A display screen detection method and device based on voltage scanning and optical imaging

Abstract

The application discloses a display screen detection method and device based on voltage scanning and optical imaging, which is used for improving the accuracy of screen detection. The method comprises the following steps: applying a preset voltage scanning signal sequence to a to-be-detected screen; synchronously collecting an optical response image sequence of the to-be-detected screen under the driving of the voltage scanning signal sequence through an optical imaging device; inputting the voltage scanning signal sequence and the optical response image sequence into a pre-trained defect identification model to obtain a target correlation between the voltage scanning signal sequence and the optical response image sequence, wherein the target correlation is a mathematical representation of a physical law describing the electrical-optical conversion relationship of the screen; and generating a defect detection result according to the target correlation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of screen detection, and in particular to a display screen detection method and device based on voltage scanning and optical imaging. BACKGROUND

[0002] In the existing screen detection technology, it is usually relied on single-frame image analysis of display pictures or brightness distribution to judge defects, such as identifying bright spots, dark spots or local light emission abnormalities through gray threshold or brightness unevenness index. However, such method is only based on static optical characteristics for judgment, and lacks the description of the dynamic relationship between the electrical response and the optical output of the screen under different driving voltages, so it is difficult to accurately reflect the root cause of pixel light emission abnormalities. Especially under complex driving conditions, the brightness change, pixel response delay and current unevenness of the screen will jointly affect the detection results, resulting in misjudgment or missed judgment of the detection method based on a single optical characteristic. In addition, some existing methods do not introduce the physical law constraint of electrical-optical conversion, and the model recognition result is sensitive to noise and environmental changes, the defect positioning accuracy is insufficient, and the stability and reliability of the detection are also reduced. Therefore, there is an urgent need for a detection scheme that can simultaneously combine voltage scanning signals and optical response information, and take the electrical-optical conversion law as the core constraint to improve the accuracy of screen defect identification. SUMMARY

[0003] The present application provides a display screen detection method and device based on voltage scanning and optical imaging, which can improve the accuracy of screen detection.

[0004] The first aspect of the present application provides a display screen detection method based on voltage scanning and optical imaging, comprising:

[0005] applying a preset voltage scanning signal sequence to a to-be-tested screen;

[0006] synchronously collecting an optical response image sequence of the to-be-tested screen under the driving of the voltage scanning signal sequence by an optical imaging device;

[0007] inputting the voltage scanning signal sequence and the optical response image sequence into a pre-trained defect identification model to obtain a target correlation relationship between the voltage scanning signal sequence and the optical response image sequence, the target correlation relationship being a mathematical representation of a physical law describing the electrical-optical conversion relationship of the screen;

[0008] generating a defect detection result according to the target correlation relationship.

[0009] Optionally, before the voltage scanning signal sequence and the optical response image sequence are input into the pre-trained defect identification model, the method further comprises:

[0010] constructing a training data set, the training data set comprising a plurality of training samples, each of the training samples comprising a voltage scanning signal sequence applied to a sample screen body, a synchronously collected optical response image sequence, and corresponding annotation information;

[0011] constructing a physical information deep learning model as an initial model;

[0012] iteratively training the initial model by using the training data set;

[0013] at the end of each training, determining whether the initial model reaches a preset convergence condition;

[0014] if yes, determining the initial model as a defect identification model; if no, updating the initial model reversely and iteratively training the initial model again.

[0015] Optionally, the iteratively training the initial model by using the training data set comprises:

[0016] inputting the voltage scanning signal sequence and the optical response image sequence in the training samples into the initial model to obtain a defect identification result predicted by the initial model;

[0017] calculating a data loss value, the data loss value representing a difference between the defect identification result predicted by the initial model and the annotation information in the training samples;

[0018] calculating a physical constraint loss value, the physical constraint loss value being determined based on a degree of violation of an electro-optical conversion physical law by an intermediate representation quantity generated by the initial model;

[0019] calculating a total loss value based on the data loss value and the physical constraint loss value.

[0020] Optionally, the preset convergence condition comprises at least one of:

[0021] the total loss value is lower than a preset first threshold value;

[0022] a change rate of the data loss value is lower than a preset second threshold value;

[0023] a preset maximum number of iterations is reached.

[0024] Optionally, the calculating the physical constraint loss value comprises:

[0025] obtaining an intermediate representation quantity generated in a process of generating the defect identification result by the initial model, the intermediate representation quantity comprising at least one of: a predicted luminance distribution, a predicted chrominance distribution, and a predicted current characteristic;

[0026] calculate a degree of violation of the intermediate representation quantity to a partial differential equation describing an electrical-optical conversion relationship of the screen body;

[0027] determine a physical constraint loss value based on the degree of violation.

[0028] Optionally, the generating a defect detection result according to the target correlation relationship comprises:

[0029] calculating a difference measure between the target correlation relationship and a pre-stored reference correlation relationship;

[0030] generating a defect detection result based on the difference measure.

[0031] Optionally, the reference correlation relationship comprises a correlation relationship conforming to the electrical-optical conversion physical law and established based on training data of good screen bodies.

[0032] or,

[0033] a classification boundary relationship established based on screen body sample training data containing multiple defect types.

[0034] The second aspect of the present application provides a display screen detection device based on voltage scanning and optical imaging, which is applied to the method in the first aspect and any one of the possible implementation manners of the first aspect. The device comprises:

[0035] a scanning unit configured to apply a preset voltage scanning signal sequence to a screen body to be measured;

[0036] a collecting unit configured to synchronously collect an optical response image sequence of the screen body to be measured under driving of the voltage scanning signal sequence by using an optical imaging device;

[0037] an input unit configured to input the voltage scanning signal sequence and the optical response image sequence into a pre-trained defect recognition model to obtain a target correlation relationship between the voltage scanning signal sequence and the optical response image sequence, the target correlation relationship being a mathematical representation of a physical law describing an electrical-optical conversion relationship of the screen body;

[0038] a generating unit configured to generate a defect detection result according to the target correlation relationship.

[0039] The third aspect of the present application provides an electronic device, which comprises:

[0040] a processor, a memory, an input / output unit and a bus;

[0041] the processor is connected with the memory, the input / output unit and the bus;

[0042] The memory stores a program, and the processor invokes the program to execute the method in the first aspect and any possible implementation manner of the first aspect.

[0043] The fourth aspect of the present application provides a computer readable storage medium, which stores a program, and the program causes a computer to execute the method in the first aspect and any possible implementation manner of the first aspect when the program is executed on the computer.

[0044] From the above technical solutions, the present application has the following advantages:

[0045] In the present application, on the basis of accurately recording the voltage timing and corresponding to the high signal-to-noise ratio optical image frame by frame, the trained defect recognition model is used to extract the electrical-optical target correlation and determine the abnormality accordingly, so that the defect positioning is more accurate and the false positive rate is lower, and the traceable timing and image evidence are provided for the defect root cause analysis under different voltage conditions, thereby significantly improving the accuracy, reliability and interpretability of the detection. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A flowchart of an embodiment of the display screen detection method based on voltage scanning and optical imaging in the present application is shown;

[0047] Figure 2 A flowchart of an embodiment of the display screen detection method based on voltage scanning and optical imaging in the present application is shown;

[0048] Figure 3 A flowchart of an embodiment of the display screen detection method based on voltage scanning and optical imaging in the present application is shown;

[0049] Figure 4 A flowchart of an embodiment of the display screen detection method based on voltage scanning and optical imaging in the present application is shown;

[0050] Figure 5 A flowchart of an embodiment of the display screen detection method based on voltage scanning and optical imaging in the present application is shown;

[0051] Figure 6 A structural diagram of an embodiment of the display screen detection device based on voltage scanning and optical imaging in the present application is shown;

[0052] Figure 7 A structural diagram of an embodiment of the electronic device in the present application is shown. DETAILED DESCRIPTION

[0053] The embodiments of the present application provide a display screen detection method and device based on voltage scanning and optical imaging, which are used to improve the accuracy of screen detection.

[0054] The method of the present application can be applied to a server, a terminal or other device with logical processing capability, and the present application is not limited thereto. For convenience of description, the following description takes the server as an example.

[0055] The embodiments of the present application will be described below with reference to the accompanying drawings.

[0056] Please refer to Figure 1 , Figure 1 One embodiment of the display screen detection method based on voltage scanning and optical imaging provided by the present application includes the following steps:

[0057] 101. applying a preset voltage scanning signal sequence to a to-be-tested screen body;

[0058] First, a preset voltage scanning signal sequence is applied to the to-be-tested screen body, which is composed of voltage points with different amplitudes and / or different pulse forms, covering the working voltage range of the screen body. The duration, rising / falling slope and interval of each voltage point are clearly set to ensure stable pixel response. The application action is accurately output by the terminal through the driving interface, and the terminal also records the time stamp and parameters of each voltage point to ensure that the subsequent image frames can correspond to a specific voltage point one by one.

[0059] 102. synchronously collecting an optical response image sequence of the to-be-tested screen body under the driving of the voltage scanning signal sequence by an optical imaging device;

[0060] During the voltage scanning process, the optical response image sequence of the screen body is synchronously collected by the optical imaging device, and the collection frequency is synchronized and aligned with the voltage sampling time sequence, so that each frame of image corresponds to a clear voltage state. To improve data quality, the collection environment adopts light shielding or reflection elimination measures, and the imaging configuration (exposure time, gain, shutter time sequence) is dynamically adjusted by the terminal to adapt to the brightness change under different voltages, and the terminal saves the collection metadata (such as frame number, time stamp, camera parameter) of each frame for subsequent registration and analysis.

[0061] 103. inputting the voltage scanning signal sequence and the optical response image sequence into a pre-trained defect recognition model to obtain a target correlation relationship between the voltage scanning signal sequence and the optical response image sequence, the target correlation relationship being a mathematical representation of the physical law describing the electrical-optical conversion relationship of the screen body;

[0062] The aligned voltage scanning signal sequence and the optical response image sequence are input into a pre-trained defect recognition model to extract the target correlation therebetween. The model is operated or called by the terminal under the scheduling of the terminal, the model structure can simultaneously process the time sequence voltage features and the image spatial features, and outputs a mapping relationship (including a pixel response curve, a nonlinear coupling term and a spatial consistency feature) describing the voltage-optical conversion, which is used to describe the difference between the electrical-optical behavior of a normal pixel and a typical abnormal mode.

[0063] 104. Generating a defect detection result according to the target correlation.

[0064] The target correlation obtained by the model is used to generate a defect detection result: the terminal compares the actually collected optical response with the ideal response inferred by the model pixel by pixel, locates an abnormal pixel group or area based on a set threshold and multi-scale consistency test, and classifies defects (such as dark spots, bright spots, drive mismatch or local failure). The terminal is also responsible for summarizing the detection result and generating a visual report, including defect position coordinates, response curves under voltage conditions and confidence evaluation, to facilitate subsequent positioning and repair.

[0065] In this embodiment, on the basis of accurately recording the voltage time sequence and corresponding to the high signal-to-noise ratio optical image frame by frame, the trained defect recognition model is used to extract the electrical-optical target correlation and determine the abnormality accordingly, so that the defect positioning is more accurate, the false positive rate is lower, and the traceable time sequence and image evidence are provided for the defect root cause analysis under different voltage conditions, thereby significantly improving the reliability and interpretability of the detection.

[0066] Please refer to Figure 2 In some embodiments of the present application, step 103 in the above embodiment of inputting the voltage scanning signal sequence and the optical response image sequence into the pre-trained defect recognition model can further include the following steps:

[0067] 201. Constructing a training data set, the training data set including a plurality of training samples, each training sample including a voltage scanning signal sequence applied to a sample screen body, a synchronously collected optical response image sequence, and corresponding annotation information;

[0068] First, a training data set is constructed, the training data set including a plurality of training samples, each sample consisting of a voltage scanning signal sequence applied to a sample screen body, a synchronously collected optical response image sequence, and corresponding annotation information. In the training sample collection stage, the terminal accurately controls the voltage scanning parameters and synchronously acquires the optical response sequence to ensure that the data is aligned in the time domain and the spatial domain; at the same time, the defect labels are generated according to the manual annotation or auxiliary detection result of the screen body defect detection, so that the training data has clear supervision information.

[0069] 202. constructing a physical information deep learning model as an initial model;

[0070] After the data preparation is completed, a physical information deep learning model is constructed as an initial model. In the model design process, the terminal combines the traditional deep network structure with the physical law of the screen electrical-optical conversion, embeds a constraint term in the network that can reflect the coupling characteristics of the driving voltage and the luminous intensity, so that the model not only relies on data feature learning, but also has physical consistency. The model usually includes multiple convolution modules, time series encoding units, and fusion layers embedded with physical parameters to enhance the model's expression ability for multi-time series and multi-space features.

[0071] 203. iteratively training the initial model through a training data set;

[0072] The initial model is iteratively trained through a training data set. The terminal calculates the loss function using the paired relationship between the voltage scanning signal and the optical response image in the training sample, and continuously optimizes the model parameters, so that the output predicted defect result gradually approaches the labeled defect result. During the training process, adaptive learning rate adjustment and regularization constraints are used to avoid overfitting and improve the generalization ability of the model.

[0073] 204. At the end of each training, it is determined whether the initial model meets the preset convergence condition; if so, the initial model is determined as a defect recognition model; if not, the initial model is updated in reverse and iteratively trained again.

[0074] At the end of each training, the terminal determines whether the initial model meets the preset convergence condition. The condition can include a validation set loss lower than a set threshold or an accuracy improvement less than a predetermined proportion in consecutive multiple training rounds. When the convergence condition is met, the terminal determines the current initial model as a defect recognition model; if the condition is not met, the model parameters are updated according to the back propagation algorithm, and the next round of training is restarted until the model performance is stable and converges.

[0075] In this embodiment, the terminal constructs a training data set containing voltage scanning signals, optical response images and defect labels, and iteratively trains and converges the deep learning model combined with physical laws, so that the final obtained defect recognition model can better reflect the real correspondence between the screen electrical drive and optical output. In the recognition process, not only the physical rationality is maintained, but also the recognition accuracy of complex defect patterns is improved, thereby significantly enhancing the reliability and adaptability of the model.

[0076] Please refer to Figure 3 In some embodiments of the present application, step 203 in the above embodiment can include the following steps:

[0077] 301、input the voltage scanning signal sequence and the optical response image sequence in the training sample into the initial model to obtain a defect identification result predicted by the initial model;

[0078] input the voltage scanning signal sequence and the optical response image sequence in the training sample into the initial model to obtain a defect identification result predicted by the initial model. The terminal is responsible for calling the model inference module in this process, and inputs each group of voltage signal and image sequence as an input vector into the network structure, and outputs a prediction result after processing by the feature extraction layer, the fusion layer and the prediction layer. The prediction result usually shows a pixel-level or region-level defect distribution map, which is used to represent the abnormal area that may exist in the screen body.

[0079] 302、calculate a data loss value, the data loss value representing a difference between the defect identification result predicted by the initial model and the labeled information in the training sample;

[0080] The data loss value is used to measure the difference between the defect identification result predicted by the initial model and the labeled information in the training sample. The terminal uses the labeled information as a supervision signal to calculate the deviation degree of the two by using indicators such as cross-entropy or mean square error. A larger data loss value indicates that there is a significant gap between the model prediction and the real defect distribution, and the model parameters need to be further optimized.

[0081] Specifically, the initial model can include two data loss functions to cope with different situations.

[0082] I. The training data set includes good screen sample data, and the following data loss function can be used to calculate the data loss value:

[0083] Formula 1

[0084] wherein, represents the data loss value under the anomaly detection paradigm; N represents the number of good training samples; represents the voltage-optical response data pair of the i-th good sample; represents the autoencoder or generative model; represents the reconstruction error, which measures the ability of the model to reconstruct the input data.

[0085] II. The training data set contains screen sample data of multiple defect types, or contains good screen sample data at the same time, and the following data loss function can be used to calculate the data loss value:

[0086] Formula 2

[0087] wherein, represents the data loss value under the supervised learning paradigm; N represents the total number of training samples; C represents the total number of defect categories (which can also include the good category). represents the true label of the i-th sample in the class C; represents the probability that the i-th sample belongs to the class C predicted by the model.

[0088] 303、calculating a physical constraint loss value, the physical constraint loss value being determined based on a degree of violation of the intermediate representation quantity generated by the initial model to the electro-optical conversion physical law;

[0089] In the process of generating the defect identification result by the initial model, the terminal extracts intermediate representation quantities, which include the predicted luminance distribution, the predicted chrominance distribution, the predicted current characteristic, and other characteristics for reflecting the mapping relationship between the screen body electrical input and the optical output. The terminal substitutes the above intermediate representation quantities into the preset electro-optical conversion physical law model, and evaluates the deviation strength between the intermediate representation quantities and the physical law description, for example, judges whether the response of the luminance prediction to the voltage change conforms to the physical increasing relationship, whether the change trend of the chrominance distribution satisfies the energy conservation characteristic, and whether the current characteristic is consistent with the electro-optical response model of the panel material. The terminal calculates the physical constraint loss value according to the deviation degree, and the greater the deviation, the higher the physical constraint loss, to measure the violation degree of the prediction result of the initial model to the actual physical law.

[0090] 304、calculating a total loss value based on the data loss value and the physical constraint loss value.

[0091] The total loss value is calculated based on the data loss value and the physical constraint loss value. The terminal performs weighted summation on the two types of losses according to the set weight, to form a comprehensive optimization target; the data loss is used to drive the model to approach the true label, and the physical constraint loss is used to keep the prediction result consistent with the actual light emitting characteristic. After obtaining the total loss value, the terminal performs the back propagation and parameter updating operation, to provide the direction for the next training iteration. In the calculation of the total loss value, the data loss value and the physical constraint loss value can be weighted and summed respectively.

[0092] In the embodiment, the terminal calculates the data loss value and the physical constraint loss value simultaneously in the model training process, and optimizes the model based on the weighted results of the two, so that the model can both fit the labeled data and keep the compliance to the electro-optical conversion law. Through this training method combined with the physical constraint, the finally obtained defect identification model can avoid the overfitting problem caused by simply relying on data fitting, and improves the generalization and stability of the model under different screens and different working conditions, so as to make the defect identification result more accurate and reliable.

[0093] Please refer to Figure 4 In some embodiments of the present application, the step 303 in the above embodiment for calculating the physical constraint loss value can include the following steps:

[0094] 401、obtaining intermediate representation generated in the process of generating defect identification result by the initial model, the intermediate representation including at least one of: predicted luminance distribution, predicted chrominance distribution, predicted current characteristic;

[0095] The intermediate representation generated in the process of generating defect identification result by the initial model is obtained. The intermediate representation includes at least one of: predicted luminance distribution, predicted chrominance distribution, predicted current characteristic. The terminal extracts these intermediate quantities between the feature extraction layer and the fusion layer of the model, to reflect the understanding degree of the model on the electrical driving characteristics and optical response characteristics of the screen body at different feature levels. These intermediate representations not only contain the internal feature expression of the model, but also directly embody the physical meaning of the screen body light emitting characteristics, providing a basis for subsequent physical constraint calculation.

[0096] 402、calculating the degree of violation of the intermediate representation to the partial differential equation describing the electrical-optical conversion relationship of the screen body;

[0097] The degree of violation of the intermediate representation to the partial differential equation describing the electrical-optical conversion relationship of the screen body is calculated. In this step, the terminal substitutes the intermediate representation into the preset physical constraint equation, which describes the continuity and energy conservation relationship between voltage, current and optical output, such as the partial derivative change trend of luminance gradient to current density. If the model predicted intermediate representation deviates from the equilibrium condition of the equation, a larger violation value is calculated; if it conforms to the physical law, the violation value is smaller. The terminal can realize the partial differential test by using numerical difference or symbolic calculation.

[0098] 403、based on the degree of violation, determining a physical constraint loss value.

[0099] Based on the degree of violation, a physical constraint loss value is determined. The terminal weights and fuses the violation values corresponding to each intermediate representation according to the set weight, to obtain the overall physical constraint loss value. The loss value is added to the total loss function as an additional term in the training, so that the model is guided by the physical consistency in parameter updating. The smaller the physical constraint loss is, the more the model prediction result conforms to the screen body electrical-optical conversion law, thereby improving the explainability and stability of the prediction result.

[0100] Specifically, the terminal can calculate the physical constraint loss value according to the following formula:

[0101] Formula 3

[0102] Wherein, represents the partial differential equation describing the electrical-optical conversion relationship of the screen body; M represents the total number of discrete points in the physical field; represents the intermediate representation of the jth point.

[0103] In one possible embodiment, the intermediate representation quantity and the corresponding partial differential equation may be implemented in one of the following forms:

[0104] 1. When the intermediate representation quantity is the predicted luminance distribution:

[0105] Equation 4

[0106] where, represents the partial differential equation residual (i.e. the violation degree) of the luminance distribution at spatial point j; represents the luminance value predicted by the model at spatial point j; t represents the time variable; represents the rate of change of luminance over time; represents the luminance diffusion coefficient; is the spatial Laplacian operator of the luminance distribution, representing the diffusion term of luminance in space; is the voltage-dependent luminous efficiency function, representing the efficiency of converting current into luminance under voltage ; represents the voltage applied at spatial point j; represents the current density predicted by the model at spatial point j.

[0107] Equation 4 represents that the change of luminance over time should be jointly determined by the luminance diffusion and the voltage-driven luminous effect. If the predicted luminance distribution satisfies this physical law, the residual is close to zero; otherwise, the magnitude of its value reflects the degree of violation of the prediction to the physical law.

[0108] 2. When the intermediate representation quantity is the predicted chrominance distribution:

[0109] Equation 5

[0110] where, represents the partial differential equation residual of the chrominance distribution at spatial point j. represents the chrominance coordinate vector predicted by the model at spatial point j, usually represented as ; k is the chrominance diffusion coefficient tensor, representing the anisotropic characteristics of the chrominance diffusion in the screen body; is the chrominance diffusion term, representing the diffusion process of chrominance in space; represents the rate of change of chrominance over time; is the chrominance adjustment coefficient, representing the rate of adjustment of chrominance to the target value; represents the target chrominance coordinate vector, determined by the screen body design and the driving signal.

[0111] Equation 5 indicates that the change of chrominance over time should be jointly determined by the chrominance diffusion and the adjustment to the target chrominance. If the model-predicted chrominance distribution satisfies this physical law, the residual approaches zero.

[0112] 3. When the intermediate representation is the predicted current characteristic:

[0113] Equation 6

[0114] wherein, represents the partial differential equation residual with respect to the current characteristic at spatial point j; represents the model-predicted current density at spatial point j; is the conductivity, representing the conductive ability of the conductive layer of the screen body; represents the voltage distribution at spatial point j; is the diffusion term of the current, representing the current due to the voltage gradient; is the capacitance, representing the capacitance per unit area; represents the rate of change of voltage over time; is the external current source density, representing the current injected by the driving circuit.

[0115] Equation 6 indicates that the current density should be jointly determined by the Ohmic current in the conductive layer, the capacitive charging current, and the external source current. If the model-predicted current characteristic satisfies this physical law, the residual approaches zero.

[0116] In this embodiment, the terminal extracts the intermediate representation such as the luminance distribution, the chrominance distribution, and the current characteristic in the model training stage, and calculates the degree of violation of these features to the electro-optical conversion partial differential equation to form a physical constraint loss, so that the model actively corrects the feature mapping that does not conform to the physical law in the optimization process, and finally improves the ability of the model to depict the real light-emitting characteristics of the screen body and the physical consistency, so that the screen body defects can still be stably and accurately identified under complex driving conditions.

[0117] Please refer to Figure 5 In some embodiments of the present application, step 104 in the above embodiment of generating a defect detection result according to the target association relationship can include the following steps:

[0118] 501. Calculate the difference measure between the target association relationship and the pre-stored reference association relationship;

[0119] The terminal first acquires a target correlation relationship formed by the target screen body under the current detection condition, and then reads a pre-stored reference correlation relationship from the storage space of the terminal. The reference correlation relationship is constructed based on good sample data and / or multiple defect type sample data in the training stage, and its structure satisfies the physical law of electrical-optical conversion, so it can be used as a reference for judging the screen body state. The terminal compares the target correlation relationship with the reference correlation relationship, and in the comparison process, three cases are distinguished according to user settings and / or rules: when the reference correlation relationship only comes from good sample data, the terminal calculates the similarity of the target correlation relationship with the good reference, and the greater the difference, the more likely it contains defects; when the reference correlation relationship only comes from defect sample data, the terminal judges the proximity of the target correlation relationship to each defect reference, and the closer to a certain defect reference, the higher the correlation between the target and the defect type; when the reference correlation relationship contains good sample data and multiple defect sample data, the terminal calculates the difference degree of the target correlation relationship with the good reference and each defect reference respectively, and judges whether the target is closer to the good or a certain defect according to the difference degree, thereby generating a comprehensive difference measure. The terminal finally obtains the difference measure data for subsequent judgment.

[0120] 502. Generating a defect detection result based on the difference measure.

[0121] The terminal generates the final defect detection result according to the preset judgment logic based on the calculated difference measure data. When the difference measure shows that the target correlation relationship deviates significantly from the good reference and is close to a certain defect reference, the terminal determines that the screen body corresponds to the defect; when the difference measure is not close to all defect references and is highly consistent with the good reference, the terminal determines that the current screen body is good; when the difference measure is in a critical range between good and defect, the terminal further judges according to the comprehensive threshold, the difference trend and the weight of the reference correlation relationship, and finally outputs a clear detection result. In this embodiment, when calculating the difference measure of the target correlation relationship with the good or defect reference, the terminal distinguishes the reference correlation relationship into three cases according to the source and processes them respectively, so that the terminal can obtain more accurate difference measure expression in the good feature, defect feature and mixed scene; then the terminal judges whether the target is closer to the good or a certain defect based on the difference measure, so that the defect detection result has higher judgment accuracy, and finally reduces the difficulty of parameter adjustment.

[0122] Please refer to Figure 6 , Figure 6 An embodiment of a display screen detection device based on voltage scanning and optical imaging provided in the present application includes:

[0123] The scanning unit 601 is configured to apply a preset voltage scanning signal sequence to the screen body to be detected.

[0124] The collection unit 602 is configured to synchronously collect, by the optical imaging device, an optical response image sequence of the to-be-tested screen body under driving of the voltage scanning signal sequence.

[0125] The input unit 603 is configured to input the voltage scanning signal sequence and the optical response image sequence into the pre-trained defect identification model to obtain a target correlation relationship between the voltage scanning signal sequence and the optical response image sequence, the target correlation relationship being a mathematical representation of a physical law describing an electrical-optical conversion relationship of the screen body.

[0126] The generation unit 604 is configured to generate a defect detection result according to the target correlation relationship.

[0127] In the embodiment, the functions of the units correspond to the steps in the foregoing Figures 1 to 5 embodiment, which will not be described herein again.

[0128] Please refer to Figure 7 , Figure 7 An embodiment of an electronic device provided in the present application includes:

[0129] The electronic device includes a processor 701, a memory 702, an input and output unit 703, and a bus 704.

[0130] The processor 701 is connected with the memory 702, the input and output unit 703, and the bus 704.

[0131] The memory 702 stores a program, and the processor 701 invokes the program to perform the steps in the foregoing Figures 1 to 5 embodiment.

[0132] In the embodiment, the functions of the processor 701 correspond to the steps in the foregoing Figures 1 to 5 embodiment, which will not be described herein again.

[0133] The present application also provides a computer readable storage medium, which stores a program, and the program causes a computer to perform the method in any one of the foregoing possible implementation manners when the program is executed on the computer. Figures 1 to 5

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the system, device, and unit described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again.

[0135] ​In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. The division of the units is merely a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0136] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0137] In addition, each functional unit in the embodiments of the present application can be integrated in a processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of a software functional unit.

[0138] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such an understanding, the technical solutions of the present application essentially, or the part that contributes to the prior art, or all or a part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), magnetic disk or optical disk, and various other media that can store program codes.

Claims

1. A display screen detection method based on voltage scanning and optical imaging, characterized in that, The method comprises the following steps: applying a preset voltage scanning signal sequence to a to-be-tested screen body; synchronously collecting, by an optical imaging device, an optical response image sequence of the to-be-tested screen body under the driving of the voltage scanning signal sequence; inputting the voltage scanning signal sequence and the optical response image sequence into a pre-trained defect identification model to obtain a target correlation between the voltage scanning signal sequence and the optical response image sequence, the target correlation being a mathematical representation of a physical law describing an electrical-optical conversion relationship of the screen body; generating a defect detection result according to the target correlation; Before the step of inputting the voltage scanning signal sequence and the optical response image sequence into the pre-trained defect identification model, the method further comprises the following steps: constructing a training data set, the training data set comprising a plurality of training samples, each training sample comprising a voltage scanning signal sequence applied to a sample screen body, a synchronously collected optical response image sequence, and corresponding annotation information; constructing a physical information deep learning model as an initial model; iteratively training the initial model by using the training data set; at the end of each training, determining whether the initial model reaches a preset convergence condition; if yes, determining the initial model as the defect identification model; if no, updating the initial model in a reverse direction and iteratively training the initial model again; the step of iteratively training the initial model by using the training data set comprises the following steps: inputting the voltage scanning signal sequence and the optical response image sequence in the training sample into the initial model to obtain a defect identification result predicted by the initial model; calculating a data loss value, the data loss value representing a difference between the defect identification result predicted by the initial model and the annotation information in the training sample; wherein the deviation degree between the two is calculated by using a cross-entropy or mean square error index; calculating a physical constraint loss value, the physical constraint loss value being determined based on a degree of violation of an electrical-optical conversion physical law by an intermediate representation quantity generated by the initial model; calculating a total loss value based on the data loss value and the physical constraint loss value; wherein the total loss value is obtained by weighted summation of the data loss value and the physical constraint loss value according to a set weight; the step of calculating the physical constraint loss value comprises the following steps: obtaining an intermediate representation quantity generated in the process of generating the defect identification result by the initial model, the intermediate representation quantity comprising at least one of the following: a predicted luminance distribution, a predicted chrominance distribution, and a predicted current characteristic; calculating a degree of violation of a partial differential equation describing an electrical-optical conversion relationship of the screen body by the intermediate representation quantity; determining a physical constraint loss value based on the degree of violation; the step of generating a defect detection result according to the target correlation comprises the following steps: calculating a difference measure between the target correlation and a pre-stored reference correlation; generating a defect detection result based on the difference measure; the reference correlation is a correlation relationship established based on the training data set and conforming to an electrical-optical conversion physical law; the training data set comprises good screen body sample data and screen body sample data containing multiple defect types.

2. The method of claim 1, wherein, The preset convergence condition comprises at least one of the following: The total loss value is lower than a preset first threshold value; The change rate of the data loss value is lower than a preset second threshold value; A preset maximum iteration number is reached.

3. A display screen inspection apparatus based on voltage scanning and optical imaging, characterized in that, The device for implementing the method of claim 1 or 2 comprises: A scanning unit configured to apply a preset voltage scanning signal sequence to a to-be-tested screen body; An acquisition unit configured to synchronously acquire, by an optical imaging device, an optical response image sequence of the to-be-tested screen body under driving of the voltage scanning signal sequence; An input unit configured to input the voltage scanning signal sequence and the optical response image sequence into a pre-trained defect recognition model to obtain a target correlation between the voltage scanning signal sequence and the optical response image sequence, the target correlation being a mathematical representation of a physical law describing an electrical-optical conversion relationship of the screen body; A generation unit configured to generate a defect detection result according to the target correlation.

4. An electronic device, comprising: Comprise: A processor, a memory, an input / output unit, and a bus; The processor is connected with the memory, the input / output unit, and the bus; The memory has a program saved therein, and the processor invokes the program to execute the method of claim 1 or 2.

5. A computer readable storage medium, characterized in that, The computer readable storage medium has a program saved thereon, and the program, when executed on a computer, causes the computer to execute the method of claim 1 or 2.

Images