Flexible touch screen polarization characteristic detection method based on multi-modal data fusion

Abstract

The present application relates to the technical field of flexible touch screen detection, and specifically discloses a flexible touch screen polarization characteristic detection method based on multi-modal data fusion. The method comprises the following steps: building a multi-parameter measurement optical path system, collecting speckle characteristic data, interferometric polarization phase information data and spectral characteristic data after the flexible touch screen is transmitted; generating a multi-modal fusion feature vector through time-space alignment and feature extraction; constructing a CNN-LSTM hybrid neural network model; dynamically adjusting the parameters of the CNN-LSTM hybrid neural network model by using an adaptive algorithm; inputting the multi-modal fusion feature vector into the adjusted CNN-LSTM hybrid neural network model, and outputting the polarization degree, the extinction ratio and the polarization axis direction parameters. Through the multi-parameter measurement optical path system cooperative detection, the multi-modal data fusion and the adaptive algorithm, the present application solves the polarization characteristic drift problem caused by the deformation of the flexible screen, and significantly improves the detection accuracy and stability.

Description

Technical Field

[0001] This invention relates to the field of flexible touchscreen testing technology, and more specifically to a method for detecting the polarization characteristics of flexible touchscreens based on multimodal data fusion. Background Technology

[0002] With the rapid development of flexible display technology, flexible touchscreens, with their foldable and bendable characteristics, are widely used in consumer electronics, automotive displays, and other fields. However, the detection of their polarization characteristics faces many challenges. Compared with traditional rigid screens, flexible touchscreens use flexible substrate materials such as polyimide (PI), which undergo anisotropic deformation during bending, leading to complex changes such as polarization axis shift and extinction ratio fluctuations. For example, when the bending radius decreases from 10mm to 5mm, the polarization axis shift can reach 8-15 degrees. At the same time, the multi-layer structure of flexible screens is prone to polarization crosstalk during deformation and is significantly affected by temperature and humidity. The thermal expansion coefficient of the PI substrate is 5-10 ppm / ℃, and temperature changes can cause optical thickness drift, thus affecting the polarization state.

[0003] Traditional rigid screens, due to their high material rigidity and low environmental sensitivity, can meet requirements without complex testing technologies, thus primarily employing single testing techniques such as photometry and ellipsometer methods. However, single testing methods cannot simultaneously capture the polarization phase changes and light intensity distribution alterations caused by deformation in flexible screens, making it difficult to meet the needs of multi-parameter collaborative testing. Furthermore, traditional fixed-parameter testing methods cannot cope with fluctuations in the mass production process of flexible screens and environmental changes during use, failing to comprehensively and accurately acquire polarization characteristics, and exhibiting insufficient generalization ability of the testing model. In contrast, multimodal data fusion technology, by integrating heterogeneous data from multiple sources, has demonstrated significant advantages in fields such as autonomous driving and biomedicine. Therefore, there is an urgent need for a testing solution specifically targeting the polarization characteristics of flexible touchscreens to improve testing accuracy and stability, and promote the development of the flexible display industry. Summary of the Invention

[0004] In view of this, the present invention aims to provide a method for detecting the polarization characteristics of flexible touch screens based on multimodal data fusion. Through collaborative detection of a multi-parameter measurement optical path system, multimodal data fusion, and adaptive algorithms, the method achieves high-precision and high-stability detection of the polarization characteristics of flexible touch screens.

[0005] To achieve the above objectives, the present invention provides a method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion, comprising the following steps:

[0006] S1: Construct a multi-parameter measurement optical path system for flexible touch screens to acquire multi-modal data after the light source is transmitted through the flexible touch screen, including speckle feature data, interferometric polarization phase information data, and spectral characteristic data.

[0007] S2: Perform spatiotemporal alignment and feature extraction on multimodal data, and generate multimodal fusion feature vectors through feature fusion;

[0008] S3: Based on multimodal fusion feature vectors, a CNN-LSTM hybrid neural network model is constructed, and the CNN-LSTM hybrid neural network model is trained using flexible touch screen sample data; the flexible touch screen sample data includes the actual measured polarization characteristic parameters and the corresponding multimodal fusion feature vectors; the CNN-LSTM hybrid neural network model uses CNN to extract the spatial features of speckle and spectrum, and uses LSTM to process the polarization phase temporal features;

[0009] S4: Real-time monitoring of environmental parameters, including ambient temperature and humidity, establishes a mapping relationship between environmental parameters and the parameters of the trained CNN-LSTM hybrid neural network model through an adaptive algorithm, dynamically adjusts the parameters of the trained CNN-LSTM hybrid neural network model, and obtains the adjusted CNN-LSTM hybrid neural network model.

[0010] S5: Input the multimodal fusion feature vector into the adjusted CNN-LSTM hybrid neural network model, and output the polarization degree, extinction ratio and polarization axis direction parameters of the flexible touch screen.

[0011] As a further improvement of the present invention:

[0012] Optionally, obtaining the speckle feature data in step S1 includes:

[0013] Based on the principle of mode superposition, when light sources are incident on a flexible touchscreen with different polarization states, the intensity distribution of the output light field is determined by the coherent superposition of multiple modes of light.

[0014]

[0015] in, To output the light field intensity distribution, For the first The amplitude of each mode, It is an exponential function. The imaginary unit, Let be the propagation constant. For wavelength, The thickness of the flexible touchscreen. The number of modes supported for flexible touchscreens;

[0016] After the output light field is scattered by the flexible screen, the speckle intensity distribution data is obtained based on the square of the modulus of the complex amplitude distribution of the light field. ,in, These are the pixel coordinates of the image sensor in the horizontal and vertical directions.

[0017] By acquiring speckle intensity distribution data through an image sensor, speckle characteristic data caused by changes in light polarization state after being affected by a flexible touchscreen are obtained.

[0018] Optionally, the acquisition of interferometric polarization phase information data in step S1 includes:

[0019] Using a Mach-Zehnder interferometer or a Sanignac interferometer, the light source is split into two beams. By setting different polarization beam-splitting paths, the two beams carry different polarization information; according to the principle of interference, the intensity of the interfering light... for:

[0020]

[0021] in, , The light intensities of the two beams, The phase difference between the two beams of light. It is a cosine function;

[0022] Based on the Jones matrix theory of polarized light, the contribution of polarization state to phase difference during the modulation of two beams is analyzed, the phase delay information introduced by polarization modulation is decoupled, and interferometric polarization phase information data is obtained.

[0023] Optionally, obtaining the spectral characteristic data in step S1 includes:

[0024] Let the spectral intensity distribution of light be... By analyzing the distribution characteristics of spectral intensity at different wavelengths and combining it with the spectral theory model of polarized light, a model is constructed based on the polarization state ( ) to spectral distribution Based on the mathematical relationships, establish a polarization-spectral correlation model:

[0025]

[0026] in, It is a polarization-sensitive characteristic spectral function. These are the weighting coefficients. Indicates degree of polarization. The angle between the polarization axis and the horizontal polarization direction. For summation functions;

[0027] The spectral characteristic data are obtained by solving the polarization-spectral correlation model parameters through algorithmic inversion.

[0028] Optionally, generating the multimodal fusion feature vector in step S2 includes:

[0029] The feature fusion adopts a vertical splicing method;

[0030] Multimodal fusion feature vector : ,in These are speckle characteristic data, interferometric polarization phase information data, and spectral characteristic data, respectively; among which... for The transpose operation. To perform spatiotemporal alignment and feature extraction of high-dimensional mathematical features with multimodal data from flexible screen polarization detection.

[0031] Optionally, the construction of the CNN-LSTM hybrid neural network model in step S3 includes:

[0032] Using mean squared error as the loss function :

[0033]

[0034] in, For the sample size, and The first The multimodal fusion feature vector and actual polarization characteristic parameters of each sample To convert complex errors into real losses, smooth gradients are provided to ensure stable model convergence;

[0035] Update the parameters of the CNN-LSTM hybrid neural network model using the stochastic gradient descent optimization algorithm. By learning the complex relationship between different data modes and the polarization characteristics of flexible touch screens, a trained CNN-LSTM hybrid neural network model is obtained.

[0036] Optionally, the adaptive algorithm in step S4 for dynamically adjusting the CNN-LSTM hybrid neural network includes: establishing a mapping function between environmental parameters and the trained CNN-LSTM hybrid neural network model parameters to obtain the adjusted model parameters. :

[0037]

[0038] in, The parameters of the CNN-LSTM hybrid neural network model before training were adjusted; For temperature, Humidity; These are mapping parameters.

[0039] Compared with the prior art, the present invention has at least the following beneficial effects:

[0040] This invention constructs a multi-parameter measurement optical path system for flexible touchscreens, comprising a speckle detection system, an interferometric polarization phase information detection system, and a spectral detection system working collaboratively to acquire polarization characteristic information of the flexible touchscreen from multiple dimensions. Combined with multimodal data fusion technology, compared to single detection methods, it achieves comprehensive collaborative detection of polarization characteristics across all dimensions, significantly improving detection accuracy. Furthermore, by employing a hybrid neural network structure, fully leveraging the advantages of CNN and LSTM, it can better process multimodal data, learn the complex relationships between different data modes and polarization characteristics, enhance the model's generalization ability, and meet the detection requirements of different types of flexible touchscreens. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without any creative effort.

[0042] Figure 1 This is a flowchart of the flexible touchscreen polarization characteristic detection method based on multimodal data fusion according to the present invention;

[0043] Figure 2 This is an output speckle pattern after light passes through a flexible touchscreen in one embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of the interferometric polarization phase information after light passes through a flexible touchscreen in one embodiment of the present invention;

[0045] Figure 4 This is a schematic diagram of the spectral information of light after it passes through a flexible touchscreen in one embodiment of the present invention; wherein, Figure 4 (a) is a schematic diagram of the spectral information of horizontally polarized light after transmission; Figure 4 (b) is a schematic diagram of the spectral information of vertically polarized light after transmission; Figure 4 (c) is a schematic diagram of the spectral information of right-handed circularly polarized light after transmission. Figure 4 (d) is a schematic diagram of the spectral information of left-handed circularly polarized light after transmission;

[0046] Figure 5 This is a flowchart of multimodal data fusion for the polarization characteristics of a flexible screen based on a CNN-LSTM hybrid neural network model. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0048] Reference Figure 1 A method for detecting the polarization characteristics of flexible touchscreens based on multimodal data fusion includes the following steps:

[0049] S1: Construct a multi-parameter measurement optical path system for a flexible touchscreen to obtain multimodal data after the light source passes through the flexible touchscreen, including speckle characteristic data, interferometric polarization phase information data, and spectral characteristic data, specifically:

[0050] S11: The constructed flexible touch screen multi-parameter measurement optical path system includes a speckle detection system, which is used to acquire speckle characteristic data after the light source passes through the flexible touch screen.

[0051] Based on the principle of mode superposition, when light of different polarization states... When incident on a flexible touchscreen, due to the propagation constants of each mode... Unlike other methods, it produces a speckle pattern at the output end of the flexible touchscreen, referring to... Figure 2 x-axis Used to describe the angular position of a speckle pattern in the horizontal direction, reflecting the angular distribution of light propagation or scattering in the horizontal dimension. The unit is milliradians, which is the unit of measurement for angle. (Vertical axis) This describes the angular position of the speckle pattern in the vertical direction, reflecting the angular distribution of light in the vertical dimension, also measured in milliradians. The color bar on the right represents the intensity value of the speckle pattern (from 0.1 to 1). The larger the value, the brighter the corresponding color, reflecting the light intensity distribution of the speckle at different angular positions after light passes through the flexible screen.

[0052] Based on the theory of mode propagation of light in a waveguide, the output light field intensity distribution is... Determined by the coherent superposition of multimode light, it can be expressed as:

[0053]

[0054] in, To output the light field intensity distribution, For the first The amplitude of each mode, It is an exponential function. The imaginary unit, For wavelength, The thickness of the flexible touchscreen. Number of modes supported for flexible touchscreens.

[0055] It should be noted that changes in the polarization characteristics of a flexible touchscreen alter the polarization state of the incident light, thereby affecting the amplitude and phase relationships of each mode and consequently impacting the speckle intensity distribution data. Changes have occurred, among which, These are the pixel coordinates of the image sensor in the horizontal and vertical directions.

[0056] After the output light field is scattered by the flexible screen, speckle intensity distribution data is acquired by an image sensor based on the modulus square of the complex amplitude distribution of the light field, thereby obtaining speckle characteristic data caused by the change in light polarization state after being affected by the flexible touch screen. .

[0057] S12: The constructed flexible touch screen multi-parameter measurement optical path system includes an interferometric polarization phase information detection system, which is used to acquire interferometric polarization phase information data after the light source passes through the flexible touch screen.

[0058] Using a Mach-Zehnder interferometer or a Saniac interferometer structure, the incident light is split into two beams. By setting different polarization beam-splitting paths, the two beams carry different polarization information, as referenced... Figure 3 x-axis This represents the pixel index position in the image along the horizontal direction. The values ​​increase from left to right; for example, a horizontal coordinate of 100 represents the 100th pixel in the horizontal direction, and 500 represents the 500th pixel in the horizontal direction. The vertical coordinate... This indicates the pixel index position in the vertical direction of the image. The values ​​increase from bottom to top. For example, a vertical coordinate of 100 represents the position of the 100th pixel in the vertical direction; 500 represents the position of the 500th pixel in the vertical direction.

[0059] According to the principle of interference, the intensity of the interfering light... It can be represented as:

[0060]

[0061] in, , The light intensities of the two beams, The phase difference between the two beams of light. It is a cosine function;

[0062] By using an image sensor to acquire the intensity distribution of the interferometric light, the phase difference between the two beams can be obtained through inversion. Based on the Jones matrix theory of polarized light, the contribution of polarization state to phase difference during the propagation / modulation of two beams can be analyzed. This allows for the decoupling of phase delay information introduced by polarization modulation, ultimately yielding interferometric polarization phase data. This enables the measurement of the polarization state of light transmitted through a flexible touchscreen.

[0063] S13: The constructed flexible touchscreen multi-parameter measurement optical path system includes a spectral detection system, used to acquire spectral characteristic data of the light source after passing through the flexible touchscreen. ;

[0064] The spectral properties of light are analyzed using a spectrometer. The polarization state of light is related to its spectral properties; for a specific polarization state, its spectral distribution exhibits characteristics. (Refer to...) Figure 4 This refers to the spectral information of light after it passes through the flexible touchscreen. Figure 4 (a) shows the spectral information of horizontally polarized light after transmission; Figure 4 (b) shows the spectral information of vertically polarized light after transmission; Figure 4 (c) shows the spectral information of right-handed circularly polarized light after transmission. Figure 4 (d) shows the spectral information of left-handed circularly polarized light after transmission.

[0065] Let the spectral intensity distribution of light be... By analyzing the distribution characteristics of spectral intensity at different wavelengths and combining the spectral theory model of polarized light, the correlation between spectral characteristics and polarization state is established, thereby obtaining information related to polarization characteristics.

[0066] Establish a polarization-spectral correlation model:

[0067]

[0068] in It is a polarization-sensitive characteristic spectral function. These are the weighting coefficients. Indicates degree of polarization. The angle between the polarization axis and the horizontal polarization direction. This is a summation function.

[0069] It should be noted that the polarization-spectral correlation model is crucial for connecting and acquiring spectral characteristic data. First, the raw spectral intensity distribution containing polarization information is collected. This model is based on polarization spectroscopy theory and constructs a model from polarization state ( ) to spectral distribution mathematical relationships, borrowing (Polarization-sensitive characteristic spectral function, reflecting the modulation law of polarization to different wavelengths) (Weighting coefficients, reflecting their contribution to the total spectrum) describe the mapping. By fitting the original data, the model parameters are solved using an algorithmic inversion method, making the model output optimally match the original spectrum. After fitting, the model output... It is the pure spectral distribution for interpreting polarization correlation, while the weighting coefficients and characteristic spectral function It is also characteristic data, which is used to extract accurate spectral characteristic data containing polarization modulation information, thus realizing the acquisition of spectral characteristic data from raw mixed data.

[0070] S2: Speckle feature data Interferometric polarization phase information data and spectral characteristic data Spatiotemporal alignment and feature extraction are performed, and a fused multimodal feature vector is generated by concatenating the feature vectors.

[0071] It should be noted that, due to differences in sampling frequency and spatial resolution among different detection modules, the collected speckle feature data needs to be processed. Polarization phase information data and spectral characteristic data Perform temporal and spatial alignment to ensure data consistency in both time and space, laying the foundation for subsequent feature extraction and fusion.

[0072] S21: Data alignment;

[0073] Let the collected speckle feature data be... ,in The number of samples for speckle feature data. For the dimensions of speckle feature data, Spatial resolution / pixel dimension of speckle feature data;

[0074] The acquired interferometric polarization phase information data is ,in The number of samples for interferometric polarization phase information data. The dimension of the interferometric polarization phase information data;

[0075] The collected spectral characteristic data are ,in The number of samples for spectral characteristic data. Dimensions of spectral characteristic data;

[0076] After data alignment, assuming That is, the number of samples for each modality is the same.

[0077] S22: Feature extraction;

[0078] Feature extraction is performed on each of the aligned modal data.

[0079] For speckle feature data, its local spatial features can be extracted through convolution operations; for interferometric polarization phase information data, its phase change features can be extracted; for spectral characteristic data, its spectral peak and wavelength distribution features can be extracted.

[0080] After feature extraction from each modal data, the speckle feature vector is obtained. ,in The extracted speckle feature dimension, where i represents a single detected object and n represents the total number of samples; interferometric polarization phase feature vector. ,in The extracted polarization phase feature dimension; spectral feature vector. ,in This represents the extracted spectral feature dimension.

[0081] S23: Feature fusion;

[0082] The feature vectors of the extracted modal data are fused according to certain rules to form a multimodal fused feature vector that can comprehensively reflect the polarization characteristics of the flexible touch screen. .

[0083] To achieve multimodal data fusion, a vector concatenation method is used to combine the modal feature vectors of each sample into a comprehensive feature vector. For the ... Each sample, the fused feature vector for:

[0084]

[0085] in, This indicates the vertical concatenation of vectors.

[0086] The multimodal fusion feature vector is obtained by combining the fused feature vectors of all samples into a matrix form. :

[0087]

[0088] in, It is a set of sample-level column vectors for multimodal fusion features. Each row corresponds to a complete fusion feature row vector of a sample, which includes speckle feature data, interferometric polarization phase information data, and splicing of spectral characteristic data. Here is the sample-level expansion matrix of the single-modal features, where Let n be the row vector of the speckle feature data of the nth sample. Let be the row vector of the interferometric polarization phase information of the nth sample. This is the row vector of the spectral feature data of the nth sample;

[0089] Further expressed as:

[0090]

[0091] in, , ,

[0092] in, These are speckle characteristic data, interferometric polarization phase information data, and spectral characteristic data, respectively. for The transpose operation. To perform spatiotemporal alignment and feature extraction of high-dimensional mathematical features with multimodal data from flexible screen polarization detection.

[0093] S3: Based on multimodal fusion feature vectors, a CNN-LSTM hybrid neural network model is constructed, and the CNN-LSTM hybrid neural network model is trained using flexible touch screen sample data; the flexible touch screen sample data includes the actual measured polarization characteristic parameters and the corresponding multimodal fusion feature vectors; the CNN-LSTM hybrid neural network model uses CNN to extract the spatial features of speckle and spectrum, and uses LSTM to process the polarization phase temporal features;

[0094] S31: Construct a CNN-LSTM hybrid neural network model based on multimodal data fusion, using a hybrid network structure that combines convolutional neural networks (CNN) and long short-term memory networks (LSTM).

[0095] CNNs are used to extract spatial features from speckle images and spectral data, while LSTMs are used to process time-series data containing polarization phase information. Let the CNN-LSTM hybrid neural network model be... ,in These are the parameters for the CNN-LSTM hybrid neural network model.

[0096] Training was performed using a large amount of sample data from flexible touchscreens with different polarization characteristics. The sample data included actually measured polarization characteristic parameters. ( This includes the number of polarization characteristic parameters (such as degree of polarization, extinction ratio, and polarization axis direction) and the corresponding multimodal fusion feature vector. .

[0097] Mean squared error (MSE) is used as the loss function. :

[0098]

[0099] in, For the sample size, and The first Multimodal fusion data and actual polarization characteristic parameters of each sample To convert complex errors into real losses, a smooth gradient is provided to ensure stable convergence of the model.

[0100] Update model parameters using the stochastic gradient descent (SGD) optimization algorithm. To study the complex relationship between different modal data and the polarization characteristics of flexible touch screens.

[0101] S4: Based on the environmental parameters, including temperature and humidity, acquired in real time by the environmental monitoring module, an adaptive algorithm is used to establish a mapping relationship between the environmental parameters and the parameters of the trained CNN-LSTM hybrid neural network model. The trained CNN-LSTM hybrid neural network model is then dynamically adjusted, including...

[0102] S41: The environmental monitoring module includes a temperature monitor and a humidity monitor for real-time monitoring of ambient temperature. and humidity Parameters provide environmental data support for subsequent adaptive adjustments;

[0103] S42: Temperature obtained from the environmental monitoring module ,humidity The environmental parameters are dynamically adjusted using an adaptive algorithm to train the CNN-LSTM hybrid neural network model.

[0104] Establish environmental parameters and parameters of the trained CNN-LSTM neural network model. mapping relationship ,in These are the mapping parameters. The mapping parameters are obtained through training with experimental data. This enables the neural network model to adaptively adjust its parameters under different environments, resulting in adjusted model parameters. for:

[0105]

[0106] The adjusted model parameters are applied to the neural network model to achieve accurate detection of the polarization characteristics of flexible touch screens under different environments.

[0107] S5: Input the multimodal fused feature vector into the adjusted CNN-LSTM hybrid neural network model, and output the polarization degree, extinction ratio and polarization axis direction parameters;

[0108] Reference Figure 5In this flexible touchscreen polarization characteristic detection process, from left to right, the process first takes spectral characteristic data, speckle feature data, and interferometric polarization phase information data as input. Then, one-dimensional local features of the spectral characteristic data are extracted through one-dimensional convolution Conv1D in the convolutional neural network (CNN), two-dimensional spatial features of the speckle feature data are extracted through two-dimensional convolution Conv2D, and interferometric polarization phase information data is extracted through long short-term memory network (LSTM). Then, the multi-modal information is integrated by feature splicing, and finally, the fully connected layer globally fuses and maps it to the output dimension to realize the detection and result output of the polarization characteristics of the flexible touchscreen.

[0109] Furthermore, it outputs parameters such as the degree of polarization, extinction ratio, and polarization axis direction of the flexible touchscreen.

[0110] Matters not covered in this invention are common knowledge.

[0111] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

[0113] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion, characterized in that, Includes the following steps: S1: Construct a multi-parameter measurement optical path system for flexible touch screens to acquire multi-modal data after the light source is transmitted through the flexible touch screen, including speckle feature data, interferometric polarization phase information data, and spectral characteristic data. S2: Perform spatiotemporal alignment and feature extraction on multimodal data, and generate multimodal fusion feature vectors through feature fusion; S3: Based on multimodal fusion feature vectors, a CNN-LSTM hybrid neural network model is constructed, and the CNN-LSTM hybrid neural network model is trained using flexible touch screen sample data; the flexible touch screen sample data includes the actual measured polarization characteristic parameters and the corresponding multimodal fusion feature vectors; the CNN-LSTM hybrid neural network model uses CNN to extract the spatial features of speckle and spectrum, and uses LSTM to process the polarization phase temporal features; S4: Real-time monitoring of environmental parameters, including ambient temperature and humidity, establishes a mapping relationship between environmental parameters and the parameters of the trained CNN-LSTM hybrid neural network model through an adaptive algorithm, dynamically adjusts the parameters of the trained CNN-LSTM hybrid neural network model, and obtains the adjusted CNN-LSTM hybrid neural network model. S5: Input the multimodal fusion feature vector into the adjusted CNN-LSTM hybrid neural network model, and output the polarization degree, extinction ratio and polarization axis direction parameters of the flexible touch screen.

2. The method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion according to claim 1, characterized in that, The acquisition of speckle feature data in step S1 includes: Based on the principle of mode superposition, when light sources are incident on a flexible touchscreen with different polarization states, the intensity distribution of the output light field is determined by the coherent superposition of multiple modes of light. in, To output the light field intensity distribution, For the first The amplitude of each mode, It is an exponential function. The imaginary unit, Let be the propagation constant. For wavelength, The thickness of the flexible touchscreen. The number of modes supported for flexible touchscreens; After the output light field is scattered by the flexible screen, the speckle intensity distribution data is obtained based on the square of the modulus of the complex amplitude distribution of the light field. ,in, These are the pixel coordinates of the image sensor in the horizontal and vertical directions. By acquiring speckle intensity distribution data through an image sensor, speckle characteristic data caused by changes in light polarization state after being affected by a flexible touchscreen are obtained.

3. The method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion according to claim 2, characterized in that, The acquisition of interferometric polarization phase information data in step S1 includes: Using a Mach-Zehnder interferometer or a Sanignac interferometer, the light source is split into two beams. By setting different polarization beam-splitting paths, the two beams carry different polarization information; according to the principle of interference, the intensity of the interfering light... for: in, , The light intensities of the two beams, The phase difference between the two beams of light. It is a cosine function; Based on the Jones matrix theory of polarized light, the contribution of polarization state to phase difference during the modulation of two beams is analyzed, the phase delay information introduced by polarization modulation is decoupled, and interferometric polarization phase information data is obtained.

4. The method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion according to claim 3, characterized in that, The acquisition of spectral characteristic data in step S1 includes: Let the spectral intensity distribution of light be... By analyzing the distribution characteristics of spectral intensity at different wavelengths and combining it with the spectral theory model of polarized light, a model is constructed based on the polarization state ( ) to spectral distribution Based on the mathematical relationships, establish a polarization-spectral correlation model: in, It is a polarization-sensitive characteristic spectral function. These are the weighting coefficients. Indicates degree of polarization. The angle between the polarization axis and the horizontal polarization direction. For summation functions; The spectral characteristic data are obtained by solving the polarization-spectrum correlation model parameters through algorithmic inversion.

5. The method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion according to claim 4, characterized in that, Step S2, generating the multimodal fusion feature vector, includes: The feature fusion adopts a vertical splicing method; Multimodal fusion feature vector : ,in These are speckle characteristic data, interferometric polarization phase information data, and spectral characteristic data, respectively. for The transpose operation. To perform spatiotemporal alignment and feature extraction of high-dimensional mathematical features with multimodal data from flexible screen polarization detection.

6. The method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion according to claim 5, characterized in that, Step S3 involves constructing the CNN-LSTM hybrid neural network model, which includes: Using mean squared error as the loss function : in, For the sample size, and The first The multimodal fusion feature vector and actual polarization characteristic parameters of each sample To convert complex errors into real losses, smooth gradients are provided to ensure stable model convergence; Update the parameters of the CNN-LSTM hybrid neural network model using the stochastic gradient descent optimization algorithm. By learning the complex relationship between different data modes and the polarization characteristics of flexible touch screens, a trained CNN-LSTM hybrid neural network model is obtained.

7. The method for detecting the polarization characteristics of a flexible touchscreen based on multimodal data fusion according to any one of claims 1-6, characterized in that, Step S4, where the adaptive algorithm dynamically adjusts the CNN-LSTM hybrid neural network, includes: establishing a mapping function between environmental parameters and the trained CNN-LSTM hybrid neural network model parameters, to obtain the adjusted model parameters. : in, The parameters of the CNN-LSTM hybrid neural network model before training were adjusted; For temperature, Humidity; These are mapping parameters.

Images