A large language model-based display panel gamma correction algorithm automatic generation method and system

Abstract

The application discloses a display panel Gamma correction algorithm automatic generation method and system based on a large language model. The method comprises the following steps: collecting luminance data of each gray scale of a to-be-tested panel; sending the structured data set obtained after preprocessing into a large language model; calling an independent numerical analysis tool by the large language model to complete curve fitting, luminance statistics and deviation quantization operation, and receiving the structured quantization indexes output by the tool; generating a correction analysis report based on the quantization indexes, and automatically generating a Gamma correction algorithm code; deploying the algorithm for retesting, calculating the correction deviation by the numerical tool, and returning the quantized deviation data to the large language model for iterative optimization if the deviation exceeds the standard; and simultaneously building a knowledge base and combining RAG to search, reuse and fit historical measurement and fitting data. The application distinguishes the logical reasoning ability of the large language model from the high-precision calculation ability of the numerical tool, solves the defect of insufficient numerical operation precision of pure LLM, converts the artificial Gamma correction into a full-automatic collaborative process, greatly shortens the development cycle of new product correction algorithms, gets rid of the dependence on the experience of experienced technicians, guarantees the consistency of correction precision of different batches of panels, and can be widely used in automatic production line detection and correction scenes of OLED, AMOLED, LCD and other display panels.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of display panel manufacturing and artificial intelligence. Specifically, it relates to a method and system for automatically analyzing optical measurement data of display panels, mining data patterns, and generating and optimizing Gamma correction algorithms using large language models to schedule numerical analysis tools. Background Technology

[0002] In the manufacturing process of various display panels such as OLED, AMOLED, and LCD, gamma calibration is a core process for controlling screen display effects. Its purpose is to ensure that the output brightness of each grayscale level of the panel conforms to the standard gamma 2.2 curve, guaranteeing the color reproduction effect and visual consistency of the entire screen. Traditionally, the entire gamma calibration process is done manually: workers use luminance meters and color analyzers to collect brightness data of the entire grayscale from 0 to 255, and then use their professional experience to analyze the inherent photoelectric differences of the panel, such as the luminous efficiency of RGB sub-pixels and the inconsistency of threshold voltage. Then, they manually select and fit a mathematical model, manually adjust and generate binding point parameters or calibration lookup tables, and finally burn the calibration scheme into the production line testing equipment for batch calibration.

[0003] This manual approach has several significant drawbacks: First, it is highly dependent on experienced process engineers. Different panel substrates, driver ICs, and pixel circuit structures result in varying Gamma characteristics. Each new panel requires manual data collection, pattern analysis, and algorithm debugging, a process that can take days or even weeks. Second, the calibration results are unstable. The selection of the fitting model and parameter adjustments rely entirely on the subjective experience of engineers. The accuracy of calibration algorithms output by different personnel varies, making it impossible to guarantee consistency in the display of batch panels. Third, iterative optimization is inefficient. If the initially generated calibration curve deviates significantly, repeated manual parameter modifications and retesting are required, leading to long iteration cycles and slowing down the pace of new product mass production. Fourth, industry experience cannot be transferred and reused. The analytical approaches and adaptation solutions accumulated by engineers during debugging remain only in their personal possession, lacking standardized storage media. Subsequent new panel development cannot directly reuse historically mature experience.

[0004] Existing gamma correction schemes based solely on large language models have inherent limitations: large language models only possess text understanding, logical reasoning, and code generation capabilities, lacking the ability for high-precision numerical fitting, statistical index calculation, and deviation quantification. If relying entirely on LLM to autonomously derive gamma coefficients and brightness differences, the calculation accuracy is poor, and the scheme cannot be stably reproduced. There is an urgent need for an automated gamma correction generation method for display panels that integrates LLM with professional numerical analysis tools, eliminates human intervention, automatically mines the patterns in panel optical data, and iteratively generates a high-precision gamma correction algorithm in a closed-loop manner. Current numerical tools such as SciPy and NumPy can perform high-precision curve fitting and statistical calculations, while large language models are responsible for scheduling tools, interpreting quantification results, and outputting correction strategies. The combination of these two provides a feasible technical path for this invention to overcome the shortcomings of traditional manual correction and pure LLM numerical calculations. Summary of the Invention

[0005] To address the shortcomings of existing Gamma correction development methods, such as high reliance on manual labor, long development cycles, unstable algorithm quality, inability to retain and reuse experience, insufficient accuracy of numerical calculations using pure large language models, and low efficiency in iterative debugging, this invention provides a method and system for automatically generating Gamma correction algorithms for display panels based on large language models. This method utilizes large language models to schedule independent numerical analysis tools to complete all quantization calculations, achieving fully automated analysis, algorithm generation, and closed-loop iterative optimization. It also constructs a reusable correction knowledge base, reducing reliance on manual labor, shortening new product development cycles, and standardizing correction accuracy.

[0006] To address the aforementioned problems and achieve the above objectives, the first aspect of this invention provides an automatic generation method for a display panel Gamma correction algorithm based on a large language model, comprising the following steps: Step S1: Data acquisition step: Control the display panel under test to sequentially display multiple preset grayscale test images, and acquire brightness data at each grayscale level through an optical measurement device to obtain a raw grayscale-brightness dataset; Step S2: Data preprocessing step: Preprocess the raw grayscale-brightness dataset, including outlier removal, data normalization, and format standardization, to generate a structured dataset to be analyzed; Step S3: Data analysis and pattern discovery step: Input the dataset to be analyzed and panel model information into a pre-configured large language model; The large language model generation tool calls instructions to send data to a preset numerical analysis tool for quantization calculation; The numerical analysis tool completes piecewise curve fitting, brightness statistics, and outlier detection, and sends back the fitted Gamma coefficient, brightness fluctuation index of each interval, and RGB values ​​to the large language model. The data includes structured quantization results such as channel differences and abnormal grayscale coordinates; the large language model generates a data analysis report based on all quantization indicators output by the tool, containing grayscale interval trends, nonlinear characteristics, channel differences, and binding point placement strategies; Step S4: Algorithm generation step: The large language model automatically generates Gamma correction algorithm code based on the data analysis report, the optimal fitting parameters output by the numerical analysis tool, and the preset target Gamma curve parameters. The algorithm code includes binding point selection strategies, interpolation calculation methods, and correction compensation parameters; Step S5: Automatic verification step: The Gamma correction algorithm code is deployed to the test environment, the correction operation is performed on the display panel under test, and the corrected grayscale-brightness data is re-acquired; the corrected data is sent to the numerical analysis tool to calculate the brightness deviation value between the correction result and the target Gamma curve; Step S6: Iterative Optimization Step: When the deviation value is greater than the preset threshold, the corrected measurement data, the deviation distribution index output by the numerical analysis tool, and the analysis results of the previous algorithm are fed back to the large language model. The large language model then adjusts and optimizes the algorithm based on the quantified deviation data, and repeats step S5 until the deviation value is no greater than the preset threshold, outputting the final Gamma correction algorithm. Further, in step S3, the large language model loads data through structured text / tables and performs multi-dimensional analysis using scene-specific prompts. The analysis dimensions include low grayscale brightness jump detection, mid-to-high grayscale smoothness evaluation, and RGB sub-pixel consistency comparison.

[0007] Furthermore, in step S3, the large language model loads data through structured text / tables, calls scene-specific prompts to generate numerical tool scheduling instructions, and completes multi-dimensional analysis based on the quantitative statistical indicators output by the numerical analysis tool. The analysis dimensions include low grayscale brightness jump detection, mid-to-high grayscale smoothness evaluation, and RGB sub-pixel consistency comparison.

[0008] Furthermore, in step S4, the numerical analysis tool outputs the fitting errors of multiple sets of mathematical models, and the large language model compares the error indices of each set to automatically match the optimal curve fitting mathematical model, determine the distribution and number of binding points, and output Python / C / C++ executable correction code.

[0009] Furthermore, in step S6, an upper limit for the number of iterations is set. If the limit is exceeded, the large language model will automatically switch the numerical tool scheduling prompt word template or change the fitting model in the numerical analysis tool to avoid the algorithm getting stuck in local optima.

[0010] Furthermore, the method also includes an S7 knowledge base construction step: storing panel hardware parameters, raw measurement data, quantitative indicators output by numerical analysis tools, LLM analysis reports, final algorithms, and iteration logs into the knowledge base; when processing new panels, RAG is used to retrieve similar historical cases to assist in large language model reasoning.

[0011] Furthermore, in S1, the grayscale coverage is 0~255, the sampling interval is ≤4 grayscales; the preset deviation threshold is 1%~5% brightness deviation percentage.

[0012] The second aspect of this invention provides an automatic generation system for display panel Gamma correction algorithms based on a large language model, comprising: a data acquisition module for controlling the display panel under test to display a grayscale test image and driving an optical measurement device to acquire brightness data; a data preprocessing module for performing outlier removal, normalization, and format standardization on the acquired raw data; a large language model analysis module configured with a pre-trained LLM, equipped with a numerical tool call interface, sending data to the numerical analysis tool, receiving quantization calculation results, and performing data pattern inference, algorithm code generation, and iterative optimization; a numerical analysis tool unit that independently completes curve fitting, statistical calculation, and correction deviation quantization, and outputs structured numerical results to the large language model; an algorithm verification module for deploying the correction algorithm; an iteration control module that receives the deviation value output by the numerical tool, determines whether it meets the standard, and if not, assembles feedback data and sends it to the large language model analysis module; and a knowledge base module that stores all historical correction data and supports RAG similar case retrieval.

[0013] Furthermore, the large language model analysis module supports cloud API calls or local open-source model deployment, and has a built-in prompt word management unit that stores dedicated prompt word templates for multiple scenarios such as numerical tool scheduling, data analysis, code generation, and deviation iteration.

[0014] In summary, this invention can significantly shorten the development cycle of new product calibration algorithms. Through the collaborative division of labor between LLM and professional numerical analysis tools: the numerical tools handle high-precision fitting, statistics, and deviation calculation, while the large language model focuses on data interpretation, strategy reasoning, and code generation, reducing the development cycle from several days to several hours, effectively accelerating the mass production and introduction of new panels. At the same time, it can completely reduce the reliance on the personal experience of senior process engineers, saving labor costs by eliminating repeated manual debugging work through standardized processes. The quantitative indicators output by the numerical tools are reproducible and free from human subjective bias, ensuring consistent calibration accuracy for batch panels. Combined with a knowledge base and RAG retrieval mechanism, historical calibration quantitative data and analysis experience can be accumulated and reused, further improving the analysis and processing efficiency of new panels. The entire process achieves full-chain automation, significantly reducing manual intervention and adapting to various batch automated testing application scenarios on production lines. Attached Figure Description

[0015] This invention is accompanied by five accompanying drawings to clearly illustrate the implementation process, system architecture, iterative logic, knowledge base mechanism, and original data collection format.

[0016] Figure 1 is the overall main flowchart of the automatic generation method of display panel Gamma correction algorithm based on large language model of the present invention;

[0017] Figure 2 is a block diagram of the module structure of the automatic generation system of display panel Gamma correction algorithm based on large language model according to the present invention;

[0018] Figure 3 is a flowchart of the algorithm iterative optimization subdivision in an embodiment of the present invention;

[0019] Figure 4 is a schematic diagram of the knowledge base RAG retrieval and reuse mechanism in an embodiment of the present invention;

[0020] Figure 5 is a table showing an example of grayscale-brightness measurement data in an embodiment of the present invention. Detailed Implementation

[0021] Example 1: Method Flow Referring to Figure 1, this embodiment provides an automatic generation method for display panel Gamma correction algorithm based on a large language model, which is applied to the Gamma correction scenario of AMOLED display panel production line.

[0022] Step S1: Data Acquisition The AMOLED panel under test was placed in a dark room, and grayscale test signals were sent to the panel via control software. The test signal covered a grayscale range from 0 to 255, and brightness data was collected every 4 grayscale levels (i.e., 65 grayscale points were collected: 0, 4, 8, 12, ..., 252, 255). A luminance meter (such as a Konica Minolta CA-410) was used to measure the brightness values ​​of the R, G, and B sub-pixel channels at each grayscale, resulting in the dataset shown in Figure 5.

[0023] Step S2: Data Preprocessing Perform the following preprocessing operations on the raw data: (a) Outlier removal: The 3σ criterion or box plot method is used to detect and remove measurement data points that are obviously abnormal. For example, if the brightness value of a grayscale point deviates from the interpolation expectation of adjacent grayscale points by more than 3 times the standard deviation, it is marked as an outlier and replaced with interpolation. (b) Data normalization: The brightness values ​​of each channel are normalized to the [0, 1] interval for easier subsequent analysis. The normalization formula is: L_norm = (L - L_min) / (L_max - L_min); (c) Format standardization: Convert the data into structured text in JSON or CSV format and attach metadata such as panel model, measuring equipment model, and measurement environment parameters.

[0024] Step S3: Data Analysis and Pattern Discovery The preprocessed dataset and dedicated prompts are input into the large language model. The prompts contain numerical analysis tool scheduling instructions. The LLM automatically packages the dataset and calls the local SciPy numerical analysis tool to perform a full set of quantization operations. The numerical analysis tool performs the following operations: 1. Piecewise power function fitting yields the optimal Gamma fitting coefficient of 2.18; 2.3σ Statistical detection of low grayscale transition range and medium-to-high grayscale fluctuation range; 3. Calculate the brightness difference between the R / G / B channels, and find that the B channel has a 15% lower brightness in low grayscale levels; 4. Output the fitting error for each interval, and recommend piecewise interpolation models and binding point positions. The numerical analysis tool sends all the above quantitative indicators back to the large language model in a structured table. After receiving the quantitative data, the large language model generates a standardized analysis report based on the prompt words: After analyzing the data, the large language model outputs an analysis report containing the following content. (a) Low grayscale area analysis: Based on the tool output indicators, identify the brightness jump of the R channel in the grayscale range of 0-16, and it is recommended to increase the binding density; (b) Analysis of medium and high grayscale areas: Using the tool to fit Gamma=2.18, it was determined that the brightness conforms to the power function law; (c) Channel consistency analysis: Based on the calculation of 15% channel difference using the tool, a low grayscale compensation strategy for channel B is proposed; (d) Abnormal range location: Based on the fluctuation coordinates output by the tool, mark the gray level fluctuation range of 128-136; (e) Model recommendation: Based on the fitting error of each group, recommend a segmentation strategy of low gray level cubic spline and medium and high gray level corrected power function.

[0025] Step S4: Algorithm Generation Based on the analysis report, the large language model automatically generates Gamma correction algorithm code in the following form (using Python as an example): Python # Automatically generated Gamma correction algorithm - Panel model XX import numpy as np from scipy.interpolate import CubicSpline def gamma_correction(gray_level, channel='R'): """ Gamma correction algorithm - Automatically generated panel model: XX Target Gamma: 2.2""" # Low grayscale range: Cubic spline interpolation if gray_level <= 32: binding_points = [0, 4, 8, 12, 16, 20, 24, 28, 32] target_luminance =calculate_target(gray_level, gamma=2.2) cs = CubicSpline(binding_points, measured_luminance[channel]) correction_value = cs(gray_level) # Medium to high grayscale range: Power function with correction term else: gamma = 2.18 a = 1.0 correction = a * (gray_level / 255.0) ** gamma # B channel compensation if channel == 'B' and gray_level < 64: correction *= 1.15 # Low grayscale B channel compensation correction_value = correction return correction_value

[0026] The generated algorithm code was deployed to the test environment, and Gamma correction was performed on the same panel. The corrected grayscale-brightness data was then remeasured. The corrected dataset was fed into a numerical analysis tool to uniformly calculate the deviation of each grayscale level: Deviation calculation formula: Δ_i = |L_corrected (i) - L_target (i)| / L_target (i) × 100%. The numerical analysis tool iterated through all grayscale levels and output the global maximum deviation Δ_max as the evaluation index. If Δ_max ≤ 3% (preset threshold), the algorithm is considered qualified; otherwise, iterative optimization steps are initiated.

[0027] Step S6: Iterative Optimization When Δ_max > 3%, the system will feed the following information back to the large language model: (a) Algorithm code generated in the previous round; (b) Corrected measured data, point-by-point deviation distribution table, and maximum deviation point output by the numerical analysis tool; (c) Deviation distribution of each gray level point; (d) Analysis report of the previous round.

[0028] Example prompt: "The Gamma correction algorithm generated in the previous round failed the verification, with a maximum deviation of 5.2%, occurring at grayscale level 24 (R channel). Below are the measured data and deviation distribution after correction. Please analyze the causes of the deviation and adjust and optimize the algorithm. Pay particular attention to the correction accuracy of the low grayscale R channel." Based on feedback, the large language model may make the following adjustments: Increase the number of binding points in the grayscale range of 20-28; Adjust the interpolation method for the low grayscale range of the R channel; Modify the compensation coefficient for channel B.

[0029] Repeat the verification and optimization until Δ_max ≤ 3% or the maximum number of iterations (e.g., 10 times) is reached.

[0030] Step S7: Knowledge Base Construction Once the algorithm passes verification, the following information will be stored in the knowledge base: Panel model and hardware parameters (glass substrate model, driver IC model, pixel arrangement, etc.); Raw measurement data and preprocessed data; Analysis reports of the large language model (each round); The final version of the Gamma correction algorithm code; Iteration log (including adjustment strategies and effects for each round).

[0031] When processing new model panels, the system first retrieves historical model data with similar hardware parameters from the knowledge base. Through the RAG mechanism, the relevant historical information is provided as context to the large language model to help it understand the possible characteristics of the new model panels more quickly, thereby accelerating the algorithm generation process.

[0032] Example 2: System Structure Reference Figure 2 This embodiment provides an automatic generation system for display panel Gamma correction algorithm based on a large language model, including: Data acquisition module (100): includes control unit (101) and optical measurement unit (102). Control unit (101) sends grayscale test signals to the panel under test through a communication interface (such as MIPI, SPI, etc.); optical measurement unit (102) includes luminance meter or color analyzer, used to acquire luminance data at each grayscale level.

[0033] The data preprocessing module (200) includes an outlier detection unit (201), a normalization processing unit (202), and a formatting unit (203), which process the raw data in sequence.

[0034] Large Language Model Analysis Module (300): LLM Service Unit (301), Prompt Word Management Unit (302), Numerical Tool Call Interface (303);

[0035] Algorithm verification module (400): includes algorithm deployment unit (401) and deviation calculation unit (402). Algorithm deployment unit (401) compiles the generated algorithm code and writes it into panel driver chip or test equipment; deviation calculation unit (402) calculates deviation index based on the corrected measurement data.

[0036] Iteration control module (500): includes a judgment unit (501) and a feedback assembly unit (502). The judgment unit (501) compares the deviation value with a preset threshold; the feedback assembly unit (502) packages relevant information into structured feedback text when iteration is required.

[0037] Knowledge base module (600): includes data storage unit (601) and retrieval unit (602). Data storage unit (601) stores complete calibration data of historical models; retrieval unit (602) supports similarity retrieval based on panel parameters and returns the most relevant historical cases.

[0038] Example 3: Optimization of Prompt Word Engineering In this embodiment, a dedicated prompt word template system was designed for different stages of the Gamma correction scenario: Data pattern discovery stage prompt word template: focuses on guiding large language models to conduct data statistical analysis, including trend identification, anomaly detection, channel comparison, etc.

[0039] Algorithm generation stage prompt word templates: Focus on guiding large language models to generate executable correction algorithm code, including mathematical model selection, parameter calculation, boundary condition handling, etc.

[0040] Deviation analysis stage prompt word template: focuses on guiding the large language model to analyze and correct the causes of deviations, and proposes targeted optimization suggestions.

[0041] The prompt word templates for each stage support parameterized configuration, and the prompt word content can be dynamically adjusted according to parameters such as panel type (OLED / LCD / MiniLED, etc.), driver IC model, and target Gamma value.

Claims

1. A method for automatically generating display panel Gamma correction algorithm based on a large language model, characterized in that, Includes the following steps: S1. Data acquisition steps: Control the display panel under test to display multiple preset grayscale test screens in sequence, and collect the brightness data of each grayscale through optical measurement equipment to obtain the original grayscale-brightness dataset. S2. Data Preprocessing Step: The original grayscale-brightness dataset is subjected to outlier removal, data normalization, and format standardization to generate a structured dataset to be analyzed. S3. Data Analysis and Pattern Discovery Step: The dataset to be analyzed and panel model information are input into a pre-configured large language model. The large language model calls a preset numerical analysis tool to perform quantization operations on the dataset to be analyzed and receives structured operation results such as curve fitting parameters, brightness statistics, and outlier region coordinates returned by the numerical analysis tool. The large language model outputs a data analysis report based on the structured quantization results from the tool. This report includes at least the brightness variation trend of each grayscale interval, nonlinear feature recognition results, RGB sub-pixel channel differences, and correction point placement strategy suggestions. S4, Algorithm Generation Step: The large language model, combining the data analysis report, the fitting parameters output by the numerical analysis tool, and the preset target Gamma curve parameters, automatically generates Gamma correction algorithm code containing point selection strategies, interpolation calculation methods, and correction parameters. S5, Automatic Verification Step: The Gamma correction algorithm code is deployed to perform calibration on the display panel under test. Corrected grayscale-brightness data is re-acquired, and the deviation between the calibration result and the target Gamma curve is calculated using the numerical analysis tool. S6, Iterative Optimization Step: When the deviation value exceeds a preset threshold, the calibrated measurement data, deviation index, and previous analysis results are fed back to the large language model. The large language model optimizes the algorithm parameters based on the quantized deviation data from the numerical analysis tool and re-executes step S5 until the deviation value meets the preset requirements, outputting the final Gamma correction algorithm.

2. The automatic generation method for display panel Gamma correction algorithm based on large language model according to claim 1, characterized in that: In step S3, the large language model loads data using structured tables and generates numerical analysis tool call commands by displaying dedicated prompts for Gamma correction on the display panel. Based on the quantitative statistical indicators output by the numerical analysis tool, multi-dimensional data analysis is completed from three dimensions: low grayscale brightness jump detection, mid-to-high grayscale brightness smoothness evaluation, and RGB sub-pixel channel consistency comparison.

3. The automatic generation method for display panel Gamma correction algorithm based on large language model according to claim 1, characterized in that: In step S4, the large language model automatically matches the optimal curve fitting mathematical model based on the fitting error index of each set of mathematical models output by the numerical analysis tool, adaptively determines the number and distribution of correction binding points, and outputs Python, C or C++ format correction algorithm code that can be run directly.

4. The automatic generation method for display panel Gamma correction algorithm based on large language model according to claim 1, characterized in that: In step S6, an upper limit is set for the number of algorithm iterations. When the number of iterations reaches the upper limit and the deviation requirement is still not met, the numerical analysis tool is automatically switched to call the prompt word template or change the fitted mathematical model to avoid the algorithm getting stuck in a local optimum.

5. The automatic generation method for display panel Gamma correction algorithm based on large language model according to claim 1, characterized in that, It also includes step S7, knowledge base construction and reuse steps: storing panel hardware parameters, raw measurement data, quantitative indicators output by numerical analysis tools, data analysis reports, final calibration algorithms and iteration logs into the calibration knowledge base; for new model panel calibration scenarios, retrieving similar historical calibration cases through the RAG retrieval mechanism to assist in the large language model reasoning optimization.

6. The automatic generation method for display panel Gamma correction algorithm based on large language model according to claim 1, characterized in that: In step S1, the grayscale test range covers the entire grayscale range from 0 to 255, and the grayscale sampling interval is no more than 4 grayscale levels; the preset threshold is the percentage of brightness deviation, and the value range is 1% to 5%.

7. An automatic generation system for display panel Gamma correction algorithm based on a large language model, characterized in that, The method for performing any one of claims 1-6 comprises: a data acquisition module for controlling the output of grayscale test images from the display panel under test and driving an optical measurement device to acquire raw grayscale-brightness data; a data preprocessing module for performing outlier removal, normalization, and format standardization on the raw dataset to generate structured data to be analyzed; a large language model analysis module with a built-in pre-trained large language model and a numerical analysis tool call interface for sending datasets to the numerical analysis tool, receiving fitting parameters and statistical indicators returned by the tool, and performing data pattern analysis, correction algorithm code generation, and iterative optimization based on the quantization results; a numerical analysis tool unit that communicates with the large language model analysis module for performing curve fitting, statistical operations, deviation calculation, and outputting structured quantized data to the large language model; an algorithm verification module for deploying the generated Gamma correction algorithm and performing panel correction; an iteration control module for receiving the deviation value output by the numerical analysis tool and determining whether the deviation meets the standard, and assembling feedback data and returning it to the large language model analysis module when the deviation exceeds the standard; and a knowledge base module for storing all historical correction data and supporting RAG. Similar case retrieval and reuse. Panel calibration and calculation of the calibrated brightness deviation value; Iterative control module, used to determine whether the deviation meets the standard, and when the deviation exceeds the standard, assemble feedback data and send it back to the large language model analysis module; Knowledge base module, used to store the full historical calibration data and support RAG similar case retrieval and reuse.

8. The automatic generation system for display panel Gamma correction algorithm based on large language model according to claim 7, characterized in that: The large language model analysis module supports cloud API calls or local open-source model deployment. It has a built-in prompt word management unit that stores dedicated prompt word templates for multiple scenarios, including data analysis, code generation, and deviation iteration optimization.

9. The automatic generation system for display panel Gamma correction algorithm based on large language model according to claim 7, characterized in that: The knowledge base module stores data including panel model parameters, raw optical measurement data, quantitative indicators output by numerical analysis tools, LLM analysis reports, calibration binding point parameters, executable algorithm code, and multi-round iteration logs.

10. The automatic generation system for display panel Gamma correction algorithm based on a large language model according to claim 7, characterized in that: The iterative control module configures an upper limit threshold for the number of iterations and a brightness deviation threshold. Combined with the deviation index output by the numerical analysis tool, it enables automatic iterative optimization, model switching, and iteration termination judgment.

Images