Satellite remote sensing image fusion method based on Wrapping-Curvelet transformation
By adopting a satellite remote sensing image fusion method based on Wrapping-Curvelet transform, the problems of block effect and large redundancy in traditional methods are solved, and more efficient image fusion results are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-03-13
AI Technical Summary
Traditional satellite remote sensing image fusion methods suffer from problems such as block effect, numerous implementation parameters, high redundancy, and complex algorithms.
The method based on Wrapping-Curvelet transform is adopted. Step 1 involves spatial registration and histogram matching, step 2 involves Wrapping-Curvelet transform, step 3 involves fusing the scale coefficients of each layer, and step 4 involves reconstructing the image using inverse Wrapping-Curvelet transform. Different fusion rules are used to reduce redundancy and simplify the algorithm.
It outperforms traditional methods in terms of entropy, mean square error, peak signal-to-noise ratio, and similarity, avoids the block effect, and achieves simpler and faster image fusion.
Smart Images
Figure CN121660904A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of satellite remote sensing image fusion, and particularly relates to a satellite remote sensing image fusion method based on Wrapping-Curvelet transform. Background Technology
[0002] Beijing-1 is an Earth observation satellite successfully launched by my country from Russia on October 27, 2005. It is a modern small satellite capable of independent operation, control, and rapid execution of specialized tasks such as disaster monitoring. It has dual remote sensors, capable of simultaneously acquiring 32m medium-resolution multispectral remote sensing image data and 4m high-resolution panchromatic remote sensing image data. As part of an international disaster monitoring constellation, Beijing-1, with its wide swath and short revisit time, will serve as an important and stable data source for urban planning, environmental monitoring, and land use monitoring. Based on the basic ideas and characteristics of Curvelet transform, its biggest advantage over wavelet transform is its high anisotropy, thus possessing a stronger ability to express edge information in images. In image processing, edges are often the most important features, crucial for further processing and analysis. However, in reality, image edges are often weakened by other factors, such as being masked by noise. In this context, the "along" edge information expressed by the Curvelet transform is more advantageous for recovering the visual features of the main structure of an image. Currently, the Curvelet transform has some successful applications in image denoising and enhancement. Research on the Curvelet transform also holds great potential in image fusion. Two-dimensional wavelets are tensor products of one-dimensional wavelets. Using separate transform kernels, a horizontal wavelet transform is first performed on the image, followed by a vertical wavelet transform. This two-dimensional wavelet transform basis is isotropic and cannot express "along" edge information. To address these shortcomings of wavelet transforms, Candès proposed the Ridgelet transform theory. The Ridgelet transform can effectively represent linear singularity features and has good approximation performance for multivariable functions with linear singularities. However, for describing the edges of curved lines in an image, its approximation performance is only equivalent to that of the wavelet transform. To seek better representation methods, Candès et al. proposed the Curvelet transform based on the Ridgelet transform and constructed a tight framework for curved waves. For objective functions with smooth singular curves, the Curvelet transform provides a stable, efficient, and near-optimal representation.
[0003] Curvelet theory has undergone two generations of development. The first generation, based on Ridgelet theory, consisted of a special filtering process and a multi-scale Ridgelet transform. The implementation of the first-generation Curvelet transform involved a series of steps, including subband decomposition, smoothing blocks, normalization, and ridge transform. This method had many parameters, significant redundancy, and inevitably suffered from block effects due to the block division. Candès et al. proposed a new framework, called the second-generation Curvelet transform. The second-generation Curvelet is structurally completely different from the first-generation Curvelet. It is unrelated to the Ridgelet transform and its implementation does not require it; the only similarity lies in the abstract mathematical meanings of tight support and framework. Candès et al. proposed two fast discrete implementation methods for the second-generation Curvelet theory: implementations based on USFFT (Unequally-Spaced Fast Fourier Transform) and Wrapping. These methods are simpler, faster, and significantly reduce the redundancy of traditional implementation algorithms. Like continuous Curvelet transform, Ridgelet transform, and wavelet transform, sparse representation of signals is achieved through the inner product of basis functions and the signal. Input is given in Cartesian coordinates. Its digital Curvelet transform coefficients are
[0004] (1) in, Represents a digital Curvelet waveform. These represent the scale, orientation, and position parameters, respectively. The Curvelet transform is implemented in the frequency domain, using a window function in the frequency domain. To achieve Representation in the frequency domain.
[0005] In the continuous-domain Curvelet transform, the window function Smooth extraction of second-order rings and angle The frequency within. This division is suitable for the Cartesian coordinate system; therefore, in the Cartesian coordinate system, concentric square regions are used. A Cartesian ring is equivalent to a concentric square (not a circle). For clusters... , Cartesian analysis can be converted to
[0006] (2) in, The inner product of a one-dimensional low-pass window is defined as... .function satisfy ,exist superior ,exist superior It can be verified.
[0007] The above is the Cartesian form of scale division. Now, let's analyze the position of the direction. Let...
[0008] (3) in, Is the domain in A real-valued function that satisfies , .
[0009] Using formula (2) and in formula (3) Define the Cartesian window as (4) Obviously, In wedge Nearby separation frequency, can be used to Consider them as basis functions on a wedge.
[0010] Therefore, the fast discrete Curvelet transform based on wrapping (or simply the wrapping-Curvelet transform) can be expressed as: (5) in, It is about , It is about The parameters represent the window function. The length and width components of the support interval, , The core idea of the Wrapping-Curvelet transform method is to wrap around the origin, meaning that in the specific implementation, any region is mapped one-to-one to a rectangular region of the origin through periodization techniques. The specific implementation steps of the Wrapping-Curvelet transform are as follows:
[0011] Step 1: For a given two-dimensional function in Cartesian coordinates, obtain its two-dimensional frequency domain representation using the two-dimensional fast Fourier transform. , , .
[0012] Step 2: In the frequency domain, for each scale / angle pair Get the product .
[0013] Step 3: Wrap around the origin and multiply by the product in Step 2 to obtain... , , .
[0014] Step 4 Perform a two-dimensional inverse fast Fourier transform to obtain discrete Curvelet coefficients. .
[0015] Currently, traditional satellite remote sensing image fusion methods suffer from problems such as the block effect caused by segmentation, numerous parameters, high redundancy, and complex algorithms. Summary of the Invention
[0016] The purpose of this invention is to provide a satellite remote sensing image fusion method based on Wrapping-Curvelet transform, which aims to solve the problems of block effect caused by block division in traditional satellite remote sensing image fusion methods, as well as the problems of many parameters, large redundancy, and complex algorithms.
[0017] The present invention is implemented as follows: a satellite remote sensing image fusion method based on Wrapping-Curvelet transform, the method comprising the following steps: Step 1, obtaining a multispectral image with the same pixel size as the panchromatic image, refers to spatially registering the multispectral image with the panchromatic image, and then resampling the multispectral image using bilinear interpolation to obtain a multispectral image with the same pixel size as the panchromatic image. Step 2, obtaining a new panchromatic image, refers to performing histogram matching on the three bands of the panchromatic image and the multispectral image to be fused to obtain a new panchromatic image; Step 3: Perform Wrapping-Curvelet Transform. This means taking the multispectral bands and their corresponding panchromatic images as examples, performing Wrapping-Curvelet Transform on each band to obtain their respective Curvelet coefficients, including Coarse scale coefficients, Detail scale coefficients, and Fine scale coefficients. Step 4, fusing the scale coefficients of each layer, refers to fusing the Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient of each band separately; Step 5, Inverse Wrapping-Curvelet Transform, refers to performing an inverse Wrapping-Curvelet transform on the layered fusion of each band to obtain the fused image data for each band.
[0018] Step six, obtaining the fused image, refers to reconstructing the fused image data after the inverse Wrapping-Curvelet transform of each band to obtain the fused image of each band.
[0019] Furthermore, in the fusion of each layer, the Coarse scale factor is the Coarse scale factor of the multispectral data, the Detail scale factor is the larger of the absolute values of the multispectral and panchromatic Detail scale factors, and the Fine scale factor is the Fine scale factor of the panchromatic data.
[0020] Furthermore, the satellite remote sensing image fusion method based on Wrapping-Curvelet transform selects a 32m multispectral remote sensing image and a 4m panchromatic remote sensing image from an artificial small satellite. The multispectral remote sensing image parameters are three bands: green B1: 520-620nm, red B2: 630-690nm, and near-infrared B3: 760-900nm. The panchromatic remote sensing image parameters are 500-800nm. The data are derived from the three-band multispectral remote sensing image dmc+4bj1l20407260035249_035709_0_2_0 and the one-band panchromatic remote sensing image dmc+4bj1l20106214034929_035437_12_0.
[0021] Furthermore, the obtained fused image includes scale, position parameters, and orientation parameters in the Wrapping-Curvelet transform. The Coarse scale coefficient after the Wrapping-Curvelet transform is taken from the Coarse scale coefficient of the multispectral image, the Detail scale coefficient is taken from the larger absolute value of the Detail scale coefficient of the multispectral and panchromatic images, and the Fine scale coefficient is taken from the Fine scale coefficient of the panchromatic image.
[0022] The satellite remote sensing image fusion method based on Wrapping-Curvelet transform provided by this invention performs Wrapping-Curvelet transform on the image, and then fuses the Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient separately using different fusion rules. Finally, it reconstructs the fused image using inverse Wrapping-Curvelet transform. It outperforms other comparative methods in terms of objective indicators such as entropy, mean square error, peak signal-to-noise ratio, and similarity. It avoids the block effect caused by block division, has fewer parameters, reduces redundancy, and is simple and fast. Attached Figure Description
[0023] Figure 1This is a flowchart of the steps of the satellite remote sensing image fusion method based on Wrapping-Curvelet transform provided in the embodiments of the present invention; Figure 2 This is a flowchart of a satellite remote sensing image fusion method based on Wrapping-Curvelet transform provided in an embodiment of the present invention. Detailed Implementation
[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0025] The application principle of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0026] like Figure 1 As shown, the present invention is implemented as follows: a satellite remote sensing image fusion method based on Wrapping-Curvelet transform. The method includes the following steps: S101 obtaining a multispectral (Ms) image with the same pixel size as the panchromatic (Pan) image; S102 obtaining new panchromatic images Pan1, Pan2, and Pan3; S103 performing Wrapping-Curvelet transform; S104 fusing each layer; S105 performing inverse Wrapping-Curvelet transform; and S106 obtaining the fused image. The aforementioned obtaining a multispectral image with the same pixel size as the panchromatic image S101 refers to spatially registering the multispectral image with the panchromatic image, and then resampling the multispectral image using bilinear interpolation to obtain a multispectral image with the same pixel size as the panchromatic image. The aforementioned obtaining new panchromatic images Pan1, Pan2, and Pan3S102 refers to performing histogram matching on the panchromatic images and the three bands Ms_B1, Ms_B1, and Ms_B3 of the multispectral image to be fused, respectively, to obtain new panchromatic images Pan1, Pan2, and Pan3. The aforementioned Wrapping-Curvelet transform S103 refers to performing Wrapping-Curvelet transform on multispectral band 1 (Ms_B1) and the corresponding panchromatic image Pan1 as examples, respectively, to obtain their respective Curvelet coefficients. and ,in Indicates scale. Indicates direction, Represents the first level on the scale layer Matrix coordinates in each direction, first layer The Coarse scale layer, also known as the low-frequency coefficient layer, contains an overview of the image. The intermediate layer... , … This is called the Detail scale layer, which contains mid-to-high frequency coefficients and mainly includes the coefficient specifications of edge feature surfaces. It is the outermost layer. This is called the Fine scale layer, which consists of high-frequency coefficients that reflect the details and edge features of the image. The aforementioned fusion of scale coefficients at each layer, S104, refers to, taking multispectral band 1 (Ms_B1) and the corresponding panchromatic image Pan1 as an example, fusing their corresponding Curvelet coefficients. and Each layer is fused separately, that is, the corresponding Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient of each layer are fused separately to obtain the fused Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient of each band. The fusion rule for the Coarse scale coefficient is to take the Coarse scale coefficient of the multispectral spectrum, the fusion rule for the Detail scale coefficient is to take the larger absolute value of the Detail scale coefficient of the multispectral spectrum and the panchromatic spectrum, and the fusion rule for the Fine scale coefficient is to take the Fine scale coefficient of the panchromatic spectrum. The inverse Wrapping-Curvelet transform S105 refers to performing the inverse Wrapping-Curvelet transform on the scale coefficients (Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient) of each layer after fusion of each band to obtain the fused image data of each band. The aforementioned obtaining the fused image S106 refers to reconstructing the fused image data after the inverse Wrapping-Curvelet transform of each band to obtain the fused image of each band.
[0027] Furthermore, the satellite remote sensing image fusion method based on Wrapping-Curvelet transform selects multispectral and panchromatic remote sensing images from the Beijing-1 microsatellite. This satellite has dual remote sensors, capable of simultaneously acquiring three bands of 32m multispectral data and 4m panchromatic data. The spatial resolution ratio of the multispectral image to the panchromatic image is 1:8. The parameters of the multispectral remote sensing image are: green B1: 520-620nm, red B2: 630-690nm, near-infrared B3: 760-900nm. The parameters of the panchromatic remote sensing image are: 500-800nm. The data are derived from the three-band multispectral remote sensing image dmc+4bj1l20407260035249_035709_0_2_0 and the one-band panchromatic remote sensing image dmc+4bj1l20106214034929_035437_12_0.
[0028] Furthermore, the obtained fused image S106 includes scale and position parameters in the Wrapping-Curvelet transform, and adds a direction parameter to achieve a more "sparse" representation of curves and edges. This results in better performance when applied to image fusion. In the selection of fusion rules, the Coarse scale coefficient is taken from the Coarse scale coefficient of the multispectral spectrum, the Detail scale coefficient is taken from the larger absolute value of the Detail scale coefficient of the multispectral spectrum and the panchromatic spectrum, and the Fine scale coefficient is taken from the Fine scale coefficient of the panchromatic spectrum.
[0029] Working principle like Figure 1As shown, a satellite remote sensing image fusion method based on Wrapping-Curvelet transform is described. The method's steps include: S101 obtaining a multispectral image with the same pixel size as the panchromatic image; S102 obtaining new panchromatic images Pan1, Pan2, and Pan3; S103 performing a Wrapping-Curvelet transform; S104 fusing the layers; S105 performing an inverse Wrapping-Curvelet transform; and S106 obtaining the fused image. The step S101, obtaining the multispectral image with the same pixel size as the panchromatic image, refers to spatially registering the multispectral image with the panchromatic image and then merging the multispectral images according to... Bilinear interpolation resampling is used to obtain a multispectral image with the same pixel size as the panchromatic image. The process of obtaining new panchromatic images Pan1, Pan2, and Pan3 (S102) involves performing histogram matching on the three bands Ms_B1, Ms_B1, and Ms_B3 of the panchromatic image and the multispectral image to be fused, respectively, to obtain new panchromatic images Pan1, Pan2, and Pan3. The process of performing a Wrapping-Curvelet transform (S103) involves performing a Wrapping-Curvelet transform on multispectral band 1 (Ms_B1) and the corresponding panchromatic image Pan1, respectively, to obtain their respective Curvelet coefficients. and ,in Indicates scale. Indicates direction, Represents the first level on the scale layer Matrix coordinates in each direction, first layer The Coarse scale layer, also known as the low-frequency coefficient layer, contains an overview of the image. The intermediate layer... , … This is called the Detail scale layer, which contains mid-to-high frequency coefficients and mainly includes the coefficient specifications of edge feature surfaces. It is the outermost layer. The Fine scale layer, also known as the high-frequency coefficients, reflects the details and edge features of the image. The fusion of scale coefficients from each layer (S104) refers to, taking multispectral band 1 (Ms_B1) and the corresponding panchromatic image Pan1 as an example, fusing their corresponding Curvelet coefficients. and Each layer is fused separately, that is, the corresponding Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient of each layer are fused separately to obtain the fused Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient of each band. The fusion rule for the Coarse scale coefficient is to take the Coarse scale coefficient of the multispectral spectrum, the fusion rule for the Detail scale coefficient is to take the larger absolute value of the Detail scale coefficient of the multispectral spectrum and the panchromatic spectrum, and the fusion rule for the Fine scale coefficient is to take the Fine scale coefficient of the panchromatic spectrum. The inverse Wrapping-Curvelet transform S105 refers to performing the inverse Wrapping-Curvelet transform on the scale coefficients (Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient) of each layer after fusion for each band to obtain the fused image data of each band. The fused image S106 refers to reconstructing the fused image data after the inverse Wrapping-Curvelet transform for each band to obtain the fused image of each band.
[0030] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A satellite remote sensing image fusion method based on Wrapping-Curvelet transform, characterized in that, The steps of this method include: Step 1, obtaining a multispectral image with the same pixel size as the panchromatic image, refers to spatially registering the multispectral image with the panchromatic image, and then resampling the multispectral image using bilinear interpolation to obtain a multispectral image with the same pixel size as the panchromatic image. Step 2, obtaining a new panchromatic image, refers to performing histogram matching on the three bands of the panchromatic image and the multispectral image to be fused to obtain a new panchromatic image; Step 3: Perform Wrapping-Curvelet Transform. This means taking the three bands of the multispectral image and the corresponding panchromatic image as examples, performing Wrapping-Curvelet Transform on each band separately to obtain their respective Curvelet coefficients, including Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient. Step 4, fusing the scale coefficients of each layer, refers to fusing the Coarse scale coefficient, Detail scale coefficient, and Fine scale coefficient of each band separately; Step 5, Inverse Wrapping-Curvelet Transform, refers to performing the inverse Wrapping-Curvelet transform on each layer after fusion of each band to obtain the fused image data for each band; Step six, obtaining the fused image, refers to reconstructing the fused image data after the inverse Wrapping-Curvelet transform of each band to obtain the fused image of each band.
2. The satellite remote sensing image fusion method based on Wrapping-Curvelet transform as described in claim 1, characterized in that, In the fusion of each layer in each band, the fusion rule for the Coarse scale coefficient is to take the Coarse scale coefficient of the multispectral spectrum, the fusion rule for the Detail scale coefficient is to take the larger absolute value of the Detail scale coefficient of the multispectral spectrum and the panchromatic spectrum, and the fusion rule for the Fine scale coefficient is to take the Fine scale coefficient of the panchromatic spectrum.
3. The satellite remote sensing image fusion method based on Wrapping-Curvelet transform as described in claim 1, characterized in that, The satellite remote sensing image fusion method based on Wrapping-Curvelet transform selects 32m multispectral remote sensing images and 4m panchromatic remote sensing images from artificial small satellites. The multispectral remote sensing image parameters are three bands: green B1: 520-620nm, red B2: 630-690nm, and near-infrared B3: 760-900nm. The panchromatic remote sensing image parameters are 500-800nm. The data comes from the three-band multispectral remote sensing image dmc+4bj1l20407260035249_035709_0_2_0 and the one-band panchromatic remote sensing image dmc+4bj1l20106214034929_035437_12_0.
4. The satellite remote sensing image fusion method based on Wrapping-Curvelet transform as described in claim 1, characterized in that, The resulting fused image is obtained by using a fusion rule that includes scale, position, and orientation parameters in the Wrapping-Curvelet transform, taking the Coarse scale coefficient of the multispectral image as the fusion factor, the Detail scale coefficient as the larger of the absolute values of the multispectral and panchromatic Detail scale coefficients, and the Fine scale coefficient as the panchromatic Fine scale coefficient.