Liquid crystal screen defect detection method, system and storage medium based on machine vision

Abstract

The present invention relates to the field of liquid crystal screen detection technology, and in particular to a method, system and storage medium for detecting defects in liquid crystal screens based on machine vision. The method comprises the following steps: obtaining panel structure data of the liquid crystal screen, including panel size data, pixel arrangement data, backlight module data and surface treatment information data; performing optical property analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping normal display states under different display states, display contents and viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data; and collecting multi-region image data of the liquid crystal panel. The present invention dynamically tracks defects of the liquid crystal screen at different refresh frequencies, especially captures defect manifestations under different display states, and provides more comprehensive defect analysis.

Description

Technical Field

[0001] The present invention relates to the field of liquid crystal screen detection technology, and in particular to a liquid crystal screen defect detection method, system and storage medium based on machine vision. Background Art

[0002] With the rapid development of liquid crystal display technology, liquid crystal screens are widely used in various display devices. However, the manufacturing process of liquid crystal screens is complex and precise, and it is easy to produce various types of defects, such as bright spots, dark spots, color difference, light leakage, etc. These defects may have a serious impact on the final display quality of the product, especially under the display requirements of high resolution, large size and high refresh rate. Traditional detection methods are difficult to quickly and accurately identify these defects.

[0003] Traditional LCD screen defect detection methods often have the following problems: The production line requires fast detection, but this may affect the accuracy of the detection. For example, large-size LCD panels (such as 65 inches or more) need to be fully inspected within a few seconds, and high refresh rate displays (such as 240Hz) need to be inspected in multiple refresh states. Handling of false positives and false negatives: It is extremely difficult to completely eliminate false detections (false positives) and missed detections (false negatives); dust or temporary stains may be misjudged as permanent defects; under certain specific display content, real defects may be ignored by the system. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a method, system and storage medium for detecting LCD defects based on machine vision to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a method for detecting LCD screen defects based on machine vision comprises the following steps:

[0006] Step S1: acquiring panel structure data of the liquid crystal screen, including panel size data, pixel arrangement data, backlight module data, and surface treatment information data; performing optical characteristic analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping different display states, display contents, and normal display states under viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data;

[0007] Step S2: collecting multi-region image data of the LCD panel; performing image preprocessing on the multi-region image data, and performing difference analysis according to the standard display characteristic diagram data to identify potential defect areas, thereby obtaining preliminary defect detection data;

[0008] Step S3: Tracking defect changes of the preliminary defect detection data at different refresh frequencies by using a multi-frame image overlay technique, thereby obtaining dynamic defect detection data; classifying defect points based on temporary and permanent defects according to the dynamic defect detection data, thereby obtaining defect point classification data;

[0009] Step S4: acquiring defect area data according to defect point classification data; acquiring spectral features of LCD screen images under different light sources for the defect area data, thereby obtaining spectral feature data of the defect area, wherein the different light sources include visible light, near infrared light and ultraviolet light; performing multi-spectral comparison on the spectral feature data of the defect area, and extracting differences between different spectra, thereby obtaining spectral difference data;

[0010] Step S5: performing false positive and false negative identification on the defect area data according to the spectral difference data, and optimizing the defect point accuracy, so as to obtain accurate defect point data.

[0011] By acquiring panel structural data such as panel size, pixel arrangement, backlight module and surface treatment information, the system of the present invention can fully understand the physical structure of the LCD screen, which provides accurate basic data for subsequent optical characteristic analysis and defect detection; optical characteristic analysis using light wave interference technology can accurately analyze the optical performance of the LCD screen under different viewing angles, display contents and display states. This analysis method can more sensitively capture the slight optical changes of the LCD screen, especially some minor defects that are difficult to detect with the naked eye. The normal display state of the LCD screen is mapped to standard display characteristic diagram data, providing an accurate reference benchmark for subsequent defect detection, which is helpful for difference analysis in subsequent steps. Multi-region image data acquisition ensures complete coverage of each area of ​​the LCD screen, and can capture each display unit of the screen in detail, thereby avoiding missing defects; through image preprocessing (such as denoising, contrast enhancement, etc.), the availability and clarity of the data can be improved, providing a more reliable data basis for subsequent defect identification and analysis. By performing a difference analysis with the standard display characteristic diagram obtained in step S1, potential defect areas can be effectively identified. This difference analysis method greatly improves the accuracy of defect detection and can detect small or hidden defects that are difficult for the human eye to detect. The multi-frame image overlay technology can capture the defects of the LCD screen in dynamic display (such as fast scene switching or brightness changes) by analyzing the defect changes at different refresh frequencies, solving the problem that traditional static detection methods are difficult to detect dynamic defects. Dynamic defect detection data can distinguish between temporary and permanent defects, avoiding misjudgment based on data at a certain moment, which helps to accurately judge the actual defect situation of the LCD screen and improve the reliability of the detection results. By imaging and analyzing the defect area under different light sources such as visible light, near infrared, and ultraviolet, the performance of the defect area can be captured from multiple spectral dimensions. This multi-spectral analysis method is more comprehensive and can identify hidden defects that are difficult to detect under a single spectrum. Extracting the spectral differences of the defect area can identify the performance differences of the defects under different spectra and help analyze the nature of the defects. This provides a more accurate basis for false positive and false negative identification in subsequent steps. By identifying false positives and false negatives in defective areas based on spectral difference data, misjudgments caused by optical errors or ambient light are effectively reduced. This method can eliminate false defect points in the detection results and improve the accuracy of defect detection. By optimizing the accuracy of defect points, the specific location and nature of the defects can be further accurately located and described, which is of great help for defect repair and quality control in the production process and helps to improve the quality of the final product. The implementation of these steps not only improves the automation level of LCD defect detection, but also significantly improves the detection accuracy and efficiency. Through technical means such as multi-spectral analysis and dynamic defect tracking, this method can capture defects that are difficult to detect with traditional methods, reduce the false alarm rate, and provide more accurate data support for quality control in production.

[0012] The present invention also provides a machine vision-based LCD defect detection system, which is used to execute the above-mentioned machine vision-based LCD defect detection method. The machine vision-based LCD defect detection system includes:

[0013] The panel characteristic modeling module is used to obtain the panel structure data of the LCD screen, including panel size data, pixel arrangement data, backlight module data and surface treatment information data; perform optical characteristic analysis of the LCD screen based on light wave interference according to the panel structure data, and map different display states, display contents and normal display states under different viewing angles according to the preset image data set, so as to obtain standard display characteristic diagram data;

[0014] The initial defect recognition module is used to collect multi-region image data of the LCD panel; perform image preprocessing on the multi-region image data, and perform difference analysis based on the standard display characteristic diagram data to identify potential defect areas, thereby obtaining preliminary defect detection data;

[0015] The dynamic defect classification module is used to track the defect changes of the preliminary defect detection data at different refresh frequencies through multi-frame image overlay technology, so as to obtain dynamic defect detection data; and classify the defect points based on temporary and permanent defects according to the dynamic defect detection data, so as to obtain defect point classification data;

[0016] The multi-spectral feature extraction module is used to obtain defect area data according to defect point classification data; obtain the spectral features of the LCD screen image under different light sources for the defect area data, so as to obtain the spectral feature data of the defect area, wherein the different light sources include visible light, near infrared and ultraviolet light; perform multi-spectral comparison on the spectral feature data of the defect area, and extract the differences between different spectra, so as to obtain spectral difference data;

[0017] The precise defect optimization module is used to identify false positives and false negatives of defect area data based on spectral difference data, and to optimize the precision of defect points, so as to obtain precise defect point data.

[0018] The present invention can fully understand the physical structure of the liquid crystal screen by acquiring data such as the size, pixel arrangement, backlight module and surface treatment information of the panel, and these data provide a reliable basis for subsequent optical analysis; optical characteristic analysis based on light wave interference can capture the optical response of liquid crystal molecule arrangement, polarizer, backlight and other aspects, and provide accurate optical characteristic data, so that the display performance of the liquid crystal screen can be carefully modeled and simulated under different states and viewing angles; by mapping the physical characteristics of the liquid crystal screen with the preset image data set, standard display characteristic diagram data under different states can be generated, providing a reference benchmark for subsequent defect detection, ensuring that the detection process has a clear comparison standard. By performing multi-region imaging on the liquid crystal panel, including the central area, edge and four corners, it is ensured that all areas of the liquid crystal screen are detected to avoid detection blind spots caused by neglect of local defects; the image preprocessing step includes geometric correction, brightness balance and noise suppression, which can effectively eliminate errors caused by equipment or environment during the imaging process, making subsequent defect identification more accurate; the preprocessed image is compared and analyzed with the standard display characteristic diagram data, which can quickly identify potential defect areas and generate preliminary defect detection data. This step lays a solid foundation for subsequent dynamic analysis. By integrating defect detection data at different refresh rates through multi-frame image overlay technology, the robustness of detection can be improved, and defects that may be missed or misjudged by a single frame of image can be avoided. Tracking the performance of defects in time and refresh rate can identify dynamic defects that change over time or behave differently at different refresh rates. This method is particularly suitable for detecting unstable defects on LCD screens; classifying defects according to their duration can distinguish temporary defects (such as temporary pixel flickering) from permanent defects (such as permanent bright spots or dark spots), providing valuable information for subsequent repair or judgment. By collecting images under multiple light sources such as visible light, near-infrared light, and ultraviolet light, the defective area of ​​the LCD screen can show different optical properties under different spectra. Some defects may not be obvious under visible light, but they will show obvious characteristics under infrared or ultraviolet light, increasing the comprehensiveness of detection. By spectrally decomposing images under different light sources and extracting spectral feature data (such as reflectivity and absorptivity) of each pixel, a unique spectral fingerprint can be generated for each defect, which provides a basis for refined defect description and classification. By analyzing the differences in different spectral bands, the special performance of certain defects in specific bands can be revealed, providing richer defect analysis data. Based on spectral difference data, the system can effectively eliminate false detection or missed detection, reduce false positive or false negative phenomena caused by equipment or environmental factors, and thus improve the reliability of defect detection. Through refined spectral analysis and difference extraction, the accuracy of the initially identified defects is further optimized to ensure that the final defect point data has higher accuracy and precision, reducing the possibility of misjudgment.By combining multi-spectral comparison, spectral difference extraction and other technologies, defect points are optimized to ensure that the final output defect detection results are more accurate and reduce the risk of defective products in production. This set of LCD screen defect detection methods based on optics, spectroscopy and dynamic analysis can comprehensively improve the accuracy and reliability of LCD screen defect detection. Through multi-step image processing, multi-spectral analysis and dynamic defect tracking, it can more accurately identify, classify and optimize defects, providing strong support for production quality control.

[0019] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program is executed to implement any one of the above-mentioned methods for detecting defects in a liquid crystal screen based on machine vision. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0021] Figure 1 The figure is a schematic diagram of the steps of the liquid crystal screen defect detection method based on machine vision of the present invention;

[0022] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0023] Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0024] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0025] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0026] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0027] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for detecting defects of a liquid crystal screen based on machine vision, the method comprising the following steps:

[0028] Step S1: acquiring panel structure data of the liquid crystal screen, including panel size data, pixel arrangement data, backlight module data, and surface treatment information data; performing optical characteristic analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping different display states, display contents, and normal display states under viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data;

[0029] Step S2: collecting multi-region image data of the LCD panel; performing image preprocessing on the multi-region image data, and performing difference analysis according to the standard display characteristic diagram data to identify potential defect areas, thereby obtaining preliminary defect detection data;

[0030] Step S3: Tracking defect changes of the preliminary defect detection data at different refresh frequencies by using a multi-frame image overlay technique, thereby obtaining dynamic defect detection data; classifying defect points based on temporary and permanent defects according to the dynamic defect detection data, thereby obtaining defect point classification data;

[0031] Step S4: acquiring defect area data according to defect point classification data; acquiring spectral features of LCD screen images under different light sources for the defect area data, thereby obtaining spectral feature data of the defect area, wherein the different light sources include visible light, near infrared light and ultraviolet light; performing multi-spectral comparison on the spectral feature data of the defect area, and extracting differences between different spectra, thereby obtaining spectral difference data;

[0032] Step S5: performing false positive and false negative identification on the defect area data according to the spectral difference data, and optimizing the defect point accuracy, so as to obtain accurate defect point data.

[0033] In the embodiment of the present invention, reference Figure 1The above is a schematic flow chart of the steps of a method for detecting defects of a liquid crystal screen based on machine vision of the present invention. In this example, the method for detecting defects of a liquid crystal screen based on machine vision includes the following steps:

[0034] Step S1: acquiring panel structure data of the liquid crystal screen, including panel size data, pixel arrangement data, backlight module data, and surface treatment information data; performing optical characteristic analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping different display states, display contents, and normal display states under viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data;

[0035] The embodiment of the present invention obtains the panel structure data of the liquid crystal screen. First, physical measurement and data collection are required. A high-precision measuring tool (such as a laser rangefinder or a digital caliper) is used to measure the panel size of the liquid crystal screen, including its width, height and thickness. Then, a high-resolution camera or microscope is used to collect pixel arrangement data, and the number of rows and columns of pixels and the arrangement of sub-pixels are analyzed. Next, the liquid crystal screen is disassembled to obtain the data of the backlight module, including the backlight type, arrangement and quantity. The brightness of the backlight module is measured to record its light intensity distribution. Finally, the surface roughness and surface treatment type (such as coating type or anti-glare treatment) of the liquid crystal screen are measured. All these data are integrated to create a panel structure data model. Based on these structural data, the optical properties of the liquid crystal screen are simulated using the principle of light wave interference, and a preset image data set (including pure color, gradient grayscale, text and image) is used for display state mapping to generate standard display characteristic diagram data.

[0036] Step S2: collecting multi-region image data of the LCD panel; performing image preprocessing on the multi-region image data, and performing difference analysis according to the standard display characteristic diagram data to identify potential defect areas, thereby obtaining preliminary defect detection data;

[0037] The embodiment of the present invention first uses a high-resolution camera array to image multiple areas (such as the center, edges, and four corners) of the LCD screen to collect detailed multi-area image data. These images are preprocessed, including geometric correction (distortion correction through a calibration plate), brightness equalization (correcting brightness differences using a histogram equalization algorithm), and noise suppression (applying methods such as Gaussian filtering). Then, these preprocessed image data are compared with the standard display characteristic diagram data, and pixel-level difference analysis (such as calculating the brightness difference of each pixel) is used to identify potential defective areas. These areas are marked as preliminary defect detection data for subsequent defect analysis.

[0038] Step S3: Tracking defect changes of the preliminary defect detection data at different refresh frequencies by using a multi-frame image overlay technique, thereby obtaining dynamic defect detection data; classifying defect points based on temporary and permanent defects according to the dynamic defect detection data, thereby obtaining defect point classification data;

[0039] In order to perform dynamic defect detection, the embodiment of the present invention first sets the LCD screen to collect multiple frames of images at different refresh frequencies (such as 30Hz, 60Hz, 120Hz and 240Hz). These image sequences are processed by multi-frame superposition to track the changes of defects at different refresh frequencies. The image registration technology is used to align the images at different frequencies to eliminate the slight displacement caused by the difference in refresh frequency. The median filtering method is used to fuse multi-frame information to enhance the visibility of defects. Dynamic defects are identified through time series analysis (such as calculating the changes of each defect point at different time points), and the defect points are classified according to their stability. Based on these data, the defect points are divided into temporary defects and permanent defects, thereby generating defect point classification data.

[0040] Step S4: acquiring defect area data according to defect point classification data; acquiring spectral features of LCD screen images under different light sources for the defect area data, thereby obtaining spectral feature data of the defect area, wherein the different light sources include visible light, near infrared light and ultraviolet light; performing multi-spectral comparison on the spectral feature data of the defect area, and extracting differences between different spectra, thereby obtaining spectral difference data;

[0041] The embodiment of the present invention first configures a multi-light source imaging system for the identified defect area, including a visible light source, a near-infrared light source, and an ultraviolet light source. These light sources are used to capture images of the defect area, and image alignment is performed to eliminate the differences between the light sources. The collected multi-spectral image data is spectrally decomposed to extract the reflectivity or emissivity information of each pixel in different bands (visible light, near-infrared, ultraviolet light). Spectral feature data of the defect area are constructed based on these spectral data, including calculating the average spectral response, standard deviation, and shape characteristics of the spectral curve of each spectral band. Multi-spectral comparative analysis is used to calculate the correlation coefficients of each defect area in different bands, and the main variation components of the spectral fingerprint are extracted through principal component analysis to obtain spectral difference data.

[0042] Step S5: performing false positive and false negative identification on the defect area data according to the spectral difference data, and optimizing the defect point accuracy, so as to obtain accurate defect point data.

[0043] The embodiment of the present invention uses spectral difference data to identify false positives and false negatives in defective areas. First, the spectral difference data is combined with the preliminary defect detection data to construct a multidimensional feature vector, and the feature vectors are classified using a support vector machine (SVM). According to the classification results, possible false positive areas and potential false negative areas are marked. The marked areas are resampled at high resolution and local contrast enhancement is performed to improve the visibility of image details. The confidence of the misjudged area is calculated by finely analyzing the enhanced image, and the false positive area is screened according to the preset confidence threshold to remove defect points with low confidence. Morphological processing and connected domain analysis are performed on potential false negative areas to identify small defects that are ignored. Finally, the optimized defect point data and the supplementary defect point data are merged, and spatial clustering analysis is performed to sort out adjacent defect points to obtain the final accurate defect point data.

[0044] By acquiring panel structural data such as panel size, pixel arrangement, backlight module and surface treatment information, the system of the present invention can fully understand the physical structure of the LCD screen, which provides accurate basic data for subsequent optical characteristic analysis and defect detection; optical characteristic analysis using light wave interference technology can accurately analyze the optical performance of the LCD screen under different viewing angles, display contents and display states. This analysis method can more sensitively capture the slight optical changes of the LCD screen, especially some minor defects that are difficult to detect with the naked eye. The normal display state of the LCD screen is mapped to standard display characteristic diagram data, providing an accurate reference benchmark for subsequent defect detection, which is helpful for difference analysis in subsequent steps. Multi-region image data acquisition ensures complete coverage of each area of ​​the LCD screen, and can capture each display unit of the screen in detail, thereby avoiding missing defects; through image preprocessing (such as denoising, contrast enhancement, etc.), the availability and clarity of the data can be improved, providing a more reliable data basis for subsequent defect identification and analysis. By performing a difference analysis with the standard display characteristic diagram obtained in step S1, potential defect areas can be effectively identified. This difference analysis method greatly improves the accuracy of defect detection and can detect small or hidden defects that are difficult for the human eye to detect. The multi-frame image overlay technology can capture the defects of the LCD screen in dynamic display (such as fast scene switching or brightness changes) by analyzing the defect changes at different refresh frequencies, solving the problem that traditional static detection methods are difficult to detect dynamic defects. Dynamic defect detection data can distinguish between temporary and permanent defects, avoiding misjudgment based on data at a certain moment, which helps to accurately judge the actual defect situation of the LCD screen and improve the reliability of the detection results. By imaging and analyzing the defect area under different light sources such as visible light, near infrared, and ultraviolet, the performance of the defect area can be captured from multiple spectral dimensions. This multi-spectral analysis method is more comprehensive and can identify hidden defects that are difficult to detect under a single spectrum. Extracting the spectral differences of the defect area can identify the performance differences of the defects under different spectra and help analyze the nature of the defects. This provides a more accurate basis for false positive and false negative identification in subsequent steps. By identifying false positives and false negatives in defective areas based on spectral difference data, misjudgments caused by optical errors or ambient light are effectively reduced. This method can eliminate false defect points in the detection results and improve the accuracy of defect detection. By optimizing the accuracy of defect points, the specific location and nature of the defects can be further accurately located and described, which is of great help for defect repair and quality control in the production process and helps to improve the quality of the final product. The implementation of these steps not only improves the automation level of LCD defect detection, but also significantly improves the detection accuracy and efficiency. Through technical means such as multi-spectral analysis and dynamic defect tracking, this method can capture defects that are difficult to detect with traditional methods, reduce the false alarm rate, and provide more accurate data support for quality control in production.

[0045] Preferably, step S1 comprises the following steps:

[0046] Step S11: measuring the physical dimensions of the liquid crystal screen and recording the dimensions of the effective display area, thereby obtaining panel dimension data;

[0047] Step S12: collecting the number of rows and columns of the pixel matrix of the liquid crystal screen and the sub-pixel arrangement, and using high-resolution imaging analysis to measure the pixel pitch and sub-pixel size, thereby obtaining pixel arrangement data;

[0048] Step S13: collecting the backlight type and the arrangement and quantity of the backlight units, and performing disassembly analysis and brightness measurement to obtain backlight module data;

[0049] Step S14: collecting the surface treatment type and measuring the surface roughness of the liquid crystal screen to obtain surface treatment information data;

[0050] Step S15: merging the panel size data, pixel arrangement data, backlight module data and surface treatment information data into panel structure data;

[0051] Step S16: performing optical characteristic analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping different display states, display contents and normal display states under different viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data.

[0052] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in the embodiment of the present invention, step S1 includes the following steps:

[0053] Step S11: measuring the physical dimensions of the liquid crystal screen and recording the dimensions of the effective display area, thereby obtaining panel dimension data;

[0054] When measuring the physical dimensions of the LCD screen in the embodiment of the present invention, a high-precision laser rangefinder or digital caliper is first used to measure the entire panel of the LCD screen, including the width, height and thickness. The measurement data is recorded, especially the size of the effective display area. The effective display area generally refers to the actual display portion after removing the border and the invalid area. Through these measurement values, the panel size data can be accurately obtained, providing a basis for subsequent optical property analysis.

[0055] Step S12: collecting the number of rows and columns of the pixel matrix of the liquid crystal screen and the sub-pixel arrangement, and using high-resolution imaging analysis to measure the pixel pitch and sub-pixel size, thereby obtaining pixel arrangement data;

[0056] When collecting pixel matrix data of a liquid crystal screen, the embodiment of the present invention uses a high-resolution microscope or a scanning microscope to photograph the pixel array of the liquid crystal screen. The number of rows and columns of the pixel matrix is ​​measured by the image analysis function of the microscope, and the arrangement of each pixel is recorded. The spacing between pixels and the size of sub-pixels are calculated using a high-resolution imaging system. These data are analyzed by image processing software to extract and record pixel arrangement data, including pixel spacing, sub-pixel size and its arrangement.

[0057] Step S13: collecting the backlight type and the arrangement and quantity of the backlight units, and performing disassembly analysis and brightness measurement to obtain backlight module data;

[0058] When obtaining the backlight module data, the embodiment of the present invention first disassembles the LCD screen to expose the backlight unit. Use a brightness meter or a photometer to measure the brightness distribution of the backlight unit, and record its type (such as LED, CCFL, etc.) and its arrangement and quantity. During the disassembly and analysis process, ensure that the structure and configuration of the backlight unit are recorded in detail to accurately obtain the backlight module data, including its arrangement and brightness characteristics.

[0059] Step S14: collecting the surface treatment type and measuring the surface roughness of the liquid crystal screen to obtain surface treatment information data;

[0060] When collecting surface treatment information in the embodiment of the present invention, the surface roughness of the LCD screen is first measured. A surface roughness meter (such as a profilometer) is used to scan the display surface of the LCD screen and record its surface roughness parameters (such as Ra, Rz, etc.). At the same time, the surface treatment type (such as anti-glare coating, anti-reflective coating, etc.) is recorded. These data can provide detailed information on the surface treatment of the LCD screen for subsequent optical property analysis.

[0061] Step S15: merging the panel size data, pixel arrangement data, backlight module data and surface treatment information data into panel structure data;

[0062] When merging the panel size data, pixel arrangement data, backlight module data and surface treatment information data, the embodiment of the present invention first organizes all independently acquired data into a unified format. The panel size data, pixel arrangement data, backlight module data and surface treatment information data are aggregated into a comprehensive panel structure data model using a data integration tool or database. The association and consistency between the data are ensured so as to facilitate subsequent optical property analysis.

[0063] Step S16: performing optical characteristic analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping different display states, display contents and normal display states under different viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data.

[0064] When the embodiment of the present invention performs optical property analysis based on the panel structure data, an optical model of the liquid crystal screen is first constructed, which includes the arrangement of liquid crystal molecules, polarizer characteristics, and backlight propagation path. The propagation of light of different wavelengths in the liquid crystal layer is simulated using the principle of light wave interference, and the optical response of the liquid crystal screen under different driving voltages is calculated. Then, a preset image data set (such as pure color, grayscale gradient, text, and image) is used to generate simulated display effect data. These simulated data are compared with standard display characteristic diagram data to generate standard display characteristic diagram data under different display states, display contents, and viewing angles, which are used to evaluate and calibrate the normal display state of the liquid crystal screen.

[0065] The present invention can provide an accurate geometric reference for subsequent defect detection by measuring the physical size of the liquid crystal screen, especially the size of the effective display area, to ensure that the detection of the entire display area does not miss any area. The effective display area is accurately measured to ensure that the detection system focuses on the actual display area of ​​the screen, avoids interference of non-display area data with the detection results, and improves detection efficiency and accuracy. By collecting the number of rows and columns of the pixel matrix and the sub-pixel arrangement of the liquid crystal screen, combined with high-resolution imaging technology, the arrangement and size of each pixel and its sub-pixel can be accurately measured. This provides high-precision data support for analyzing pixel-level defects (such as bright spots, dark spots, and uneven colors). This detailed pixel arrangement and size information provides an accurate reference for subsequent defect detection, which can ensure consistency detection in various parts of the screen, which is particularly important on high-resolution liquid crystal screens. By collecting the backlight type, arrangement and quantity, combined with brightness measurement and disassembly analysis, the structure and optical properties of the backlight module can be accurately understood. This helps detect defects caused by backlight problems (such as unevenness and brightness reduction) and improves the detection capability of backlight-related problems; brightness measurement can provide effective data support for the brightness consistency of LCD screens, help detect the brightness changes of backlight modules under different display states, and identify possible brightness unevenness defects. By collecting the surface treatment type of LCD screens (such as anti-glare, mirror treatment, etc.) and measuring its surface roughness, optical defects caused by surface treatment problems can be identified, helping to improve the detection accuracy of surface irregularities or texture problems; measuring surface roughness provides more detailed optical analysis data, which can better evaluate the optical performance of LCD screens under different lighting conditions and avoid misjudgments caused by surface treatment defects. The panel size, pixel arrangement, backlight module and surface treatment information are combined into panel structure data to form a comprehensive LCD screen structure data set, which provides a complete physical and structural benchmark for subsequent optical property analysis. The integration of these data can ensure a comprehensive analysis of every detail of the LCD screen and avoid misjudgments caused by the lack of a certain local data during defect detection. Analysis based on light wave interference can accurately evaluate the optical performance of LCD screens, especially the fine capture of complex optical phenomena (such as interference, reflection, and refraction). This high-precision analysis provides a basis for identifying tiny defects (such as uneven color, bright spots and dark spots). Mapping the standard display characteristic diagram according to different display states, display content, and viewing angle conditions provides a reliable reference data set. With this data set, the inspection system can more effectively identify differences from normal display states in subsequent steps, improving the accuracy and reliability of defect detection. These steps gradually build a complete inspection framework from physical size, pixel arrangement to optical characteristic analysis. Through precise data acquisition, physical and optical characteristic analysis, this method can effectively improve the comprehensiveness and accuracy of LCD defect detection, especially in high resolution and multi-display states, ensuring the reliability and consistency of inspection results.

[0066] Preferably, step S16 comprises the following steps:

[0067] Step S161: constructing a liquid crystal screen optical model according to the panel structure data, wherein the liquid crystal screen optical model includes liquid crystal molecule arrangement, polarizer characteristics, and backlight propagation path;

[0068] When constructing the optical model of the liquid crystal screen in the embodiment of the present invention, the panel structure data, including the arrangement of liquid crystal molecules, the characteristics of the polarizer, and the backlight propagation path, is first obtained, and a three-dimensional optical model of the liquid crystal screen is established using computer-aided design (CAD) software or optical simulation software (such as OptiSystem, LightTools, etc.). The molecular arrangement characteristics of the liquid crystal layer (such as vertical arrangement or tilted arrangement), the optical characteristics of the polarizer (such as polarization direction and transmittance), and the propagation path of the backlight module are defined in the model. By setting the optical characteristics of each layer in the model in detail, it is ensured that the model accurately reflects the actual structure of the liquid crystal screen, providing basic data for subsequent optical response simulations.

[0069] Step S162: simulating the propagation characteristics of light of different wavelengths in the liquid crystal layer according to the optical model of the liquid crystal screen by using the principle of light wave interference, and calculating the optical response of the liquid crystal screen under different driving voltages, thereby obtaining optical response simulation data;

[0070] When the embodiment of the present invention uses the principle of light wave interference to simulate the propagation characteristics of light of different wavelengths in the liquid crystal layer, the input parameters of the light wave interference simulation are first set, including the wavelength of light, the incident angle of the light source, and the thickness of the liquid crystal layer, and the optical simulation software is used to perform interference calculation to simulate the propagation and interference effect of light in the liquid crystal layer. The optical response of the liquid crystal screen under different driving voltages is calculated, the intensity distribution and phase change of the light are recorded, and the optical response simulation data is generated. These data reflect the optical behavior of the liquid crystal screen under actual working conditions and provide a basis for the generation of theoretical display characteristic data.

[0071] Step S163: generating theoretical display characteristic data at different grayscales, colors and viewing angles according to the optical response simulation data;

[0072] When the embodiment of the present invention generates theoretical display characteristic data at different grayscales, colors, and viewing angles based on the optical response simulation data, the simulation data is first matched with a standard color space (such as sRGB or Adobe RGB). Image processing software is used to generate display characteristic data at different grayscales and colors, which represent the theoretical performance of the LCD screen under different display states. In addition, the optical characteristics at different viewing angles are also considered, and the theoretical display characteristic data is created through the optical response data generated by multi-angle simulation. These data reflect the expected performance of the LCD screen under various display conditions and provide a reference for the subsequent generation of standard display characteristic diagrams.

[0073] Step S164: mapping and converting the preset image data set according to the theoretical display characteristic data, thereby generating simulated display effect data under different display states, wherein the preset image data set includes display contents of pure color, grayscale gradient, text, and image;

[0074] When the embodiment of the present invention maps and converts the preset image data set according to the theoretical display characteristic data, the preset image data set is first defined, including pure color, grayscale gradient, text and image content. The theoretical display characteristic data is mapped using image processing software, and the characteristics under different display states are applied to the preset image. The image effects under different display states are generated by simulation software, including color accuracy, grayscale transition smoothness, and text clarity. The generated simulated display effect data can reflect the performance of the LCD screen in actual use, ensuring the consistency between the theoretical data and the actual display effect.

[0075] Step S165: Generate standard display characteristic diagram data including different display states, display contents and viewing angle information according to the simulated display effect data and the theoretical display characteristic data.

[0076] When the embodiment of the present invention generates standard display characteristic diagram data containing different display states, display contents, and viewing angle information, the theoretical display characteristic data and the simulated display effect data obtained in steps S163 and S164 are first integrated. The effect diagrams under different display states are synthesized using image synthesis software to generate a standard display characteristic diagram. These image data should cover a variety of display contents (such as text, images, and gradients) and information at different viewing angles (such as front viewing angle and side viewing angle). By summarizing and organizing these data, a comprehensive standard display characteristic diagram data set is created to evaluate the display performance of the LCD screen under various conditions.

[0077] The present invention constructs an optical model of a liquid crystal screen according to data such as the arrangement of liquid crystal molecules, polarizer characteristics, and backlight propagation paths, and can comprehensively capture the optical characteristics of each link of the liquid crystal screen from backlight to display. This model provides an accurate simulation of the optical behavior of the liquid crystal screen, which helps to accurately analyze each influencing factor of the optical response. The optical model takes into account multiple physical and optical parameters of the liquid crystal screen, and can better evaluate the root causes of display anomalies (such as color distortion and uneven brightness), providing an accurate analysis basis for subsequent defect detection. The propagation characteristics of light of different wavelengths in the liquid crystal layer are simulated by the principle of light wave interference, and the performance of the liquid crystal screen under different optical conditions can be analyzed in detail. For tiny optical defects (such as color shift or uneven brightness caused by reflection, refraction, and interference), this simulation provides accurate data support. By calculating the optical response of the liquid crystal screen under different driving voltages, the performance of the screen under different operating conditions can be comprehensively evaluated. In particular, when the driving voltage is unstable, the screen may have dynamic defects or flickering phenomena, and the simulated optical response can predict these potential problems in advance. By generating theoretical display characteristic data under different grayscales, colors, and viewing angles, the optical characteristics of the liquid crystal screen under various display conditions can be comprehensively modeled. This extensive data support ensures that the display effect of the LCD screen can be predicted and evaluated in different usage scenarios. Grayscale and color data are directly related to the color reproduction and brightness performance of the LCD screen, while the viewing angle data evaluates the display effect of the screen at different viewing angles, which is particularly suitable for detecting viewing angle-related defects such as color cast and brightness reduction. By mapping and converting the preset image data set according to the theoretical display characteristic data, simulated display effect data under different display states can be generated. This allows the display performance of the LCD screen to be pre-tested and evaluated before the actual production of the LCD screen, avoiding problems after production, saving costs and time. The preset image data set covers pure colors, grayscale gradients, text, and complex images. These rich test scenarios can detect the performance of the screen in a variety of actual usage situations, which helps to comprehensively evaluate the display effect and defect risks of the LCD screen. By generating standard display characteristic diagram data containing different display states, display content, and viewing angle information, an accurate benchmark is provided for subsequent defect detection. These standard characteristic diagrams not only cover the ideal display state of the LCD screen, but also take into account the performance under various external conditions to ensure the accuracy of defect detection. The multi-dimensional data of the standard display characteristic diagram (including grayscale, color, and viewing angle) enables the defect detection system to compare and analyze the actual display effect, quickly identify abnormal areas, reduce false alarms and missed alarms, and greatly improve the reliability of defect detection. These steps gradually provide comprehensive and accurate data support for LCD screen defect detection through precise optical modeling, simulation, and display effect generation. From the physical characteristics of the LCD screen to the optical response to the generation of the standard display characteristic diagram, each step is designed to provide a reliable reference and benchmark for subsequent detection.Through these steps, the quality control capability of LCD screens can be significantly improved, potential defects can be predicted and discovered in advance, and an efficient and accurate solution can be provided for defect detection in mass production.

[0078] Preferably, step S2 comprises the following steps:

[0079] Step S21: using a high-resolution camera array to perform multi-region imaging on the LCD panel, and collecting multi-region image data of the center region, the edge region, and the four corner regions;

[0080] Step S22: performing geometric correction on the multi-region image data to eliminate image distortion caused by camera position and angle, thereby obtaining multi-region image correction data;

[0081] Step S23: performing brightness equalization processing on the multi-region image correction data to obtain multi-region image equalization data, wherein the brightness equalization processing includes compensating for brightness differences between different regions;

[0082] Step S24: performing noise suppression on the multi-region image equalization data and performing contrast enhancement to improve the visibility of image details, thereby obtaining multi-region image comparison data;

[0083] Step S25: performing pixel-level comparison on the multi-region image comparison data according to the standard display characteristic map data, and calculating the deviation value of each pixel point, thereby obtaining a pixel deviation map;

[0084] Step S26: Identify potential defect areas based on the pixel deviation map to obtain preliminary defect detection data.

[0085] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed step flow diagram of step S2 in the embodiment of the present invention, step S2 includes the following steps:

[0086] Step S21: using a high-resolution camera array to perform multi-region imaging on the LCD panel, and collecting multi-region image data of the center region, the edge region, and the four corner regions;

[0087] When the embodiment of the present invention utilizes a high-resolution camera array to perform multi-region imaging on a liquid crystal panel, a plurality of high-resolution cameras are first arranged to ensure that the camera array covers the center area, edge area, and four corner areas of the liquid crystal screen. Each camera should shoot with the same settings (such as focal length, aperture, and shutter speed) to ensure image consistency. When shooting the LCD screen, the camera should be aimed at different areas of the LCD screen to collect image data for each area. Ensure that the relative position of the camera and the LCD screen is stable to reduce errors caused by differences in shooting angles and positions. The acquired image data includes images of the center area, edge area, and four corner areas to cover all key areas of the LCD screen.

[0088] Step S22: performing geometric correction on the multi-region image data to eliminate image distortion caused by camera position and angle, thereby obtaining multi-region image correction data;

[0089] When the embodiment of the present invention performs geometric correction on multi-region image data, a geometric correction algorithm is used to eliminate image distortion caused by differences in camera position and angle. First, the camera is calibrated using a calibration plate (such as a chessboard) to obtain the camera's internal parameters and distortion coefficients. These parameters are then used to correct the multi-region image data using a geometric correction algorithm (such as OpenCV's cv2.undistort function), repair the distortion in the image, and obtain geometric correction data. Ensure that the perspective effect of the corrected image is consistent in all regions, providing an accurate basis for subsequent image processing.

[0090] Step S23: performing brightness equalization processing on the multi-region image correction data to obtain multi-region image equalization data, wherein the brightness equalization processing includes compensating for brightness differences between different regions;

[0091] When the embodiment of the present invention performs brightness equalization processing on multi-region image data, the image data is first processed using a histogram equalization technique. The brightness histogram of each region is calculated, and the brightness of the image is adjusted using an equalization algorithm (such as the cv2.equalizeHist function of OpenCV) to minimize the brightness difference between different regions. The image of each region is equalized to ensure that the brightness distribution of all regions is consistent, and multi-region image equalization data is obtained. This process helps to eliminate image deviations caused by brightness differences in different regions, making image processing and analysis results more accurate.

[0092] Step S24: performing noise suppression on the multi-region image equalization data and performing contrast enhancement to improve the visibility of image details, thereby obtaining multi-region image comparison data;

[0093] When performing noise suppression and contrast enhancement on multi-region image equalization data, the embodiment of the present invention first applies a noise removal algorithm (such as Gaussian filtering or median filtering) to suppress the image noise. A filter is used to reduce random noise in the image to ensure image quality. Next, a contrast enhancement algorithm (such as adaptive histogram equalization or CLAHE) is applied to improve the visibility of image details and enhance the contrast in the image to make defects more obvious. Ultimately, the obtained multi-region image comparison data has higher image quality and clearer details, which is helpful for subsequent defect detection.

[0094] Step S25: performing pixel-level comparison on the multi-region image comparison data according to the standard display characteristic map data, and calculating the deviation value of each pixel point, thereby obtaining a pixel deviation map;

[0095] When the embodiment of the present invention performs pixel-level comparison on the multi-region image comparison data according to the standard display characteristic map data, the standard display characteristic map data and the multi-region image comparison data are first aligned to the same coordinate system. A pixel-level comparison algorithm (such as mean square error or difference map generation) is used to calculate the brightness difference of each pixel in the standard display characteristic map and the comparison map. A pixel deviation map is generated to display the deviation value of each pixel. These deviation values ​​indicate the change in the display state of each pixel, providing a basis for identifying potential defect areas.

[0096] Step S26: Identify potential defect areas based on the pixel deviation map to obtain preliminary defect detection data.

[0097] When the embodiment of the present invention identifies potential defect areas based on the pixel deviation map, a threshold segmentation algorithm (such as Otsu's method or adaptive threshold) is first applied to mark the significant deviation areas in the deviation map as potential defect areas. Connected domain analysis is performed on the identified defect areas to merge adjacent defect areas and eliminate isolated small areas. According to these steps, preliminary defect detection data is generated, in which the marked areas are possible defect areas. These data can be used for further detailed analysis and defect confirmation.

[0098] The present invention uses a high-resolution camera array to image the central area, edge area and four corner areas of the liquid crystal screen, which can ensure that each area is fully monitored. The optical performance of each area of ​​the liquid crystal screen may be different, especially the edges and corners are prone to uneven brightness or color distortion. This step helps to fully collect the image data of the liquid crystal screen. The use of the camera array can capture high-resolution images of multiple areas at the same time to ensure consistent image quality. This high-resolution imaging method can detect tiny defects and improve the accuracy of detection. When the liquid crystal screen is imaged, due to the different camera positions and angles, the image may be geometrically distorted, especially in the edge and four corner areas. Geometric correction can eliminate these distortions and ensure the accuracy of subsequent image processing and analysis. Through geometric correction, the image data of all areas can be uniformly processed under an accurate coordinate system, providing a stable basis for subsequent pixel-level comparison and defect detection. Different areas of the liquid crystal screen may have brightness differences due to uneven backlight or changes in camera viewing angles during the imaging process; brightness equalization processing can ensure the brightness consistency of the entire image by compensating for these differences, and avoid local brightness being too high or too low to affect the results of defect detection. The image after brightness equalization can more clearly show the true performance of the screen, help identify subtle differences in the screen surface and backlight structure, and improve the accuracy of subsequent detection. When the LCD screen is imaged, it may be affected by external noise, resulting in unnecessary noise in the image; through noise suppression, these interference factors can be removed to ensure the clarity of the image, making defect detection more reliable. Contrast enhancement technology can highlight subtle differences in the image, making potential defects more obvious, especially small brightness unevenness, color distortion or sub-pixel arrangement problems, to ensure that these details are not ignored in subsequent analysis. By comparing the processed multi-region image with the standard display characteristic map generated in step S16 at the pixel level, the difference between the performance of each pixel and the standard can be accurately identified. This process ensures that any minor anomalies, such as brightness or color shift of a single pixel, can be found, greatly improving the accuracy of defect detection. The calculation of the deviation value can quantify the degree of deviation of each pixel and provide detailed defect data. These deviation values ​​provide an important reference for defect analysis, classification and repair in subsequent steps. By analyzing the pixel deviation map, potential defect areas can be automatically identified. This automated identification method reduces the error and time of manual inspection and improves the efficiency of defect detection. The identified potential defect areas lay the foundation for subsequent detailed analysis and classification. The preliminary defect detection data can help determine the nature of the defect (such as brightness problems, color distortion, etc.), providing effective data support for quality control of the production line. Through the implementation of these steps, the LCD screen's multi-area imaging, correction, equalization, noise suppression and contrast enhancement processing methods have greatly improved the image quality and defect detection accuracy.The resulting pixel deviation map and potential defect area identification results provide reliable data support for subsequent defect classification and repair. This method can efficiently and accurately identify potential defects in LCD screens, reduce misjudgments, and improve production quality control levels.

[0099] Preferably, step S26 comprises the following steps:

[0100] Step S261: marking the areas in the pixel deviation map where the deviation exceeds a preset threshold as potential defect areas by using an adaptive threshold segmentation algorithm, thereby obtaining a preliminary defect area map;

[0101] When the embodiment of the present invention uses an adaptive threshold segmentation algorithm to mark the area in the pixel deviation map where the deviation exceeds a preset threshold as a potential defect area, the local mean and standard deviation of the deviation map are first calculated. Then, the threshold of each pixel is determined using an adaptive threshold segmentation method (such as an adaptive Gaussian threshold or an adaptive mean threshold). Each pixel in the deviation map is compared with its adaptive threshold. If the pixel deviation exceeds the preset threshold, it is marked as a potential defect area. Finally, a preliminary defect area map is generated, in which the area where the deviation exceeds the threshold is marked as a potential defect area, which is convenient for further processing and analysis.

[0102] Step S262: performing morphological processing on the preliminary defect region map to eliminate noise and connect adjacent defect regions, thereby obtaining a preferred defect region map;

[0103] When the embodiment of the present invention performs morphological processing on the preliminary defect region map, morphological filtering techniques (such as opening and closing operations) are first applied to remove noise. The opening operation removes small isolated noise points by first corroding and then dilating; the closing operation connects adjacent defect regions by first dilating and then corroding. These operations are performed using structural elements (such as circular or rectangular kernels) to enhance the main defect regions in the image and eliminate small artifacts. After morphological processing, a preferred defect region map is obtained, which significantly reduces noise and connects adjacent defect regions, thereby improving the accuracy of defect detection.

[0104] Step S263: performing connected domain analysis on the preferred defect region map, and marking and numbering each independent defect region, thereby obtaining a defect region marking map;

[0105] When the embodiment of the present invention performs connected domain analysis on the preferred defect area map, a connected component labeling algorithm (such as eight-neighborhood connection) is first applied to identify and label all independent defect areas in the image. The algorithm traverses each pixel in the image, labels the connected pixels as the same defect area, and assigns a unique number. A labeling algorithm (such as Blob analysis) is used to count the properties of each connected domain, including area, shape, and boundary information, to generate a defect area labeling map. Each labeled area represents an independent defect area, providing detailed information for subsequent feature extraction and analysis.

[0106] Step S264: extracting features of each marked defect area according to the defect area marking map and the multi-area image data, thereby obtaining preliminary defect detection data.

[0107] When the embodiment of the present invention extracts features of each marked defect area based on the defect area marking map and the multi-region image data, each marked defect area is first located in the multi-region image data. Feature extraction is performed on each defect area, including calculating the geometric features (such as area, perimeter, shape factor) and photometric features (such as average brightness, contrast, color distribution) of the area. These feature data are extracted using image analysis tools (such as OpenCV's cv2.contourArea function and cv2.mean function). These feature data are compared with the standard display characteristic map data to obtain preliminary defect detection data. In this way, the detailed information of each defect area is identified and recorded to provide basic data for subsequent analysis.

[0108] The present invention can dynamically adjust the threshold according to the numerical value of the pixel deviation map through an adaptive threshold segmentation algorithm, and mark the area where the deviation exceeds the preset threshold as a potential defect area; this adaptive threshold method adapts to different display environments and optical conditions, improves the sensitivity of the detection system to various defects, and ensures that slight but potentially meaningful defects can be identified. The adaptive algorithm can dynamically adjust the threshold according to the overall optical characteristics of the image, avoid the problem of false alarm or omission caused by a single fixed threshold, and improve the accuracy of defect identification. In the preliminary defect area map, pseudo defect areas may appear due to noise interference. Through morphological processing, such as corrosion and expansion operations, isolated noise points can be effectively removed to ensure the reliability of the detection results and reduce unnecessary pseudo defect markings. Morphological processing can connect adjacent defect areas to ensure that defects close to each other are reasonably merged into a larger area. This helps to detect those scattered defects, especially small but related defects at the edge of the LCD screen or in complex patterns, thereby improving the overall defect detection effect. Through connected domain analysis, each independent defect area in the image can be identified and numbered to ensure that each defect area is clearly marked. This method helps the subsequent classification and analysis of different types of defects and provides a basis for accurately evaluating the properties of each defect area. The numbered defect area marking map provides an intuitive visual reference for the distribution of all defect areas on the LCD screen, allowing the inspection system to quickly locate and analyze defects in different areas, which helps to classify and repair defects more efficiently. Based on the defect area marking map and multi-region image data, the system can extract the key features of each defect area, including defect area, shape, brightness difference, color anomaly, etc. This detailed feature extraction process helps to gain a deep understanding of the nature of each defect, thereby providing data support for further classification (such as temporary defects and permanent defects). By extracting features from all marked defect areas, the system generates preliminary defect detection data and provides a comprehensive inspection report. This data can be effectively used for LCD screen quality assessment to help production lines quickly decide whether a certain screen needs to be further repaired or eliminated. These four steps gradually and accurately mark, optimize and classify defect areas through technical means such as adaptive threshold segmentation, morphological processing, connected domain analysis and feature extraction. Each step enhances the accuracy and reliability of defect detection, ensuring that all potential defects in the LCD screen can be identified and classified, and high-quality inspection data can be generated. The final preliminary defect detection data lays a solid foundation for subsequent defect classification and repair, and improves the quality control efficiency in the production process.

[0109] Preferably, step S3 comprises the following steps:

[0110] Step S31: setting a plurality of different LCD screen refresh frequencies, and acquiring a plurality of frames of images at each refresh frequency, thereby obtaining a multi-refresh frequency image sequence, wherein the refresh frequencies include 30 Hz, 60 Hz, 120 Hz and 240 Hz;

[0111] When setting multiple different LCD screen refresh frequencies and collecting multiple frames of images in the embodiment of the present invention, the LCD screen display is first configured with different refresh frequencies (such as 30Hz, 60Hz, 120Hz and 240Hz) to ensure that the display content is consistent under each frequency setting. Use a high-resolution camera array to capture the screen to ensure that the entire screen area is covered, and capture several frames of images at each refresh frequency. The image sequence at each frequency will be stored separately to form a multi-refresh frequency image sequence data set. This ensures that sufficient data is collected at different refresh frequencies to analyze the performance of defects under various conditions.

[0112] Step S32: performing the preprocessing of step S2 and defect area recognition on the multi-refresh rate image sequence to obtain corresponding preliminary defect detection data, thereby forming a multi-refresh rate defect detection data set;

[0113] When the embodiment of the present invention preprocesses the multi-refresh frequency image sequence, the image preprocessing algorithm (such as denoising and brightness equalization) is first applied to improve the image quality. Then, the image processing method in step S2 is used to identify the defect area of ​​the image sequence at each refresh frequency. This includes calculating the pixel deviation map of the image and performing pixel-level comparison based on the standard display characteristic map data to identify the potential defect area. These processing results are summarized to form a defect detection data set at multiple refresh frequencies, providing comprehensive data on defect information at different refresh frequencies.

[0114] Step S33: performing time series analysis on each defect area in the multi-refresh frequency defect detection data set, and tracking the change of defects at different refresh frequencies and time points, so as to obtain defect change trajectory data;

[0115] When the embodiment of the present invention performs time series analysis on the defect area in the multi-refresh frequency defect detection data set, the defect data at each refresh frequency is first sorted to construct a time series. Use time series analysis tools (such as dynamic time warping or recursive neural networks) to model the changes in the defect area at different time points. Track and record the change trend of the defect area over time to form defect change trajectory data. This can reveal the dynamic changes of defects at different time points and refresh frequencies, providing support for further analysis.

[0116] Step S34: using image registration technology to align the multi-refresh frequency defect detection data set to the same coordinate system, and eliminate the small displacement caused by screen refresh, so as to obtain a registered multi-frequency defect map;

[0117] When the embodiment of the present invention uses image registration technology to align multi-refresh frequency defect detection data sets to the same coordinate system, a reference frequency image is first selected as a benchmark. An image registration algorithm (such as registration based on feature point matching or registration method based on mutual information) is used to align images at different refresh frequencies. The slight displacement caused by screen refresh is eliminated to ensure that image data at all frequencies are aligned to the same coordinate system. Finally, the registered multi-frequency defect map is obtained, which is convenient for subsequent image processing and analysis.

[0118] Step S35: superimposing multiple frames of the multi-frequency defect image, and fusing the multi-frame information using a median filtering method, thereby obtaining an enhanced defect image;

[0119] When the embodiment of the present invention performs multi-frame image superposition on the multi-frequency defect map, all aligned image sequences are first stacked. Multi-frame information fusion is performed using the median filtering method, which can reduce noise and enhance the real defect features in the image. The values ​​of each pixel in all frames are sorted, and the median is taken as the pixel value of the final image to generate an enhanced defect map. This can improve the visibility of defects and reduce interference introduced by noise or image capture errors.

[0120] Step S36: performing variance analysis on the enhanced defect image in the time dimension, and identifying regions with inconsistent performance at different refresh frequencies, thereby obtaining dynamic defect candidate regions;

[0121] When performing variance analysis on the enhanced defect image in the time dimension, the embodiment of the present invention first calculates the variance of the brightness or pixel value of the defect area at different time points and refresh frequencies. A statistical analysis tool is used to calculate the variance of each area in the time dimension to identify areas with inconsistent performance. Variance analysis can reveal the dynamic characteristics of defects at different refresh frequencies, thereby determining dynamic defect candidate areas.

[0122] Step S37: performing time characteristic defect analysis based on dynamic time warping on the dynamic defect candidate area, thereby obtaining dynamic defect detection data;

[0123] When the embodiment of the present invention performs time characteristic defect analysis based on dynamic time warping on dynamic defect candidate areas, the time series data of each candidate area is first extracted. The dynamic time warping (DTW) algorithm is applied to analyze the time series data to evaluate the change pattern and time consistency of the defects. In this way, dynamic defect detection data is obtained to identify and analyze those defect areas that show significant changes in time.

[0124] Step S38: performing stability evaluation of defect points at different refresh frequencies and time points based on the dynamic defect detection data and defect change trajectory data, thereby obtaining defect persistence score data;

[0125] When the embodiment of the present invention performs stability assessment of defect points based on dynamic defect detection data and defect change trajectory data, the defect points in the dynamic defect detection data are first compared with their change trajectory data. The stability score of each defect point at different refresh frequencies and time points is calculated. The defect points are scored using a stability scoring model (such as a statistically based stability assessment or a machine learning model) to obtain defect persistence scoring data. This helps to identify which defects are persistent and which may be temporary.

[0126] Step S39: Classify defect points into temporary defects and permanent defects according to the defect persistence score data, thereby obtaining defect point classification data, wherein defects with persistence scores higher than a preset threshold are classified as permanent defects, and defects with persistence scores lower than or equal to the preset threshold are classified as temporary defects.

[0127] When the embodiment of the present invention classifies defect points into temporary defects and permanent defects according to the defect persistence score data, a confidence threshold is first set to distinguish temporary defects from permanent defects. The stability score of each defect point is compared with the threshold, and defect points with scores higher than the threshold are classified as permanent defects, and defect points with scores lower than or equal to the threshold are classified as temporary defects. Finally, defect point classification data is obtained to provide a basis for defect management and repair.

[0128] The performance of the liquid crystal screen of the present invention may be different at different refresh frequencies. By setting multiple frequencies such as 30Hz, 60Hz, 120Hz, 240Hz, and collecting multiple frames of images, the performance of the liquid crystal screen at various refresh frequencies can be fully understood. This multi-frequency acquisition method helps to capture dynamic defects that appear at specific frequencies. Some defects may not be obvious at low refresh frequencies, but they will become more obvious or recur at high refresh frequencies. Therefore, by covering multiple frequencies, the coverage of defect detection can be improved to ensure the comprehensiveness of the detection results. The same image preprocessing and defect area identification process as step S2 is performed at each refresh frequency, and preliminary defect detection data at multiple frequencies can be obtained. By analyzing the defect performance at different frequencies, the system can find some frequency-sensitive defects and avoid missing detection. After generating a multi-frequency defect detection data set, a multi-dimensional rich data set can be provided for subsequent dynamic defect analysis, which improves the understanding of the dynamic behavior of defects. By performing time series analysis on each defect area and tracking the changes of defects at different refresh frequencies, the system can generate defect change trajectories. This is very important for understanding the behavior of dynamic defects, such as some defects may expand, shrink or disappear with changes in refresh frequency. The defect change trajectory data provides a basis for subsequent defect classification, which can be used to determine the stability and persistence of defects and help distinguish temporary defects from permanent defects. The screen may have slight displacements at different refresh frequencies. Using image registration technology to align multi-frequency images to the same coordinate system can eliminate these displacement errors and ensure the accuracy of defect analysis results. Through the registered multi-frequency defect map, it can be ensured that the images collected at different frequencies are in the same reference frame, which is crucial for the subsequent superposition and fusion of multi-frequency defect maps. By superimposing multiple frames of images at different refresh frequencies and fusing the information using the median filtering method, the details of the defect area can be effectively enhanced and noise interference can be reduced. The superimposed image can more clearly show the outline and range of the defect. Median filtering can not only remove errors caused by external noise, but also improve the signal-to-noise ratio of the defect image, so that the enhanced defect map generated in the end can more reliably reflect the real defects in the LCD screen. Through variance analysis in the time dimension, areas with inconsistent performance at different refresh frequencies can be identified, which are often candidate areas for dynamic defects. Defects that are inconsistent at different frequencies may reflect problems with the dynamic response of the screen, which is particularly important for quality assessment of high refresh rate displays. Variance analysis helps to further accurately locate dynamic defect areas, reduce misjudgments, and ensure that only those areas that truly show dynamic instability are marked as candidate defect areas. Through dynamic time warping (DTW) technology, the time series of defect areas can be deeply analyzed to identify the characteristic changes of dynamic defects at different refresh frequencies.DTW technology can capture the nonlinear changes of defect areas over time or frequency, and effectively analyze complex dynamic defects. The temporal feature analysis of dynamic defects can reveal how defects change over time and refresh frequency, providing in-depth insights for subsequent classification and repair. By analyzing the stability of defect points at different refresh frequencies and time points, the system can score the defects for persistence, which can quantify the stability of defects and provide a clear indicator to evaluate the severity and persistence of each defect. The persistence score of defects provides a scientific basis for distinguishing temporary defects from permanent defects. The higher the score, the more likely it is to be a permanent problem, which helps to optimize the repair and processing strategy of LCD screens. According to the persistence score data of defects, defect points can be divided into temporary defects and permanent defects. Through this classification method, the production line can take corresponding treatment measures for different types of defects. Temporary defects may be caused by environmental factors, while permanent defects require further repair or replacement. After defect classification, it can help the production line to screen and repair screens more effectively, reduce waste caused by misjudgment, and improve production efficiency and product quality. Through the implementation of these steps, dynamic defects of LCD screens at different refresh rates can be efficiently and accurately identified and classified. Each step enhances the detection and understanding of dynamic defects, especially the distinction between temporary defects and permanent defects, making the overall inspection process more intelligent and automated, which helps to improve the quality control efficiency of LCD screens.

[0129] Preferably, step S4 comprises the following steps:

[0130] Step S41: extracting the location information and surrounding area of ​​each defect point according to the defect point classification data, thereby obtaining defect area data;

[0131] When extracting the location information and surrounding area of ​​each defect point according to the defect point classification data, the embodiment of the present invention first extracts the coordinate position of each defect point from the defect point classification data. The bounding box of the defect area is defined using these coordinates, which is usually a rectangular area centered on the defect point. The size of the bounding box can be adjusted according to preset expansion parameters (such as expanding a certain number of pixels to cover the surrounding area of ​​the defect). Then, by cropping the image area within each bounding box, image data containing the defect point and its surrounding area is obtained to form defect area data.

[0132] Step S42: configuring a multi-light source imaging system, including a visible light source, a near-infrared light source, and an ultraviolet light source, thereby obtaining a multi-light source illumination configuration;

[0133] When configuring a multi-light source imaging system in an embodiment of the present invention, first select and install a visible light source, a near-infrared light source, and an ultraviolet light source on a fixed camera system. Ensure that the position and angle of each light source are fixed to avoid introducing changes during the image acquisition process. Use a light source controller to adjust the brightness and illumination angle of each light source separately to cover the display area of ​​the entire LCD screen. Calibrate each light source to ensure their consistency in spectral range and brightness, thereby obtaining a complete multi-light source lighting configuration.

[0134] Step S43: performing multi-light source image acquisition on the defect area data according to the multi-light source illumination configuration, and performing image alignment, thereby obtaining multi-spectral image data;

[0135] When the embodiment of the present invention performs image acquisition of defect area data according to the multi-light source illumination configuration, the visible light source, the near-infrared light source and the ultraviolet light source are first activated one by one on each defect area. A high-resolution camera is used to capture images of the defect area under each light source condition to ensure that a clear image can be obtained under each light source illumination condition. Afterwards, the acquired images are aligned, and image registration technology (such as feature point matching or grayscale-based registration algorithm) is used to align the images under different light source conditions to the same coordinate system, thereby obtaining multispectral image data.

[0136] Step S44: performing spectral decomposition on the multispectral image data, extracting the reflectivity or emissivity information of each pixel point in different bands, thereby obtaining spectral characteristic data of the defect area;

[0137] When the embodiment of the present invention performs spectral decomposition on multispectral image data, a spectral decomposition algorithm (such as principal component analysis or independent component analysis) is first applied to separate the spectral information of each pixel from multiple bands. For each pixel, its reflectivity or emissivity data in the visible light, near infrared and ultraviolet bands are extracted. Then, the data of each band is analyzed to form spectral feature data of each pixel. In this way, spectral feature data of the defect area can be obtained, reflecting the spectral characteristics of different bands.

[0138] Step S45: performing statistical analysis on the average spectral response, standard deviation and spectral curve shape characteristics of the defect area spectral characteristic data in different bands, and constructing the spectral fingerprint of each defect area, thereby obtaining the spectral fingerprint data of the defect area, wherein the spectral fingerprint includes the reflection peak, absorption valley and spectral ratio of the characteristic band;

[0139] When the embodiment of the present invention performs statistical analysis on the spectral characteristic data of the defect area, the average spectral response and standard deviation of each defect area in different bands are first calculated. The main features of the spectral curve, including reflection peaks, absorption valleys and spectral ratios, are extracted using spectral curve analysis methods (such as smoothing and curve fitting). Combined with these characteristic data, the spectral fingerprint of each defect area is constructed to generate spectral fingerprint data of the defect area, which can be used for further spectral comparison analysis.

[0140] Step S46: performing multi-spectral comparative analysis on the defect area spectral fingerprint data to obtain multi-spectral comparative result data, wherein the multi-spectral comparative analysis includes calculating the correlation coefficient between each defect area in the visible light, near infrared and ultraviolet spectra, quantifying the similarity between different bands, and extracting the main change components in the defect area spectral fingerprint data using the principal component analysis method;

[0141] When the embodiment of the present invention performs multi-spectral comparative analysis on the spectral fingerprint data of the defect area, the correlation coefficient between the visible light, near infrared and ultraviolet spectra of each defect area is first calculated. The similarity between different bands is quantified using statistical tools (such as correlation analysis or regression analysis). The principal component analysis (PCA) method is used to extract the main variation components in the spectral fingerprint data, thereby identifying the main patterns and differences in the spectral features. These comparison results can be used to identify the type and characteristics of the defect.

[0142] Step S47: extracting the differences between different spectra based on the multi-spectrum comparison result data, thereby obtaining spectrum difference data.

[0143] When extracting spectral differences based on multi-spectral comparison result data, the embodiment of the present invention first compares the spectral fingerprint of each defect area with the spectral fingerprint of other areas. The difference between each band is calculated, and the spectral difference data is extracted using a difference calculation method (such as absolute difference, relative difference or ratio difference). These differences are analyzed to identify significant changes under different spectral conditions, thereby obtaining detailed spectral difference data for further defect analysis and classification.

[0144] The present invention can provide an accurate spatial range for subsequent multi-light source imaging and analysis by extracting the location information and surrounding area of ​​each defect point. This helps to concentrate resources in key areas, avoid redundant processing of the entire LCD screen, and improve detection efficiency. The accurately extracted defect area data provides a reliable basis for spectral analysis in subsequent steps, ensuring that the data collected under different light sources can accurately reflect the optical characteristics of the area. By introducing multiple light sources such as visible light, near-infrared light, and ultraviolet light, the defect area of ​​the LCD screen can show different characteristics under light of different wavelengths, thereby providing richer optical feature data. Different light sources have different sensitivities to different types of defects, and defects can be detected more comprehensively. Multi-light source imaging helps to reveal defects that are not obvious under a single light source, especially tiny defects that are not easy to find under visible light may be more obvious in the near-infrared or ultraviolet bands, thereby improving the accuracy of overall detection. Under the irradiation of multiple light sources, the defect area is imaged and aligned, and spectral data covering multiple wavelengths can be generated. In this way, the comprehensive performance of the defect area in different bands can be obtained, thereby more deeply analyzing the optical characteristics and defect characteristics of the LCD screen. The images are aligned to ensure that the images collected under different light sources can accurately correspond to the same defect area, avoid errors caused by displacement, and ensure the reliability of subsequent spectral analysis results. By extracting the reflectivity or emissivity information of each pixel in different bands through spectral decomposition, the optical characteristics of the defect area can be understood in more detail. The spectral response of different bands can reveal the differences in material and surface characteristics and help identify potential defect types. The data after spectral decomposition can reveal the response differences of the defect area under different spectra, providing detailed data support for subsequent spectral fingerprint analysis. The spectral fingerprint of each area is constructed by statistically analyzing the average spectral response, standard deviation and spectral curve shape of each defect area in different bands. These spectral fingerprints include reflection peaks, absorption valleys and spectral ratios, which can generate unique optical "fingerprints" for each defect area and improve the ability to distinguish different types of defects. Spectral fingerprints provide multi-dimensional data for LCD screen defects, which can not only identify defects, but also further analyze their deep-level characteristics, such as material or process defects. By calculating the correlation coefficient between different bands and quantifying the similarity of defect areas under visible light, near-infrared and ultraviolet spectra, subtle changes between different bands can be found. This comparative analysis helps to reveal the differences in defect performance from multiple spectral dimensions and comprehensively evaluate defect characteristics. By extracting the main variation components in the spectral fingerprint data through principal component analysis (PCA), data redundancy can be reduced, the main characteristics of the defects can be highlighted, and the analysis efficiency and accuracy can be improved. Spectral difference data can reveal areas that perform inconsistently under different spectra. Some defects may not be obvious in the visible spectrum, but have obvious reflection or absorption characteristics in the near-infrared or ultraviolet bands.By extracting these differences, the system can more accurately identify and locate defects; the extraction of spectral differences can help eliminate false positive or false negative detection results, optimize the accuracy and precision of defect identification, and thus obtain more accurate detection results. LCD screen defect detection based on multi-light source and multi-spectral analysis technology can deeply explore the performance characteristics of defect areas in different bands, and use spectral fingerprint construction, spectral comparison and difference analysis technologies to comprehensively improve the accuracy and meticulousness of defect detection. This series of steps can effectively eliminate interference factors within the spectral range, accurately identify potential defects of LCD screens, and provide detailed spectral data support.

[0145] Preferably, step S5 comprises the following steps:

[0146] Step S51: constructing a multidimensional feature vector according to the spectral difference data and its corresponding preliminary defect detection data, thereby obtaining a defect spectral feature vector;

[0147] When the embodiment of the present invention constructs a multidimensional feature vector based on the spectral difference data and its corresponding preliminary defect detection data, the feature values ​​of each defect point in different spectral bands are first extracted from the spectral difference data. These feature values ​​include the reflectivity or emissivity data of each pixel point in each band, the shape characteristics of the spectral curve, etc. These feature values ​​are combined with the preliminary defect detection data (such as the location, size, type, etc. of the defect) to form a feature vector containing multiple dimensions. Each dimension represents a specific spectral feature or detection parameter, thereby obtaining a defect spectral feature vector for subsequent classification and analysis.

[0148] Step S52: classifying the defect spectrum feature vector based on the support vector machine to obtain a defect authenticity prediction result;

[0149] When the embodiment of the present invention classifies the defect spectral feature vector based on the support vector machine (SVM), the defect spectral feature vector is first used as input data, and the corresponding label data (such as real defects and false positives or false negatives) is prepared. A suitable kernel function (such as a linear kernel or a radial basis kernel) is selected to train the support vector machine model. Through the training process, the SVM model will learn how to distinguish different types of defects based on the spectral feature vector. The trained SVM model is used to predict the new data to obtain the defect authenticity prediction result, that is, each defect point is predicted as a real defect, a false positive or a false negative.

[0150] Step S53: According to the defect authenticity prediction result, the defect points predicted as false positive are marked as potential false positive areas, and the areas that are not detected but have abnormal spectral characteristics are marked as potential false negative areas, so as to obtain potential misjudgment area data;

[0151] When the embodiment of the present invention marks the defect points predicted as false positives as potential false positive areas according to the defect authenticity prediction results, the defect points marked as false positives are first screened out from the SVM classification results. These defect points may not actually exist but are mistaken for defects. These areas are marked as potential false positive areas. At the same time, from the areas that have not been detected but have abnormal spectral characteristics, these areas are identified and marked as potential false negative areas, that is, areas where defects may exist but have not been detected; finally, the potential misjudgment area data is sorted out, including false positive areas and false negative areas.

[0152] Step S54: resampling the potential misjudgment area data with high resolution and performing local contrast enhancement to obtain an enhanced misjudgment area image;

[0153] When the embodiment of the present invention performs high-resolution resampling on the data of the potential misjudgment area, a high-resolution imaging device is first selected to resample these areas to obtain more detailed image data. The resolution of the image is improved by using an interpolation method (such as bilinear interpolation or cubic interpolation). Next, local contrast enhancement processing is performed, and the visibility of details in the image is enhanced by an image processing algorithm (such as histogram equalization or contrast-limited adaptive histogram equalization), thereby obtaining an enhanced misjudgment area image, which will help improve the accuracy of subsequent analysis.

[0154] Step S55: performing a refined analysis on the enhanced misjudged area image to obtain misjudged area confidence data, and screening potential false positive areas according to the misjudged area confidence data and a preset confidence threshold, removing defect points with confidence values ​​lower than the threshold, thereby obtaining optimized defect point data;

[0155] When the embodiment of the present invention performs a refined analysis on the enhanced misjudged area image, it first uses an image processing algorithm (such as edge detection and texture analysis) to perform an in-depth analysis of the image to determine the detailed features of each misjudged area. The confidence data of these areas are calculated, including the probability of defect existence and the accuracy of recognition. According to the preset confidence threshold, the potential false positive areas are screened, and the areas with confidence below the threshold are removed, so as to obtain optimized defect point data, ensuring that the retained defect points have a high recognition accuracy.

[0156] Step S56: performing morphological processing and connected domain analysis on the potential false negative areas in the enhanced misjudged area image, and identifying small or low-contrast defects that may be ignored, thereby obtaining supplementary defect point data;

[0157] When performing morphological processing on potential false negative areas in the enhanced misjudged area image, the embodiment of the present invention first applies morphological operations (such as dilation and erosion) to highlight small or low-contrast defects in the image. Then, a connected domain analysis is performed to identify and mark small defect areas that may be ignored through algorithms (such as connected component marking). Based on the analysis results, supplementary defect point data is generated, that is, the identified defect areas are supplemented to make up for the omissions in the detection.

[0158] Step S57: Merge the optimized defect point data and the supplemented defect point data, perform spatial cluster analysis, merge adjacent defect points, and thus obtain accurate defect point data.

[0159] When the embodiment of the present invention merges the optimized defect point data and the supplemented defect point data, the optimized defect point data and the supplemented defect point data are first combined to form a comprehensive defect data set. Then, a spatial clustering analysis (such as K-means clustering or DBSCAN algorithm) is performed to cluster the defect points and merge adjacent or similar defect points into a single defect area. In this way, repeated defect point records can be reduced, accurate defect point data can be obtained, and the actual defect situation of the LCD screen can be accurately reflected.

[0160] The present invention constructs spectral difference data and preliminary defect detection data into multi-dimensional feature vectors, and the system can comprehensively consider the performance of defects in multiple dimensions, such as spectrum, spatial position, brightness difference, etc. In this way, a richer and more detailed defect representation can be formed, which helps to improve the accuracy of defect identification. The detection data from different sources are integrated into feature vectors to ensure that different types of data can be processed and analyzed under a unified standard, providing reliable input for subsequent classification algorithms. Support vector machines (SVMs) can efficiently process multi-dimensional feature data, especially suitable for small samples and nonlinear problems. Through SVM classification, the authenticity of defect points can be accurately predicted to reduce the possibility of misjudgment. The powerful classification ability of SVM can effectively distinguish true and false defects, reduce the occurrence of false positives and false negatives, and ensure that the detection results are more accurate. Through the classification results, possible false positive areas and potential false negative areas that have not been detected but have abnormal spectral characteristics can be marked. This process can further reduce misjudgments and missed judgments and improve the comprehensiveness of detection. Marking potential misjudgment areas can provide direction for subsequent refined analysis, concentrate resources to process these key areas, and improve detection efficiency. Through high-resolution resampling and local contrast enhancement, the visibility of details in potential misjudgment areas can be improved, so that subsequent analysis can capture tiny defects that might have been overlooked. Image enhancement technology can help better identify low-contrast defects, making detection more sensitive and accurate. Based on confidence analysis, a confidence value can be assigned to each potential false positive area to reflect the possibility of its authenticity. This helps to distinguish possible real defects from misjudged areas. By setting a confidence threshold, defect points with low confidence can be effectively removed, thereby reducing false positives and further improving the accuracy of defect detection. For false negative areas that may be ignored, morphological processing and connected domain analysis can be performed to effectively identify defects that have not been initially detected due to small size or low contrast, thereby increasing the comprehensiveness of detection. This process ensures that even tiny or imperceptible defects can be detected, further improving the coverage of LCD defect detection. By merging optimized defect point data and supplementary defect point data, it is ensured that all detected defect points are fully considered to avoid omissions. Using spatial clustering analysis to merge adjacent defect points can reduce redundant information, optimize the number and distribution of defect points, and further improve detection efficiency and accuracy. Through multi-dimensional feature vector construction, support vector machine classification, and refined identification and analysis of false positives and false negatives, this step can effectively reduce misjudgments, optimize defect detection results, and supplement undetected potential defect points, ultimately obtaining high-precision defect detection data. This method not only improves detection accuracy, but also enhances sensitivity to tiny and dynamic defects, providing more reliable quality assurance for LCD production.

[0161] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0162] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for detecting defects in a liquid crystal screen based on machine vision, characterized in that: The following steps are involved: Step S1: acquiring panel structure data of the liquid crystal screen, including panel size data, pixel arrangement data, backlight module data, and surface treatment information data; performing optical characteristic analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping different display states, display contents, and normal display states under viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data; Step S2: collecting multi-region image data of the LCD panel; performing image preprocessing on the multi-region image data, and performing difference analysis according to the standard display characteristic diagram data to identify potential defect areas, thereby obtaining preliminary defect detection data; Step S3: Tracking defect changes of preliminary defect detection data at different refresh frequencies by using a multi-frame image overlay technique, thereby obtaining dynamic defect detection data; According to the dynamic defect detection data, the defect points are classified based on temporary and permanent defects, so as to obtain defect point classification data; Step S4: Obtain defect area data according to defect point classification data; Acquiring spectral characteristics of the LCD screen image under different light sources for the defect area data, thereby obtaining spectral characteristic data of the defect area, wherein the different light sources include visible light, near infrared light and ultraviolet light; Perform multi-spectral comparison on the spectral feature data of the defect area, and extract the differences between different spectra to obtain spectral difference data; Step S5: performing false positive and false negative identification on the defect area data according to the spectral difference data, and optimizing the defect point accuracy, so as to obtain accurate defect point data.

2. The method for detecting defects of a liquid crystal display based on machine vision according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: measuring the physical dimensions of the liquid crystal screen and recording the dimensions of the effective display area, thereby obtaining panel dimension data; Step S12: collecting the number of rows and columns of the pixel matrix of the liquid crystal screen and the sub-pixel arrangement, and using high-resolution imaging analysis to measure the pixel pitch and sub-pixel size, thereby obtaining pixel arrangement data; Step S13: collecting the backlight type and the arrangement and quantity of the backlight units, and performing disassembly analysis and brightness measurement to obtain backlight module data; Step S14: collecting the surface treatment type and measuring the surface roughness of the liquid crystal screen to obtain surface treatment information data; Step S15: merging the panel size data, pixel arrangement data, backlight module data and surface treatment information data into panel structure data; Step S16: performing optical characteristic analysis of the liquid crystal screen based on light wave interference according to the panel structure data, and mapping different display states, display contents and normal display states under different viewing angles according to a preset image data set, thereby obtaining standard display characteristic diagram data.

3. The method for detecting LCD screen defects based on machine vision according to claim 2, characterized in that: Step S16 includes the following steps: Step S161: constructing a liquid crystal screen optical model according to the panel structure data, wherein the liquid crystal screen optical model includes liquid crystal molecule arrangement, polarizer characteristics, and backlight propagation path; Step S162: simulating the propagation characteristics of light of different wavelengths in the liquid crystal layer according to the optical model of the liquid crystal screen by using the principle of light wave interference, and calculating the optical response of the liquid crystal screen under different driving voltages, thereby obtaining optical response simulation data; Step S163: generating theoretical display characteristic data at different grayscales, colors and viewing angles according to the optical response simulation data; Step S164: mapping and converting the preset image data set according to the theoretical display characteristic data, thereby generating simulated display effect data under different display states, wherein the preset image data set includes display contents of pure color, grayscale gradient, text, and image; Step S165: Generate standard display characteristic diagram data including different display states, display contents and viewing angle information according to the simulated display effect data and the theoretical display characteristic data.

4. The method for detecting defects of a liquid crystal display based on machine vision according to claim 3, characterized in that: Step S2 includes the following steps: Step S21: using a high-resolution camera array to perform multi-region imaging on the LCD panel, and collecting multi-region image data of the center region, the edge region, and the four corner regions; Step S22: performing geometric correction on the multi-region image data to eliminate image distortion caused by camera position and angle, thereby obtaining multi-region image correction data; Step S23: performing brightness equalization processing on the multi-region image correction data to obtain multi-region image equalization data, wherein the brightness equalization processing includes compensating for brightness differences between different regions; Step S24: performing noise suppression on the multi-region image equalization data and performing contrast enhancement to improve the visibility of image details, thereby obtaining multi-region image comparison data; Step S25: performing pixel-level comparison on the multi-region image comparison data according to the standard display characteristic map data, and calculating the deviation value of each pixel point, thereby obtaining a pixel deviation map; Step S26: Identify potential defect areas based on the pixel deviation map to obtain preliminary defect detection data.

5. The method for detecting defects of a liquid crystal display based on machine vision according to claim 4, characterized in that: Step S26 includes the following steps: Step S261: marking the areas in the pixel deviation map where the deviation exceeds a preset threshold as potential defect areas by using an adaptive threshold segmentation algorithm, thereby obtaining a preliminary defect area map; Step S262: performing morphological processing on the preliminary defect region map to eliminate noise and connect adjacent defect regions, thereby obtaining a preferred defect region map; Step S263: performing connected domain analysis on the preferred defect region map, and marking and numbering each independent defect region, thereby obtaining a defect region marking map; Step S264: extracting features of each marked defect area according to the defect area marking map and the multi-area image data, thereby obtaining preliminary defect detection data.

6. The method for detecting defects of a liquid crystal display based on machine vision according to claim 5, characterized in that: Step S3 includes the following steps: Step S31: setting a plurality of different LCD screen refresh frequencies, and acquiring a plurality of frames of images at each refresh frequency, thereby obtaining a multi-refresh frequency image sequence, wherein the refresh frequencies include 30 Hz, 60 Hz, 120 Hz and 240 Hz; Step S32: performing the preprocessing of step S2 and defect area recognition on the multi-refresh rate image sequence to obtain corresponding preliminary defect detection data, thereby forming a multi-refresh rate defect detection data set; Step S33: performing time series analysis on each defect area in the multi-refresh frequency defect detection data set, and tracking the change of defects at different refresh frequencies and time points, so as to obtain defect change trajectory data; Step S34: using image registration technology to align the multi-refresh frequency defect detection data set to the same coordinate system, and eliminate the small displacement caused by screen refresh, so as to obtain a registered multi-frequency defect map; Step S35: superimposing multiple frames of the multi-frequency defect image, and fusing the multi-frame information using a median filtering method, thereby obtaining an enhanced defect image; Step S36: performing variance analysis on the enhanced defect image in the time dimension, and identifying regions with inconsistent performance at different refresh frequencies, thereby obtaining dynamic defect candidate regions; Step S37: performing time characteristic defect analysis based on dynamic time warping on the dynamic defect candidate area, thereby obtaining dynamic defect detection data; Step S38: performing stability evaluation of defect points at different refresh frequencies and time points based on the dynamic defect detection data and defect change trajectory data, thereby obtaining defect persistence score data; Step S39: Classify defect points into temporary defects and permanent defects according to the defect persistence score data, thereby obtaining defect point classification data, wherein defects with persistence scores higher than a preset threshold are classified as permanent defects, and defects with persistence scores lower than or equal to the preset threshold are classified as temporary defects.

7. The method for detecting defects of a liquid crystal display based on machine vision according to claim 6, characterized in that: Step S4 includes the following steps: Step S41: extracting the location information and surrounding area of ​​each defect point according to the defect point classification data, thereby obtaining defect area data; Step S42: configuring a multi-light source imaging system, including a visible light source, a near-infrared light source, and an ultraviolet light source, thereby obtaining a multi-light source illumination configuration; Step S43: performing multi-light source image acquisition on the defect area data according to the multi-light source illumination configuration, and performing image alignment, thereby obtaining multi-spectral image data; Step S44: performing spectral decomposition on the multispectral image data, extracting the reflectivity or emissivity information of each pixel point in different bands, thereby obtaining spectral characteristic data of the defect area; Step S45: performing statistical analysis on the average spectral response, standard deviation and spectral curve shape characteristics of the defect area spectral characteristic data in different bands, and constructing the spectral fingerprint of each defect area, thereby obtaining the spectral fingerprint data of the defect area, wherein the spectral fingerprint includes the reflection peak, absorption valley and spectral ratio of the characteristic band; Step S46: performing multi-spectral comparative analysis on the defect area spectral fingerprint data to obtain multi-spectral comparative result data, wherein the multi-spectral comparative analysis includes calculating the correlation coefficient between each defect area in the visible light, near infrared and ultraviolet spectra, quantifying the similarity between different bands, and extracting the main change components in the defect area spectral fingerprint data using the principal component analysis method; Step S47: extracting the differences between different spectra based on the multi-spectrum comparison result data, thereby obtaining spectrum difference data.

8. The method for detecting defects of a liquid crystal display based on machine vision according to claim 7, characterized in that: Step S5 includes the following steps: Step S51: constructing a multidimensional feature vector according to the spectral difference data and its corresponding preliminary defect detection data, thereby obtaining a defect spectral feature vector; Step S52: classifying the defect spectrum feature vector based on the support vector machine to obtain a defect authenticity prediction result; Step S53: According to the defect authenticity prediction result, the defect points predicted as false positive are marked as potential false positive areas, and the areas that are not detected but have abnormal spectral characteristics are marked as potential false negative areas, so as to obtain potential misjudgment area data; Step S54: resampling the potential misjudgment area data with high resolution and performing local contrast enhancement to obtain an enhanced misjudgment area image; Step S55: performing a refined analysis on the enhanced misjudged area image to obtain misjudged area confidence data, and screening potential false positive areas according to the misjudged area confidence data and a preset confidence threshold, removing defect points with confidence values ​​lower than the threshold, thereby obtaining optimized defect point data; Step S56: performing morphological processing and connected domain analysis on the potential false negative areas in the enhanced misjudged area image, and identifying small or low-contrast defects that may be ignored, thereby obtaining supplementary defect point data; Step S57: Merge the optimized defect point data and the supplemented defect point data, perform spatial cluster analysis, merge adjacent defect points, and thus obtain accurate defect point data.

9. A liquid crystal screen defect detection system based on machine vision, characterized in that: Used to perform the LCD screen defect detection method based on machine vision as claimed in claim 1, the LCD screen defect detection system based on machine vision comprises: The panel characteristic modeling module is used to obtain the panel structure data of the LCD screen, including panel size data, pixel arrangement data, backlight module data and surface treatment information data; perform optical characteristic analysis of the LCD screen based on light wave interference according to the panel structure data, and map different display states, display contents and normal display states under different viewing angles according to the preset image data set, so as to obtain standard display characteristic diagram data; The initial defect recognition module is used to collect multi-region image data of the LCD panel; perform image preprocessing on the multi-region image data, and perform difference analysis based on the standard display characteristic diagram data to identify potential defect areas, thereby obtaining preliminary defect detection data; The dynamic defect classification module is used to track the defect changes of the preliminary defect detection data at different refresh frequencies through multi-frame image overlay technology, so as to obtain dynamic defect detection data; and classify the defect points based on temporary and permanent defects according to the dynamic defect detection data, so as to obtain defect point classification data; The multi-spectral feature extraction module is used to obtain defect area data according to defect point classification data; obtain the spectral features of the LCD screen image under different light sources for the defect area data, so as to obtain the spectral feature data of the defect area, wherein the different light sources include visible light, near infrared and ultraviolet light; perform multi-spectral comparison on the spectral feature data of the defect area, and extract the differences between different spectra, so as to obtain spectral difference data; The precise defect optimization module is used to identify false positives and false negatives of defect area data based on spectral difference data, and to optimize the precision of defect points, so as to obtain precise defect point data.

10. A computer-readable storage medium storing a computer program, characterized in that: The computer program is executed to implement the liquid crystal screen defect detection method based on machine vision as described in any one of claims 1 to 8.

Images