Method and system for preventing mistaken touch of touch screen and computer equipment

Abstract

The invention relates to the technical field of touch screen control, and discloses a touch screen mistaken touch prevention method and system and computer device.The method comprises the steps that touch input signals of a user are collected on the basis of a sensor in a touch screen and filtered and normalized, and target touch signals are obtained; inputting the target touch signal into a sliding mode compensation function observer for touch signal interference estimation to obtain false touch interference information; performing feature extraction on the target touch signal according to the false touch interference information to obtain a touch feature vector; processing through the touch signal classification model to obtain a touch signal classification result; performing statistical analysis on a touch signal classification result, and judging an effective touch input event; according to the effective touch input event and the touch signal classification result, the touch input signal is processed, and the touch response signal of the touch screen is generated, so that dynamic adjustment is performed according to the use habits of different users and the use environment of equipment, and the mistaken touch prevention performance of the touch screen is improved.

Description

Technical Field

[0001] The present application relates to the field of touch screen control technology, and in particular to a touch screen accidental touch prevention method, system and computer equipment. Background Art

[0002] With the widespread application of touch screen technology, the user experience of smart devices is increasingly dependent on the accuracy and smoothness of touch interaction. However, in actual use, touch screens often face various forms of false touch problems, which not only affects the user's operating efficiency, but may also cause unexpected device responses and cause trouble to users. Traditional touch recognition methods mainly rely on simple threshold judgment or fixed pattern recognition, which is difficult to adapt to complex and changing usage environments and user habits. Summary of the invention

[0003] The present application provides a touch screen anti-mistouch method, system and computer device, which can be dynamically adjusted according to the usage habits of different users and the usage environment of the device, thereby improving the touch screen anti-mistouch performance.

[0004] In a first aspect, the present application provides a touch screen anti-mistouch method, the touch screen anti-mistouch method comprising: Collecting a user's touch input signal based on a sensor in the touch screen, and filtering and normalizing the touch input signal to obtain a target touch signal; Inputting the target touch signal into a sliding mode compensation function observer to perform touch signal interference estimation to obtain false touch interference information; Extracting features of the target touch signal according to the false touch interference information to obtain a touch feature vector; Inputting the touch feature vector into a preset touch signal classification model for processing to obtain a touch signal classification result; Performing statistical analysis on the touch signal classification results within the sliding time window, and determining a valid touch input event; Processing the touch input signal and generating a touch response signal of the touch screen according to the valid touch input event and the touch signal classification result A second aspect of the present application provides a touch screen accidental touch prevention system, the touch screen accidental touch prevention system comprising: An acquisition module, used for acquiring a user's touch input signal based on a sensor in the touch screen, and filtering and normalizing the touch input signal to obtain a target touch signal; an estimation module, configured to input the target touch signal into a sliding mode compensation function observer to perform touch signal interference estimation to obtain false touch interference information; An extraction module, configured to extract features of the target touch signal according to the false touch interference information to obtain a touch feature vector; A processing module, used for inputting the touch feature vector into a preset touch signal classification model for processing to obtain a touch signal classification result; An analysis module, used for performing statistical analysis on the touch signal classification results within a sliding time window and determining a valid touch input event; A generating module is used to process the touch input signal and generate a touch response signal of the touch screen according to the valid touch input event and the touch signal classification result.

[0005] The third aspect of the present application provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned touch screen anti-mistouch method.

[0006] Compared with the prior art, the present application has the following beneficial effects: by filtering and normalizing the touch input signal, combined with the Butterworth low-pass filter and outlier removal techniques, the quality of the original touch data is effectively improved, and the sliding mode compensation function observer is used to estimate the touch signal interference, which can capture various forms of false touch interference in real time and accurately, and improve the system's adaptability to complex touch scenes. Touch features are extracted by various methods such as time domain analysis, frequency domain transformation and wavelet transformation, and the principal component analysis is combined for dimensionality reduction, which not only retains the key information of the touch signal, but also uses a dual-stream attention enhanced convolutional neural network and a multi-head self-calibration classifier to make full use of the advantages of deep learning in feature extraction and classification tasks, and improve the accuracy of touch signal classification. Through statistical analysis and multiple judgment criteria within the sliding time window, the probability of false touch and misjudgment is effectively reduced, and the stability and reliability of the system are improved. The present invention can dynamically adjust the feature extraction and classification strategies according to the false touch interference information, and has good environmental adaptability and user personalization capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0008] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0009] Figure 1 1 is a schematic diagram of a flow chart of a touch screen accidental touch prevention method provided by an embodiment of the present invention; Figure 2 It is a schematic block diagram of the structure of a touch screen accidental touch prevention system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0011] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0012] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0013] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In the embodiment of the present application, an embodiment of the touch screen anti-mistouch method includes: Step 100: collecting a user's touch input signal based on a sensor in the touch screen, and filtering and normalizing the touch input signal to obtain a target touch signal; It is understandable that the execution subject of the present application may be a touch screen anti-mistouch system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0014] Specifically, the surface of the touch screen is scanned at high frequency to collect raw touch data including touch position coordinates, touch pressure values, touch duration and touch frequency. The raw touch data is input into a Butterworth low-pass filter to filter out high-frequency noise, eliminate high-frequency interference signals therein, and obtain touch data after preliminary filtering. Outliers and impulse noise are removed from the touch data after preliminary filtering to obtain more accurate and clean noise-reduced touch data. The touch pressure value is converted to a range of 0 to 1 through a linear mapping function to obtain a normalized touch pressure value. At the same time, for the data of touch duration, since its numerical range is usually large and difficult to process, its numerical range is reduced by logarithmic transformation to obtain the touch duration after logarithmic transformation, thereby improving the robustness and computational efficiency of the algorithm. When processing the touch frequency, an exponential smoothing method is used to reduce the influence of short-term fluctuations, obtain the touch frequency after smoothing, make the touch frequency data more stable and continuous, and reduce the possibility of false touches. The touch position coordinates in the noise-reduced touch data are track-smoothed and predicted to eliminate possible jitter and inaccuracy, and generate optimized touch position coordinates. A complete target touch signal is obtained by combining the normalized touch pressure value, the touch duration after logarithmic transformation, the touch frequency after smoothing, and the optimized touch position coordinates.

[0015] Step 200: Input the target touch signal into the sliding mode compensation function observer to estimate the touch signal interference and obtain false touch interference information; Specifically, the target touch signal is represented as x1, and the first-order derivative is calculated to obtain the change rate of the touch signal x2. Considering the possible false touch interference factors in the touch operation, the false touch interference variable x3 is introduced to construct the system state vector x, which contains x1, x2 and x3, and is expressed as x=[x1, x2, x3] T , where x1 represents the amplitude of the touch signal, x2 represents the rate of change of the touch signal, and x3 represents the false touch interference. Based on the system state vector, the sliding mode compensation function observer equation group is constructed. The specific equation is: x1 hat =z1+β1|x1-z1| (α1) sign(x1-z1); x2 hat =z2+β2|x1-z1| (α2) sign(x1-z1); x3 hat =z3+β3|x1-z1 |(α3) sign(x1-z1); Among them, x1 hat , x2 hat , x3hat are the estimated values ​​of x1, x2, and x3, respectively; z1, z2, and z3 are auxiliary state variables; β1, β2, and β3 are positive design parameters used to adjust the convergence speed; α1, α2, and α3 are constants satisfying 0<α1<α2<1<α3, used to ensure the finite time convergence of the observer. The sliding surface s is constructed based on the equation, and the specific form is s=c1(x1-x1 hat )+c2(x2-x2 hat )+(x3-x3 hat ), where c1 and c2 are positive constants, which are used to adjust the convergence characteristics of the sliding surface. In order to ensure the stability of the entire system, the Lyapunov function V=(1 / 2)s² is constructed, and its derivative dV / dt is calculated to ensure that dV / dt<0. The satisfaction of this condition ensures the stability of the system, that is, the observer can converge stably. The Lyapunov function is used as an energy function in sliding mode control, which describes the energy characteristics of the system state over time. By solving its derivative, it is determined whether the system converges. Based on the stability condition, β1, β2, and β3 are iteratively optimized until the optimal parameter value β1 is found opt , β2 opt , β3 opt , so that the sliding mode compensation function observer reaches the best state in terms of interference estimation performance. After the parameter optimization is completed, the target touch signal x1 is input into the optimized sliding mode compensation function observer, and the observer obtains the state estimation value x1 through iterative calculation. hat , x2 hat and x3 hat , where x3 hat This is the preliminary estimated false touch interference value. Using Kalman filtering, the preliminary estimated false touch interference value x3 hat Noise suppression and smoothing are performed. In Kalman filtering, the state transfer matrix F and the observation matrix H are pre-set according to the characteristics of the system. Kalman filtering can reduce the random noise in the system and obtain a smoother interference estimate x3 smooth . Set the corrected false touch interference value x3 smooth Compared with the preset interference threshold Th, when x3 smooth When the threshold value Th is exceeded, the false touch interference information is set to 1, indicating that a false touch interference has occurred; otherwise, the false touch interference information is set to 0, indicating that there is no false touch interference. The threshold value Th is set in advance according to the characteristics of the touch screen and the user's usage habits to ensure that the system can accurately determine the occurrence of false touch events in actual applications, thereby effectively preventing erroneous operations caused by false touches.

[0016] The target touch signal x1 is sampled to generate a discrete time series x1[n], where n represents the discrete time index. Based on the discrete time series, the central difference method is used to calculate the discrete derivative of x1, that is, the rate of change of the touch signal x2[n], and its calculation formula is x2[n] = (x1[n+1]- x1[n-1]) / (2Δt), where Δt is the sampling time interval. This derivative x2[n] reflects the rate of change of the touch signal at time n. Substitute x1[n] and x2[n] into the optimized sliding mode compensation function observer equation group, and initialize the auxiliary state variables z1[0], z2[0] and z3[0] to zero vectors to provide initial conditions for subsequent iterative calculations. In order to enable the observer to operate in discrete time, the observer equation group is discretized to obtain the discrete time observer equation. The specific discretization formula is: z1[n+1]=z1[n]+Δt*(z2[n]+β1|x1[n]-z1[n]| (α1) sign(x1[n]-z1[n])); z2[n+1]=z2[n]+Δt*(z3[n]+β2|x1[n]-z1[n]| (α2) sign(x1[n]-z1[n])); z3[n+1]=z3[n]+Δt*β3|x1[n]-z1[n]| (α3) sign(x1[n]-z1[n]); Among them, β1, β2 and β3 are positive design parameters used to adjust the convergence speed, while α1, α2 and α3 are constants that satisfy 0<α1<α2<1<α3, ensuring that the observer can converge within a finite time. Through these formulas, the time series of z1[n], z2[n] and z3[n] are iteratively calculated. As time goes by, the estimated values ​​of these auxiliary state variables gradually converge. According to the output equation of the sliding mode compensation function observer, the state estimate is calculated. The specific calculation formula is: x1 hat [n]=z1[n]+β1|x1[n]-z1[n]| (α1) sign(x1[n]-z1[n]); x2 hat [n]=z2[n]+β2|x1[n]-z1[n]| (α2) sign(x1[n]-z1[n]); x3 hat [n]=z3[n]+β3|x1[n]-z1[n]| (α3) sign(x1[n]-z1[n]); Among them, x1 hat[n], x2 hat [n], x3 hat [n] represents the estimated values ​​of touch signal, signal change rate and false touch interference at discrete time index n. Through these estimated values, preliminary estimates of touch signal and false touch interference are gradually obtained. In order to ensure the accuracy of the estimation, the state estimation value x1 hat [n], x2 hat [n], x3 hat [n] Perform convergence analysis and calculate each estimation error: e1[n]=x1[n]-x1 hat [n]、e2[n]=x2[n]-x2 hat [n]、e3[n]=x3[n]-x3 hat [n], these errors represent the estimated errors of the touch signal, signal change rate, and false touch interference, respectively. In order to quantify the overall size of the error, the root mean square error (RMSE) is calculated based on the estimated errors e1[n], e2[n], and e3[n], reflecting the overall error level of the observer at each time step. When the RMSE is less than the preset threshold ε or the number of iterations reaches the maximum value Nmax, the observer stops iterating and outputs the final state estimate x1 hat , x2 hat and x3 hat , where RMSE represents the root mean square of the estimation error, ε is the preset convergence threshold, and Nmax is the maximum number of iterations allowed.

[0017] Step 300: extract features of the target touch signal according to the false touch interference information to obtain a touch feature vector; It should be noted that the target touch signal is analyzed in the time domain to extract the maximum, minimum and average values ​​of the touch pressure values, describe the range of change and average strength of the touch pressure, and obtain the touch pressure characteristics. Based on the target touch signal, the touch duration and touch frequency are calculated to obtain the time characteristics. The touch duration reflects the length of time the user touches the screen, while the touch frequency reflects the repetition rate of the touch. The target touch signal is transformed in the frequency domain to extract the main frequency components and energy distribution of the signal to obtain the frequency domain characteristics, which reflect the vibration characteristics and interference characteristics in the touch signal. Based on the touch position coordinates in the target touch signal, the curvature and speed of the touch trajectory are calculated to obtain the spatial characteristics. The curvature of the touch trajectory can characterize the changing trend of the user's touch path, while the touch speed reflects the fast and slow changes of the touch. Therefore, the spatial characteristics can describe the geometric characteristics and dynamic characteristics of the touch action. Based on the false touch interference information, the interference intensity and duration are calculated to measure the degree of external interference and its impact on the touch signal. The interference intensity indicates the strength of the false touch, and the interference duration reflects the length of time the interference exists. Perform wavelet transform on the target touch signal to extract the time-frequency characteristics of the signal at multiple scales. Wavelet transform can provide detailed frequency information of the touch signal at different scales and times. The wavelet coefficients can capture the subtle changes in the signal and enhance the system's ability to identify false touches. The touch pressure features, time features, frequency domain features, spatial features, interference features and wavelet coefficients are combined to construct the initial feature vector. In order to improve computational efficiency and reduce redundant information, principal component analysis is performed on the initial feature vector. Several principal components with the largest contribution rate are selected through dimensionality reduction, and those components that are irrelevant to the final classification and processing are eliminated. Generate a touch feature vector with strong discrimination ability.

[0018] Step 400: input the touch feature vector into a preset touch signal classification model for processing to obtain a touch signal classification result; Specifically, a multi-scale time-frequency transformation is performed on the touch feature vector to obtain a multi-scale time-frequency feature matrix. This matrix can capture the characteristics of the touch signal at different time and frequency scales. The multi-scale time-frequency feature matrix is ​​input into a two-stream attention-enhanced convolutional neural network. The convolutional neural network with a two-stream structure contains two independent branches, the time stream and the frequency stream. Each branch consists of three convolutional layers, and the convolution kernel sizes are 3x3, 3x3 and 5x5 respectively. These convolutional layers extract local features in the time and frequency streams by sliding convolution kernels of different sizes to form convolution output feature maps. The convolution output feature maps of the time stream and the frequency stream are processed by the channel attention mechanism. The channel attention mechanism assigns weights to different feature channels, strengthens those more important features, weakens irrelevant or redundant features, and obtains a weighted time-frequency feature representation. The weighted time-frequency feature representation is input into a cross-modal self-attention module to capture the long-range dependency between the time domain and the frequency domain. Through this interaction, a deeper feature fusion is achieved to obtain a fused feature. The fused feature is residually connected and layer normalized to construct a residual attention block. Residual connection allows the network to retain part of the original information during information transmission, avoiding the problem of gradient disappearance. At the same time, layer normalization accelerates the training process of the model and improves the generalization ability. The residual attention block stacks three layers, each of which extracts deep feature representations through residual structure and attention mechanism, which can better characterize the complex characteristics of touch signals. The deep feature representation is input into the global average pooling layer for processing, and the features of each channel are mapped to a feature vector of fixed dimension, thereby simplifying the structure of the network and reducing the risk of overfitting of the model to obtain a feature vector of fixed dimension. The feature vector of fixed dimension is input into the multi-head self-calibration classifier for classification. The classifier consists of three classification heads, each of which outputs two categories of probability distribution. Weighted voting is performed on the set of category probability distributions output by the three classification heads, and the classification result with the highest weight is selected as the final touch signal classification result.

[0019] The multi-scale time-frequency feature matrix is ​​decomposed into a time feature matrix and a frequency feature matrix. The time feature matrix is ​​subjected to the first convolution operation, and the convolution kernel size used is 3x3. This operation can extract local features in the time stream and obtain the first convolution layer output feature map of the time stream. At the same time, the frequency feature matrix is ​​subjected to the first convolution operation, and the convolution kernel size is 3x3 to extract local features in the frequency stream and obtain the first convolution layer output feature map of the frequency stream. The output feature map of the first convolution layer of the time stream is input into the batch normalization layer for normalization to speed up the convergence of the model and reduce overfitting during training. The normalized time stream feature map is input into the ReLU activation function layer for nonlinear activation to generate the first activation feature map of the time stream. Similarly, the output feature map of the first convolution layer of the frequency stream passes through the batch normalization layer and the ReLU activation function layer in sequence to obtain the first activation feature map of the frequency stream. The role of the activation function is to introduce nonlinearity so that the network can learn more complex feature representations. The first activation feature map of the time stream is convolved for the second time, with a convolution kernel size of 3x3, to further extract deeper features in the time stream and generate the second convolution layer output feature map of the time stream. The first activation feature map of the frequency stream is convolved for the second time through the same convolution kernel to extract more detailed frequency domain features in the frequency stream and output the second convolution layer output feature map of the frequency stream. The second convolution layer output feature maps of the time stream and the frequency stream are input to the batch normalization layer for normalization again, and are nonlinearly activated through the ReLU activation function layer to generate the second activation feature map of the time stream and the second activation feature map of the frequency stream. The second activation feature map of the time stream is convolved for the third time, with a convolution kernel size of 5x5. The 5x5 convolution kernel can cover a larger receptive field and extract more global features, generating the third convolution layer output feature map of the time stream. The second activation feature map of the frequency stream is convolved for the third time through the same size 5x5 convolution kernel to obtain the third convolution layer output feature map of the frequency stream. Output the third convolution layer output feature map of the time stream and the frequency stream.

[0020] Step 500: Perform statistical analysis on the touch signal classification results within the sliding time window, and determine the valid touch input event; Specifically, the size and step length of the sliding time window are set. The size of the sliding time window determines the time range of each analysis, while the step length determines the speed of the window movement. The classification results of the touch signal are segmented and assigned to different time windows to form a time window sequence. The touch signal classification results in each time window are counted and counted to calculate the frequency of occurrence of each category (such as valid touch or false touch). Based on the frequency of occurrence of each category, the proportion of the valid touch category in each time window is calculated to obtain a valid touch ratio sequence, which reflects the fluctuation of the proportion of the valid touch signal in the overall signal in different time windows. A trigger threshold is set for the valid touch ratio sequence. When the proportion of the valid touch signal in a certain time window exceeds the trigger threshold, the system marks the time window as a potential valid touch window. The setting of the threshold is adjusted according to the specific touch screen application scenario and the user's operating habits to accurately capture the valid touch operation. The continuous potential valid touch windows are merged to obtain a candidate touch event segment. The candidate event segment represents a series of possible valid touch inputs. The duration of each candidate touch event segment is calculated and compared with the preset minimum valid touch duration. Only those candidate event segments that last longer than the minimum valid touch time will be considered as true valid touch event segments, avoiding mistaking short false touches or invalid touches for valid touches. Feature parameters are extracted from valid touch event segments, including start time, end time, average touch pressure, and touch trajectory. The extracted feature parameters are matched with the predefined touch pattern, which is a standard pattern pre-set based on the user's operating habits and common interactive behaviors. When the extracted touch event segment matches the predefined touch pattern to the preset standard, the touch event is confirmed to be a valid touch input, and the valid touch input event is finally output.

[0021] Step 600: Process the touch input signal and generate a touch response signal of the touch screen according to the valid touch input event and the touch signal classification result.

[0022] Specifically, the touch input signals determined as valid touch events are tracked and analyzed to ensure that these touch signals are indeed derived from the user's intended operation, rather than caused by false touches or other interference factors. Combined with the touch signal classification results, the distinction between valid touch categories and false touch categories helps the system make dynamic decisions when processing input signals. If some touch signals are misclassified as valid touches, they are reclassified as false touches by comprehensively analyzing their time, pressure, trajectory and other parameters to avoid incorrect responses. At the same time, the nature of the touch is evaluated based on the characteristic parameters of the touch signal (such as touch pressure, trajectory and duration). For example, a large touch pressure represents a strong user operation intention, while a complex trajectory indicates a specific gesture. In the process of processing touch signals, the response logic is set according to the specific application scenario of the touch screen. For example, in a game application, the user's touch may require a fast and sensitive response, while in a text input scenario, the touch may require de-jittering and filtering to avoid false operations. Based on different scenarios, according to the characteristics of the touch input event, a corresponding response strategy is generated. The strategy determines the feedback method of the touch screen, such as whether a virtual button needs to be popped up, a picture needs to be scaled, or a character needs to be moved in a game. When the touch input signal is processed and the response strategy is determined, a corresponding touch response signal is generated. The touch response signal includes relevant parameters of touch feedback, such as the coordinate position, pressure value, operation type, etc. of the response, and generates corresponding screen operations according to the user's input. This response signal is transmitted to the touch screen driver layer, prompting the screen to make corresponding feedback, such as displaying the corresponding operation effect, executing the action on the screen, or triggering a specific function.

[0023] In the embodiment of the present application, by filtering and normalizing the touch input signal, combined with the Butterworth low-pass filter and outlier removal techniques, the quality of the original touch data is effectively improved, and the sliding mode compensation function observer is used to estimate the touch signal interference, which can capture various forms of false touch interference in real time and accurately, and improve the system's adaptability to complex touch scenes. Touch features are extracted by various methods such as time domain analysis, frequency domain transformation and wavelet transformation, and the principal component analysis is combined for dimensionality reduction, which not only retains the key information of the touch signal, but also uses a dual-stream attention enhanced convolutional neural network and a multi-head self-calibration classifier to make full use of the advantages of deep learning in feature extraction and classification tasks, and improve the accuracy of touch signal classification. Through statistical analysis and multiple judgment criteria within the sliding time window, the probability of false touch and misjudgment is effectively reduced, and the stability and reliability of the system are improved. The present invention can dynamically adjust the feature extraction and classification strategy according to the false touch interference information, and has good environmental adaptability and user personalization capabilities.

[0024] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Perform high-frequency scanning on the touch screen surface to collect raw touch data including touch position coordinates, touch pressure value, touch duration and touch frequency; Inputting the original touch data into a Butterworth low-pass filter to filter out high-frequency noise to obtain preliminary filtered touch data, and removing outliers and impulse noise from the preliminary filtered touch data to obtain noise-reduced touch data; The touch pressure value in the noise reduction touch data is converted into a range of 0-1 by a linear mapping function to obtain a normalized touch pressure value, and the touch duration in the noise reduction touch data is logarithmically transformed to obtain a touch duration after logarithmic transformation; The touch frequency in the noise reduction touch data is processed by exponential smoothing to obtain the smoothed touch frequency, and the touch position coordinates in the noise reduction touch data are trajectory smoothed and predicted to obtain the optimized touch position coordinates; The normalized touch pressure value, the touch duration after logarithmic transformation, the touch frequency after smoothing, and the optimized touch position coordinates are combined to obtain a target touch signal.

[0025] Specifically, the touch screen surface is scanned at high frequency to quickly and accurately capture the user's touch operation on the touch screen, including the coordinate position of the touch, the pressure applied during the touch, the duration of the touch, and the frequency of the user's repeated touches, to form an original touch data matrix, which includes the position coordinates (x, y), the pressure value P, the duration T, and the touch frequency f. The original touch data is input into the Butterworth low-pass filter for processing. The Butterworth filter is a filter with good smoothness. Its design feature is that the frequency response is very flat near the cutoff frequency, so it can effectively filter out high-frequency noise while retaining the main signal characteristics. The transfer function of the Butterworth low-pass filter is expressed as: ; in, is the transfer function of the filter, is the complex frequency, is the cutoff frequency of the filter, is the order of the filter. By choosing a suitable cutoff frequency , effectively remove high-frequency noise in the touch signal, retain the main frequency components of the touch operation, and obtain preliminary filtered touch data. Remove outliers and impulse noise from the touch data after preliminary filtering. Outliers usually refer to abnormal values ​​that deviate significantly from other data points, caused by signal interference or sensor errors. Impulse noise is a short, sudden, high-amplitude noise that comes from electromagnetic interference or other external factors in the environment. Through statistical analysis and signal processing algorithms, these abnormal data points are detected and removed to generate cleaner noise-reduced touch data. The touch pressure value in the noise-reduced touch data is converted to the range of 0-1 through a linear mapping function, and the value of the touch pressure is standardized to a unified scale for subsequent processing. At the same time, the touch duration in the noise-reduced touch data is logarithmically transformed to reduce its numerical range and enhance the processability of the data. The formula for logarithmic transformation is: ; in, is the touch duration after logarithmic transformation, is the original touch duration, and +1 is to avoid the problem of logarithmic function processing 0 value. Through logarithmic transformation, the duration is compressed to a smaller value range while retaining its main characteristics. For the touch frequency f, exponential smoothing is used. The touch frequency represents the repeatability of the user's touch, and exponential smoothing can smooth out short-term fluctuations and make the signal more stable. The formula for exponential smoothing is: ; in, is the smooth touch frequency at the nth moment, is the original touch frequency, is a smoothing coefficient (usually between 0 and 1) that controls the intensity of the smoothing effect. Through this method, the fluctuation of the touch frequency is reduced, and a more stable frequency value is obtained. The touch position coordinates (x, y) are smoothed and predicted to eliminate noise and improve the accuracy of the touch trajectory. Trajectory smoothing usually uses a filter or regression model, and prediction calculates future position points through the model to optimize the touch trajectory. The optimized touch position coordinates obtained can more accurately reflect the user's touch path and reduce errors such as hand shaking.

[0026] In a specific embodiment, the process of executing step 200 may specifically include the following steps: The target touch signal is represented as x1, the first-order derivative of the target touch signal x1 is obtained to obtain x2, the false touch interference variable x3 is introduced, and the system state vector x=[x1, x2, x3] is constructed. T , where x1 represents the touch signal amplitude, x2 represents the touch signal change rate, and x3 represents the false touch interference amount; According to the system state vector x, the sliding mode compensation function observer equation group is constructed: x1 hat =z1+β1|x1-z1| (α1) sign(x1-z1); x2 hat =z2+β2|x1-z1| (α2) sign(x1-z1); x3 hat =z3+β3|x1-z1 |(α3) sign(x1-z1); Among them, x1 hat , x2 hat , x3 hat are the estimated values ​​of x1, x2, and x3, respectively; z1, z2, and z3 are auxiliary state variables; β1, β2, and β3 are positive design parameters used to adjust the convergence speed; α1, α2, and α3 are constants satisfying 0<α1<α2<1<α3, used to ensure the finite time convergence of the observer; Construct the sliding surface s=c1(x1-x1 hat )+c2(x2-x2 hat )+(x3-x3 hat ), where c1 and c2 are positive constants used to adjust the convergence characteristics of the sliding surface; Construct Lyapunov function V=(1 / 2)s 2 , calculate its derivative dV / dt, and ensure that dV / dt<0 to meet the system stability condition, where V represents the energy function of the system; Based on the stability condition, the parameters β1, β2, and β3 are iteratively optimized to obtain the optimal parameter value β1 opt , β2 opt , β3 opt , so that the observer has the best disturbance estimation performance; The target touch signal x1 is input into the optimized sliding mode compensation function observer, and the state estimation value x1 is obtained by iterative calculation. hat , x2 hat and x3 hat , where x3 hat This is the preliminary estimated false touch interference value; Preliminary estimated false touch interference value x3 hat Perform noise suppression and smoothing to obtain the corrected false touch interference value x3 smooth , where the state transfer matrix F and the observation matrix H of the Kalman filter are preset according to the system characteristics; Set the corrected false touch interference value x3 smooth Compared with the preset interference threshold Th, when x3 smooth>Th, the false touch interference information is set to 1, otherwise it is set to 0 to obtain the false touch interference information, where Th is a false touch judgment threshold predetermined according to the touch screen characteristics and user habits.

[0027] Specifically, the target touch signal is represented as ,in Indicates the amplitude of the touch signal, reflecting the intensity or pressure of the touch operation. Take the first-order derivative of the signal to get the rate of change of the signal . Rate of change Describes the rate of change of the touch signal over time. For example, a rapid change in pressure indicates the suddenness or instability of the touch gesture. In order to distinguish between valid touch and false touch operations, the false touch interference variable is introduced. , which is used to represent the amount of interference that may be introduced during the touch process. Through these three variables, a system state vector is constructed , which integrates the amplitude, rate of change and false touch interference of the touch signal. The equation group of the sliding mode compensation function observer is constructed based on the state vector. The main purpose of the observer is to estimate the true state of the touch signal and distinguish false touch interference from normal touch. The basic equation group of the observer is as follows: ; ; ; in, are the touch signal amplitudes , rate of change and false touch interference The estimated value of These are auxiliary state variables that are adjusted according to changes in touch signals. is a positive design parameter used to adjust the convergence speed of the observer. is a constant used to ensure that the observer can converge in a finite time, satisfying Through this design, the observer can approach the real touch signal state at a faster speed while ensuring stability. In order to control the convergence and stability of the system, the sliding surface equation is constructed. The purpose of the sliding surface is to establish a feedback relationship between the state estimation and the real signal, so as to adjust the system state. The sliding surface equation is as follows: ; in, is the sliding surface, and is a constant used to adjust the convergence characteristics of the sliding surface. The sliding surface provides a feedback control signal by combining the error of the signal amplitude, rate of change and disturbance to ensure that the observer can converge to the correct estimate as quickly as possible. In order to ensure the stability of the system, the Lyapunov function is introduced, which is in the form of: ; Lyapunov function Represents the energy function of the system, describing the degree to which the system deviates from the stable state. By calculating its derivative, it is ensured that the system is stable in time. To meet the stability condition, ensure that dv / dt , that is, the energy of the system gradually decreases over time, thus achieving stable convergence. In order to optimize the performance of the observer, the parameters Perform iterative optimization. By continuously adjusting these parameters, we can find the optimal , so that the observer has the best interference estimation performance. The optimization process is implemented through gradient descent or other optimization algorithms to ensure that the observer can operate effectively in different touch scenarios. After the optimization is completed, the target touch signal Input into the optimized sliding mode compensation function observer, and obtain the state estimation value through iterative calculation .in, It is the preliminary estimated false touch interference value, which reflects the possibility of false touch in the touch signal. The Kalman filter is used to suppress noise and smooth the preliminary estimated false touch interference value. Kalman filtering is a state estimation method that can optimize state variables in the presence of noise. The state transfer matrix of Kalman filtering is and the observation matrix Preset according to the characteristics of the system to ensure that the filter can accurately track the changes in the false touch interference. Through Kalman filtering, the corrected false touch interference value is obtained. Compare the corrected false touch interference value with the preset interference threshold For comparison. When the current touch behavior is judged as false touch interference, the false touch interference information is set to 1; otherwise, the false touch interference information is set to 0, indicating that the current touch is a valid operation. Threshold It is pre-set according to the characteristics of the touch screen and the user's operating habits to ensure the accuracy of false touch judgment.

[0028] In a specific embodiment, the step of inputting the target touch signal x1 into the optimized sliding mode compensation function observer is executed, and the state estimation value x1 is obtained by iterative calculation. hat , x2 hat and x3 hat , where x3 hat That is, the process of preliminarily estimating the false touch interference value may specifically include the following steps: The target touch signal x1 is sampled to obtain a discrete time series x1[n], where n represents a discrete time index; Based on the discrete time series x1[n], the discrete derivative x2[n]=(x1[n+1]-x1[n-1]) / (2Δt) of x1 is calculated by the central difference method, where Δt is the sampling time interval and x2[n] represents the rate of change of the touch signal at time n. Substitute x1[n] and x2[n] into the optimized sliding mode compensation function observer equations, initialize z1[0], z2[0], z3[0] as zero vectors, where z1, z2, z3 are auxiliary state variables; Discretize the observer equations to obtain the discrete-time observer equation: z1[n+1]=z1[n]+Δt*(z2[n]+β1|x1[n]-z1[n]| (α1) sign(x1[n]-z1[n])); z2[n+1]=z2[n]+Δt*(z3[n]+β2|x1[n]-z1[n]| (α2) sign(x1[n]-z1[n])); z3[n+1]=z3[n]+Δt*β3|x1[n]-z1[n]| (α3) sign(x1[n]-z1[n]); Among them, β1, β2, and β3 are positive design parameters used to adjust the convergence speed, and α1, α2, and α3 are constants satisfying 0<α1<α2<1<α3, which are used to ensure the finite time convergence of the observer; Based on the discrete time observer equation, the time series of z1[n], z2[n], and z3[n] are obtained through iterative calculation, where n represents the discrete time index; According to the output equation of the sliding mode compensation function observer, the state estimation value is calculated: x1 hat [n]=z1[n]+β1|x1[n]-z1[n]| (α1) sign(x1[n]-z1[n]); x2 hat [n]=z2[n]+β2|x1[n]-z1[n]| (α2) sign(x1[n]-z1[n]); x3 hat [n]=z3[n]+β3|x1[n]-z1[n]| (α3) sign(x1[n]-z1[n]); Among them, x1 hat [n], x2hat [n], x3 hat [n] represents the estimated values ​​of touch signal, signal change rate and false touch interference at discrete time index n respectively; Estimated value of state x1 hat [n], x2 hat [n], x3 hat [n] performs convergence analysis and calculates the estimated error e1[n]=x1[n]-x1 hat [n]、e2[n]=x2[n]-x2 hat [n]、e3[n]=x3[n]-x3 hat [n], where e1[n], e2[n], and e3[n] represent the estimated errors of touch signal, signal change rate, and false touch interference, respectively; Based on the estimated errors e1[n], e2[n], and e3[n], the root mean square error RMSE is calculated. When the RMSE is less than the preset threshold ε or the number of iterations reaches the maximum value Nmax, the final state estimate x1 is output. hat , x2 hat and x3 hat , where RMSE represents the root mean square of the estimation error, ε is the preset convergence threshold, and Nmax is the maximum number of iterations.

[0029] Specifically, the continuous touch signal is discretized. By sampling the touch signal, the continuous time series is converted into a discrete time series. The sampling interval is set to , the sampled discrete-time signal is ,in Represents a discrete time index. Based on sampling , the discrete derivative of the signal is calculated by the central difference method to obtain the rate of change of the touch signal The central difference method is a method for numerically calculating derivatives. It approximates the derivative by calculating at the points before and after the current time step. The formula is: ; in, Indicates that the touch signal is The rate of change at a time point, is the sampling time interval. By this method, the change rate of the touch signal is estimated. and Substitute into the optimized sliding mode compensation function observer equations and initialize the auxiliary state variables is a zero vector. Auxiliary state variables are intermediate variables used in the observer to estimate the signal state. As time goes by, they will be gradually adjusted to approach the true signal state. In order to adapt to the processing of discrete time, the sliding mode compensation function observer equations are discretized to obtain the discrete time observer equations. By discretizing the original equations, the following discrete time recursive formula is obtained: ; ; ; in, is a positive design parameter, which is mainly used to adjust the convergence speed of the observer; are constants that ensure that the observer converges in finite time, and they satisfy the condition These parameters control the update rate of each state variable, so that the observer can estimate the true state of the touch signal within a reasonable time. Based on the discrete time observer equation, the state variables are gradually obtained through iterative calculation. The update of state variables is an estimation process that gradually approaches the real touch signal over time. The state estimation value of each time step is calculated according to the output equation of the sliding mode compensation function observer. The specific calculation formula is as follows: ; ; ; in, , and Respectively represent the discrete time index The estimated values ​​of the touch signal amplitude, signal change rate and false touch interference are obtained. In order to ensure the accuracy and stability of the estimation, the convergence analysis of the state estimation is performed. Calculate the estimation error: ; ; ; in, , and They represent the estimated errors of touch signal amplitude, signal change rate and false touch interference respectively. Through error analysis, the convergence of the observer is judged. The root mean square error (RMSE) is calculated based on the estimated error, and the formula is as follows: ; in, is the total number of time steps, and RMSE represents the root mean square value of the estimation error, which is a common indicator for measuring the accuracy of observer estimation. By calculating RMSE, it is dynamically evaluated whether the current estimation has achieved sufficient accuracy. When RMSE is less than the preset convergence threshold When , or the number of iterations reaches the maximum value When , the system stops iterating and outputs the final state estimate These final estimated values ​​can accurately reflect the actual state of the touch signal and the interference caused by false touches, providing an important basis for the anti-false touch function of the touch screen.

[0030] In a specific embodiment, the process of executing step 300 may specifically include the following steps: Perform time domain analysis on the target touch signal, extract the maximum, minimum and average values ​​of the touch pressure value, and obtain the touch pressure characteristics; Based on the target touch signal, the touch duration and touch frequency are calculated to obtain the time feature, and the target touch signal is transformed into the frequency domain to extract the main frequency components and energy distribution of the signal to obtain the frequency domain feature; Based on the touch position coordinates in the target touch signal, the curvature and speed of the touch track are calculated to obtain the spatial features, and based on the false touch interference information, the interference intensity and duration are calculated to obtain the interference features; Perform wavelet transform on the target touch signal, extract multi-scale time-frequency features, and obtain wavelet coefficients; Constructing an initial feature vector based on touch pressure features, time features, frequency domain features, spatial features, interference features and wavelet coefficients; Perform principal component analysis on the initial feature vector, select several principal components with the largest contribution rate, and obtain the touch feature vector.

[0031] Specifically, the touch signal is analyzed in the time domain to extract the key features of the touch pressure value. Touch pressure is one of the direct interaction signals between the user and the touch screen. The change in pressure can reflect the intensity of the touch and the intention of the operation. In the time domain analysis, the maximum, minimum and average values ​​of the touch pressure value are extracted from the touch signal. The maximum pressure value indicates the maximum force applied by the user during the touch process, the minimum pressure value reflects the lightest touch, and the average pressure value represents the typical pressure intensity during the entire touch process. By extracting these values, the touch pressure feature is constructed. The touch duration and touch frequency are calculated based on the target touch signal to obtain the time feature. The touch duration refers to the length of time the user's finger stays on the screen, which is used to measure the continuity of the touch operation. For example, a long press operation often corresponds to a longer touch duration. The touch frequency indicates the number of repeated touches by the user within a certain period of time. A higher touch frequency indicates that the user performs a quick operation or multiple clicks. Through these two time features, the user's operation mode can be analyzed more accurately. In order to obtain the characteristics of the touch signal in the frequency domain, the target touch signal is transformed in the frequency domain to extract the main frequency components and energy distribution of the touch signal, revealing the vibration or periodic changes in the touch signal, and helping the system identify unstable operations such as false touches. Discrete Fourier transform is a common method of frequency domain transformation. It converts the time domain signal into the frequency domain and extracts the frequency component of the signal. The specific formula is as follows: ; in, It is The amplitude of the frequency component, For the The signal value at a time point, is the total length of the signal. Through this formula, the amplitude of each frequency component in the touch signal and its corresponding energy distribution are calculated. Based on the touch position coordinates in the target touch signal, the curvature and speed of the touch trajectory are calculated to extract spatial features. The touch trajectory reflects the path of the user's finger sliding on the screen. The calculation of the curvature can reveal the degree of curvature of the touch path, while the speed reflects the speed of the user's sliding. The curvature of the touch trajectory is usually calculated by analyzing the change in the tangent direction of the trajectory at different points, while the speed is obtained by dividing the displacement of the trajectory by the time interval. The speed calculation formula is: ; in, Indicates the touch speed, is the displacement between touch points, is the time interval. Through these spatial features, the geometric characteristics of the touch action are understood. For example, a trajectory with a large curvature indicates that the user is performing a complex curved sliding operation, while a high speed means that the user is sliding quickly. Assuming that the curvature of a touch is 0.5 and the speed is 300 pixels per second, this indicates that the user has performed a relatively smooth and fast sliding operation. According to the false touch interference information, the interference characteristics are calculated. Interference intensity and interference duration are important indicators for evaluating false touches. Interference intensity is estimated by analyzing the amplitude change of false touches in the touch signal, while interference duration indicates the length of time the interference signal exists. For example, a long period of high interference intensity may indicate that the user's palm accidentally touches the screen. Through these interference characteristics, invalid touch signals are filtered out and the anti-false touch function is enhanced. When performing multi-scale analysis on touch signals, wavelet transform is used to extract multi-scale time-frequency features. Wavelet transform is a tool that can simultaneously analyze the time and frequency characteristics of a signal and is suitable for processing non-stationary signals. Through wavelet transform, the local characteristics of the touch signal are analyzed at different time scales and the wavelet coefficients are extracted. The formula for wavelet transform is: ; in, is the wavelet coefficient, For touch signal, is the wavelet basis function, is the scale parameter, is the time translation parameter. By analyzing the wavelet coefficients, the feature changes on different time scales in the touch signal are captured. The touch pressure features, time features, frequency domain features, spatial features, interference features and wavelet coefficients are combined to construct the initial feature vector. The initial feature vector is subjected to principal component analysis, and the high-dimensional features are projected into a low-dimensional space through linear transformation, while retaining the main information of the original data as much as possible. Through principal component analysis, several principal components with the largest contribution rate are selected to obtain the touch feature vector. Assuming that the dimension of the initial feature vector is 100, the first 10 principal components are retained after dimensionality reduction through principal component analysis. These principal components occupy more than 90% of the information of the original features, thereby greatly reducing the computational complexity while retaining the main features.

[0032] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Transform the touch feature vector to obtain a multi-scale time-frequency feature matrix; The multi-scale time-frequency feature matrix is ​​input into the two-stream attention-enhanced convolutional neural network in the touch signal classification model to obtain the convolution output feature maps of the time stream and frequency stream, where the time stream and frequency stream contain three convolution layers respectively, and the convolution kernel sizes are 3x3, 3x3 and 5x5 respectively; The convolution output feature maps of the time stream and frequency stream are processed by the channel attention mechanism to obtain the weighted time-frequency feature representation; The weighted time-frequency feature representation is input into the cross-modal self-attention module to capture the long-range dependency between the time and frequency domains and obtain the fused feature; Perform residual connection and layer normalization on the fused features, build residual attention blocks, and extract deep feature representations. The residual attention blocks are stacked in 3 layers. The deep feature representation is input into the global average pooling layer for processing to obtain a feature vector of fixed dimension; The fixed-dimensional feature vector is input into the multi-head self-calibration classifier to obtain a set of category probability distributions. The multi-head self-calibration classifier is set with three classification heads, and each classification head outputs the probability distribution of two categories. A weighted vote is performed on the category probability distribution set to obtain the touch signal classification result.

[0033] Specifically, the touch feature vector is transformed to analyze the touch signal at different time and frequency scales. The features in the touch signal are presented in different ways in the time and frequency domains, and time-frequency transformation of the touch feature vector can help capture these changes. Using a multi-scale wavelet transform, the feature vector is decomposed to obtain frequency domain features at different time scales, and a multi-scale time-frequency feature matrix is ​​constructed, which contains the features of the touch signal in multiple time windows and frequency ranges. The multi-scale time-frequency feature matrix is ​​input into the two-stream attention enhanced convolutional neural network in the touch signal classification model. The design of the two-stream convolutional neural network includes two branches, one branch processes the time stream features, and the other branch processes the frequency stream features. Each stream consists of three convolutional layers, where the convolution kernel size of the first two convolutional layers is , and the convolution kernel size of the third convolution layer is Smaller convolution kernels capture local features, while larger convolution kernels can capture more global features. Through these convolution layers, the network extracts features in the time and frequency domains and generates convolution output feature maps of the time stream and frequency stream. Assume that the input multi-scale time-frequency feature matrix is , the output after the first layer of convolution operation is: ; in, Represents a 3x3 convolution operation, and ReLU is the activation function used to introduce nonlinearity. After the first layer of convolution captures the local time-frequency features, it passes through the second and third layers of convolution and finally outputs the feature maps of the time stream and frequency stream. After the convolution layer is processed, the convolution output feature maps of the time stream and frequency stream are processed by the channel attention mechanism. The channel attention mechanism highlights key features and weakens unimportant information by assigning weights to different channels, thereby weighting the time-frequency features and enhancing the classification model's perception of important features. The output of the channel attention mechanism is expressed as: ; in, The weighted time-frequency feature representation is used. The Attention operation performs weighted processing according to the importance of the channel features. The weighted time-frequency feature representation is input into the cross-modal self-attention module to capture the long-range dependency between the time domain and the frequency domain. The cross-modal self-attention module can fuse the features of the time and frequency domains by calculating the mutual influence between different modalities to generate a more comprehensive fusion feature. The calculation of cross-modal self-attention is expressed as: ; in, For the fused features, the self-attention mechanism extracts long-range dependency features by calculating the correlation between time and frequency streams. The fused features are processed with residual connections and layer normalization to construct residual attention blocks. The role of residual connections is to retain the original feature information to prevent the gradient vanishing problem in deep networks, while layer normalization helps to accelerate the convergence process of the network and stabilize model training. The stacking of residual attention blocks can gradually extract deeper feature representations while retaining the original information. The specific formula is: ; in, is the deep feature representation after residual connection and normalization. The deep feature representation is input to the global average pooling layer for processing to convert it into a feature vector of fixed dimension. The function of the global average pooling layer is to take the average of the feature map of each channel to generate a feature vector with fixed dimension. This operation can effectively reduce the dimension of the feature map and retain the global information. Assume that the input deep feature representation is , the output of global average pooling is: ; in, and is the height and width of the feature map, is a fixed-dimensional feature vector after global pooling. The fixed-dimensional feature vector is input into the multi-head self-calibration classifier. The classifier contains three classification heads, each of which outputs the probability distribution of two categories. The multi-head self-calibration classifier uses multiple independent classification heads to judge the features from different angles to improve the accuracy of classification. The output of the classifier is expressed as: ; in, For the The output probability distribution of the classification head is and are the weights and biases of the classifier, and the Softmax function is used to convert the output into a probability distribution. Weighted voting is performed on the category probability distribution set output by the three classification heads, and the judgment results of different classification heads are combined to generate the final classification result. The output of each classification head is assigned different weights according to its accuracy, and the classification result of the touch signal is finally obtained. The specific calculation formula is: ; in, is the final classification probability distribution, For the The weight of each classification head is dynamically adjusted according to the performance of each classification head. When the weight is the largest, the corresponding category is determined as the final classification result.

[0034] In a specific embodiment, the step of inputting the multi-scale time-frequency feature matrix into the dual-stream attention-enhanced convolutional neural network in the touch signal classification model to obtain the convolution output feature map of the time stream and the frequency stream, wherein the time stream and the frequency stream respectively include 3 convolution layers, and the convolution kernel sizes are 3x3, 3x3 and 5x5 respectively, can specifically include the following steps: Decomposing the multi-scale time-frequency feature matrix into a time feature matrix and a frequency feature matrix to obtain a time feature matrix and a frequency feature matrix; Perform the first convolution operation on the time feature matrix with a convolution kernel size of 3x3 to obtain the output feature map of the first convolution layer of the time stream, and perform the first convolution operation on the frequency feature matrix with a convolution kernel size of 3x3 to obtain the output feature map of the first convolution layer of the frequency stream; Input the output feature map of the first convolutional layer of the time stream into the batch normalization layer and the ReLU activation function layer to obtain the first activation feature map of the time stream, and input the output feature map of the first convolutional layer of the frequency stream into the batch normalization layer and the ReLU activation function layer to obtain the first activation feature map of the frequency stream; Perform a second convolution operation on the first activated feature map of the time stream, with a convolution kernel size of 3x3, to obtain the output feature map of the second convolution layer of the time stream, and perform a second convolution operation on the first activated feature map of the frequency stream, with a convolution kernel size of 3x3, to obtain the output feature map of the second convolution layer of the frequency stream; Input the output feature map of the second convolutional layer of the time stream and the output feature map of the second convolutional layer of the frequency stream into the batch normalization layer and the ReLU activation function layer respectively to obtain the second activation feature map of the time stream and the second activation feature map of the frequency stream; A third convolution operation is performed on the second activated feature map of the time stream, and the convolution kernel size is 5x5 to obtain the third convolution layer output feature map of the time stream. A third convolution operation is performed on the second activated feature map of the frequency stream, and the convolution kernel size is 5x5 to obtain the third convolution layer output feature map of the frequency stream. The output feature map of the third convolution layer of the time stream and the output feature map of the third convolution layer of the frequency stream are output to obtain the convolution output feature maps of the time stream and the frequency stream.

[0035] Specifically, the multi-scale time-frequency feature matrix is ​​decomposed into a time feature matrix and a frequency feature matrix. The time feature matrix contains the information about the signal changes in the time domain, while the frequency feature matrix contains the information about the signal changes in the frequency domain. The time feature matrix is ​​subjected to the first convolution operation to extract local features in the time domain. The size of the convolution kernel is set to 3×3. Small convolution kernels can effectively capture short-term feature changes in the time series, and the output feature map of the first convolution layer of the time stream is obtained. At the same time, the system performs the same operation on the frequency feature matrix, using the same size of convolution kernel (3×3) to extract local features in the frequency domain and generate the output feature map of the first convolution layer of the frequency stream. The output feature map of the first convolution layer of the time stream is input to the batch normalization layer. By normalizing the data to a distribution with a mean of 0 and a variance of 1, the convergence speed of training is accelerated and the gradient vanishing problem is avoided. Normalization The normalized output feature map of the time stream is input to the ReLU activation function layer. ReLU (Rectified Linear Unit) is a commonly used activation function used to introduce nonlinearity so that neural networks can process complex patterns. The output after activation is the first activated feature map of the time stream. The output feature map of the first convolution layer of the frequency stream is also processed in the same way: after the batch normalization layer, and then after the ReLU activation function layer, the first activated feature map of the frequency stream is obtained. Through batch normalization and activation processing, the important features in the feature map are retained, and the training process of the network is more stable. The second convolution operation is performed on the first activated feature map of the time stream. The size of the convolution kernel is set to 3×3 to further extract deeper local features in the time domain. Through the convolution operation, more complex temporal features than the first convolution layer are captured to generate the output feature map of the second convolution layer of the time stream. The second convolution operation is performed on the first activated feature map of the frequency stream, using a 3×3 convolution kernel to extract deeper features in the frequency domain, and the output feature map of the second convolution layer of the frequency stream is obtained. The output feature map of the second convolution layer of the time stream and the output feature map of the second convolution layer of the frequency stream are respectively input into the batch normalization layer to maintain the stability of the data and ensure that the network can maintain a good gradient flow during the training process. The normalized output feature map is input into the ReLU activation function layer to obtain the second activated feature map of the time stream and the second activated feature map of the frequency stream. In order to extract features in a wider range, a third convolution operation is performed on the second activated feature map of the time stream. The convolution kernel size of the convolution operation is set to 5×5, covering a larger receptive field, capturing the global features in the time series, extracting more global time features, and generating the output feature map of the third convolution layer of the time stream. At the same time, the same operation is performed on the second activated feature map of the frequency stream, using a 5×5 convolution kernel to extract global features in the frequency domain, and generating the output feature map of the third convolution layer of the frequency stream. The output feature map of the third convolution layer of the time stream and the output feature map of the third convolution layer of the frequency stream respectively contain the time domain and frequency domain features extracted by three layers of convolution. These feature maps are represented as: ; ; in, express The convolution operation, express The convolution operation is , BatchNorm is the batch normalization layer, and ReLU is the activation function. and Represent the input time feature matrix and frequency feature matrix respectively. Through multi-layer convolution operations, the deep features of the touch signal in the time and frequency domains are extracted. The feature map output by the third convolution layer of the time stream and the feature map output by the third convolution layer of the frequency stream are used as the final convolution feature output.

[0036] In a specific embodiment, the process of executing step 500 may specifically include the following steps: The size and step length of the sliding time window are set, and the touch signal classification results are processed in segments to obtain a time window sequence. The touch signal classification results in each time window are counted and statistically analyzed to obtain the occurrence frequency of each category. Based on the occurrence frequency of each category, calculate the proportion of effective touch categories in each time window to obtain the effective touch proportion sequence; A trigger threshold is set for the effective touch ratio sequence, and when the effective touch ratio in the time window exceeds the trigger threshold, the time window is marked as a potential effective touch window; Merging consecutive potential valid touch windows to obtain candidate touch event segments; Calculate the duration of each candidate touch event segment, and compare it with the preset minimum valid touch duration, to filter out valid touch event segments that meet the duration requirement; Extract characteristic parameters of the effective touch event segment, including start time, end time, average touch pressure and touch trajectory; The extracted characteristic parameters are matched with the predefined touch pattern, and when the matching degree reaches the preset standard, the valid touch input event is confirmed and output.

[0037] Specifically, the size and step length of the sliding time window are set. The size of the sliding time window determines the time range of each processing, while the step length controls the speed and frequency of the window movement. Assuming that the sliding time window size is set to 1 second and the step length is 0.5 seconds, it means that the system slides once every 0.5 seconds and classifies the touch signals within the time period. By sliding the window, the continuous changes of the touch signal are captured. The setting of the time window should be determined according to the actual application scenario. For example, in a fast touch operation, a smaller time window and step length can more sensitively capture the user's operation details. After setting the time window and step length, the classification results of the touch signal are segmented and the entire touch signal classification result sequence is divided into multiple time window segments. These segments form a time window sequence, and each time window corresponds to a fixed time period. The touch signal classification results in each time window are counted and counted. The classification results are divided into different categories such as valid touch and false touch. In each time window, the frequency of occurrence of each category is counted, and their distribution in the time window is calculated. Based on the statistical results, the proportion of valid touch categories in each time window is calculated. The percentage of effective touches reflects the strength of effective operations in the time window. Assuming that there are 100 touch signal classification results in a certain time window, of which 70 are effective touches and 30 are false touch interferences, the percentage of effective touches in the time window is: ; in, The total number of signals. By performing similar calculations on each time window, a series of valid touch ratios are obtained. These ratio values ​​constitute a valid touch ratio sequence, which reflects the changes in the effectiveness of touch signals in different time periods. In order to distinguish between valid touches and false touches, a trigger threshold is set for the valid touch ratio sequence. The trigger threshold is a key criterion for determining whether a time window contains a valid touch operation. Assuming that the set trigger threshold is 0.6, if the valid touch ratio in a time window exceeds 0.6, then the time window is marked as a potential valid touch window. If the valid touch ratio is lower than the threshold, it is considered that the time period may be mainly affected by false touches or other interference. Through threshold judgment, the time window that may contain a valid touch operation is preliminarily screened out. After marking the potential valid touch window, the continuous potential valid touch windows are merged. If multiple continuous time windows are marked as potential valid touch windows, these windows are merged into a complete candidate touch event segment to form a continuous touch operation area. After the merger is completed, the duration of each candidate touch event segment is calculated. The duration refers to the time difference between the start time and the end time of the candidate event segment. The duration is compared with the preset minimum valid touch duration. If the duration of the candidate touch event segment is greater than or equal to the minimum valid touch duration, the touch operation is considered to be of sufficient length and is regarded as a valid touch event; otherwise, it is excluded and considered to be insufficiently continuous. After filtering out valid touch event segments that meet the duration requirements, key information is extracted from these event segments, including the start time, end time, average touch pressure, and touch trajectory. The start time and end time reflect the time span of the touch event, while the average touch pressure is calculated based on the pressure values ​​of all touch operations in the event segment. At the same time, the touch trajectory in the touch event segment is calculated. The trajectory is composed of the position information of the touch points. By processing these points, the movement path of the user's finger on the screen is obtained. The extracted feature parameters are matched with predefined touch patterns. Predefined touch patterns are usually set in advance according to the user's actual operating habits or application scenarios, such as different operating modes such as light touch, sliding or long press. By comparing the extracted touch features with these patterns, it is determined whether the current touch event meets a certain valid touch pattern. If the matching degree meets the preset standard, the valid touch input event is confirmed and output, and finally fed back to the touch screen application.

[0038] The above describes the touch screen accidental touch prevention method in the embodiment of the present application. The following describes the touch screen accidental touch prevention system 10 in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the touch screen anti-mistouch system 10 includes: The acquisition module 11 is used to acquire the user's touch input signal based on the sensor in the touch screen, and filter and normalize the touch input signal to obtain a target touch signal; An estimation module 12, configured to input the target touch signal into a sliding mode compensation function observer to perform touch signal interference estimation to obtain false touch interference information; An extraction module 13, used to extract features of a target touch signal according to false touch interference information to obtain a touch feature vector; The processing module 14 is used to input the touch feature vector into a preset touch signal classification model for processing to obtain a touch signal classification result; An analysis module 15, used to perform statistical analysis on the touch signal classification results within the sliding time window and determine valid touch input events; The generating module 16 is used to process the touch input signal and generate a touch response signal of the touch screen according to the valid touch input event and the touch signal classification result.

[0039] Through the synergy of the above-mentioned components, by filtering and normalizing the touch input signal, combined with Butterworth low-pass filter and outlier removal techniques, the quality of the original touch data is effectively improved, and the sliding mode compensation function observer is used to estimate the touch signal interference, which can capture various forms of false touch interference in real time and accurately, and improve the system's adaptability to complex touch scenes. Touch features are extracted by various methods such as time domain analysis, frequency domain transformation and wavelet transform, and the principal component analysis is combined for dimensionality reduction, which not only retains the key information of the touch signal, but also uses a dual-stream attention enhanced convolutional neural network and a multi-head self-calibration classifier to make full use of the advantages of deep learning in feature extraction and classification tasks, and improve the accuracy of touch signal classification. Through statistical analysis and multiple judgment criteria within the sliding time window, the probability of false touch and misjudgment is effectively reduced, and the stability and reliability of the system are improved. The present invention can dynamically adjust the feature extraction and classification strategies according to the false touch interference information, and has good environmental adaptability and user personalization capabilities.

[0040] The present application also provides a computer device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the touch screen anti-mistouching method in the above embodiments.

[0041] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0042] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0043] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A touch screen anti-mistouch method, characterized in that: The touch screen anti-mistouch method comprises: Collecting a user's touch input signal based on a sensor in the touch screen, and filtering and normalizing the touch input signal to obtain a target touch signal; Inputting the target touch signal into a sliding mode compensation function observer to perform touch signal interference estimation to obtain false touch interference information; Extracting features of the target touch signal according to the false touch interference information to obtain a touch feature vector; Inputting the touch feature vector into a preset touch signal classification model for processing to obtain a touch signal classification result; Performing statistical analysis on the touch signal classification results within the sliding time window, and determining a valid touch input event; According to the valid touch input event and the touch signal classification result, the touch input signal is processed and a touch response signal of the touch screen is generated.

2. The touch screen anti-mis-touch method according to claim 1, characterized in that: The method of collecting a user's touch input signal based on a sensor in the touch screen, filtering and normalizing the touch input signal to obtain a target touch signal includes: Perform high-frequency scanning on the touch screen surface to collect raw touch data including touch position coordinates, touch pressure value, touch duration and touch frequency; Inputting the original touch data into a Butterworth low-pass filter to filter out high-frequency noise to obtain preliminary filtered touch data, and removing outliers and impulse noise from the preliminary filtered touch data to obtain noise-reduced touch data; The touch pressure value in the noise reduction touch data is converted into a range of 0-1 by a linear mapping function to obtain a normalized touch pressure value, and the touch duration in the noise reduction touch data is logarithmically transformed to obtain a touch duration after logarithmic transformation; Performing exponential smoothing on the touch frequency in the noise reduction touch data to obtain a smoothed touch frequency, and performing trajectory smoothing and prediction on the touch position coordinates in the noise reduction touch data to obtain optimized touch position coordinates; The target touch signal is obtained by combining the normalized touch pressure value, the touch duration after logarithmic transformation, the touch frequency after smoothing, and the optimized touch position coordinates.

3. The touch screen anti-mis-touch method according to claim 1, characterized in that: The step of inputting the target touch signal into a sliding mode compensation function observer to perform touch signal interference estimation to obtain false touch interference information includes: The target touch signal is represented as x1, the first-order derivative of the target touch signal x1 is obtained to obtain x2, the false touch interference variable x3 is introduced, and the system state vector x=[x1, x2, x3] is constructed. T , where x1 represents the touch signal amplitude, x2 represents the touch signal change rate, and x3 represents the false touch interference amount; According to the system state vector x, the sliding mode compensation function observer equation group is constructed: x1 hat =z1+β1|x1-z1| (α1) sign(x1-z1); x2 hat =z2+β2|x1-z1| (α2) sign(x1-z1); x3 hat =z3+β3|x1-z1 |(α3) sign(x1-z1); Among them, x1 hat , x2 hat , x3 hat are the estimated values ​​of x1, x2, and x3, respectively; z1, z2, and z3 are auxiliary state variables; β1, β2, and β3 are positive design parameters used to adjust the convergence speed; α1, α2, and α3 are constants satisfying 0<α1<α2<1<α3, used to ensure the finite time convergence of the observer; Construct the sliding surface s=c1(x1-x1 hat )+c2(x2-x2 hat )+(x3-x3 hat ), where c1 and c2 are positive constants used to adjust the convergence characteristics of the sliding surface; Construct Lyapunov function V=(1 / 2)s 2 , calculate its derivative dV / dt, and ensure that dV / dt<0 to meet the system stability condition, where V represents the energy function of the system; Based on the stability condition, the parameters β1, β2, and β3 are iteratively optimized to obtain the optimal parameter value β1 opt , β2 opt , β3 opt , so that the observer has the best disturbance estimation performance; The target touch signal x1 is input into the optimized sliding mode compensation function observer, and the state estimation value x1 is obtained by iterative calculation. hat , x2 hat and x3 hat , where x3 hat This is the preliminary estimated false touch interference value; The preliminary estimated false touch interference value x3 hat Perform noise suppression and smoothing to obtain the corrected false touch interference value x3 smooth , where the state transfer matrix F and the observation matrix H of the Kalman filter are preset according to the system characteristics; The corrected false touch interference value x3 smooth Compared with the preset interference threshold Th, when x3 smooth >Th, the false touch interference information is set to 1, otherwise it is set to 0 to obtain the false touch interference information, where Th is a false touch judgment threshold predetermined according to the touch screen characteristics and user habits.

4. The touch screen anti-mis-touch method according to claim 3, characterized in that: The target touch signal x1 is input into the optimized sliding mode compensation function observer, and the state estimation value x1 is obtained by iterative calculation. hat , x2 hat and x3 hat , where x3 hat This is the preliminary estimated false touch interference value, including: Sampling the target touch signal x1 to obtain a discrete time sequence x1[n], where n represents a discrete time index; Based on the discrete time series x1[n], the discrete derivative x2[n]=(x1[n+1]-x1[n-1]) / (2Δt) of x1 is calculated by the central difference method, where Δt is the sampling time interval, and x2[n] represents the rate of change of the touch signal at time n; Substitute the x1[n] and x2[n] into the optimized sliding mode compensation function observer equation group, initialize z1[0], z2[0], z3[0] as zero vectors, where z1, z2, z3 are auxiliary state variables; Discretize the observer equations to obtain the discrete-time observer equation: z1[n+1]=z1[n]+Δt*(z2[n]+β1|x1[n]-z1[n]| (α1) sign(x1[n]-z1[n]))) z2[n+1]=z2[n]+Δt*(z3[n]+β2|x1[n]-z1[n]| (α2) sign(x1[n]-z1[n])); z3[n+1]=z3[n]+Δt*β3|x1[n]-z1[n]| (α3) sign(x1[n]-z1[n]); Among them, β1, β2, and β3 are positive design parameters used to adjust the convergence speed, and α1, α2, and α3 are constants satisfying 0<α1<α2<1<α3, which are used to ensure the finite time convergence of the observer; Based on the discrete time observer equation, a time series of z1[n], z2[n], z3[n] is obtained by iterative calculation, where n represents a discrete time index; According to the output equation of the sliding mode compensation function observer, the state estimation value is calculated: x1 hat [n]=z1[n]+β1|x1[n]-z1[n]| (α1) sign(x1[n]-z1[n]): x2 hat [n]=z2[n]+β2|x1[n]-z1[n]| (α2) sign(x1[n]-z1[n]): x3 hat [n]=z3[n]+β3|x1[n]-z1[n]| (α3) sign(x1[n]-z1[n]): Among them, x1 hat [n], x2 hat [n], x3 hat [n] represents the estimated values ​​of touch signal, signal change rate and false touch interference at discrete time index n respectively; The estimated value x1 of the state hat [n], x2 hat [n], x3 hat [n] performs convergence analysis and calculates the estimated error e1[n]=x1[n]-x1 hat [n]、e2[n]=x2[n]-x2 hat [n]、e3[n]=x3[n]-x3 hat [n], where e1[n], e2[n], and e3[n] represent the estimated errors of touch signal, signal change rate, and false touch interference, respectively; Based on the estimated errors e1[n], e2[n], and e3[n], the root mean square error RMSE is calculated. When the RMSE is less than the preset threshold ε or the number of iterations reaches the maximum value Nmax, the final state estimation value x1 is output. hat , x2 hat and x3 hat , where RMSE represents the root mean square of the estimation error, ε is the preset convergence threshold, and Nmax is the maximum number of iterations.

5. The touch screen anti-mis-touch method according to claim 1, characterized in that: The extracting features of the target touch signal according to the false touch interference information to obtain a touch feature vector includes: Performing time domain analysis on the target touch signal, extracting the maximum value, the minimum value and the average value of the touch pressure value, and obtaining the touch pressure feature; Based on the target touch signal, the touch duration and the touch frequency are calculated to obtain a time feature, and the target touch signal is transformed into a frequency domain to extract a main frequency component and energy distribution of the signal to obtain a frequency domain feature; Based on the touch position coordinates in the target touch signal, the curvature and speed of the touch track are calculated to obtain a spatial feature, and according to the false touch interference information, the interference intensity and duration are calculated to obtain an interference feature; Performing wavelet transform on the target touch signal, extracting multi-scale time-frequency features, and obtaining wavelet coefficients; constructing an initial feature vector based on the touch pressure feature, the time feature, the frequency domain feature, the space feature, the interference feature and the wavelet coefficient; The initial feature vector is subjected to principal component analysis, and several principal components with the largest contribution rates are selected to obtain a touch feature vector.

6. The touch screen anti-mis-touch method according to claim 1, characterized in that: The step of inputting the touch feature vector into a preset touch signal classification model for processing to obtain a touch signal classification result includes: Transforming the touch feature vector to obtain a multi-scale time-frequency feature matrix; Inputting the multi-scale time-frequency feature matrix into a dual-stream attention-enhanced convolutional neural network in a touch signal classification model to obtain convolution output feature maps of the time stream and the frequency stream, wherein the time stream and the frequency stream respectively contain 3 convolution layers, and the convolution kernel sizes are 3x3, 3x3, and 5x5, respectively; Performing channel attention mechanism processing on the convolution output feature maps of the time stream and the frequency stream to obtain a weighted time-frequency feature representation; The weighted time-frequency feature representation is input into the cross-modal self-attention module to capture the long-range dependency between the time and frequency domains and obtain the fused feature; Perform residual connection and layer normalization on the fused features, build residual attention blocks, and extract deep feature representations. The residual attention blocks are stacked in 3 layers. Inputting the deep feature representation into the global average pooling layer for processing to obtain a feature vector of fixed dimension; Inputting the fixed-dimensional feature vector into a multi-head self-calibration classifier to obtain a set of category probability distributions, wherein the multi-head self-calibration classifier is provided with three classification heads, and each classification head outputs probability distributions of two categories; A weighted vote is performed on the category probability distribution set to obtain a touch signal classification result.

7. The touch screen anti-mis-touch method according to claim 6, characterized in that: The multi-scale time-frequency feature matrix is ​​input into the dual-stream attention-enhanced convolutional neural network in the touch signal classification model to obtain the convolution output feature map of the time stream and the frequency stream, wherein the time stream and the frequency stream respectively contain 3 convolution layers, and the convolution kernel sizes are 3x3, 3x3 and 5x5 respectively, including: Decomposing the multi-scale time-frequency feature matrix into a time feature matrix and a frequency feature matrix to obtain a time feature matrix and a frequency feature matrix; Performing a first convolution operation on the time feature matrix with a convolution kernel size of 3x3 to obtain a first convolution layer output feature map of the time stream, and performing a first convolution operation on the frequency feature matrix with a convolution kernel size of 3x3 to obtain a first convolution layer output feature map of the frequency stream; Inputting the output feature map of the first convolutional layer of the time stream into a batch normalization layer and a ReLU activation function layer to obtain a first activated feature map of the time stream, and inputting the output feature map of the first convolutional layer of the frequency stream into a batch normalization layer and a ReLU activation function layer to obtain a first activated feature map of the frequency stream; Performing a second convolution operation on the first activated feature map of the time stream, with a convolution kernel size of 3x3, to obtain a second convolution layer output feature map of the time stream, and performing a second convolution operation on the first activated feature map of the frequency stream, with a convolution kernel size of 3x3, to obtain a second convolution layer output feature map of the frequency stream; Inputting the output feature map of the second convolutional layer of the time stream and the output feature map of the second convolutional layer of the frequency stream into a batch normalization layer and a ReLU activation function layer respectively, to obtain a second activation feature map of the time stream and a second activation feature map of the frequency stream; A third convolution operation is performed on the second activated feature map of the time stream, with a convolution kernel size of 5x5, to obtain a third convolution layer output feature map of the time stream, and a third convolution operation is performed on the second activated feature map of the frequency stream, with a convolution kernel size of 5x5, to obtain a third convolution layer output feature map of the frequency stream, and the third convolution layer output feature map of the time stream and the third convolution layer output feature map of the frequency stream are output to obtain the convolution output feature map of the time stream and the frequency stream.

8. The touch screen anti-mis-touch method according to claim 1, characterized in that: The performing statistical analysis on the touch signal classification results within the sliding time window and determining a valid touch input event includes: Setting the size and step length of the sliding time window, performing segmented processing on the touch signal classification results to obtain a time window sequence, and performing counting and statistics on the touch signal classification results in each time window to obtain the occurrence frequency of each category; Based on the occurrence frequency of each category, calculate the proportion of effective touch categories in each time window to obtain the effective touch proportion sequence; Setting a trigger threshold for the effective touch ratio sequence, and when the effective touch ratio in a time window exceeds the trigger threshold, marking the time window as a potential effective touch window; Merging consecutive potential valid touch windows to obtain candidate touch event segments; Calculate the duration of each candidate touch event segment, and compare it with the preset minimum valid touch duration, to filter out valid touch event segments that meet the duration requirement; Extracting characteristic parameters from the effective touch event segment, including start time, end time, average touch pressure and touch trajectory; The extracted characteristic parameters are matched with the predefined touch pattern, and when the matching degree reaches the preset standard, the valid touch input event is confirmed and output.

9. A touch screen anti-mistouch system, characterized in that: Used to execute the touch screen anti-mistouch method according to any one of claims 1 to 8, the touch screen anti-mistouch system comprises: An acquisition module, used for acquiring a user's touch input signal based on a sensor in the touch screen, and filtering and normalizing the touch input signal to obtain a target touch signal; an estimation module, configured to input the target touch signal into a sliding mode compensation function observer to perform touch signal interference estimation to obtain false touch interference information; An extraction module, configured to extract features of the target touch signal according to the false touch interference information to obtain a touch feature vector; A processing module, used for inputting the touch feature vector into a preset touch signal classification model for processing to obtain a touch signal classification result; An analysis module, used for performing statistical analysis on the touch signal classification results within a sliding time window and determining a valid touch input event; A generating module is used to process the touch input signal and generate a touch response signal of the touch screen according to the valid touch input event and the touch signal classification result.

10. A computer device, characterized in that: The computer device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the computer device to execute the touch screen mistake-touch prevention method according to any one of claims 1 to 8.

Images