Image non-uniformity correction method for imaging spectrometer using uniform bright and dark areas

Abstract

This invention proposes a method for correcting image non-uniformity in an imaging spectrometer using uniform bright and dark regions. The method includes: obtaining correction coefficients in the laboratory for preliminary detector pixel calibration; selecting uniform bright and dark regions from the spectral band unaffected by banding noise after coefficient correction; applying a threshold constraint using an improved iterative method to remove abnormal signals in the uniform bright and dark regions to obtain a stable reference value; and using the reference value to calculate the gain and bias of all pixels in the spatial dimension, ultimately correcting non-uniformity across the entire image and spectral band. This method requires only a single non-uniform image that meets the requirements as input to quickly obtain good correction results, with low computational complexity and high correction speed.

Description

Technical Field

[0001] The present invention relates to the technical field of non-uniform noise removal, and in particular to a method for correcting image non-uniformity of an imaging spectrometer using uniform bright and dark areas. The method is an adaptive non-uniformity compensation method based on image push-scan characteristics, and is used to remove fixed pattern noise on a focal plane array. Background Art

[0002] Image sensing devices using single-point photoelectric sensors can produce excellent digital images because their sensitivity, resolution, and gain or transfer function are identical at all points in the image. However, the speed of such devices is limited by mechanical motion, and they cannot collect a sufficient number of photons, making it difficult to accurately measure scene radiance. Therefore, the use of linear sensors has become the mainstream spectral imaging method. However, at the current stage of technological development, sensor pixels suffer from inconsistent response to the imaging target, pixel size errors caused by process problems, and uneven optical coatings on the detector components. Infrared focal plane detectors, due to their low energy acquisition and higher sensitivity requirements, are more susceptible to striping noise in hyperspectral images during push-broom imaging. Striping noise reduces the image quality of hyperspectral images, corrupts image information, and severely impacts subsequent information extraction and target recognition. Therefore, removing striping noise is crucial.

[0003] Currently, there are three main types of methods for removing stripe noise: spectral filtering-based methods, total variation and improved algorithms, and methods based on spatial information statistics. Spectral filtering-based methods mainly include frequency domain and spatial domain joint filtering and wavelet transform filtering. However, since the frequency of the noise cannot be determined, image details will be lost and the image will be blurred during filtering, and the overall image structure will be seriously damaged. Various algorithms based on total variation construct energy functionals with constraints, but the constraint parameters are difficult to select, especially for irregular stripe noise with excessive noise, which makes the staircase effect of the image more serious. Methods based on spatial information statistics mainly include histogram matching, moment matching and improved algorithms. These methods directly use the original information in the original image. The reasonable assumption of the reference value not affected by stripe noise will directly determine the quality of denoising.

[0004] Therefore, in order to address the shortcomings of these general methods, a more suitable non-uniformity correction method is developed for push-broom imaging spectrometers with bright and dark area imaging characteristics to solve the non-uniformity correction problem of push-broom hyperspectral remote sensing images with bright and dark uniform areas. Summary of the Invention

[0005] The present invention describes a method for correcting image non-uniformity of an imaging spectrometer using uniform bright and dark areas. The technical problem to be solved by this method is to provide a method with good accuracy, convenient calculation, no signal distortion and strong spectral fidelity for push-broom imaging images with bright and dark area imaging characteristics, in the case where general methods have poor effects.

[0006] The technical solution of the present invention is a method for correcting non-uniformity of an imaging spectrometer image using uniform light and dark areas, comprising the following steps: step 1, initial correction based on laboratory non-uniformity correction coefficients; step 2, selection of preliminary uniform light and dark areas; step 3, iteratively obtaining uniform area reference values for the entire spectrum of the image; and step 4, calculating the non-uniformity correction coefficients using the reference values.

[0007] Step 1: The initial calibration process based on the laboratory non-uniformity correction coefficient is as follows: Using a uniform light source (the central area of a large-aperture integrating sphere or a black body) as the reference source, align the instrument with the reference source and record imaging data for different brightness reference sources at different integration times. Subtract the dark background from the data. Finally, the response signal value of the detection pixel in each band to the different brightness reference sources is obtained. Calculate the non-uniformity correction coefficient:

[0008]

[0009]

[0010]

[0011] in is the signal value after subtracting the dark background from the original data cube; S i，k is the signal value of a pixel in a certain band; is the standard reference value for all pixels in a band; A i，k is the non-uniformity correction coefficient: N S is the sampling number, that is, the number of pixels in each band. The least square method is used to calculate the non-uniformity correction coefficient based on the measured data. In the imaging data processing process, the obtained non-uniformity correction coefficient matrix can be used to perform non-uniformity correction on the data to be corrected. The calculation method is

[0012]

[0013] in is the pixel value after uniformity correction; I i，j，kThe pixel values of the data cube to be corrected. Step 2: Initial uniform brightness and darkness region selection requires that push-scan imaging have distinct bright and dark regions. The bright and dark regions in the image represent the overall coverage of the linear array detector's detection pixels. These bright and dark regions are defined within the range of pixel value fluctuations, not necessarily with absolutely uniform brightness and darkness. Initial uniform brightness and darkness region selection involves selecting an appropriate, unaffected spectral segment based on the actual detection situation.

[0014] Step 2 The selection process of the initial uniform light and dark area is as follows: In order to obtain the uniform area, for one track detection image, we first merge the spectral dimension, and then accumulate the row direction to obtain a column mean I im Set the window size N and the step size i to move the window. For each window C i , get N row mean data, the window center position is assumed to be W mid , and then find the mean n in each window ave and variance n var , the mean indicates the level of the value within the window range, and the smaller the variance, the more uniform the distribution of the random variable and the smaller the variability.

[0015]

[0016]

[0017] Where:

[0018] Then, for the mean N of the entire track image ave , using this mean as the dividing line, we can find the range of the evenly light and dark area:

[0019] S v1 ={max(..., C i ,…)|n ave -n var ≤n ave ≤n ave +n var , n ave >N ave}

[0020] S v2 ={max(..., C i ,…)|n ave -n var ≤n ave ≤n ave +n var , n ave <N ave}

[0021] Where S v1 With S v2Respectively expressed as ave Up and down, including window C i The most uniform area.

[0022] Step 3 Iterate the full spectrum of the image to obtain a uniform area reference value. The steps are as follows: set an initial threshold T0, divide all the pixels of the image into two categories, one category has pixel values less than or equal to T0 (background area), and the other category has pixel values greater than T0 (foreground area). Calculate the average pixel values m0 and m1 of the background area and foreground area respectively. In order to get a threshold value closer to the pixel value of the high-frequency area, let

[0023] T1 i =m0+(m1-m0)*0.9=m0*0.1+m1*0.9

[0024] Then calculate the average pixel values n0 and n1 of the background and foreground areas, so that the threshold is closer to the pixel values of the shadow and other lower frequency areas.

[0025] T2 i =n0+(n1-n0)*0.1=n0*0.9+n1*0.1

[0026] The conditions for ending two iterations are the same, even if the threshold obtained by the iteration is less than 10 of the threshold of the previous iteration. -4 The pixels between the thresholds T1 and T2 are the corresponding true background, that is, the ideal uniform area value.

[0027] Step 4: Calculate the non-uniformity correction coefficient using the reference value. The steps are as follows: Calculate the average of the uniform areas of light and dark:

[0028]

[0029] When in bright areas:

[0030] Y H =G ijk R ijk (E H )+O ijk

[0031] When in a dark area:

[0032] Y L =G ijk R ijk (E L )+O ijk

[0033] Then we get from the formula

[0034]

[0035]

[0036] Among them, R ijk (E H ) and R ijk (E L ) are the responses of pixel (i, j, k) to the background of uniform light and dark area radiation. H and Y L are the mean values under the background of uniform radiation in bright and dark areas respectively. i，j，k and O i，j，k is the gain and bias of pixel (i, j, k), which is the correction coefficient.

[0037] Finally, the uniformity-corrected image is obtained according to the formula:

[0038] D′ i，j，k =G i，j，k *D i，j，k +O i，j，k

[0039] D′ i，j，k is the pixel value after correction; D i，j，k is the original pixel value; BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart for correcting streak noise non-uniformity using uniform bright and dark areas;

[0041] Figure 2 is the image before non-uniformity correction.

[0042] Figure 3 is the image after non-uniformity correction. DETAILED DESCRIPTION

[0043] The present invention describes a method for correcting image non-uniformity of an imaging spectrometer using uniform light and dark areas. This method can be used more conveniently and accurately for specific imaging, is easy to calculate, has no signal distortion, and has high spectral fidelity. The following, combined with the accompanying drawings, provides a clear and complete description of the technical solution of the present invention through specific embodiments. Please refer to Figure 1 , is a flow chart of a method for correcting image non-uniformity of an imaging spectrometer using uniform bright and dark areas according to an embodiment of the present invention, comprising:

[0044] Step 1: initial calibration based on laboratory non-uniformity correction coefficient;

[0045] Step 2: preliminary selection of evenly lit and dark areas;

[0046] Step 3, iterate the entire spectrum of the image to obtain a uniform area reference value;

[0047] Step 4: Calculate the non-uniformity correction coefficient using the reference value.

[0048] Specifically, perform step 1. In this example, we explore the capabilities and response characteristics of each pixel of the detector itself, and obtain the detector correction coefficient to preliminarily improve the uniformity of the response of each detection unit of the sensor hardware to the entrance pupil radiation. The process is as follows: Use a uniform light source (the central area of a large-aperture integrating sphere or a black body) as the reference source, align the instrument with the reference source, and record the imaging data of different brightness reference sources at different integration times. Subtract the dark background from the data cube. Finally, the response signal value of the detection pixel of each band to the reference source of different brightness is obtained. Calculate the non-uniformity correction coefficient:

[0049]

[0050]

[0051]

[0052] in is the signal value after subtracting the dark background from the original data cube; S i，k is the signal value of a pixel in a certain band; is the standard reference value for all pixels in a band; A i，k is the non-uniformity correction coefficient: N S is the sampling number, that is, the number of pixels in each band. The least square method is used to calculate the non-uniformity correction coefficient based on the measured data. In the imaging data processing process, the obtained non-uniformity correction coefficient matrix can be used to perform non-uniformity correction on the data to be corrected. The calculation method is

[0053]

[0054] Where I^_(i, j, k) is the pixel value after uniformity correction; I_(i, j, k) is the pixel value of the data cube to be corrected.

[0055] Execute step 2. In this example, the imaging spectrum image is a spectrum image after push-scan imaging. It is required that the imaging spectrum data obtained for one track is a segment with a relatively obvious light and dark distribution when expressed in DN value or radiant brightness. The light and dark segments in the image are the overall coverage of the linear array detection pixels of the linear array detector, and the light and dark areas are areas within the fluctuation range of the pixel values, and do not represent absolute constant light and dark. Images with such characteristics are the required processing data. Select the near-infrared spectrum segment with less atmospheric aerosol influence and no band noise impression for preliminary uniform area selection. The process is as follows: In order to preliminarily obtain a uniform area, for one track detection image, first merge the spectral dimensions, and then accumulate the row direction to obtain a column of row means I imSet the window size N and move it in steps of i. For each window C i , get N row mean data, the window center position is assumed to be W mid , and then find the mean n in each window ave and variance n var , the mean indicates the level of the value within the window range, and the smaller the variance, the more uniform the distribution of the random variable and the smaller the variability.

[0056]

[0057]

[0058] Where:

[0059] Then, for the mean N of the entire track image ave , using this mean as the dividing line, we can find the range of the evenly light and dark area:

[0060] S v1 ={max(..., C i ,…)|n ave -n var ≤n ave ≤n ave +n var , n ave >N ave}

[0061] S v2 ={max(..., C i ,…)|n ave -n var ≤n ave ≤n ave +n var , n ave <N ave}

[0062] Where S v1 With S v2 Respectively expressed as ave Up and down, including window C i The most uniform area.

[0063] Execute step 3. In this example, after using near-infrared to obtain a preliminary uniform light and dark area, the position of this area can be used to iteratively threshold the full spectrum to obtain the pixels of the precise uniform area. The steps are as follows: set an initial threshold T0, divide all the pixels of the image into two categories, one category has pixel values less than or equal to T0 (background area), and the other category has pixel values greater than T0 (foreground area). Calculate the average pixel values m0 and m1 of the background area and foreground area respectively. In order to obtain a threshold value closer to the pixel value of the high-frequency area, let

[0064] T1 i =m0+(m1-m0)*0.9=m0*0.1+m1*0.9

[0065] Then calculate the average pixel values n0 and n1 of the background and foreground areas, so that the threshold is closer to the pixel values of the shadow and other lower frequency areas.

[0066] T2 i =n0+(n1-n0)*0.1=n0*0.9+n1*0.1

[0067] The conditions for ending two iterations are the same, even if the threshold obtained by the iteration is less than 10 of the threshold of the previous iteration. -4 The pixels between the thresholds T1 and T2 are the corresponding true background, that is, the ideal uniform area value.

[0068] Execute step 4. In this example, after obtaining the ideal uniform area, the sensor spectral response is linear. Therefore, the gain and offset are solved using the bright and dark uniform areas to perform non-uniformity correction. The steps are as follows: Calculate the average of the uniform areas of light and dark:

[0069]

[0070] When in bright areas:

[0071] Y H =G ijk R ijk (E H )+O ijk

[0072] When in a dark area:

[0073] Y L =G ijk R ijk (E L )+O ijk

[0074] Then we get from the formula

[0075]

[0076]

[0077] Among them, R ijk (E H ) and R ijk (E L ) are the responses of pixel (i, j, k) to the background of uniform light and dark area radiation. H and Y Lare the mean values under the background of uniform radiation in bright and dark areas respectively. i，j，k and O i，j，k are the gain and bias of pixel (i, j, k).

[0078] Finally, the uniformity-corrected image is obtained according to the formula:

[0079] D′ i，j，k =G i，j，k *D i，j，k +O i，j，k

[0080] D′ i，j，k is the pixel value after correction; D i，j，k is the original pixel value;

[0081] D′ i，j，k That is the imaging spectrum image after the non-uniformity correction we need. The method of using the bright and dark uniform area to correct the stripe noise non-uniformity of the embodiment of the present invention can quickly process the infrared non-uniform image and obtain a very good correction result. Please refer to Figure 2 and Figure 3 , Figure 2 This is the initial image, not corrected, with obvious stripes. Figure 3 The image is corrected and the streaks are almost gone.

Claims

1. A method for correcting image non-uniformity of an imaging spectrometer using uniform light and dark areas, characterized in that The following steps are involved: Step 1: Initial calibration based on laboratory non-uniformity correction coefficient; Step 2: Preliminary selection of evenly lit and dark areas; Step 3: Iterate the entire spectrum of the image to obtain a uniform area reference value; Step 4: Calculate the non-uniformity correction coefficient using the reference value; The uniform area reference value is obtained by iterating the full spectrum of the image as described in step 3. The steps are as follows: set an initial threshold T0, divide all the pixels of the background area image into two categories, one category has pixel values less than or equal to T0, and the other category has foreground area pixel values greater than T0, calculate the average pixel values m0 and m1 of the background area and foreground area respectively, and make T1 i =m0+(m1-m0)*0.9=m0*0.1+m1 * 0.9 Then calculate the average pixel values n0 and n1 of the background and foreground areas, so that the threshold is closer to the pixel values of the shadow and other lower frequency areas. T2 i =n0+(n1-n0)*0.1=n0*0.9+n1 * 0.1 The conditions for ending two iterations are the same, even if the threshold obtained by the iteration is less than 10 of the threshold of the previous iteration. -4 ; The pixels between the thresholds T1 and T2 are the corresponding true background, that is, the ideal uniform area value; The non-uniformity correction coefficient is calculated using the reference value as described in step 4. The steps are as follows: Calculate the average of the uniform areas of light and dark: When in bright areas: Y H =G ijk R ijk (E H )+O ijk When in a dark area: Y L =G ijk R ijk (E L )+O ijk Then we get from the formula: Among them, R ijk (E H ) and R ijk (E L ) are the responses of pixel (i, j, k) to the background of uniform light and dark area radiation; Y H and Y L are the mean values under the background of uniform radiation in bright and dark areas respectively; G i,j,k and O i,j,k is the gain and bias of pixel (i, j, k), which is the correction coefficient; Finally, the uniformity-corrected image is obtained according to the formula: D′ i,j,k =G i,j,k *D i,j,k +O i,j,k D′ i,j,k is the pixel value after correction; D i,j,k is the original pixel value.

2. The method for correcting image non-uniformity of an imaging spectrometer using uniform light and dark areas according to claim 1, wherein: The initial calibration process based on the laboratory non-uniformity correction coefficient described in step 1 is as follows: using a uniform light source as the reference source, aligning the instrument with the reference source, and recording imaging data for reference sources of different brightness at different integration times; subtracting the dark background from the data; finally, obtaining the response signal value of the detection pixel in each band to the reference source of different brightness; and calculating the non-uniformity correction coefficient: in is the signal value after subtracting the dark background from the original data cube; S i,k is the signal value of a pixel in a certain band; is the standard reference value for all pixels in a band; A i,k is the non-uniformity correction coefficient: N S is the sampling number, that is, the number of pixels in each band; the non-uniformity correction coefficient is calculated based on the measured data using the least squares method; in the imaging data processing process, the obtained non-uniformity correction coefficient matrix is used to perform non-uniformity correction on the data to be corrected; the calculation method is in is the pixel value after uniformity correction; I i,j,k are the pixel values of the data cube to be corrected.

3. The method for correcting image non-uniformity of an imaging spectrometer using uniform light and dark areas according to claim 1, wherein: The selection of the preliminary uniform light and dark areas described in step 2 requires that there are obvious light and dark areas after push-scan imaging; the light and dark areas in the image are the overall coverage of the linear array detection pixels of the linear array detector, and the light and dark areas are within the range of pixel value fluctuation, not absolutely uniform light and dark.

4. The method for correcting image non-uniformity of an imaging spectrometer using uniform light and dark areas according to claim 1, wherein: The selection of the preliminary uniform light and dark area in step 2 is to select a suitable spectrum segment that is not affected by stripe noise according to the actual detection situation.

5. The method for correcting image non-uniformity of an imaging spectrometer using uniform light and dark areas according to claim 1, wherein: The selection process of the preliminary uniform light and dark area in step 2 is as follows: In order to obtain the uniform area initially, for one track of detection image, the spectral dimension is first merged, and then the row direction is accumulated to obtain a column mean I im ; Set the window size N, set the step size to i to move the window; for each window C i , get N row mean data, the window center position is assumed to be W mid , and then find the mean n in each window ave and variance n var , the mean indicates the level of the value within the window range, and the smaller the variance, the more uniform the distribution of the random variable and the smaller the variability; that is, Where: Then, for the mean N of the entire track image ave , using this mean as the dividing line, we can find the range of the evenly light and dark area: S v1 ={max(…,C i ,…)|n ave -n var ≤n ave ≤n ave +n var ,n ave >N ave } S v2 ={max(…,C i ,…)|n ave -n vavr ≤n ave ≤n ave +n vavr ,n ave <N ave } Where S v1 With S v2 Respectively expressed as ave Up and down, including window C i Most uniform areas.

Images