CN118196099B

CN118196099B - Manufacturing method of fiber mechanical reinforced LED display array

Info

Publication number: CN118196099B
Application number: CN202410616067.1A
Authority: CN
Inventors: 王国宏; 李璟; 郭德博
Original assignee: Nanjing Ajibi Information Technology Co ltd
Current assignee: Nanjing Ajibi Information Technology Co ltd
Priority date: 2024-05-17
Filing date: 2024-05-17
Publication date: 2024-07-23
Anticipated expiration: 2044-05-17
Also published as: CN118196099A

Abstract

The invention relates to the technical field of photoelectric devices, in particular to a manufacturing method of a fiber mechanical reinforced LED display array. Firstly, acquiring an LED image after glass fibers are scattered; analyzing the LED image based on Bayer array characteristics and RGB channels to determine a high-frequency texture region; determining a main direction distribution feature descriptor of each high-frequency texture region and an anisotropic feature descriptor of the LED image based on a Gabor kernel; and inputting the LED image, the main direction distribution feature descriptors and the anisotropic feature descriptors into a trained neural network to obtain a quality label of the LED image. The invention can automatically evaluate the quality of the LED after the glass fiber is scattered, save the labor time and improve the production efficiency.

Description

Manufacturing method of fiber mechanical reinforced LED display array

Technical Field

The invention relates to the technical field of photoelectric devices, in particular to a manufacturing method of a fiber mechanical reinforced LED display array.

Background

The technical progress of LEDs is the greatest driving force to expand market demand and application. LEDs were originally only used as miniature indicator lights in high-end devices such as computers, audio and video recorders, and with the continued advancement of large-scale integrated circuits and computer technology, LED displays are rapidly rising and gradually expanding into the stock market, digital cameras, PDAs and cell phones. The LED display integrates the microelectronic technology, the computer technology and the information processing, has the advantages of bright color, wide dynamic range, high brightness, long service life, stable and reliable work and the like, becomes a new generation of display media with the most advantages, is widely applied to scenes such as large squares, commercial advertisements, stadiums, information propagation, news release, securities trade and the like at present, and can meet the requirements of different environments. The LED display array is the minimum unit of an outdoor LED screen, and in an outdoor use environment, a strong fiber mechanical reinforcement technology is necessary to achieve the surface strength so as to resist the impact of external force and reduce the damage rate.

The glass fibers may cover the LED lamp beads when the glass fibers are scattered, so that scattering and uneven illumination may occur, LED display may be affected, and shielding problems occur.

Disclosure of Invention

In order to solve the technical problem that the LED display is affected due to the fact that glass fibers are scattered randomly to possibly cause the fibers to shade LED lamp beads, scattering and uneven illumination can be generated, the invention aims to provide a manufacturing method of a fiber mechanical reinforced LED display array, which comprises the following steps:

acquiring an LED image after glass fibers are scattered;

Analyzing the LED image based on Bayer array characteristics and RGB channels to determine a high-frequency texture region;

determining a main direction distribution feature descriptor of each high-frequency texture region and an anisotropic feature descriptor of the LED image based on a Gabor kernel;

Inputting the LED image, the main direction distribution feature descriptors and the anisotropic feature descriptors into a trained neural network to obtain a quality label of the LED image;

The training process of the neural network comprises the following steps: partitioning each LED image, and determining an equally partitioned area; based on the characteristics of the equally divided areas, lamp bead shielding distribution characteristics of the LED image are constructed; based on the anisotropic feature descriptors, anomaly detection is carried out on different LED images, and local anomaly factors of the LED images are determined; and determining the comprehensive score of the glass fiber scattering quality of the LED by combining the lamp bead shielding distribution characteristics and the anisotropic characteristic descriptors, and obtaining the quality label of the LED image according to the comprehensive score.

Preferably, the analyzing the LED image based on Bayer array characteristics and RGB channels to determine a high-frequency texture region includes:

Analyzing channel values of a blue channel and a red channel in an LED image, and determining a high-frequency region;

Analyzing a channel value of a green channel in the LED image, and determining a supplementary region;

and forming a high-frequency texture region by the high-frequency region and the supplementary region.

Preferably, the analyzing the channel values of the blue channel and the red channel in the LED image, and determining the high frequency region includes:

for pixel points in the LED image, taking the ratio of the channel value of the red channel to the channel value of the blue channel as a first texture judgment value;

Averaging the first texture judgment value of each pixel point of the LED image to obtain a first judgment average value;

the high frequency region in the LED image is constituted by pixels of the first texture determination value that are larger than the first determination mean value.

Preferably, the analyzing the channel value of the green channel in the LED image, determining the complementary region includes:

Calculating the green channel mean value of all pixel points in the LED image in the green channel;

Taking the difference value of the channel value of each pixel point in the green channel and the mean value of the green channel as a high-frequency pixel value; and reserving the pixel point with the high-frequency pixel value being a positive value as a complementary pixel point, and forming a complementary region by the complementary pixel point.

Preferably, the determining, based on the Gabor kernel, a main direction distribution feature descriptor of each high-frequency texture region includes:

dividing a 180-degree direction angle into a preset number of Gabor kernels;

Calculating the characteristic response of the pixel point in each direction in each high-frequency texture area based on the Gabor kernel; for the characteristic response of each direction, setting all the characteristic responses with negative values to zero, and reserving the characteristic responses with positive values; for each direction, taking the sum of the characteristic responses of the positive values corresponding to the pixel points in the high-frequency texture area as the texture direction in the direction; taking the direction angle of the Gabor kernel corresponding to the maximum texture direction as the main direction angle of the high-frequency texture region;

And obtaining the number of high-frequency texture areas corresponding to different main direction angles, and constructing a main direction distribution feature descriptor.

Preferably, the method for acquiring the anisotropic feature descriptors of the LED image comprises the following steps:

dividing a 180-degree direction angle into a preset number of Gabor kernels;

determining regional characteristic response of each high-frequency texture region under Gabor kernels at different angles based on Gabor kernels at different angles;

Taking the sum value of the regional characteristic response of the high-frequency texture region under Gabor kernels at all angles as the anisotropic characteristic of the high-frequency texture region;

and constructing an anisotropic characteristic descriptor from the anisotropic characteristic of the high-frequency texture region in the LED image.

Preferably, the constructing the bead shielding distribution feature of the LED image based on the feature of the equally divided region includes:

for the equal division areas in the LED image, calculating the maximum value and the median of the equal division areas;

and forming a characteristic binary group by the maximum value and the median of all the equal division areas in the LED image, and marking the characteristic binary group as a lamp bead shielding distribution characteristic.

Preferably, the anomaly detection is performed on different LED images based on the anisotropic feature descriptors, and the determining the local anomaly factor of the LED images includes:

And determining local anomaly factors of each LED image by using an outlier factor detection algorithm.

Preferably, the determining the composite score of the glass fiber distribution quality of the LED by combining the bead shielding distribution feature and the anisotropic feature descriptor includes:

The product of the sum of the elements in the bead shading distribution feature and the LOF value of the anisotropic descriptor is used as a comprehensive score of the glass fiber spreading quality of the LED.

Preferably, the obtaining the quality label of the LED image according to the composite score includes:

and taking the LED image with the comprehensive score higher than the preset scoring threshold value as a sample with qualified scattering quality, marking the quality label as 1, and otherwise marking the quality label as 0.

The embodiment of the invention has at least the following beneficial effects:

the embodiment of the invention firstly acquires an LED image after glass fibers are scattered; analyzing the LED image based on Bayer array characteristics and RGB channels to determine a high-frequency texture region; determining a main direction distribution feature descriptor of each high-frequency texture region and an anisotropic feature descriptor of the LED image based on a Gabor kernel; and inputting the LED image, the main direction distribution feature descriptors and the anisotropic feature descriptors into a trained neural network to obtain a quality label of the LED image. The invention can automatically evaluate the quality of the LED after the glass fiber is scattered, save the labor time and improve the production efficiency.

The invention adopts a mode of dispersing the front panel of the LED display array based on glass fibers. Scattering and uneven illumination may occur because random scattering may cause the fibers to shade the LED beads. The embodiment of the invention provides a method for rapidly optimizing random scattering conditions and identifying scattering, which comprises two stages: a random spreading phase and a manual adjustment phase. In the random scattering stage, quantitative glass fibers are scattered in front of an LED panel through a feeding mechanism, then an LED light source is started, the scattered area is evaluated according to the condition of the light source, only a small amount of fibers are ensured to be attached to the front of the LED lamp beads, obvious shielding cannot be formed, in the identification scattering stage, the scattered position is identified according to the lighting position of the lamp beads, the quality label of the LED panel corresponding to an LED image is determined, the scattering quality of the LED panel is further obtained, the automatic evaluation of the scattering quality is realized, the labor time is saved, the production efficiency is improved, and the situation that a large amount of glass fibers are attached to the front of the LED lamp beads and cause obvious shielding is avoided.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions and advantages of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a method for fabricating a fiber-mechanical-reinforced LED display array according to one embodiment of the present invention;

Fig. 2 is a flowchart illustrating steps of a neural network training process according to an embodiment of the present invention.

Detailed Description

In order to further describe the technical means and effects adopted by the invention to achieve the preset aim, the following is a detailed description of a specific implementation, structure, characteristics and effects of a manufacturing method of a fiber mechanical reinforced LED display array according to the invention with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

The embodiment of the invention provides a specific implementation method of a manufacturing method of a fiber mechanical reinforced LED display array, which is suitable for manufacturing scenes of the LED display array needing glass fiber reinforcement. The embodiment of the invention provides a special LED chip packaging structure. The packaging structure has an asymmetric hexahedral design, forms a trapezoid structure, can directly identify the light emitting surface and the positive and negative poles of the LED chip, avoids the braiding process, and greatly reduces the process and the cost of product manufacturing. The embodiment of the invention adopts a mode of dispersing the front panel of the LED display array based on glass fibers. Scattering and uneven illumination may occur because the randomly dispersed glass fibers may cause the fibers to shade the LED beads. The embodiment of the invention provides a method for rapidly optimizing random scattering conditions and identifying scattering, which comprises two stages: a random spreading phase and an identification scattering phase. In the random spreading stage, a certain amount of glass fibers are spread in front of the LED panel through the feeding mechanism, then the LED light source is started, the spreading area is evaluated according to the condition of the light source, and therefore, only a small amount of fibers are attached in front of the LED lamp beads, and obvious shielding cannot be formed. Finally, in the scattering identification stage, the scattering quality is automatically evaluated according to the scattering position identification according to the lighting position of the lamp beads, and whether the scattering needs to be re-scattered is further evaluated by combining the time adjusted manually, so that the labor time is saved, and the production efficiency is improved.

The following specifically describes a specific scheme of a manufacturing method of a fiber mechanical reinforced LED display array provided by the present invention with reference to the accompanying drawings.

Referring to fig. 1, a flowchart illustrating steps of a method for fabricating a fiber-reinforced LED display array according to an embodiment of the invention is shown, the method includes the steps of:

in step S100, an LED image obtained after glass fibers are scattered is acquired.

The special-shaped LEDs can further leave a space for embedding glass fibers, and a structure of the fiber mechanical reinforced LED display array is constructed first.

The embodiment of the invention provides an LED lattice layout which is used as a basis of a display unit and a physical hardware platform. Each LED can be individually controlled and the overall system includes circuitry to drive and control the point sources of the LEDs. The use of shaped LEDs in embodiments of the present invention, particularly compared to rectangular LEDs, may be considered as being cut out with an additional portion, thus leaving more room for embedding the glass fibers. The LED package has an asymmetric hexahedral design and forms a trapezoid structure, and the design can directly identify the anode and the cathode of the LED chip and the light emitting surface. The structure is specially designed, the distance between the anode and cathode of the LED chip and the light emitting surface and the packaging shell is moderate, and meanwhile, enough space is reserved for the subsequent embedding of glass fibers so as to enhance the strength.

The glass fibers are cut and spread over the LED array. The glass fiber in the embodiment is transparent, has good light transmittance and mechanical strength, is controlled to be about 20 microns on average, has certain elasticity, and is easier to embed in a reserved space of the LED panel.

The glass fibers of the selected length were first cut using a device to cut into pieces of 5cm average length.

Then pouring the cut glass fiber into a device provided with a feeding mechanism, in particular to a hopper which is inclined towards a screen and is positioned right above the luminous surface of the screen, and the luminous surface of the screen is horizontally upwards. The spreading device spreads the glass fibers uniformly in front of the LED panel, and evaluates the spread area according to the condition of the LED light source, but cannot control only a small amount of fibers to adhere in front of the LED beads.

Before curing, lighting tests and evaluations are required. The invention uses data of a plurality of batches as reference data, and specifically, the production process of 1000 LED display arrays is used as test data in the embodiment of the invention due to the initialization of a machine learning model.

First, an LED screen in which glass fibers are scattered is acquired by an industrial camera, recorded as an LED image, and the LED image is lighted in accordance with different patterns, and the image is analyzed based on the characteristics of the Bayer array.

And taking 500 LED images collected historically as a training set. And the LED images of the front and rear sides of the dispersion glass to be processed are collected in real time.

And step 200, analyzing the LED image based on the Bayer array characteristic and the RGB channel to determine a high-frequency texture region.

Because glass fibers are extremely fine, the resolution of an industrial camera cannot accurately distinguish textures of the glass fibers, and in order to acquire a picture of the whole module, the embodiment of the invention utilizes the characteristics of a Bayer Array (Bayer Array) and the result of RGB channels to perform the following processing to obtain a high-frequency region, analyze the main direction of the textures, and then obtain a main direction distribution feature descriptor B and an anisotropic feature descriptor V.

Firstly, channel values of three RGB channels of the obtained LED image and a Bayer array are analyzed, and a high-frequency texture area is determined. Specific:

firstly, the screen is lightened, because the LEDs are designed in three primary colors, and one lamp bead contains the LEDs in three primary colors, the LEDs in the three primary colors are required to be lightened respectively, then the LEDs are lightened simultaneously, 3 single-color pictures and 4 alternate white pictures are obtained respectively, and the 7 pictures can reflect the effect that glass fibers shield the LEDs and scatter the LEDs respectively.

First, the LEDs of the three primary colors are turned on, so that the screen displays three colors of red, green and blue. Three solid-color images were taken separately, noted as LED images.

Since the number of the green microlenses of the bayer array is larger, the perception of green light to high frequency is slightly higher than that of red and blue, and the resolution is slightly higher, but the glass fiber correspondingly causes a certain degree of moire or a part of light to be refracted, so that the color is more obvious on a green image, and is marked as。

Since the red filter and the blue filter are displaced to some extent in the overall arrangement, i.e., displacement of one pixel unit in the diagonal direction, this displacement further highlights the detail of the high frequency region, i.e., the difference in texture of moire portions.

Therefore, the embodiment of the invention carries out the following processing based on the pictures of the blue channel and the red channel so as to find out the texture caused by the glass fiber with extremely high frequency, namely, the channel values of the blue channel and the red channel in the LED image are analyzed to determine a high-frequency region, and the specific is:

Firstly, dividing a blue channel and a red channel for a pixel point in an LED image, namely taking the ratio of the channel value of the red channel to the channel value of the blue channel as a first texture judgment value, wherein the division can obtain a flat texture in a region with consistent amplitude relative difference, namely a low-frequency part, but is very sensitive to the amplitude change of a high-frequency difference part because of a certain brightness difference of a picture of blue response and red response. Forming a difference image by the first texture judgment value of each pixel point of the LED image; averaging the first texture judgment value of each pixel point of the LED image to obtain a first judgment average value; the high frequency region in the LED image is constituted by pixels of the first texture determination value that are larger than the first determination mean value.

And taking the region formed by the pixels of the first texture judgment value larger than the first judgment mean value as a region with larger difference, marking the pixels of the first texture judgment value larger than the first judgment mean value as 1, and otherwise marking the pixels of the first texture judgment value larger than the first judgment mean value as 0. In this way, the high frequency characteristic of the red-blue channel due to the small displacement is obtained.

Further, the channel value of the green channel in the LED image is analyzed to determine the complementary region, specifically:

Calculating the green channel mean value of all pixel points in the LED image in the green channel; taking the difference value of the channel value of each pixel point in the green channel and the mean value of the green channel as a high-frequency pixel value; and reserving the pixel point with the high-frequency pixel value being a positive value as a complementary pixel point, and forming a complementary region by the complementary pixel point.

And step S300, determining a main direction distribution feature descriptor of each high-frequency texture region and an anisotropic feature descriptor of the LED image based on the Gabor kernel.

Further, based on the region map containing the high-frequency texture region, the scattered main direction distribution descriptor B and the anisotropic feature descriptor V are obtained.

First, based on Gabor kernel, a main direction distribution feature descriptor of each of the high frequency texture regions is determined. The main direction is obtained, in the embodiment of the present invention, the 180 degree direction angle is divided into a preset number of Gabor kernels, in the embodiment of the present invention, the preset number of directions is 60, and in other embodiments, the value can be adjusted by an implementer according to the actual situation. By summing up the Gabor characteristic responses, the direction of the largest response is chosen as the main direction, more specifically:

Since the main direction is a scattering pattern representing the natural arrangement of voids based on shaped LEDs, in theory all fibers need to be arranged in this direction for optimal arrangement, however the arrangement is random and the presence of other voids results in a main direction which may be near the angle of the other hypotenuse of a vertical, parallel, shaped LED, thus the main direction represents important information for the pattern of high frequency textures caused by glass fibers which in this case cross in other directions.

For the entire LED, embodiments of the present invention use a block wise approach to count the principal directions and record the grain intensities for different regions. Taking a high-frequency texture region as an example, the texture intensity is calculated as follows:

The characteristic response of the pixel point in each high-frequency texture region in each direction is calculated firstly based on the Gabor kernel, and the characteristic response can be realized by carrying out convolution operation on the input image containing the high-frequency texture region. And setting the positive value reserved in the obtained image to zero, namely setting all the characteristic responses with the negative values to zero for the characteristic response of each direction, and reserving the characteristic responses with the positive values. For the obtained texture, the response value of the LED area is set to zero, so that the texture response caused by the edge of the special-shaped LED is avoided.

Specifically, the shade is obtained by manual marking by an operator according to the acquisition conditions, so that the LED positions of the subsequent LED screens are ensured to be matched with the shade. It should be noted that, since the specific labeling method for the mask is a well-known technology of those skilled in the art, the detailed description will not be repeated here, where the mask refers to the region formed by the pixels after the feature response is zeroed.

Then, the feature responses of the pixels in the non-mask area in the high-frequency texture area are summed, that is, for each direction, the sum of feature responses of positive values corresponding to the pixel points in the high-frequency texture area is taken as the texture direction in the direction, and since the maximum exists when the direction of the Gabor kernel is perpendicular to the texture direction, the true angle is not actually recorded, so that the direction angle of the Gabor kernel corresponding to the maximum texture direction is taken as the main direction angle of the high-frequency texture area in the embodiment of the invention. The direction angle of the Gabor core refers to an angle formed by the direction of the Gabor core and a horizontal line which is horizontal to the right.

And obtaining the number of high-frequency texture areas corresponding to different main direction angles, and constructing a main direction distribution feature descriptor. According to the main direction angles of the respective areas, the main direction distribution feature descriptor B is a numerical value of a statistical histogram of 60 Gabor kernels, specifically, each main direction angle is matched with an angle of a Gabor kernel, and then the main direction distribution feature descriptor B records the number of Gabor kernels corresponding to the respective angles.

For the final statistical histogram, the count of most angles is still 0, for the comparison of the subsequent distribution characteristics, the invention uses a one-dimensional mean filtering mode, and uses a convolution kernel of 1x5 to slide the dimension of the whole statistical angle, so that the smoothed histogram value is obtained, and the number is 60, but 4 angles near the main direction angle and the angle are not 0, and the average effect is realized.

Based on the Gabor kernel, an anisotropic feature descriptor of the LED image is determined. Specific: determining the regional characteristic response of each high-frequency texture region under the Gabor kernels of different angles based on the Gabor kernels of different angles, and taking the sum of the regional characteristic responses of the high-frequency texture regions under the Gabor kernels of different angles as the anisotropic characteristic of the high-frequency texture region; and constructing an anisotropic characteristic descriptor from the anisotropic characteristic of the high-frequency texture region in the LED image. The anisotropic characterization is capable of representing the characteristics of the various angular textures, and the distribution of the numerical values of the summation result of a high frequency texture region represents the degree of significance of the distribution of glass fibers in various directions.

And step S400, inputting the LED image, the main direction distribution feature descriptors and the anisotropic feature descriptors into a trained neural network to obtain a quality label of the LED image.

According to the embodiment of the invention, VGG16 is used as a trunk of a CNN network, two input layers are added at the rear end of the VGG16 to form a fully-connected neural network with bypass input, a main direction distribution feature descriptor B, an anisotropic descriptor V and an LED image are used as inputs, wherein the LED image is processed into a 256-dimensional high-dimensional vector through the VGG16, and is spliced with the distribution feature descriptor B and the anisotropic descriptor V in dimensions to obtain a high-dimensional input, and classification results are output through an implicit layer and an output layer of 3 layers of 256 layers, wherein the classified output layers are two neurons. And taking the historical LED image, the corresponding main direction distribution feature descriptors, the corresponding anisotropic feature descriptors and the LED quality labels which are determined through analysis and calculation as a training set of the neural network.

Specifically, an Adam optimizer is used in the training process, the learning rate is 0.002, dropout parameters are not used, and the input image of CNN is mirrored randomly to amplify the sample.

More specifically, referring to fig. 2, fig. 2 is a flowchart illustrating steps of a training process of a neural network, specifically:

Step S301, partitioning each LED image, and determining equally partitioned areas; and constructing the bead shielding distribution characteristics of the LED image based on the characteristics of the equally divided areas.

The main directional distribution characteristics and the anisotropic characteristics of the glass fibers of the display have been described above, since the number of glass fibers scattered each time is of the same quality, there must be cases where the special-shaped space between the LEDs falls well and cases where the special-shaped space between the LEDs is worse, depending on the distribution characteristics, and the distribution between the high-frequency texture regions represents the overall quality of the LEDs.

To reveal the actual quality of the LEDs and identify which beads need to be manually processed, the bead occlusion distribution characteristics C need to be calculated for the display array, specifically:

For an LED lamp bead, there is a possibility that a glass fiber covers the LED lamp bead to cause a halo with a larger area after the LED lamp bead is lighted, that is, the glass fiber is lighted by the LED and transmitted to other glass fibers, and because the glass fiber not only has the effect of conducting the internal refraction of light, but also can scatter the light, the glass fiber can be lighted by the LED, and for the obvious shielding condition, the peripheral brightness of the LED is larger, and the crosstalk between pixels is caused. If the LED is severely blocked, then the LED needs to manually dispose of the blocked fiber.

This step takes a longer time to test, but does not require manual handling of the glass fibers. It has been mentioned above that this detection process is long, and a problem to be solved by the embodiments of the present invention is to be able to judge the dispersion quality of the glass fibers of the LED in advance before the long-time detection.

In order to accurately evaluate the shielding condition of the LEDs, the embodiment of the invention processes all the LEDs in the image into equal-sized subareas in the area of the picture, namely, the image of each LED is partitioned, equal-partitioned areas are determined, and the equal-partitioned areas are determined by a manual labeling mode, so that each LED can acquire the picture.

And constructing the bead shielding distribution characteristics of the LED image based on the characteristics of the equally divided areas.

And calculating the maximum value and the median of the equal-divided areas in the LED image, and then forming a characteristic binary group of the current LED by the maximum value and the median of all the equal-divided areas in the LED image, and marking the characteristic binary group as a lamp bead shielding distribution characteristic. The reason for this is that if there is more glass fiber shielding around one LED, the maximum value can represent the reference brightness of the LED, and the median represents the typical brightness of the halo, so that the corresponding bead shielding distribution feature C exists in the LED image corresponding to each LED. For example, for one LED image, if there are equal division areas a1, a2, and a3 in the LED image, the maximum value and the median corresponding to the equal division area a1 are b1 and c1, the maximum value and the median corresponding to the equal division area a2 are b2 and c2, and the maximum value and the median corresponding to the equal division area a3 are b3 and c3, respectively, the corresponding bead shielding distribution features are { (b 1, c 1), (b 2, c 2), (b 3, c 3) }.

Step S302, based on the anisotropic feature descriptors, anomaly detection is carried out on different LED images, and local anomaly factors of the LED images are determined.

According to the invention, the quality of the LED screen is further calculated by using the lamp bead shielding distribution characteristics and combining with LOF values of the anisotropic characteristic descriptors.

Therefore, based on the anisotropic feature descriptors, anomaly detection is performed on different LED images, and local anomaly factors of the LED images are determined, specifically: and determining local anomaly factors of each LED image by using an outlier factor detection algorithm.

Firstly, for a plurality of LED display arrays for testing, each LED display array corresponds to one historical LED image, in the embodiment of the invention, 1000 LED display arrays are used, namely, 1000 historical LED images are arranged in a training set of a training neural network, local anomaly factors of each anisotropic descriptor V, namely LOF values of the anisotropic descriptors V, and a distance function of the LOF values uses normalized cosine similarity.

Step S303, determining a comprehensive score of the glass fiber scattering quality of the LED by combining the lamp bead shielding distribution characteristics and the anisotropic characteristic descriptors, and obtaining a quality label of the LED image according to the comprehensive score.

The sum of the elements within the bead occlusion distribution profile is further calculated, multiplied by the LOF value of the anisotropic descriptor V as a composite score for the glass fiber distribution quality of the LED.

And presetting a scoring threshold by an implementer, taking the LED image with the comprehensive score higher than the preset scoring threshold as a sample with qualified scattering quality, marking the quality label as 1, otherwise marking the quality label as 0. That is, a screen above the preset scoring threshold is marked as a sample with qualified spreading quality, which means that the number of manually adjusted LEDs can be reduced, and otherwise marked as 0, which is considered as a situation that the glass fibers need to be re-spread, and this situation consumes more manpower to adjust the arrangement of the glass fibers.

In summary, the present invention relates to the technical field of optoelectronic devices. Firstly, acquiring an LED image after glass fibers are scattered; analyzing the LED image based on Bayer array characteristics and RGB channels to determine a high-frequency texture region; determining a main direction distribution feature descriptor of each high-frequency texture region and an anisotropic feature descriptor of the LED image based on a Gabor kernel; and inputting the LED image, the main direction distribution feature descriptors and the anisotropic feature descriptors into a trained neural network to obtain a quality label of the LED image. The invention can automatically evaluate the quality of the LED after the glass fiber is scattered, save the labor time and improve the production efficiency.

It should be noted that: the sequence of the embodiments of the present invention is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.

Claims

1. A method of fabricating a fiber-mechanically reinforced LED display array, the method comprising the steps of:

acquiring an LED image after glass fibers are scattered;

2. The method of manufacturing a fiber-reinforced LED display array of claim 1, wherein analyzing the LED image based on Bayer array characteristics and RGB channels to determine a high-frequency texture region comprises:

3. The method of fabricating a fiber-reinforced LED display array of claim 2, wherein analyzing the channel values of the blue and red channels in the LED image to determine the high frequency region comprises:

4. A method of fabricating a fiber-mechanically reinforced LED display array according to claim 2, wherein analyzing the channel value of the green channel in the LED image to determine the supplemental area comprises:

5. The method of fabricating a fiber-mechanical-reinforced LED display array of claim 1, wherein the determining the main direction distribution feature descriptors for each of the high-frequency texture regions based on Gabor kernels comprises:

dividing a 180-degree direction angle into a preset number of Gabor kernels;

6. The method for manufacturing the fiber mechanical reinforced LED display array according to claim 1, wherein the method for obtaining the anisotropic characteristic descriptor of the LED image is as follows:

dividing a 180-degree direction angle into a preset number of Gabor kernels;

7. The method for manufacturing the fiber-reinforced LED display array according to claim 1, wherein the constructing the bead shielding distribution feature of the LED image based on the feature of the equally divided region comprises:

8. The method for manufacturing the fiber-reinforced LED display array according to claim 1, wherein the anomaly detection is performed on different LED images based on the anisotropic feature descriptors, and determining local anomaly factors of the LED images comprises:

9. The method of claim 1, wherein the step of determining a composite score for the glass fiber distribution quality of the LED by combining the bead occlusion distribution feature and the anisotropic feature descriptor comprises:

10. The method for manufacturing a fiber-reinforced LED display array of claim 1, wherein said obtaining a quality label of an LED image from said composite score comprises: