A flatness testing device and system for LED displays

Abstract

This invention discloses a flatness detection device and system for LED displays, comprising: an acquisition module acquiring several standard images; a marking module marking pits, protrusions, and creases on each standard display image to obtain marked display images; a training module introducing an initial model and retraining to obtain a non-flatness recognition model; a recognition module recognizing the RGB values ​​of pixels in each recognition area of ​​pits, protrusions, and creases to obtain a first RGB set for each recognition area; a prediction module introducing an initial detection model and retraining to obtain a flatness detection model; and an optimization module using several external flatness detection devices to detect several flatness data on the test display, processing the relationship between each flatness data and the actual flatness to obtain several corresponding correction coefficients, and adjusting the proportion of each first RGB set according to each correction coefficient to obtain a flatness optimized detection model. This invention improves the accuracy of flatness detection for LED displays.

Description

Technical Field

[0001] This invention relates to the field of display screen testing technology, and in particular to a flatness testing device and system for LED displays. Background Technology

[0002] An LED display is a flat panel display composed of small LED module panels used to display various information such as text, images, and videos. The flatness of an LED display is a crucial indicator of its quality. The surface flatness of an LED display should be controlled within ±1mm to ensure that the displayed image is not distorted. When the surface flatness exceeds ±1mm, local bulges or depressions will occur, leading to blind spots in the viewing angle and affecting usability. Therefore, flatness testing of LED displays is particularly important. Currently, existing technologies generally use infrared, laser, or ultrasonic ranging devices to test the flatness of LED displays. However, these methods are limited by the accuracy of the devices themselves. When the accuracy of these devices is insufficient, accurate flatness testing of the LED display cannot be guaranteed. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a flatness detection device and detection system for LED displays, so as to improve the accuracy of flatness detection of LED displays.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a flatness detection system for an LED display screen, comprising:

[0005] The acquisition module is used to acquire several test images of the test display screen and several standard images of the standard display screen. The flatness of each standard image is known and is called standard flatness.

[0006] A marking module, connected to the acquisition module, is used to mark pits, bumps and creases on the standard image to obtain a marked display image;

[0007] The training module, connected to the acquisition module and the labeling module respectively, is used to introduce an initial model, take several standard images before labeling as input, take several corresponding labeled display images as output, and retrain the initial model to obtain a non-flat part recognition model. The non-flat part recognition model is used to output a first labeled image based on the input standard image and a second labeled image based on the input test image. In both the first labeled image and the second labeled image, the pits, the protrusions and the creases are labeled with different colors.

[0008] The recognition module, connected to the training module, is used to divide the first marked image and the second marked image into several recognition regions, and to recognize the RGB values ​​of the pixels of the pits, bumps and creases in each recognition region, so as to obtain a first RGB set and a second RGB set in each recognition region, wherein the first RGB set corresponds to the first marked image and the second RGB set corresponds to the second marked image;

[0009] The prediction module, connected to the recognition module, is used to introduce an initial detection model, and is used to take the first RGB set in each of the recognition regions as input, take the standard flatness of the standard image as output, and retrain the initial detection model to obtain a flatness detection model.

[0010] An optimization module, connected to the prediction model, is used to perform flatness detection on the standard display screen using several external flatness detection devices to obtain several flatness data. Based on the relationship between each of the flatness data and the standard flatness, several corresponding correction coefficients are obtained. Based on each correction coefficient, the proportion of each of the first RGB sets is adjusted to obtain a flatness optimization detection model. The flatness optimization detection model is used to output flatness optimization detection data based on the second RGB set in each of the recognition areas.

[0011] Furthermore, the external flatness detection equipment includes an infrared ranging device, a laser ranging device, and an ultrasonic ranging device. The test display screen is pre-divided into several detection areas, each detection area corresponding to each identification area. The infrared ranging device is used to detect the first flatness data in each detection area, the laser ranging device is used to detect the second flatness data in each detection area, and the ultrasonic ranging device is used to detect the third flatness data in each detection area.

[0012] The optimization module includes:

[0013] The first calculation unit calculates a first correlation coefficient, a second correlation coefficient, and a third correlation coefficient based on the relationship between the first flatness data, the second flatness data, and the third flatness data and the standard flatness, respectively, and inputs the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient into a coefficient conversion formula to obtain a first correction coefficient, a second correction coefficient, and a third correction coefficient.

[0014] The second calculation unit, connected to the first calculation unit, is used to input the first correction coefficient, the second correction coefficient and the third correction coefficient into a preset comprehensive calculation formula to obtain the comprehensive correction coefficient;

[0015] An adjustment unit, connected to the second calculation unit, is used to adjust the weight parameters between the input layer and the hidden layer in the flatness detection model according to the comprehensive correction coefficient.

[0016] Furthermore, the coefficient correction formula is configured as follows:

[0017]

[0018] Among them, used for C c1 This represents the first correlation coefficient;

[0019] C c2 Used to represent the second correlation coefficient;

[0020] C c3 Used to represent the third correlation coefficient;

[0021] C o1 Used to represent the first correction coefficient;

[0022] C o2 Used to represent the second correction coefficient;

[0023] C o3 Used to represent the third correction coefficient;

[0024] k1, k2, and k3 are used to represent the preset first coefficient, second coefficient, and third coefficient, respectively.

[0025] Furthermore, the comprehensive calculation formula is configured as follows:

[0026]

[0027] Among them, C se Used to represent the comprehensive correction coefficient.

[0028] Furthermore, the weight parameters are configured as follows:

[0029]

[0030] Where W represents the weight parameter;

[0031] W0 is used to represent the preset initial value of the weight parameters.

[0032] Furthermore, the adjustment unit includes:

[0033] The association subunit is used to associate each of the comprehensive correction coefficients with the corresponding detection areas;

[0034] The adjustment subunit, connected to the associated subunit, is used to adjust the weight parameters of the first RGB set in the corresponding recognition area of ​​the detection area in the flatness detection model according to the comprehensive correction coefficient.

[0035] Furthermore, the first coefficient, the second coefficient, and the third coefficient can be dynamically adjusted based on the area of ​​the pits, the protrusions, and the creases. Therefore, the first coefficient, the second coefficient, and the third coefficient are configured as follows:

[0036]

[0037] Wherein, k1', k2' and k3' are used to represent the preset first initial coefficient, second initial coefficient and third initial coefficient, respectively;

[0038] S1, S2, and S3 are used to represent the area of ​​the pit, the area of ​​the protrusion, and the area of ​​the crease in each of the identification areas, respectively.

[0039] Furthermore, the training module includes:

[0040] A partitioning unit is used to divide the standard display images before labeling and the corresponding labeled display images into a training set and a test set according to a preset ratio;

[0041] The training unit, connected to the partitioning unit, is used to retrain the initial model based on each of the standard display images and the corresponding labeled display images in the training set to obtain an initial recognition model;

[0042] The comparison unit, connected to the training unit, is used to input each of the standard display images in the test set into the initial recognition model to obtain the corresponding initial prediction images, and to compare each of the initial prediction images with the corresponding labeled display images in the training set to obtain the comparison similarity.

[0043] An optimization unit, connected to the comparison unit, is used to adjust the weight parameters of the initial recognition model and retrain the initial recognition model when the comparison similarity is lower than a preset similarity threshold, until the comparison similarity is not lower than the similarity threshold, and then output the initial recognition model at this time as the non-flat part recognition model.

[0044] Furthermore, the preset ratio between the training set and the test set is 6:1.

[0045] A flatness detection device for an LED display screen is applied to the aforementioned flatness detection system, wherein the acquisition module and several external flatness detection devices are mounted on the flatness detection device.

[0046] The beneficial effects of this invention are:

[0047] This invention first trains a non-flatness recognition model to output a first marker image based on the input standard image. Then, the recognition module analyzes the first RGB set of the first marker image in each recognition region. Subsequently, a flatness prediction model is trained using each first RGB set and the standard image. Finally, the proportion of each first RGB set in the flatness prediction model is adjusted by using the relationship between the flatness data detected by the external detection device and the standard flatness, resulting in a flatness optimization detection model. This achieves dynamic correction of the weight parameters between the model's input layer and hidden layer, effectively improving the model's prediction accuracy and thus enhancing the accuracy of flatness detection for LED displays. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the structure of the present invention.

[0049] Reference numerals in the attached figures: 1. Acquisition module; 2. Marking module; 3. External flatness detection equipment; 4. Training module; 41. Division unit; 42. Training unit; 43. Comparison unit; 44. Optimization unit; 5. Recognition module; 6. Prediction module; 7. Optimization module; 71. First calculation unit; 72. Second calculation unit; 73. Adjustment unit; 731. Association subunit; 732. Adjustment subunit. Detailed Implementation

[0050] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0051] like Figure 1As shown in this embodiment, an LED display flatness detection system includes:

[0052] Acquisition module 1 is used to acquire several test images of the test display screen and several standard images of the standard display screen. The flatness of each standard image is known and is called standard flatness.

[0053] Marking module 2, connected to acquisition module 1, is used to mark pits, bumps and creases on a standard image to obtain a marked display image;

[0054] Training module 4 is connected to acquisition module 1 and labeling module 2 respectively. It is used to introduce the initial model, take several standard images before labeling as input, take several corresponding labeled display images as output, and retrain the initial model to obtain the non-flat part recognition model. The non-flat part recognition model is used to output the first labeled image according to the input standard image and the second labeled image according to the input test image. In both the first labeled image and the second labeled image, pits, bumps and creases are marked with different colors.

[0055] The recognition module 5, connected to the training module 4, is used to divide the first and second labeled images into several recognition regions, and to recognize the RGB values ​​of the pixels of pits, bumps and creases in each recognition region, so as to obtain the first RGB set and the second RGB set in each recognition region. The first RGB set corresponds to the first labeled image, and the second RGB set corresponds to the second labeled image.

[0056] Prediction module 6, connected to recognition module 5, is used to introduce the initial detection model, and is used to take the first RGB set in each recognition area as input, the standard flatness of the standard image as output, and retrain the initial detection model to obtain the flatness detection model.

[0057] The optimization module 7 is connected to the prediction model. It is used to use several external flatness detection devices 3 to perform flatness detection on the standard display screen to obtain several flatness data. Based on the relationship between each flatness data and the standard flatness, it processes the data to obtain several corresponding correction coefficients. Based on each correction coefficient, it adjusts the proportion of each first RGB set to obtain the flatness optimization detection model. The flatness optimization detection model is used to output flatness optimization detection data based on the second RGB set in each recognition area.

[0058] Specifically, in this embodiment, both the standard display screen and the test display screen are LED displays. The standard display screen is a display screen with known standard flatness, and each recognition area on the standard display screen contains pits, bumps, and creases to facilitate subsequent data acquisition and training. The acquisition module 1 is an image acquisition device, preferably an industrial camera, such as the MV-CS060-10GM. The acquisition module 1 first acquires several standard images of the standard display screen for model training, and then acquires several test images of the test display screen. The marking module 2 can be automatic marking software installed on the terminal device, or it can be manually marked by professional testers. The initial model introduced by the training module 4 can be an OpenCV model, and image recognition is performed based on the OpenCV model. The training module 4 uses each standard image and the corresponding marked display image to train an uneven part recognition model. This uneven part recognition model first processes the input standard image to obtain a first marked image, and then processes the input test image to obtain a second marked image. The first and second marked images can be distinguished by red, green, and blue, respectively, to facilitate subsequent analysis and differentiation. The recognition module 5 can be color recognition software installed on the terminal device, used to recognize the RGB values ​​of pixels occupied by pits, bumps, and creases in each recognition area, obtaining a first RGB set and a second RGB set for each recognition area. The first RGB set contains the RGB values ​​of each pixel in a recognition area of ​​the first marked image, and the second RGB set contains the RGB values ​​of each pixel in a recognition area of ​​the second marked image. The prediction module 6 introduces an initial detection model that can be the PSPnet model. The PSPnet model has low computational cost and high prediction accuracy, laying a good foundation for improving the accuracy of LED display flatness detection in this technical solution. The prediction module 6 uses the first RGB set of each recognition area in the standard image as input and the known standard flatness of the standard image as output to retrain the PSPnet model to obtain a flatness detection model. This flatness detection model is used to input the second RGB set of each recognition area in the test image to predict the initial flatness detection data. The prediction accuracy of this initial flatness detection data is not high enough, therefore, the optimization module 7 is needed to adjust the parameters of the flatness detection model to improve the prediction accuracy.The optimization module 7 detects the flatness data of a standard display screen using several external flatness testing devices 3. It then performs correlation analysis between each flatness data point and the standard flatness to obtain several correlation coefficients. These correlation coefficients are then processed to obtain correction coefficients. These correction coefficients are used to adjust the weight of each first RGB set in the flatness prediction model, resulting in an optimized flatness detection model. This achieves dynamic correction of the weight parameters between the model's input and hidden layers, effectively improving the model's prediction accuracy and thus enhancing the accuracy of LED display screen flatness detection. Finally, each second RGB set is input into the optimized flatness detection model, which then outputs optimized flatness detection data corresponding to the test image.

[0059] Preferably, the external flatness testing device 3 includes an infrared ranging device, a laser ranging device, and an ultrasonic ranging device. The test display screen is pre-divided into several testing areas, each testing area corresponding to a recognition area. The infrared ranging device is used to detect the first flatness data in each testing area, the laser ranging device is used to detect the second flatness data in each testing area, and the ultrasonic ranging device is used to detect the third flatness data in each testing area.

[0060] Optimization module 7 includes:

[0061] The first calculation unit 71 calculates the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient based on the relationship between the first flatness data, the second flatness data, and the third flatness data and the standard flatness, respectively, and inputs the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient into the coefficient conversion formula to obtain the first correction coefficient, the second correction coefficient, and the third correction coefficient.

[0062] The second calculation unit 72 is connected to the first calculation unit 71 and is used to input the first correction coefficient, the second correction coefficient and the third correction coefficient into a preset comprehensive calculation formula to obtain the comprehensive correction coefficient.

[0063] The adjustment unit 73 is connected to the second calculation unit 72 and is used to adjust the weight parameters between the input layer and the hidden layer in the flatness detection model according to the comprehensive correction coefficient.

[0064] Preferably, the coefficient correction formula is configured as follows:

[0065]

[0066] Among them, used for C c1 This represents the first correlation coefficient;

[0067] C c2 Used to represent the second correlation coefficient;

[0068] C c3Used to represent the third correlation coefficient;

[0069] C o1 Used to represent the first correction factor;

[0070] C o2 Used to indicate the second correction factor;

[0071] C o3 Used to represent the third correction factor;

[0072] k1, k2, and k3 are used to represent the preset first coefficient, second coefficient, and third coefficient, respectively.

[0073] Specifically, in this embodiment, the infrared ranging device can be an infrared ranging sensor, and the model of the infrared ranging sensor can be GP2Y0E03. The laser ranging device can be a laser ranging sensor, and the model of the laser ranging sensor can be ZX / DLS-C15. The ultrasonic ranging device can be an ultrasonic ranging sensor, and the model of the ultrasonic ranging sensor can be JCS2503. The infrared ranging sensor, laser ranging sensor, and ultrasonic ranging sensor are used to detect the first flatness data, second flatness data, and third flatness data within each detection area, respectively. The covariance formula is used to calculate the first correlation coefficient, second correlation coefficient, and third correlation coefficient between the first flatness data, second flatness data, and third flatness data and the standard flatness. Then, the coefficient conversion formula is used to convert the first correlation coefficient, second correlation coefficient, and third correlation coefficient into a first correction coefficient, second correction coefficient, and third correction coefficient. Then, based on the first, second, and third correction coefficients, a comprehensive correction coefficient is calculated to dynamically adjust the weight parameters between the input layer and the hidden layer in the flatness detection model. This achieves the correction of the proportion of each first RGB set in the flatness prediction model, effectively improving the prediction accuracy of the model.

[0074] The preferred configuration for the comprehensive calculation formula is as follows:

[0075]

[0076] Among them, C se Used to represent the overall correction coefficient.

[0077] Preferably, the weight parameters are configured as follows:

[0078]

[0079] Where W represents the weight parameter;

[0080] W0 is used to represent the preset initial value of the weight parameters.

[0081] Preferably, the adjustment unit 73 includes:

[0082] The associated subunit 731 is used to associate each comprehensive correction coefficient with the corresponding detection area;

[0083] Adjustment subunit 732, connected to association subunit 731, is used to adjust the weight parameters of the first RGB set in the corresponding recognition area of ​​the detection area in the flatness detection model according to the comprehensive correction coefficient.

[0084] Preferably, the first, second, and third coefficients can be dynamically adjusted based on the area of ​​the pits, bumps, and creases. Therefore, the first, second, and third coefficients are configured as follows:

[0085]

[0086] Wherein, k1', k2' and k3' are used to represent the preset first initial coefficient, second initial coefficient and third initial coefficient, respectively;

[0087] S1, S2, and S3 are used to represent the area of ​​the pit, the area of ​​the protrusion, and the area of ​​the crease in each recognition area, respectively.

[0088] Specifically, in this embodiment, by introducing the area of ​​pits, bumps and creases in each recognition area to dynamically adjust the first coefficient, the second coefficient and the third coefficient, dynamic optimization of the comprehensive correction coefficient is achieved, which in turn further optimizes the weight parameters between the input layer and the hidden layer in the flatness detection model, thereby further improving the prediction accuracy of the model.

[0089] Preferably, training module 4 includes:

[0090] The partitioning unit 41 is used to divide the standard display images before the labeling and the corresponding label display images into a training set and a test set according to a preset ratio;

[0091] The training unit 42 and the connection partitioning unit 41 are used to retrain the initial model based on the standard display images and the corresponding labeled display images in the training set to obtain the initial recognition model.

[0092] The comparison unit 43 is connected to the training unit 42. It is used to input each standard display image in the test set into the initial recognition model to obtain the corresponding initial prediction image, and compare each initial prediction image with the corresponding labeled display images in the training set to obtain the comparison similarity.

[0093] The optimization unit 44 is connected to the comparison unit 43. When the comparison similarity is lower than the preset similarity threshold, the weight parameters of the initial recognition model are adjusted and the initial recognition model is retrained until the comparison similarity is not lower than the similarity threshold. Then, the initial recognition model at this time is output as the non-flat part recognition model.

[0094] Specifically, in this embodiment, the similarity threshold can be 95%. By setting the optimization unit 44, the prediction accuracy of the non-flat part recognition model is initially limited, so that the prediction accuracy of the non-flat part recognition model is at least 95%, laying the foundation for subsequent accuracy optimization.

[0095] Preferably, the preset ratio between the training set and the test set is 6:1.

[0096] A flatness testing device for an LED display screen is applied to the aforementioned flatness testing system. The acquisition module 1 and several external flatness testing devices 3 are mounted on the flatness testing device.

[0097] Specifically, in this embodiment, by setting up a flatness detection device, the acquisition module 1 and several external flatness detection devices 3 are installed and fixed, so that the acquisition module 1 can face the standard display screen and the test display screen to acquire images, while ensuring that each external flatness detection device 3 can acquire flatness data.

[0098] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A flatness detection system of an LED display screen, characterized in that, The application relates to a display flatness detection system, which comprises: a collection module (1) for collecting a plurality of test images of a test display screen and a plurality of standard images of a standard display screen, the flatness of each standard image being known and being a standard flatness; a marking module (2) connected to the collection module (1) and used for marking pits, protrusions and folds on the standard images to obtain marked display images; a training module (4) connected to the collection module (1) and the marking module (2) respectively, used for introducing an initial model, taking a plurality of standard images before marking as input and taking a plurality of corresponding marked display images as output, retraining the initial model to obtain a non-flat part recognition model, the non-flat part recognition model being used for outputting a first marked image according to the input standard image and outputting a second marked image according to the input test image, the first marked image and the second marked image both marking the pits, the protrusions and the folds according to different colors; an identification module (5) connected to the training module (4) and used for dividing the first marked image and the second marked image into a plurality of identification regions and identifying the RGB values of the pixels of the pits, the protrusions and the folds in each identification region to obtain a first RGB set and a second RGB set in each identification region, the first RGB set corresponding to the first marked image and the second RGB set corresponding to the second marked image; a prediction module (6) connected to the identification module (5) and used for introducing an initial detection model, taking the first RGB set in each identification region as input and taking the standard flatness of the standard image as output, retraining the initial detection model to obtain a flatness detection model; an optimization module (7) connected to the prediction module (6) and used for obtaining a plurality of flatness data by using a plurality of external flatness detection devices (3) to detect the flatness of the standard display screen, processing a plurality of correction coefficients according to the relationship between each flatness data and the standard flatness, adjusting the proportion of each first RGB set according to each correction coefficient to obtain a flatness optimization detection model, and the flatness optimization detection model being used for outputting flatness optimization detection data according to the second RGB set in each identification region. 2.The flatness detection system of the LED display screen according to claim 1, characterized in that: The external flatness detection device (3) comprises an infrared distance measuring device, a laser distance measuring device and an ultrasonic distance measuring device, the test display screen is divided into a plurality of detection regions in advance, each detection region corresponds to each identification region, the infrared distance measuring device is used for detecting first flatness data in each detection region, the laser distance measuring device is used for detecting second flatness data in each detection region, and the ultrasonic distance measuring device is used for detecting third flatness data in each detection region; The optimization module (7) comprises: The first calculation unit (71) calculates a first correlation coefficient, a second correlation coefficient and a third correlation coefficient according to the relationship between the first flatness data, the second flatness data, the third flatness data and the standard flatness respectively, and inputs the first correlation coefficient, the second correlation coefficient and the third correlation coefficient into a coefficient conversion formula to obtain a first correction coefficient, a second correction coefficient and a third correction coefficient; The second calculation unit (72) is connected with the first calculation unit (71) and is configured to input the first correction coefficient, the second correction coefficient and the third correction coefficient into a preset comprehensive calculation formula to obtain a comprehensive correction coefficient; The adjustment unit (73) is connected with the second calculation unit (72) and is configured to adjust the weight parameter between the input layer and the hidden layer in the flatness detection model according to the comprehensive correction coefficient.

3. The flatness detection system of the LED display screen according to claim 2, characterized in that: The coefficient conversion formula is configured as: wherein the first correction factor is used for C c1 denotes the first correction factor; C c2 for representing the second correction coefficient; C c3 for representing the third correction coefficient; C o1 for representing the first correlation coefficient; C o2 for representing the second correlation coefficient; C o3 for representing the third correlation coefficient; k1, k2 and k3 respectively represent the preset first coefficient, the second coefficient and the third coefficient.

4. The flatness detection system of the LED display screen according to claim 3, characterized in that: The comprehensive calculation formula is configured as: wherein C se for representing the comprehensive correction coefficient.

5. The flatness detection system of the LED display screen according to claim 2, characterized in that: The adjustment unit (73) comprises: The association sub-unit (731) is configured to associate each of the comprehensive correction coefficients with the corresponding detection area; The adjustment sub-unit (732) is connected with the association sub-unit (731) and is configured to adjust the weight parameter of the first RGB set in the corresponding identification area of the detection area in the flatness detection model according to the comprehensive correction coefficient.

6. The flatness detection system of the LED display screen according to claim 3, characterized in that: The first coefficient, the second coefficient and the third coefficient can be dynamically adjusted according to the area of the pit, the protrusion and the crease, and the first coefficient, the second coefficient and the third coefficient are configured as: Wherein, k1', k2' and k3' respectively represent the preset first initial coefficient, the second initial coefficient and the third initial coefficient; S1, S2 and S3 respectively represent the area of the pit, the area of the protrusion and the area of the crease in each of the identification areas.

7. The flatness detection system of the LED display screen according to claim 1, characterized in that: The training module (4) comprises: The division unit (41) is configured to divide each of the standard images before marking and the corresponding each of the marked display images into a training set and a test set according to a preset proportion; The training unit (42) is connected with the division unit (41) and is configured to retrain the initial model to obtain an initial identification model according to each of the standard images in the training set and the corresponding each of the marked display images; The comparison unit (43) is connected with the training unit (42) and is configured to input each of the standard images in the test set into the initial identification model to obtain a corresponding initial predicted image, and compare each of the initial predicted images with the corresponding each of the marked display images in the training set to obtain a comparison similarity; The optimization unit (44) is connected with the comparison unit (43) and is configured to adjust the weight parameter of the initial identification model and retrain the initial identification model when the comparison similarity is lower than a preset similarity threshold, and output the initial identification model at this time as the non-flat part identification model when the comparison similarity is not lower than the similarity threshold.

8. The flatness detection system of the LED display screen according to claim 7, characterized in that: The preset ratio between the training set and the test set is 6:

1.

9. The flatness detection system of the LED display screen according to claim 4, characterized in that: The weight parameter is configured as: Wherein, W is used to represent the weight parameter; W0 is used to represent the preset weight parameter initial value.

10. A flatness detection device of an LED display screen, applied to the flatness detection system of any one of claims 1-9, characterized in that, The acquisition module (1) and a plurality of external flatness detection devices (3) are arranged on the flatness detection device.

Images