information and the spectral scale residual information are added to the scale regression, and an iterative convergent hybrid scale regression model is generated by calculating the hybrid weight coefficient through mutual information to obtain the final panchromatic sharpening result. The process of completing the multi-source remote sensing data fusion includes: Step 4.1: The detail image I adjusted by full-scale regression PL The high residual scale information of the covariance with P, and considering I PL and The low residual scale information of the covariance between is added to the scale regression, resulting in: Among them, MI is based on the calculated detail image I PL With P L The mutual information of Step 4.2: Introduce the final injection coefficient, calculate the mixing weight coefficient through mutual information to generate an iterative convergent mixed-scale regression model, and obtain a closed solution through the iterative process.

Images