A method for detecting bad pixels of an infrared detector

Abstract

The application relates to the field of infrared detection technology and discloses an infrared detector dead point detection method, which comprises the following steps: acquiring a current scene image during the working period of an infrared detector and multiple frames of isothermal uniform surface background images collected by the infrared detector under different working temperatures to form a background image sequence; adopting an improved wavelet transform algorithm to respectively perform singular point detection on the current scene image and each frame of image in the background image sequence, screening out effective singular points and calculating the singular values of the effective singular points; based on the singular point detection result, the singular value difference and historical background image data, distinguishing static dead points, dynamic dead points and noise through multiple layer judgment rules to obtain a dead point detection result; regularly evaluating the accuracy of the dead point detection result and dynamically calibrating the detection parameters according to the evaluation result. Through the cooperative analysis of the current working image and the multiple temperature background image sequence, the application can accurately detect dead points in real time during the normal working process of the detector.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of infrared detection technology, and particularly relates to a method for detecting bad points of an infrared detector. BACKGROUND

[0002] An infrared detector is a core sensor of an infrared imaging system, and its performance directly affects the imaging quality. Due to manufacturing process defects, material non-uniformity and long-term work in a complex temperature change environment, bad points will be generated on a focal plane array of the detector. The bad points are divided into two categories: static bad points (fixed position, not changing with time and temperature) and dynamic bad points (intermittently appearing or disappearing with the working temperature of the detector). The bad points are shown as abnormal bright / dark pixels on an image, which seriously reduces the signal-to-noise ratio and usability of the image.

[0003] The current mainstream bad point detection methods mainly have the following deficiencies:

[0004] 1. Low detection accuracy: Most methods rely on spatial domain filtering (such as median filtering) to pre-de-noise, but the filtering process is easy to smooth the real bad points (especially weak response bad points), leading to missed detection; at the same time, strong random noise may be misjudged as a bad point, leading to false alarm.

[0005] 2. Unable to effectively detect dynamic bad points: Traditional methods mostly calibrate bad points at a single temperature, and cannot capture dynamic bad points changing with temperature. Dynamic bad points widely exist in actual application, and are an important factor leading to unstable imaging quality.

[0006] 3. Dependent on special test environment: High-precision bad point detection usually needs to be carried out in a laboratory environment, with the help of a large-area blackbody source for uniform irradiation calibration. This method cannot be integrated into the normal working process of the detector, and cannot realize real-time and online bad point monitoring and compensation.

[0007] 4. Detection and compensation are disconnected: Existing methods mostly focus on the detection link, and use a single interpolation compensation strategy for the detected bad points, without considering the differentiated needs of bad point types (static / dynamic) and image regions (edge / non-edge) for compensation effect, leading to possible introduction of new image distortion after compensation, especially in the edge region.

[0008] Therefore, there is an urgent need for an infrared detector bad point detection method which can online, real-time, high-precision distinguish static / dynamic bad points and noise, and can adaptively perform differentiated compensation. SUMMARY

[0009] The present application aims to overcome the defects of the prior art, and provides an infrared detector bad point detection method. The method can realize the acquisition of multi-temperature background images by using the shutter of the detector itself without relying on external black body calibration environment, and can accurately distinguish static bad points, dynamic bad points and random noise by combining improved wavelet transform and multi-layer decision logic. Further, the present application integrates adaptive parameter calibration and targeted bad point compensation strategy, realizes self-optimization of detection accuracy and intelligent improvement of image quality.

[0010] The present application is realized by the following technical solutions:

[0011] An infrared detector bad point detection method, comprising:

[0012] S1, acquiring a current scene image A during the working period of an infrared detector, and a plurality of frames of isothermal uniform surface background images collected by the infrared detector at different working temperatures to form a background image sequence , wherein i is the collection number and ;

[0013] S2, performing singular point detection on the current scene image A and each frame of image in the background image sequence respectively by using an improved wavelet transform algorithm, screening out effective singular points and calculating the singularity values of each effective singular point; the improved wavelet transform algorithm comprises adaptive selection of wavelet base, two-dimensional wavelet transform and extraction of high-frequency detail components, adaptive threshold setting and singular point screening, and singularity value calculation;

[0014] S3, based on the singular point detection results, singularity value differences and historical background image data of corresponding pixel positions in the current scene image A and the background image sequence , distinguishing static bad points, dynamic bad points and noise by using multi-layer decision rules to obtain a bad point detection result containing bad point types, positions and corresponding temperature parameters;

[0015] S4, periodically evaluating the accuracy of the bad point detection result, dynamically calibrating the detection parameters according to the evaluation result, and feeding back the bad point detection result to a detector control unit.

[0016] As an optimization, the specific process of S1 is:

[0017] S1.1, acquiring the working scene image of the detector according to a preset frame rate, and forming the current scene image A and storing after pixel gray value calibration and image size normalization preprocessing;

[0018] S1.2, triggering the shutter to close by a linkage control mechanism when the working temperature of the detector changes by a preset temperature step, and collecting a single frame of background image and pre-processed, and stored in a background image database after associating a temperature parameter to form the sequence of background images The background image database adopts a cyclic covering mechanism, and retains the latest N frames of background images, N being a preset positive integer.

[0019] As an optimization, the preset frame rate is 10-30 frames / second; and / or,

[0020] The preset temperature step size is ; and / or,

[0021] The collection accuracy of the temperature parameter is ; and / or,

[0022] .

[0023] As an optimization, the improved wavelet transform algorithm specifically includes the following sub-steps:

[0024] S2.1, adaptive selection of wavelet basis: selecting a biorthogonal wavelet as the wavelet basis, and dynamically determining a wavelet decomposition scale of 3-5 according to the gray mean and variance of the image;

[0025] S2.2, two-dimensional wavelet transform and detail component extraction: performing two-dimensional wavelet transform on the image to obtain a low-frequency detail component LL, and three high-frequency detail components LH, HL, and HH;

[0026] S2.3, adaptive threshold setting and singular point screening: calculating the gray standard deviation of the non-edge region in each of the high-frequency detail components As an indicator of noise intensity, an adaptive threshold value is set to be wherein k is an empirical coefficient of 2.5-3.5; a first neighborhood window is used to traverse each of the high-frequency detail components, and candidate singular points whose center pixel gray value is a local maximum and exceeds the corresponding threshold value T are screened out, and the intersection of the candidate singular points in the three high-frequency detail components is taken as the effective singular points, thereby obtaining an effective singular point set;

[0027] S2.4, calculation of singularity value: extracting the wavelet coefficient modulus value at the position of the effective singular point under different wavelet decomposition scales, obtaining a decay index by fitting the decay curve of the modulus value with respect to the scale, and taking the decay index as the singularity value.

[0028] As an optimization, after obtaining the effective singular points, step S2 further includes, before step S3 is performed:

[0029] For each of the effective singular points, the effective singular point is taken as a to-be-verified point;

[0030] adopting at least two second neighborhood windows of different sizes to perform secondary singular point judgment at the position of the to-be-verified point respectively;

[0031] counting the number of times that the to-be-verified point is judged as a singular point under all second neighborhood windows, and calculating a confidence value C of the to-be-verified point according to the number of times;

[0032] comparing the confidence value C with a preset confidence threshold value C0, if , determining that the to-be-verified point passes the verification, and retaining the to-be-verified point as a final valid singular point; if , determining that the to-be-verified point fails the verification, and eliminating the to-be-verified point from the valid singular point set.

[0033] As an optimization, the multi-layer judgment rule is a three-layer judgment rule, and the following three-layer judgment rule is performed based on the valid singular points obtained after the screening and verification in step S2:

[0034] First layer judgment: if a pixel position is a valid singular point in the current scene image A, and is not a valid singular point in the latest single-frame background image in the background image sequence , it is determined that the pixel position is a noise point;

[0035] Second layer judgment: if a pixel position is a valid singular point in the current scene image A and the latest single-frame background image , the difference value between the singular values corresponding to the pixel position in the current scene image A and the latest single-frame background image is calculated ; a difference value threshold value is set, if , it is determined that the pixel position is a static bad point, otherwise it is determined that the pixel position is a noise point, wherein, is the singular value corresponding to the pixel position in the current scene image A, is the singular value corresponding to the pixel position in the single-frame background image ;

[0036] Third layer judgment: if a pixel position is not a valid singular point in the current scene image A, and is a valid singular point in the latest single-frame background image , the detection result of the pixel position in the historical background images to is retrieved; if the pixel position was a valid singular point in any one of the historical background images, it is determined that the pixel position is a dynamic bad point, otherwise it is determined that the pixel position is a noise point.

[0037] As an optimization, the difference value threshold value The value range of the parameter is 0.1-0.3.

[0038] As an optimization, the specific process of S4 is:

[0039] S4.1, parameter adaptive calibration:

[0040] S4.1.1, periodically statistics the accuracy rate of the bad point detection result, and compares the accuracy rate with a preset accuracy rate threshold;

[0041] S4.1.2, if the accuracy rate is lower than the preset accuracy rate threshold, at least one of the wavelet decomposition scale, the singular point detection threshold coefficient k, the bad point judgment difference threshold , the background image acquisition temperature step size is adjusted in the direction and degree of deviation of the accuracy rate according to a preset adjustment step;

[0042] S4.1.3, using the newly collected verification image of a preset number M frames after parameter adjustment, re-executing steps S2 to S3, calculating a new accuracy rate to verify the parameter adjustment effect, until the accuracy rate reaches or exceeds the preset accuracy rate threshold, M is a positive integer;

[0043] S4.2, bad point adaptive compensation:

[0044] According to the type, position information and corresponding temperature parameter in the bad point detection result, the corresponding compensation strategy is called to perform pixel compensation.

[0045] As an optimization, the corresponding compensation strategy is called to perform pixel compensation in S4.2, specifically including:

[0046] S4.2.1, compensation strategy initialization:

[0047] A differentiated compensation strategy library associated with bad point types and image regions is pre-stored; the strategy library includes: a regular neighborhood interpolation strategy for non-edge area static bad points, a weighted neighborhood interpolation strategy for edge area static bad points, and a temperature prediction based pre-compensation strategy for dynamic bad points;

[0048] S4.2.2, static bad point compensation execution:

[0049] For the pixel of the static bad pixel type, the image region where the static bad pixel is located is determined; if the static bad pixel is located in a non-edge region, the regular neighborhood interpolation strategy is called to take the gray scale values of all normal pixels in a preset compensation neighborhood window of the static bad pixel as the center of the static bad pixel, and interpolation calculation is performed to obtain a compensation value; if the static bad pixel is located in an edge region, the weighted neighborhood interpolation strategy is called to take the gray scale values of pixels in a preset compensation neighborhood window of the static bad pixel as the center of the static bad pixel, and interpolation calculation is performed in combination with a preset weight coefficient to obtain a compensation value, wherein pixels belonging to an edge direction are given a higher weight.

[0050] S4.2.3, dynamic bad pixel compensation execution:

[0051] For the pixel of the dynamic bad pixel type, the pre-compensation strategy is called to query a temperature-bad pixel mapping relationship established based on historical data according to a temperature parameter associated with the dynamic bad pixel; when an absolute value of a difference between a real-time working temperature of the detector and a temperature threshold value at which the dynamic bad pixel is prone to appear and indicated in the mapping relationship is less than or equal to a preset early warning temperature difference , a compensation parameter pre-stored for the temperature threshold value is automatically called to pre-compensate the pixel of the dynamic bad pixel.

[0052] As an optimization, the regular neighborhood interpolation strategy is:

[0053] The gray scale values of all normal pixels in a preset compensation neighborhood window of the static bad pixel to be compensated are taken as the center of the static bad pixel, an arithmetic mean value of the gray scale values is calculated, and the arithmetic mean value is taken as a compensation value to replace an original gray scale value of the bad pixel.

[0054] The weighted neighborhood interpolation strategy is:

[0055] The pixels in a preset compensation neighborhood window of the static bad pixel to be compensated are taken as the center of the static bad pixel.

[0056] After a weight coefficient is given to the gray scale values of the pixels in the compensation neighborhood window, a weighted average value is calculated as a compensation value; wherein the weight coefficient is given according to the following rule: the weight coefficient of a neighboring pixel in a direction of an edge tangent line to which the static bad pixel belongs is higher than that of a pixel in another direction.

[0057] The pre-compensation strategy is:

[0058] According to the temperature threshold value at which the dynamic bad pixel is prone to appear obtained through the query, a compensation parameter verified to be effective at the temperature threshold value is called from historical compensation records; the compensation parameter at least includes a size of a compensation neighborhood window, a weight coefficient distribution, or a direct optimal compensation value; when the pre-compensation trigger condition is met, the compensation parameter is directly applied to replace the gray scale value of the pixel of the dynamic bad pixel.

[0059] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0060] The present application can detect bad points in real time and accurately during the normal working process of the detector by the cooperative analysis of the current working image and the multi-temperature background image sequence and the improved wavelet transform algorithm, without interrupting the work or relying on an external black body source, and the detection accuracy is significantly higher than that of the traditional filtering denoising method.

[0061] The present application creatively uses the multi-temperature background image sequence and its historical data to design three layers of judgment rules, which can effectively distinguish static bad points, dynamic bad points and random noise, and solve the industry problem that the traditional method is not sensitive to dynamic bad points.

[0062] The present application periodically evaluates the detection accuracy and feeds back to adjust the key algorithm parameters (such as wavelet decomposition scale, detection threshold, etc.), so that the method can adapt to different types of detectors and the performance drift of the detector over time, and has strong universality and robustness.

[0063] According to the type and position of the bad point, the present application performs a differentiated compensation strategy. For the edge area static bad point, a weighted interpolation is used to protect the details, and for the dynamic bad point, a pre-compensation is performed based on temperature prediction, so that the original details and spatial continuity of the image are maintained to the greatest extent while eliminating the bad points, and the overall imaging quality is improved.

[0064] The present application adds a multi-scale neighborhood verification step after the core detection, further filters out false singular points through confidence evaluation, so that the data input into the judgment link is more pure, and the misjudgment rate is fundamentally reduced. BRIEF DESCRIPTION OF DRAWINGS

[0065] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0066] Figure 1 A flowchart of an infrared detector bad point detection method according to the present application. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below with reference to the embodiments and drawings, and the illustrative embodiments of the present application and their descriptions are only used to explain the present application, and do not constitute a limitation on the present application.

[0068] Embodiment 1 of the present application provides an infrared detector bad point detection method, as shown in FIG. 1, which includes steps S1-S4, and next, the four steps will be introduced in detail. Figure 1 As shown in FIG. 1, which includes steps S1-S4, and next, the four steps will be introduced in detail.

[0069] S1, acquiring a current scene image A during the working period of an infrared detector, and a plurality of frames of background images of an isothermal uniform surface collected by the infrared detector at different working temperatures to form a background image sequence where i is an acquisition serial number and .

[0070] This step is image acquisition, and the core is to synchronously acquire a current scene image A reflecting actual scene information and a background image sequence reflecting the background response of the detector itself at different temperatures .

[0071] In some embodiments, the specific process of S1 is as follows:

[0072] S1.1, acquiring a scene image of the working period of the detector according to a preset frame rate, and forming and storing the current scene image A after pixel gray value calibration and image size normalization preprocessing.

[0073] The current scene image A is the normal imaging result of the detector on the observed target scene, and is used to analyze the possible bad points under the real working condition.

[0074] The infrared detector works continuously at its nominal frame frequency. After the output of the image sensor (focal plane array) is converted into digital, an original digital video stream is formed. The image acquisition unit (usually FPGA or a special image interface chip) captures a single frame image from the video stream at a preset frame rate (20 frames per second in this embodiment). The selection of this frame rate takes into account the real-time requirement and the calculation load of the processing system. The captured original image needs to go through a standardized preprocessing pipeline to ensure the accuracy and consistency of subsequent algorithm processing:

[0075] 1. Non-uniformity correction and gray scale calibration: eliminate the fixed pattern noise caused by the inconsistency of the response rate and dark current of each pixel of the detector. The correction coefficient calibrated at the factory of the detector is applied. Usually, two-point correction method is used, and the calculation formula is: ;

[0076] where (x, y) is the pixel coordinate, is the original gray value, Gain and Offset are the gain and offset coefficient matrices calibrated in advance, is the gray value after pixel coordinate correction. After this step, the response of each pixel to the same radiation input is calibrated to be consistent.

[0077] 2. Image size normalization: Ensure all input images have the same spatial resolution, which is convenient for subsequent pixel-level comparison and indexing. Determine the original image resolution. If it is consistent with the preset resolution (640x512 in this embodiment), it is directly passed. If it is not consistent, the image is scaled to the preset resolution using a bilinear interpolation algorithm. This preset resolution needs to match or be an integer multiple of the effective pixel count of the detector.

[0078] The image after preprocessing is officially marked as the current scene image A. This image is sent to a first-in-first-out (FIFO) temporary image buffer area. This buffer area is usually designed to accommodate 1-5 frames of images, and its update period is synchronized with or slightly faster than the acquisition period of the subsequent background image, ensuring that the latest scene image can always be obtained when analysis is needed.

[0079] S1.2, trigger the shutter to close through a linkage control mechanism when the detector operating temperature changes by a preset temperature step, and acquire a single frame of background image and pre-process, store in the background image database after associating the temperature parameter, to form the background image sequence The background image database uses a circular coverage mechanism to retain the last N frames of background images, and N is a preset positive integer.

[0080] Background image sequence is the key innovative data source of the present application. It acquires images of uniform radiation background at different operating temperature points through the structure of the detector itself, and is used to separate the inherent defects (bad pixels) of the detector from the scene content.

[0081] A platinum resistance (PT100 / PT1000) or an integrated digital temperature sensor (such as TI TMP117) is used to monitor the core temperature of the infrared detector chip package in real time with an accuracy of . The installation position of the sensor needs to be thermally designed to ensure that its temperature measurement point can quickly and accurately reflect the actual temperature of the focal plane array. A normally open mechanical shutter is integrated in the optical path of the detector. The shutter blade surface is coated with high-emissivity black paint, so that it can act as an isothermal radiation surface close to a black body when closed.

[0082] During system initialization, the initial temperature of the detector is recorded and a temperature change step is set (the embodiment sets ). The system continuously monitors . When is detected (the temperature at which the last successful background image was acquired), the following ordered operations are triggered immediately:

[0083] 1. Shutter control: send close command to shutter driver. Shutter closes completely within ≤ 50 ms, blocking external scene light and leaving the detector focal plane to receive only the IR radiation from the closed shutter blades.

[0084] 2. Steady state wait: after shutter closing, wait for a short steady time (e.g. 100 ms) to eliminate mechanical vibrations and transient responses in the detector output caused by shutter action.

[0085] 3. Image acquisition: under steady state conditions, acquire a frame of the detector output image with shutter closed. This image is the isothermal uniform field background image at the current detector temperature .

[0086] 4. Shutter reset: immediately after acquisition, control shutter to reopen, restoring normal observation function of the detector. The entire shutter operation cycle (close-acquire-open) is controlled within 200-300 ms, having minimal impact on normal observation.

[0087] 5. Preprocessing of background image and information association:

[0088] The acquired raw background image goes through exactly the same non-uniformity correction and size normalization preprocessing as the current scene image A. After processing, a usable single-frame background image is generated.

[0089] The is strongly associated with the detector temperature value at the time of this acquisition, forming a data pair . The associated data pair is stored in a dedicated background image database. This database can be physically a ring buffer in memory, or a circular storage area in flash memory. Each entry contains: image data , temperature value , time stamp and acquisition sequence number i. i is used to identify the i-th background image acquisition event triggered by temperature change reaching a step size.

[0090] The database has a fixed capacity, i.e. it retains the most recent N frames of background images. The lower limit of this embodiment is set to N = 10. When a new background image needs to be stored: if the current number of stored entries is less than N, it is directly appended. If N entries have already been reached, the oldest record (i.e. the record with acquisition sequence number i-N) is overwritten. This mechanism ensures that the database is always dynamically maintaining a sample of the detector's background response over the most recent temperature change history. For example, if N = 10, the database covers a temperature change history of approximately temperature range, sufficient to capture the "appearance-disappearance" temperature characteristics of most dynamic bad pixels.

[0091] The system maintains a logical pointer always pointing to the latest background image in the database (i.e. the frame of imax). When making bad pixel judgment, the current scene image A will be compared with this latest background image in real time. Meanwhile, the historical sequence of the entire database will be used for the retrospective judgment of dynamic bad pixels. This design ensures that the background data for analysis is highly relevant to the current working state (temperature) of the detector.

[0092] Through the above detailed implementation process, the application ingeniously uses the shutter of the detector itself without introducing an external blackbody source, and automatically constructs a high-quality uniform background image library that is updated synchronously with the working temperature, thereby laying an irreplaceable data foundation for subsequent high-precision bad pixel discrimination.

[0093] S2, using an improved wavelet transform algorithm to detect singular points in each frame of image in the current scene image A and the background image sequence , screen out effective singular points and calculate the singularity value of each effective singular point; the improved wavelet transform algorithm includes wavelet basis adaptive selection, two-dimensional wavelet transform and high-frequency detail component extraction, adaptive threshold setting and singular point screening, singularity value calculation.

[0094] Bad pixels appear as local gray level abrupt points on the image, and the core idea of the method is to use the improved wavelet transform to enhance the extraction of such abrupt features, and to improve the reliability of detection through multi-scale space verification.

[0095] In some embodiments, the improved wavelet transform algorithm specifically includes the following sub-steps:

[0096] S2.1, wavelet basis adaptive selection: selecting biorthogonal wavelet as the wavelet basis, and dynamically determining the wavelet decomposition scale of 3-5 levels according to the gray mean value and variance of the image.

[0097] The purpose of this step is to select an adaptive time-frequency analysis basis function for wavelet transform operation and dynamically configure the depth of multi-resolution analysis.

[0098] Wavelet basis selection principle: infrared images usually have a flat background, low signal-to-noise ratio, and bad pixels appear as impulse-like singular points. Biorthogonal wavelets (such as bior2.4, bior3.5, etc.) can more accurately locate singular points and reduce reconstruction distortion due to their linear phase characteristics and good regularity, so they are selected as the wavelet basis. In this embodiment, bior3.5 wavelet is selected by default.

[0099] Decomposition scale adaptive determination (refinement of "according to the mean and variance of the image's gray level"): Calculate the global mean and variance of the input image I and variance The determination of decomposition scale J follows the following rules:

[0100] If the overall contrast of the image is low and the details are gentle ( ), a deeper decomposition is needed to capture the weak bad point features that may be overwhelmed by noise at a higher scale (coarser resolution), so set J = 5.

[0101] If the image contrast is moderate ( ), set J = 4. This is typical for most outdoor scenes.

[0102] If the image contrast is high and the details are rich ( ), a shallower decomposition can effectively extract significant singular features while avoiding redundant calculations, so set J = 3.

[0103] This dynamic mechanism ensures the adaptive ability of the algorithm to different scene contents, always analyzing in the appropriate scale space.

[0104] S2.2, two-dimensional wavelet transform and detail component extraction: Perform two-dimensional wavelet transform on the image, and decompose to obtain a low-frequency detail component LL, and three high-frequency detail components LH, HL, and HH.

[0105] This step decomposes the image into different frequency subbands to separate the background and details.

[0106] Transformation operation: Perform J-level two-dimensional discrete wavelet transform on the preprocessed image. Each level of decomposition decomposes the image into four subbands:

[0107] Low-frequency approximation component ( ): Carries the main energy of the image, reflecting large-area background and slowly varying information.

[0108] Horizontal high-frequency detail component ( ): Highlights the edges and abrupt changes of the image in the horizontal direction.

[0109] Vertical high-frequency detail component ( ): Highlights the edges and abrupt changes of the image in the vertical direction.

[0110] Diagonal high-frequency detail component ( ): Highlights the edges and abrupt changes of the image in the diagonal direction.

[0111] Where j represents the decomposition level, from 1 (finest) to J (coarsest).

[0112] Bad pixels, as local isolated gray level jumps, have their energy mainly contained in the high frequency detail components (LH, HL, HH). The low frequency component LL mainly contains background and uniform region information, in which the bad pixel feature is severely diluted. Therefore, the subsequent singular point detection (S2.3 and S2.4) will be completely and only based on these three high frequency detail components, discarding the low frequency component LL. This strategy greatly reduces the data processing amount and focuses on the target feature.

[0113] S2.3, adaptive threshold setting and singular point screening: calculate the gray level standard deviation of the non-edge region in each of the high frequency detail components Set an adaptive threshold as a noise intensity indicator , where k is an empirical coefficient of 2.5-3.5; traverse each of the high frequency detail components using a first neighborhood window, screen out candidate singular points whose center pixel gray level value in the first neighborhood window is a local maximum and exceeds the corresponding threshold T, and take the intersection of the candidate singular points in the three high frequency detail components as the effective singular points, thereby obtaining an effective singular point set.

[0114] This step aims to identify real singular point candidates from high frequency noise.

[0115] Noise intensity estimation: Directly calculating the standard deviation of the entire detail component will be disturbed by real edges and structures in the image, resulting in overestimation of the noise level. First, for each high frequency detail component image (such as ), calculate its gradient amplitude using the Sobel edge detection operator, and set a lower threshold to mark out the obvious edge pixels to generate a binary edge mask. Ignore the pixel regions marked as edges by the mask, and only calculate the standard deviation of the gray level values of the remaining non-edge region pixels, denoted as , , for each scale j. This value more purely reflects the intensity of the image noise at this scale. For simplicity, the average of at each scale is often taken as the final noise indicator , , of the component.

[0116] For each high frequency detail component (LH, HL, HH), according to its noise intensity , calculate the detection threshold of the component: . Where k is an empirical coefficient, and the initial value is set to 3.0 in this embodiment. This threshold dynamically adjusts with the image noise level, ensuring the stability of the detection sensitivity under different imaging conditions.

[0117] A 3x3 first-neighborhood window is used to slide over the whole image for each of the three high-frequency detail component images. For the window center pixel (x, y) and its gray value I(x, y), it is determined whether it satisfies the following conditions:

[0118] Local extremum condition: I(x, y) is the maximum of the gray values of the 9 pixels in the current 3x3 window.

[0119] Significance condition: (T is the adaptive threshold corresponding to the detail component where the pixel is located).

[0120] If it is satisfied, the position (x, y) is marked as 1 (a candidate singular point) in a binary marker map of the current component.

[0121] After obtaining the candidate singular point marker maps of the three detail components LH, HL and HH, the three maps are subjected to logical AND operation. That is, only when a pixel position is marked as a candidate point in the three LH, HL and HH maps, it is finally confirmed as a valid singular point. This strict intersection operation requires that the singular point must show significant mutations in the horizontal, vertical and diagonal directions, thereby greatly suppressing false alarms caused by noise stripes or textures in a single direction. All points screened out through this step constitute an initial set of valid singular points, and their pixel coordinates are recorded.

[0122] S2.4, Singularity value calculation: Extract the wavelet coefficient modulus value of the position where the valid singular point is located under different wavelet decomposition scales, obtain the decay index by fitting the decay curve of the modulus value with the scale change, and take the decay index as the singularity value.

[0123] This step calculates a quantitative index for each valid singular point, representing the significant degree and stability of its "bad point" characteristics.

[0124] For each point P (coordinates ) in the set of valid singular points, the wavelet coefficients of its corresponding positions in the three high-frequency subbands LH, HL and HH are located at all J decomposition scales obtained by previous wavelet transform (due to downsampling, the coordinates need to be converted according to the scale). Take the maximum absolute value of the three coefficients as the wavelet coefficient modulus value of the point at the scale j, denoted as (j = 1, 2,..., J).

[0125] In theory, an ideal singular point (such as a bad point), its wavelet coefficient modulus value will show a power-law decay with the increase of the decomposition scale j (i.e. the observation becomes "coarse"). Plot For the scatter plot of scale j, linear fitting is performed by least square method to obtain the slope of the fitting straight line .

[0126] The quantized value of the point singularity is defined as The greater the S value, the slower the wavelet coefficient modulus value of the point decays with scale, that is, the singular feature remains significant at multiple scales, and it is more likely to be a real, stable bad point. Conversely, the singular feature of a point with a small S value may only appear at the finest scale, and it is more likely to be random noise.

[0127] In some embodiments, after obtaining the effective singular points, before performing step S3, step S2 further includes:

[0128] For each of the effective singular points, the effective singular point is taken as a to-be-verified point;

[0129] At least two second neighborhood windows of different sizes are used to perform secondary singular point determination at the location of the to-be-verified point;

[0130] The number of times the to-be-verified point is determined to be a singular point under all second neighborhood windows is counted, and a confidence value C of the to-be-verified point is calculated according to the number of times;

[0131] The confidence value C is compared with a preset confidence threshold C0, if , the to-be-verified point is determined to pass the verification, and the to-be-verified point is retained as a final effective singular point; if , the to-be-verified point is determined to fail the verification, and the to-be-verified point is removed from the set of effective singular points.

[0132] This step is an enhancement and de-aliasing process of the core detection algorithm, which is performed after obtaining the initial set of effective singular points and before formally entering the bad point determination step S3. The purpose is to use spatial multi-scale consistency to remove accidental and unstable detection results.

[0133] For example, a set of second neighborhood windows is predefined, and the size of the second neighborhood windows is greater than the size of the first neighborhood window used in the detection phase. In this embodiment, three sizes are used: 3x3, 5x5, and 7x7. At the same time, a confidence threshold C0 = 0.67 (i.e., about 2 / 3) is set.

[0134] Point-by-point verification process:

[0135] Initialization: traverse the initial set of effective singular points, and mark each point P as a to-be-verified point.

[0136] Multi-scale spatial verification: On the original gray image I (i.e. the image after preprocessing but before wavelet transform), take the point P to be verified as the center, and use 3x3, 5x5, 7x7 windows in turn to verify:

[0137] Verification criterion: Calculate the gray mean value of all pixels in the current window and the standard deviation . Determine whether the gray value I(P) of the center point P satisfies: . Wherein, is a loose threshold, for example, set to 2.0. This criterion aims to test whether point P is a statistical outlier in the local window.

[0138] Record: If the criterion is met, record the determination result of the window as "yes" (singular point), otherwise as "no".

[0139] Confidence calculation: Count the number of determination results as "yes" in the three windows, denoted as n. Calculate the confidence value C of the point to be verified P as n / 3.

[0140] If , consider that point P shows singularity in multiple spatial scales and is a high-confidence candidate point, which is retained and added to the final set of effective singular points.

[0141] If , consider that the singularity of point P is not stable or only appears accidentally at a specific scale, and it is likely to be noise or texture interference, which is eliminated.

[0142] By requiring that the singular point not only shows significant in the frequency domain (wavelet transform) but also maintains statistical abnormality in the multi-scale neighborhood in the spatial domain, its authenticity is confirmed twice. Experiments show that this step can filter out more than 90% of false singular points, significantly improving the input data quality of subsequent bad point judgment, and is a key link to reduce the overall false detection rate.

[0143] At this point, for image A and each background image , a final set of effective singular points containing pixel positions and corresponding singularity values S is obtained. These data will be directly input to the next stage of "bad point judgment".

[0144] S3, based on the singular point detection results, singularity value differences, and historical background image data of the corresponding pixel positions in the current scene image A and the background image sequence , distinguish static bad points, dynamic bad points and noise through multi-layer judgment rules, and obtain bad point detection results containing bad point type, location and corresponding temperature parameters.

[0145] The application fully utilizes the difference between the working image and the multi-temperature background image and the historical information in the time dimension by designing three layers of determination rules.

[0146] Before starting the pixel-by-pixel determination, the system needs to complete the following data preparation:

[0147] 1. The effective singular point set of the current scene image A: from the output of step S2 and the verification through the second field window, the final effective singular point set of the current scene image A is obtained, denoted as Each element in the set contains the pixel coordinates (x, y) and the corresponding singular value .

[0148] 2. The effective singular point set of the latest single-frame background image : from the background image database, the final effective singular point set corresponding to the background image with the latest time stamp (i.e., the maximum acquisition serial number i) is obtained, denoted as (e.g. ). Similarly, the set contains the coordinates and the singular value .

[0149] 3. The historical background image effective singular point record: for each historical image (e.g. ) in the background image database, the corresponding effective singular point set record is maintained. These records constitute a historical singular point appearance archive indexed by temperature and / or time.

[0150] The system will traverse each pixel position (x, y) of the detector and sequentially execute the following three-layer determination logic.

[0151] First layer determination: noise point rapid filtering:

[0152] Determination condition: the pixel position (x, y) belongs to (i.e., it is determined as an effective singular point in the current working image A), but does not belong to (i.e., it is not an effective singular point in the new uniform background image ).

[0153] Determination logic: a real bad point is a defect inherent to the detector hardware, and its abnormal response characteristics should be stable. In the uniform radiation background ( ), the interference of the scene content is excluded, and the bad point should consistently exhibit as a singular point. Therefore, if a point exhibits as a singular point only in A containing complex scenes, but disappears in the uniform , its singular characteristics are most likely caused by local high-contrast objects (such as point heat sources, edges) or transient random noise in the scene, rather than detector defects. ​

[0154] Decision: The pixel position is determined as a noise. Its type is marked as NOISE, and the subsequent decision flow for this pixel is ended.

[0155] Second layer decision: Static dead pixel identification and noise depth discrimination:

[0156] Entry condition: The pixel position (x, y) belongs to both and (i.e. it is detected as a valid outlier in both the current scene image A and the latest single-frame background image B).

[0157] Core calculation: Obtain the corresponding outlier values of this point in the two sets, respectively and . Calculate the absolute value of the difference between the two .

[0158] Decision threshold: Set a key difference threshold . In this embodiment, according to a large amount of experimental data, it is set to .

[0159] Decision logic:

[0160] If : It indicates that there is a significant difference in the outlier values of the point under the scene image and the uniform background. Although it behaves as an outlier under both conditions, its outlier degree is unstable. This instability usually means that the abnormal response of the point may be modulated or superimposed by specific structures or non-uniform illumination in the scene, and it is more likely to be an abnormal pixel sensitive to the background (such as a nonlinear response pixel), rather than a completely failed dead pixel. To simplify classification, this method classifies it as an abnormality that needs attention, but to distinguish from fixed dead pixels, it is still determined as a static dead pixel, but can be treated differently in subsequent compensation strategies (this part can be expanded).

[0161] If : It indicates that the quantified values of the point are highly consistent under two completely different imaging conditions (with scene and without scene). This strongly suggests that the abnormal response of the point is inherent and stable, and does not depend on the pattern of external incident radiation. This is consistent with the core characteristics of static dead pixels (such as dead pixels, hot pixels), i.e. their defects are fixedly present.

[0162] Decision result:

[0163] When , it is determined as a static dead pixel, and the type is marked as STATIC_DEAD (which can be further divided into bright or dark points).

[0164] When ​If so, it is determined as a noise point, and the type is labeled as NOISE. This step eliminates those points which are detected in both images but are unstable in feature (may be slight noise or coincidental weak scene structure) from the bad pixel candidate list, further reducing the false alarm rate.

[0165] Third tier decision: dynamic bad pixel capture

[0166] Entry condition: pixel position (x, y) does not belong to but belongs to (i.e. only appears as a valid outlier in the latest uniform background image, and is normal in the current scene image A).

[0167] Historical trace: the system automatically queries the detection record of this pixel position (x, y) in the historical frame set of the background image database. Check whether it has been recorded as a valid outlier in any historical background image in the past.

[0168] Decision logic:

[0169] 1. If there is a history record: this indicates that the "bad pixel" characteristics of this pixel are not a one-time accidental event at the current temperature . It has also shown abnormalities at one or more temperature points in the past (corresponding to the acquisition temperature of the historical background image). This intermittent behavior with temperature change is the typical feature of dynamic bad pixels. Dynamic bad pixels are usually related to the physical properties of the detector material or structure (such as resistance, tunneling current, etc.) that change with temperature.

[0170] If there is no history record: the pixel only behaves as an outlier at the current temperature , and is normal before the temperature history. This is more likely to be a one-time transient interference, readout circuit transient noise or accidental error during background acquisition.

[0171] Decision result:

[0172] If there is a history record, it is determined as a dynamic bad pixel, and the type is labeled as DYNAMIC, and the current "activated" temperature and the associated historical appearance temperature are recorded.

[0173] If there is no history record, it is determined as a noise point, and the type is labeled as NOISE.

[0174] After completing the three-tier decision for all pixels, the system integrates to generate the final bad pixel detection result. The result is a structured data list, which at least contains the following fields:

[0175] Pixel coordinates (x, y); ​

[0176] Bad pixel type: STATIC_DEAD, DYNAMIC, NOISE, NORMAL;

[0177] Associated temperature: for DYNAMIC type, record the current active temperature ; for all types, attach the current detector temperature as reference.

[0178] Singularity value: and / or (Optional, for subsequent analysis or compensation parameter calculation)

[0179] Timestamp.

[0180] This bad pixel detection result list, both as the final output of this period detection, also as the input of the next part "adaptive optimization and compensation" module, and can be used to update the historical archives, to provide more abundant basis for future decision. Through this progressive, multi-dimensional (space, frequency, temperature, time) judgment logic, the invention realizes the reliable identification of static and dynamic bad pixels with high precision and low misjudgment.

[0181] S4, periodically evaluate the accuracy of the bad pixel detection results, and dynamically calibrate the detection parameters according to the evaluation results, and feed back the bad pixel detection results to the detector control unit.

[0182] This step forms a complete closed loop from detection to processing to optimization, which is the key to the high practicality and long-term stability of the method.

[0183] S4.1 Parameter adaptive calibration:

[0184] This step aims to solve the problem of manual adjustment of algorithm parameters due to individual differences, aging drift or environmental changes of the detector. By establishing a closed-loop control loop based on performance feedback, automatic optimization of key parameters is realized.

[0185] S4.1.1 Accuracy statistics and evaluation period:

[0186] Evaluation period setting: the system performs performance evaluation at fixed time or frame interval. In this embodiment, it is set to every 100 frames of current scene image A (about 5 seconds of working time) as an evaluation period. This period should be much longer than the acquisition period of the background image to ensure the statistical significance of the evaluation sample.

[0187] Accuracy definition and statistics: in each evaluation period, the system has generated 100 bad pixel detection results. In order to evaluate its accuracy, a combination of manual assisted labeling and automatic statistics is adopted:

[0188] 1. From the 100 frames of image corresponding to the detection results, 20 frames are randomly selected as evaluation samples.

[0189] 2. By experienced operators, in the 20 frames of image, with the help of high-precision display and auxiliary tools, all confident bad points (including static and dynamic) are manually visually distinguished and labeled to form a reference true value list.

[0190] 3. The 20 detection results generated by the system are compared with the reference true value list. The precision (Precision) is calculated as the accuracy index:

[0191] Accuracy P = (TP) / (TP + FP);

[0192] Where TP (True Positive) is the number of bad points correctly detected by the system (consistent with manual labeling), and FP (False Positive) is the number of normal pixels misjudged as bad points by the system (i.e. false alarm).

[0193] Threshold comparison: compare the calculated accuracy P with a preset accuracy threshold . The threshold value of this embodiment is set to 0.9. The threshold value represents the minimum acceptable performance standard of the system.

[0194] S4.1.2 Parameter adjustment decision and execution:

[0195] If the accuracy P is lower than , the parameter calibration process is triggered. The system needs to determine the main reason for performance degradation (missed detection or false alarm), and decide the adjustment direction and target accordingly.

[0196] If (TP + FN) / GT is low (where FN is the missed detection and GT is the total number of manual labeling), i.e. the detection rate is low, indicating that the system is too conservative, there may be a large number of missed detections. The main contradiction is that the bad points are not effectively identified.

[0197] If the number of FP is significantly high, i.e. the false alarm rate is high, indicating that the system is too sensitive, a large number of noise or texture are misjudged as bad points. The main contradiction is that there are too many misjudgments.

[0198] Adjustment strategy and step size: according to the diagnosis result, at least one key parameter of wavelet decomposition scale, singular point detection threshold coefficient k, bad point judgment difference threshold , background image acquisition temperature step size is iteratively adjusted in a targeted and small amplitude. The preset adjustment step size is as follows:

[0199] ​Wavelet decomposition scale J: adjust step size by ±1. If there are too many missed detections, increase J (e.g., from 4 to 5) to capture features at coarser scales; if there are too many false alarms, decrease J (e.g., from 4 to 3) to reduce noise interference.

[0200] Singular point detection threshold coefficient k: adjust step size by ±0.2. If there are too many missed detections, decrease k (e.g., from 3.0 to 2.8) to lower the detection threshold; if there are too many false alarms, increase k (e.g., from 3.0 to 3.2) to raise the detection threshold.

[0201] Bad point determination difference threshold : adjust step size by ±0.05. If there are too many missed detections (especially static bad points), they may be misjudged as noise points due to being slightly larger than the threshold, so it is appropriate to increase ( e.g., from 0.2 to 0.25); if there are too many false alarms (the second layer of judgment judges noise points as static bad points), decrease ( e.g., from 0.2 to 0.15) to tighten the judgment conditions.

[0202] Background image acquisition temperature step size : adjust step size by ±0.1°C. If dynamic bad point detection is not ideal, it may be that the temperature sampling is not dense enough, so decrease ( e.g., from 1.0°C to 0.9°C) to obtain denser temperature samples; if the system load is too high or the background image changes too frequently, it is appropriate to increase.

[0203] Execution of adjustment: usually only one parameter that has the most significant impact on the current accuracy is selected for adjustment each time the calibration is performed, following the principle of easy first and difficult later (e.g., preferentially adjusting k and ). The adjustment instruction is issued to the corresponding algorithm module.

[0204] S4.1.3 Adjustment effect verification and iteration:

[0205] Parameter adjustment cannot be performed blindly and must be verified online to ensure that the adjustment is effective.

[0206] Verification set collection: after the parameter adjustment is completed, the system immediately starts collecting a new, independent verification image set. This verification set consists of M=10 frames of continuous new scene images A. These images are collected after the new parameters take effect, ensuring the relevance of the verification data to the adjusted state.

[0207] Re-execution and evaluation: for these M frames of verification images, steps S2 (singular point detection) and S3 (bad point determination) are completely re-executed to generate new detection results. Similarly, the accuracy of these M frames of results is calculated.

[0208] Iteration termination condition:

[0209] If , it indicates that the parameter adjustment is effective, and the system performance has recovered to the standard. The calibration process ends, and the new parameters are solidified until the next periodic evaluation.

[0210] If , it indicates that a single adjustment has failed to solve the problem. The system will perform the diagnosis and adjustment of S4.1.2 again according to the gap between and , and then continue to verify with new M-frame images. This evaluation-adjustment-verification cycle will continue until the accuracy meets the standard, ensuring the robustness and reliability of the calibration.

[0211] S4.2 Bad Pixel Adaptive Compensation:

[0212] This step converts the detection results into actual image quality improvement actions. The core idea is to take the most targeted compensation strategy according to the physical characteristics and spatial context of the bad pixels. The specific process is as follows:

[0213] S4.2.1 Compensation Strategy Library Initialization:

[0214] At startup or initialization, a compensation strategy lookup table is constructed in the differentiated compensation strategy library (implemented through memory), which is the basis for subsequent compensation execution.

[0215] Strategy Index: The strategy library is indexed by bad pixel type and image region type as a joint primary key.

[0216] Strategy Content:

[0217] 1. Regular Neighborhood Interpolation Strategy: Associated key is (STATIC_DEAD, non-edge region). This strategy is defined as the basic operation.

[0218] 2. Weighted Neighborhood Interpolation Strategy: Associated key is (STATIC_DEAD, edge region). This strategy includes additional weight distribution rules.

[0219] 3. Pre-compensation strategy based on temperature prediction: Associated key is (DYNAMIC, *) (* indicates no distinction between regions). This strategy includes temperature query, trigger judgment, and parameter call logic.

[0220] S4.2.2 Static Bad Pixel Compensation Execution:

[0221] For each pixel judged as STATIC_DEAD:

[0222] 1. Region determination: use Canny edge detection algorithm to extract edges from the current scene image A, and generate a binary edge map. Take the bad pixel coordinates as the center, check whether there are edge pixels in the 3x3 check neighborhood window. If so, it is determined that the bad pixel is located in the edge region; otherwise, it is determined to be a non-edge region.

[0223] 2. Non-edge region compensation (regular neighborhood interpolation):

[0224] Operation: take the bad pixel p as the center, and take its preset 5x5 compensation neighborhood window (window size can be configured).

[0225] Calculation: traverse all pixels in the window except p, if the pixel is not marked as any type of bad pixel, it is considered as a "normal pixel". Collect the gray values of all normal pixels and calculate their arithmetic mean.

[0226] Replacement: assign this arithmetic mean directly to pixel p to complete the compensation.

[0227] 2. Edge region compensation (weighted neighborhood interpolation):

[0228] Operation: also take the bad pixel p as the center, and take the 5x5 compensation neighborhood window.

[0229] Edge direction analysis: calculate the image gradient direction of the position of pixel p (through Sobel operator), which is the normal direction of the edge. Then the tangent direction perpendicular to the normal direction is the edge direction (i.e. edge direction).

[0230] Pixel screening: traverse all pixels in the compensation neighborhood window. Only the pixels that are not marked as any type of bad pixel (i.e. normal pixels) are included in the subsequent weight assignment and weighted average calculation. The bad pixel to be compensated and other bad pixels in the neighborhood are excluded.

[0231] Weight assignment:

[0232] For two adjacent pixels in the compensation neighborhood window located in the edge tangent direction (i.e. along the edge direction), give them higher weights .

[0233] For other direction pixels in the compensation neighborhood window, give them standard weights .

[0234] The bad pixel itself and other pixels in the window that are marked as bad pixels have a weight of 0 and do not participate in the calculation.

[0235] Calculation and replacement: Calculate the weighted average of all normal pixel gray values in the compensation neighborhood window, and replace the original value of p with the weighted average as the compensation value. This method can effectively utilize edge continuity information and avoid the edge from becoming blurred or broken after compensation. The calculation formula is as follows: wherein i is to traverse all the screened normal pixels.

[0236] S4.2.3 Dynamic bad pixel compensation execution:

[0237] For each pixel q determined as DYNAMIC, the compensation is preventive and data-driven.

[0238] 1. Temperature-bad pixel mapping relationship construction: The system continuously learns in the background. Whenever a dynamic bad pixel q is identified, its activation temperature is recorded. After accumulation for multiple working cycles (hours or days), a set of activation temperature samples of q can be obtained. By kernel density estimation or clustering analysis (such as DBSCAN) on these samples, the main active temperature interval (temperature threshold) of q is determined. This temperature interval is the temperature-bad pixel mapping relationship of q, that is, the temperature threshold at which the dynamic bad pixel is prone to appear.

[0239] 2. Pre-compensation parameter learning: Within the above temperature interval , when q behaves as a bad pixel, the system attempts to apply different compensation parameters (such as different window sizes, different weight schemes) for compensation, and evaluates the compensation effect (such as the difference between the compensated pixel value and the background) in the next frame of background image. The set of compensation parameters with the best evaluation effect (such as the smallest difference) is saved as the historical optimal compensation parameters for q in this temperature interval. These parameters may be a special window shape of 7x7, a set of asymmetric weight distribution, or even a fixed gray compensation value learned directly (i.e., a direct optimal compensation value).

[0240] 3. Pre-compensation triggering and execution:

[0241] Real-time monitoring: The system continuously reads the real-time temperature of the detector .

[0242] Triggering judgment: Set a pre-warning temperature difference . When is met or , it is considered that the temperature has approached the active interval of the dynamic bad pixel q.

[0243] Execute pre-compensation: Once the triggering condition is met, the system immediately calls the stored compensation parameters for q in The optimal compensation parameter pre-stored in the interval is directly applied to the pixel q in the current frame image to replace the gray value thereof.

[0244] The generation and application of the optimal compensation parameter specifically include:

[0245] A. Parameter learning stage:

[0246] In the temperature interval , when the dynamic bad pixel q actually appears in the background image, the compensation learning module performs the following steps:

[0247] 1. Multi-strategy parallel test: simultaneously use multiple candidate compensation strategies to repair q, including:

[0248] (a) Direct assignment strategy: attempt to replace it with multiple candidate fixed gray values {V1, V2,...}.

[0249] (b) Fixed weight interpolation strategy: attempt to use multiple sets of pre-set neighborhood pixel weight coefficients {W1, W2,...} for weighted averaging.

[0250] (c) Modeling interpolation strategy: attempt to call different interpolation algorithms (such as Gaussian weighted interpolation, bicubic interpolation, anisotropic diffusion repair) and adjust their parameters.

[0251] 2. Effect quantitative evaluation: in the subsequent acquisition of uniform background images whose temperature is still in the interval, calculate the absolute difference or squared difference between the pixel value of q position after compensation by each strategy and the ideal background value.

[0252] 3. Optimal parameter selection: select the compensation strategy with the optimal evaluation result (smallest difference) and its corresponding specific parameters (i.e., fixed , or weight set , or algorithm identifier and parameter set ), bind and store it as the optimal compensation parameter for the bad pixel q under the temperature interval.

[0253] B. Parameter application stage (pre-compensation execution):

[0254] When the real-time temperature meets the pre-compensation trigger condition, the system performs:

[0255] 1. Parameter call: read the optimal compensation parameter pre-stored for the bad pixel q in the temperature interval.

[0256] 2. Parameter analysis and execution:

[0257] 2.1, if the parameter is a direct optimal compensation value , then the gray value of pixel q in the current frame is directly replaced by .

[0258] 2.2, if the parameter is a weight coefficient distribution , then take the neighborhood centered at q, and calculate the weighted average of the real-time gray values of the normal pixels in the neighborhood, and assign the result to q.

[0259] 2.3, if the parameter is an algorithm identifier and a parameter set , then call the corresponding interpolation algorithm, input and the neighborhood information of q in the current image, calculate the compensation value and assign it to q.

[0260] Specifically, if the parameter is an algorithm identifier and a parameter set , then call the corresponding interpolation algorithm, input and the neighborhood information of q in the current image, calculate the compensation value and assign it to q, the implementation process is as follows:

[0261] 2.3.1, a compensation algorithm registry is established, the specific process is as follows:

[0262] 2.3.1.1, algorithm registration: in the system initialization stage, all supported compensation algorithms (such as Gaussian weighted interpolation, bicubic interpolation, anisotropic diffusion repair, etc.) are registered with the central registry. Each registration entry contains:

[0263] algorithm identifier: a unique string or number code, used to refer to the algorithm (such as "GAUSSIAN_INTERP").

[0264] algorithm function pointer: points to the memory address of the specific calculation function of the algorithm.

[0265] parameter template: defines the type and structure of the parameters required by the algorithm (such as "window size" and "standard deviation σ" for Gaussian algorithm).

[0266] 2.3.1.2, unified interface: all algorithms follow the same calling interface specification, that is: input parameters and image data, output a compensated pixel value. This allows the system to schedule different algorithms in a unified manner.

[0267] 2.3.2, the composition and storage of parameter set In the learning stage, when the system determines that the optimal compensation method for the bad point q in a certain temperature interval is a certain specific algorithm, the

[0268] stored by it contains two parts of information: algorithm identifier: specifies which registered algorithm to use (for example, identifier "GAUSSIAN_7x7").

[0269]

[0270] ​​Algorithm-specific parameters: concrete configuration values needed when the algorithm is computed. These parameters are stored in a structured key-value pair form. For example:

[0271] For the "GAUSSIAN_7x7" algorithm, the parameters can be: {"window_size": 7, "sigma": 1.5};

[0272] For the "BICUBIC" (bicubic interpolation) algorithm, the parameters can be: {"b":0.0,"c":0.75} (parameters of the bicubic function);

[0273] For the "ANISO_DIFFUSION" (anisotropic diffusion repair) algorithm, the parameters can be: {"iterations":5,"kappa":30};

[0274] After these information are serialized, they are associated with the coordinates of the bad pixel q and its corresponding temperature interval, and stored in the system's non-volatile memory (such as flash memory).

[0275] 2.3.3 Real-time pre-compensation execution flow:

[0276] When the real-time temperature of the detector meets the pre-compensation trigger condition, the system performs the following steps for the bad pixel q:

[0277] Step 1: Parameter loading and parsing:

[0278] The system retrieves the corresponding data packet from the storage according to the coordinates of q and the current temperature interval. Then it parses it to separate the algorithm identifier and the specific parameter key-value pairs.

[0279] Step 2: Algorithm lookup and instantiation:

[0280] The system matches the parsed algorithm identifier with the compensation algorithm registry. After finding it, the system knows which specific function should be called for calculation, and at the same time, according to the parameter template defined by the algorithm, it converts the key-value pairs in into a memory data structure (such as a struct) that can be directly used by the function.

[0281] Step 3: Neighborhood data preparation:

[0282] The system takes the coordinates of the bad pixel q in the current real-time image as the center, and according to the The system extracts a 7x7 pixel region from the image buffer, centered at the pixel q. Meanwhile, the system checks each pixel in this region: if a pixel is also marked as bad (static or dynamic), it is marked as "invalid"; only pixels marked as "normal" have their intensity values used in the subsequent calculations.

[0283] Step 4: Algorithm dispatch and computation:

[0284] The system passes the prepared neighborhood pixel data (with valid / invalid flags) and the instantiated parameter structure to the algorithm computation function found in Step 2.

[0285] This function computes according to its internal logic and the passed-in parameters. For example:

[0286] A. If the algorithm is Gaussian-weighted interpolation, it generates a 7x7 Gaussian weight matrix with sigma=1.5, then only for pixels in the neighborhood marked as "normal", it multiplies their intensity values by the corresponding weights and sums them up, finally divides by the total weight to get the compensation value.

[0287] The core of this implementation is to build a dynamic, valid Gaussian weight system based only on normal pixels.

[0288] Input: algorithm parameters ; ImagePatch structure of the pixel q to be compensated (contains 7x7 neighborhood intensity data and pixel status flags).

[0289] Output: compensated pixel intensity value.

[0290] Specific steps:

[0291] 1. Generate the base Gaussian kernel: take the 7x7 window center as the coordinate origin (0,0), calculate the Gaussian weight at each integer coordinate (u,v) (u,v∈[-3,3]):

[0292] where . Then normalize so that its sum is 1.

[0293] 2. Build the valid weight mask: generate a binary mask matrix M of the same size according to ImagePatch.pixel_status. If pixel_status[i][j] is marked as "normal", then M(i,j)=1; if marked as "bad / invalid" (including the center pixel q itself), then M(i,j)=0.

[0294] 3. Calculate effective Gaussian weight matrix: multiply the base Gaussian kernel with the mask element-wise to get the effective weight matrix This operation forces the weight of the bad pixel position to zero, preventing it from participating in the calculation.

[0295] 4. Calculate compensation value:

[0296] Numerator: calculate the sum of the product of the gray value of all normal pixels in the neighborhood and their corresponding weight in . where I(i,j) is the gray value.

[0297] Denominator: calculate the sum of all weight values in . .

[0298] Compensation value: .

[0299] The denominator is dynamically changing, ensuring that even if there are multiple bad pixels in the neighborhood causing the total effective weight to be less than 1, the final compensation value can maintain brightness consistency through normalization, avoiding dark or bright bias.

[0300] B. If the algorithm is bicubic interpolation algorithm.

[0301] The key of this implementation is to simulate the fitting process of bicubic surface in the local neighborhood with invalid pixels, and use the valid pixel sample points to estimate the center missing value.

[0302] Input: algorithm parameters (standard bicubic parameters); ImagePatch of the pixel to be compensated q.

[0303] Output: compensated pixel gray value.

[0304] Specific steps:

[0305] 1. Determine the interpolation region and sample points: treat the 7x7 ImagePatch as a continuous local coordinate plane, with the center point q at (0, 0). Define the 4x4 core region required for interpolation (usually the nearest 16 pixels). Check the pixel_status corresponding to the 16 preset sample point positions.

[0306] 2. Handle invalid sample points:

[0307] If a preset sample point is marked as "invalid", do not directly use the gray value of that point.

[0308] Search for the nearest pixel with "normal" status in a smaller neighborhood (e.g. 3x3) around the invalid point, and use its gray value as the replacement value for the sample point.

[0309] If no substitute pixel is found, the weight of the sample point is set to zero in the subsequent calculation.

[0310] 3. Perform bicubic interpolation calculation:

[0311] Based on the bicubic interpolation formula, the value of the interpolation point (0, 0) is obtained by weighted summation of the gray values of the surrounding 4x4 sample points. The weight is determined by the horizontal and vertical distances (dx, dy) of the sample point from (0, 0), and adjusted by parameters b, c.

[0312] The weight function R(x) is usually: , for , for ; 0, otherwise}, where a is usually -0.75 (corresponding to b=0, c=0.75).

[0313] Final compensation value where (i,j) iterates through the 4x4 sample points processed in step 2, and I(i,j) is the gray value of each sample point.

[0314] Transform the global image bicubic interpolation into a local adaptive sample point repair technique that can bypass invalid pixels. By dynamically finding substitute sample points, even in the case of incomplete local samples (with bad points), a smooth fitting surface can still be constructed to estimate the center value, which can better maintain details than simple neighborhood interpolation.

[0315] C. If the algorithm is anisotropic diffusion, it will take the neighborhood of the bad point q as the initial state, and perform several iterations (iterations=5) of diffusion calculation according to parameters such as kappa=30, and finally take the value of the center point after the iteration is stable as the compensation value.

[0316] The core of this implementation is to constrain anisotropic diffusion within a very small local window, and to take the center bad point as the repair target, with diffusion flux only coming from reliable pixels.

[0317] Input: algorithm parameters ; ImagePatch of the pixel q to be compensated.

[0318] Output: compensated pixel gray value.

[0319] Specific steps:

[0320] 1. Initialization: copy the gray data in ImagePatch to the working buffer (t=0 represents the 0th iteration). Set the initial value of the center bad point q Set to the median value of the gray levels of all normal pixels in its neighborhood (not the mean value, to increase robustness against outliers).

[0321] 2. Perform custom diffusion iterations (for t = 0 to iterations - 1):

[0322] a. Gradient computation and correction: For each normal pixel p in the window except the center point q, compute its gradient with the four immediate neighbors (up, down, left, right). Key correction: if a neighbor pixel in a certain direction is marked as "invalid", ignore the gradient contribution in that direction (treat the gradient as zero).

[0323] b. Diffusion coefficient computation: For each valid gradient direction d of pixel p, compute the Perona-Malik diffusion coefficient: where kappa = 30 controls the edge sensitivity, denotes the gradient of the image gray level I in a certain direction d.

[0324] c. Pixel value update (non-center point): Update the value of a non-center normal pixel p according to the diffusion equation: where is the step size, and the sum is only over valid gradient directions. denotes the new gray level of pixel p after the t+1th iteration. denotes the new gray level of pixel p after the tth iteration. denotes the diffusion coefficient in direction d.

[0325] d. Special update for the center bad point q: The value update of the center point q is only driven by its surrounding normal neighbors N(q): where is the gradient from normal pixel s pointing to center q, and the diffusion coefficient c is computed from this gradient. This ensures that the repair information only "flows in" from reliable neighbors. denotes the summation operator over the set denotes the diffusion coefficient from normal pixel s pointing to center q. denotes the gradient from normal pixel s pointing to center q during the tth iteration. 3. Output the compensation value: After the pre-set 5 iterations are completed, take the final center point value

[0326] as the compensation value.

[0327] ​The present application converts a partial differential equation method for global image smoothing and edge preservation into a precise point repair tool with directional protection for isolated pixel defects. By strictly controlling the gradient source (only from normal pixels) and the center point update rule, it can effectively fill in bad points while strictly avoiding damage to the original image structure near complex textures or edges.

[0328] Step 5: Pixel value replacement

[0329] The algorithm function returns the calculated compensation value. The system immediately replaces the original gray value at the position of the bad pixel q in the current frame output image with this value.

[0330] 2.3.4, Abnormal processing and degradation strategy

[0331] To ensure system robustness, a backup mechanism is designed:

[0332] Algorithm search failure: If the algorithm identifier in is not found in the registry (for example, the algorithm changes after software upgrade), the system will automatically degrade, call a preset, reliable default algorithm (such as a simple 3x3 mean interpolation) to complete the compensation, and record the log.

[0333] Calculation failure: If an error occurs during algorithm calculation (such as illegal parameters, all invalid pixels in the neighborhood data), the system will adopt a conservative strategy, such as replacing the value with the nearest normal pixel value around the bad pixel in the current frame.

[0334] The system takes out the specified algorithm and parameters from the algorithm registry through pre-learned , and then performs real-time and accurate calculation on the bad pixel neighborhood, finally completing pixel repair. This mechanism enables the compensation strategy to upgrade from simple fixed value replacement to complex, data-driven adaptive image repair, greatly improving the technical upper limit of compensation effect.

[0335] In summary, before the dynamic bad pixel q actually appears as an abnormal bright / dark spot in the image due to reaching the critical point of temperature, the system has already corrected it. The transient defects caused by dynamic bad pixels are completely eliminated in the image observed by the end user, achieving seamless image output.

[0336] Through the cooperative work of S4.1 and S4.2 modules, the present application not only can accurately detect bad pixels, but also can continuously optimize the detection system itself and repair image defects in the most intelligent way, finally ensuring that the infrared detector outputs stable and high-quality images throughout its life cycle.

[0337] To make the specific embodiments of the present application and its technical effects more clear, the following will be described in detail in combination with a specific detector model (a certain type of 640x512 uncooled infrared focal plane detector). This case will fully demonstrate the whole process from image acquisition, bad pixel detection and judgment to adaptive compensation.

[0338] I. System deployment and initialization

[0339] 1. Hardware platform: integrate the software system of the present application method in the embedded image processing board matched with the detector (such as the platform based on TI TDA4VM or Xilinx Zynq series FPGA+ARM architecture).

[0340] 2. Parameter initialization:

[0341] Image parameters: preset frame rate FPS = 20, current image buffer capacity is 1 frame, and image normalization resolution is set to 640x512.

[0342] Background acquisition parameters: temperature acquisition accuracy , temperature change step , background image database capacity N = 10.

[0343] Detection algorithm parameters: wavelet basis is initialized to bior3.5; decomposition scale J = 4; singular point detection coefficient k = 3.0; difference threshold for judgment .

[0344] Compensation parameters: static bad pixel compensation neighborhood window is 5x5; edge direction pixel weight in weighted interpolation , others ; dynamic bad pixel warning temperature difference .

[0345] II. Workflow

[0346] Assume that the detector starts working from the environment temperature of 20.0℃, and the target scene is outdoor park monitoring.

[0347] Step A: image synchronous acquisition

[0348] The system continuously outputs the park scene image (current scene image A) after non-uniformity correction at 20 frames / second.

[0349] At the same time, the temperature sensor monitors the chip temperature. When the temperature rises to (the change reaches ), the shutter closes, and the first frame of uniform background image (correlation temperature ) is collected and stored in the database. The subsequent temperature changes by , that is, it triggers the collection of ​ ], and ( )…, the database always keeps the last 10 frames.

[0350] Step B: Outlier detection (take the current scene image A and background ( ) as an example)

[0351] 1. Wavelet transform: 4-level 2D wavelet transform is performed on A and B_10 respectively, and the high-frequency detail components LH, HL, HH are obtained.

[0352] 2. Adaptive threshold screening: calculate the standard deviation of the non-edge area in the LH component of A , then the threshold T = 3.0 * 4.2 = 12.6. Traverse the LH component, find the local gray maximum value and > 12.6. Perform the same operation on HL and HH components, take the intersection, and preliminarily locate 150 candidate outliers in A.

[0353] 3. Multi-scale verification: for the above 150 points, use 3x3, 5x5, 7x7 windows to verify on the original image. For example, a point (100, 200) is determined to be an outlier in 3 windows, then the confidence C = 1.0, and it is retained; another point (150, 300) is only in the 3x3 window, C = 0.33, and it is rejected. Finally, the effective outlier set of A leaves 138 points.

[0354] 4. Calculate the singularity value: for each point in A , extract its 4-scale wavelet coefficient modulus value, and fit the decay curve to get the singularity value S. Assume that the of point (100, 200) is 0.5.

[0355] Step C: Bad point judgment:

[0356] For any pixel position (x, y) on the detector:

[0357] Scenario one: if (100, 200) is also an effective outlier in A , then enter the second layer judgment. Calculate . Since , it is determined to be a noise point.

[0358] Scenario two: if (150, 250) is an effective outlier in A and B_10, and , , then , it is also determined to be a noise point.​​​​

[0359] Scenario three: if (300, 400) is normal in A, but is an effective singular point in B, then enter the third layer of judgment. Query the historical database to find that this point has also been a singular point in C and D. Therefore, it is judged to be a dynamic bad point, and its activation temperature interval is recorded as (300, 400).

[0360] Scenario four: if a point is an effective singular point in A and B, and

[0361] Step D: adaptive compensation execution

[0362] 1. Static bad point compensation: for pixels judged to be static bad points, first use the Canny operator to determine that they are located in a non-edge region. Call the regular 5x5 neighborhood interpolation, and replace it with the average gray value 125 of the surrounding 24 normal pixels.

[0363] 2. Dynamic bad point pre-compensation: for dynamic bad points (300, 400), the system has learned that it is prone to appear in , and the optimal compensation parameter is "Gaussian weighted interpolation ( )". When the detector temperature rises to 27.5℃ ( , meeting the condition), the system automatically performs Gaussian weighted interpolation pre-compensation on this point in real-time video stream, and the user watches the screen without any abnormal flicker.

[0364] Three, adaptive optimization cases

[0365] After the system runs for a period of time, the accuracy rate decreases to 92% (below the threshold of 95%) in the evaluation period. Analysis finds that the false alarm increases, mainly due to the recent foggy weather causing image noise to increase.

[0366] Calibration action: the system automatically adjusts the singular point detection threshold coefficient k from 3.0 to 3.2, making the threshold T larger and the screening more stringent.

[0367] Using the newly collected 10 frames of verification images after adjustment, the accuracy rate rises to 96%, and the parameter adjustment is successful.

[0368] Four, technical effect verification data

[0369] Continuous testing on this model detector for 48 hours (environmental temperature​​​​​​​​​​ After the cycle, the following data were measured:

[0370] Performance index The method of the present application Traditional single temperature calibration + median filtering method Promotion effect Static bad pixel detection accuracy 98.5% 85% Increased by 13.5 percentage points Dynamic bad pixel detection accuracy 97.2% 70% (and cannot be stably identified) Increased by 27.2 percentage points Bad pixel false detection rate 1.8% 15% Reduced by 13.2 percentage points Edge bad pixel compensation PSNR 42.5 dB 40.7 dB Increased by 1.8 dB Dynamic bad pixel visible time 0 frame 2-3 frames / time Completely eliminated Single-frame full-process processing time ≤ 50ms ≤ 20ms Meet the real-time requirement of 20fps

[0371] The embodiment shows that the method of the application realizes the super-high precision detection and intelligent compensation of static and dynamic bad points without the black body calibration working state for the detector, significantly improves the output image quality and stability of the infrared imaging system, and has important engineering application value.

[0372] The above-described specific embodiments further illustrate the purposes, technical solutions and beneficial effects of the application. It should be understood that the above-described embodiments are merely specific embodiments of the application and are not intended to limit the protection scope of the application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for detecting bad pixels of an infrared detector, characterized in that, Comprise: S1, acquire a current scene image A during working of an infrared detector, and a plurality of frames of isothermal uniform surface background images collected by the infrared detector at different working temperatures to form a background image sequence wherein i is an acquisition serial number and i≥1; The specific process of S1 is: S1.1, collect the detector working scene image according to the preset frame rate, after pixel gray value calibration and image size normalization preprocessing, form the current scene image A and store; S1.2, triggering shutter closing by linkage control mechanism when the working temperature of the detector changes to reach preset temperature step, collecting a single frame of background image and pre-processing, storing to background image database after associating temperature parameter to form the background image sequence ; The background image database adopts a cyclic coverage mechanism, retaining the last N frames of background images, N is a preset positive integer; S2, adopt improved wavelet transform algorithm to detect singular points in each frame image in the current scene image A and the background image sequence respectively, filter out effective singular points and calculate the singular value of each effective singular point; the improved wavelet transform algorithm comprises wavelet base adaptive selection, two-dimensional wavelet transform and high-frequency detail component extraction, adaptive threshold setting and singular point screening, and singular value calculation. S3, based on the current scene image A and the background image sequence The singular point detection result, the singularity value difference and the historical background image data of the corresponding pixel position in the middle are distinguished by a multi-layer judgment rule to obtain a bad point detection result containing a bad point type, a position and a corresponding temperature parameter. The multi-layer decision rule is a three-layer decision rule, based on the effective singular points obtained after step S2 screening and verification, the following three-layer decision rule is executed: First layer decision: if a pixel position is a valid outlier in the current scene image A, and is not a valid outlier in the latest single frame background image in the background image sequence then the pixel position is determined to be a noise point. Second layer judgment: if a pixel position is valid singular point in both the current scene image A and the latest single-frame background image , then calculate the difference of the singular value of the pixel position in the current scene image A and the corresponding singular value in the latest single-frame background image ; set the difference threshold value , if , then determine that the pixel position is a static bad point, otherwise determine that the pixel position is a noise point, wherein is the singular value of the pixel position in the current scene image A, is the corresponding singular value of the pixel position in the single-frame background image .​ Third layer decision: if a pixel position is not a valid outlier in the current scene image A, but is a valid outlier in the latest single-frame background image B, then the detection result of the pixel position in the historical background images C is called. If the pixel position was a valid outlier in any one of the historical background images, it is determined that the pixel position is a dynamic bad pixel, otherwise it is determined that the pixel position is a noise point. ​​​ S4, periodically evaluate the accuracy of the bad pixel detection result, and dynamically calibrate the detection parameters according to the evaluation result, and feed back the bad pixel detection result to the detector control unit.

2. The method of claim 1, wherein the step of detecting the bad pixel is performed by using a method of detecting a bad pixel of an infrared detector, the method comprising the steps of: The preset frame rate is 10-30 frames / second; and / or, ​ The preset temperature step is ; and / or, The collection accuracy of the temperature parameter is ; and / or, 。 3. The method of claim 1, wherein the step of detecting the bad pixel is performed by using a method of detecting a bad pixel of an infrared detector, comprising the steps of: The improved wavelet transform algorithm specifically includes the following sub-steps: ​ S2.1, adaptive selection of wavelet base: select biorthogonal wavelet as wavelet base, and dynamically determine 3-5 level wavelet decomposition scale according to the gray mean and variance of the image; S2.2, two-dimensional wavelet transform and detail component extraction: two-dimensional wavelet transform is performed on the image, and a low-frequency detail component LL and three high-frequency detail components LH, HL and HH are obtained by decomposition; S2.3, adaptive threshold setting and outlier screening: calculate the gray scale standard deviation of non-edge region in each high-frequency detail component As a noise intensity indicator, an adaptive threshold is set wherein k is an empirical coefficient of 2.5-3.5; using a first neighborhood window to traverse each high-frequency detail component, screening out candidate outliers whose center pixel gray scale value of the first neighborhood window is a local maximum and exceeds the corresponding threshold T, and taking the intersection of candidate outliers in the three high-frequency detail components as effective outliers, thereby obtaining an effective outlier set; S2.4, singular value calculation: extract the wavelet coefficient modulus value at the position of the effective singular point under different wavelet decomposition scales, obtain the decay index by fitting the decay curve of the modulus value with the scale, and take the decay index as the singular value.

4. The method for detecting defective pixels in an infrared detector according to claim 3, characterized in that, After step S2 obtains the effective singular points, before step S3 is executed, it further includes: For each of the effective singular points, the effective singular point is taken as a to-be-verified point; At least two second neighborhood windows of different sizes are used to perform secondary singular point judgment at the position of the to-be-verified point; The number of times that the to-be-verified point is judged as a singular point in all second neighborhood windows is counted, and the confidence value C of the to-be-verified point is calculated according to the number of times; The confidence value C is compared with a preset confidence threshold value C0, if the to-be-verified point is determined to pass the verification, and the to-be-verified point is kept as a final valid singular point; if the to-be-verified point is determined to fail the verification, and the to-be-verified point is removed from the valid singular point set.

5. The method for detecting defective pixels in an infrared detector according to claim 1, characterized in that, the difference threshold value is in the range of 0.1-0.

3.

6. The method of claim 1, wherein the step of detecting the bad pixel is performed by using a method of detecting a bad pixel of an infrared detector, the method comprising the steps of: The specific process of S4 is: ​ S4.1, parameter adaptive calibration: S4.1.1, periodically count the accuracy of the bad pixel detection result, and compare the accuracy with a preset accuracy threshold; S4.1.2, if the accuracy is lower than the preset accuracy threshold, then according to the deviation direction and degree of the accuracy, the wavelet decomposition scale, the singular point detection threshold coefficient k, the bad point judgment difference threshold , the background image acquisition temperature step at least one parameter is adjusted. S4.1.3, use the newly collected verification image of preset number M frames after parameter adjustment to re-execute steps S2 to S3, calculate a new accuracy to verify the parameter adjustment effect, until the accuracy reaches or exceeds the preset accuracy threshold, M is a positive integer; S4.2, adaptive compensation of bad pixels: According to the type, position information and corresponding temperature parameter in the bad pixel detection result, the corresponding compensation strategy is called to perform pixel compensation.

7. The method for detecting defective pixels in an infrared detector according to claim 6, characterized in that, In S4.2, the corresponding compensation strategy is called to perform pixel compensation, specifically including: S4.2.1, compensation strategy initialization: Pre-store a differentiated compensation strategy library associated with bad pixel types and image regions; the strategy library includes: a regular neighborhood interpolation strategy for non-edge area static bad pixels, a weighted neighborhood interpolation strategy for edge area static bad pixels, and a temperature prediction based pre-compensation strategy for dynamic bad pixels; S4.2.2, static bad pixel compensation execution: For the pixel of the type of static bad pixel, the image region where the static bad pixel is located is determined; if located in a non-edge region, the regular neighborhood interpolation strategy is called to take the gray scale values of all normal pixels in the preset compensation neighborhood window of the static bad pixel as the center of the static bad pixel, and interpolation calculation is performed to obtain a compensation value; if located in an edge region, the weighted neighborhood interpolation strategy is called to take the gray scale values of pixels in the preset compensation neighborhood window of the static bad pixel as the center, and interpolation calculation is performed in combination with a preset weight coefficient to obtain a compensation value, wherein pixels belonging to an edge direction are given a higher weight; S4.2.3, dynamic bad pixel compensation execution: For the pixel of the type of dynamic bad point, the pre-compensation strategy is called, a temperature-bad point mapping relationship based on historical data is queried according to a temperature parameter associated with the dynamic bad point; when an absolute value of a difference between a real-time working temperature of the detector and a temperature threshold value, at which the dynamic bad point is prone to appear, indicated in the mapping relationship is less than or equal to a preset early warning temperature difference , a compensation parameter preset for the temperature threshold value is automatically called to pre-compensate the pixel of the dynamic bad point.

8. The method of claim 7, wherein the step of detecting the bad pixel is performed by using a method of detecting a bad pixel of an infrared detector, comprising the steps of: The regular neighborhood interpolation strategy is: ​ Taking the static bad pixel to be compensated as the center, the gray scale values of all normal pixels in the preset compensation neighborhood window of the static bad pixel are taken, the arithmetic mean value of the gray scale values is calculated, and the arithmetic mean value is taken as a compensation value to replace the original gray scale value of the bad pixel; The weighted neighborhood interpolation strategy is: Taking the static bad pixel to be compensated as the center, the pixels in the preset compensation neighborhood window of the static bad pixel are taken; After a weight coefficient is given to the gray scale values of the pixels in the compensation neighborhood window, a weighted average value is calculated as a compensation value; wherein the weight coefficient assignment rule is that the adjacent pixels in the edge tangent direction of the static bad pixel are given a weight coefficient higher than that of the pixels in other directions; The pre-compensation strategy is: According to the temperature threshold value where the dynamic bad pixel is prone to appear obtained by querying, compensation parameters verified effective at the temperature threshold value are called from historical compensation records; the compensation parameters at least include compensation neighborhood window size, weight coefficient distribution or direct optimal compensation value; when the pre-compensation trigger condition is met, the compensation parameters are directly applied to replace the gray scale value of the pixel of the dynamic bad pixel.

Images