A method and system for optimizing the distribution of light guide plate dots based on deep learning

Abstract

The present invention belongs to the technical field of light guide plate dot design. The present invention provides a light guide plate dot distribution optimization design method and system based on deep learning. The type analysis of multiple dots on the light guide plate is performed through a convolutional neural network to identify the main light guide dots and the auxiliary light guide dots. The light guide plate is divided into regions according to the coordinate positions of the main light guide dots and the auxiliary light guide dots on the light guide plate using a convolutional neural network to obtain divided sub-regions. The light guide intensity is matched and analyzed in combination with the actual light guide sub-regions of the light guide plate to screen out non-matching sub-regions and matching sub-regions, thereby reflecting the degree of matching of the light guide intensity of different regions on the light guide plate under specific conditions. This helps to quickly locate the areas with poor light guide performance of the light guide plate used in specific environments, and provide direction for subsequent optimization or adjustment.

Description

Technical Field

[0001] The present invention belongs to the technical field of light guide plate dot design, and specifically provides a light guide plate dot distribution optimization design method and system based on deep learning. Background Art

[0002] In today's display technology and lighting fields, light guide plates (LGPs), as key components, have a direct impact on the product's display effects and lighting quality due to their light-guiding performance. Accurately identifying and optimizing dot distribution to improve light-guiding performance has always been a key focus of the industry during the design and manufacturing of LGPs. Currently, the identification of dot types on LGPs relies mainly on traditional algorithms or manual experience, making it difficult to accurately distinguish between primary and auxiliary light guide dots. This makes it impossible to provide sufficient targeted data for subsequent dot distribution design optimization, resulting in a lack of scientific basis for optimization and low efficiency.

[0003] The existing technology lacks effective means for analyzing the matching degree of light guide intensity in different areas of a light guide plate. It is impossible to rationally divide the light guide plate into different areas based on the coordinate positions of the main and auxiliary light guide points, and to accurately analyze the light guide intensity matching. This makes it difficult to quickly locate areas with poor light guide performance, resulting in a lack of clear direction for subsequent optimization or adjustment.

[0004] In addition, the light-guiding performance-related models constructed by existing technologies lack accuracy. When constructing the models, the complex relationship between the dot spacing and the light-guiding intensity is not fully considered, resulting in that when the model is used to optimize the light-guiding performance of non-matching sub-areas, the optimization accuracy is difficult to guarantee, and the problem of insufficient light-guiding performance of individual sub-areas on the light guide plate under specific conditions cannot be effectively solved. It is difficult to reduce the light-guiding intensity and angle differences, which in turn affects the light-guiding uniformity and overall quality.

[0005] To this end, the present invention provides a method and system for optimizing the design of light guide plate dot distribution based on deep learning. Summary of the Invention

[0006] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0007] The technical solution adopted by the present invention to solve its technical problem is:

[0008] In a first aspect, a method for optimizing the dot distribution of a light guide plate based on deep learning is provided, comprising:

[0009] The convolutional neural network is used to analyze the types of dots on the light guide plate and identify the main light guide dots and auxiliary light guide dots.

[0010] Based on the coordinate positions of the main light guide points and auxiliary light guide points on the light guide plate, the light guide plate is divided into sub-areas using a convolutional neural network. Combined with the required light guide intensity of the light guide sub-areas on the actual light guide plate, a light guide intensity matching analysis is performed to screen out non-matching sub-areas and matching sub-areas;

[0011] The dot spacing between the main light-guiding dots and the auxiliary light-guiding dots in the anastomotic sub-region is obtained respectively. Combined with the light-guiding intensity in the anastomotic sub-region, a deep learning method based on convolutional neural network is used to construct a spacing-light-guiding intensity model.

[0012] According to the constructed spacing-light guiding intensity model, the non-matching spacing optimization value is obtained, and the optimization strategy is formulated according to the non-matching spacing optimization value.

[0013] As a further solution of the present invention, the type of the dots on the light guide plate is analyzed by a convolutional neural network, and the process is as follows:

[0014] Randomly select a dot on the light guide plate as the target analysis dot, obtain the scattering intensity of the target analysis dot under different incident angles, and sort the scattering intensities from large to small to obtain a scattering intensity convolution set;

[0015] In the scattering intensity convolution set, the scattering intensities in adjacent convolution layers are combined to obtain the convolution intensity combination;

[0016] The convolution-emphasized combination is input into the Euclidean distance formula, and the target intensity value is output.

[0017] As a further solution of the present invention, the identification process of the main light guide network point and the auxiliary light guide network point is as follows:

[0018] If the target intensity value is greater than the target intensity threshold, it is recorded as a light guide auxiliary network point;

[0019] If the target intensity is less than or equal to the target intensity threshold, it is recorded as the main light guide point.

[0020] As a further solution of the present invention, a convolutional neural network is used to divide the light guide plate into regions according to the coordinate positions of the main light guide points and the auxiliary light guide points on the light guide plate. The process is as follows:

[0021] Use convolutional neural network to perform spatial transformation on the light guide area of ​​the light guide plate to obtain a two-dimensional spatial model;

[0022] In the two-dimensional space model, coordinate points corresponding to the main light-guiding points and the auxiliary light-guiding points are respectively obtained, and the two-dimensional space model is divided to obtain divided sub-areas.

[0023] As a further solution of the present invention, a matching analysis of the light guide intensity is performed in combination with the actual light guide sub-region of the light guide plate to obtain the light guide angle matching ratio. The process is as follows:

[0024] The scattering intensity of the main light guide points in the divided sub-areas at different incident angles is averaged and the main light guide intensity is output;

[0025] The maximum scattering intensity and minimum scattering intensity of the light-guiding auxiliary points in the divided sub-areas at different incident angles are extracted respectively, and averaged, and the light-guiding auxiliary intensity is output;

[0026] The main light guiding intensity and the auxiliary light guiding intensity are summed and calculated to obtain the light guiding intensity value, which is then subtracted from the light guiding intensity corresponding to the actual light guiding sub-area and the absolute value is taken to obtain the light guiding intensity matching value. The proportion of the light guiding intensity corresponding to the actual light guiding sub-area is then obtained to obtain the light intensity matching ratio.

[0027] As a further solution of the present invention, the screening process of the non-anastomotic subregions and the anastomotic subregions is as follows:

[0028] Extract the scattering angle of the main light-guiding point in the divided sub-area, and subtract it from the light-guiding angle corresponding to the actual light-guiding sub-area, take the absolute value, and obtain the light-guiding angle matching value. Then obtain the proportion of the light-guiding angle corresponding to the actual light-guiding sub-area occupied to obtain the light-guiding angle matching ratio;

[0029] The light guide intensity matching ratio and the light guide angle matching ratio are calculated by geometric product method, and the matching evaluation value is output;

[0030] If the matching evaluation value is greater than the matching evaluation threshold, it is recorded as a non-matching sub-region;

[0031] If the matching evaluation value is less than or equal to the matching evaluation threshold, it is recorded as a matching sub-region.

[0032] As a further solution of the present invention, deep learning is performed through a convolutional neural network to obtain a suspected correlation coefficient. The process is as follows:

[0033] In the two-dimensional space model, the coordinates of the main light guide points and the auxiliary light guide points in each matching sub-area are extracted and input into the coordinate point distance formula to obtain the dot spacing;

[0034] The matching evaluation value corresponding to each matching sub-region is extracted, combined with the corresponding dot spacing, and incorporated into the convolution layer within the convolutional neural network to obtain multiple spacing matching analysis layers. The multiple spacing matching analysis layers are then sorted according to the size of the matching evaluation value to obtain a spacing matching sequence;

[0035] The spacing matching sequence is divided into a deep training set and a deep validation set, and adjacent spacing matching analysis layers in the deep training set are combined to obtain multiple adjacent analysis groups;

[0036] In the adjacent analysis group, the matching evaluation values ​​corresponding to the adjacent spacing matching analysis layer and the dot spacing are respectively subjected to difference processing to obtain the unit matching difference and the unit spacing difference;

[0037] The ratio of the unit matching difference and the unit spacing difference corresponding to each adjacent analysis group was calculated and averaged to obtain the suspected correlation coefficient.

[0038] As a further solution of the present invention, the construction process of the spacing-light guide intensity model is as follows:

[0039] Based on the suspected correlation coefficient, the product calculation is performed with the dot spacing corresponding to the spacing matching analysis layer in the depth verification set to output the suspected matching evaluation value;

[0040] Combine all suspected matching evaluation values ​​with the matching evaluation values ​​in the interval matching analysis layer in the corresponding depth verification set to obtain multiple model verification groups, which are input into the Euclidean calculation formula and output as verification difference values;

[0041] If the verification difference value is less than or equal to the verification difference threshold, it is recorded as the model correlation coefficient. Based on the model correlation coefficient, the X-axis is the dot spacing, and the Y-axis is the matching evaluation value corresponding to the matching sub-region, a spacing-light guide intensity model is constructed, and the corresponding model equation is: ,in, Expressed as correlation coefficient.

[0042] As a further solution of the present invention, an optimization value of the non-matching spacing is obtained, and an optimization strategy is formulated according to the optimization value of the non-matching spacing. The process is as follows:

[0043] Extract the model equation corresponding to the spacing-light guide intensity model ;

[0044] Subtract all matching evaluation values ​​from the matching evaluation threshold, take the absolute value, and output the matching evaluation difference;

[0045] Input into the model equation corresponding to the spacing-light guiding intensity model, and output the non-matching spacing optimization value;

[0046] The non-matching spacing optimization values ​​corresponding to all non-matching sub-regions are compared, and the non-matching sub-regions are optimized in descending order until all non-matching sub-regions on the light guide plate are optimized.

[0047] In the second aspect, a light guide plate dot distribution optimization design system based on deep learning includes:

[0048] Dot type recognition module: Uses convolutional neural networks to analyze the types of dots on the light guide plate and identify the main light guide dots and auxiliary light guide dots;

[0049] Matching analysis and evaluation module: Based on the coordinate positions of the main and auxiliary light guide points on the light guide plate, the convolutional neural network is used to divide the light guide plate into sub-areas. Combined with the light guide intensity required by the light guide sub-areas on the actual light guide plate, the light guide intensity matching analysis is performed to screen out non-matching sub-areas and matching sub-areas;

[0050] Deep learning construction module: The distance between the main light-guiding dots and the auxiliary light-guiding dots in the anastomotic sub-region is obtained respectively. Combined with the light-guiding intensity in the anastomotic sub-region, deep learning is performed through a convolutional neural network to construct a distance-light-guiding intensity model.

[0051] Model optimization design module: According to the constructed spacing-light guide intensity model, the non-matching spacing optimization value is obtained, and the optimization strategy is formulated based on the non-matching spacing optimization value.

[0052] The beneficial effects of the present invention are as follows:

[0053] 1. The present invention uses a convolutional neural network to perform type analysis on multiple dots on a light guide plate, identifies main light guide dots and auxiliary light guide dots, and thus identifies the type of dots on the light guide plate according to the target intensity value, providing targeted design optimization data for subsequent distribution design optimization of dots on the light guide plate. In addition, the present invention uses a convolutional neural network to divide the light guide plate into regions according to the coordinate positions of the main light guide dots and auxiliary light guide dots on the light guide plate, obtaining sub-regions, and performing matching analysis on the light guide intensity in combination with the actual light guide sub-regions of the light guide plate, screening out non-matching sub-regions and matching sub-regions, thereby reflecting the degree of matching of the light guide intensity of different regions on the light guide plate under specific conditions, helping to quickly locate regions with poor light guide performance of the light guide plate used in specific conditions, and providing direction for subsequent optimization or adjustment.

[0054] 2. The present invention obtains the dot spacing between the main light-guiding dots and the auxiliary light-guiding dots in the matching sub-area, combines the light-guiding intensity in the matching sub-area, performs deep learning through a convolutional neural network, and constructs a spacing-light-guiding intensity model, thereby not only improving the accuracy of the constructed model, but also improving the accuracy when optimizing the light-guiding performance of the non-matching sub-area according to the model. According to the constructed spacing-light-guiding intensity model, the non-matching spacing optimization value is obtained, and according to the non-matching spacing optimization value, an optimization strategy is formulated, which effectively solves the problem of insufficient light-guiding performance of individual sub-areas on the light guide plate under specific conditions, improves the light-guiding performance of the light guide plate, reduces the light-guiding intensity and angle differences, and improves the light-guiding uniformity and overall quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention will be further described below with reference to the accompanying drawings.

[0056] Figure 1 This is a flowchart of the steps of a method for optimizing the distribution of light guide plate dots based on deep learning in the present invention;

[0057] Figure 2 This is a schematic diagram of a light guide plate dot distribution optimization design system based on deep learning in the present invention. DETAILED DESCRIPTION

[0058] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0059] Example 1

[0060] See also Figure 1 As shown, a method for optimizing the dot distribution of a light guide plate based on deep learning according to an embodiment of the present invention includes the following steps:

[0061] Step 1: Use a convolutional neural network to analyze the types of multiple dots on the light guide plate and identify the main light guide dots and auxiliary light guide dots;

[0062] In some embodiments, a dot on the light guide plate is arbitrarily selected as a target analysis dot;

[0063] Perform scattering intensity analysis on the target analysis point to obtain the target intensity value. The process is as follows:

[0064] Use a simulated light source to emit light at different incident angles to the target analysis point, and record the scattering intensity corresponding to each emission. Then, according to the order of each simulated light source emission, the recorded scattering intensity is input into the convolution layer to obtain the scattering intensity convolution set;

[0065] Among them, each convolutional layer includes only one scattering intensity;

[0066] In the scattering intensity convolution set, the scattering intensities in adjacent convolution layers are combined to obtain the convolution intensity combination;

[0067] The convolution-emphasized combination is input into the Euclidean distance formula, and the output is the target intensity value;

[0068] Specifically, the Euclidean distance formula is: , calculate the target intensity value , q represents the total number of convolution intensity combinations, Represented as the qth convolution intensity combination sorted The scattering intensity within the convolutional layer, Represented as the qth convolution intensity combination sorted Scattering intensity within the convolutional layer;

[0069] It should be noted that the purpose of using the Euclidean distance formula is that the Euclidean distance formula is essentially to quantify the distance between two points in the spatial dimension. Now, the scattering intensity of the light guide plate at two different incident angles is regarded as two points in the spatial dimension. The Euclidean distance formula can be used to quantify the difference between the scattering intensity of the light guide plate at different incident angles.

[0070] Furthermore, since the primary LGP dots are designed to minimize the differences in scattering intensity at different incident angles, and the auxiliary LGP dots are designed to fill the dark areas of the primary LGP dots, the Euclidean distance formula is used to analyze the differences in scattering intensity at different incident angles. This improves the accuracy of LGP dot type identification and provides targeted design optimization data for subsequent optimization of the LGP dot distribution design.

[0071] It is understood that the target emphasis value represents the value calculated by the Euclidean distance formula. Its core meaning is to quantify the degree of difference in the scattering intensity of the target analysis point at different incident angles. Specifically, a larger target intensity value indicates that the scattering intensity of the point at different incident angles varies greatly. A smaller target intensity value indicates that the scattering intensity of the point at different incident angles varies less. The target intensity value can be used to identify the type of point on the light guide plate, providing targeted design optimization data for subsequent optimization of the distribution design of the point on the light guide plate.

[0072] If the target intensity value is greater than the target intensity threshold, it means that the light received by the analyzed light guide plate network point at different incident angles has a large deviation in the scattered intensity after scattering, and it is recorded as a light guide auxiliary network point;

[0073] If the target intensity is less than or equal to the target intensity threshold, it means that the light received by the analyzed light guide plate mesh point at different incident angles has a small deviation in the scattered intensity after scattering, and it is recorded as the main light guide mesh point;

[0074] Step 2: Based on the coordinate positions of the main light guide points and the auxiliary light guide points on the light guide plate, the light guide plate is divided into sub-areas using a convolutional neural network. Combined with the light guide intensity required by the light guide sub-areas on the actual light guide plate, a light guide intensity matching analysis is performed to screen out non-matching sub-areas and matching sub-areas;

[0075] In some embodiments, a convolutional neural network is used to perform spatial transformation on the light guide area of ​​the light guide plate to obtain a two-dimensional spatial model;

[0076] In the two-dimensional space model, coordinate points corresponding to the main light guide grid points and the auxiliary light guide grid points are obtained respectively, and the two-dimensional space model is divided to obtain divided sub-areas;

[0077] It should be noted that there is only one main light guide point and one auxiliary light guide point in the divided sub-area;

[0078] Get the light guide intensity value corresponding to the divided sub-area and compare it with the actual light guide intensity value. The process is as follows:

[0079] The scattering intensity of the main light guide points in the divided sub-areas at different incident angles is averaged and the main light guide intensity is output;

[0080] The maximum scattering intensity and minimum scattering intensity of the light-guiding auxiliary points in the divided sub-areas at different incident angles are extracted respectively, and averaged, and the light-guiding auxiliary intensity is output;

[0081] The sum of the main light guide intensity and the auxiliary light guide intensity is calculated and output as the light guide intensity value, which is then subtracted from the light guide intensity corresponding to the actual light guide sub-area and the absolute value is taken to obtain the light guide intensity matching value;

[0082] Calculate the ratio of the light guide intensity matching value to the light guide intensity corresponding to the actual light guide sub-area, and output the light guide intensity matching ratio;

[0083] Extract the scattering angle of the main light-guiding point in the divided sub-area, and subtract it from the light-guiding angle corresponding to the actual light-guiding sub-area, take the absolute value, and obtain the light-guiding angle matching value;

[0084] Calculate the ratio of the light guide angle matching value to the light guide angle corresponding to the actual light guide sub-region to obtain the light guide angle matching ratio;

[0085] The light guide intensity matching ratio and the light guide angle matching ratio are calculated by geometric product method, and the matching evaluation value is output;

[0086] It can be explained that the meaning of the matching evaluation value is: comprehensively considering the difference between the light guiding intensity of the light guiding main network point in the divided sub-region of the light guide plate and the light guiding intensity of the actual light guiding sub-region (light guiding intensity matching ratio), as well as the difference between the light guiding angle and the light guiding angle of the actual light guiding sub-region. Specifically, if the matching evaluation value is larger, it means that the difference between the divided sub-region and the actual light guiding sub-region in light guiding intensity and light guiding angle is larger, and the matching degree is lower. If the matching evaluation value is smaller, it means that the difference between the divided sub-region and the actual light guiding sub-region in light guiding intensity and light guiding angle is smaller, and the matching degree is higher.

[0087] Furthermore, the geometric product formula multiplies the light guide intensity matching ratio by the light guide angle matching ratio to reflect the matching degree of light guide performance in different areas of the light guide plate. This helps to quickly locate areas with poor light guide performance and provides direction for subsequent optimization or adjustment.

[0088] If the matching evaluation value is greater than the matching evaluation threshold, it means that the light guide intensity matching degree between the divided sub-region and the actual light guide sub-region is low, and it is recorded as a non-matching sub-region;

[0089] If the matching evaluation value is less than or equal to the matching evaluation threshold, it means that the light guide intensity of the divided sub-region and the actual light guide sub-region has a high matching degree, and is recorded as a matching sub-region;

[0090] The specific scheme of the embodiment of the present invention is: using a convolutional neural network to perform type analysis on multiple dots on the light guide plate, identify the main light guide dots and the auxiliary light guide dots, and then identify the type of dots on the light guide plate according to the target intensity value, providing targeted design optimization data for the subsequent distribution design optimization of the dots on the light guide plate, and using a convolutional neural network to divide the light guide plate into regions according to the coordinate positions of the main light guide dots and the auxiliary light guide dots on the light guide plate to obtain divided sub-regions, and combining the actual light guide plate light guide sub-regions to perform matching analysis on the light guide intensity, screen out non-matching sub-regions and matching sub-regions, and then reflect the degree of matching of the light guide intensity of different regions on the light guide plate under specific conditions, which helps to quickly locate the areas with poor light guide performance of the light guide plate used in specific environments, and provide direction for subsequent optimization or adjustment.

[0091] Example 2

[0092] See also Figure 1 As shown, the method for optimizing the dot distribution of a light guide plate based on deep learning according to an embodiment of the present invention further includes the following steps:

[0093] Step 3: Obtain the dot spacing between the main light-guiding dots and the auxiliary light-guiding dots in the anastomotic sub-region respectively, and build a spacing-light-guiding intensity model by performing deep learning through a convolutional neural network based on the light-guiding intensity in the anastomotic sub-region.

[0094] In some embodiments, in the two-dimensional space model, the coordinates of the main light guide grid points and the coordinates of the auxiliary light guide grid points in each matching sub-region are extracted;

[0095] The coordinate distance between the main light guide dot coordinates and the auxiliary light guide dot coordinates is obtained by using the coordinate point distance formula, and the ratio is calculated with the perimeter of the matching sub-area to obtain the dot spacing;

[0096] The matching evaluation value corresponding to each matching sub-region is extracted, combined with the corresponding dot spacing, and incorporated into the convolution layer within the convolutional neural network to obtain multiple spacing matching analysis layers;

[0097] According to the size of the matching evaluation value, multiple interval matching analysis layers are sorted to obtain an interval matching sequence;

[0098] Split the gap matching sequence into a deep training set and a deep validation set;

[0099] It should be noted that the number of spacing matching analysis layers in the deep training set accounts for 70% of the total number of spacing matching analysis layers in the spacing matching sequence, and the spacing matching analysis layers in the deep training set are selected according to the order of spacing matching analysis layers in the spacing matching sequence;

[0100] The number of spacing matching analysis layers in the deep validation set accounts for 30% of the total number of spacing matching analysis layers in the spacing matching sequence, and the spacing matching analysis layers in the deep validation set are also selected according to the order of spacing matching analysis layers in the spacing matching sequence;

[0101] It is further explained that the total number of spacing matching analysis layers in the spacing matching sequence is a whole number of ten. For example, if the total number of spacing matching analysis layers in the spacing matching sequence is 10, then the number of spacing matching analysis layers in the deep training set is the first 7 spacing matching analysis layers, and the number of spacing matching analysis layers in the deep validation set is the last 3;

[0102] Combine adjacent spacing matching analysis layers in the deep training set to obtain multiple adjacent analysis groups;

[0103] In adjacent analysis groups, the matching evaluation values ​​corresponding to adjacent spacing matching analysis layers are subjected to difference processing, and the unit matching difference is output;

[0104] Similarly, the dot spacing corresponding to the adjacent spacing matching analysis layer is subjected to difference processing, and the output is the unit spacing difference;

[0105] Calculate the ratio of the unit matching difference to the unit spacing difference, and output the unit correlation ratio;

[0106] The unit correlation coefficients corresponding to all adjacent analysis groups are averaged and the suspected correlation coefficients are output;

[0107] Based on the suspected correlation coefficient, the product calculation is performed with the dot spacing corresponding to the spacing matching analysis layer in the depth verification set to output the suspected matching evaluation value;

[0108] All suspected matching evaluation values ​​are combined with the matching evaluation values ​​in the interval matching analysis layer in the corresponding depth validation set to obtain multiple model validation groups;

[0109] Multiple model validation groups are input into the Euclidean calculation formula, and the validation difference value is output;

[0110] Specifically, the Euclid calculation formula is: , calculate the verification difference value , Expressed as The suspected matching evaluation value within the model validation group, Expressed as Matching assessment value within the model validation group;

[0111] If the validation difference value is greater than the validation difference threshold, it indicates that the suspected correlation coefficient has limitations and is recorded as a non-model correlation coefficient;

[0112] If the verification difference value is less than or equal to the verification difference threshold, it means that the suspected correlation coefficient has no limitations and is recorded as the model correlation coefficient;

[0113] Based on the model correlation coefficient, the X-axis is the dot spacing, and the Y-axis is the matching evaluation value corresponding to the matching sub-area, a spacing-light guide intensity model is constructed, and the corresponding model equation is: ,in, Expressed as correlation coefficient;

[0114] Step 4: Obtain the non-matching spacing optimization value based on the constructed spacing-light guide intensity model, and formulate an optimization strategy based on the non-matching spacing optimization value;

[0115] In some embodiments, the model equation corresponding to the spacing-light guide intensity model is extracted ;

[0116] Subtract all matching evaluation values ​​from the matching evaluation threshold, take the absolute value, and output the matching evaluation difference;

[0117] Input into the model equation corresponding to the spacing-light guiding intensity model, and output the non-matching spacing optimization value;

[0118] Compare the non-matching spacing optimization values ​​corresponding to all non-matching sub-regions, and optimize the non-matching sub-regions in descending order until all non-matching sub-regions on the light guide plate are optimized;

[0119] The specific scheme of the embodiment of the present invention is: respectively obtain the dot spacing between the main light-guiding dots and the auxiliary light-guiding dots in the matching sub-area, combine the light-guiding intensity in the matching sub-area, perform deep learning through a convolutional neural network, and construct a spacing-light-guiding intensity model, thereby not only improving the accuracy of the constructed model, but also improving the accuracy when optimizing the light-guiding performance of the non-matching sub-area according to the model. According to the constructed spacing-light-guiding intensity model, the non-matching spacing optimization value is obtained, and according to the non-matching spacing optimization value, an optimization strategy is formulated, which effectively solves the problem of insufficient light-guiding performance of individual sub-areas on the light guide plate under specific conditions, improves the light-guiding performance of the light guide plate, reduces the light-guiding intensity and angle differences, and improves the light-guiding uniformity and overall quality.

[0120] Example 3

[0121] Based on the same inventive concept as the light guide plate dot distribution optimization design method based on deep learning in the aforementioned embodiment, Figure 2 As shown, the present application provides a light guide plate dot distribution optimization design system based on deep learning, wherein the system specifically includes:

[0122] Dot type recognition module: Uses convolutional neural networks to analyze the types of dots on the light guide plate and identify the main light guide dots and auxiliary light guide dots;

[0123] Matching analysis and evaluation module: Based on the coordinate positions of the main light guide points and auxiliary light guide points on the light guide plate, the convolutional neural network is used to divide the light guide plate into sub-areas. The light guide intensity is then matched with the actual light guide sub-areas of the light guide plate to screen out non-matching and matching sub-areas.

[0124] Deep learning construction module: The distance between the main light-guiding dots and the auxiliary light-guiding dots in the anastomotic sub-region is obtained respectively. Combined with the light-guiding intensity in the anastomotic sub-region, deep learning is performed through a convolutional neural network to construct a distance-light-guiding intensity model.

[0125] Model optimization design module: According to the constructed spacing-light guide intensity model, the non-matching spacing optimization value is obtained, and the optimization strategy is formulated based on the non-matching spacing optimization value.

[0126] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the dot distribution of light guide plates based on deep learning, characterized by: include: The convolutional neural network is used to analyze the types of dots on the light guide plate and identify the main light guide dots and auxiliary light guide dots. Based on the coordinate positions of the main light guide points and auxiliary light guide points on the light guide plate, the light guide plate is divided into sub-areas using a convolutional neural network. Combined with the required light guide intensity of the light guide sub-areas on the actual light guide plate, a light guide intensity matching analysis is performed to screen out non-matching sub-areas and matching sub-areas; The dot spacing between the main light-guiding dots and the auxiliary light-guiding dots in the anastomotic sub-region is obtained respectively. Combined with the light-guiding intensity in the anastomotic sub-region, a deep learning method based on convolutional neural network is used to construct a spacing-light-guiding intensity model. According to the constructed spacing-light guide intensity model, the non-matching spacing optimization value is obtained, and the optimization strategy is formulated according to the non-matching spacing optimization value; Through deep learning of convolutional neural network, the suspected correlation coefficient is obtained. The process is as follows: In the two-dimensional space model, the coordinates of the main light guide points and the auxiliary light guide points in each matching sub-area are extracted, and the coordinate distance between the main light guide point coordinates and the auxiliary light guide point coordinates is obtained using the coordinate point distance formula. The ratio of the distance to the perimeter of the matching sub-area is then calculated to obtain the dot spacing; The matching evaluation value corresponding to each matching sub-region is extracted, combined with the corresponding dot spacing, and incorporated into the convolution layer within the convolutional neural network to obtain multiple spacing matching analysis layers. The multiple spacing matching analysis layers are then sorted according to the size of the matching evaluation value to obtain a spacing matching sequence; The spacing matching sequence is divided into a deep training set and a deep validation set, and adjacent spacing matching analysis layers in the deep training set are combined to obtain multiple adjacent analysis groups; In the adjacent analysis group, the matching evaluation values ​​corresponding to the adjacent spacing matching analysis layer and the dot spacing are respectively subjected to difference processing to obtain the unit matching difference and the unit spacing difference; After calculating the ratio of the unit matching difference and the unit spacing difference corresponding to each adjacent analysis group, the ratio is averaged to obtain the suspected correlation coefficient; The construction process of the spacing-light guide intensity model is as follows: Based on the suspected correlation coefficient, the product calculation is performed with the dot spacing corresponding to the spacing matching analysis layer in the depth verification set to output the suspected matching evaluation value; Combine all suspected matching evaluation values ​​with the matching evaluation values ​​in the interval matching analysis layer in the corresponding depth verification set to obtain multiple model verification groups, which are input into the Euclidean calculation formula and output as verification difference values; If the verification difference value is less than or equal to the verification difference threshold, it is recorded as the model correlation coefficient. Based on the model correlation coefficient, the X-axis is the dot spacing, and the Y-axis is the matching evaluation value corresponding to the matching sub-region, a spacing-light guide intensity model is constructed, and the corresponding model equation is: ,in, Expressed as the correlation coefficient, Expressed as dot spacing, Expressed as a match evaluation value.

2. The method for optimizing the dot distribution of a light guide plate based on deep learning according to claim 1, characterized in that: The type of dots on the light guide plate is analyzed using a convolutional neural network. The process is as follows: Randomly select a dot on the light guide plate as the target analysis dot, obtain the scattering intensity of the target analysis dot under different incident angles, and sort the scattering intensities from large to small to obtain a scattering intensity convolution set; In the scattering intensity convolution set, the scattering intensities in adjacent convolution layers are combined to obtain the convolution intensity combination; The convolution intensity combination is input into the Euclidean distance formula, and the target intensity value is output.

3. The method for optimizing the dot distribution of a light guide plate based on deep learning according to claim 2, wherein: The identification process of the main light guide points and auxiliary light guide points is as follows: If the target intensity value is greater than the target intensity threshold, it is recorded as a light guide auxiliary network point; If the target intensity is less than or equal to the target intensity threshold, it is recorded as the main light guide point.

4. The method for optimizing the dot distribution of a light guide plate based on deep learning according to claim 1, wherein: According to the coordinate positions of the main light guide points and auxiliary light guide points on the light guide plate, the convolutional neural network is used to divide the light guide plate into regions. The process is as follows: Use convolutional neural network to perform spatial transformation on the light guide area of ​​the light guide plate to obtain a two-dimensional spatial model; In the two-dimensional space model, coordinate points corresponding to the main light-guiding points and the auxiliary light-guiding points are respectively obtained, and the two-dimensional space model is divided to obtain divided sub-areas.

5. The method for optimizing the dot distribution of a light guide plate based on deep learning according to claim 1, wherein: Combined with the actual light guide plate light guide area, the light guide intensity matching analysis is performed to obtain the light guide intensity matching ratio and light guide angle matching ratio. The process is as follows: The scattering intensity of the main light guide points in the divided sub-areas at different incident angles is averaged and the main light guide intensity is output; The maximum scattering intensity and minimum scattering intensity of the light-guiding auxiliary points in the divided sub-areas at different incident angles are extracted respectively, and averaged, and the light-guiding auxiliary intensity is output; The sum of the main light guide intensity and the auxiliary light guide intensity is calculated and output as the light guide intensity value. The sum is then subtracted from the light guide intensity corresponding to the actual light guide sub-area, and the absolute value is taken to obtain the light guide intensity matching value. The proportion of the light guide intensity corresponding to the actual light guide sub-area is obtained to obtain the light guide intensity matching ratio. The scattering angle of the main light-guiding point in the divided sub-area is extracted, and the difference is made with the light-guiding angle corresponding to the actual light-guiding sub-area. The absolute value is taken to obtain the light-guiding angle matching value, and the proportion of the light-guiding angle corresponding to the actual light-guiding sub-area is obtained to obtain the light-guiding angle matching ratio.

6. The method for optimizing the dot distribution of a light guide plate based on deep learning according to claim 5, characterized in that: The screening process of non-anastomotic subregions and anastomotic subregions is as follows: The light guide intensity matching ratio and the light guide angle matching ratio are calculated by geometric product method, and the matching evaluation value is output; If the matching evaluation value is greater than the matching evaluation threshold, it is recorded as a non-matching sub-region; If the matching evaluation value is less than or equal to the matching evaluation threshold, it is recorded as a matching sub-region.

7. The method for optimizing the dot distribution of a light guide plate based on deep learning according to claim 6, characterized in that: Obtain the optimal value of the non-matching spacing and formulate an optimization strategy based on the optimal value of the non-matching spacing. The process is as follows: Extract the model equation corresponding to the spacing-light guide intensity model ; Subtract all matching evaluation values ​​from the matching evaluation threshold, take the absolute value, output the matching evaluation difference, and input it into the model equation corresponding to the spacing-light guide intensity model to output the non-matching spacing optimization value; The non-matching spacing optimization values ​​corresponding to all non-matching sub-regions are compared, and the non-matching sub-regions are optimized in descending order until all non-matching sub-regions on the light guide plate are optimized.

8. A system for optimizing the dot distribution of a light guide plate based on deep learning, the system being configured to implement the method for optimizing the dot distribution of a light guide plate based on deep learning as claimed in any one of claims 1 to 7, characterized in that: include: Dot type recognition module: Uses convolutional neural networks to analyze the types of dots on the light guide plate and identify the main light guide dots and auxiliary light guide dots; Matching analysis and evaluation module: Based on the coordinate positions of the main and auxiliary light guide points on the light guide plate, the convolutional neural network is used to divide the light guide plate into sub-areas. Combined with the light guide intensity required by the light guide sub-areas on the actual light guide plate, the light guide intensity matching analysis is performed to screen out non-matching sub-areas and matching sub-areas; Deep learning construction module: The distance between the main light-guiding dots and the auxiliary light-guiding dots in the anastomotic sub-region is obtained respectively. Combined with the light-guiding intensity in the anastomotic sub-region, deep learning is performed through a convolutional neural network to construct a distance-light-guiding intensity model. Model optimization design module: According to the constructed spacing-light guide intensity model, the non-matching spacing optimization value is obtained, and the optimization strategy is formulated based on the non-matching spacing optimization value.

Images