Remote Sensing Image Preprocessing Methods

By processing orthogonal correction of remote sensing images in parallel on the GPU and Gram-Schmidt fusion algorithm, the problem of slow preprocessing of remote sensing images in the prior art is solved, and the ability to quickly process high-resolution images is realized.

CN111383158BActive Publication Date: 2024-11-01ZHONGKE XINGTU SPACE TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN201811613709.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2018-12-27
Publication Date
2024-11-01
Estimated Expiration
2038-12-27

AI Technical Summary

Technical Problem

The existing remote sensing image preprocessing methods are limited by the number of CPU cores and cannot quickly process high-resolution remote sensing images, especially when the data volume reaches 10GB level, which cannot meet the requirements of quickly processing multi-scape images.

Method used

Using GPU for parallel processing, through CUDA's general parallel computing architecture, orthogonal correction and Gram-Schmidt fusion algorithm are improved, image preprocessing is performed in parallel on thousands of threads, calculation methods and intermediate result preservation strategies are optimized, and data is reduced in memory and Exchange between system memory.

Benefits of technology

It greatly improves the speed of remote sensing image processing, meets the requirements of quickly processing multi-view images, and reduces the usage of GPU video memory, reducing the additional time consumption caused by data transmission.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111383158B_ABST
    Figure CN111383158B_ABST
Patent Text Reader

Abstract

The present invention provides a remote sensing image preprocessing method, comprising: a step of orthorectifying an input remote sensing image and a step of performing Gram-Schmidt fusion on the orthorectified image, wherein for each band of the image, multiple computing threads are executed in parallel on a GPU, and the image includes a single-scene low-resolution multispectral image MSS and a high-resolution single-band image PAN. The present invention is based on the CUDA (Compute Unified Device Architecture) general parallel computing architecture, utilizes the parallel processing capability of the GPU to preprocess the remote sensing image, and improves the orthorectification and Gram-Schmidt fusion algorithms in the processing, thereby realizing parallel processing of remote sensing images on thousands of threads on the GPU, greatly improving the processing speed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of remote sensing technology, and in particular to a remote sensing image preprocessing method. Background Art

[0002] Image preprocessing is the first and most important step in remote sensing applications. It generally includes several links such as geometric (ortho) correction, image fusion, image mosaicking, and atmospheric correction. Among them, the low-resolution multispectral image MSS (usually 4 bands of red, green, blue and near-infrared) and the high-resolution single-band image PAN are orthorectified based on geometric correction and digital elevation model respectively, and then resampled and fused to generate a high-resolution multispectral image (referred to as "ortho + fusion") is one of the most basic and commonly used processing procedures in image preprocessing. As the spatial resolution of remote sensing images is improved to the meter level or even sub-meter level, the data volume of the multispectral image after single-view fusion has reached the order of 10GB. The existing method is to process it through the CPU. Due to the limitation of the number of cores of a single machine processor (desktop computers or general graphics workstations have a maximum of several cores, high-end workstations have a maximum of more than ten cores, and high-end multi-way servers have a maximum of only dozens of cores), when faced with such an order of magnitude, the preprocessing time increases significantly and cannot meet the requirements of fast processing of multi-view images. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art or related art, and provide a solution that can greatly improve the remote sensing image processing speed.

[0004] To this end, according to a first aspect of the present invention, a remote sensing image preprocessing method is provided, comprising:

[0005] The step of orthorectifying the input image, and

[0006] Steps for Gram-Schmidt fusion of orthorectified images,

[0007] In each step, a plurality of computing threads are executed in parallel on an image processor GPU, and the image includes a single-scene low-resolution multispectral image MSS and a high-resolution single-band image PAN.

[0008] Furthermore, the orthorectification step includes:

[0009] For each band of the image, threads for orthorectification polynomial calculation are executed in parallel on the GPU according to the number of horizontal and vertical pixels of the output file;

[0010] The steps of Gram-Schmidt fusion include:

[0011] According to the number of horizontal and vertical pixels of the orthorectified PAN, the threads for calculating the simulated single-band image SPAN, the threads for calculating APAN, the threads for calculating the band covariance and the threads for fusion calculation are executed in parallel on the GPU in sequence. The SPAN is a simulated single-band image generated by weighted averaging the MSS according to the weight values ​​of each band, and the APAN is an image after brightness correction is performed on each pixel in the PAN according to the mean and standard deviation of the orthorectified PAN and SPAN.

[0012] Furthermore, the orthorectification step includes:

[0013] a. Read the image and the corresponding rational polynomial parameters RPC;

[0014] b. Calculate the latitude and longitude range of the image, the horizontal and vertical pixel numbers and the geographic conversion coefficient of the output file according to the horizontal and vertical pixel numbers, the geographic conversion coefficient and the RPC parameters of the image;

[0015] c. Read the digital elevation model DEM file and copy the DEM data within the latitude and longitude range of the image to the video memory;

[0016] d. For each band of the image, multiple threads of orthorectification polynomial calculation are executed in parallel on the GPU according to the horizontal and vertical pixel numbers of the output file, and each thread calculates the brightness value of the corresponding pixel in the band.

[0017] Furthermore, for each pixel in the band, step d includes:

[0018] d1. Calculate the longitude and latitude corresponding to the pixel according to the geographic conversion coefficient;

[0019] d2, calculating the elevation corresponding to the pixel from the DEM data by bilinear interpolation;

[0020] d3. Calculate the pixel coordinates of the pixel corresponding to the input image using a polynomial using the RPC parameters;

[0021] d4. Calculate the brightness value of the pixel through bilinear interpolation.

[0022] Furthermore, in the step a, the image data and RPC parameters are copied to a video memory; and / or in the step b, the geographic conversion coefficients are copied to a video memory.

[0023] Furthermore, the step of Gram-Schmidt fusion includes:

[0024] e. Calculate the polynomial coefficients of pixel coordinate conversion between MSS and PAN through the geographic conversion coefficients of the orthorectified MSS and PAN;

[0025] f. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads for calculating SPAN are executed in parallel on the GPU. Each thread calculates the pixel value of the pixel corresponding to SPAN. The pixel values ​​of PAN and SPAN and the sum of their square values ​​are calculated through reduction operations and atomic operations to obtain the mean and standard deviation of PAN and SPAN.

[0026] g. According to the number of horizontal and vertical pixels of the orthorectified PAN, multiple threads for calculating APAN are executed in parallel on the GPU, and each thread calculates the pixel value of the image APAN after the corresponding pixel is corrected according to the mean and standard deviation of the orthorectified PAN and SPAN;

[0027] h. For each band of the image, according to the number of horizontal and vertical pixels of the PAN after orthorectification, multiple threads for calculating band covariance and multiple threads for fusion calculation are executed in parallel on the GPU in sequence to output the fusion result of the band.

[0028] Further, the step h comprises:

[0029] h1. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads for calculating band covariance are executed in parallel on the GPU. Each thread calculates the covariance of the corresponding pixel SPAN and APAN, and calculates the sum of the covariance through reduction operations and atomic operations to obtain the conversion coefficient of SPAN and APAN.

[0030] h2. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads of fusion calculation are executed in parallel on the GPU, and each thread calculates the final fusion result of the pixel. The fusion result is calculated based on the gray value of the corresponding MSS after bilinear interpolation, the gray value of the pixel SPAN and APAN, and the conversion coefficient;

[0031] h3. Output the band fusion result.

[0032] Further, in step f, the pixel value of the SPAN is not retained in the video memory.

[0033] Furthermore, the multiple threads are There are thread blocks, each of which contains 16*16 threads. X is the number of horizontal pixels of the orthorectified PAN, and Y is the number of vertical pixels of the orthorectified PAN.

[0034] The present invention is based on the CUDA (Compute Unified Device Architecture) general parallel computing architecture, utilizes the parallel processing capability of the GPU to pre-process remote sensing images, and improves the orthorectification and Gram-Schmidt fusion algorithms in the processing, thereby realizing parallel processing of remote sensing images on thousands of threads on the GPU, greatly improving the processing speed; in addition, by optimizing the calculation method and the intermediate calculation result storage strategy, the peak value of the GPU video memory usage is reduced as much as possible, avoiding multiple exchanges of data between the GPU video memory and the system memory, and reducing the extra time consumption caused by data transmission.

[0035] Additional aspects and advantages of the present invention will be given in part in the following description, and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0037] Figure 1 is a flow chart of a method according to the present invention.

[0038] Figure 2 is a flow chart of a method according to one embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0040] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0041] Existing preprocessing methods mainly use multiple (core) CPUs of servers / workstations to perform parallel calculations of polynomials and interpolations. Limited by the number of CPU cores, the number of parallel threads is at most dozens. When the spatial resolution of remote sensing images is improved and the amount of single-view data increases to 10GB, the processing time increases significantly, and it cannot meet the requirements of fast processing of multi-view images. However, the number of parallel processing of high-end graphics cards (GPUs, i.e. image processors) can reach thousands. For polynomials and interpolation operations that account for the vast majority of calculations in remote sensing image preprocessing algorithms, GPUs can be used for parallel calculations because the data type and calculation method are fixed.

[0042] refer to Figure 1 , which shows a remote sensing image preprocessing method proposed by the present invention, comprising:

[0043] S11, performing orthorectification on the input image;

[0044] S12, performing Gram-Schmidt fusion on the orthorectified image;

[0045] In each step, a plurality of computing threads are executed in parallel on an image processor GPU, and the image includes a single-scene low-resolution multispectral image MSS and a high-resolution single-band image PAN.

[0046] Furthermore, the orthorectification step S11 includes:

[0047] For each band of the image, threads for orthorectification polynomial calculation are executed in parallel on the GPU according to the number of horizontal and vertical pixels of the output file;

[0048] In the step S12 of the Gram-Schmidt fusion, it includes:

[0049] According to the number of horizontal and vertical pixels of the orthorectified PAN, the threads for calculating the simulated single-band image SPAN, the threads for calculating APAN, the threads for calculating the band covariance and the threads for fusion calculation are executed in parallel on the GPU in sequence. The SPAN is a simulated single-band image generated by weighted averaging the MSS according to the weight values ​​of each band, and the APAN is an image after brightness correction is performed on each pixel in the PAN according to the mean and standard deviation of the orthorectified PAN and SPAN.

[0050] Furthermore, the orthorectification step includes:

[0051] a. reading the image and the corresponding rational polynomial parameters RPC, and further copying the image data and RPC parameters to a video memory;

[0052] b. Calculating the latitude and longitude range of the image, the horizontal and vertical pixel numbers of the output file and the geographic conversion coefficient according to the horizontal and vertical pixel numbers of the image, the geographic conversion coefficient and the RPC parameters, and further copying the geographic conversion coefficient to the video memory;

[0053] c. Read the digital elevation model DEM file and copy the DEM data within the latitude and longitude range of the image to the video memory;

[0054] d. For each band of the image, multiple threads of orthorectification polynomial calculation are executed in parallel on the GPU according to the horizontal and vertical pixel numbers of the output file, and each thread calculates the brightness value of the corresponding pixel in the band.

[0055] For each pixel in the band, step d comprises:

[0056] d1. Calculate the longitude and latitude corresponding to the pixel according to the geographic conversion coefficient;

[0057] d2, calculating the elevation corresponding to the pixel from the DEM data by bilinear interpolation;

[0058] d3. Calculate the pixel coordinates of the pixel corresponding to the input image using a polynomial using the RPC parameters;

[0059] d4. Calculate the brightness value of the pixel through bilinear interpolation.

[0060] Furthermore, the step of Gram-Schmidt fusion includes:

[0061] e. Calculate the polynomial coefficients of pixel coordinate conversion between MSS and PAN through the geographic conversion coefficients of the orthorectified MSS and PAN;

[0062] f. According to the number of horizontal and vertical pixels of the orthorectified PAN, multiple threads for calculating SPAN are executed in parallel on the GPU, each thread calculates the pixel value of the pixel corresponding to the SPAN, and the sum of the pixel values ​​of PAN and SPAN and their square values ​​are calculated through reduction operations and atomic operations to obtain the mean and standard deviation of PAN and SPAN. Furthermore, the pixel value of the SPAN is not retained in the video memory;

[0063] g. According to the number of horizontal and vertical pixels of the orthorectified PAN, multiple threads for calculating APAN are executed in parallel on the GPU, and each thread calculates the pixel value of the image APAN after the corresponding pixel is corrected according to the mean and standard deviation of the orthorectified PAN and SPAN;

[0064] h. For each band of the image, according to the number of horizontal and vertical pixels of the PAN after orthorectification, multiple threads for calculating band covariance and multiple threads for fusion calculation are executed in parallel on the GPU in sequence to output the fusion result of the band.

[0065] The step h comprises:

[0066] h1. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads for calculating band covariance are executed in parallel on the GPU. Each thread calculates the covariance of the corresponding pixel SPAN and APAN, and calculates the sum of the covariance through reduction operations and atomic operations to obtain the conversion coefficient of SPAN and APAN.

[0067] h2. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads of fusion calculation are executed in parallel on the GPU, and each thread calculates the final fusion result of the pixel. The fusion result is calculated based on the gray value of the corresponding MSS after bilinear interpolation, the gray value of the pixel SPAN and APAN, and the conversion coefficient;

[0068] h3. Output the band fusion result.

[0069] Furthermore, the multiple threads described in the method are There are thread blocks, each of which contains 16*16 threads. X is the number of horizontal pixels of the orthorectified PAN, and Y is the number of vertical pixels of the orthorectified PAN.

[0070] Therefore, using multiple threads in the GPU to perform parallel calculations on polynomials and interpolation operations with fixed data types and calculation methods in preprocessing can greatly improve the processing speed, reduce the processing time, and meet the requirements of fast processing of multi-view images.

[0071] refer to Figure 2 , which further illustrates an embodiment of the remote sensing image preprocessing method of the present invention, wherein the input remote sensing image includes a single-scene low-resolution multispectral image MSS (taking 4 bands as an example) and a high-resolution single-band image PAN. In this embodiment, the orthorectification steps are performed on the MSS and PAN in sequence. Optionally, the orthorectification steps can also be performed on the MSS and PAN in parallel.

[0072] MSS orthorectification steps:

[0073] S211, read the input MSS and the corresponding rational polynomial parameters RPC (Rational Polynomial Coefficients), and copy the MSS data and RPC parameters to the video memory;

[0074] S212, calculating the latitude and longitude range of the image, the horizontal and vertical pixel numbers and the geographic conversion coefficient of the output file according to the horizontal and vertical pixel numbers of the MSS, the geographic conversion coefficient and the RPC parameters, and copying the geographic conversion coefficient to the video memory;

[0075] S213, reading the digital elevation model DEM file, and copying the DEM data within the latitude and longitude range of the image to the video memory;

[0076] S214, sequentially executing orthorectification polynomial calculation threads of bands 1-4 on the GPU.

[0077] For each band of the image, multiple threads of orthorectification polynomial calculation are executed in parallel on the GPU according to the horizontal and vertical pixel numbers X and Y of the output file, with a total of There are 16*16 threads in each thread block. Each thread calculates the brightness value of the corresponding pixel in this band.

[0078] For each pixel in the band, step S214 includes:

[0079] Calculate the longitude and latitude corresponding to the pixel according to the geographic conversion coefficient;

[0080] Calculate the elevation corresponding to the pixel from the DEM data by bilinear interpolation;

[0081] The pixel coordinates of the pixel corresponding to the input image are calculated by a polynomial using the RPC parameters;

[0082] The brightness value of the pixel is calculated by bilinear interpolation.

[0083] PAN orthorectification steps:

[0084] S221, read the input PAN and the corresponding rational polynomial parameter RPC, and copy the MSS data and RPC parameter to the video memory;

[0085] S222, calculating the latitude and longitude range of the image, the horizontal and vertical pixel numbers and the geographic conversion coefficient of the output file according to the horizontal and vertical pixel numbers of the PAN, the geographic conversion coefficient and the RPC parameters, and copying the geographic conversion coefficient to the video memory;

[0086] S223, read the digital elevation model DEM file, and copy the DEM data within the latitude and longitude range of the image to the video memory;

[0087] S224, executing orthorectification polynomial calculation threads on the GPU. According to the horizontal and vertical pixel numbers X and Y of the output file, multiple threads of orthorectification polynomial calculation are executed in parallel on the GPU, with a total of There are 16*16 threads in each thread block. Each thread calculates the brightness value of the corresponding pixel in this band.

[0088] Since PAN is a single-band image, step S224 is a calculation step for one band. The specific calculation method is the same as step S214 and will not be repeated here.

[0089] Gram-Schmidt fusion steps:

[0090] S231, calculating the polynomial coefficients of pixel coordinate conversion between the MSS and PAN through the geographic conversion coefficients of the orthorectified MSS and PAN;

[0091] S232, according to the horizontal and vertical pixel numbers X and Y of the PAN after orthorectification, multiple threads of calculating SPAN are executed in parallel on the GPU, with a total of Thread blocks, each containing 16*16 threads. Each thread calculates the pixel value of the corresponding pixel of SPAN, and calculates the sum of the pixel values ​​of PAN and SPAN and their square values ​​through reduction and atomic operations to obtain the mean and standard deviation of PAN and SPAN; since the calculation of the corresponding pixel of SPAN is only a simple bilinear interpolation and weighted accumulation, the calculation takes less time. In order to reduce the usage of video memory, the data of SPAN is not retained in the video memory, and needs to be recalculated when needed in subsequent steps;

[0092] S233, according to the horizontal and vertical pixel numbers X and Y of the PAN after orthorectification, multiple threads of calculating APAN are executed in parallel on the GPU, with a total of Thread blocks, each containing 16*16 threads. Each thread calculates the pixel value of the image APAN after the corresponding pixel is corrected according to the mean and standard deviation of the orthorectified PAN and SPAN;

[0093] S234. For each band of the image, according to the number of horizontal and vertical pixels of the PAN after orthorectification, multiple threads for calculating band covariance and multiple threads for fusion calculation are executed in parallel on the GPU in sequence, and the band fusion result is output.

[0094] For each band of the image, step S234 includes:

[0095] According to the horizontal and vertical pixel numbers X and Y of the orthorectified PAN, multiple threads for calculating band covariance are executed in parallel on the GPU, with a total of Thread blocks, each containing 16*16 threads. Each thread calculates the covariance of the corresponding pixel SPAN and APAN, calculates the sum of the covariances through reduction operations and atomic operations, and obtains the conversion coefficients of SPAN and APAN;

[0096] According to the horizontal and vertical pixel numbers X and Y of the orthorectified PAN, multiple threads of fusion calculation are executed in parallel on the GPU, with a total of Thread blocks, each thread block contains 16*16 threads. Each thread calculates the final fusion result of the pixel, and the fusion result is calculated based on the gray value of the corresponding MSS after bilinear interpolation, and the gray value and conversion coefficient of the pixel SPAN and APAN are calculated according to the polynomial;

[0097] Output the band fusion result.

[0098] Since the number of horizontal and vertical pixels X, Y of the output file in the orthorectification step is the same as the number of horizontal and vertical pixels X, Y of the PAN after orthorectification, each multi-threaded calculation in the preprocessing process includes thread blocks, each thread block contains 16*16 threads.

[0099] In this embodiment, after the fusion results of the four bands are outputted in sequence, the preprocessing process of the remote sensing image is terminated.

[0100] The present invention is based on the CUDA (Compute Unified Device Architecture) general parallel computing architecture, uses the parallel processing capability of the GPU to pre-process remote sensing images, and improves the orthorectification and Gram-Schmidt fusion algorithms in the processing, thereby realizing parallel processing of remote sensing images on thousands of threads on the GPU, and greatly improving the processing speed; further, by keeping intermediate data with a large amount of calculation in the video memory, it does not exchange with the system memory before the final fusion output, thereby reducing the transmission consumption time; intermediate data with a low amount of calculation is selected not to be saved, and is recalculated when needed later, thereby reducing the total peak usage of the video memory, so that the present algorithm is suitable for processing satellite images of larger sizes.

[0101] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For a person skilled in the art, the present invention can have various changes and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A remote sensing image preprocessing method, It is characterized in that include: The step of orthorectifying the input image, wherein: a. Read the image and the corresponding rational polynomial parameters RPC, and copy the image data and RPC parameters to the video memory; b. Calculate the latitude and longitude range of the image, the horizontal and vertical pixel numbers of the output file and the geographic conversion coefficient according to the horizontal and vertical pixel numbers of the image, the geographic conversion coefficient and the RPC parameters, and copy the geographic conversion coefficient to the video memory; c. Read the digital elevation model DEM file and copy the DEM data within the latitude and longitude range of the image to the video memory; d. For each band of the image, multiple threads of orthorectification polynomial calculation are executed in parallel on the GPU according to the number of horizontal and vertical pixels of the output file, and each thread calculates the brightness value of the corresponding pixel in the band; In the step of performing Gram-Schmidt fusion on the orthorectified image, the intermediate data with large computational complexity is kept in the video memory and is not exchanged with the system memory before the final fusion output; the intermediate data with low computational complexity is not saved and will be recalculated when needed later. The Gram-Schmidt fusion refers to executing in parallel on the GPU the threads for calculating the simulated single-band image SPAN, the threads for calculating APAN, the threads for calculating the band covariance, and the threads for fusion calculation according to the horizontal and vertical pixel numbers of the high-resolution single-band image PAN after orthorectification, wherein: The SPAN is a simulated single-band image generated by weighting and averaging the MSS according to the weight values ​​of each band, and the APAN is an image after brightness correction is performed on each pixel in the PAN according to the mean and standard deviation of the orthorectified PAN and SPAN; The specific steps include: e. Calculate the polynomial coefficients of pixel coordinate conversion between MSS and PAN through the geographic conversion coefficients of the orthorectified MSS and PAN; f. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads for calculating SPAN are executed in parallel on the GPU. Each thread calculates the pixel value of the pixel corresponding to SPAN. The pixel values ​​of PAN and SPAN and the sum of their square values ​​are calculated through reduction operations and atomic operations to obtain the mean and standard deviation of PAN and SPAN. g. According to the number of horizontal and vertical pixels of the orthorectified PAN, multiple threads for calculating APAN are executed in parallel on the GPU, and each thread calculates the pixel value of the image APAN after the corresponding pixel is corrected according to the mean and standard deviation of the orthorectified PAN and SPAN; h. For each band of the image, according to the number of horizontal and vertical pixels of the PAN after orthorectification, multiple threads for calculating band covariance and multiple threads for fusion calculation are sequentially executed in parallel on the GPU, and the band fusion result is output, wherein the step h includes: h1. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads for calculating band covariance are executed in parallel on the GPU. Each thread calculates the covariance of the corresponding pixel SPAN and APAN, and calculates the sum of the covariance through reduction operations and atomic operations to obtain the conversion coefficient of SPAN and APAN. h2. According to the number of horizontal and vertical pixels of PAN after orthorectification, multiple threads of fusion calculation are executed in parallel on the GPU, and each thread calculates the final fusion result of the pixel. The fusion result is calculated based on the gray value of the corresponding MSS after bilinear interpolation, the gray value of the pixel SPAN and APAN, and the conversion coefficient; h3, output the band fusion result; The images include a single-view low-resolution multispectral image MSS and a high-resolution single-band image PAN.

2. The method according to claim 1, It is characterized in that For each pixel in the band, step d comprises: d1. Calculate the longitude and latitude corresponding to the pixel according to the geographic conversion coefficient; d2, calculating the elevation corresponding to the pixel from the DEM data by bilinear interpolation; d3. Calculate the pixel coordinates of the pixel corresponding to the input image using a polynomial using the RPC parameters; d4. Calculate the brightness value of the pixel through bilinear interpolation.

3. The method according to claim 1, It is characterized in that In step f, the pixel values ​​of the SPAN are not retained in the video memory.

4. The method according to any one of claims 1 to 3, It is characterized in that The multiple threads are There are thread blocks, each of which contains 16*16 threads. X is the number of horizontal pixels of the orthorectified PAN, and Y is the number of vertical pixels of the orthorectified PAN.

Citation Information

Patent Citations

  • Method for correcting distortion in real time based on GPU camera and video camera

    CN101572828A

  • Automatic and moderate orthographic projection correction method of satellite remote sensing image

    CN101750606A

  • Method for realizing fusion and enhancement of remote sensing image

    CN102063710A

  • Remote-sensing image ortho-rectification method based on central processing unit (CPU) and graphics processing unit (GPU) coprocessing

    CN103020620A

  • Method for processing remote sensing image by adopting parallel pipeline

    CN107526573A