Infrared focal plane detector nonlinear calibration and parameter calculation method based on flat field data

Abstract

The invention discloses an infrared focal plane detector nonlinear calibration and parameter calculation method based on flat field data, which comprises the following steps: collecting flat field and dark field sequences at different integration times, and calculating a group mean value and a group difference for each pixel; fitting a linear baseline by using a low-brightness area, constructing a ratio sequence, representing deviation of a mean value by using a quadratic multiplicative model, and linearizing a measurement signal of a pixel; counting the covariance of the pixel and the neighborhood offset of the pixel in each integration time, fitting to obtain a space coupling model changing along with a mean value, and adding back the variance dispersed by the neighborhood to obtain a corrected variance; photon transfer curve fitting is carried out on the corrected variance-mean value data, and pixel-by-pixel gain and readout noise are obtained; pixel response heterogeneity is calculated on the highlight unsaturation and the flat field, and the full-well charge and the dynamic range are deduced according to the saturated DN value. According to the invention, key parameters of pixel-by-pixel gain, readout noise, PRNU, full-well charge, dynamic range and the like can be stably obtained.

Description

Technical Field

[0001] This invention relates to the fields of infrared imaging and photoelectric testing technology, and in particular to a nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data. Background Technology

[0002] Infrared focal plane array detectors (IR FPAs) are widely used in astronomical observation, remote sensing payloads, and precision temperature measurement. For example, Chinese patent document CN104897282A discloses a practical thermal infrared hyperspectral imager payload system, and Chinese patent document CN116170656A discloses an infrared focal plane array for high-orbit remote sensing satellites, both of which are applied in the field of remote sensing.

[0003] To ensure the accuracy and cross-platform consistency of radiometric calibration, engineering typically requires obtaining key parameters on a pixel-by-pixel or partition-by-partition basis, including gain (e - / DN), readout noise (e - Pixel response nonuniformity (PRNU), full well charge (FWC), and dynamic range (DR) are all factors to consider. A common approach is to use the photon transfer curve (PTC) to calculate the mean-variance relationship under different integration times or illumination levels, and then derive the gain and read noise from the slope and intercept. Under ideal conditions of "linearity and pixel independence", this method is simple and effective.

[0004] However, many practical devices (especially cooled HgCdTe area arrays and infrared FPAs with CMOS readout structures) exhibit significant response nonlinearity and spatial coupling: the former manifests as nonlinearity in integral time-DN or electron-DN conversion, with significant differences between pixels; the latter originates from charge redistribution among neighboring pixels (bright fat effect, BFE) or readout mutual capacitance (IPC), which compresses single-pixel variance and increases neighborhood covariance. Taking astronomical systems as an example, such as HST / WFC3-IR and JWST / NIRCam, pixel-by-pixel nonlinearity correction and spatial coupling correction are included in the calibration process; while in engineering remote sensing / laboratory calibration, the common process still directly calculates the gain using "uncorrected PTC" or relies on dedicated devices such as point sources / knife edges for coupling characterization, resulting in complex equipment, high cost and time investment, and difficulty in promoting it on mass-produced devices.

[0005] Affected by the above factors, the traditional method using only PTC is prone to the following problems when nonlinearity and coupling exist: (1) the slope drifts with the brightness range and the gain is unstable; (2) the reading noise estimation fluctuates with the fitting range; (3) it cannot output a complete parameter map for each pixel, which limits the accuracy of digital twin / imaging simulation. Summary of the Invention

[0006] This invention discloses a nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data. Under the condition of relying solely on the flat-field sequence, it solves the problems of PTC deviation caused by response nonlinearity and variance compression caused by spatial coupling in the prior art, thereby stably obtaining key parameters such as pixel-by-pixel gain, readout noise, PRNU, full-well charge and dynamic range.

[0007] A nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data, characterized by the following steps: (1) Acquire flat field and dark field sequences at different integration times, and calculate the group mean and group variance for each pixel; (2) Fit a linear baseline to the low brightness area, then construct a ratio sequence to represent the deviation of the mean using a quadratic multiplicative model, and linearize the pixel measurement signal accordingly. (3) Calculate the covariance between the pixel and its neighborhood offset in each integration time, fit the spatial coupling model that varies with the mean, and add back the variance scattered by the neighborhood according to the spatial coupling model to obtain the corrected variance. (4) The photon transfer curve is fitted on the corrected variance-mean data to obtain the pixel-by-pixel gain and readout noise; (5) Calculate the pixel response nonuniformity on the bright unsaturated flat field, and derive the full-well charge and dynamic range based on the saturated DN value.

[0008] In step (1), flat field and dark field sequences are acquired at different integration times, and each set of sequences is acquired. Frame, in which .

[0009] In step (1), the group mean and group variance are calculated for each pixel, using the following formula: ; ; in, and They represent the first Integral Time Pixel The group mean and variance, For the first Integral time and the first Frame pixels The observed values, The number of frames in each sequence; if the original pixel value If the value is close to saturation, it will be discarded in subsequent fitting. This represents the pixel saturation value.

[0010] In step (2), a linear baseline is fitted using the low-brightness region, and the formula is: ; in, For pixels The average response value obtained by linear fitting in the low brightness region For the fitting intercept term, For linear slope terms, For the first Each integration time point; Construct a ratio sequence The deviation from the mean is characterized by a quadratic multiplicative model, as shown in the following formula: ; in, For pixels In the The average response value over several integration times. For the normalized ratio sequence, and For pixel-by-pixel nonlinear coefficients; The measurement signal of a pixel is linearized using the following formula: ; in, For measuring signals, This is the signal after nonlinear correction.

[0011] In step (3), the covariance between the pixel and its neighborhood offset is calculated in each integration time, as shown in the following formula: ; in, Represents pixels and its neighborhood Sample covariance in direction; Indicates the first At the time of integration, the first... Frame-time pixels The grayscale value, in DN; Indicated in pixels During the points time The average gray value below; Indicates the neighborhood offset; The number of frames used to calculate variance and covariance; Indicates the first At the time of integration, the first... Frame and Pixel Gray values ​​of adjacent pixels; This represents the time mean of the neighboring pixels at the integration time. The fitted spatial coupling model that varies with the mean is as follows: ; in, A statistical model representing the variation of neighborhood covariance with pixel mean. The coefficients of the linear term represent the linear correlation between neighborhoods under low signal conditions. The coefficients are quadratic terms, reflecting the variance compression or diffusion trend caused by the Bright Fat Effect (BFE) under high signal conditions.

[0012] In step (3), the variance dispersed in the neighborhood is added back according to the spatial coupling model to obtain the corrected variance, as shown in the following formula: ; in, The corrected variance, Indicates the first Integral Time Pixel The group variance; This represents the average pixel value after non-linear correction; This represents the neighborhood set used for statistics, containing four- or eight-neighbor pixels; Indicates the corrected average value Below, the neighborhood covariance value is calculated by the model.

[0013] The specific process of step (4) is as follows: In point set Perform linear regression on it to obtain ; Further calculation of gain With readout noise The formula is as follows: ; ; in, Let be the slope of the linear regression. This is the intercept for linear regression.

[0014] The specific process of step (5) is as follows: Pixel-wise averaging is obtained on a bright, unsaturated flat field. Compared to the overall average Calculate pixel response nonuniformity ; ; And by pixel saturation value Calculate pixel full-well charge and dynamic range; ; ; in, For pixels Relative response deviation, i.e., pixel response non-uniformity (PRNU); This represents the total number of pixels used for statistical purposes. The standard deviation of the response nonuniformity of the entire image; For pixels The saturation value, i.e., the digital output at saturation; For pixels The electronic gain; For pixels The full charge of the well; For pixels Readout noise; For pixels The dynamic range.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention only requires flat field data and does not rely on point sources or dedicated optical devices to complete pixel-by-pixel nonlinear correction and spatial coupling correction; 2. This invention corrects the systemic bias of classic PTC in the presence of nonlinearity / BFE by adding back the variance dispersed in the neighborhood through a covariance-mean quadratic model. The gain and read noise are more stable and it is not sensitive to the brightness range.

[0016] 3. This invention can output pixel-by-pixel parameter maps (gain, read noise, PRNU, FWC, DR), providing high-resolution input for digital twins and imaging simulation.

[0017] 4. The process of this invention is simple and uniform, easy to implement in batches, and significantly improves calibration accuracy and consistency, and has good engineering applicability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat field data according to the present invention.

[0020] Figure 2 This is a schematic diagram of the experimental apparatus according to an embodiment of the present invention.

[0021] Figure 3 As described in the embodiments of the present invention A schematic diagram of a linear baseline.

[0022] Figure 4 As described in the embodiments of the present invention And a schematic diagram of the quadratic fitting.

[0023] Figure 5 As described in the embodiments of the present invention A full-scale statistical diagram.

[0024] Figure 6 As described in the embodiments of the present invention A full-scale statistical diagram.

[0025] Figure 7 This is a schematic diagram of domain covariance modeling in an embodiment of the present invention.

[0026] Figure 8 This is a schematic diagram of the comparison of PTC before and after and the weighted linear regression in an embodiment of the present invention.

[0027] Figure 9 This is a schematic diagram of pixel gain statistics in an embodiment of the present invention.

[0028] Figure 10 This is a PRNU pixel map in an embodiment of the present invention.

[0029] Figure 11 This is a schematic diagram of the FWC distribution of the full-well charge in an embodiment of the present invention.

[0030] Figure 12 This is a schematic diagram of the dynamic range (DR) distribution in an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] It should be noted that, unless otherwise specified, the features in the following embodiments and implementation methods can be combined with each other.

[0033] like Figure 1 As shown, a nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data includes the following steps: Step S101: Data acquisition and preprocessing.

[0034] like Figure 2As shown, the experimental setup in this embodiment includes: a temperature-adjustable blackbody source, a standard optical channel, a focal plane array under test and readout electronics, a data acquisition unit, and control software. The blackbody temperature and integration time remain stable within each exposure group; to avoid saturation, an upper limit threshold is set for each integration time. A central ROI (e.g., 128×128) is preferably set for fitting and comparison. Data is collected group by group on a preset integration time schedule, and the group mean is calculated pixel by pixel. and variance Blind elements can be removed using thresholding and connected component analysis; bias differences between different channels can be estimated in the dark field and then uniformly corrected. The original data is only denoised and checked for consistency, without any sharpening or texturing to avoid affecting the statistics.

[0035] Different integration times The following sequences of flat and dark fields were acquired, with each group of acquisitions... Frame (recommended) For each pixel Calculate the group mean and group variance: ; ; in, and They represent the first Integral Time Pixel The group mean and variance, For the first Integral time and the first Frame pixels The observed values, The number of frames in each sequence; if the original pixel value If the value is close to saturation, it will be discarded in subsequent fitting. This represents the pixel saturation value.

[0036] Step S102: Nonlinear correction.

[0037] For pixels, use a linear baseline in the low-signal region: ; in, For pixels The average response value obtained by linear fitting in the low brightness region For the fitting intercept term, For linear slope terms, For the first Each integration time point.

[0038] Construct a ratio sequence and with the mean A quadratic approximation is fitted to the independent variable to characterize the nonlinearity of the pixel-wise mean: ; ; in, For pixels In the The average response value over several integration times. For the normalized ratio sequence, and For pixel-by-pixel nonlinear coefficients.

[0039] Alternatively, an equivalent approximation can be made using a pixel-by-pixel monotonic polynomial (preferably quadratic or cubic). … To facilitate subsequent processing, this step uses a multiplicative correction method to linearize the measurement signal.

[0040] ; in, For measuring signals, This is the signal after nonlinear correction.

[0041] Thus, the corrected mean is obtained. (Unit: DN; multiply by gain if necessary to convert to electron count). Where: , For pixel-by-pixel nonlinear coefficients; This represents the pixel measurement signal (DN).

[0042] This step is performed in the DN domain, avoiding additional unit conversions and numerical inverse kinematics. Typical curves and fitting residuals are shown below. Figure 3 and Figure 4 See pixel-level coefficient distribution Figure 5 and Figure 6 Specifically, Figure 3 Showing With a linear baseline; Figure 4 Showing and quadratic fitting; Figure 5 and Figure 6 They respectively showed the and Full statistics.

[0043] Step S103: Fitting the neighborhood covariance.

[0044] Pixel statistics within each integration time offset from neighborhood covariance ; in, Represents pixels and its neighborhood Sample covariance in direction; Indicates the first At the time of integration, the first... Frame-time pixels Gray value (unit: DN); Indicated in pixels During the points time The average gray value below; Represents the neighborhood offset (e.g., the four neighbors: top, bottom, left, and right). or ; The number of frames used to calculate variance and covariance; Indicates the first At the time of integration, the first... Frame and Pixel Adjacent pixels (by) (Definition) Grayscale value; This represents the time mean of the neighboring pixels at the integration time.

[0045] For each offset, bins are divided according to several average values. Fitted neighborhood covariance with respect to mean The quadratic model: ; in, A statistical model representing the variation of neighborhood covariance with pixel mean. The coefficients of the linear term represent the linear correlation between neighborhoods under low signal conditions. The coefficients are quadratic terms, reflecting the variance compression or diffusion trend caused by the Bright Fat Effect (BFE) under high signal conditions.

[0046] In this step, to correct variance compression caused by bright pixel effects, the covariance between the four-neighbor pixels and their neighbors is statistically analyzed, and then... The model fits the variation with the mean. It only "adds back the variance carried away by the neighborhood" and does not involve the estimation of point spread or optical PSF. Neighborhood definition and fitting examples are shown below. Figure 7 As shown in (a) and (b).

[0047] Step S104: PTC-corrected regression.

[0048] Based on the model in step 3, the variance "scattered" by the neighborhood is added back to the single-pixel variance to obtain the corrected variance: ; in, The corrected variance, Indicates the first Integral Time Pixel The group variance; This represents the average pixel value after non-linear correction; This represents the neighborhood set used for statistics (usually containing four or eight neighboring pixels). Indicates the corrected average value Below, the neighborhood covariance value is calculated by the model.

[0049] In point set Perform linear regression on ; Further calculation of gain With readout noise : ; ; in, Let be the slope of the linear regression. This is the intercept for linear regression.

[0050] In this step, based on the previous step The variance of each point is "added back" to obtain the result. Subsequently in ( The plane is subjected to weighted / robust linear regression. The reciprocal of the PTC slope is the pixel gain. Intercept converted to readout noise High residual points and points near saturation are automatically masked to avoid bias. The comparison before / after PTC and weighted linear regression are shown below. Figure 8 See pixel gain statistics Figure 9 .

[0051] Step S105: PRNU / Full Well / Dynamic Range.

[0052] Pixel-wise averaging is obtained on a bright, unsaturated flat field. Compared to the overall average Calculate pixel response nonuniformity: ; ; And by pixel saturation value Calculate pixel full-well charge and dynamic range: ; ; in, For pixels Relative response deviation (i.e., pixel response non-uniformity, Photo Response Non-Uniformity, PRNU); This represents the total number of pixels used for statistical purposes. The standard deviation of the response nonuniformity of the entire image; For pixels The saturation value (i.e., the digital output at saturation); For pixels Electronic gain (unit e) - / DN); For pixels Full well capacity; For pixels Readout noise (in e) - ); For pixels The dynamic range.

[0053] See example results Figure 10 , Figure 11 and Figure 12 .

[0054] Step S106: Output storage parameters.

[0055] Output in matrix or key-value pair format For use in radiation calibration and simulation.

[0056] The embodiments described above provide a detailed explanation of the technical solutions and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for nonlinear calibration and parameter calculation of an infrared focal plane detector based on flat-field data, characterized in that, Includes the following steps: (1) Acquire flat field and dark field sequences at different integration times, and calculate the group mean and group variance for each pixel; (2) Fit a linear baseline to the low brightness area, then construct a ratio sequence to represent the deviation of the mean using a quadratic multiplicative model, and linearize the pixel measurement signal accordingly. (3) Calculate the covariance between the pixel and its neighborhood offset in each integration time, fit the spatial coupling model that varies with the mean, and add back the variance scattered by the neighborhood according to the spatial coupling model to obtain the corrected variance. (4) The photon transfer curve is fitted on the corrected variance-mean data to obtain the pixel-by-pixel gain and readout noise; (5) Calculate the pixel response nonuniformity on the bright unsaturated flat field, and derive the full-well charge and dynamic range based on the saturated DN value.

2. The nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data according to claim 1, characterized in that, In step (1), flat field and dark field sequences are acquired at different integration times, and each set of sequences is acquired. Frame, in which .

3. The nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data according to claim 1, characterized in that, In step (1), the group mean and group variance are calculated for each pixel, using the following formula: ； ； in, and They represent the first Integral Time Pixel The group mean and variance, For the first Integral time and the first Frame pixels The observed values, The number of frames in each sequence; if the original pixel value If the value is close to saturation, it will be discarded in subsequent fitting. This represents the pixel saturation value.

4. The nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data according to claim 1, characterized in that, In step (2), a linear baseline is fitted using the low-brightness region, and the formula is: ； in, For pixels The average response value obtained by linear fitting in the low brightness region For the fitting intercept term, For linear slope terms, For the first Each integration time point; Construct a ratio sequence The deviation from the mean is characterized by a quadratic multiplicative model, as shown in the following formula: ； in, For pixels In the The average response value over several integration times. For the normalized ratio sequence, and For pixel-by-pixel nonlinear coefficients; The measurement signal of a pixel is linearized using the following formula: ； in, For measuring signals, This is the signal after nonlinear correction.

5. The nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data according to claim 1, characterized in that, In step (3), the covariance between the pixel and its neighborhood offset is calculated in each integration time, as shown in the following formula: ； in, Represents pixels and its neighborhood Sample covariance in direction; Indicates the first At the time of integration, the first... Frame-time pixels The grayscale value, in DN; Indicated in pixels During the points time The average gray value below; Indicates the neighborhood offset; The number of frames used to calculate variance and covariance; Indicates the first At the time of integration, the first... Frame and Pixel Gray values ​​of adjacent pixels; This represents the time mean of the neighboring pixels at the integration time. The fitted spatial coupling model that varies with the mean is as follows: ； in, A statistical model representing the variation of neighborhood covariance with pixel mean. The coefficients of the linear term represent the linear correlation between neighborhoods under low signal conditions. The coefficients of the quadratic term reflect the variance compression or diffusion trend caused by the bright fat effect under high signal conditions.

6. The nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data according to claim 5, characterized in that, In step (3), the variance dispersed in the neighborhood is added back according to the spatial coupling model to obtain the corrected variance, as shown in the following formula: ； in, The corrected variance. Indicates the first Integral Time Pixel The group variance; This represents the average pixel value after non-linear correction; This represents the neighborhood set used for statistics, containing four- or eight-neighbor pixels; Indicates the corrected average value Below, the neighborhood covariance value is calculated by the model.

7. The nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data according to claim 6, characterized in that, The specific process of step (4) is as follows: In point set Perform linear regression on it to obtain ； Further calculation of gain With readout noise The formula is as follows: ； ； in, Let be the slope of the linear regression. This is the intercept for linear regression.

8. The nonlinear calibration and parameter calculation method for an infrared focal plane detector based on flat-field data according to claim 7, characterized in that, The specific process of step (5) is as follows: Pixel-wise averaging is obtained on a bright, unsaturated flat field. Compared to the overall average Calculate pixel response nonuniformity ； ； And by pixel saturation value Calculate pixel full-well charge and dynamic range; ； ； in, For pixels Relative response deviation, i.e., pixel response non-uniformity (PRNU); This represents the total number of pixels used for statistical purposes. The standard deviation of the response nonuniformity of the entire image; For pixels The saturation value, i.e., the digital output at saturation; For pixels The electronic gain; For pixels The full charge of the well; For pixels Readout noise; For pixels The dynamic range.

Images