A detector array anomaly data repair method, system, and program product

Abstract

This invention discloses a method, system, and program product for repairing abnormal data in a detector array, relating to the fields of image processing and optical imaging computation. Addressing the problem of persistent missing measurements due to permanent defects or bad regions in a target spectrometer array within a measurement sequence, this invention utilizes temporal context information to reconstruct the missing measurement region in the center frame. Compared to single-frame interpolation or local neighborhood completion methods, it exhibits higher reconstruction accuracy and better temporal stability under conditions of larger missing measurement areas. Furthermore, this invention employs an observation hard fusion mechanism based on a permanent failure mask during the inference stage, ensuring that the effective observed pixels strictly maintain their original measurement values ​​in the repair result. This reduces the risk of additional offsets to the actual measurement data during the repair process, thereby improving the reliability of the measurement results.

Description

Technical Field

[0001] This invention relates to the field of image processing and optical imaging computing, and in particular to a method, system, and program product for repairing abnormal data of a detector array. Background Technology

[0002] The detector array target is the core component of laser parameter metrology, responsible for converting the spatial and temporal intensity distribution of incident laser light into calculable data. The arrayed photodetectors directly sample the light spot on the target surface. After signal amplification and analog-to-digital conversion, key indicators such as total power, spot centroid, equivalent diameter, and beam mass factor β can be stably calculated. Compared to scanning or single-point measurements, the array target can cover the entire light spot at once, improving timeliness and reducing uncertainties caused by collimation errors and scene changes. This makes it a fundamental platform for far-field spot evaluation, performance criteria for laser emission systems, and research on atmospheric transmission effects.

[0003] Engineering implementations typically employ a hierarchical readout architecture to balance channel size and throughput. Detectors on the target surface operate independently; if one component fails, the local impact is limited to a small area. However, if the analog switches or subsequent circuits corresponding to a group of channels malfunction, the impact amplifies hierarchically, forming groups or blocks of invalid pixels. Under the combined effects of high-energy irradiation, thermal load accumulation, and device aging, continuous bad regions may also appear, coinciding with high-gradient areas of the light spot. Invalid pixels not only create gaps in the intensity map but also systematically distort parameter estimations, manifesting as total power deviation, centroid drift, and unstable aperture and β estimations. Furthermore, the errors vary with time and operating conditions, exhibiting significant non-stationarity.

[0004] Existing processing methods mainly fall into two categories. The first is a lightweight method based on a pre-calibrated bad pixel table and neighborhood interpolation, relying on mean, bilinear, or spline methods to fill in missing pixels. Its advantages include simplicity and low latency, but it is prone to blurring and energy leakage near strong edges and peaks. The repaired centroid and peak positions may shift considerably, and its adaptability to bad regions evolving over time is insufficient. The second method borrows from general image inpainting and completion techniques, including iso-illuminance line extension for small-scale defects and structure-texture decomposition or block-based texture synthesis for large-area defects. These methods perform well on natural images, but suffer from prior mismatch and lack of constraints when transferred to laser measurement data. Laser spot repair should maintain approximate energy conservation and peak stability while preserving geometric contours and edge sharpness; measurement noise and quantization errors should also be included in the constraints. Conventional methods ignore these physical and system priors, and coupled with limitations in real-time performance and edge computing resources, they are often difficult to implement stably in engineering scenarios.

[0005] Based on the above situation, there is an urgent need for an adaptive repair technology for abnormal data of detector arrays. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to provide a method, system, and program product for repairing abnormal data in a detector array, thereby improving the accuracy and stability of far-field parameter estimation.

[0007] The first aspect discloses a method for repairing abnormal data in a detector array, the method comprising:

[0008] The measurement data sequence output by the target spot array detector and the corresponding permanent failure mask are obtained. The measurement data sequence includes multiple consecutive time points, and each time point corresponds to one frame of array data. The permanent failure mask represents the location of the permanently failed pixels and has the same dimension as the detector output array. For any time point to be repaired, array data frames from k adjacent time points are selected and, together with the center frame corresponding to the time point to be repaired and the corresponding permanent failure mask, constitute a context input sequence, where k is an integer greater than 1. The context input sequence is input into a pre-trained mask-aware reconstruction network to repair the center frame and obtain the center frame prediction result. Based on the permanent failure mask, the prediction result of the center frame is hard-fused with the original measurement array data to obtain the repaired center frame array data.

[0009] The second aspect discloses a detector array abnormal data repair system, the system comprising:

[0010] The module includes an acquisition module for acquiring the measurement data sequence and the corresponding permanent failure mask output by the target speckle array detector. The measurement data sequence includes multiple consecutive time points, with each time point corresponding to one frame of array data. The permanent failure mask represents the location of the permanently failed, missing pixel and has the same dimension as the detector output array. A context input sequence construction module is used to select the array data frames of k adjacent time points for any time point to be repaired. These frames, together with the center frame corresponding to the time point to be repaired and the corresponding permanent failure mask, constitute a context input sequence, where k is an integer greater than 1. A mask-aware reconstruction network module is used to input the context input sequence into a pre-trained mask-aware reconstruction network to repair the center frame and obtain a center frame prediction result. A hard fusion module is used to perform hard fusion of the center frame prediction result with the original measurement array data based on the permanent failure mask to obtain the repaired center frame array data.

[0011] The third aspect discloses a computer program product that, when run on a computer, causes the computer to perform the detector array abnormal data repair method disclosed in the first aspect or any possible implementation of the first aspect.

[0012] As can be seen from the above technical solutions, the present invention has the following beneficial effects:

[0013] This invention addresses the problem of persistent missing measurements in target spot arrays where permanent bad spots or bad regions exist within the measurement sequence. It utilizes temporal context information to reconstruct the missing measurement region in the center frame. Compared to single-frame interpolation or local neighborhood completion methods, it exhibits higher reconstruction accuracy and better temporal stability under conditions of larger missing measurements.

[0014] Furthermore, this invention employs an observation hard fusion mechanism based on a permanent failure mask during the inference phase, ensuring that the effective observed pixels strictly maintain their original measurement values ​​in the repair results. This reduces the risk of additional offsets to the actual measurement data caused by the repair process, thereby improving the reliability of the measurement results. Further, by introducing total ADU consistency constraints and centroid consistency constraints, this invention reduces the systematic impact of the repair process on energy estimation and spot position determination, improving the metrological consistency and engineering applicability of the repair results in high-energy laser diagnostic tasks. This invention can consistently map the repaired array data to the continuous spot reconstruction domain and output quality assessment indicators and quality reports, thus supporting online monitoring, threshold alarms, and closed-loop verification of the diagnostic link, demonstrating significant engineering deployment value. Attached Figure Description

[0015] Figure 1 The present invention provides an overall flowchart of a method for repairing abnormal data in a detector array.

[0016] Figure 2 This is a schematic diagram of a mask that has permanently failed.

[0017] Figure 3 This is a schematic diagram of the hard fusion process.

[0018] Figure 4 This is a comparison image showing the effect of light spot before and after restoration.

[0019] Figure 5 This is a schematic diagram of the architecture of a detector array abnormal data repair system provided by the present invention. Detailed Implementation

[0020] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be thorough and complete.

[0021] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly on the other element or there may be an intervening element. When an element is considered to be "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," "up," "down," and similar expressions used herein are for illustrative purposes only and are not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention.

[0023] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances. The term "and / or" as used herein includes any and all combinations of one or more of the related listed items.

[0024] Existing speckle array detectors may experience permanent measurement omissions such as bad pixels, bad areas, entire row failures, or entire column failures during long-term operation or due to hardware damage. Within a single measurement sequence, the location of these failures typically remains unchanged, resulting in a continuous lack of valid observations for the corresponding pixel location throughout the sequence. For this type of permanent measurement omission problem, traditional single-frame interpolation, neighborhood completion, or general image inpainting methods are prone to causing the reconstructed results to deviate from the true measurement distribution when the missing area is large. This leads to problems such as total ADU (Analog-to-Digital Unit) shift, centroid drift, peak position deviation, and timing jitter, making it difficult to meet the metrological consistency and engineering stability requirements of high-energy laser diagnostic scenarios.

[0025] The purpose of this invention is to provide a method, system, and program product for repairing abnormal data of a detector array, in order to solve the technical problem that existing technologies cannot simultaneously guarantee the fidelity of effective observations, the consistency of key diagnostic quantities, and the stability of time-series reconstruction when permanent bad spots or bad areas persist and the missing range is large. This improves the metrological reliability, engineering availability, and online deployment capability of target spot array data repair.

[0026] In one embodiment, the present invention provides a method for repairing abnormal data in a detector array, such as... Figure 1 As shown, the specific steps include:

[0027] S101. Obtain the measurement data sequence and the corresponding permanent failure mask output by the target spot array detector. The measurement data sequence includes multiple consecutive time moments, and each time moment corresponds to one frame of array data. The permanent failure mask represents the position of the permanently failed pixel. The permanent failure mask has the same dimension as the detector output array.

[0028] Specifically, the object processed in this embodiment is the time-series measurement data output by the speckle array detector. During high-energy laser atmospheric transmission, disturbance propagation, or experimental diagnosis, the speckle array samples the beam irradiance distribution at continuous moments, with each moment corresponding to a frame of two-dimensional array intensity matrix.

[0029] Suppose the target spot array time-series measurement data sequence is as follows:

[0030] (1)

[0031] in, Indicates time array data matrix, and These represent the array height and array width, respectively. Indicates the length of the measurement sequence.

[0032] Due to detector aging, device damage, readout link failure, or local hardware anomalies, some pixels in the array may lose their effective response capability, forming bad pixels, bad areas, or entire rows or columns of failure. Furthermore, the location of such failures typically remains unchanged within the same measurement sequence. Therefore, a permanent failure mask is defined as:

[0033] (2)

[0034] in, This indicates that the location is a valid observation pixel. This indicates that the pixel at this location is a permanently failed, missing pixel. An index representing the array height. The index represents the array width. The permanent failure mask is a pre-generated binary mask for permanent failures, constructed based on multiple frames of data collected by the detector array during the calibration or historical operation phases. Specifically, based on dark field response, uniform illumination response, historical measurement sequences, or equipment self-test results, statistical analysis can be performed on pixel response amplitude, temporal stability, neighborhood bias, and connectivity anomalies to identify long-term stable failure pixels or failure regions, and generate a binary mask corresponding to the spatial dimensions of the detector array, such as... Figure 2 As shown. Since this failure state remains stable within the same sequence, the same mask can be used for data repair at all times within that sequence.

[0035] This invention takes the array timing measurement data sequence output by the target spot instrument and the permanent failure mask as input. During the online inference process, the permanent failure mask is directly read as a priori input to guide the repair of missing measurement areas and subsequent observation consistency fusion, and outputs the repaired complete array sequence.

[0036] S102. For any time to be repaired, select the array data frames of the k adjacent times before and after it, and together with the center frame corresponding to the time to be repaired and the corresponding permanent failure mask, form a context input sequence, where k is an integer greater than 1.

[0037] Specifically, for any time to be repaired , in time The corresponding center frame is used as the reconstruction target, and the construction length is [missing information]. time window :

[0038] (3)

[0039] in, The time window is half the width.

[0040] Time window The array data of each frame is stacked according to the channel dimension and together with the permanent failure mask, they form the context input sequence. :

[0041] (4)

[0042] It should be noted that when constructing the context input sequence, the time to be repaired is the time that meets the construction time window. For the time that does not meet the construction time window, the array data of the time that cannot be repaired can be deleted. Since the time that cannot be repaired in the measurement data sequence is a minority, it will not affect the subsequent calculation of mean absolute error, mean square error of missing measurement area, relative error of total ADU, centroid offset, peak position deviation and / or inter-frame temporal stability index.

[0043] In one embodiment, the invention further includes:

[0044] In the presence of overall inter-frame drift, the observation centroid of each frame array is calculated within the effective observation area, which is the array area after removing the missing measurement area;

[0045] Based on the observed centroid of each frame, each frame is translated to a position approximately aligned with the observed centroid of the central frame.

[0046] Specifically, under actual operating conditions, the target spot may experience overall inter-frame drift due to atmospheric disturbances, platform jitter, or changes in pointing. To reduce the adverse effects of this type of overall translation on temporal context fusion, the observation centroid of each frame can be calculated using the effective observation area, and coarse alignment can be performed on each frame within the time window.

[0047] For a certain frame The centroid of the effective observation region can be represented as:

[0048] (5)

[0049] (6)

[0050] Each frame is shifted to a position approximately aligned with the centroid of the observation in the center frame. Coarse alignment is only used to enhance the efficiency of temporal context utilization and does not change the mechanism for retaining the original valid observations in the final output.

[0051] Through this invention, in the case of overall inter-frame drift, the observation centroid of each frame array can be calculated based on the effective observation area, and coarse alignment of the array domain can be performed on each frame within the time window to reduce the adverse effects of overall translation on time-series fusion.

[0052] S103. Input the context input sequence into the pre-trained mask perception reconstruction network to repair the center frame and obtain the center frame prediction result.

[0053] Specifically, in the network structure implementation stage, this embodiment uses a mask-aware reconstruction network to reconstruct the missing measurement region in the center frame. The mask-aware reconstruction network can be a U-Net-like structure, or other encoder-decoder structures with multi-scale feature extraction and upsampling recovery capabilities.

[0054] Input context sequence The input is a mask-aware reconstruction network, and the output is the center frame prediction result, as shown in the formula:

[0055] (7)

[0056] in, The parameter is Mask perception reconstruction network, This indicates the prediction result for the center frame.

[0057] In another embodiment, the training process of the mask-aware reconstruction network includes:

[0058] S1031. Construct a training sample set, which includes multiple array data frame sequences. Each array data frame sequence is obtained by injecting noise into clean array data samples according to the detector statistical model and superimposing quantization operators. Each array data frame sequence corresponds to a permanent failure mask, and the permanent failure masks corresponding to different array data frame sequences may be the same or different.

[0059] Specifically, array timing measurement data of the target spot instrument are collected under different exposure parameters, gain parameters, temperature conditions, and typical atmospheric transmission conditions. For each measurement sequence, the device configuration and acquisition conditions are recorded, and historical statistical information or calibration information of bad spots, bad areas, and entire row or column failures are obtained. Based on this, a permanent failure mask is generated. The permanent failure mask can generate various defect patterns such as point, block, row, and column based on historical statistics, and remain constant within the same sequence.

[0060] When there are no fault-free array samples, or when approximately clean samples can be obtained by removing local abnormal frames and faulty pixels, noise and quantization processes consistent with the detector readout mechanism can be injected into the clean samples based on the detector's statistical characteristics. Then, point-like, block-like, row-level, or column-level permanent failure masks can be superimposed to construct a training dataset combining measured and synthetic samples. The training, validation, and test datasets are preferably divided according to the measurement sequence level to avoid evaluation bias caused by adjacent frames within the same measurement sequence being placed in different datasets.

[0061] For example, let the clean sample be... Then synthesize the sample It can be generated using the following formula:

[0062] (8)

[0063] in, This represents the process of Poisson noise. This indicates Gaussian readout noise. This represents the quantization operator. A permanent failure mask is then superimposed on the synthetic samples to form the training input.

[0064] In constructing training samples, this invention injects Poisson and Gaussian noise into clean array samples according to the detector statistical model and superimposes quantization operators, ensuring that synthetic and measured samples share a unified dynamic range and normalization rules. The permanent failure mask generates point-like, block-like, row-level, or column-level missing measurement patterns based on historical failure statistics and remains constant within the same measurement sequence. Furthermore, the measured and synthetic samples undergo unified preprocessing: bias subtraction, background correction, dynamic range clipping, and amplitude normalization to unify the data format and dynamic range of samples under different operating conditions. Samples with abnormal exposure, obvious saturation, or strong nonlinear response are screened or labeled to improve the stability of the training process and the model's adaptability under multiple operating conditions.

[0065] S1032. For each array data frame sequence, construct an input tensor. The input tensor consists of the center frame corresponding to any time to be repaired, the array data frames at k adjacent times before and after the center frame, and the corresponding permanent failure mask.

[0066] S1033. Reconstruct the input tensor input mask perception reconstruction network to obtain the prediction result;

[0067] Specifically, it includes:

[0068] The mask-aware coding module extracts multi-scale spatial features of the center frame and the array data frames at k adjacent time points, and suppresses the interference of missing measurement regions on feature statistics through explicit mask input and gated convolution mechanism.

[0069] The temporal feature fusion module is used to perform temporal correlation modeling on the coding features at each time step in order to restore the temporal continuity of the missing measurement area in the center frame;

[0070] The spatial resolution is restored step by step by the feature decoding module, and the high-resolution feature representation corresponding to the center frame is obtained by combining the shallow detail features with the skip connection.

[0071] The center frame prediction result is obtained by performing convolutional mapping through the output reconstruction module.

[0072] Specifically, the mask-aware reconstruction network includes a mask-aware coding module, a temporal feature fusion module, a feature decoding module, and an output reconstruction module.

[0073] The network consists of a mask-aware reconstruction network that receives array measurement data from the center frame and its adjacent time points, and combines this data with an input tensor constructed from a permanently failed mask. A mask-aware coding module extracts multi-scale spatial features and suppresses interference from anomalous regions on feature statistics through explicit mask input, gated convolution, and explicit missing measurement indication propagation mechanisms. A temporal feature fusion module performs temporal correlation modeling on the coded features from multiple time points to restore the temporal continuity of the missing measurement region in the center frame. A feature decoding module progressively restores spatial resolution and fuses shallow detail features using skip connections to obtain a high-resolution feature representation corresponding to the center frame. An output reconstruction module obtains the center frame prediction result through convolutional mapping.

[0074] Furthermore, the mask-aware mechanism can be implemented in the following ways: a permanently failed mask is used as an independent explicit input channel; a gated convolution mechanism is adopted in the convolution module. Specifically, during feature extraction and fusion, the gate function for the missing measurement region is set to 0, and the gate function for the effective observation region is set to 1, thereby effectively suppressing the interference of missing measurement values ​​on feature statistics during feature extraction and feature fusion.

[0075] To reduce the interference of missing values ​​on the feature extraction process, this invention employs a mask-aware mechanism, specifically treating a permanently failed mask as an independent input channel. In local convolution operations, the effective region and the missing region are distinguished based on the permanently failed mask. In the multi-scale feature fusion process, the missing region is gated and suppressed. In the encoding and decoding stages, the missing indicator information is explicitly transmitted.

[0076] In the continuous domain reconstruction and quality assessment implementation phase, this embodiment requires the network not only to recover pixel values ​​in the missing test areas during the training phase, but also to ensure that its output meets the consistency of observation and key diagnostic quantities, thereby reducing the systematic bias introduced by the repair.

[0077] S1034. Construct a target loss function based on the prediction results, wherein the target loss function is obtained based on the missing region reconstruction loss, observation consistency loss, energy consistency loss and centroid consistency loss;

[0078] Specifically, this invention uses the missing region reconstruction loss as the main supervision target, and constructs a composite loss function by combining observation consistency loss and diagnostic consistency loss to train the mask perception reconstruction network.

[0079] Let the target truth array corresponding to the center frame be... The reconstruction loss in the missing measurement area This can be represented as shown in the formula:

[0080] (9)

[0081] in, This represents element-wise multiplication. Indicates a permanently failed mask. This represents the prediction result for the center frame. Missing region reconstruction is used to drive the network to focus on learning pixel recovery in the failed regions.

[0082] To suppress unnecessary shifts in the network's effective observation area, an observation consistency loss is introduced within the effective observation area. Its specific definition is shown in the formula:

[0083] (10)

[0084] in, This represents element-wise multiplication. Indicates a permanently failed mask. This indicates the center frame prediction result. This represents the original center frame array data. This loss is used to guide the network to maintain consistency with the true observations during the training phase, i.e., to suppress the network's output offset from the effective observation area.

[0085] The total ADU is defined in the formula shown below:

[0086] (11)

[0087] in, Indicates the position at time t. The array response value at that location, express The corresponding total ADU.

[0088] Energy consistency loss The specific definition is shown in the formula:

[0089] (12)

[0090] in, This represents the total ADU corresponding to the center frame prediction result. This represents the total ADU corresponding to the target ground truth array of the center frame; this loss is used to reduce the total energy statistical bias after reconstruction and reduce the systematic impact of the repair process on the energy proxy estimation.

[0091] The centroid coordinates are defined as follows:

[0092] (13)

[0093] (14)

[0094] in, , Represent the x and y coordinates of the centroid;

[0095] Centroid Consistency Loss The specific definition is shown in the formula:

[0096] (15)

[0097] in, , This represents the x and y coordinates of the centroid of the center frame prediction result. , The x and y coordinates represent the centroid of the target ground truth array corresponding to the center frame; this loss is used to reduce the additional errors caused by the reconstruction results in the analysis of the spot center position, pointing and drift judgment.

[0098] Therefore, the target loss function is defined as shown in the formula:

[0099] (16)

[0100] in, These are the weighting coefficients for each loss term. This indicates the reconstruction loss in the missing measurement area. This indicates the loss of observation consistency. Indicates energy consistency loss, This represents the loss of centroid consistency. In a preferred set of embodiments, it can be taken as... , , , .

[0101] In one embodiment, in the case of anomalies where saturated pixels exist, a saturation mask is constructed for the location of the abnormal pixels;

[0102] Based on the saturation mask, calculate the energy consistency loss and centroid consistency loss between the center frame prediction result and the target ground truth array data.

[0103] Specifically, in the presence of saturated pixels, a saturation mask can be introduced. The contributions of relevant pixels in the loss function calculation and diagnostic statistics are eliminated or downweighted to avoid additional bias caused by erroneous conservation constraints. Let the saturation mask be:

[0104] (17)

[0105] in, Indicates position For saturated pixels, Indicates position These are unsaturated pixels.

[0106] For example, the total ADU statistic can be rewritten as:

[0107] (18)

[0108] Correspondingly, centroid statistics are only performed in the unsaturated pixel region.

[0109] S1035. The mask-aware reconstruction network is iteratively optimized according to the target loss function to obtain a pre-trained mask-aware reconstruction network.

[0110] Specifically, the network training cutoff condition can be at least one of the following: 1. The number of training epochs reaches a preset maximum value; 2. The composite loss on the validation set decreases by less than a preset threshold within a consecutive preset number of epochs; 3. The diagnostic error on the validation set no longer improves within a consecutive preset number of epochs; 4. The network parameter update magnitude is lower than a preset optimization threshold. Preferably, when the relative decrease in the composite loss on the validation set is less than 10 over 10 to 30 consecutive training epochs... -4 Up to 10 -3 Alternatively, when the total number of training rounds reaches 100 to 300, training is terminated and the model parameters with the best performance on the validation set are retained, thus obtaining the pre-trained mask-aware reconstruction network.

[0111] S104. Based on the permanent failure mask, perform hard fusion of the prediction result of the center frame with the original measurement array data to obtain the repaired center frame array data.

[0112] In the reasoning stage, such as Figure 3 As shown, the center frame prediction results and the original measurement array data are hard-fused using a permanent failure mask to obtain the repaired center frame array data. The specific formula is as follows:

[0113] (19)

[0114] This method ensures that all valid observation pixels strictly maintain their original measurements in the repair results, and only uses network prediction results to fill in the missing measurements in permanently failed areas, thereby eliminating the risk of additional offsets caused by the network to the valid observations at the output end.

[0115] The goal of this embodiment is to determine the original array sequence. and permanent failure mask Obtain the repaired complete array sequence The specific formula is as follows:

[0116] (20)

[0117] in, This represents the repaired center frame array data. For example... Figure 4 As shown, the present invention repairs damaged light spots, resulting in a repaired complete light spot.

[0118] In one embodiment, the invention further includes:

[0119] S105. The repaired center frame array data is mapped into a continuous spot image through a preset upsampling reconstruction operator;

[0120] S106. Calculate the mean absolute error of the missing area, the mean square error of the missing area, the relative error of the total ADU, the centroid shift, the peak position deviation and / or the inter-frame temporal stability index based on the continuous spot image.

[0121] S107. Generate a quality assessment report based on the aforementioned indicators for online monitoring and threshold alarms.

[0122] Specifically, the repaired complete array data The image is mapped to a continuous light spot image using a preset upsampling reconstruction operator, which is then used for visualization and reconstruction domain quality assessment. The upsampling reconstruction operator can be bicubic interpolation, spline interpolation, radial basis function reconstruction, or a reconstruction method based on the system calibration point spread function or kernel function.

[0123] The quality indicators of the array domain and the reconstructed domain are calculated based on the repair results, and a quality assessment report is generated for online operation monitoring, threshold alarms, and diagnostic link consistency verification.

[0124] In the array domain, the following metrics can be calculated for the repair results: mean absolute error of the missing region; mean square error of the missing region; relative error of total ADU; centroid shift; peak position deviation; and inter-frame temporal stability metrics. Among these, the relative error of total ADU... Defined as:

[0125] (twenty one)

[0126] in, This represents the total ADU corresponding to the target truth array of the center frame; This indicates the repaired center frame array data. The corresponding total ADU.

[0127] In addition, the centroid shift Defined as:

[0128] (twenty two)

[0129] in, , This indicates the repaired center frame array data. The x and y coordinates of the centroid; , Represents the target truth array corresponding to the center frame. The x and y coordinates of the centroid.

[0130] In addition, temporal stability can be evaluated by the difference in reconstruction between adjacent frames, the variance of diagnostic fluctuations, or by sliding window statistics.

[0131] In one alternative implementation, the mask-aware reconstruction network employs a shared backbone and dual-output branch structure; specifically, in the input tensor... After processing by the mask-aware coding module, the temporal feature fusion module, and the feature decoding module, a high-resolution feature representation corresponding to the center frame is obtained. Based on this, the first output branch obtains the prediction result of the center frame through convolutional mapping. The second output branch obtains a pixel-level uncertainty map with the same spatial size as the center frame through convolution mapping. The specific formula is as follows:

[0132] (twenty three)

[0133] in, This represents the pixel-level uncertainty map corresponding to the center frame prediction result. The pixel-level uncertainty map can represent pixel-level variance, logarithmic variance, or other non-negative reliability characteristics. Used to characterize the reliability of the prediction results for each pixel location, where, The larger the value, the higher the uncertainty of the repair result at the corresponding location.

[0134] Specifically, the steps include the following:

[0135] When the mask-aware reconstruction network outputs a pixel-level uncertainty map, the pixel-level uncertainty map is used to characterize the reliability of the prediction results at each pixel position. The larger the value of each pixel position, the higher the uncertainty of the corresponding position repair result.

[0136] If the pixel value of the uncertainty map of the untested region is greater than a preset threshold, the region is marked as a high-risk reconstruction region, and at least one preset enhancement strategy is executed to optimize the inference process.

[0137] It should be noted that when the mask-aware reconstruction network outputs a pixel-level uncertainty map... In such cases, the missing data areas can be classified according to the magnitude of uncertainty. Let the preset threshold be... When a certain missing measurement location satisfies: .

[0138] The location is then marked as a high-risk reconstruction area, and at least one pre-defined enhancement strategy is executed: increasing the half-width k of the time window or invoking the mask-aware reconstruction network for secondary inference. At the same time, maintain the normal rapid inference process. This approach helps improve the robustness of repairing high-risk missing test areas while ensuring that the overall processing latency is controllable.

[0139] In one implementation, the pixel-level uncertainty map is normalized to [0,1], and a preset threshold is used. Set the threshold to 0.5. Furthermore, the preset threshold can also be adjusted within the range of 0.4 to 0.7 based on the statistical results of the validation set, in order to balance the coverage and reliability requirements of the repair results.

[0140] This invention outputs a pixel-level uncertainty map and performs adaptive post-processing on the missing regions based on the magnitude of uncertainty. For high uncertainty regions, the robustness of inference can be improved by increasing the time window or performing secondary inference; for low uncertainty regions, a fast inference strategy is maintained to reduce the overall processing latency.

[0141] This invention addresses the problem of persistent missing measurements in target spot arrays where permanent bad spots or bad regions exist within the measurement sequence. It utilizes temporal context information to reconstruct the missing measurement region in the center frame. Compared to single-frame interpolation or local neighborhood completion methods, it exhibits higher reconstruction accuracy and better temporal stability under conditions of larger missing measurements.

[0142] Furthermore, this invention employs an observation hard fusion mechanism based on a permanent failure mask during the inference phase, ensuring that the effective observed pixels strictly maintain their original measurement values ​​in the repair results. This reduces the risk of additional offsets to the actual measurement data caused by the repair process, thereby improving the reliability of the measurement results. Further, by introducing total ADU consistency constraints and centroid consistency constraints, this invention reduces the systematic impact of the repair process on energy estimation and spot position determination, improving the metrological consistency and engineering applicability of the repair results in high-energy laser diagnostic tasks. This invention can consistently map the repaired array data to the continuous spot reconstruction domain and output quality assessment indicators and quality reports, thus supporting online monitoring, threshold alarms, and closed-loop verification of the diagnostic link, demonstrating significant engineering deployment value.

[0143] This application also provides a repair system corresponding to the method embodiments described above. Since the system embodiments are basically similar to the method embodiments, the description is relatively simple. For details of the relevant technical features and their effects, please refer to the corresponding descriptions of the method embodiments provided above. The present invention provides a detector array abnormal data repair system, such as... Figure 5 As shown, this system specifically includes:

[0144] The acquisition module is used to acquire the measurement data sequence and the corresponding permanent failure mask output by the target spot array detector. The measurement data sequence includes multiple consecutive time points, and each time point corresponds to one frame of array data. The permanent failure mask represents the position of the permanently failed pixel and has the same dimension as the detector output array.

[0145] The context input sequence construction module is used to select the array data frames of the k adjacent times before and after any time to be repaired, and together with the center frame corresponding to the time to be repaired and the corresponding permanent failure mask, they form the context input sequence, where k is an integer greater than 1.

[0146] The mask-aware reconstruction network module is used to input the context input sequence into the pre-trained mask-aware reconstruction network to repair the center frame and obtain the center frame prediction result.

[0147] The hard fusion module is used to perform hard fusion of the prediction result of the center frame with the original measurement array data based on the permanent failure mask, so as to obtain the repaired center frame array data.

[0148] This application also provides an electronic device, which includes a processor and a memory. The memory stores at least one instruction or at least one program, which is loaded by the processor and executed to perform the detector array abnormal data repair method provided in the above method embodiments.

[0149] Furthermore, the electronic device may participate in or include the apparatus or system provided in the embodiments of this application. The electronic device may include one or more processors (processors may include, but are not limited to, processing devices such as microprocessors (MCUs) or programmable logic devices (FPGAs), memory for storing data, and transmission devices for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera.

[0150] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits can be implemented wholly or partially as software, hardware, firmware, or any other combination. Furthermore, the data processing circuits can be a single, independent processing module, or wholly or partially integrated into any other element within a device (or mobile device). As involved in the embodiments of this application, the data processing circuit serves as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0151] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method described in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the above-mentioned data processing method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to electronic devices via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0152] The transmission device is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the device's communication provider. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0153] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows users to interact with the user interface of an electronic device (or mobile device).

[0154] This application also provides a computer storage medium storing at least one instruction or at least one program, which is loaded and executed by a processor to implement the detector array abnormal data repair method provided in the above method embodiments.

[0155] Optionally, in this embodiment, the aforementioned computer storage medium may be located at at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the aforementioned storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0156] This application also provides a computer program product or computer program, which includes computer instructions stored in a computer storage medium. The processor of an electronic device reads the computer instructions from the computer storage medium and executes the computer instructions, causing the electronic device to perform the detector array abnormal data repair method provided in the above-described method embodiments.

[0157] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0158] It should be understood that the above description of the preferred embodiments is quite detailed and should not be construed as a limitation on the scope of protection of the present invention. Those skilled in the art can make substitutions or modifications under the guidance of the present invention without departing from the scope of protection of the claims of the present invention, and all such substitutions or modifications fall within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for repairing abnormal data of a detector array, characterized in that, The method includes: The measurement data sequence output by the target spot array detector and the corresponding permanent failure mask are obtained. The measurement data sequence includes multiple consecutive time points, and each time point corresponds to one frame of array data. The permanent failure mask represents the location of the permanently failed pixel. The permanent failure mask has the same dimension as the detector output array. For any time to be repaired, select the array data frames of the k adjacent times before and after it, and together with the center frame corresponding to the time to be repaired and the corresponding permanent failure mask, they form a context input sequence, where k is an integer greater than 1. The context input sequence is input into a pre-trained mask-aware reconstruction network to repair the center frame and obtain the center frame prediction result. Based on the permanent failure mask, the prediction result of the center frame is hard fused with the original measurement array data to obtain the repaired center frame array data.

2. The method of claim 1, wherein, The method further includes: The repaired center frame array data is mapped into a continuous spot image through a preset upsampling reconstruction operator; Calculate the mean absolute error of the missing area, the mean square error of the missing area, the relative error of the total ADU, the centroid shift, the peak position deviation and / or the inter-frame temporal stability index based on the continuous spot image. A quality assessment report is generated based on the aforementioned indicators and used for online monitoring and threshold alarms.

3. The method of claim 1, wherein, The training process of the mask-aware reconstruction network includes: A training sample set is constructed, which includes multiple array data frame sequences. Each array data frame sequence is obtained by injecting noise into clean array data samples according to the detector statistical model and superimposing quantization operators. Each array data frame sequence corresponds to a permanent failure mask, and the permanent failure masks corresponding to different array data frame sequences may be the same or different. For each array data frame sequence, an input tensor is constructed. The input tensor consists of the center frame corresponding to any time to be repaired, the array data frames at k adjacent times before and after the center frame, and the corresponding permanent failure mask. The input tensor input mask perception reconstruction network is used to reconstruct the prediction result; A target loss function is constructed based on the prediction results, wherein the target loss function is obtained based on the missing region reconstruction loss, observation consistency loss, energy consistency loss and centroid consistency loss; The mask-aware reconstruction network is iteratively optimized based on the target loss function to obtain a pre-trained mask-aware reconstruction network.

4. The method of claim 3, wherein, The process of reconstructing the input tensor input mask perception reconstruction network to obtain the prediction result includes: The mask-aware coding module extracts multi-scale spatial features of the center frame and the array data frames at k adjacent time points, and suppresses the interference of missing measurement regions on feature statistics through explicit mask input and gated convolution mechanism. The temporal feature fusion module is used to perform temporal correlation modeling on the coding features at each time step in order to restore the temporal continuity of the missing measurement area in the center frame; The spatial resolution is restored step by step by the feature decoding module, and shallow detail features are fused by skip connections to obtain the high-resolution feature representation corresponding to the center frame. The center frame prediction result is obtained by performing convolutional mapping through the output reconstruction module.

5. The method of claim 3, wherein, The target loss function includes: ； wherein, is a preset weight coefficient, denotes a missing region reconstruction loss, denotes an observation consistency loss, denotes an energy consistency loss, denotes a centroid consistency loss.

6. The method according to claim 3, characterized in that, The method further includes: In the case of anomalies where saturated pixels exist, a saturation mask is constructed for the location of the abnormal pixels; The energy consistency loss and centroid consistency loss are calculated based on the saturation mask and the center frame prediction results.

7. The method according to claim 1, characterized in that, The method further includes: In the presence of overall inter-frame drift, the observation centroid of each frame array data is calculated within the effective observation area, which is the array area after removing the missing measurement area. Based on the observed centroid of each frame, each frame is translated to a position approximately aligned with the observed centroid of the central frame.

8. The method according to claim 1, characterized in that, The method further includes: When the mask-aware reconstruction network outputs a pixel-level uncertainty map, the pixel-level uncertainty map is used to characterize the reliability of the prediction results at each pixel position. The larger the value of each pixel position, the higher the uncertainty of the corresponding position repair result. If the pixel value of the uncertainty map of the untested region is greater than a preset threshold, the region is marked as a high-risk reconstruction region, and at least one preset enhancement strategy is executed to optimize the inference process.

9. A system for repairing abnormal data from a detector array, characterized in that, The system includes: The acquisition module is used to acquire the measurement data sequence and the corresponding permanent failure mask output by the target spot array detector. The measurement data sequence includes multiple consecutive time points, and each time point corresponds to one frame of array data. The permanent failure mask represents the position of the permanently failed pixel and has the same dimension as the detector output array. The context input sequence construction module is used to select the array data frames of the k adjacent times before and after any time to be repaired, and together with the center frame corresponding to the time to be repaired and the corresponding permanent failure mask, they form the context input sequence, where k is an integer greater than 1. The mask-aware reconstruction network module is used to input the context input sequence into the pre-trained mask-aware reconstruction network to repair the center frame and obtain the center frame prediction result. The hard fusion module is used to perform hard fusion of the prediction result of the center frame with the original measurement array data based on the permanent failure mask, so as to obtain the repaired center frame array data.

10. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium, which a processor reads from and executes to implement the method as described in any one of claims 1 to 8.

Images