Intelligent POB module integration platform and control method thereof

Abstract

The invention provides an intelligent POB module integration platform and a control method thereof, and the method comprises the steps: detecting the state data of each pixel point of an LED display screen through an optical sensing array, obtaining a damaged pixel point in the LED display screen, recording the damaged pixel point to a storage unit, and maintaining a dead pixel position table in the storage unit; compensating the brightness of the bad pixel points through the neighborhood pixel points; the display content of the LED display screen is predicted through AI during compensation, and the refresh rate of the LED display screen is dynamically driven according to the predicted display content; according to the physical pixel density, the LED display screen is divided into grid partitions, the grid partitions are sorted according to priorities, the real-time power consumption of each partition is calculated after the LED display screen is refreshed, and if the real-time power consumption is larger than a preset threshold value, the current and the duty ratio of the low-priority area are reduced. Through the control method, a compensation scheme can be started in time after a dead pixel occurs, the maintenance and replacement cost is reduced, and the operation and maintenance efficiency is improved.

Description

Technical Field

[0001] The present invention provides an intelligent POB module integration platform and a control method thereof, relating to LED display technology. Background Art

[0002] There are three Mini LED backlight technology solutions: POB (Package-on-Board), COB (Chip On Board, fixing the IC die on the printed circuit board through bonding), and COG (Chip On Glass, fixing the chip on the glass). Currently, the mainstream application solutions in the market are POB and COB. Some pixel points of LED displays are prone to damage. The existing POB module control methods often replace the entire LED display, and cannot compensate for the damaged pixel points through the normal pixel points around them to make the entire LED display "look" normal, or the existing POB cannot perform heat dissipation control according to the LED display area, or cannot predict the display content of the LED display in advance and control the refresh rate of the LED display based on the display content. Summary of the Invention

[0003] The present invention provides an intelligent POB module integration platform and a control method thereof to solve the above-mentioned problems: An intelligent POB module control method provided by the present invention, the method includes: Detecting the state data of each pixel point of the LED display through an optical sensing array, obtaining the damaged pixel points in the LED display based on the state data of each pixel point, recording the damaged pixel points in a storage unit, and maintaining a bad point position table in the storage unit; Compensating the brightness of the damaged pixel points through neighboring pixel points according to the damage situation of the damaged pixel points; While compensating, predicting the display content of the LED display through AI, and dynamically driving the refresh rate of the LED display according to the predicted display content; Dividing the LED display into grid partitions according to the physical pixel density, sorting the grid partitions according to the priority, calculating the real-time power consumption of each partition after refreshing the LED display, and if the real-time power consumption is greater than a preset threshold, reducing the current and duty cycle of the low-priority area.

[0004] Further, detecting the state data of each pixel point of the LED display through an optical sensing array, and obtaining the damaged pixel points in the LED display based on the state data of each pixel point, including: Detecting the state data of each pixel point of the LED display through an optical sensing array, where the pixel point state data includes the brightness of the pixel point; Compare the actually detected brightness value with the expected display brightness value. If the difference between the actually detected brightness value and the expected display brightness value exceeds the first preset threshold, mark the pixel points with abnormal brightness; Record the coordinates of the pixel points with abnormal brightness into the storage unit, and maintain a real-time updated bad point position table in the storage unit.

[0005] Furthermore, compensate for the brightness of the bad pixel points through the neighboring pixel points, including: Check whether the four pixels around the bad pixel point are in the bad point table. The four pixels include the upper left B11, the upper right B21, the lower left B12, and the lower right B22. If all are normal pixels, calculate the bad point compensation brightness in advance through the bad point brightness compensation model in the FPGA. Specifically, the bad point brightness compensation model is:

[0006] Wherein, represents the coordinates of the bad point to be compensated, represents the boundary coordinates of the four surrounding normal pixels, (for example, the upper left corner is , and the lower right corner is ), , and represent the brightness values of the four normal pixels; If there are multiple neighboring bad points in the neighborhood pixel points of the bad pixel point (such as both B11 and B21 are bad): Area averaging method: Taking the bad point as the center, expand the search range to a 3×3 area, and take the average value of the normal pixels:

[0007] If the reference pixels still contain bad points after expansion, perform recursive compensation on the bad points of the reference pixels, first repair their values, and then use them for the interpolation of the original bad points. If B11 is a bad point, first compensate B11 through the surrounding normal pixels, and then substitute the compensated value of B11 into the original formula to calculate B(x,y); If the deviation between the compensated value and the actual value exceeds the preset threshold, add the pixel points with compensation failure to the high-priority maintenance list. When there are pixel points in the high-priority maintenance queue, trigger a reminder for manual maintenance inspection.

[0008] Furthermore, while compensating, predict the display content of the LED display through AI, and adjust the refresh rate of the LED display according to the predicted display content, including: Real-time collect the display content video stream and metadata through the HDMI signal analyzer of the LED display. The metadata includes the resolution and frame rate of the display content, and synchronously record the time stamp; Obtain video stream data and image data from public datasets, where the video stream data covers action, game, and movie scenes; Normalize the frame rate of the video stream data, compensate for inter-frame motion offsets through optical flow method, encode the resolution and frame rate into feature vectors, and splice them with the video stream data; Label the content type of the spliced video stream data, construct a 3D-CNN + LSTM model, and divide the labeled data into a training set, a validation set, and a test set according to the ratio of 7:1:2; Input the training set data into the 3D-CNN + LSTM model, train for 200 epochs, monitor the accuracy loss on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, terminate the training in advance. The loss function of the validation set is total loss = α·classification loss + β·regression loss, where α and β are weight coefficients, the classification loss is the cross-entropy loss for classifying based on the video content type, and the regression loss is the mean square error loss based on the motion speed and color complexity of the video content; Obtain the trained model, input the real-time display content into the trained model, predict the motion speed and color complexity of the LED display content, and query the preset refresh rate mapping table based on the predicted motion speed and color complexity.

[0009] Further, divide the LED display screen into grid partitions according to the physical pixel density, sort the grid partitions according to the priority, calculate the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than the preset threshold, reduce the current and duty cycle of the low-priority areas, including: Divide the LED display screen into grid partitions, and each partition contains the same number of pixels; Sort the grid partitions according to the priority, and the priority is calculated by the priority model. Specifically, the priority model is:

[0010] Calculate the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than the preset threshold, reduce the current and duty cycle of the low-priority areas.

[0011] An intelligent POB module integration platform proposed by the present invention, the system includes: A storage bad pixel point module, which is used to detect the state data of each pixel point of the LED display screen through an optical sensing array, obtain the damaged pixel points in the LED display screen based on the state data of each pixel point, record the damaged pixel points in the storage unit, and maintain a bad point position table in the storage unit; A compensation module, which is used to compensate the brightness of the damaged pixel points through the neighboring pixel points according to the damage situation of the damaged pixel points; A prediction adjustment module, which is used to compensate and predict the display content of the LED display through AI at the same time, and dynamically drive the refresh rate of the LED display according to the predicted display content; A priority-based adjustment module, which is used to divide the LED display into grid partitions according to the physical pixel density, sort the grid partitions according to the priority, calculate the real-time power consumption of each partition after refreshing the LED display, and if the real-time power consumption is greater than the preset threshold, reduce the current and duty cycle of the low-priority area.

[0012] Furthermore, the storage bad pixel point module includes: Detect the state data of each pixel point of the LED display through an optical sensing array, and the pixel point state data includes the brightness of the pixel point; Compare the actually detected brightness value with the expected display brightness value. If the difference between the actually detected brightness value and the expected display brightness value exceeds the first preset threshold, mark it as a pixel point with abnormal brightness; Record the coordinates of the pixel points with abnormal brightness into the storage unit, and maintain a real-time updated bad point position table in the storage unit.

[0013] Furthermore, the compensation module includes: A storage bad point brightness compensation model module, which is used to check whether the four pixels around the bad pixel are in the bad point table. The four pixels include the upper left B11, the upper right B21, the lower left B12, and the lower right B22. If they are all normal pixels, calculate the bad point compensation brightness in advance through the bad point brightness compensation model in the FPGA. Specifically, the bad point brightness compensation model is:

[0014] Wherein, represents the coordinates of the bad point to be compensated, represents the boundary coordinates of the four surrounding normal pixels, (for example, the upper left corner is , and the lower right corner is ), , and represent the brightness values of the four normal pixels; An area averaging module, which is used to expand the search range to a 3×3 area centered on the bad point and take the average value of the normal pixels if there are multiple neighboring bad points (such as B11 and B21 are both bad) in the neighboring pixels of the bad pixel point:

[0015] A recursive compensation module, which is used to recursively compensate for the bad pixels of the reference pixels if the extended reference pixels still contain bad pixels. First, repair their values and then use them for the interpolation of the original bad pixels. If B11 is a bad pixel, first compensate B11 through the surrounding normal pixels, and then substitute the compensated value of B11 into the original formula to calculate B(x, y). A trigger for manual maintenance reminder module, which is used to add the pixel points with compensation failure to the high-priority maintenance list if the deviation between the compensation value and the actual value exceeds the preset threshold. When there are pixel points in the high-priority maintenance queue, a reminder for manual maintenance inspection is triggered.

[0016] Furthermore, the prediction adjustment module includes: A real-time display content acquisition module, which is used to analyze the HDMI signal of the LED display screen to acquire the display content video stream and metadata in real time. The metadata includes the resolution and frame rate of the display content, and the time stamp is synchronously recorded. An open data acquisition module, which is used to obtain video stream data and image data from the open data set. The video stream data covers action, game, and movie scenes. A preprocessing module, which is used to standardize the frame rate of the video stream data, compensate for the inter-frame motion offset by the optical flow method, encode the resolution and frame rate into a feature vector, and splice it with the video stream data. A labeling and partitioning module, which is used to label the content type of the spliced video stream data, construct a 3D-CNN + LSTM model, and divide the labeled data into a training set, a validation set, and a test set according to the ratio of 7:1:2. A training module, which is used to input the training set data into the 3D-CNN + LSTM model, train for 200 epochs, monitor the accuracy loss on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, the training is terminated in advance. The loss function of the validation set is the total loss = α·classification loss + β·regression loss, where α and β are weight coefficients, the classification loss is the cross-entropy loss for classifying based on the video content type, and the regression loss is the mean square error loss based on the motion speed and color complexity of the video content. A query module, which is used to obtain the trained model, input the real-time display content into the trained model, predict the motion speed and color complexity of the LED display content, and query the preset refresh rate mapping table based on the predicted motion speed and color complexity.

[0017] Furthermore, the priority-based adjustment module includes: A partitioning module, which is used to divide the LED display screen into grid partitions, and each partition contains the same number of pixels. A priority sorting module, which is used to sort the grid partitions according to the priority. The priority is calculated by the priority model. Specifically, the priority model is:

[0018] Among them, represents the content type weight coefficient, represents the mapping of the content type where the pixel coordinates (x, y) are located, represents the position weight coefficient, (x, y) represents the pixel coordinates, and σ represents the attenuation parameter, represents the number of times the content in the partition has changed within the past 1 hour, represents the number of times the content in the smallest partition within the entire LED display screen has changed, represents the number of times the content in the largest partition within the entire LED display screen has changed; The current and duty cycle adjustment module is used to calculate the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than the preset threshold, the current and duty cycle of the low-priority area are reduced.

[0019] Beneficial effects of the present invention: Dead pixel detection and recording, optical sensing array. A high-density photosensitive sensor array is deployed on the back or surface of the LED display screen to collect the luminous state data of each pixel point in real time, and the damaged pixels are accurately located through a threshold comparison algorithm; the dead pixel coordinates and damage types are recorded in a non-volatile memory to form a global dead pixel map; Display content prediction, capture the current frame of image, video stream and metadata (resolution, frame rate) through an HDMI signal analyzer, and combine with historical content patterns; extract spatio-temporal features (such as motion vectors, color complexity), and predict the content type (such as text, video, animation) and dynamic characteristics (such as motion speed) of the next 1-10 frames. Refresh rate mapping, the dynamic mapping table adjusts the refresh rate according to the predicted content (such as 60Hz for text scenes, 240Hz for game scenes), matches the human eye's critical fusion frequency (CFF), and realizes hardware-level control of refresh rate switching through FPGA, with a switching delay <5ms; Combine the content type (such as text 0.8, video 0.5), position weight (central Gaussian decay), and historical activity (update frequency normalization) to calculate the priority, and calculate the partition power consumption in real time. If the total power consumption exceeds 80% of the rated value, trigger a power consumption reduction process, reduce the load of low-priority areas, and select 20% of the partitions in ascending order of priority, and reduce the power consumption by adjusting the current and duty cycle. Improve display reliability and reduce screen damage caused by dead pixels. Through the real-time detection of the optical sensing array, dead pixels can be quickly discovered and recorded; on this basis, neighborhood compensation can be carried out to effectively reduce the impact of dead pixels on the overall display screen and ensure the viewing effect and display quality; Automatic dead pixel processing improves operation and maintenance efficiency. The traditional method often relies on manual inspection or low-frequency monitoring methods. This technical solution uses a sensing array to automatically detect and record the dead pixel positions, eliminating the cumbersome steps of manual inspection. At the same time, a compensation scheme can be started in time after a dead pixel occurs, reducing maintenance and replacement costs and improving operation and maintenance efficiency; combined with AI prediction, reduce unnecessary energy consumption and improve dynamic display quality. Predict the upcoming content through AI, and dynamically adjust the refresh rate according to the actual scene: output high refresh rates for rapidly changing images to ensure smoothness and clarity; reduce the refresh rate for relatively static images, reducing power consumption without affecting the visual effect and prolonging the service life of the LED; Precise energy consumption management is achieved through partition priority control. According to the real-time power consumption and priority of the grid partitions, it is possible to ensure the brightness and clarity of key, core or high-value display areas first when the peak power consumption or heat is too high. By appropriately reducing the current and duty cycle of low-priority areas, the overall energy consumption can be kept at a controllable level, reducing the risk of overheating and prolonging the service life of the device; comprehensively improve the display quality and service life. This solution can reduce the device's energy consumption and heat load, delay the aging speed of components, and thus improve the overall service life and stability of the LED display screen through comprehensive strategies such as dead pixel compensation, dynamic refresh rate adjustment, and partition priority management while maintaining the display quality and viewing experience. Brief Description of the Drawings

[0020] Figure 1 Schematic diagram of a control method for an intelligent POB module according to the present invention. Detailed implementation manners

[0021] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0022] In the following description, many specific details are set forth in order to fully understand the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments, and are not intended to limit the present invention.

[0024] An embodiment of the present invention, an intelligent POB module control method, the method includes: Detecting the state data of each pixel point of the LED display screen through an optical sensing array, obtaining the damaged pixel points in the LED display screen based on the state data of each pixel point, recording the damaged pixel points in a storage unit, and maintaining a bad point position table in the storage unit; Compensating the brightness of the damaged pixel points through the neighboring pixel points according to the damage situation of the damaged pixel points; While compensating, predicting the display content of the LED display screen through AI, and dynamically driving the refresh rate of the LED display screen according to the predicted display content; Dividing the LED display screen into grid partitions according to the physical pixel density, sorting the grid partitions according to the priority, calculating the real-time power consumption of each partition after refreshing the LED display screen, and if the real-time power consumption is greater than a preset threshold, reducing the current and duty cycle of the low-priority area.

[0025] The working principle and effect of the above technical solution are: bad point detection and recording, optical sensing array, deploying a high-density photosensitive sensor array on the back or surface of the LED display screen, collecting the light-emitting state data of each pixel point in real time, and accurately positioning the damaged pixels through a threshold comparison algorithm; recording the bad point coordinates and damage types in a non-volatile memory to form a global bad point map; Display content prediction: Capture the image, video stream, and metadata (resolution, frame rate) of the current frame through an HDMI signal analyzer, and combine with the historical content pattern; Extract spatio-temporal features (such as motion vectors, color complexity), and predict the content type (such as text, video, animation) and dynamic characteristics (such as motion speed) of the next 1-10 frames. Refresh rate mapping: The dynamic mapping table adjusts the refresh rate according to the predicted content (such as 60Hz for text scenes, 240Hz for game scenes), matching the human eye's critical fusion frequency (CFF), and realizes hardware-level control of refresh rate switching through FPGA, with a switching delay <5ms; Calculate the priority by combining the content type (such as text 0.8, video 0.5), position weight (central Gaussian decay), and historical activity (update frequency normalization), and calculate the partition power consumption in real time. If the total power consumption exceeds 80% of the rated value, trigger a power consumption reduction process, unload the low-priority areas, select 20% of the partitions in ascending order of priority, and reduce the power consumption by adjusting the current and duty cycle. Improve display reliability and reduce screen damage caused by dead pixels. Through real-time detection by the optical sensing array, dead pixels can be quickly discovered and recorded; On this basis, neighborhood compensation is performed, which can effectively reduce the impact of dead pixels on the overall display screen, ensuring the viewing effect and display quality; Automatic dead pixel processing improves operation and maintenance efficiency. The traditional method often relies on manual inspection or low-frequency monitoring methods. This technical solution uses the sensing array to automatically detect and record the positions of dead pixels, eliminating the cumbersome steps of manual inspection. At the same time, the compensation scheme can be started in a timely manner after the occurrence of dead pixels, reducing maintenance and replacement costs and improving operation and maintenance efficiency; Cooperate with AI prediction to reduce unnecessary energy consumption and improve dynamic display quality. Predict the upcoming content through AI, and dynamically adjust the refresh rate according to the actual scene: Output high refresh rates for rapidly changing images to ensure smoothness and clarity; For relatively static images, the refresh rate can be reduced, reducing power consumption without affecting the visual effect and extending the service life of the LED; Partition priority control realizes precise energy consumption management. Regulate according to the real-time power consumption and priority of the grid partitions. When the peak power consumption or heat is too high, the brightness and clarity of critical, core, or high-value display areas can be preferentially guaranteed. By appropriately reducing the current and duty cycle of low-priority areas, the overall energy consumption can be kept at a controllable level, reducing the risk of overheating and extending the device life; Comprehensively improve the display quality and service life. This solution can reduce the device's energy consumption and heat load, delay the aging speed of components, and thus improve the overall service life and stability of the LED display screen through comprehensive strategies such as dead pixel compensation, dynamic refresh rate adjustment, and partition priority management while maintaining the display quality and viewing experience.

[0026] In one embodiment of the present invention, the state data of each pixel of the LED display screen is detected by an optical sensing array, and the damaged pixels in the LED display screen are obtained based on the state data of each pixel, including: Detect the status data of each pixel of the LED display screen through an optical sensing array, and the pixel status data includes the brightness of the pixel; Compare the actually detected brightness value with the expected display brightness value. If the difference between the actually detected brightness value and the expected display brightness value exceeds the first preset threshold, mark the pixel as a pixel with abnormal brightness; Record the coordinates of the pixels with abnormal brightness into the storage unit, and maintain a real-time updated bad pixel position table in the storage unit.

[0027] The working principle and effect of the above technical solution are as follows: An optical sensing array is arranged in front of (or behind) the LED display screen to perform real-time or periodic brightness detection on each pixel of the screen to obtain the actual brightness value; compare with the expected brightness to determine abnormal brightness, compare the actually detected brightness value of the pixel with the preset expected display brightness value. If the difference exceeds the first preset threshold, it is regarded that the pixel has abnormal brightness; record the abnormal pixels and maintain the bad pixel position table, store the coordinate information of the pixels determined to have abnormal brightness into the storage unit, and correspondingly maintain a real-time updated bad pixel position table for subsequent compensation or maintenance decisions. Discover bad pixels in a timely and accurate manner. By independently detecting each pixel through the optical sensing array, the brightness data can be accurately collected; once the pixel brightness deviates from the expected value, its abnormality can be determined and recorded immediately; reduce the difficulty of manual inspection and maintenance costs. The bad pixels of traditional LED display screens often rely on manual visual inspection or low-frequency testing, which is time-consuming and laborious. This solution realizes automatic detection, significantly reduces the manual workload, and can monitor the pixel health status for a long time and stably; improve the display quality and visual experience, can quickly locate and record bad pixels, provide a basis for subsequent automatic or manual compensation and repair, avoid the long-term existence of bad pixels affecting the overall display effect, and improve the viewing experience of viewers or users; by using the optical sensing array for brightness detection and threshold comparison, the brightness abnormal pixels in the LED display screen can be quickly identified and recorded, realizing automatic and precise bad pixel management and maintenance, and ultimately effectively ensuring the display quality and use reliability of the LED display screen.

[0028] In one embodiment of the present invention, the brightness of a bad pixel is compensated by neighboring pixels, including: Check whether the four pixels around the bad pixel are in the bad pixel table. The four pixels include the upper left B11, the upper right B21, the lower left B12, and the lower right B22. If they are all normal pixels, calculate the bad pixel compensation brightness in advance through the bad pixel brightness compensation model in the FPGA. Specifically, the bad pixel brightness compensation model is:

[0029] Among them, represents the coordinates of the bad pixel to be compensated, Indicates the boundary coordinates of the four surrounding normal pixels (for example, the upper left corner is , and the lower right corner is ), , and represent the brightness values of the four normal pixels; If there are multiple neighboring bad pixels in the neighborhood of the bad pixel (such as both B11 and B21 are bad): Regional averaging method: Taking the bad pixel as the center, expand the search range to a 3×3 area, and take the average value of the normal pixels:

[0030] If the reference pixels still contain bad pixels after expansion, perform recursive compensation on the bad pixels of the reference pixels. First, repair their values, and then use them for the interpolation of the original bad pixels. If B11 is a bad pixel, first compensate B11 through the surrounding normal pixels, and then substitute the compensated value of B11 into the original formula to calculate B(x,y); If the deviation between the compensated value and the actual value exceeds the preset threshold, add the pixel point with compensation failure to the high-priority maintenance list. When there are pixel points in the high-priority maintenance queue, trigger a reminder for manual maintenance inspection.

[0031] The working principle and effect of the above technical solution are as follows: Fine and flexible bad pixel compensation. Through the interpolation model based on four normal neighboring pixels, more accurate bad pixel compensation can be achieved when the bad pixels are scattered and the neighboring pixels are relatively healthy; if there are also some bad pixels in the neighboring pixels, the system can adopt the regional averaging method and recursive compensation method to ensure that a relatively smooth transition of the screen brightness can still be maintained when the bad pixel density is relatively high; significantly improve the overall visual experience of the display, timely correct the brightness of the bad pixels, greatly reduce the screen defects caused by single or local bad pixels, and for large-size LED displays with high visual requirements, it can effectively mask or weaken the visual impact of bad pixels and improve the viewing comfort; automated and hierarchical maintenance mechanism. The system indexes the "compensation failure" pixels to the high-priority maintenance list according to the deviation of the compensation results, and issues a manual maintenance prompt when necessary; this hierarchical maintenance mechanism enables maintenance personnel to prioritize solving the most serious or irreparable bad pixels automatically, saving operation and maintenance costs and improving work efficiency; extend the service life of the LED module and the drive system. Through fast and effective bad pixel detection and compensation technology, the good picture quality of the LED display can be maintained for a longer time, and the replacement cycle of some modules can be postponed; combined with the high-priority maintenance list, reduce unnecessary overall disassembly and replacement, which helps to save operation and maintenance costs and extend the overall life of the equipment; through the comprehensive strategy of "multi-mode interpolation compensation + neighborhood expansion + recursive processing + deviation detection and hierarchical maintenance", rapid and accurate repair and hierarchical operation and maintenance of bad pixels on the LED display are realized, greatly improving the display quality and reliability, and reflecting the advantages of high efficiency and low cost in management and manpower operation and maintenance.

[0032] In one embodiment of the present invention, while compensating, the display content of the LED display is predicted by AI, and the refresh rate of the LED display is adjusted according to the predicted display content, including: The display content video stream and metadata are collected in real time through the HDMI signal analyzer of the LED display. The metadata includes the resolution and frame rate of the display content, and the timestamp is recorded synchronously; Video stream data and image data are obtained from the public dataset. The video stream data covers action, game, and movie scenes; The frame rate of the video stream data is normalized, the inter-frame motion offset is compensated by the optical flow method, the resolution and frame rate are encoded as feature vectors, and are concatenated with the video stream data; The content type of the concatenated video stream data is labeled, a 3D-CNN + LSTM model is constructed, and the labeled data is divided into a training set, a validation set, and a test set according to a ratio of 7:1:2; The training set data is input into the 3D-CNN + LSTM model, and trained for 200 epochs. The accuracy loss is monitored on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, the training is terminated early. The loss function of the validation set is total loss = α·classification loss + β·regression loss, where α and β are weight coefficients, the classification loss is the cross-entropy loss based on classifying the video content type, and the regression loss is the mean square error loss based on the motion speed and color complexity of the video content; The trained model is obtained, the real-time display content is input into the trained model, the motion speed and color complexity of the LED display content are predicted, and based on the predicted motion speed and color complexity, the preset refresh rate mapping table is queried.

[0033] The working principle and effects of the above technical solution are as follows: The HDMI signal analyzer captures in real time the video stream being played on the LED display and the corresponding metadata information (resolution, frame rate, timestamp, etc.). This process ensures the complete monitoring and precise synchronization of the displayed content, facilitating subsequent analysis of the resolution, frame rate, and motion characteristics of the video; obtaining and preprocessing the training data, retrieving rich and diverse video stream data (including scenes such as actions, games, movies, etc.) from public datasets to ensure the adaptability of the model to various types of video content; standardizing the frame rate of the captured video stream and using the optical flow method to estimate the inter-frame motion offset, thereby unifying the video time sequence and motion characteristics; encoding the resolution and frame rate into feature vectors and concatenating them with the image / video data compensated by optical flow to enrich the input feature dimensions; annotating the concatenated video stream data with the content type (actions, static, games, movies, etc.), and constructing classification targets and regression targets respectively; training the annotated data using a 3D-CNN + LSTM model, and splitting the data according to the ratio of 7:1:2 (training set: validation set: test set); setting 200 epochs for the training process and monitoring the accuracy and loss on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, the training is terminated early; the overall loss function consists of two parts: classification loss (cross-entropy loss based on the video content type), and regression loss (mean squared error based on the motion speed and color complexity of the video content), which are combined with α and β as weight coefficients; inputting the video stream of the displayed content obtained in real time into the trained 3D-CNN + LSTM model, and the model outputs the motion speed and color complexity; the system dynamically selects an appropriate refresh rate for the LED display according to the preset mapping table of "motion speed + color complexity" → "refresh rate", thereby achieving intelligent control. By improving the matching degree of the LED display content and the refresh rate, and obtaining the motion speed and color complexity of the video content through the model, it is possible to give a higher refresh rate to high-speed motion scenes or content with drastic color changes; for relatively static or low-motion scenes, the refresh rate can be appropriately reduced to avoid waste of energy consumption and hardware performance; enhancing the user's visual experience and viewing experience, increasing the refresh rate in high-speed scenes can significantly reduce motion blur and screen tearing, improving the smoothness; in complex color scenes, the appropriate frame rate can also be accurately matched to effectively ensure the restoration degree and visual quality of the image; saving energy consumption and reducing heat accumulation, dynamic refresh rate adjustment can effectively reduce unnecessary high-frequency refreshing on the premise of ensuring the display effect, thereby saving power consumption; at the same time, it can reduce the power load of the LED driver and related circuits, extending the service life of the device.By combining real-time HDMI video stream analysis with deep learning, using methods such as optical flow to extract video motion features, and then through classification and regression prediction of the 3D-CNN + LSTM model, the goal of dynamically adjusting the refresh rate of the LED display according to the content is finally achieved. This method can not only effectively improve the visual experience, but also take into account energy conservation and stability. In one embodiment of the present invention, according to the physical pixel density, the LED display is divided into grid partitions, the grid partitions are sorted according to priority, and the real-time power consumption of each partition is calculated after refreshing the LED display. If the real-time power consumption is greater than a preset threshold, the current and duty cycle of the low-priority area are reduced, including: The LED display screen is divided into grid partitions, and each partition contains the same number of pixels; The grid partitions are sorted according to priority, and the priority is calculated through a priority model. Specifically, the priority model is:

[0034] Among them, represents the content type weight coefficient, represents the content type mapping where the pixel coordinates (x, y) are located, represents the position weight coefficient, (x, y) represents the pixel coordinates, and σ represents the attenuation parameter, represents the number of times the content in the partition has changed in the past 1 hour, represents the number of times the content in the smallest partition within the entire LED display screen has changed, represents the number of times the content in the largest partition within the entire LED display screen has changed; After refreshing the LED display, the real-time power consumption of each partition is calculated. If the real-time power consumption is greater than a preset threshold, the current and duty cycle of the low-priority area are reduced.

[0035] The working principle and effects of the above technical solution are as follows: For refined energy consumption control, through grid management and differential priority scheduling, the system can implement current reduction and duty cycle strategies for low-priority areas under high load, thereby accurately controlling the overall power consumption of the screen; reduce energy consumption as much as possible while meeting the core display requirements, reduce unnecessary power expenditure and heat generation, and ensure the display quality of key areas. High-priority partitions (such as important information, central images, or high-frequency change areas) can maintain normal or higher brightness and refresh levels; even when the total power consumption exceeds the limit, it can prioritize ensuring the integrity of the overall visual effect and information transmission; improve operation and maintenance efficiency, reduce the load of the LED driving device while reducing energy consumption, and extend the service life of the corresponding modules and electronic components; enhance the user experience and viewing comfort. Since the [brightness / refresh rate] of the core image and high-priority content is prioritized, users have a good visual perception; although the power consumption of low-priority areas is reduced, it has little impact on the overall viewing effect and can effectively avoid problems such as large-scale brightness attenuation or image flickering; through the comprehensive method of "grid partitioning + priority model + real-time power consumption monitoring + dynamic scheduling", on the one hand, it realizes precise control of the local power consumption of the LED screen, and on the other hand, it prioritizes ensuring the display quality of key image areas when the total power consumption exceeds the limit. This method has significant application value in energy conservation, maintaining the effect of the core image, and extending the device life, and reflects good flexibility and scalability.

[0036] An embodiment of the present invention, an intelligent POB module integration platform, the platform includes: A storage bad pixel point module, configured to detect the state data of each pixel point of the LED display screen through an optical sensing array, obtain the damaged pixel points in the LED display screen based on the state data of each pixel point, record the damaged pixel points in a storage unit, and maintain a bad point position table in the storage unit; A compensation module, configured to compensate the brightness of the bad pixel points through neighboring pixel points according to the bad conditions of the bad pixel points; A prediction adjustment module, configured to predict the display content of the LED display screen through AI while compensating, and dynamically drive the refresh rate of the LED display screen according to the predicted display content; A priority-based adjustment module, configured to divide the LED display screen into grid partitions according to the physical pixel density, sort the grid partitions according to the priority, calculate the real-time power consumption of each partition after refreshing the LED display screen, and if the real-time power consumption is greater than a preset threshold, reduce the current and duty cycle of the low-priority area.

[0037] An embodiment of the present invention, the storage bad pixel point module includes: Detect the status data of each pixel of the LED display screen through an optical sensing array, and the pixel status data includes the brightness of the pixel; Compare the actually detected brightness value with the expected display value of the brightness. If the difference between the actually detected brightness value and the expected display value of the brightness exceeds the first preset threshold, mark the pixel as a pixel with abnormal brightness; Record the coordinates of the pixels with abnormal brightness in the storage unit, and maintain a real-time updated bad pixel position table in the storage unit.

[0038] In one embodiment of the present invention, the compensation module includes: A storage bad pixel brightness compensation model module for checking whether the four pixels around the bad pixel are in the bad pixel table. The four pixels include the upper left B11, the upper right B21, the lower left B12, and the lower right B22. If they are all normal pixels, the bad pixel compensation brightness is calculated in advance through the bad pixel brightness compensation model in the FPGA. Specifically, the bad pixel brightness compensation model is:

[0039] Wherein, represents the coordinates of the bad pixel to be compensated, represents the boundary coordinates of the four surrounding normal pixels, (for example, the upper left corner is , and the lower right corner is ), , and represent the brightness values of the four normal pixels; An area averaging module for, if there are multiple neighboring bad pixels in the neighboring pixels of the bad pixel (such as both B11 and B21 are bad), expanding the search range to a 3×3 area centered on the bad pixel and taking the average value of the normal pixels:

[0040] A recursive compensation module for, if the reference pixels still contain bad pixels after expansion, performing recursive compensation on the bad pixels of the reference pixels, first repairing their values and then using them for the interpolation of the original bad pixels. If B11 is a bad pixel, first compensate B11 through the surrounding normal pixels, and then substitute the compensated value of B11 into the original formula to calculate B(x,y); A trigger for manual maintenance reminder module for, if the deviation between the compensation value and the actual value exceeds the preset threshold, adding the pixel with compensation failure to the high-priority maintenance list, and triggering a reminder for manual maintenance inspection when there are pixels in the high-priority maintenance queue.

[0041] In one embodiment of the present invention, the prediction adjustment module includes: A real-time display content acquisition module, which is used to acquire the display content video stream and metadata in real time through the HDMI signal analyzer of the LED display screen. The metadata includes the resolution and frame rate of the display content, and the timestamp is synchronously recorded; An open data acquisition module, which is used to acquire video stream data and image data from an open data set. The video stream data covers action, game, and movie scenes; A preprocessing module, which is used to standardize the frame rate of the video stream data, compensate for the inter-frame motion offset by the optical flow method, encode the resolution and frame rate into feature vectors, and splice them with the video stream data; A labeling and partitioning module, which is used to label the content type of the spliced video stream data, construct a 3D-CNN + LSTM model, and divide the labeled data into a training set, a validation set, and a test set according to a ratio of 7:1:2; A training module, which is used to input the training set data into the 3D-CNN + LSTM model, train for 200 epochs, monitor the accuracy loss on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, the training is terminated in advance. The loss function of the validation set is the total loss = α·classification loss + β·regression loss, where α and β are weight coefficients, the classification loss is the cross-entropy loss for classifying based on the video content type, and the regression loss is the mean square error loss based on the motion speed and color complexity of the video content; A query module, which is used to obtain the trained model, input the real-time display content into the trained model, predict the motion speed and color complexity of the LED display content, and query the preset refresh rate mapping table based on the predicted motion speed and color complexity.

[0042] In an embodiment of the present invention, the priority adjustment module includes: A partitioning module, which is used to divide the LED display screen into grid partitions, and each partition contains the same number of pixels; A priority sorting module, which is used to sort the grid partitions according to the priority. The priority is calculated by a priority model. Specifically, the priority model is:

[0043] Wherein, represents the content type weight coefficient, represents the content type mapping where the pixel coordinates (x, y) are located, represents the position weight coefficient, (x, y) represents the pixel coordinates, and σ represents the attenuation parameter, represents the number of times the content in the partition has changed within the past 1 hour, represents the number of times the content in the smallest partition within the entire LED display screen has changed, Indicates the maximum number of content changes in the largest partition within the entire LED display screen; The current and duty cycle adjustment module is used to calculate the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than the preset threshold, the current and duty cycle of the low-priority area are reduced.

[0044] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and its equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. An intelligent POB module control method, characterized in that, The method includes: Detecting the state data of each pixel of the LED display through an optical sensing array, obtaining the damaged pixels in the LED display based on the state data of each pixel, recording the damaged pixels in a storage unit, and maintaining a bad pixel position table in the storage unit; Compensating the brightness of the bad pixels through neighboring pixels according to the bad conditions of the bad pixels; While compensating, predicting the display content of the LED display through AI, and dynamically driving the refresh rate of the LED display according to the predicted display content; Dividing the LED display into grid partitions according to the physical pixel density, sorting the grid partitions according to the priority, calculating the real-time power consumption of each partition after refreshing the LED display, and if the real-time power consumption is greater than the preset threshold, reducing the current and duty cycle of the low-priority area.

2. The intelligent POB module control method according to claim 1, characterized in that Detecting the state data of each pixel of the LED display through an optical sensing array, and obtaining the damaged pixels in the LED display based on the state data of each pixel, including: Detecting the state data of each pixel of the LED display through an optical sensing array, where the pixel state data includes the brightness of the pixel; Comparing the actually detected brightness value with the expected display brightness value, and if the difference between the actually detected brightness value and the expected display brightness value exceeds the first preset threshold, marking it as a pixel with abnormal brightness; Recording the coordinates of the pixels with abnormal brightness in the storage unit, and maintaining a bad pixel position table that is updated in real time in the storage unit.

3. The intelligent POB module control method according to claim 1, wherein Compensating the brightness of the bad pixels through neighboring pixels, including: Checking whether the four pixels around the bad pixel are in the bad pixel table. The four pixels include the upper left B11, the upper right B21, the lower left B12, and the lower right B22. If all are normal pixels, the bad pixel compensation brightness is calculated in advance through the bad pixel brightness compensation model in the FPGA. Specifically, the bad pixel brightness compensation model is: ； Among them, represents the coordinates of the defective pixel to be compensated, represents the boundary coordinates of the four surrounding normal pixels, , and represent the brightness values of the four normal pixels; If there are multiple neighboring bad pixels in the neighboring pixels of the bad pixel, expand the search range to a 3×3 area centered on the bad pixel, and take the average value of the normal pixels; ； If the reference pixels still contain bad pixels after expansion, recursively compensate the bad pixels of the reference pixels, first repair their values, and then use them for the interpolation of the original bad pixels. If B11 is a bad pixel, first compensate B11 through the normal pixels around it, and then substitute the compensation value of B11 into the original formula to calculate B(x,y); If the deviation between the compensation value and the actual value exceeds the preset threshold, add the pixel with compensation failure to the high-priority maintenance list. When there are pixels in the high-priority maintenance queue, trigger a reminder for manual maintenance inspection.

4. The intelligent POB module control method according to claim 1, wherein While compensating, predicting the display content of the LED display through AI, and adjusting the refresh rate of the LED display according to the predicted display content, including: Real-time collecting the display content video stream and metadata through the HDMI signal analyzer of the LED display. The metadata includes the resolution and frame rate of the display content, and synchronously recording the time stamp; Obtaining video stream data and image data from a public data set. The video stream data covers action, game, and movie scenes; Normalize the frame rate of the video stream data, compensate for the inter-frame motion offset by the optical flow method, encode the resolution and frame rate as feature vectors, and splice them with the video stream data; Annotate the content type of the spliced video stream data, construct a 3D-CNN + LSTM model, and divide the annotated data into a training set, a validation set, and a test set according to a ratio of 7:1:2; Input the training set data into the 3D-CNN + LSTM model, train for 200 epochs, monitor the accuracy loss on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, terminate the training early. The loss function of the validation set is total loss = α·classification loss + β·regression loss, where α and β are weight coefficients, the classification loss is the cross-entropy loss for classifying based on the video content type, and the regression loss is the mean square error loss based on the motion speed and color complexity of the video content; Obtain the trained model, input the real-time display content into the trained model, predict the motion speed and color complexity of the LED display content, and query the preset refresh rate mapping table based on the predicted motion speed and color complexity; 5. The intelligent POB module control method according to claim 1, characterized in that Divide the LED display screen into grid partitions according to the physical pixel density, sort the grid partitions according to the priority, calculate the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than the preset threshold, reduce the current and duty cycle of the low-priority areas, including: Divide the LED display screen into grid partitions, and each partition contains the same number of pixels; Sort the grid partitions according to the priority, and the priority is calculated by the priority model. Specifically, the priority model is: ； Among them, represents the content type weight coefficient, represents the content type mapping where the pixel coordinates (x, y) are located, represents the position weight coefficient, (x, y) represents the pixel coordinates, and σ represents the attenuation parameter, represents the number of times the content in the partition has changed within the past 1 hour, represents the change number weight coefficient, represents the number of times the content in the smallest partition within the entire LED display screen has changed, represents the number of times the content in the largest partition within the entire LED display screen has changed; Calculate the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than the preset threshold, reduce the current and duty cycle of the low-priority areas.

6. An intelligent POB module integration platform, characterized in that, The system includes: A storage bad pixel point module for detecting the state data of each pixel point of the LED display screen through an optical sensing array, obtaining the damaged pixel points in the LED display screen based on the state data of each pixel point, recording the damaged pixel points in the storage unit, and maintaining a bad point position table in the storage unit; A compensation module for compensating the brightness of the bad pixel points through the neighboring pixel points according to the bad situation of the bad pixel points; A prediction adjustment module for predicting the display content of the LED display screen through AI while compensating, and dynamically driving the refresh rate of the LED display screen according to the predicted display content; A priority-based adjustment module for dividing the LED display screen into grid partitions according to the physical pixel density, sorting the grid partitions according to the priority, calculating the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than the preset threshold, reduce the current and duty cycle of the low-priority areas.

7. The intelligent POB module integration platform according to claim 6, characterized in that, The storage bad pixel point module includes: Detect the state data of each pixel point of the LED display screen through an optical sensing array, and the pixel point state data includes the brightness of the pixel point; Compare the actually detected brightness value with the expected display brightness value. If the difference between the actually detected brightness value and the expected display brightness value exceeds the first preset threshold, mark the pixel points with abnormal brightness; Record the coordinates of the pixel points with abnormal brightness in the storage unit, and maintain a real-time updated bad pixel position table in the storage unit.

8. The intelligent POB module integration platform according to claim 6, characterized in that, The compensation module includes: A module for storing the bad pixel brightness compensation model, which is used to check whether the four pixels around the bad pixel are in the bad pixel table. The four pixels include the upper left B11, the upper right B21, the lower left B12, and the lower right B22. If they are all normal pixels, calculate the bad pixel compensation brightness in advance through the bad pixel brightness compensation model in the FPGA. Specifically, the bad pixel brightness compensation model is: ； Among them, represents the coordinates of the defective pixel to be compensated, represents the boundary coordinates of the four surrounding normal pixels, , and represent the luminance values of the four normal pixels; A regional average module, which is used to expand the search range to a 3×3 area centered on the bad pixel and take the average value of the normal pixels if there are multiple neighboring bad pixels in the neighboring pixels of the bad pixel; ； A recursive compensation module, which is used to recursively compensate the bad pixels of the reference pixels if the reference pixels still contain bad pixels after expansion. First, repair its value and then use it for the interpolation of the original bad pixels. If B11 is a bad pixel, first compensate B11 through the normal pixels around it, and then substitute the compensation value of B11 into the original formula to calculate B(x,y); A module for triggering an artificial maintenance reminder, which is used to add the pixel points with compensation failure to the high-priority maintenance list if the deviation between the compensation value and the actual value exceeds the preset threshold. When there are pixel points in the high-priority maintenance queue, trigger a reminder for artificial maintenance inspection.

9. The intelligent POB module integration platform according to claim 6, wherein The prediction adjustment module includes: A module for collecting real-time display content, which is used to collect the display content video stream and metadata in real time through the HDMI signal analyzer of the LED display screen. The metadata includes the resolution and frame rate of the display content, and synchronously record the time stamp; A module for obtaining public data, which is used to obtain video stream data and image data from the public data set. The video stream data covers action, game, and movie scenes; A preprocessing module, which is used to standardize the frame rate of the video stream data, compensate the inter-frame motion offset through the optical flow method, encode the resolution and frame rate into feature vectors, and splice them with the video stream data; A labeling and partitioning module, which is used to label the content type of the spliced video stream data, construct a 3D-CNN + LSTM model, and divide the labeled data into a training set, a validation set, and a test set according to a ratio of 7:1:2; A training module, which is used to input the training set data into the 3D-CNN + LSTM model, train for 200 epochs, monitor the accuracy loss on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, terminate the training in advance. The loss function of the validation set is the total loss = α·classification loss + β·regression loss, where α and β are weight coefficients, the classification loss is the cross-entropy loss for classifying based on the video content type, and the regression loss is the mean square error loss based on the motion speed and color complexity of the video content; A query module, which is used to obtain a trained model, input the real-time display content into the trained model, predict the motion speed and color complexity of the LED display content, and query a preset refresh rate mapping table based on the predicted motion speed and color complexity.

10. The intelligent POB module integration platform according to claim 6, characterized in that, The priority adjustment module includes: A partitioning module, which is used to divide the LED display screen into grid partitions, and each partition contains the same number of pixels; A priority sorting module, which is used to sort the grid partitions according to the priority, and the priority is calculated by a priority model. Specifically, the priority model is: ； Among them, represents the content type weight coefficient, represents the content type mapping where the pixel coordinates (x, y) are located, represents the position weight coefficient, (x, y) represents the pixel coordinates, and σ represents the attenuation parameter, represents the number of times the content in the partition has changed within the past 1 hour, represents the change number weight coefficient, represents the number of times the content in the smallest partition within the entire LED display screen has changed, represents the number of times the content in the largest partition within the entire LED display screen has changed; An adjusting current and duty cycle module, which is used to calculate the real-time power consumption of each partition after refreshing the LED display screen. If the real-time power consumption is greater than a preset threshold, the current and duty cycle of the low-priority area are reduced.

Images