Search

CN-122023358-A - Star spot dead spot detection and compensation algorithm based on Gaussian quantization geometric model

CN122023358ACN 122023358 ACN122023358 ACN 122023358ACN-122023358-A

Abstract

The invention discloses a star point detection and compensation algorithm based on a Gaussian quantization geometric model, and belongs to the field of star sensor dead point detection and compensation. The method provides a star-spot pixel gray level Gaussian quantization equal ratio model which does not depend on specific Gaussian model parameters, discloses a constraint relation of pixel gray level values in the star spots, and overcomes interference of actual background and noise on the constraint relation through background estimation and signal uncertainty evaluation. Based on the quantitative geometric model and by combining background estimation and signal uncertainty estimation, calculating gray distribution consistency residual errors of pixels, and realizing accurate detection of dead points in the star spots. On the basis, joint estimation is carried out by constructing a plurality of pixel combinations containing dead pixels, and the optimal compensation of the dead pixels is realized by adopting inverse variance weighted fusion. The invention realizes the satellite spot dead spot detection and compensation method with accurate dead spot detection, high compensation precision and low operation complexity, and is suitable for on-orbit real-time processing of the satellite sensor.

Inventors

  • WEI XINGUO
  • LI XIWANG
  • FAN QIAOYUN
  • WANG GANGYI
  • LI JIAN

Assignees

  • 北京航空航天大学

Dates

Publication Date
20260512
Application Date
20260202

Claims (7)

  1. 1. A design method of a star-spot pixel gray level Gaussian quantization geometric model is characterized by comprising the following steps of: Based on the energy distribution of the star spots The characteristics of the two-dimensional Gaussian function are met, and the effective starlight signals of pixels in the star spots are obtained Modeled as being in the pixel coverage area Internal pair of star spot energy Is expressed as: formula (1.2); using the separability of the two-dimensional gaussian function in the x and y directions, separating equation (1.2) into the product of the component related to the x-axis only and the component related to the y-axis only yields: formula (1.4); Wherein the constant is ; Is the energy gray coefficient; is Gaussian dispersion radius; The coordinates of the mass center of the star spot with sub-pixel precision; one-dimensional integral components in the x-direction and y-direction are recorded as And The following formula is shown: formula (1.5); Based on the formulas (1.4) and (1.5), four pixels forming rectangular vertices in the star are selected , , And Wherein The abscissa value representing the pixel, Representing the ordinate value of a pixel to be represented by said four pixels The combination of the four pixels is defined as a four-tuple, and the four pixels correspond to the effective starlight signals , , , Expressed as: (1.7) Based on the separation characteristic of the effective signals of the pixel starlight, a proportional constraint relation between pixels in the same quadruple in a star spot which does not depend on Gaussian model parameters is provided, namely a star spot pixel gray Gaussian quantization equal ratio model is provided as follows: Formula (1.8).
  2. 2. A method for star point image background and pixel effective star signal estimation and star signal estimation uncertainty analysis, comprising: (1) The method comprises the steps of constructing a satellite spot pixel gray scale model shown in a formula (1.3), wherein a satellite spot gray scale observation value I is formed by overlapping an effective starlight signal S, a background B and noise N: formula (1.3); (2) The distribution of the star point gray observation value I satisfies the following Gaussian distribution: Formula (1.16); Wherein, the For the overall gain of the system of the imaging device, Is the standard deviation of Gaussian noise; (3) A window covering the star point and the surrounding background thereof is arranged in the star point image by taking the star point as the center; (4) Selecting a background candidate region formed by pixels with preset line numbers at the edge of the window Calculating pixels in the region As a background estimate of the current star point : Formula (1.9); (5) Subtracting the background estimate from the gray level observer I Obtaining an unbiased estimated value of the effective starlight signal S of the star point pixel : Formula (1.17); (6) Estimating the effective starlight signal Conversion to logarithmic space, the estimator defined in logarithmic space is : Formula (1.18); (7) Analyzing the effective starlight signal log estimate Standard uncertainty of (2) : Formula (1.20).
  3. 3. A method for detecting dead pixels in a star point is characterized by adopting a model obtained by the design method of the star point pixel gray Gaussian quantization equal ratio model as claimed in claim 1, and comprising the following steps: (1) Taking the natural logarithm of both sides of the gaussian quantized geometric model, linearizing it as: Formula (1.21); (2) The definition of formula (1.18) in the method of the star image background and pixel effective star signal estimation and star signal estimation uncertainty analysis according to claim 2, recording the logarithm of the theoretical effective star signal From equation (1.21), it can be seen that theoretically, pixels of two diagonals in the same quadruple in a satellite spot The sum of the values is equal; (1.22) (3) A "gray distribution consistency residual" was introduced to describe the extent to which each patch pixel gray deviates from the gaussian quantized geometric model. The residual error Defined as two diagonal pixels in the same quad as the current pixel within a star point The absolute value of the difference of the sum of values is represented by formula (1.23): formula (1.23); when the four groups are all normal pixels, residual errors Approaching 0, otherwise, if there is residual error The larger the absolute value of (a) shows that the more obvious the pixel gray level in the quaternion deviates from the Gaussian quantization equal ratio model degree, the quaternion Is likely to contain dead pixels; (4) For each pixel to be detected Constructing N groups of pixels within a star spot Thereby obtaining N gray distribution consistency residuals of the pixels For comprehensively characterizing pixels Degree of abnormality of (1) if pixel Is a bad point, then the residual sequence Is generally larger than if the pixel Is a normal pixel, then the residual is Overall smaller, however, due to noise, the estimated amount actually involved in the calculation Uncertainty of (1.24) resulting in calculated values of the residuals of each set being inconsistent in scale and fluctuation level, lacking direct comparability, residuals of each set of four-tuple according to equation (1.24) Normalizing to obtain N pixels Normalized gray distribution consistency residual of (2) : (1.24) Wherein, the Representing the nth residual error Due to the standard deviation of (2) Is two diagonal pixels in the same quadruple in the star point A linear combination of values, according to the error propagation law, Calculated according to the formula (1.25), wherein Representing the pixel of the corresponding position of the nth quadruple calculated by the formula (1.20) An estimated uncertainty of the value; formula (1.25); (5) After N pixels to be detected are obtained Normalized residual of (2) On the basis of the above, considering that the median is insensitive to individual extreme anomaly values, relatively robust statistics can be provided, and the interference of individual anomaly residuals on the results is effectively suppressed, the normalized residuals are selected Is the median value of (1) as the pixel Final gray distribution consistency residual of (2) As shown in formula (1.26): Formula (1.26); (6) The dead pixel breaks the gray distribution constraint of the star spot area, which The value is usually larger, whereas for normal pixels The value is relatively small. Based on the method, the invention adopts the average value of the residual squares of the gray distribution consistency of all pixels in the star spots And standard deviation of Calculating self-adaptive discrimination threshold value and pixel to be detected Is of a bad point state of (2) The criterion is shown in formula (1.27): formula (1.27).
  4. 4. The method for compensating the gray level of the satellite spot dead spot, which adopts the model obtained by the design method of the Gaussian quantization equal ratio model of the pixel gray level of the satellite spot as set forth in claim 1, is characterized by comprising the following steps: (1) For already detected spot of (1) Deducing bad points based on the Gaussian quantization geometric model of the star point pixel gray scale Log of theoretical effective starlight signal Can be formed by the other three pixels in the same quad The specific relation is determined as formula (1.28): formula (1.28); (2) For detected dead pixels Constructing N groups of quadruplets in a satellite spot area to estimate dead pixels To obtain the independent estimated values of N effective starlight signals in logarithmic space ; (3) Using maximum likelihood estimation method to estimate the N independent estimation values Data fusion is carried out to calculate the dead pixel Optimal estimated value of starlight signal in logarithmic space As shown in formula (1.33): Formula (1.33); The optimal estimated value is obtained for all independent estimated values Wherein the weight of each estimator is proportional to the inverse of the variance of the estimate; (4) Superimposing the background of the star image of claim 2 with an estimated value of the background estimated value of the current star in a method for estimating the effective star signal of pixels and analyzing uncertainty of the estimated value of the star signal, wherein the final gray optimal compensation value of dead pixels in the star is shown in formula (1.34): Formula (1.34).
  5. 5. A star sensor comprising the method as claimed in claims 3 and 4, wherein the method is used for detecting and compensating dead pixels in a star point area in a star map preprocessing link so as to improve the positioning accuracy of the centroid of the star point.
  6. 6. A computer readable storage medium, on which a computer program is stored which, when being executed by a processor, implements the method according to claim 3.
  7. 7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of detecting a dead pixel as claimed in claim 3 and the method as claimed in claim 4 when executing the program.

Description

Star spot dead spot detection and compensation algorithm based on Gaussian quantization geometric model Technical Field The invention relates to a dead pixel detection and compensation algorithm, in particular to a dead pixel detection and compensation algorithm in a star point based on a Gaussian quantization equal ratio model. Background The star sensor is a high-precision attitude measuring instrument widely applied to space aircrafts, wherein the high-precision extraction of the centroid of a star spot is used as a front key link to directly determine the final attitude measuring precision of the star sensor. However, during long-term storage or in-orbit operation, the image sensor is susceptible to high-energy particle radiation damage, cosmic ray impact, single-particle breakdown and other factors, so that a large number of pixels with abnormal gray response, namely dead pixels, are generated. When a dead pixel appears in a star spot area, the abnormal gray level of the dead pixel can seriously influence the gray level distribution of the star spot, so that the centroid of the star spot is obviously deviated, and the attitude measurement precision and the operation stability of the star sensor are seriously influenced. Therefore, under the above conditions, effective detection and high-fidelity compensation of dead pixels in the star spots are realized, and the method has become a key technical problem for guaranteeing the performance of the high-precision star sensor. According to different detection mechanisms, the existing dead pixel detection method is mainly divided into a statistical method based on a neighborhood rule and a multi-frame joint sequence method. The basic principle of the statistics method based on the neighborhood rule is that in a single frame image, the deviation degree of the current pixel relative to the neighborhood statistic is used as a dead pixel judgment basis. For example, tchendjou et al （Tchendjou G. T., Simeu E. Detection, Location and Concealment of Defective Pixels in Image Sensors[J]. IEEE Transactions on Emerging Topics in Computing, 2021,9(2): 664-679） have devised three online methods of detecting dead pixels from the perspective of distance measurement, local median and local dispersion of neighboring pixels. The method has low operand and is easy to realize by hardware, however, the detection performance is sensitive to threshold selection, and false detection and omission detection are easy to occur when a highlight target exists in an image or the background is complex. In order to improve the accuracy of dead pixel detection, a multi-frame joint sequence method is adopted in some researches. The basic principle of the method is that the response of the same pixel in a multi-frame image is subjected to statistical analysis, and the pixel which continuously shows abnormality in the time dimension is judged to be a dead pixel. However, such methods generally rely on multi-scene or multi-exposure data, have poor real-time performance, and consume large amounts of storage and computing resources. In addition, wang et al （Wang W., Wei X., Li J.,et al. Noise Suppression Algorithm of Short-Wave Infrared Star Image for DaytimeStar Sensor[J]. Infrared Physics & Technology, 2017,85: 382-394） utilized gradient and gaussian weighting based one-dimensional feature descriptors in combination with a clustering method to detect and distinguish between plaque pixels and dead pixels. In a comprehensive view, the existing dead pixel detection method is suitable for dead pixel detection in a star map background, but due to the fact that obvious gray scale differences exist between pixels in a star spot and between the star spot and the background, normal star spot pixels are easily misjudged as dead pixels in a star spot area, and real dead pixels in the star spot are difficult to accurately detect. According to the difference of the dead pixel compensation mechanisms, the existing dead pixel compensation method is mainly divided into a neighborhood interpolation compensation method and a data driven learning compensation method. The basic principle of the neighborhood interpolation compensation method is to use gray statistics of neighborhood pixels to perform certain linear compensation operations. For example Tchendjou et al （Tchendjou G. T., Simeu E. Detection, Location and Concealment of Defective Pixels in Image Sensors[J]. IEEE Transactions on Emerging Topics in Computing, 2021,9(2): 664-679） replace the dead pixel gray value with a neighborhood gray median. In order to maintain the image edge and detail structure as much as possible, some methods add a direction self-adaptive strategy on the basis of the conventional linear neighborhood interpolation compensation method. For example, peng et al （Peng L., Huang Y., Wang M., et al. Defective Pixel Corrector for Line Scan and Area Scan Image Sensors[J]. IEEE Transactions on Circuits and Systems I: Regular Papers, 2025,72(7