Remote sensing image processing method and device

Abstract

The present disclosure relates to the general field of image data processing, and particularly to a method and apparatus for remote sensing image processing. The remote sensing image processing method includes the following steps: obtaining an area image with a preset size around a pixel point in a hyperspectral image or a panchromatic image as a feature description of the pixel point; generating a feature descriptor of the pixel point according to the feature description of the pixel point; obtaining a pixel matching point pair of the hyperspectral image and the panchromatic image according to the feature descriptor of the pixel point in the hyperspectral image and the feature descriptor of the pixel point in the panchromatic image; substituting the horizontal grid basis function, the vertical grid basis function, and the pixel matching point pair of the panchromatic image into a preset matching point pair mapping function to obtain the actual position of the pixel point in the hyperspectral image mapped to the panchromatic image. The remote sensing image processing method provided by the present disclosure can realize the fusion of the distorted hyperspectral image and the panchromatic image, so that the fusion of the hyperspectral image and the panchromatic image can be completed more accurately.

Description

Technical Field

[0001] The present disclosure relates to the general field of image data processing, and in particular to a remote sensing image processing method and apparatus. Background Art

[0002] Remote sensing refers to an image data acquisition technology that detects and records the radiation and reflection characteristics of electromagnetic waves of a target through a hyperspectral camera and an area array camera without contacting the target.

[0003] The area array camera acquires high-resolution panchromatic images, and the hyperspectral camera acquires ground object spectral images. In the prior art, when fusing the two types of images, generally, image feature extraction algorithms such as SIFT or ORB are used to extract image features from the two types of images respectively, and then feature matching and pose relationship estimation are performed based on the extracted image features.

[0004] Since the hyperspectral images acquired by the hyperspectral camera are distorted and there are very large differences in the data between the two, the same image features cannot be directly subjected to affine transformation, resulting in the method of the prior art being unable to accurately complete the fusion of hyperspectral images and panchromatic images. Summary of the Invention

[0005] To solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a remote sensing image processing method and apparatus, which can more accurately complete the fusion of hyperspectral images and panchromatic images.

[0006] In a first aspect, the present disclosure provides a remote sensing image processing method, and the remote sensing image processing method includes the following steps:

[0007] Obtain a regional image with a preset size around a pixel point in a hyperspectral image or a panchromatic image as a feature description of the pixel point;

[0008] Generate a feature descriptor of the pixel point according to the feature description of the pixel point;

[0009] Obtain pixel point matching point pairs of the hyperspectral image and the panchromatic image according to the feature descriptors of the pixel points in the hyperspectral image and the feature descriptors of the pixel points in the panchromatic image;

[0010] Perform n horizontal and m vertical grid divisions on the panchromatic image to obtain horizontal and vertical grid basis functions of the panchromatic image, and substitute the horizontal grid basis function, the vertical grid basis function, and the pixel point matching point pairs of the hyperspectral image and the panchromatic image into a preset matching point pair mapping function to obtain the actual positions of the pixel points in the hyperspectral image mapped to the panchromatic image.

[0011] Optionally, the preset matching point pair mapping function is:

[0012]

[0013] Among them, p(u, v) is the actual position of the pixel point of the hyperspectral image corresponding to the panchromatic image, and M ip (u) is the grid basis function in the vertical direction of the panchromatic image, and M jq (v) is the grid basis function in the horizontal direction of the panchromatic image, and T ij is the pixel point of the hyperspectral image, and S ij is the weight factor of the feature descriptor of the hyperspectral image;

[0014] Among them,

[0015]

[0016]

[0017]

[0018]

[0019] Among them, i is the grid ordinate of the pixel point of the panchromatic image corresponding to the pixel point T of the hyperspectral image ij of the panchromatic image, and j is the grid abscissa of the pixel point of the panchromatic image corresponding to the pixel point T of the hyperspectral image ij of the panchromatic image. H is the height of the panchromatic image, and W is the width of the panchromatic image.

[0020] Optionally, generating a feature descriptor of a pixel point according to the feature description of the pixel point includes the following steps:

[0021] Performing a transformation on the feature description of the pixel point through a preset anti-rotation and anti-illumination transformation formula to obtain a first processed image;

[0022] Scaling the first processed image through a pyramid scaling model with a preset scaling ratio to obtain multiple first processed images with different sizes;

[0023] Inputting the multiple first processed images with different sizes into a feature descriptor generator to obtain a feature descriptor with a preset dimension.

[0024] Optionally, the preset anti-rotation and anti-illumination transformation formula is:

[0025]

[0026]

[0027] Among them, Ω is the feature description of the pixel point, I(x, y) is the pixel value with abscissa x and ordinate y in the feature description, is the pixel average value of the feature description of the pixel point, and xp is the horizontal coordinate pixel value after transformation of the pixel point with the horizontal coordinate x and the vertical coordinate y, y p is the vertical coordinate pixel value after transformation of the pixel point with the horizontal coordinate x and the vertical coordinate y.

[0028] Optionally, the number of layers of the pyramid scaling model is four, and the preset scaling ratio is 1.5 times.

[0029] Optionally, the feature descriptor generator includes a first image feature extraction network and a second image feature extraction network, and the first image feature extraction network and the second image feature extraction network jointly output a feature descriptor of a preset dimension;

[0030] Among them, the first image feature extraction network includes a first number of first image feature extraction layers in series, the second image feature extraction network includes a second number of second image feature extraction layers in series, the first image feature extraction layer includes a first convolutional layer, a first pooling layer and a first activation layer in series, the second image feature extraction layer includes a second convolutional layer, a second pooling layer, a third pooling layer, a second activation layer and a merging layer, the second convolutional layer, the second pooling layer and the second activation layer are in series, the third pooling layer is used to perform maximum pooling on the input of the second convolutional layer and then output, and the merging layer is used to merge the outputs of the second activation layer and the third pooling layer.

[0031] Optionally, the first convolutional layer uses a 3×3 convolutional kernel, the first pooling layer uses 2×2 maximum pooling, and the first activation layer uses the ReLu function as the activation function;

[0032] The second convolutional layer uses a 1×1 convolutional kernel, the second pooling layer and the third pooling layer use 2×2 maximum pooling, and the activation function of the second activation layer is:

[0033] Optionally, the preset dimension is 128 dimensions.

[0034] Optionally, obtaining the pixel matching point pairs of the hyperspectral image and the panchromatic image according to the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image includes the following steps:

[0035] Training to obtain a mapping network for the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image;

[0036] Inputting the feature descriptor of the pixel point of the hyperspectral image or the feature descriptor of the pixel point of the panchromatic image into the mapping network to obtain the pixel matching point pairs of the hyperspectral image and the panchromatic image.

[0037] In a second aspect, the present disclosure provides a remote sensing image processing apparatus, which includes a processor and a memory. At least one instruction, at least one program, a code set or an instruction set is stored in the memory, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the remote sensing image processing method according to any one of the first aspect.

[0038] The technical solution provided by the present disclosure has the following advantages compared with the prior art:

[0039] The remote sensing image processing method provided by the present disclosure extracts the regional images with a preset size around the hyperspectral image and the panchromatic image respectively to obtain the feature descriptions of each pixel point of the hyperspectral image and the panchromatic image. Then, according to the feature descriptions, the feature descriptors of each pixel point of the hyperspectral image and the panchromatic image are obtained, and according to the feature descriptors, the pixel point matching pairs of the hyperspectral image and the panchromatic image are obtained. Finally, according to the pixel point matching pairs, the actual positions of the pixel points of the hyperspectral image mapped onto the panchromatic image are obtained through the matching point pair mapping function provided by the present disclosure. The remote sensing image processing method provided by the present disclosure can realize the fusion of the distorted hyperspectral image and the panchromatic image, so the fusion of the hyperspectral image and the panchromatic image can be completed more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a scenario diagram of the remote sensing image processing method provided by an embodiment of the present disclosure;

[0041] Figure 2 It is a flowchart of the remote sensing image processing method provided by an embodiment of the present disclosure;

[0042] Figure 3 It is a schematic diagram of the network structure of the feature descriptor generator provided by an embodiment of the present disclosure;

[0043] Figure 4 It is a block diagram of the structure of the remote sensing image processing apparatus provided by an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] In order to more clearly understand the above objects, features and advantages of the present disclosure, the solutions of the present disclosure will be further described below. It should be noted that, without conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0045] Many specific details are set forth in the following description in order to fully understand the present disclosure, but the present disclosure may be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of the present disclosure, rather than all the embodiments.

[0046] Figure 1This is an application scenario diagram of the remote sensing image processing method provided by the embodiments of the present disclosure. As Figure 1 shown, the remote sensing image processing method provided by the embodiments of the present disclosure is used to process the fusion of the distorted hyperspectral image 12 and the panchromatic image 11. Since the hyperspectral image 12 obtained by the hyperspectral camera is distorted, there are very large differences in data between the two, and the same image features cannot be directly subjected to affine transformation. Therefore, in the prior art, image feature extraction algorithms such as SIFT or ORB are used to extract image features from the two images respectively, and then feature matching is performed based on the extracted image features. The obtained pose relationship is inaccurate, and the prior art cannot accurately complete the fusion of the distorted hyperspectral image 12 and the panchromatic image 11.

[0047] The remote sensing image processing method provided by the present disclosure extracts the regional images with a preset size around the hyperspectral image 12 and the panchromatic image 11 respectively to obtain the feature descriptions of each pixel point of the hyperspectral image 12 and the panchromatic image 11. Then, according to the feature descriptions, the feature descriptors of each pixel point of the hyperspectral image 12 and the panchromatic image 11 are obtained, and according to the feature descriptors, the pixel point matching pairs of the hyperspectral image 12 and the panchromatic image 11 are obtained. Finally, according to the pixel point matching pairs, the actual positions of the pixel points of the hyperspectral image 12 mapped to the panchromatic image 11 are obtained through the matching point pair mapping function provided by the present disclosure. The remote sensing image processing method provided by the present disclosure can realize the fusion of the distorted hyperspectral image 12 and the panchromatic image 11, so the fusion of the hyperspectral image 12 and the panchromatic image 11 can be completed more accurately.

[0048] Figure 2 This is a flowchart of the remote sensing image processing method provided by the embodiments of the present disclosure.

[0049] S201: Obtain the regional image with a preset size around the pixel point in the hyperspectral image or the panchromatic image as the feature description of the pixel point.

[0050] Specifically, taking a certain pixel point in the hyperspectral image or the panchromatic image as the center, the regional image with a preset size around it is used as the feature description of the pixel point, and the preset size is 47×47 pixels.

[0051] For the pixel points within 47 pixels from the image edge in the hyperspectral image or the panchromatic image, the feature description of the pixel point can be obtained by filling zeros in the missing parts of the surrounding regional image corresponding to the pixel point.

[0052] By analogy, for each pixel point in the hyperspectral image or the panchromatic image in the embodiments of the present disclosure, the regional image with a preset size around the pixel point as the center is obtained as its feature description.

[0053] In this embodiment, pixels within 47 pixels from the image edge of the hyperspectral image or the panchromatic image are prohibited from being used as the pixels for registration and fusion, that is, the pixels within 47 pixels from the image edge of the hyperspectral image or the panchromatic image do not participate in the fusion of the hyperspectral image or the panchromatic image. Therefore, the step of padding zeros to obtain the feature description of the pixels within 47 pixels from the image edge can be discarded.

[0054] S202: Generate a feature descriptor for the pixel based on the feature description of the pixel.

[0055] In the embodiments of the present disclosure, generating a feature descriptor for the pixel based on the feature description of the pixel includes the following steps:

[0056] Transform the feature description of the pixel through a preset anti-rotation and anti-illumination transformation formula to obtain a first processed image.

[0057] Specifically, according to the different sources of the pixels, the feature descriptions of the pixels of the hyperspectral image and the panchromatic image are respectively transformed through a preset anti-rotation and anti-illumination transformation formula to obtain a first processed image of the feature description of the pixels of the hyperspectral image and a first processed image of the feature description of the pixels of the panchromatic image.

[0058] Specifically, the preset anti-rotation and anti-illumination transformation formula is:

[0059]

[0060]

[0061] where Ω is the feature description of the pixel, I(x, y) is the pixel value at the abscissa x and ordinate y in the feature description, is the pixel average value of the feature description of the pixel, x p is the abscissa pixel value after transformation of the pixel at the abscissa x and ordinate y, y p is the ordinate pixel value after transformation of the pixel at the abscissa x and ordinate y.

[0062] By processing the feature description of the pixel through the anti-rotation and anti-illumination transformation formula provided by the present disclosure, a first processed image with the same reference direction can be obtained to describe the feature description of the pixel, realizing the function of anti-rotation of the feature description.

[0063] At the same time, after processing the feature description of the pixel through the anti-rotation and anti-illumination transformation formula provided by the present disclosure, the influence of light intensity on the calculation result can be reduced, realizing the function of anti-illumination of the feature description.

[0064] The first processed image is scaled by a pyramid scaling model with a preset scaling ratio to obtain multiple first processed images of different sizes.

[0065] Specifically, the number of layers of the pyramid scaling model is four, and the scaling ratio between adjacent layers is 1.5 times, that is, the preset scaling ratio is 1.5 times. By importing the first processed image into the pyramid scaling model, multiple first processed images with different scaling ratios are obtained. In this embodiment, since the number of layers of the pyramid scaling model is four, there are a total of five first processed images.

[0066] Figure 3 It is a schematic diagram of the network structure of the feature descriptor generator provided by an embodiment of the present disclosure. Refer to Figure 3 As shown, the multiple first processed images 33 of different sizes are input into the feature descriptor generator to obtain a feature descriptor 34 of a preset dimension.

[0067] Specifically, the feature descriptor generator provided by the embodiment of the present disclosure includes a first image feature extraction network 31 and a second image feature extraction network 32, and the first image feature extraction network 31 and the second image feature extraction network 32 jointly output a feature descriptor 34 of a preset dimension.

[0068] Specifically, in this embodiment, the first image feature extraction network 31 and the second image feature extraction network 32 respectively output feature descriptors 34 with half of the preset dimension quantity.

[0069] Among them, the first image feature extraction network 31 includes a first number of first image feature extraction layers 311 in series, the second image feature extraction network 32 includes a second number of second image feature extraction layers 321 in series, the first image feature extraction layer 311 includes a first convolutional layer 3111, a first pooling layer 3112, and a first activation layer 3113 in series, the second image feature extraction layer 321 includes a second convolutional layer 3211, a second pooling layer 3212, a third pooling layer 3213, a second activation layer 3214, and a merging layer 3215. The second convolutional layer 3211, the second pooling layer 3212, and the second activation layer 3214 are in series. The third pooling layer 3213 is used to perform maximum pooling on the input of the second convolutional layer 3211 and then output, and the merging layer 3215 is used to merge the outputs of the second activation layer 3214 and the third pooling layer 3213.

[0070] Specifically, the first convolutional layer 3111 uses a 3×3 convolutional kernel, the first pooling layer 3112 uses 2×2 maximum pooling, and the first activation layer 3113 uses the ReLu function as the activation function.

[0071] Specifically, the second convolutional layer 3211 uses a 1×1 convolutional kernel, the second pooling layer 3212 and the third pooling layer 3213 use 2×2 maximum pooling, and the activation function of the second activation layer 3214 is:

[0072] Specifically, in this embodiment, the first quantity is 3 and the second quantity is 3.

[0073] Specifically, the merging layer 3215 directly merges the matrices output by the second activation layer 3214 and the third pooling layer 3213.

[0074] Specifically, in this embodiment, the preset dimension is 128 dimensions. It should be noted that the value of the preset dimension can be adjusted by the staff, and only the final output quantity of the feature descriptor generator needs to be adjusted synchronously.

[0075] S203: Obtain the pixel matching point pairs of the hyperspectral image and the panchromatic image according to the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image.

[0076] In the embodiments of the present disclosure, obtaining the pixel matching point pairs of the hyperspectral image and the panchromatic image according to the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image includes the following steps:

[0077] Train a mapping network for the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image.

[0078] Specifically, the mapping network is a fully connected neural network. By using the pre-calibrated and paired feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image as the training set, the fully connected neural network is trained to obtain a mapping network for the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image.

[0079] Specifically, the mapping network is a fully connected neural network composed of one input layer, four hidden layers and one output layer. The input layer of the fully connected neural network is 128 dimensions, the first hidden layer is 256 dimensions, the second hidden layer is 512 dimensions, the third hidden layer is 512 dimensions, the fourth hidden layer is 256 dimensions, and the output layer is 128 dimensions.

[0080] In this embodiment, the mapping network is used to convert the feature descriptors of the pixel points of the hyperspectral image into the type of feature descriptors of the pixel points of the panchromatic image, so as to compare the feature descriptors of the pixel points of the hyperspectral image with the feature descriptors of the pixel points of the panchromatic image.

[0081] In this embodiment, after converting the feature descriptors of the pixel points of the hyperspectral image into the type of feature descriptors of the pixel points of the panchromatic image, the feature descriptors of the pixel points of the hyperspectral image are compared with the feature descriptors of the pixel points of the panchromatic image one by one, and finally the matching point pairs between the pixel points of the hyperspectral image and the pixel points of the panchromatic image are obtained.

[0082] Specifically, in this embodiment, by calculating the distance between the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image, since the expression of the feature descriptor is actually a matrix, the distance between the feature descriptors of the pixel points of the hyperspectral image and the feature descriptors of the pixel points of the panchromatic image can be obtained by calculating the distance between the matrices. The combination of the pixel points of the hyperspectral image and the pixel points of the panchromatic image corresponding to the feature descriptors with the shortest distance is used as the matching point pair of the pixel points of the hyperspectral image and the panchromatic image.

[0083] In other embodiments, the staff can train a mapping network that converts the type of feature descriptors of the pixel points of the panchromatic image into the type of feature descriptors of the pixel points of the hyperspectral image through the same training set as in the above embodiment.

[0084] S204: Substitute the grid basis function and the pixel point matching point pair into a preset matching point pair mapping function to obtain the actual position where the pixel points of the hyperspectral image are mapped to the panchromatic image.

[0085] Specifically, the panchromatic image is divided into grids in n horizontal directions and m vertical directions to obtain the horizontal and vertical grid basis functions of the panchromatic image. Substitute the horizontal grid basis function, the vertical grid basis function, and the pixel point matching point pair of the hyperspectral image and the panchromatic image into a preset matching point pair mapping function to obtain the actual position where the pixel points of the hyperspectral image are mapped to the panchromatic image. The preset matching point pair mapping function is:

[0086]

[0087] where p(u, v) is the actual position of the pixel point of the hyperspectral image corresponding to the panchromatic image, M ip (u) is the vertical grid basis function of the panchromatic image, M jq (v) is the horizontal grid basis function of the panchromatic image, T ij is the pixel point of the hyperspectral image, S ij is the weight factor of the feature descriptor of the hyperspectral image;

[0088] where

[0089]

[0090]

[0091]

[0092]

[0093] wherein, i is the grid ordinate of the pixel point T of the panchromatic image corresponding to the hyperspectral image ij and j is the grid abscissa of the pixel point T of the panchromatic image corresponding to the hyperspectral image ij , H is the height of the panchromatic image, and W is the width of the panchromatic image.

[0094] Specifically, i and j are obtained by the pixel matching pairs of the hyperspectral image and the panchromatic image to get the pixel point T of the hyperspectral image ij corresponding to the pixel point on the panchromatic image, and then the pixel point of the panchromatic image is converted into the grid coordinates of the panchromatic image, that is, the coordinate position of the pixel point of the panchromatic image on the divided grid is found.

[0095] Specifically, the weight factor S ij is set to the constant 1. In other embodiments, when the weight factor S ij is set to the constant 0, the pixel point T of the hyperspectral image corresponding to the weight factor S ij does not participate in the fusion matching of the hyperspectral image and the panchromatic image. ij

[0096] Figure 4 is the structural block diagram of the remote sensing image processing device provided by the embodiment of the present disclosure. Referring to Figure 4 as shown, the remote sensing image processing device provided by the embodiment of the present disclosure includes a processor 41 and a memory 43. At least one instruction, at least one program, a code set or an instruction set is stored in the memory 43, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor 41 to implement the functions of the remote sensing image processing method in the above method embodiment. The functions can be implemented by hardware or by hardware executing corresponding software.

[0097] In a specific application, each component of the remote sensing image processing device is coupled together through a bus system 42. The bus system 42 may include a power bus, a control bus, a status signal bus, etc. in addition to a data bus. However, for the sake of clarity, various buses are labeled as the bus system 42 in the figure.

[0098] The method disclosed in the embodiments of the present invention above can be applied to or implemented by the processor 41. The processor 41 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit in the hardware of the processor 41 or the instructions in the form of software. The above-mentioned processor 41 may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by the combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 43, and the processor 41 reads the information in the memory 43 and combines its hardware to complete the steps of the above method.

[0099] The above remote sensing image processing device can be understood by referring to the description in the method embodiment section, and will not be elaborated here too much.

[0100] It should be noted that in this article, relational terms such as "first" and "second" 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 term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.

[0101] The above are only specific implementation manners of the present disclosure, enabling those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments described herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing image processing method, characterized in that, The remote sensing image processing method includes the following steps: Obtain the regional image with a preset size around the pixel point in the hyperspectral image or panchromatic image as the feature description of the pixel point; Generate a feature descriptor of the pixel point according to the feature description of the pixel point; Obtain the pixel matching point pairs of the hyperspectral image and the panchromatic image according to the feature descriptors of the pixel points in the hyperspectral image and the feature descriptors of the pixel points in the panchromatic image; Divide the panchromatic image into n horizontal and m vertical grids to obtain the horizontal and vertical grid basis functions of the panchromatic image, and substitute the horizontal grid basis function, the vertical grid basis function, and the pixel matching point pairs of the hyperspectral image and the panchromatic image into a preset matching point pair mapping function to obtain the actual position where the pixel point of the hyperspectral image is mapped to the panchromatic image.

2. The remote sensing image processing method according to claim 1, wherein the preset matching point pair mapping function is: Wherein, p(u, v) is the actual position in the panchromatic image corresponding to the pixel of the hyperspectral image, M ip M(u) is the grid basis function in the vertical direction of the panchromatic image, M jq M(v) is the grid basis function in the horizontal direction of the panchromatic image, T ij S is the pixel of the hyperspectral image ij is the weight factor of the feature descriptor of the hyperspectral image; Wherein, where i is the grid ordinate of the pixel point T of the panchromatic image corresponding to the hyperspectral image ij and j is the grid abscissa of the pixel point T of the panchromatic image corresponding to the hyperspectral image ij H is the height of the panchromatic image, and W is the width of the panchromatic image.

3. The remote sensing image processing method according to claim 1, characterized in that, Generating a feature descriptor of the pixel point according to the feature description of the pixel point includes the following steps: Transform the feature description of the pixel point through a preset anti-rotation and anti-illumination transformation formula to obtain a first processed image; Scale the first processed image through a pyramid scaling model with a preset scaling ratio to obtain multiple first processed images with different sizes; Input the multiple first processed images with different sizes into a feature descriptor generator to obtain a feature descriptor with a preset dimension.

4. The remote sensing image processing method according to claim 3, characterized in that, The preset anti-rotation and anti-illumination transformation formula is: Among them, Ω is the feature description of the pixel point, I(x, y) is the pixel value of the pixel with abscissa x and ordinate y in the feature description, is the pixel average value of the feature description of the pixel point, x p is the abscissa pixel value after transformation of the pixel point with abscissa x and ordinate y, y p is the ordinate pixel value after transformation of the pixel point with abscissa x and ordinate y.

5. The remote sensing image processing method according to claim 3, characterized in that, The number of layers of the pyramid scaling model is four, and the preset scaling ratio is 1.5 times.

6. The remote sensing image processing method according to claim 3, characterized in that, The feature descriptor generator includes a first image feature extraction network and a second image feature extraction network, and the first image feature extraction network and the second image feature extraction network jointly output a feature descriptor with a preset dimension; Wherein, the first image feature extraction network includes a first number of first image feature extraction layers in series, the second image feature extraction network includes a second number of second image feature extraction layers in series, the first image feature extraction layer includes a first convolutional layer, a first pooling layer, and a first activation layer in series, the second image feature extraction layer includes a second convolutional layer, a second pooling layer, a third pooling layer, a second activation layer, and a merging layer, the second convolutional layer, the second pooling layer, and the second activation layer are in series, the third pooling layer is used to perform maximum pooling on the input of the second convolutional layer and then output, and the merging layer is used to merge the outputs of the second activation layer and the third pooling layer.

7. The remote sensing image processing method according to claim 6, characterized in that, The first convolutional layer uses a 3×3 convolutional kernel, the first pooling layer uses 2×2 maximum pooling, and the first activation layer uses the ReLu function as the activation function; The second convolutional layer uses a 1×1 convolutional kernel, the second pooling layer and the third pooling layer use 2×2 maximum pooling, and the activation function of the second activation layer is:

8. The remote sensing image processing method according to claim 3, characterized in that, The preset dimension is 128 dimensions.

9. The remote sensing image processing method according to claim 1, characterized in that, Obtaining the pixel matching point pairs of the hyperspectral image and the panchromatic image according to the feature descriptors of the pixel points in the hyperspectral image and the feature descriptors of the pixel points in the panchromatic image includes the following steps: Train a mapping network for the feature descriptors of the pixel points in the hyperspectral image and the feature descriptors of the pixel points in the panchromatic image; Input the feature descriptor of the pixel point in the hyperspectral image or the feature descriptor of the pixel point in the panchromatic image into the mapping network to obtain the pixel matching point pairs of the hyperspectral image and the panchromatic image.

10. A remote sensing image processing apparatus, characterized in that, The remote sensing image processing device includes a processor and a memory. At least one instruction, at least one program, a code set, or an instruction set is stored in the memory. The at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the remote sensing image processing method according to any one of claims 1-9.

Images