Machine Learning-Based Method for Accidental Touch Detection and Intelligent Correction of LCD Touchscreens

Abstract

This invention provides a machine learning-based method for detecting and intelligently correcting accidental touches on LCD touchscreens. It is applicable to scenarios where frequent operations such as returning, swiping, and menu access occur in the edge zone, and where high-consequence operations involve clickable controls that are close to or overlap with the edge zone. This method constructs an edge spatial risk skeleton structure as a unified carrier of spatial context and long-term statistics without adding extra hardware, utilizing only touch data acquisition and terminal sensor information. During maintenance triggers, the edge risk grid is adaptively split or merged based on recent statistics, and the protective zone coverage relationship is updated accordingly, achieving statistical caliber correction without writing back to historical events. When the interface version changes, long-term statistics and tags are migrated and initialized according to the version change relationship, ensuring consistent data for accidental touch detection and correction action selection. This reduces the risk of triggering high-consequence accidental touches and minimizes interference with normal edge operations.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction and touch input processing technology, and more specifically, to a machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen. Background Technology

[0002] Human-computer interaction and touch input processing technology involves the prevention and correction of accidental touches on the edges of LCD touchscreens. In smart terminals, the edge area is often used for high-frequency operations such as returning, swiping, and menu access. However, the clickable areas of high-consequence controls such as payment confirmation and deletion of important data may be close to or overlap with the edge area. This means that a single touch within the edge area could be either a normal edge operation or trigger a high-consequence control. Existing terminals typically use touch acquisition and sensor information, employing methods such as edge rejection zone settings, gesture recognition rules, click judgment thresholds, and secondary confirmation in the event distribution chain to identify and intervene in edge area touches.

[0003] The existing technology has the following shortcomings:

[0004] Existing technologies generally suffer from two main problems: First, most solutions rely primarily on the instantaneous characteristics of a single touch and the current control's hit relationship, lacking event-level traceability of spatial attribution and write-back associations. This makes it difficult to stably loop the judgment result, action result, and user feedback to the same event and the same spatial caliber. Consequently, when edge gesture areas and high-consequence control areas are adjacent or overlap, it's difficult to simultaneously reduce the risk of high-consequence accidental touches and maintain the continuity of edge operations. Second, updates to statistical calibers and control strategies often depend on the current interface layout. Changes in control positions or clickable areas after application interface version updates can cause historical statistics and judgment criteria to drift. This results in the need for cold starts or costly recalculations, making it difficult to maintain the continuity and reusability of statistical data on resource-constrained terminals over the long term. To address these problems, this invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Machine learning-based methods for detecting and intelligently correcting accidental touches on LCD touchscreens include:

[0008] A single touch event falling within the edge band is registered for spatial attachment, the grid to which the event belongs is determined, an associated protective band identifier and an associated high-consequence operation control identifier are generated, and the spatial attachment result is written to the edge touch event record set;

[0009] Based on the edge touch event record set, the accidental touch judgment result and risk level are output according to the preset accidental touch judgment rules and preset classification rules and written back to the edge touch event record set, and an anti-accidental touch action command set is generated based on the risk level;

[0010] When the maintenance cycle is triggered, the edge touch event record set is aggregated by grid to form a window caliber statistical summary. The structural adjustment action decision model outputs the grid adjustment results for maintaining, splitting, or merging adjacent grids of the edge risk grid structure. The structural adjustment action decision model is a machine learning model. Before implementation, constraint verification related to the coverage relationship of the high-consequence operation area protection zone structure is performed to avoid merging across the protection zone boundary. The correspondence between grids before and after splitting or merging is recorded. The window caliber statistical summary is migrated or reassigned to the sub-grids obtained from splitting or the new grids obtained from merging, without writing back the grid identifiers to which the historical touch events belong.

[0011] When the results of the anti-accidental touch action are available and user feedback is available, incremental updates are performed on the long-term statistics in the edge risk grid structure and the high-consequence operation area protection zone structure; and when the interface version migration configuration structure indicates that the application interface version has changed, migration initialization is performed on the long-term statistics and version tags, and the migration only applies to the long-term statistics and version tags and does not directly rewrite the protection zone coverage relationship and area status.

[0012] In a preferred embodiment, the maintenance cycle is triggered by a sliding statistical time window structure; within its defined time range, the number of touch events and the risk level distribution are aggregated according to the grid identifier to which the event belongs to obtain a statistical summary of the grid dimension window caliber; and when the associated protection zone identifier is valid, it is further aggregated according to the associated protection zone identifier to obtain a statistical summary of the protection zone dimension window caliber.

[0013] In a preferred embodiment, when there are no event records within the statistical range that have been completed for accidental touch judgment and risk classification writing back, the window caliber statistical summary is set to empty or generated according to the default conservative caliber; and only based on the sample size threshold constraint and the regional status label constraint, the output retains or merges candidates, does not output split candidates based on the risk level distribution skew, and does not form or update the window caliber statistical summary of the protection zone dimension.

[0014] In a preferred embodiment, the structural adjustment action decision model is a machine learning model; it consists of a grid-level maintenance feature group composed of sample size, window caliber statistical summary, guardrail dimension window caliber statistical summary, grid adjacency relationship, guardrail coverage constraint and conflict state, and region state label constraint, and outputs structural adjustment action results of maintaining, splitting or merging; and after satisfying the sample size threshold, risk distribution skew threshold and consistency and conflict resolution rules in the edge space skeleton maintenance parameters and state set, it implements grid adjustment.

[0015] In a preferred embodiment, the grid splitting outputs a splitting candidate when the splitting sample size threshold is met and the risk distribution skewness reaches the threshold. Before the splitting is implemented, the continuity of the protective strip, the minimum width, and the conflict of the protection level are checked. If the check fails, the output is maintained, and the region status of the corresponding grid is marked as a conflict pending state.

[0016] In a preferred embodiment, the adjacent grid merging is output as a merging candidate when the merging sample size threshold is met and the difference in risk level between adjacent grids does not exceed the merging similarity threshold; and when adjacent grids belong to different protection zone protection levels, or one side is within the coverage area indicated by the protection zone coverage relationship while the other side is not within the coverage area, the output is maintained to avoid merging across the protection zone boundary.

[0017] In a preferred embodiment, when performing grid splitting, the candidate grids are subdivided into sub-grids according to their spatial layout, and new grid records are written into the edge risk grid structure; when performing adjacent grid merging, new merged grid records are generated and the window diameter statistical summaries of each grid participating in the merging are summarized; and the correspondence between grids before and after splitting or merging is recorded in the above splitting or merging, and the window diameter statistical summaries are migrated or redistributed to the sub-grids obtained from splitting or the new grids obtained from merging.

[0018] In a preferred embodiment, the edge risk grid structure is a status marker for each grid maintenance area. The area status marker includes at least a stable state, a high-risk state, a sample shortage state, and a conflict pending state. Furthermore, the transition from a stable state to a high-risk state must meet the condition of reaching the high-risk threshold for multiple consecutive maintenance cycles. The area status marker and grid division are jointly written into the edge space risk skeleton structure for reference in accidental touch judgment and action decision.

[0019] In a preferred embodiment, the migration initialization is achieved through a risk skeleton migration mechanism. When a change in the application interface version is detected, the risk skeleton migration mechanism determines the source control area and the target control area based on the semantic correspondence of the controls registered in the interface version migration configuration structure, and calculates the spatial overlap metric as the migration weight. When the spatial overlap metric is not lower than the effective overlap threshold, the long-term statistics of the edge risk grid structure and the high-consequence operation area protection zone structure are subjected to weighted migration initialization and written to the version tag.

[0020] In a preferred embodiment, when there is no valid semantic correspondence between controls or the spatial overlap metric is lower than the effective overlap threshold, the risk skeleton migration mechanism determines that there is no valid migration source, and performs stronger attenuation, zeroing or reset initialization on the long-term statistics corresponding to the target control area according to the migration attenuation rule and writes a new version tag; and the migration processing is limited to only acting on the long-term statistics and version tag, and does not directly rewrite the protection zone coverage relationship and area status.

[0021] The advantages and effectiveness of the machine learning-based method for detecting and intelligently correcting accidental touches on LCD touchscreens in this invention are as follows:

[0022] This invention constructs an edge spatial risk skeleton structure, unifying the spatial attachment of edge-related touch events, window caliber statistics, and long-term statistics. This allows the terminal to continuously provide a consistent spatial context and statistical basis for accidental touch detection and anti-accidental touch action selection without adding extra hardware, utilizing only existing touch acquisition and terminal sensor capabilities. When a maintenance cycle is triggered, the edge spatial skeleton adaptive maintenance mechanism performs adaptive splitting and merging of the edge risk grid based on aggregated statistics within the window, and simultaneously updates the protection zone coverage relationship. This completes statistical caliber correction without writing back historical events, thereby reducing the cost of repeated statistics and recalculations caused by granularity mismatch. When the interface version changes, the interface version-aware risk skeleton migration mechanism migrates long-term statistics and tags according to the version change relationship and initializes them according to rules, ensuring that long-term statistics remain reusable and continuous after the interface update. Based on the above consistent spatial context and reusable statistical basis, it can reduce the risk of high-consequence operations near the edge being triggered by accidental touches while reducing the interception and interference with normal edge operations. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0024] Figure 2 This is a schematic diagram of the spatial skeleton adaptive evolution mechanism of the present invention;

[0025] Figure 3 This is a schematic diagram of the structural adjustment action decision model of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] This invention provides a machine learning-based method for detecting and intelligently correcting accidental touches on LCD touchscreens. This method is applicable to scenarios where users frequently perform edge operations such as returning, swiping, and menu access in the edge zone area of ​​a smart LCD touchscreen. Because the clickable areas of controls for high-consequence operations such as payment confirmation and important data deletion may be close to or overlap with the edge zone, a single touch within the edge zone could be either an intentional edge operation or an accidental trigger of a high-consequence operation. The technical problem this invention aims to solve is: without adding additional hardware, and utilizing only existing touch acquisition and terminal sensor capabilities, how to continuously provide consistent spatial context and statistical data for touches within the edge zone, and maintain the reusability of this data after maintenance updates and interface version changes, thereby reducing the risk of high-consequence operations being triggered by accidental touches and minimizing interference with normal edge operations.

[0028] To address this, this invention constructs an edge space risk skeleton structure as a unified carrier for edge zone spatial context and long-term statistics. When a maintenance cycle is triggered, the edge space skeleton's adaptive maintenance mechanism adaptively adjusts the raster granularity based on recent statistics, including raster splitting of areas with significant internal differences and raster merging of adjacent areas with similar characteristics. The splitting and merging processes synchronously update the protection zone coverage relationship, enabling the skeleton to complete statistical caliber correction without rewriting historical events. When the interface version changes, the interface version-aware risk skeleton migration mechanism migrates long-term statistics and tags according to the version change relationship and initializes them according to migration rules.

[0029] Based on the above design, this invention constructs a complete process for a machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen, consisting of steps S101 to S105 sequentially. (Refer to...) Figure 1 , Figure 1 This is a schematic diagram of the method flow of the present invention.

[0030] Based on the above design, this invention constructs a complete process for a machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen, consisting of steps S101 to S105 sequentially. (Refer to...) Figure 1 , Figure 1 This is a schematic diagram of the method flow of the present invention.

[0031] Step S101, Event Acquisition and Spatial Attribution Registration, is used to acquire, parse, and spatially attach single touch events falling within the edge band of the LCD touchscreen during the event acquisition and spatial attribution registration phase, and complete the registration of event-level basic data. This step reads the edge touch raw event input set X101, obtains high-consequence operation controls and their spatial locations according to the high-consequence operation control configuration table C101, determines the grid to which the event belongs according to the edge risk grid structure Z101, and determines the protection zone hit and associated protection zone according to the high-consequence operation area protection zone structure W101. A spatial attachment result containing the grid identifier, protection zone hit mark, associated protection zone identifier, high-consequence control hit mark, and associated high-consequence operation control identifier is written into the edge touch event record set R101. This result is used for subsequent step S102's grid and protection zone aggregation maintenance and step S103's event-level risk determination and reading. It can also perform incremental updates to the basic operational statistics fields of Z101 and W101 based on the grid identifier and protection zone hit mark. Among them, X101 provides raw touch attribute input, C101 provides spatial positioning of high-consequence operation controls, Z101 provides grid boundary positioning, W101 provides the basis for determining the range of the protection zone and the protection level, and R101 is used for event-level information writing back and tracing.

[0032] Step S102, skeleton maintenance and statistical aggregation, is used in the skeleton maintenance phase triggered by the maintenance cycle. Based on the structural adjustment action decision model M1, it performs window aperture statistical aggregation, grid splitting and merging, and regional status maintenance on the edge risk grid structure Z101 and the high-consequence operation area protection zone structure W101 covered by the edge space risk skeleton structure K101, and completes the adaptive evolution of the edge space risk skeleton during operation. This step reads the edge touch event record set R101, and forms a grid-level sample size and window size statistical summary within its limited time window according to the sliding statistical time window structure W102. Under the constraints of the edge spatial skeleton maintenance parameters and the maintenance threshold of the state set Y102, as well as the regional state consistency and conflict resolution rules, the model M1 determines the actions of maintaining, splitting, or merging Z101 and updates the protective zone coverage relationship and regional state record of W101 simultaneously. The grid identifier of the historical event record in R101 is not written back. The structural evolution result is written back to Z101 and W101 and K101 is refreshed. At the same time, the statistical time window is advanced according to the advancement strategy of W102, and the window size statistical summary that exceeds the window range is attenuated or cleared in Z101 and W101 according to its attenuation configuration. This is used for the subsequent steps S103 and S104 to directly reference the spatial skeleton and window size statistical summary. Among them, the sliding statistical time window structure W102 is used to limit the time range of window caliber statistics and provide a unified caliber for maintenance cycle triggering and window advancement; the edge space risk skeleton structure K101 is the spatial skeleton result carrier, used to uniformly carry the window caliber statistical summary, regional status marker and version marker, and is composed of Z101, W101 and maintenance-related configurations; the model M1 is used to map the grid-level maintenance feature group within the maintenance cycle to the structural adjustment action result, and update the model parameters or decision thresholds based on the effect evaluation results after the maintenance cycle.

[0033] Step S103, Accidental Touch Judgment and Risk Classification, is used to perform accidental touch judgment and risk classification on a single touch event falling within the edge band of the LCD touchscreen during the accidental touch judgment and risk classification stage, and to determine whether the event constitutes an accidental touch and its event-level risk level. This step reads the edge touch event record set R101, constructs a grid-level spatial risk context based on the edge spatial risk skeleton structure K101 and the edge risk grid structure Z101, and supplements the protection level and spatial relationship information of the protection zone based on the high-consequence operation area protection zone structure W101 when the protection zone is hit. Combines the touch behavior feature input, device posture information input, and user long-term usage habit index input to form a fused input feature, and performs accidental touch judgment and event-level risk base value calculation on the fused input feature according to preset rules to obtain the accidental touch judgment result and event-level risk base value. According to the preset classification rules, the event-level risk base value is mapped to the risk level and written back to R101 for subsequent action decision-making in step S104 and long-term statistical update in step S105. This step supplements the registration of accidental touch judgment results, event-level risk baseline values, and risk levels in the risk judgment field group of R101, and serves as the source of event-level risk input for subsequent action decisions and long-term statistics.

[0034] Step S104, the anti-mistouch action decision, is used in the action decision stage to determine the execution range and action decision of a single touch event falling within the edge band of the LCD touch screen in the candidate anti-mistouch action set, and to complete the determination of the action type and action intensity. This step reads the edge touch event record set R101, obtains the area status mark of the grid based on the edge space risk skeleton structure K101 and incorporates the window aperture statistical summary when necessary, obtains the protection level of the protection zone based on the high-consequence operation area protection zone structure W101, and determines the selectable range of action type and action intensity for this event by combining the edge space skeleton maintenance parameters, the area-side constraints and strategy parameters of the state set Y102, and the user's acceptable level of strong intervention actions. Within the selectable range, the target action and action intensity are determined according to the preset action selection rules, written into the anti-mistouch action instruction set R104, and simultaneously written back to R101 for feedback statistics and long-term updates in the subsequent step S105. Among them, the candidate set of anti-accidental touch actions includes at least four categories: release, interception, confirmation and position correction. R104 is used to register the action type, action intensity and execution parameters and provide a traceability carrier for backfilling the execution results.

[0035] Step S105, closed-loop update and long-term statistics, is used in the closed-loop update stage to perform feedback write-back and long-term statistical incremental update on single edge touch events that have completed the anti-mistouch action and have been approved by the user. It also generates maintenance status and threshold correction information and writes it into the edge space skeleton maintenance parameters and status set Y102. This step reads the edge touch event record set R101 and the anti-accidental touch action instruction set R104. Based on the long-term statistical inclusion range of the edge risk grid structure Z101, the high-consequence operation area protection zone structure W101, and the sliding statistical time window structure W102, it configures and limits the long-term statistical inclusion caliber. Combining the risk assessment results and action decision results of R101 with the execution result backfill information of R104, it generates a long-term statistical increment and writes it to Z101 and W101. Under the constraint of the interface version migration configuration structure C105, the risk skeleton migration mechanism M2 migrates the long-term statistics and writes them to the version tag when an interface version change is detected. This is used by the subsequent step S102 to update and refresh the unified landing area status and protection zone coverage relationship within the maintenance cycle, and the resulting K101 is referenced by steps S103 and S104. The interface version migration configuration structure C105 is used to register the control correspondence between the old and new versions, the spatial overlap judgment caliber, and the migration attenuation rules to constrain the scope and attenuation intensity of the long-term statistical migration.

[0036] The implementation process and operational effects of the method of the present invention will be described in detail below with reference to specific embodiments. It should be understood that the embodiments are only used to illustrate the technical solution of the present invention, and not to limit it. The relevant steps, parameters and module divisions can be appropriately adjusted without changing the essence of the invention.

[0037] In an optional embodiment, step S101 is used to perform original attribute extraction and spatial attachment registration processing on a single touch event within the edge band based on X101, so as to provide event-level input for subsequent step S102 of grid-based and protective band aggregation maintenance and step S103 of event-level risk determination.

[0038] To complete the spatial attachment determination and write-back, the high-consequence operation control configuration table C101, the edge risk grid structure Z101, and the high-consequence operation area protection zone structure W101 are defined as follows in this step: C101 is a control-level configuration carrier, used to register the control identifier, high-consequence attribute, and control spatial positioning information of high-consequence operation controls that require key protection. The control spatial positioning information includes at least the control display area boundary or equivalent locatable information, and serves as the standard for touch coordinate standardization, ensuring that touch coordinates and control spatial positioning information are under the same coordinate standard; Z101 is an edge zone spatial discretization structure, used to divide the edge zone of the LCD touch screen into multiple grid units and provide the positioning basis for the grid to which the event belongs, including at least the grid identifier and its grid spatial boundary information; W101 is a protection zone description structure established around the adjacent area of ​​the high-consequence operation control, used to provide the basis for determining the protection zone hit and associated protection zone and to maintain the protection zone coverage relationship, including at least the protection zone identifier, protection zone spatial range information, and protection zone protection level information.

[0039] This step involves event reading and standardized parsing, spatial attachment determination and association generation, and statistical updates and event recording. In event reading and standardized parsing, the current event is read from X101, and raw touch attributes such as touch coordinates, timestamp, touch size, and contact duration are extracted. The touch coordinates are converted into a standardized event space description under a unified edge space coordinate system based on the interface coordinate caliber. The control identifier and spatial positioning information of the high-consequence operation control are read from C101 as the basis for subsequent association determination. In spatial attachment determination and association generation, the grid identifier is determined based on the grid space boundary information of Z101, and the protective zone hit mark is determined based on the protective zone spatial range information of W101. Upon hit, an associated protective zone identifier is generated. A valid associated protective zone identifier means that the associated protective zone identifier is not empty and can be retrieved from the protective zone identifier set or coverage relationship record of the high-consequence operation area protective zone structure W101. Then, based on the control spatial positioning information of C101, the high-consequence control hit mark is determined and an associated high-consequence operation control identifier is generated when a hit occurs. In the statistics update and event log, based on the grid identifier and the guard strip hit mark, the basic operation statistics fields such as the number of touches are updated in Z101 and W101, and an event log is added in R101. The original touch attributes and spatial attachment results are written into the corresponding field group, and the risk judgment field group, action decision field group and user feedback field group are reserved in R101 as a unified write-back location for subsequent steps.

[0040] The grid identifier and associated protective zone identifier serve as subsequent aggregation indexes by grid and protective zone. High-consequence control hit markers and associated high-consequence operation control identifiers serve as event-level positioning bases for subsequent differentiated judgments and action decisions. The touch count is a basic running statistics field, not a window-scope statistics field constrained by the sliding statistics time window structure W102, nor does it replace the incremental maintenance of subsequent long-term statistics fields. Through R101 written in this step, step S102 can directly use the grid identifier and associated protective zone identifier as aggregation indexes to perform window maintenance and skeleton maintenance. Step S103 can directly construct event-level judgment inputs based on the original touch attributes and spatial attachment information and write back the risk judgment result, avoiding repeated parsing of touch events and repeated positioning of spatial attribution.

[0041] In an optional embodiment, step S102 is triggered when W102 indicates the start of a maintenance cycle, and aggregates and maintains historical events within the time window defined by W102 to ensure that subsequent events can reference the latest spatial skeleton when executing steps S103 and S104. The structural adjustment action decision model M1 is used in this step to aggregate events and perform grid statistical calculations within the time window of R101. Based on the formation of grid-level maintenance feature groups, it outputs structural adjustment action results for each grid, where the structural adjustment action result is one of maintaining, splitting, or merging. Subsequently, regional state update calculations are performed according to the action results, and the structural evolution results are written back to Z101 and W101 to form an updated K101. K101 is composed at least of the grid division and regional state markings of Z101, the protective zone coverage relationship and protection level of W101, and maintenance-related configuration results. Step S102 updates Z101 and W101 and simultaneously refreshes K101, ensuring that subsequent steps reference a consistent spatial risk context.

[0042] In this embodiment, the structural adjustment action decision model M1 is a machine learning model used to map the grid-level maintenance feature set formed during the maintenance cycle to the structural adjustment action result. The grid-level maintenance feature set includes at least the sample size, window aperture statistical summary, protective strip dimension window aperture statistical summary, grid adjacency relationship, protective strip coverage constraints and conflict status, and area state-related constraints. The training samples for model M1 come from historical maintenance cycles, and the sample labels are the final structural adjustment action results implemented in that maintenance cycle. The model parameters or decision thresholds can be updated based on the effect evaluation results after the maintenance cycle. Touch behavior feature input, device posture information input, and user long-term usage habit index input are provided by the terminal acquisition module when a touch event occurs or when the window is updated. These are obtained by aligning the touch event timestamp and event identifier with the corresponding event in the edge touch event record set R101, serving as supplementary input for accidental touch judgment and risk classification. (Refer to...) Figure 2, Figure 2 This is a schematic diagram of the structural adjustment action decision model of the present invention.

[0043] The implementation process of this step includes event aggregation and grid statistical calculation within the time window, determination of structural adjustment actions and implementation of grid splitting and merging, and implementation and updating of protective belt coverage relationship and statistical time window.

[0044] In the event aggregation and grid statistical calculation within the time window, edge touch events located within the time range defined by W102 are selected from R101. The number of events in each grid within the current time window is counted based on the grid identifier to which the event belongs as the aggregation index. The risk level distribution statistics are obtained by aggregating the risk judgment field groups corresponding to the event records that have completed the risk judgment write-back in step S103 within the time window, forming the risk level distribution statistics of the sample size and window caliber corresponding to each grid. When a touch event has a protection zone hit and the associated protection zone identifier is valid, the number of touch events is further aggregated using the associated protection zone identifier as the aggregation index. When the risk level distribution statistics are available, aggregation is performed to form a window caliber statistical summary of the protection zone dimension to support the subsequent maintenance of protection zone coverage relationship and the construction of action decision constraints. When there are no event records that have completed the risk judgment write-back in step S103 within the time window, the risk level distribution statistics in the window caliber statistical summary are set to empty or generated according to the default conservative caliber, and the risk level distribution aggregation statistics of the protection zone dimension are not formed or updated. In this scenario, this maintenance cycle only maintains or merges candidate filters based on sample size threshold constraints and regional state constraints, without performing grid splitting decisions based on risk level distribution skewness. The sample size and risk level distribution statistics for each grid are written into the window caliber statistics field group of Z101, and when a protective zone hits and the associated protective zone identifier is valid, the number of touch events in the protective zone dimension is written into the window caliber statistics field group of W101.

[0045] After the statistics for this time window are completed, the structural adjustment action decision model M1 will organize the sample size, risk level distribution statistics, protective belt dimension window caliber statistical summary, grid spatial location and adjacency relationship of Z101, and protective belt coverage constraints and conflict status of W101 into a grid-level maintenance feature group, and input it into model M1 to output the structural adjustment action results.

[0046] In the structural adjustment action determination and grid splitting and merging implementation, based on the aforementioned grid-level maintenance feature group, the grid spatial location and adjacency relationship of Z101, and the protective belt coverage constraint of W101, the structural adjustment action decision model M1 outputs the structural adjustment action result under the premise of satisfying the sample size threshold constraints, risk distribution skewness threshold constraints, and regional state consistency and conflict resolution rules registered in Y102, and determines the splitting candidate grid and merging candidate grid pairs accordingly; after the candidate grids are constrained and verified, they are split or merged. For grids that meet the splitting conditions, they are subdivided into sub-grids according to their spatial layout and new grid records are written in Z101. The grid correspondence before and after splitting is recorded and the window aperture statistical summary is migrated or redistributed. For adjacent grid pairs that meet the merging conditions, new merged grid records are generated under the condition of not destroying the protective belt continuity and minimum width constrained by W101, and the window aperture statistical summary is summarized and the grid correspondence before and after merging is recorded. At the same time, step S102 does not write back the grid identifier of the historical event record in R101. The constraint verification includes at least the continuity of the protective zone, minimum width, and conflict verification of the protection level. If the verification fails, the output is retained and the corresponding grid is marked as a conflict pending state. Merging candidates must meet the following requirements: the difference in risk level between adjacent grids does not exceed the merging similarity threshold and they do not cross the boundary of the protective zone. The protection level conflict verification is used to determine whether splitting or merging candidates causes adjacent grids to cross the boundary of different protective zones' protection levels, or whether one side is within the coverage area indicated by the protective zone coverage relationship while the other side is not. When a conflict is established, the verification is deemed to have failed and the corresponding grid is marked as a conflict pending state. The risk distribution skew is used to characterize the degree of imbalance in the risk level distribution in the window's statistical summary. It can be determined by the difference between the proportion of high-risk level events and the proportion of low-risk level events, or by the variance and entropy measure of the risk level distribution. The risk level difference or merging similarity threshold between adjacent grids can be determined by the risk level distribution distance or the difference in the risk level mean between adjacent grids. The sample size threshold is determined by the number of events within the window that have completed risk level write-back.

[0047] In the process of updating the coverage relationship and statistical time window of the protective belt, the coverage grid set of each protective belt in W101 is recalculated based on the latest grid division results, and its protection level related records are updated synchronously when necessary. At the same time, the statistical time window is advanced according to the advancement strategy of W102, and the window caliber statistics that exceed the window range are attenuated or cleared in Z101 and W101 according to its attenuation configuration. The window status field group of W102 is updated so that subsequent maintenance cycles can continue to perform aggregation and skeleton maintenance in the new statistical window. The attenuation or clearing process does not affect the basic operation statistical field reach count of Z101 and W101, nor does it affect the long-term statistical fields of incremental maintenance in step S105.

[0048] This step updates the unified landing area status markers and protective zone coverage relationships during the maintenance cycle. Step S105 does not directly rewrite the area status markers and protective zone coverage relationships. Through the write-back and update of K101 in this step, step S103 can directly perform event-level risk determination based on the updated grid division, area status markers, and window aperture statistical summary, and write back the risk determination results. Step S104 can directly construct action decision inputs based on the protective zone coverage relationship and area status constraints, and generate a set of anti-accidental touch action instructions, avoiding repeated execution of spatial skeleton reconstruction, coverage relationship recalculation, and window aperture aggregation statistical calculation.

[0049] In an optional embodiment, step S103 is triggered when there is a pending event record in R101 that has not yet been written back to the risk determination field group. Based on the event record and the spatial risk context provided by K101, the determination and classification are performed, providing a unified event-level risk input for subsequent steps S104 and S105.

[0050] The implementation process of this step includes reading event logs and constructing spatial risk context, constructing fused input features and determining classification, and writing back the determination results.

[0051] In the event log reading and spatial risk context construction, the original touch attributes such as touch coordinates, timestamp, touch size, and contact duration of the current event log to be judged are read from R101, as well as spatial attachment information such as the grid identifier, guardrail hit mark, associated guardrail identifier, high-consequence control hit mark, and associated high-consequence operation control identifier. Based on the grid identifier, the sample size and risk level distribution statistics of the area status and window size corresponding to the grid are searched in Z101 to form a grid-level spatial risk context. When the guardrail is hit and the associated guardrail identifier is not null, the guardrail protection level and spatial relationship information are further obtained from W101 and incorporated into the spatial risk context. When the guardrail hit mark is not hit, only the grid-level spatial risk context is used.

[0052] In the construction and classification of fused input features, behavioral features, posture features, and habit features aligned with the event are extracted from touch behavior feature input, device posture information input, and user long-term usage habit index input. These features are then normalized and encoded according to preset feature specifications. Combined with the aforementioned spatial risk context and the spatial relationship information related to controls introduced when high-consequence controls are hit, fused input features are formed. Based on preset rules, the fused input features are used to perform accidental touch judgment and calculate the event-level risk baseline value, obtaining the accidental touch judgment result and the event-level risk baseline value. The event-level risk baseline value is then mapped to a risk level according to preset classification rules. Preset accidental touch judgment rules, preset classification rules, preset action selection rules, preset feature specifications, as well as mapping and pruning calibers and correction conditions and correction magnitudes are uniformly registered in the edge space skeleton maintenance parameters and state set Y102, and take effect for different interface versions under the constraints of the interface version migration configuration structure C105. Unconfigured items adopt the system default rules. The touch behavior characteristics, device posture information, and long-term user habit indicators are all obtained by aligning the event timestamps and event identifiers with R101. Their value caliber and update strategies are preset by the system or registered by Y102 and configurable and updateable within the maintenance cycle. Specifically, the preset rules are used to consistently integrate spatial context such as high-consequence control hits, protective zone protection levels, grid area status, and window size statistical summaries with touch behavior characteristics, device posture information, and long-term user habit indicators. This also applies conservative or increased corrections to the event-level risk baseline values. The relevant conditions and correction magnitudes are registered by Y102 or preset by the system and configurable and updateable within the maintenance cycle.

[0053] In the write-back of the judgment results, the risk judgment field group of the event record is updated in R101 based on the judgment and classification results. The accidental touch judgment result, the event-level risk base value, and the risk level are written into the corresponding field group, while keeping the original touch attribute field group and spatial attachment field group in R101 unchanged. At the same time, the action decision field group and user feedback field group are not written back, so that the write-back results of this step can be used as the action selection input for the subsequent step S104 and the long-term statistical update basis for step S105, and as the event-level basis for window-caliber risk distribution statistical aggregation when executing step S102 in the subsequent maintenance cycle. The window-caliber statistical summary of step S102 can be directly aggregated based on the risk judgment field group written back to R101 in step S103 according to the grid identifier and associated protection zone identifier to obtain the risk level distribution statistics item summarized by time window, thereby ensuring that step S102 can complete the window-caliber statistical aggregation without reconstructing the event-level judgment within the maintenance cycle.

[0054] This step only writes back the risk assessment field group of R101, without updating the basic operational statistics fields of Z101 and W101. It does not replace the window caliber statistical aggregation and skeleton maintenance in step S102, nor does it replace the long-term statistical incremental maintenance in step S105. Through the R101 written in this step, step S104 can directly construct the action decision input based on the accidental touch assessment result, the event-level risk baseline value, and the risk level, and write it back to the action decision field group, avoiding the repeated execution of spatial risk context construction and assessment grading in subsequent steps.

[0055] In an optional embodiment, step S104 is triggered when the risk determination field group of R101 has been written back and the action decision field group has not yet written the action decision result. Based on the spatial context such as the area status and protection level of the protective belt provided by K101, combined with the area-side constraints and experience-side constraints of Y102, the execution range of the candidate anti-mistouch action set is determined and the rule decision is processed, providing action-side input for the subsequent step S105.

[0056] The implementation process of this step includes summarizing event risk information and spatial context, determining the range of optional actions and making rule decisions, and writing back instructions to the ground: In the summarizing event risk information and spatial context, the risk judgment field group and spatial attachment information of the current event are read from R101. Based on the grid identifier, the area status marker of the grid is searched in Z101 and, if necessary, the sample size and risk level distribution statistics of the window aperture corresponding to the grid are incorporated. When the protection zone hit mark is hit and the associated protection zone identifier is not null, the protection zone protection level and spatial relationship information are further searched based on W101 and incorporated into the spatial context to form an event-level context summary for action constraint determination and rule decision-making.

[0057] In determining the range of selectable actions and making rule decisions, regional constraints are determined by combining the policy parameters corresponding to the regional state in Y102, and experience-side constraints are formed by reading the user's acceptable level of strong intervention actions registered in Y102. The regional constraints and experience-side constraints are merged to obtain the selectable range of action types and action intensities for this event. Within the selectable range, the target anti-accidental touch action and action intensity are determined according to preset action selection rules, and a strong intervention marker is generated. The candidate anti-accidental touch action set includes at least four action types: release action, interception action, confirmation action, and position correction action. The release action is applied to inappropriate... Additional intervention is applied to the previous touch event. The interception action is used to prevent the current touch event from triggering a high-consequence operation. The confirmation action is used to require the user to confirm again before triggering. The position correction action is used to correct the touch coordinates or trigger target to avoid hitting the high-consequence operation control area. The preset action selection rule maps the action type and action intensity within the candidate anti-accidental touch action set, taking into account the risk level, the protection level of the protective zone, the area status, and the user experience constraints. The upper limit of the intensity is pruned to meet the user experience constraints. The relevant mapping and pruning criteria are registered by Y102 or preset by the system and can be configured and updated within the maintenance cycle.

[0058] During the instruction write-back process, based on the target's anti-accidental touch action, an action instruction record corresponding to the event is generated in R104. The action type, action intensity, and execution parameters are written into the corresponding record. An execution result backfill field group is reserved or written in R104 for subsequent writing of execution result markers. At the same time, the action type, action intensity, and strong intervention marker are written into the action decision field group in R101 to provide action-side input for subsequent execution result backfilling and long-term statistical updates.

[0059] This step does not update the basic operational statistics fields of Z101 and W101, nor does it replace the window-based statistical aggregation in step S102 or the long-term incremental statistical maintenance in step S105. Through the action decision field group of R104 and R101 written in this step, step S105 can directly perform feedback statistics and long-term updates based on action type, action intensity, and strong intervention markers combined with user feedback information. It can also locate the corresponding action instruction record based on the event index of R104, avoiding the need for subsequent steps to repeatedly perform event-level spatial context aggregation, optional action range determination, and rule decision processing.

[0060] In an optional embodiment, step S105 is triggered when the execution result backfill information of the event in R104 is available, and when user feedback information is received, the user feedback field group of R101 is completed; step S105 only performs incremental updates on the long-term statistical fields of Z101 and W101 and writes the correction information into Y102 for subsequent step S102 to uniformly update the status of the landing area and the protection zone coverage relationship within the maintenance cycle. Simultaneously, when an interface version change is detected, mechanism M2 migrates the long-term statistics according to C105 and writes them into the version tag. Mechanism M2 is a risk skeleton migration mechanism, used to perform migration initialization on the long-term statistics and version tag according to the semantic correspondence of controls and spatial overlap measurement when an application interface version change is detected. (Refer to...) Figure 3 , Figure 3 This is a schematic diagram of the risk skeleton migration mechanism of the present invention.

[0061] The interface version migration configuration structure C105 is used to register the semantic correspondence of controls between the old and new versions, the spatial overlap metric and its effective threshold, and the migration attenuation rules, so as to constrain the range and attenuation intensity of long-term statistical migration.

[0062] The implementation process of this step includes writing back the event execution results and feedback, recording and accumulating long-term statistical increments, and generating correction information and processing version migration.

[0063] In the event execution result and feedback write-back, the event record that has completed the anti-accidental touch processing is read from R101 and the risk judgment field group, action decision field group and spatial attachment information of the event are obtained. The action type, action intensity, execution parameters and execution result flag in the execution result backfill field group corresponding to the event are read from R104. Feedback flags such as undo, retry, and continuous click in the user feedback information input are received. The execution result flags and feedback flags are written into the user feedback field group of R101 as a traceable basis for subsequent long-term statistical updates and correction information generation.

[0064] In the long-term statistical increment calculation and accumulation, the grid corresponding to the current event and the associated protective band at the time of the hit are determined according to the grid identifier of R101 and the protection band hit mark. The long-term statistical calculation range configuration of W102 determines whether the current event is included in the long-term statistics. The risk judgment result, risk level, whether the anti-mistouch action is triggered, and whether it belongs to the strong intervention action of the event are mapped to the long-term statistical increment. In Z101, the increment of long-term mistouch statistics and strong intervention usage statistics is accumulated for the grid. In W101, the corresponding long-term statistical increment is accumulated for the area covered by the associated protective band, forming the incremental contribution of the event to the long-term statistical scope and writing it into the corresponding long-term statistical field.

[0065] In the correction information generation and version migration process, the incrementally updated Z101 and W101 are used as inputs, and combined with the regional status threshold configuration and maintenance status information registered in Y102, the risk level and strong intervention usage of each grid and its associated protective zone on a long-term time scale are evaluated. The withdrawal or acceptance tendency in the user feedback field group of R101 is used as the correction direction criterion to generate maintenance status and threshold correction information related to the regional status adjustment direction and threshold correction, which is written into Y102. Then, the current application interface version identifier is read and compared with the version mark recorded in Z101 and W101 to determine whether an interface version change has occurred. When an interface version change is detected, mechanism M2, according to C105, migrates the long-term statistics in the old version Z101 and W101 that have spatial overlap with the control area of ​​the new version to the grid and protective zone area corresponding to the new version, and writes the migrated long-term statistical results and version mark. The migration process is limited to the long-term statistical fields and version mark fields of Z101 and W101 and does not directly rewrite the protective zone coverage relationship and regional status.

[0066] In a further optional embodiment, when mechanism M2 detects a change in the application interface version, it first determines the source control area and the target control area based on the control semantic correspondence registered in C105. If no valid control semantic correspondence exists, it is determined that there is no valid migration source. Subsequently, mechanism M2 calculates the spatial overlap metric between the source and target control areas and uses it as the migration weight metric. The spatial overlap metric can be the intersection-union ratio (IU), and C105 registers a valid overlap threshold. If the spatial overlap metric is lower than the valid overlap threshold, it is also determined that there is no valid migration source. In the weighted migration initialization, it can be done according to... Calculate the migration initialization values, where For spatial overlap measurement, The migration attenuation coefficient registered in the interface version migration configuration structure C105. This is a long-term statistical measure for the source region. This serves as the initialization amount for long-term statistics in the target area; when it is determined that there is no valid migration source, the stronger attenuation, zeroing, or reset initialization rules registered in C105 are applied. Perform the processing and write the new version tag.

[0067] When no valid migration source is determined, Mechanism M2 performs stronger attenuation, zeroing, or reset initialization on the long-term statistics corresponding to the target control area according to the migration attenuation rules registered in C105, and writes the new version tag. When the semantic correspondence is valid and the spatial overlap meets the overlap validity threshold, Mechanism M2 performs weighted migration initialization on the long-term statistics of the old version source area according to the migration weight, and attenuates the migration initialization value according to the attenuation caliber registered in C105 before writing it to the long-term statistics and version tag of the target area. The above migration processing is limited to the long-term statistics field and the version tag field, and does not directly rewrite the protection band coverage relationship and area status, nor does it affect the window caliber statistics summary field and the basic operation statistics field.

[0068] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0069] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0070] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0071] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0072] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine learning-based method for detecting and intelligently correcting accidental touches on an LCD touchscreen, characterized in that, include: A single touch event falling within the edge band is registered for spatial attachment, the grid to which the event belongs is determined, an associated protective band identifier and an associated high-consequence operation control identifier are generated, and the spatial attachment result is written to the edge touch event record set; Based on the edge touch event record set, the accidental touch judgment result and risk level are output according to the preset accidental touch judgment rules and preset classification rules and written back to the edge touch event record set, and an anti-accidental touch action command set is generated based on the risk level; When the maintenance cycle is triggered, the edge touch event record set is aggregated by grid to form a window caliber statistical summary. The structural adjustment action decision model outputs the grid adjustment results for maintaining, splitting, or merging adjacent grids of the edge risk grid structure. The structural adjustment action decision model is a machine learning model. Before implementation, constraint verification related to the coverage relationship of the high-consequence operation area protection zone structure is performed to avoid merging across the protection zone boundary. The correspondence between grids before and after splitting or merging is recorded. The window caliber statistical summary is migrated or reassigned to the sub-grids obtained from splitting or the new grids obtained from merging, without writing back the grid identifiers to which the historical touch events belong. When the results of the anti-accidental touch action are available and user feedback is available, incremental updates are performed on the long-term statistics in the edge risk grid structure and the high-consequence operation area protection zone structure; and when the interface version migration configuration structure indicates that the application interface version has changed, migration initialization is performed on the long-term statistics and version tags, and the migration only applies to the long-term statistics and version tags and does not directly rewrite the protection zone coverage relationship and area status.

2. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 1, characterized in that, The maintenance cycle is triggered by a sliding statistical time window structure; within its limited time range, the number of touch events and the risk level distribution are aggregated according to the grid identifier to which the event belongs, to obtain a statistical summary of the grid dimension window caliber; and when the associated protection zone identifier is valid, it is further aggregated according to the associated protection zone identifier to obtain a statistical summary of the protection zone dimension window caliber.

3. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 2, characterized in that, When there are no event records within the statistical scope that have been completed for accidental touch judgment and risk classification writing back, the window caliber statistical summary is set to empty or generated according to the default conservative caliber; and only based on sample size threshold constraints and regional status label constraints, the output retains or merges candidates, does not output split candidates based on risk level distribution skew, and does not form or update the window caliber statistical summary of the protection zone dimension.

4. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 2, characterized in that, The structural adjustment action decision model is a machine learning model; it consists of a grid-level maintenance feature group composed of sample size, window caliber statistical summary, protective belt dimension window caliber statistical summary, grid adjacency relationship, protective belt coverage constraints and conflict status, and regional state label constraints, and outputs structural adjustment action results of maintaining, splitting or merging; and after satisfying the sample size threshold, risk distribution skew threshold and consistency and conflict resolution rules in the edge space skeleton maintenance parameters and state set, it implements grid adjustment.

5. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 4, characterized in that, The grid splitting process outputs a splitting candidate when the splitting sample size threshold is met and the risk distribution skewness reaches the threshold. Before the splitting is implemented, the continuity of the protective strip, the minimum width, and the conflict of the protection level are checked. If the check fails, the output is retained, and the region status of the corresponding grid is marked as a conflict pending state.

6. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 4, characterized in that, The adjacent grid merging is output as a merging candidate when the merging sample size threshold is met and the difference in risk level between adjacent grids does not exceed the merging similarity threshold; and when adjacent grids belong to different protection zone protection levels, or one side is within the coverage area indicated by the protection zone coverage relationship while the other side is not within the coverage area, the output is kept to avoid merging across the protection zone boundary.

7. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 4, characterized in that, When performing grid splitting, the candidate grids are subdivided into sub-grids according to their spatial layout, and new grid records are written into the edge risk grid structure. When performing adjacent grid merging, new merged grid records are generated and the window caliber statistical summary of each grid participating in the merging is summarized. The correspondence between grids before and after splitting or merging is recorded in the above splitting or merging process, and the window caliber statistical summary is migrated or redistributed to the sub-grids obtained from splitting or the new grids obtained from merging.

8. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 4, characterized in that, The edge risk grid structure is a maintenance area status label for each grid, and the area status label includes at least a stable state, a high-risk state, a sample shortage state, and a conflict pending state. Furthermore, the transition from a stable state to a high-risk state requires meeting the condition of reaching the high-risk threshold for multiple consecutive maintenance cycles. The region status markers and grid divisions are jointly written into the edge space risk skeleton structure for reference in accidental touch detection and action decision-making.

9. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 1, characterized in that, The migration initialization is achieved through a risk skeleton migration mechanism. When a change in the application interface version is detected, the risk skeleton migration mechanism determines the source control area and the target control area based on the control semantic correspondence registered in the interface version migration configuration structure, and calculates the spatial overlap metric as the migration weight. When the spatial overlap metric is not lower than the effective overlap threshold, the long-term statistics of the edge risk grid structure and the high-consequence operation area protection zone structure are subjected to weighted migration initialization and written to the version tag.

10. The machine learning-based method for detecting and intelligently correcting accidental touches on a liquid crystal touchscreen according to claim 9, characterized in that, When there is no valid semantic correspondence between controls or the spatial overlap metric is lower than the valid overlap threshold, the risk skeleton migration mechanism determines that there is no valid migration source, and performs stronger attenuation, clearing, or resetting initialization on the long-term statistics corresponding to the target control area according to the migration attenuation rule and writes a new version tag; and the migration processing is limited to only acting on the long-term statistics and version tag, and does not directly rewrite the protection zone coverage relationship and area status.

Images