Dislocation short circuit signal decoupling method and device, electronic equipment and storage medium
By acquiring the signal matrix from the touch screen, locating the peak point, and processing it using a neural network model and logistic regression function, the problem of inaccurate touch positioning caused by misaligned short-circuited sensors was solved, achieving a high-precision touch positioning effect.
Patent Information
- Application Number
- CN202511668335.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-02-24
AI Technical Summary
Existing touch screens, by employing a misaligned short-circuit sensor design, cause touch signal fusion, resulting in a false center signal. This severely affects the accuracy of touch positioning, and existing calibration methods are unable to effectively compensate for this hardware-introduced error.
By acquiring the global sensor signal matrix of the touch screen, locating the signal peak points, extracting local patches and normalizing them, and then inputting them into a pre-trained neural network model, a hard constraint mask is generated and processed by a logistic regression function to calculate the centroid coordinates to achieve high-precision touch positioning.
The fused signal after shorting is accurately restored to the ideal signal, achieving high-precision touch positioning and improving the user experience.
Smart Images
Figure CN121560183A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of touch signal processing technology, and in particular to a method, apparatus, electronic device and storage medium for decoupling misaligned short-circuit signals. Background Technology
[0002] Touchscreens, as the core human-computer interaction interface, are widely used in various electronic devices. In market competition, how to reduce costs while ensuring touch performance has become a key challenge for manufacturers. One strategy to reduce hardware costs is to adopt a staggered short-circuit sensor design, which involves short-circuiting physically adjacent sensor units to share a signal channel. However, this design causes the original touch signals to fuse, forming a "pseudo-center" signal that deviates from the actual touch point, severely undermining the signal distribution assumptions upon which traditional centroid positioning algorithms rely.
[0003] Existing calibration methods, such as lookup tables or linear interpolation, struggle to address the highly nonlinear, position-dependent systematic errors introduced by shorting, leading to inaccurate touch positioning and a poor user experience. Therefore, there is an urgent need in the field for a new method that can effectively compensate for these hardware-introduced errors, thereby achieving high-precision touch control on low-cost hardware. Summary of the Invention
[0004] Therefore, it is necessary to provide a method, apparatus, electronic device, and storage medium for decoupling misaligned short-circuit signals that can effectively compensate for the introduced errors, in order to address the above-mentioned technical problems.
[0005] A method for decoupling misaligned short-circuit signals, the method comprising:
[0006] S1: Obtain the global sensor signal matrix of the touch screen and locate the signal peak point;
[0007] S2: Using the signal peak point as the center, extract a local patch of a preset size from the global sensor signal matrix;
[0008] S3: Normalize the local map pieces to make them conform to the probability distribution;
[0009] S4: Input the normalized local map patch into the pre-trained neural network model to obtain the original output logic value;
[0010] S5: Generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0011] S6: Apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints;
[0012] S7: Calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0013] In one embodiment, it further includes:
[0014] S10: Obtain the training dataset, which contains multiple training samples. Each training sample includes a short-circuit signal patch, a corresponding ideal signal patch, and associated physical prior information.
[0015] S20: Normalize the short-circuit signal block to generate an input probability distribution; and normalize the ideal signal block to generate a target probability distribution;
[0016] S30: Input the normalized local map patch into the neural network model to be trained to obtain the original output logic value;
[0017] S40: Generate a hard constraint mask corresponding to the current training sample based on the associated physical prior information;
[0018] S50: The original output logic value is corrected using the hard constraint mask, and a logistic regression function is applied to the corrected logic value to generate a predicted probability distribution;
[0019] S60: Calculate the composite loss value between the predicted probability distribution and the target probability distribution;
[0020] S70: Based on the composite loss value, update the parameters of the neural network model using the backpropagation algorithm;
[0021] S80: Repeat steps S20 to S70 until the neural network model meets the preset convergence condition.
[0022] In one embodiment, the neural network model is a convolutional neural network, and its architecture includes:
[0023] The feature extraction layer, consisting of at least two consecutive 3x3 convolutional blocks, is used to extract features from the input and expand the number of feature channels;
[0024] The feature integration layer, consisting of at least one 1x1 convolutional block, is used to integrate features;
[0025] The output layer, consisting of a 1x1 convolutional layer, projects the integrated features back to the single-channel original output logic value with the same size as the input patch.
[0026] In one embodiment, the specific process of obtaining the training dataset includes:
[0027] Acquire paired global short-circuit signal matrices and global ideal signal matrices;
[0028] Locate the signal peak point in each of the global short-circuit signal matrices;
[0029] Centered on the signal peak point, a local patch of a preset size is extracted as the short-circuit signal patch;
[0030] At the same location in the global ideal signal matrix, local patches of the same size are extracted as the ideal signal patches.
[0031] In one embodiment, when extracting the short-circuit signal block, the method further includes:
[0032] Random horizontal jitter is applied to the center column of the local patch to perform data augmentation.
[0033] In one embodiment, the composite loss value L is composed of geometric loss. Distribution loss and contour loss The weighted calculation yields the following formula:
[0034]
[0035] in, , , The preset weighting coefficients, The L1 distance is the centroid coordinate between the predicted probability distribution and the target probability distribution. The KL divergence between the predicted probability distribution and the target probability distribution; The minimum "movement cost" between the predicted probability distribution and the target probability distribution projected in the horizontal direction.
[0036] In one embodiment, step S50 specifically includes:
[0037] Set the original output logic value corresponding to the invalid position in the hard constraint mask to negative infinity;
[0038] The logistic regression function is applied to the set logical values to generate the predicted probability distribution.
[0039] A decoupling device for misaligned short-circuit signals, the device comprising:
[0040] The signal acquisition module is used to acquire the global sensor signal matrix of the touch screen and locate the signal peak points;
[0041] The patch extraction module is used to extract a local patch of a preset size from the global sensor signal matrix, centered on the signal peak point.
[0042] The distribution prediction module is used to normalize the local map tiles to make them conform to a probability distribution;
[0043] The logic value acquisition module is used to input the normalized local map patch into a pre-trained neural network model to obtain the original output logic value;
[0044] The signal constraint module is used to generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0045] The logic value correction module is used to apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints.
[0046] The position output module is used to calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0047] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:
[0048] S1: Obtain the global sensor signal matrix of the touch screen and locate the signal peak point;
[0049] S2: Using the signal peak point as the center, extract a local patch of a preset size from the global sensor signal matrix;
[0050] S3: Normalize the local map pieces to make them conform to the probability distribution;
[0051] S4: Input the normalized local map patch into the pre-trained neural network model to obtain the original output logic value;
[0052] S5: Generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0053] S6: Apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints;
[0054] S7: Calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0055] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:
[0056] S1: Obtain the global sensor signal matrix of the touch screen and locate the signal peak point;
[0057] S2: Using the signal peak point as the center, extract a local patch of a preset size from the global sensor signal matrix;
[0058] S3: Normalize the local map pieces to make them conform to the probability distribution;
[0059] S4: Input the normalized local map patch into the pre-trained neural network model to obtain the original output logic value;
[0060] S5: Generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0061] S6: Apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints;
[0062] S7: Calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0063] The aforementioned method, apparatus, electronic device, and storage medium for decoupling misaligned short-circuit signals acquire the global sensor signal matrix of the touch screen and locate the signal peak point; extract a local patch of a preset size from the global sensor signal matrix with the signal peak point as the center; normalize the local patch to make it conform to a probability distribution; input the normalized local patch into a pre-trained neural network model to obtain the original output logic value; generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix, the hard constraint mask being used to identify the sensor positions where the signal is physically allowed; apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to physical constraints; calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the high-precision touch point position after decoupling. This application combines the powerful fitting ability of deep learning with the physical prior knowledge of sensor layout, which can accurately restore the fused signal after short-circuiting to the ideal non-short-circuit signal, thereby achieving high-precision touch positioning. Attached Figure Description
[0064] Figure 1 This is a schematic diagram of a touch sensor being shorted in one embodiment;
[0065] Figure 2 This is a flowchart illustrating a method for decoupling misaligned short-circuit signals in one embodiment;
[0066] Figure 3 A flowchart illustrating a method for training a neural network model in another embodiment;
[0067] Figure 4 A schematic diagram of the network model in one embodiment;
[0068] Figure 5 This is a structural block diagram of a misaligned short-circuit signal decoupling device in one embodiment;
[0069] Figure 6 This is a diagram of the internal structure of an electronic device in one embodiment. Detailed Implementation
[0070] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0071] It should be noted that staggered shorting of touchscreen sensors refers to physically connecting sensor driving or sensing channels that are not directly adjacent but are distributed according to a specific topological pattern, making these channels an equivalent unit at the circuit level. For example, channels in the 2nd and 3rd columns of odd-numbered rows can be connected, and channels in the 1st and 2nd columns of even-numbered rows can be connected, and so on, with staggered shorting as follows. Figure 1 As shown.
[0072] In one embodiment, such as Figure 2 As shown, a method for decoupling misaligned short-circuit signals is provided, which can be applied to... Figure 1 Taking an electronic device as an example, the explanation includes the following steps:
[0073] Step S1: Obtain the global sensor signal matrix of the touch screen and locate the signal peak point.
[0074] Step S2: Using the signal peak point as the center, extract a local patch of a preset size from the global sensor signal matrix.
[0075] The preset size is 3x7. A 3×7 local patch is extracted from the global signal matrix. The extraction process centers on the signal peak points in the short-circuit signal matrix to ensure that the most critical short-circuit information is completely captured.
[0076] Step S3: Normalize the local map patch to make it conform to the probability distribution.
[0077] Step S4: Input the normalized local map patch into the pre-trained neural network model to obtain the original output logic value.
[0078] Step S5: Based on the position information of the local patch in the global sensor signal matrix, generate a corresponding hard constraint mask. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0079] Step S6: Apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints.
[0080] Step S7: Calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0081] In the above-mentioned method for decoupling misaligned short-circuit signals, the global sensor signal matrix of the touch screen is obtained, and the signal peak point is located. A local patch of a preset size is extracted from the global sensor signal matrix, centered on the signal peak point. The local patch is normalized to conform to a probability distribution. The normalized local patch is input into a pre-trained neural network model to obtain the original output logic value. Based on the position information of the local patch in the global sensor signal matrix, a corresponding hard constraint mask is generated. The hard constraint mask is used to identify the sensor positions where signals are physically allowed. The hard constraint mask is applied to the original output logic value and processed through a logistic regression function to obtain a predicted probability distribution that conforms to physical constraints. The centroid coordinates of the predicted probability distribution are calculated, and these centroid coordinates are used as the output of the high-precision touch point position after decoupling. This application combines the powerful fitting ability of deep learning with the physical prior knowledge of sensor layout, enabling precise restoration of the fused signal after short-circuiting to an ideal non-short-circuit signal, thereby achieving high-precision touch positioning.
[0082] In one embodiment, such as Figure 3 As shown, prior to step S1 above, the method further includes:
[0083] S10: Obtain the training dataset, which contains multiple training samples. Each training sample includes a short-circuit signal patch, a corresponding ideal signal patch, and associated physical prior information.
[0084] The training dataset includes paired shorted signal matrices, non-shorted ideal signal matrices, and associated physical prior information. To focus on local features, 3×7 local patches are extracted from the global signal matrix. The extraction process centers on the signal peaks in the shorted signal matrix, ensuring that the most critical shorting information is fully captured. A width of 3×7 is chosen to fully encompass the lateral diffusion and fusion features of the signal.
[0085] Specifically, the process of obtaining the training dataset includes:
[0086] Acquire paired global short-circuit signal matrices and global ideal signal matrices;
[0087] Locate the signal peak point in each of the global short-circuit signal matrices;
[0088] Centered on the signal peak point, a local patch of a preset size is extracted as the short-circuit signal patch; preferably, when extracting the short-circuit signal patch, the method further includes:
[0089] Random horizontal jitter is applied to the center column of the local patch to augment the data. During training, random horizontal jitter is applied to the center column of the patch matrix to improve the model's robustness to slight shifts in peak positions.
[0090] At the same location in the global ideal signal matrix, local patches of the same size are extracted as the ideal signal patches.
[0091] S20: Normalize the short-circuit signal block to generate an input probability distribution; and normalize the ideal signal block to generate a target probability distribution.
[0092] The extracted Patch M (whether it is a shortened matrix or a non-shortened matrix) is converted into a probability distribution p, which is calculated using the following formula:
[0093]
[0094] Where (p,c) are the coordinates within the Patch. It is a very small constant to prevent the denominator from being zero. This operation ensures that all values are non-negative and that their sum is 1.
[0095] S30: Input the normalized local map patch into the neural network model to be trained to obtain the original output logic value.
[0096] The neural network model is a convolutional neural network, specifically a lightweight convolutional neural network designed for efficient processing of small input sizes (3×7 patches). Its core is a feedforward sequence of convolutional kernels, and its architecture includes:
[0097] The feature extraction layer, consisting of at least two consecutive 3x3 convolutional blocks, is used to extract features from the input and expand the number of feature channels;
[0098] The feature integration layer, consisting of at least one 1x1 convolutional block, is used to integrate features;
[0099] The output layer, consisting of a 1x1 convolutional layer, projects the integrated features back to the single-channel original output logic value with the same size as the input patch.
[0100] Each convolutional layer is followed by a batch normalization layer and a ReLU activation function, as shown in the model structure. Figure 4 As shown.
[0101] S40: Generate a hard constraint mask corresponding to the current training sample based on the associated physical prior information.
[0102] S50: The original output logic value is corrected using the hard constraint mask, and a logistic regression function is applied to the corrected logic value to generate a predicted probability distribution;
[0103] The physical design knowledge of the sensor is incorporated into the deep learning model. This constraint is enforced in the final step of the model's forward propagation using a masked logistic regression Softmax function. The calculation of the original output logistic value Logits Z is divided into two steps:
[0104] Apply mask: Set the Logit value of the corresponding position in the hard constraint mask H that is 0 (i.e. the position of the physical non-short-circuited sensor) to negative infinity.
[0105]
[0106] Calculate Softmax: Apply the standard Softmax function to the processed Logits Z' to obtain the final probability distribution P.
[0107]
[0108] By forcing the model to assign probabilities only at physically short-circuited locations, the possibility of generating spurious signals in invalid regions can be avoided, allowing the network to focus more on learning how to correctly allocate signal weights between valid locations.
[0109] S60: Calculate the composite loss value between the predicted probability distribution and the target probability distribution.
[0110] To optimize the prediction results from multiple dimensions, a weighted composite loss function is defined, where the composite loss value L is composed of geometric loss. Distribution loss and contour loss The weighted calculation yields the following formula:
[0111]
[0112] in, , , The preset weighting coefficients, The L1 distance is the centroid coordinate between the predicted probability distribution and the target probability distribution. The KL divergence between the predicted probability distribution and the target probability distribution; The minimum "movement cost" between the predicted probability distribution and the target probability distribution projected in the horizontal direction.
[0113] Geometric loss (Centroid L1 Loss): The core loss function, directly aimed at improving positioning accuracy. It first calculates the centroid coordinates of the predicted distribution P and the target distribution t. :
[0114]
[0115] Then, the L1 distance (Manhattan distance) between the two centroids is calculated as the loss:
[0116]
[0117] in It is a weighting coefficient used to balance the importance of the x and y directions. This loss function directly penalizes the bias in coordinate predictions as it drives the model to learn to generate a signal distribution with the correct centroid.
[0118] Distributed loss (KL divergence): Used to measure the overall shape similarity between two probability distributions. It ensures that the model not only focuses on the centroid location but also learns to generate signals with a shape similar to the target distribution. The formula is:
[0119]
[0120] Where t(r, c) is the target probability distribution on the coordinate system. The probability value at point ɛ is a very small constant to prevent the denominator from being zero.
[0121] Contour loss (1D Wasserstein distance): This loss function provides a more robust measure of the signal profile. It first considers the two-dimensional distribution... Projecting onto the x-axis yields a one-dimensional contour distribution. Then calculate its cumulative distribution function.
[0122]
[0123] The 1D Wasserstein distance is calculated as follows:
[0124]
[0125] Wasserstein distance measures the “cost” required to “transfer” one distribution to another, and can be used to optimize the overall shape and energy distribution of a signal.
[0126] S70: Based on the composite loss value, update the parameters of the neural network model using the backpropagation algorithm.
[0127] S80: Repeat steps S20 to S70 until the neural network model meets the preset convergence condition.
[0128] Repeat the above steps until the model's performance on the validation set (using centroid error as the primary metric) converges. During training, when the validation set error reaches a new minimum, save the current model parameters as the optimal model.
[0129] It should be understood that, although Figure 2-3 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are executed, and they can be performed in other orders. Furthermore, Figure 2-3 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.
[0130] In one embodiment, such as Figure 5 As shown, a decoupling device for misaligned short-circuit signals is provided, comprising: a signal acquisition module 51, a patch extraction module 52, a distribution prediction module 53, a logic value acquisition module 54, a signal constraint module 55, a logic value correction module 56, and a position output module 57, wherein:
[0131] The signal acquisition module 51 is used to acquire the global sensor signal matrix of the touch screen and locate the signal peak point;
[0132] The patch extraction module 52 is used to extract a local patch of a preset size from the global sensor signal matrix, centered on the signal peak point.
[0133] The distribution prediction module 53 is used to normalize the local map patch so that it conforms to the probability distribution;
[0134] The logic value acquisition module 54 is used to input the normalized local map patch into a pre-trained neural network model to obtain the original output logic value;
[0135] The signal constraint module 55 is used to generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0136] The logic value correction module 56 is used to apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints.
[0137] The position output module 57 is used to calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0138] Furthermore, the device also includes:
[0139] The dataset acquisition module is used to acquire the training dataset, which contains multiple training samples. Each training sample includes a short-circuit signal patch, a corresponding ideal signal patch, and associated physical prior information.
[0140] The data processing module is used to normalize the short-circuit signal patch to generate an input probability distribution; and to normalize the ideal signal patch to generate a target probability distribution.
[0141] The network learning module is used to input the normalized local map patch into the neural network model to be trained to obtain the original output logical value;
[0142] The physics prior module is used to generate a hard constraint mask corresponding to the current training sample based on the associated physics prior information.
[0143] The logic value correction module is used to correct the original output logic value using the hard constraint mask, and apply a logistic regression function to the corrected logic value to generate a predicted probability distribution.
[0144] The loss value calculation module is used to calculate the composite loss value between the predicted probability distribution and the target probability distribution;
[0145] The parameter update module is used to update the parameters of the neural network model based on the composite loss value using a backpropagation algorithm.
[0146] The model convergence module is used to repeatedly execute steps S20 to S70 until the neural network model meets the preset convergence conditions.
[0147] Furthermore, the neural network model is a convolutional neural network, and its architecture includes:
[0148] The feature extraction layer, consisting of at least two consecutive 3x3 convolutional blocks, is used to extract features from the input and expand the number of feature channels;
[0149] The feature integration layer, consisting of at least one 1x1 convolutional block, is used to integrate features;
[0150] The output layer, consisting of a 1x1 convolutional layer, projects the integrated features back to the single-channel original output logic value with the same size as the input patch.
[0151] Furthermore, the specific process of obtaining the training dataset includes:
[0152] Acquire paired global short-circuit signal matrices and global ideal signal matrices;
[0153] Locate the signal peak point in each of the global short-circuit signal matrices;
[0154] Centered on the signal peak point, a local patch of a preset size is extracted as the short-circuit signal patch;
[0155] At the same location in the global ideal signal matrix, local patches of the same size are extracted as the ideal signal patches.
[0156] Furthermore, when extracting the short-circuit signal block, the device further includes:
[0157] The data augmentation module is used to randomly jitter the center column of the local map tile to perform data augmentation.
[0158] Furthermore, the composite loss value L is composed of geometric loss. Distribution loss and contour loss The weighted calculation yields the following formula:
[0159]
[0160] in, , , The preset weighting coefficients, The L1 distance is the centroid coordinate between the predicted probability distribution and the target probability distribution. The KL divergence between the predicted probability distribution and the target probability distribution; The minimum "movement cost" between the predicted probability distribution and the target probability distribution projected in the horizontal direction.
[0161] Furthermore, the loss value calculation module is specifically used to set the original output logic value corresponding to the invalid position in the hard constraint mask to negative infinity;
[0162] The logistic regression function is applied to the set logical values to generate the predicted probability distribution.
[0163] Specific limitations regarding the misaligned short-circuit signal decoupling device can be found in the limitations of the misaligned short-circuit signal decoupling method described above, and will not be repeated here. Each module in the aforementioned misaligned short-circuit signal decoupling device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the electronic device, or stored in software in the memory of the electronic device, so that the processor can call and execute the corresponding operations of each module.
[0164] In one embodiment, an electronic device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown, the electronic device includes a processor, memory, and a display screen connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the computer program is executed by the processor, it implements a method for decoupling misaligned short-circuit signals. The display screen can be a liquid crystal display (LCD) or an electronic ink display, and the input device can be a touch layer covering the display screen.
[0165] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0166] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the following steps:
[0167] S1: Obtain the global sensor signal matrix of the touch screen and locate the signal peak point;
[0168] S2: Using the signal peak point as the center, extract a local patch of a preset size from the global sensor signal matrix;
[0169] S3: Normalize the local map pieces to make them conform to the probability distribution;
[0170] S4: Input the normalized local map patch into the pre-trained neural network model to obtain the original output logic value;
[0171] S5: Generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0172] S6: Apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints;
[0173] S7: Calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0174] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:
[0175] S1: Obtain the global sensor signal matrix of the touch screen and locate the signal peak point;
[0176] S2: Using the signal peak point as the center, extract a local patch of a preset size from the global sensor signal matrix;
[0177] S3: Normalize the local map pieces to make them conform to the probability distribution;
[0178] S4: Input the normalized local map patch into the pre-trained neural network model to obtain the original output logic value;
[0179] S5: Generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist.
[0180] S6: Apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints;
[0181] S7: Calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
[0182] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0183] 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.
[0184] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. 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 all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
Claims
1. A method for decoupling misaligned short-circuit signals, characterized in that, The method includes: S1: Obtain the global sensor signal matrix of the touch screen and locate the signal peak point; S2: Using the signal peak point as the center, extract a local patch of a preset size from the global sensor signal matrix; S3: Normalize the local map pieces to make them conform to the probability distribution; S4: Input the normalized local map patch into the pre-trained neural network model to obtain the original output logic value; S5: Generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist. S6: Apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints; S7: Calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
2. The method according to claim 1, characterized in that, Prior to step S1, the method further includes: S10: Obtain the training dataset, which contains multiple training samples. Each training sample includes a short-circuit signal patch, a corresponding ideal signal patch, and associated physical prior information. S20: Normalize the short-circuit signal block to generate an input probability distribution; and normalize the ideal signal block to generate a target probability distribution; S30: Input the normalized local map patch into the neural network model to be trained to obtain the original output logic value; S40: Generate a hard constraint mask corresponding to the current training sample based on the associated physical prior information; S50: The original output logic value is corrected using the hard constraint mask, and a logistic regression function is applied to the corrected logic value to generate a predicted probability distribution; S60: Calculate the composite loss value between the predicted probability distribution and the target probability distribution; S70: Based on the composite loss value, update the parameters of the neural network model using the backpropagation algorithm; S80: Repeat steps S20 to S70 until the neural network model meets the preset convergence condition.
3. The method according to claim 1 or 2, characterized in that, The neural network model is a convolutional neural network, and its architecture includes: The feature extraction layer, consisting of at least two consecutive 3x3 convolutional blocks, is used to extract features from the input and expand the number of feature channels; The feature integration layer, consisting of at least one 1x1 convolutional block, is used to integrate features; The output layer, consisting of a 1x1 convolutional layer, projects the integrated features back to the single-channel original output logic value with the same size as the input patch.
4. The method according to claim 2, characterized in that, In step S10, the specific process of obtaining the training dataset includes: Acquire paired global short-circuit signal matrices and global ideal signal matrices; Locate the signal peak point in each of the global short-circuit signal matrices; Centered on the signal peak point, a local patch of a preset size is extracted as the short-circuit signal patch; At the same location in the global ideal signal matrix, local patches of the same size are extracted as the ideal signal patches.
5. The method according to claim 4, characterized in that, When extracting the short-circuit signal block, the method further includes: Random horizontal jitter is applied to the center column of the local patch to perform data augmentation.
6. The method according to claim 2, characterized in that, In step S60, the composite loss value L is composed of geometric loss. Distribution loss and contour loss The weighted calculation yields the following formula: in, , , The preset weighting coefficients, The L1 distance is the centroid coordinate between the predicted probability distribution and the target probability distribution. The KL divergence between the predicted probability distribution and the target probability distribution; The minimum "movement cost" between the predicted probability distribution and the target probability distribution projected in the horizontal direction.
7. The method according to claim 2, characterized in that, Step S50 specifically includes: Set the original output logic value corresponding to the invalid position in the hard constraint mask to negative infinity; The logistic regression function is applied to the set logical values to generate the predicted probability distribution.
8. A decoupling device for misaligned short-circuit signals, characterized in that, The device includes: The signal acquisition module is used to acquire the global sensor signal matrix of the touch screen and locate the signal peak points; The patch extraction module is used to extract a local patch of a preset size from the global sensor signal matrix, centered on the signal peak point. The distribution prediction module is used to normalize the local map tiles to make them conform to a probability distribution; The logic value acquisition module is used to input the normalized local map patch into a pre-trained neural network model to obtain the original output logic value; The signal constraint module is used to generate a corresponding hard constraint mask based on the position information of the local patch in the global sensor signal matrix. The hard constraint mask is used to identify the sensor positions where signals are physically allowed to exist. The logic value correction module is used to apply the hard constraint mask to the original output logic value and process it through a logistic regression function to obtain a predicted probability distribution that conforms to the physical constraints. The position output module is used to calculate the centroid coordinates of the predicted probability distribution and output the centroid coordinates as the decoupled high-precision touch point position.
9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.