A control method and system of an intelligent mouse

Abstract

This invention relates to the field of human-computer interaction and intelligent control technology, and discloses a control method and system for an intelligent mouse. The method comprises five steps: user action acquisition, action feature encoding, semantic intent recognition, mouse behavior mapping, and feedback adjustment and optimization. The system comprises a motion acquisition and feature extraction module, an intent recognition and control signal generation module, and a behavior feedback and adaptive optimization module. By integrating an inertial measurement unit and a pressure sensor to acquire three-dimensional hand motion data, combining temporal convolutional coding and a Transformer network to recognize user operation intent, and generating mouse control signals based on context, precise control and adaptive optimization of the mouse are achieved. This invention enables complex operations without physical buttons, improving the personalization and intelligence of mouse control. It is applicable to intelligent interactive devices, auxiliary input systems, and other scenarios.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction and intelligent control technology, and in particular to a control method and system for an intelligent mouse. Background Technology

[0002] As the most commonly used input device for computers, the mouse's basic structure and interaction methods have remained largely unchanged since its invention in the last century. It primarily relies on the user's fingers operating physical buttons and a scroll wheel, combined with an optical positioning module to achieve two-dimensional coordinate control. While this control method is mature and stable, its input dimension is low, and its expressive capabilities are limited, making it difficult to meet the demands of complex human-computer interaction scenarios such as multi-tasking collaboration, immersive interaction, and personalized operation. In emerging interactive environments such as augmented reality and virtual reality, users require more natural control methods, making the limitations of traditional mouse devices increasingly apparent.

[0003] To address the limitations of traditional interaction methods, some research has introduced input mechanisms based on gesture or motion recognition, such as using inertial measurement units, cameras, or pressure sensors to capture users' dynamic behaviors and attempting to map them into mouse control signals. However, existing technologies suffer from numerous problems in practical applications, such as low accuracy and stability of motion recognition, significant input latency, and difficulty in accurately understanding user intentions, especially prone to misjudgments in continuous or complex operations. Furthermore, most systems neglect the correlation between actions and context, resulting in a lack of judgment and adjustment capabilities in dynamic scenarios.

[0004] On the other hand, current mainstream action recognition methods mostly focus on processing single frames or short time sequences, lacking modeling of long-term dependencies and contextual logic, making it difficult to effectively mine deep semantic intentions of users. Meanwhile, traditional mouse control behaviors are mostly fixed mappings, unable to dynamically adjust according to user habits, custom preferences, or scene states, severely limiting their development potential in intelligent interaction. Therefore, there is an urgent need for an intelligent mouse control solution that combines high-dimensional sensor data, semantic recognition, and behavior optimization mechanisms to achieve a more efficient and natural human-computer interaction process. Summary of the Invention

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solution: a control method for an intelligent mouse, comprising the following steps:

[0006] Step S1: User motion acquisition, acquiring user hand micro-motion data through the motion sensing module. And perform preliminary filtering to extract effective action feature vectors. ;

[0007] Step S2: Action feature encoding, encoding the action feature vector. Perform timing encoding to generate an action-state matrix. To express the temporal evolution relationship between consecutive actions;

[0008] Step S3: Semantic intent recognition, which involves analyzing the action state matrix. Input is fed into a deep semantic recognition network, and the output is the user intent. ;

[0009] Step S4: Mouse behavior mapping, based on user intent With contextual data Generate multidimensional control signals It is used to control mouse movement, click, and scrolling behavior;

[0010] Step S5: Feedback, adjustment, and optimization; collect execution results. Through feedback learning modules and historical user preference data Jointly perform adaptive model updates to form an adjustment parameter matrix. And optimize the processing path of steps S1-S4 in reverse.

[0011] Preferably, step S1 includes:

[0012] A motion sensing module, consisting of an integrated inertial measurement unit and a capacitive pressure sensor, acquires triaxial acceleration data in real time. Angular velocity data and finger pressure data ;

[0013] The original signal is then denoised and smoothed to generate an initial set of action features. Then, a multi-channel weight normalization algorithm is introduced to extract effective action feature vectors. The formula is as follows:

[0014] ;

[0015] in, , To adjust the weights, and satisfy =1, Represents the first in the initial set of action features One portion, Represents action feature vector No. Each component.

[0016] Preferably, step S2 includes:

[0017] Based on the Finite Impulse Response Convolutional Coding Network, the action feature vector is... Transform into an action-state matrix :

[0018] ;

[0019] in, For convolution kernel, For timing frame numbering, The width of the window;

[0020] This structure is used to construct an action-time evolution graph, which is then further utilized using a position encoding function. :

[0021] ;

[0022] in, Action feature vector Dimensions For dimensional indexing.

[0023] Preferably, step S3 includes:

[0024] A multi-head attention mechanism based on the Transformer structure is adopted to process the action-state matrix. As input, an intent vector set is formed. The user intent is output using the following formula. :

[0025] ;

[0026] in, Represents the set of intent vectors transpose, This represents the convolution operation. Action feature vector Dimensions.

[0027] Preferably, step S4 includes:

[0028] Constructing a set of intent-behavior mapping functions :

[0029] ;

[0030] in, This represents contextual data derived from user input. Indicates multidimensional control signals. , Indicates mouse movement. Indicates a click. Indicates scrolling behavior;

[0031] Combined with the current screen distribution matrix of the user With window focus position Mouse movement can be optimized using the following adjustment functions:

[0032] ;

[0033] in, , It is a dynamic adjustment coefficient, and satisfies + =1, The focus position of the window is reflected on the computer. The current screen distribution matrix is ​​obtained through integrated frame buffer access.

[0034] Preferably, step S5 includes:

[0035] Collect behavioral response data ,Will With multidimensional control signals and historical user preference data Input the regression residual module to construct the error feedback matrix. :

[0036] ;

[0037] Among them, behavioral response data Historical user preference data captured in real time by the Windows system's Raw Input API. Obtained through long-term user interaction statistics;

[0038] Using the backpropagation mechanism, based on the error feedback matrix Generate adjustment parameter matrix And reverse-optimize the processing path of steps S1-S4, adjusting the parameter matrix. The algorithm formula is as follows:

[0039] ;

[0040] in, For user intent action feature vector The first-order partial derivative, Action feature vector Initial set of action features The first-order partial derivative, Represents the initial set of action features To user intent How does the entire path affect the final behavior? For the transpose of multidimensional control signals, This is historical user preference data.

[0041] This invention also provides a control system for an intelligent mouse, comprising the following modules:

[0042] The motion acquisition and feature extraction module is used to collect user hand motion data in real time through motion sensing devices, including acceleration, angular velocity and finger pressure information. It preprocesses the collected raw data, removes noise signals, smooths motion curves, and extracts high-dimensional motion feature vectors that reflect the user's operation intentions. This completes the structured expression of the user's micro-motion behavior and provides basic data support for subsequent temporal coding and semantic analysis.

[0043] The intent recognition and control signal generation module is used to receive the action feature vector output by the action acquisition and feature extraction module. By embedding a temporal analysis mechanism and a deep semantic understanding network, it performs state modeling and semantic reasoning on continuous action behaviors, extracts the user's intent information within a specific time period, and generates multi-dimensional mouse control signals, including displacement, click, and scrolling, in combination with the current application context environment, so as to drive the mouse device to perform operation behaviors that match the user's intent.

[0044] The behavior feedback and adaptive optimization module is used to collect system response data after the smart mouse performs actual operations, including cursor position changes, click results, and scrolling feedback. By comparing and analyzing the execution results with the original control signals and historical user preferences, a behavior error feedback mechanism is constructed. Based on this, the internal parameters of the system are dynamically updated, and the action feature extraction method, intent recognition model, and control signal generation strategy are adaptively adjusted to gradually improve the accuracy and personalized response capability of mouse control, forming a continuously evolving user interaction experience.

[0045] The motion acquisition and feature extraction module includes:

[0046] The motion sensing acquisition unit is used to acquire dynamic data of the user's hand in three-dimensional space in real time through an integrated inertial measurement unit and capacitive sensor combination device, including acceleration information, angular velocity information and pressure information applied to various points on the mouse surface;

[0047] The feature extraction and processing unit is used to perform denoising, filtering and normalization processing on the original sensing data to generate an initial feature set and form a high-dimensional feature vector that can be used to express the user's behavioral intention.

[0048] The intent recognition and control signal generation module includes:

[0049] The temporal coding unit is used to perform time series modeling on the input action feature vector. Through the sliding window mechanism and convolutional coding network, it captures the temporal evolution relationship in the user's continuous operation and generates the action state matrix.

[0050] The intent reasoning and control mapping unit is used to receive the action state matrix, perform intent recognition using the Transformer architecture, output a semantic vector representing the current operation purpose, and then combine it with external contextual environment data to construct a behavior mapping model and generate corresponding multi-dimensional mouse control signals.

[0051] The intent recognition and control signal generation module includes:

[0052] The temporal coding unit is used to perform time series modeling on the input action feature vector. Through the sliding window mechanism and convolutional coding network, it captures the temporal evolution relationship in the user's continuous operation and generates the action state matrix.

[0053] The intent reasoning and control mapping unit is used to receive the action state matrix, perform intent recognition using the Transformer architecture, output a semantic vector representing the current operation purpose, and then combine it with external contextual environment data to construct a behavior mapping model and generate corresponding multi-dimensional mouse control signals.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] Enhancing the naturalness and multidimensionality of human-computer interaction: This invention introduces action recognition, sensor data fusion, and semantic parsing mechanisms, enabling the mouse to not only perceive the user's traditional click behavior, but also accurately recognize the user's gestures, inertial movements, and intentions, significantly enhancing the user's expressive ability and achieving a more natural and fluid multimodal input experience.

[0056] Improving the accuracy and robustness of action recognition: This invention adopts a multi-source sensor collaborative acquisition strategy and combines it with a feature extraction algorithm based on time series modeling. This can effectively suppress background interference and posture noise during the recognition process, improve the accuracy and stability of action classification, and ensure accurate control and low misjudgment rate in dynamic environments.

[0057] Achieving personalized and scenario-adaptive control: By constructing a multi-layer semantic intent mapping model and a user behavior preference database, this invention can dynamically adjust the control logic according to the user's current operation context, realize self-learning and adaptive optimization of mouse control methods, and significantly improve the system's interaction efficiency and user satisfaction in different application scenarios. Attached Figure Description

[0058] Figure 1 A flowchart illustrating the method steps provided in this application;

[0059] Figure 2 A schematic diagram of the system modules provided in this application. Detailed Implementation

[0060] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0061] refer to Figure 1 This invention provides a method for controlling a smart mouse, comprising the following steps:

[0062] Step S1: User motion acquisition, acquiring user hand micro-motion data through the motion sensing module. And perform preliminary filtering to extract effective action feature vectors. ;

[0063] Step S2: Action feature encoding, encoding the action feature vector. Perform timing encoding to generate an action-state matrix. To express the temporal evolution relationship between consecutive actions;

[0064] Step S3: Semantic intent recognition, which involves analyzing the action state matrix. Input is fed into a deep semantic recognition network, and the output is the user intent. ;

[0065] Step S4: Mouse behavior mapping, based on user intent With contextual data Generate multidimensional control signals It is used to control mouse movement, click, and scrolling behavior;

[0066] Step S5: Feedback, adjustment, and optimization; collect execution results. Through feedback learning modules and historical user preference data Jointly perform adaptive model updates to form an adjustment parameter matrix. And optimize the processing path of steps S1-S4 in reverse.

[0067] In step S1, the user's hand movements are first dynamically monitored by a motion sensing module constructed from an integrated inertial measurement unit and a capacitive pressure sensor. The integrated inertial measurement unit includes a three-axis accelerometer and a three-axis gyroscope, which are used to acquire acceleration data during the user's hand movements. With angular velocity data Capacitive pressure sensors are used to collect touch pressure values ​​from key fingers such as the thumb and index finger. This is to enhance the stability and accuracy of behavioral perception.

[0068] After data acquisition, the system uses wavelet denoising and an exponential moving average filter to process the raw data, removing sampling errors, jitter, and environmental interference. The three types of processed data are combined to form the initial set of action features. To enhance the contribution of signals from different channels to overall behavior recognition, the system introduces a multi-channel weight normalization algorithm to generate action feature vectors. , its first The formula for calculating each component is:

[0069] ;

[0070] in, , As weight parameters, they must satisfy... =1, used to balance the importance of channels such as acceleration, angular velocity and pressure in motion recognition.

[0071] In step S2, based on the action feature vector generated in the previous step... The system constructs a finite impulse response convolutional coding network to model the dynamic state sequence of a user's hand movements over time. This network contains multiple time windows, each with convolutional kernel parameters of [parameter values ​​missing]. The continuous action state matrix is ​​generated using the following formula. :

[0072] ;

[0073] in, Indicates the width of the convolution window. For the current time frame, convolution operations can preserve the temporal continuity of action changes.

[0074] To further enhance the feature representation capability in the time dimension, the system introduces a positional encoding mechanism, through a positional encoding function. This embeds the location of each time frame into the feature matrix, compensating for the limitation of convolutional kernels in capturing absolute temporal location information. Here... For feature dimension, For dimensional indexing.

[0075] The final matrix It fully preserves the evolution pattern of user actions over time and can accurately represent the transition relationship between different action types (click, swipe, etc.).

[0076] In step S3, the system utilizes a multi-head attention mechanism based on the Transformer architecture to process the action-state matrix. This process generates a set of user intent vectors. The Transformer module can extract contextual relationships between multiple features in parallel and capture semantic changes under subtle differences in actions.

[0077] Each intent vector This represents the system's understanding of a potential action (click, right-click, scroll wheel, etc.). The end-user intent is calculated using the following attention score function. :

[0078] ;

[0079] in, Given the vector dimension, multiplying the dot product by the normalization coefficient can prevent gradient explosion caused by high-dimensional data.

[0080] Step S4: The system constructs a set of intent-behavior mapping functions. User intent With contextual data Combined, a set of control signals is generated. .in, Information such as the current application type, the state of the focused window, and the screen layout is dynamically obtained through the system API.

[0081] The system is designed with an intelligent adjustment mechanism for screen space, based on the screen distribution matrix. With window focus position Generate optimized displacement signals:

[0082] ;

[0083] parameter and To adjust the coefficients in real time, they are dynamically calculated based on the operation frequency and the size of the target area to meet the requirements. + =1 ensures that the mouse pointer moves quickly to the area the user expects, avoiding inaccurate positioning or delays.

[0084] In step S5, the system constructs a behavior feedback mechanism to continuously optimize the aforementioned processing path based on the user's operation response. The system first collects execution behavior response data. This includes the actual mouse movement path, click response, and scroll range, all captured in real time via Windows' Raw Input API.

[0085] Combined with the generated control signals and historical user preference data Construct the error feedback matrix :

[0086] ;

[0087] This error matrix is ​​used to quantify the deviation between the current control behavior and the intended purpose. The system introduces a residual regression analysis model to attribute the deviation and constructs the adjustment parameter matrix using the following formula. :

[0088] ;

[0089] Through the above matrix multiplication combination, the system can clearly track the initial action signal. To the final intention decision The system identifies sensitive paths and pinpoints the most critical areas for optimization (e.g., unreasonable feature weights in a specific dimension, insufficient convolutional kernel learning, etc.). Parameters are then adjusted. The system is applied in real time to the aforementioned steps S1 to S4, completing a round of dynamic optimization iterations to ensure that the system continues to adapt as user behavior habits evolve, thereby improving the overall interactive experience.

[0090] refer to Figure 2 This invention provides a control system for an intelligent mouse, comprising the following modules:

[0091] The motion acquisition and feature extraction module is used to collect user hand motion data in real time through motion sensing devices, including acceleration, angular velocity and finger pressure information. It preprocesses the collected raw data, removes noise signals, smooths motion curves, and extracts high-dimensional motion feature vectors that reflect the user's operation intentions. This completes the structured expression of the user's micro-motion behavior and provides basic data support for subsequent temporal coding and semantic analysis.

[0092] The intent recognition and control signal generation module is used to receive the action feature vector output by the action acquisition and feature extraction module. By embedding a temporal analysis mechanism and a deep semantic understanding network, it performs state modeling and semantic reasoning on continuous action behaviors, extracts the user's intent information within a specific time period, and generates multi-dimensional mouse control signals, including displacement, click, and scrolling, in combination with the current application context environment, so as to drive the mouse device to perform operation behaviors that match the user's intent.

[0093] The behavior feedback and adaptive optimization module is used to collect system response data after the smart mouse performs actual operations, including cursor position changes, click results, and scrolling feedback. By comparing and analyzing the execution results with the original control signals and historical user preferences, a behavior error feedback mechanism is constructed. Based on this, the internal parameters of the system are dynamically updated, and the action feature extraction method, intent recognition model, and control signal generation strategy are adaptively adjusted to gradually improve the accuracy and personalized response capability of mouse control, forming a continuously evolving user interaction experience.

[0094] The motion acquisition and feature extraction module includes:

[0095] The motion sensing acquisition unit is used to acquire dynamic data of the user's hand in three-dimensional space in real time through an integrated inertial measurement unit and capacitive sensor combination device, including acceleration information, angular velocity information and pressure information applied to various points on the mouse surface;

[0096] The feature extraction and processing unit is used to perform denoising, filtering and normalization processing on the original sensing data to generate an initial feature set and form a high-dimensional feature vector that can be used to express the user's behavioral intention.

[0097] The intent recognition and control signal generation module includes:

[0098] The temporal coding unit is used to perform time series modeling on the input action feature vector. Through the sliding window mechanism and convolutional coding network, it captures the temporal evolution relationship in the user's continuous operation and generates the action state matrix.

[0099] The intent reasoning and control mapping unit is used to receive the action state matrix, perform intent recognition using the Transformer architecture, output a semantic vector representing the current operation purpose, and then combine it with external contextual environment data to construct a behavior mapping model and generate corresponding multi-dimensional mouse control signals.

[0100] The behavior feedback and adaptive optimization module includes:

[0101] The behavior acquisition and error feedback unit is used to acquire the system execution results after each round of mouse control operation. The system execution results include the final cursor position, operation success rate, and click response data. These results are compared with the original control signal and the user's long-term preference data to construct an error feedback matrix and quantify the deviation between the operation execution and the user's intention.

[0102] The parameter adjustment and model optimization unit is used to dynamically update the key parameters in the action feature extraction strategy, temporal modeling structure and control mapping function based on the error feedback matrix and using the gradient backpropagation mechanism. It also continuously learns the user's preference features in different application scenarios to achieve system adaptive optimization, thereby improving the personalization and accuracy of mouse response.

[0103] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.

[0104] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.

Claims

1. A control method for an intelligent mouse, characterized in that, Includes the following steps: Step S1: User motion acquisition, acquiring user hand micro-motion data through the motion sensing module. And perform preliminary filtering to extract effective action feature vectors. ; Step S2: Action feature encoding, encoding the action feature vector. Perform timing encoding to generate an action-state matrix. To express the temporal evolution relationship between consecutive actions; Step S3: Semantic intent recognition, which involves analyzing the action state matrix. Input is fed into a deep semantic recognition network, and the output is the user intent. ; Step S4: Mouse behavior mapping, based on user intent With contextual data Generate multidimensional control signals It is used to control mouse movement, click, and scrolling behavior; Step S5: Feedback, adjustment, and optimization; collect execution results. Through feedback learning modules and historical user preference data Jointly perform adaptive model updates to form an adjustment parameter matrix. And optimize the processing path of steps S1-S4 in reverse.

2. The control method for an intelligent mouse according to claim 1, characterized in that, Step S1 includes: A motion sensing module, consisting of an integrated inertial measurement unit and a capacitive pressure sensor, acquires triaxial acceleration data in real time. Angular velocity data and finger pressure data ; The original signal is then denoised and smoothed to generate an initial set of action features. Then, a multi-channel weight normalization algorithm is introduced to extract effective action feature vectors. The formula is as follows: ； in, , To adjust the weights, and satisfy =1, Represents the first in the initial set of action features One portion, Represents action feature vector No. Each component.

3. The control method for an intelligent mouse according to claim 1, characterized in that, Step S2 includes: Based on the Finite Impulse Response Convolutional Coding Network, the action feature vector is... Transform into an action-state matrix : ； in, For convolution kernel, For timing frame numbering, The width of the window; An action-time evolution graph is constructed using the finite impulse response convolutional coding network, and a positional coding function is further introduced. To compensate for the limitation of convolutional kernels in capturing absolute temporal location information, the location encoding function... Specifically as follows: ； in, Action feature vector Dimensions For dimensional indexing.

4. The control method for an intelligent mouse according to claim 1, characterized in that, Step S3 includes: A multi-head attention mechanism based on the Transformer structure is adopted to process the action-state matrix. As input, an intent vector set is formed. The user intent is output using the following formula. : ； in, Represents the set of intent vectors transpose, This represents the convolution operation. Action feature vector Dimensions.

5. The control method for an intelligent mouse according to claim 1, characterized in that, Step S4 includes: Constructing a set of intent-behavior mapping functions : ； in, This represents contextual data derived from user input. Indicates multidimensional control signals. , Indicates mouse movement. Indicates a click. Indicates scrolling behavior; Combined with the current screen distribution matrix of the user With window focus position Mouse movement can be optimized using the following adjustment functions: ； in, , It is a dynamic adjustment coefficient, and satisfies + =1, The focus position of the window is reflected on the computer. The current screen distribution matrix is ​​obtained through integrated frame buffer access.

6. The control method for an intelligent mouse according to claim 1, characterized in that, Step S5 includes: Collect behavioral response data ,Will With multidimensional control signals and historical user preference data Input the regression residual module to construct the error feedback matrix. : ； Among them, behavioral response data Historical user preference data captured in real time by the Windows system's Raw Input API Obtained through long-term user interaction statistics; Using the backpropagation mechanism, based on the error feedback matrix Generate adjustment parameter matrix And reverse-optimize the processing path of steps S1-S4, adjusting the parameter matrix. The algorithm formula is as follows: ； in, For user intent For action feature vectors The first-order partial derivative, Action feature vector Initial set of action features The first-order partial derivative, Represents the initial set of action features To user intent How does the entire path affect the final behavior? For the transpose of multidimensional control signals, This is historical user preference data.

7. A control system for an intelligent mouse, characterized in that, Includes the following modules: The motion acquisition and feature extraction module is used to collect user hand motion data in real time through motion sensing devices, including acceleration, angular velocity and finger pressure information. It preprocesses the collected raw data, removes noise signals, smooths motion curves, and extracts high-dimensional motion feature vectors that reflect the user's operation intentions. This completes the structured expression of the user's micro-motion behavior and provides basic data support for subsequent temporal coding and semantic analysis. The intent recognition and control signal generation module is used to receive the action feature vector output by the action acquisition and feature extraction module. By embedding a temporal analysis mechanism and a deep semantic understanding network, it performs state modeling and semantic reasoning on continuous action behaviors, extracts the user's intent information within a specific time period, and generates multi-dimensional mouse control signals, including displacement, click, and scrolling, in combination with the current application context environment, so as to drive the mouse device to perform operation behaviors that match the user's intent. The behavior feedback and adaptive optimization module is used to collect system response data after the smart mouse performs actual operations, including cursor position changes, click results, and scrolling feedback. By comparing and analyzing the execution results with the original control signals and historical user preferences, a behavior error feedback mechanism is constructed. Based on this, the internal parameters of the system are dynamically updated, and the action feature extraction method, intent recognition model, and control signal generation strategy are adaptively adjusted to gradually improve the accuracy and personalized response capability of mouse control, forming a continuously evolving user interaction experience.

8. The control system for an intelligent mouse according to claim 7, characterized in that, The motion acquisition and feature extraction module includes: The motion sensing acquisition unit is used to acquire dynamic data of the user's hand in three-dimensional space in real time through an integrated inertial measurement unit and capacitive sensor combination device, including acceleration information, angular velocity information and pressure information applied to various points on the mouse surface; The feature extraction and processing unit is used to denoise, filter and normalize the raw sensing data acquired by the motion sensing acquisition unit to generate an initial feature set and form a high-dimensional feature vector that can be used to express the user's behavioral intention.

9. The control system for an intelligent mouse according to claim 7, characterized in that, The intent recognition and control signal generation module includes: The temporal coding unit is used to perform time series modeling on the input action feature vector. Through the sliding window mechanism and convolutional coding network, it captures the temporal evolution relationship in the user's continuous operation and generates the action state matrix. The intent reasoning and control mapping unit is used to receive the action state matrix, perform intent recognition using the Transformer architecture, output a semantic vector representing the current operation purpose, and then combine it with external contextual environment data to construct a behavior mapping model and generate corresponding multi-dimensional mouse control signals.

10. The control system for an intelligent mouse according to claim 7, characterized in that, The behavior feedback and adaptive optimization module includes: The behavior acquisition and error feedback unit is used to acquire the system execution results after each round of mouse control operation. The system execution results include the final cursor position, operation success rate, and click response data. These results are compared with the original control signal and the user's long-term preference data to construct an error feedback matrix and quantify the deviation between the operation execution and the user's intention. The parameter adjustment and model optimization unit is used to dynamically update the key parameters in the action feature extraction strategy, temporal modeling structure and control mapping function based on the error feedback matrix and using the gradient backpropagation mechanism. It also continuously learns the user's preference features in different application scenarios to achieve system adaptive optimization, thereby improving the personalization and accuracy of mouse response.

Images