Man-machine interaction identity intelligent identification method and system based on multi-mode perception

Abstract

The invention provides a man-machine interaction identity intelligent identification method and system based on multi-modal perception, and relates to the technical field of man-machine interaction, and the method comprises the steps: firstly capturing a behavior signal level sequence of an interaction main body in man-machine interaction, triggering a multi-dimensional perception feature derivative transmission generation feature derivative transmission unit, and generating a feature derivative transmission unit; the feature derivation transmission unit contains derivation and transmission relation data of physiology, behavior patterns, response feedback and other features, then the units are connected in series in a chain manner according to a progressive sequence to form a feature derivation transmission time sequence chain, a feature dynamic evolution track is recorded, and then a preset identity reference chain library is called; according to the method, scene parameters and historical data are combined to dynamically update and generate a scene adaptation reference chain, then cross-dimension core matching features are extracted, bidirectional track alignment is performed by using a dynamic weight distribution strategy, finally, an identity recognition conclusion containing information such as identity attribution is generated based on an alignment result, and the method improves the accuracy, adaptability and intelligent level of identity recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of human-computer interaction, in particular to a human-computer interaction identity intelligent recognition method and system based on multi-modal perception. BACKGROUND

[0002] In the field of human-computer interaction, accurately identifying the identity of the interaction subject is crucial for ensuring system security, providing personalized services, and achieving precise interaction control. Traditional human-computer interaction identity recognition methods mainly rely on single biometric recognition technology, such as fingerprint recognition, facial recognition, etc. Although the above methods can achieve identity recognition to some extent, they have obvious limitations.

[0003] Single biometric recognition is easily affected by environmental factors, for example, fingerprint recognition accuracy will decrease significantly when the finger is wet or has stains; facial recognition also has difficulty in accurately working when the lighting conditions are poor or the face is obscured. Moreover, single features are easily forged or imitated, and the security is low. In addition, existing methods do not fully consider the dynamic behavior characteristics of the interaction subject in the human-computer interaction process and the influence of different interaction scenarios on identity recognition, and cannot dynamically adjust and optimize according to the real-time state of the interaction and historical interaction data, resulting in insufficient accuracy and adaptability of identity recognition, making it difficult to meet the increasingly complex and diversified human-computer interaction needs. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a human-computer interaction identity intelligent recognition method based on multi-modal perception, which comprises: Capturing the behavior signal hierarchical sequence of the interaction subject in the human-computer interaction process, triggering multi-dimensional perception feature derivation transmission based on the gradient features of the behavior signal hierarchical sequence, generating a feature derivation transmission unit, which contains derived data and transmission relationship data of physiological perception features, behavior pattern features, response feedback features progressing with the interaction level; According to the progressive order of the behavior signal hierarchical sequence, the feature derivation transmission unit is processed in chain series to form a feature derivation transmission time sequence chain, which records the dynamic evolution trajectory data of the feature derivation relationship data and transmission path data with the advancement of the interaction level; Accessing a preset identity reference chain library, the preset identity reference chain library stores initial reference chains corresponding to known identities, and dynamically updates the initial reference chains by a self-learning evolution model combined with current interaction scene parameters and historical interaction data to generate a scene-adapted reference chain; Extract the cross-dimensional core matching features of the feature derivation and transmission time-series chain and the scene adaptation benchmark chain, perform bidirectional trajectory alignment processing based on the dynamic weight allocation strategy, and generate trajectory alignment results. The dynamic weight allocation strategy adjusts the matching weight values ​​in real time according to the importance of the features. An identity recognition conclusion is generated based on the trajectory alignment result. The identity recognition conclusion includes the identity attribution information of the interactive subject, feature derivation and transmission matching data, and weight distribution data.

[0005] In another aspect, embodiments of the present invention also provide a human-computer interaction intelligent identification system based on multimodal perception, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to run the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0006] Based on the above, this invention captures the hierarchical sequence of behavioral signals of the interactive subject during human-computer interaction, and triggers multi-dimensional perceptual feature derivation and transmission based on its gradient features. This generates feature derivation and transmission units containing rich derivation data and transmission relationship data. These units are then chained together to form a feature derivation and transmission time-series chain, recording the dynamic evolution trajectory data of feature derivation relationships and transmission paths as the interaction level progresses. This accurately reflects the feature change patterns of the interactive subject throughout the entire interaction process. A preset identity benchmark chain library is retrieved, and the initial benchmark chain is dynamically updated using current interaction scenario parameters and historical interaction data to generate a scenario-adaptive benchmark chain. This allows the benchmark chain to better adapt to different interaction scenarios, improving the scenario adaptability of identity recognition. Cross-dimensional core matching features are extracted, and bidirectional trajectory alignment is performed based on a dynamic weight allocation strategy. This allows for real-time adjustment of matching weight values ​​according to feature importance, ensuring that key features play a greater role in identity recognition and improving the accuracy and reliability of identity recognition. Finally, based on the trajectory alignment results, an identity recognition conclusion containing identity attribution information, feature derivation and transmission matching data, and weight distribution data is generated, providing accurate results for identity recognition and effectively improving the overall performance and intelligence level of human-computer interaction identity recognition. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the human-computer interaction identity intelligent recognition method based on multimodal perception provided in the embodiments of the present invention.

[0008] Figure 2 This is a schematic diagram of the hardware architecture of the human-computer interaction intelligent identity recognition system based on multimodal perception provided in an embodiment of the present invention. Detailed Implementation

[0009] Figure 1 This is a flowchart illustrating a human-computer interaction identity intelligent recognition method based on multimodal perception, provided in one embodiment of the present invention. A detailed description follows.

[0010] Step S110: Capture the hierarchical sequence of behavioral signals of the interactive subject during human-computer interaction, trigger multi-dimensional perception feature derivation and transmission based on the gradient features of the hierarchical sequence of behavioral signals, and generate feature derivation and transmission units. The feature derivation and transmission units include physiological perception features, behavioral pattern features, response feedback features, derivation data of interaction levels, and transmission relationship data.

[0011] In this embodiment, the process of unlocking a smartphone and subsequent operations is used as the human-computer interaction scenario. The user performs a series of operations on the smartphone, from unlocking to opening a specific application. During this process, the system continuously captures various behavioral signals generated by the user and processes them in a hierarchical sequence.

[0012] Step S111: Continuously capture the hierarchical sequence of behavioral signals generated by the interactive subject in the human-computer interaction interface. The hierarchical sequence of behavioral signals includes the hierarchical electrical signal sequence formed by touch operation, the hierarchical sound wave signal sequence formed by voice interaction, and the hierarchical visual signal sequence formed by posture and movement. Each signal level corresponds to a behavioral complexity level of the interactive subject.

[0013] In the aforementioned smartphone interaction scenario, the user first performs a screen touch unlock operation, which generates a sequence of touch electrical signals. After unlocking, the user uses the voice command "Open Email Application" to generate a sequence of hierarchical sound wave signals. Simultaneously, the front-facing camera captures the user's head posture and movements when uttering the voice, forming a sequence of hierarchical visual signals. These signals are divided into different levels according to the complexity of the behavior. For example, touch unlock is the basic level, voice command is the intermediate level, and posture and movements are the auxiliary level, which together constitute the hierarchical sequence of behavioral signals.

[0014] Step S112: Perform hierarchical decomposition processing on the behavior signal hierarchy sequence, and divide it into interaction level one signal, interaction level two signal, and interaction level three signal according to the signal complexity classification standard. Each level signal includes corresponding signal strength parameters, change rate parameters, and duration period parameters.

[0015] The captured behavioral signal sequence is broken down into three levels. The first level signal is the touch electrical signal generated by the touch unlock operation. Its signal strength parameter represents the range of touch pressure, the rate of change parameter reflects the change of touch pressure over time, and the duration parameter is the time from when the finger touches the screen to when the unlock operation is completed. The second level signal is the sound wave signal corresponding to the voice command "Open Email Application." Its signal strength parameter reflects the decibel range of the sound, the rate of change parameter is how fast the sound wave frequency changes, and the duration parameter is the duration of the voice command. The third level signal is the visual signal of head posture movements. Its signal strength parameter is the significance of posture changes in the image, the rate of change parameter is the speed of posture change, and the duration parameter is the time from the start to the end of the posture movement.

[0016] Step S113: Extract the gradient features of each level signal. The gradient features include signal complexity gradient, intensity change gradient, and period change gradient. The gradient features record the data of the progressive process of interactive behavior.

[0017] For the interaction level one signal, its signal complexity gradient is calculated, which represents the trend of complexity change of the touch electrical signal from simple touch to the completion of the unlock pattern during the unlocking process; the intensity change gradient is the rate of change of touch pressure from initial contact to stable pressing and then to lifting; and the periodic change gradient reflects the rhythmic change of the unlocking operation time. For the interaction level two signal, the signal complexity gradient reflects the change in the complexity of word combinations in the voice command; the intensity change gradient is the rate of change of sound decibels from the start of utterance to the stabilization of sound and then to its end; and the periodic change gradient is the rhythmic feature of the speech duration. For the interaction level three signal, the signal complexity gradient represents the increasing complexity of head posture movements from simple rotation to specific postures; the intensity change gradient is the rate of change of the amplitude of posture changes; and the periodic change gradient is the rhythmic change of the duration of the posture movement.

[0018] Step S114: Based on the gradient features, set the feature derivation direction corresponding to each level signal. The feature derivation direction points to the new feature type that needs to be derived from the existing perceptual features at that level. The interaction level one signal corresponds to the initial feature derivation, and the interaction level two signal and the interaction level three signal correspond to the composite feature derivation.

[0019] Based on the gradient characteristics of the interaction level one signal, its feature derivation direction is set to the derivation of initial physiological characteristics and initial behavioral characteristics, because touch unlocking operation mainly reflects the physiological characteristics of the interactive subject, such as fingerprints, and the basic behavioral characteristics of finger operation. The gradient characteristics of the interaction level two signal show higher complexity, and its feature derivation direction is determined to be the derivation of complex physiological characteristics (such as voiceprint characteristics) and complex behavioral characteristics (such as pronunciation habits of voice commands). The interaction level three signal serves as an auxiliary signal, and its feature derivation direction is the derivation of complex response characteristics (such as the coordinated response of head posture and voice commands).

[0020] Step S115: For each feature derivation direction, call the corresponding multi-dimensional perception acquisition module. Each perception acquisition module is responsible for acquiring one initial feature. Multiple perception acquisition modules are started synchronously based on the continuous cycle of hierarchical signals to acquire initial physiological features, initial behavioral features, and initial response features.

[0021] For the initial feature derivation direction of the interaction level one signal, the fingerprint acquisition module and the touch acquisition module are invoked. The fingerprint acquisition module is responsible for acquiring fingerprint texture data from the initial physiological features, and the touch acquisition module is responsible for acquiring operation rhythm data from the initial behavioral features. These two modules start synchronously at the start of the touch operation, based on the duration of the interaction level one signal, and the acquisition duration is consistent with the duration of touch unlocking. For the composite feature derivation direction corresponding to the interaction level two signal, the audio acquisition module is invoked to acquire voiceprint frequency data from the initial physiological features, as well as voice interaction interval data from the operation rhythm data. The audio acquisition module starts at the start of the voice command, and the acquisition duration is the duration of the voice command. For the composite response feature derivation direction of the interaction level three signal, the visual acquisition module is invoked to acquire head posture feedback delay data from the initial response features. The visual acquisition module starts at the start of the head posture movement, and the acquisition duration is the same as the duration of the posture movement.

[0022] Step S1151: Based on the characteristic derivation direction of each level of signal, set the initial feature type to be collected. The initial feature type includes fingerprint texture data, facial contour data, voiceprint frequency data, and palm vein data in the initial physiological features; operation rhythm data, movement amplitude data, and interaction interval data in the initial behavioral features; and feedback delay data, response content data, and operation accuracy data in the initial response features.

[0023] In the feature derivation direction of the interaction level one signal, the initial feature types to be collected are fingerprint texture data from the initial physiological features and operation rhythm data from the initial behavioral features. Fingerprint texture data is used for physiological feature comparison in subsequent identity verification, while operation rhythm data reflects the touch operation habits of the interacting subject. In the feature derivation direction of the interaction level two signal, it is necessary to collect voiceprint frequency data from the initial physiological features and interaction interval data from the initial behavioral features. Voiceprint frequency data is an important physiological feature for identity recognition, and interaction interval data reflects the time interval habits between voice commands. In the feature derivation direction of the interaction level three signal, it is necessary to collect feedback delay data from the initial response features, i.e., the response time of head posture to voice commands. This feedback delay data reflects the behavioral response habits of the interacting subject.

[0024] Step S1152: Set up a dedicated perception acquisition module for each initial feature type. Fingerprint texture data corresponds to the fingerprint acquisition module, facial contour data corresponds to the visual acquisition module, voiceprint frequency data corresponds to the audio acquisition module, and operation rhythm data corresponds to the touch acquisition module. Each module has corresponding acquisition functions and technical parameters.

[0025] Fingerprint texture data is collected by the smartphone's fingerprint acquisition module, which has fingerprint image acquisition capabilities and technical parameters including acquisition resolution and recognition speed. Facial contour data is collected by the front-facing vision acquisition module, which has dynamic image capture capabilities and technical parameters including image frame rate and pixel size. Voiceprint frequency data is collected by the audio acquisition module, which has sound wave signal acquisition and analysis capabilities and technical parameters including sampling rate and frequency response range. Operation rhythm data is collected by the touch acquisition module, which has the ability to collect data such as touch position, pressure, and time, and technical parameters including sampling frequency and pressure sensing accuracy.

[0026] Step S1153: Retrieve the device operating parameters of each sensing and acquisition module, including acquisition accuracy parameters, signal response speed parameters, data transmission rate parameters, and operating voltage range parameters.

[0027] Before activating each sensing and acquisition module, retrieve the acquisition accuracy parameter of the fingerprint acquisition module, which indicates the clarity of the acquired fingerprint image; the signal response speed parameter reflects the time from receiving the command to starting acquisition; the data transmission rate parameter is the speed at which the module transmits the acquired data to the processing unit; and the operating voltage range parameter ensures that the module operates normally within the power supply range of the smartphone. Similarly, retrieve the above-mentioned operating parameters for the vision acquisition module, audio acquisition module, and touch acquisition module.

[0028] Step S1154: Calculate the acquisition duration of each sensing acquisition module based on the duration of the current level signal.

[0029] The duration of the interaction level one signal is the time of the touch unlock operation, assuming this duration is T1. The acquisition duration of both the fingerprint acquisition module and the touch acquisition module is set to T1 to fully capture the feature data of the entire touch unlock process. The duration of the interaction level two signal is the duration of the voice command, T2, and the acquisition duration of the audio acquisition module is set to T2. The duration of the interaction level three signal is the duration of the head posture movement, T3, and the acquisition duration of the visual acquisition module is set to T3.

[0030] Step S1155: Calculate the start-up advance time of each sensing and acquisition module by combining the signal response speed parameters of the sensing and acquisition modules.

[0031] The signal response speed parameter of the fingerprint acquisition module is t1. To ensure that the module can acquire data in a timely manner when a touch operation begins, its start-up advance time is set to t1. Similarly, the signal response speed parameter of the touch acquisition module is t2, and its start-up advance time is t2. The signal response speed parameter of the audio acquisition module is t3, and its start-up advance time is t3. The signal response speed parameter of the vision acquisition module is t4, and its start-up advance time is t4. Through the above calculations, it is ensured that each module can immediately enter the acquisition state when the corresponding level signal begins.

[0032] Step S1156: According to the calculated start advance time, send start commands to the corresponding sensing and acquisition modules in sequence. The start commands include acquisition duration data, acquisition frequency data, and data transmission format data.

[0033] At time t1 before the expected start of the interaction level one signal, a start command is sent to the fingerprint acquisition module. This command includes the acquisition duration T1, acquisition frequency (e.g., A acquisitions per second), and data transmission format (e.g., JPEG image format). Simultaneously, at time t2 before the expected start of the interaction level one signal, a start command is sent to the touch acquisition module. This command includes the acquisition duration T1, acquisition frequency (e.g., B acquisitions per second), and data transmission format (e.g., CSV format). For the interaction level two signal, at time t3 before the expected start, a start command is sent to the audio acquisition module. This command includes the acquisition duration T2, acquisition frequency (e.g., C samplings per second), and data transmission format (e.g., WAV format). At time t4 before the expected start of the interaction level three signal, a start command is sent to the visual acquisition module. This command includes the acquisition duration T3, acquisition frequency (e.g., D frames per second), and data transmission format (e.g., MP4 format).

[0034] Step S1157: After all sensing and acquisition modules are started, monitor the operating status of each sensing and acquisition module in real time, determine whether the module is in normal acquisition state through the device status feedback signal, and immediately send a restart command if an abnormality occurs.

[0035] After each sensing and acquisition module is activated, the system receives device status feedback signals from them in real time. The fingerprint acquisition module's feedback signals include whether it has successfully started and whether a fingerprint has been detected; the touch acquisition module's feedback signals include whether a touch signal has been received and whether the acquired data is normal; the audio acquisition module's feedback signals include whether a sound wave signal has been received and whether the signal strength is within the normal range; the vision acquisition module's feedback signals include whether the camera is properly turned on and whether the image acquisition is clear. If the feedback signal from any module is abnormal, such as the fingerprint acquisition module not detecting a fingerprint, the system immediately sends a restart command to that module to ensure the continuity of the acquisition process.

[0036] Step S1158: Adjust the acquisition frequency of the sensing acquisition module according to the intensity change rate parameter of the hierarchical signal. When the intensity change rate parameter exceeds the preset range, increase the acquisition frequency value. When the intensity change rate parameter is within the preset range, maintain the initial acquisition frequency.

[0037] The intensity change rate parameter of the interaction level one signal reflects how quickly the touch pressure changes. When this parameter exceeds the preset range, i.e., the touch pressure changes rapidly, the system controls the touch acquisition module to increase the acquisition frequency from B times per second to B+E times per second to capture pressure change details in more detail. When the intensity change rate parameter is within the preset range, the initial acquisition frequency of B times / second is maintained. Similarly, for the interaction level two signal, if the intensity change rate parameter of the voiceprint frequency exceeds the preset range, the audio acquisition module increases the acquisition frequency; when the posture change intensity change rate parameter of the interaction level three signal is abnormal, the vision acquisition module increases the acquisition frequency.

[0038] Step S1159: Each sensing acquisition module acquires the corresponding initial feature data in real time according to the set acquisition parameters. The fingerprint acquisition module acquires the image data of fingerprint texture, the visual acquisition module acquires the dynamic image data of facial contour, the audio acquisition module acquires the sound wave data of voiceprint frequency, and the palm vein acquisition module acquires palm vein data.

[0039] The fingerprint acquisition module collects fingerprint texture image data of the user's finger in the fingerprint recognition area according to the set acquisition duration T1, acquisition frequency A times / second, and JPEG data transmission format. This fingerprint texture image data includes information such as the fingerprint's ridge direction and node features. The visual acquisition module collects dynamic image data of the user's head posture according to the set acquisition duration T3, acquisition frequency D frames / second, and MP4 format, recording posture features such as head rotation angle and speed. The audio acquisition module collects sound wave data of the user's voice command "Open Email Application" according to the set acquisition duration T2, acquisition frequency C times / second, and WAV format, including features such as voiceprint frequency and amplitude. The touch acquisition module collects operation rhythm data such as the position, pressure, and time of touch operations according to the set acquisition duration T1, acquisition frequency B times / second (or an adjusted frequency), and CSV format.

[0040] Step S11510: Receive the initial feature data transmitted by each sensing and acquisition module in real time, add acquisition timestamp data and module identification data to each data block, associate it with the corresponding hierarchical signal and acquisition module, and complete the initial feature acquisition operation.

[0041] The system receives initial feature data transmitted from each sensing and acquisition module in real time. For fingerprint texture image data blocks transmitted by the fingerprint acquisition module, a timestamp (accurate to milliseconds) and the module identifier "Fingerprint Acquisition Module" are added, and the data is associated with the first interaction level signal. Similarly, a timestamp, the module identifier "Touch Acquisition Module," and the first interaction level signal are added to the operation rhythm data blocks of the touch acquisition module. A timestamp and the module identifier "Audio Acquisition Module" are added to the sound wave data blocks of the audio acquisition module, and the data is associated with the second interaction level signal. A timestamp and the module identifier "Visual Acquisition Module" are added to the dynamic image data blocks of the visual acquisition module, and the data is associated with the third interaction level signal. Through the above processing, the initial feature acquisition operation is completed.

[0042] Step S116: According to the preset feature transfer rules, the initial features collected at the current level are transferred and fused with the features derived from the previous level to generate the derived features of the current level. The transfer and fusion process includes feature attribute superposition operation, data range expansion operation, and correlation strengthening operation.

[0043] In the second-level interaction signal processing, the initial features acquired at this level are voiceprint frequency data and interaction interval data. Features derived from the previous level (first-level interaction) include fingerprint texture-derived features and operation rhythm-derived features. Following preset feature transfer rules, a feature attribute overlay operation is first performed, overlaying the identity attributes from the fingerprint texture-derived features onto the voiceprint frequency data, giving the voiceprint frequency data an identity-related attribute. Then, a data range expansion operation is performed, extending the data range of the operation rhythm-derived features to the time dimension of the voiceprint frequency data, forming a broader time-series data. Finally, a correlation strengthening operation is performed, strengthening the correlation between fingerprint texture and voiceprint frequency in terms of identity features, generating second-level interaction derived features, such as composite physiological features that fuse fingerprints and voiceprints.

[0044] Step S117: Calculate the transmission strength parameter in the feature transmission process. The transmission strength parameter is derived based on the degree of correlation between features at different levels. The degree of correlation is determined by the overlap ratio of feature data and the attribute similarity parameter, respectively. The overlap ratio is used to measure the inheritance ratio at the data level, and the attribute similarity parameter is used to measure the similarity at the semantic level.

[0045] Calculate the feature transfer strength parameter from interaction level one to interaction level two. First, calculate the overlap ratio of feature data, that is, the proportion of overlap in data content between the derived features of interaction level one and the initial features of interaction level two. For example, some feature points in fingerprint texture derived features overlap with feature parameters in voiceprint frequency data; calculate the proportion of these overlapping data to the total data volume. Then, calculate the attribute similarity parameter to analyze the semantic similarity between the identity attributes of fingerprint texture derived features and the identity attributes of voiceprint frequency data, such as whether they both point to the same physiological characteristics of the same identity. Assign certain weights to the overlap ratio and attribute similarity parameter, and obtain the degree of correlation by weighted summation, thus obtaining the transfer strength parameter.

[0046] Step S118: Bind the initial features, derived features, transmission intensity parameters, and time sequence markers of the current level to form feature transmission binding data. The time sequence markers record the hierarchical order data and time node data generated by the features.

[0047] For interaction level two, the initial features such as the collected voiceprint frequency data and interaction interval data are bound together with derived features such as the composite physiological features generated through transmission fusion, the calculated transmission intensity parameters, and a time sequence marker recording the level as level two and the time node as T1+T2 after the start of the interaction. This data is combined through data structure association to form feature transmission binding data, ensuring a clear correspondence between the data and facilitating subsequent processing and analysis.

[0048] Step S119: Perform structural standardization processing on the feature transfer binding data, integrate the standardized feature transfer binding data, and generate feature derivative transfer units containing initial features, derived features, transfer path data, transfer intensity parameters, and time sequence information. Each feature derivative transfer unit corresponds to a feature derivative transfer process data of an interaction level.

[0049] The feature transfer binding data of interaction level two undergoes structural standardization, unifying the data format, data type, and data storage method. For example, the data format of initial and derived features is unified into a specific vector form, the transfer strength parameter is standardized to a value between 0 and 1, and the temporal information is converted into a unified timestamp format. After standardization, the above data is integrated to form a feature derivation transfer unit. This feature derivation transfer unit contains the initial features of interaction level two (voiceprint frequency data, interaction interval data), derived features (composite physiological features fusing fingerprint and voiceprint), transfer path data (feature transfer path from interaction level one to interaction level two), transfer strength parameters, and temporal information (hierarchical order and time nodes), completely recording the entire process of feature derivation transfer data at this interaction level.

[0050] Step S120: According to the progressive order of the behavior signal hierarchy sequence, the feature derivation transmission units are chained together to form a feature derivation transmission time sequence chain. The feature derivation transmission time sequence chain records the dynamic evolution trajectory data of feature derivation relationship data and transmission path data as the interaction level progresses.

[0051] After generating the feature derivation and transmission units for interaction levels one, two, and three, these units are chained together according to the progressive order of interaction levels one, two, and three in the behavioral signal hierarchy sequence. By establishing the connection relationship between the units, the feature derivation and transmission data of each level are organized in a temporal and logical order to form a feature derivation and transmission time sequence chain. This feature derivation and transmission time sequence chain can demonstrate the dynamic changes of the derivation relationship and transmission path of features between different levels as the interaction process progresses.

[0052] Step S121: Extract the global hierarchical marker of the behavior signal hierarchy sequence. The global hierarchical marker includes the start number, end number, hierarchical progression interval time data, and hierarchical association type data of each hierarchical signal. The hierarchical association type data records the behavioral logical relationship data between the preceding and following hierarchical levels.

[0053] Global hierarchical markers are extracted from the behavioral signal hierarchy sequence. The starting number is the beginning sequence number of each hierarchical signal in the sequence; for example, the starting number of interaction level one signal is 1, interaction level two signal is 2, and interaction level three signal is 3. The ending number corresponds to the ending sequence number of each hierarchical signal, and is 100, 200, and 300 respectively (assuming each hierarchical signal contains 100 data points). The hierarchical progression interval data is the interval between the start times of adjacent hierarchical signals; for example, the interval from the end of interaction level one to the start of interaction level two is t5, and the interval from the end of interaction level two to the start of interaction level three is t6. The hierarchical association type data records the logical relationship between preceding and following hierarchical levels; for example, interaction level one and interaction level two have a sequential execution relationship, while interaction level two and interaction level three have a parallel execution relationship.

[0054] Step S122: Add a corresponding hierarchical marker to each feature derivation and transmission unit. The hierarchical marker is consistent with the hierarchical signal number in the global hierarchical marker, and marks the interaction hierarchical position data corresponding to each feature derivation and transmission unit.

[0055] A layer marker "Layer 1" is added to the feature derivation and transmission unit of interaction layer one, corresponding to the start number 1 and end number 100 of the interaction layer one signal in the global layer marker, indicating that this unit belongs to interaction layer one. A layer marker "Layer 2" is added to the feature derivation and transmission unit of interaction layer two, corresponding to the start number 2 and end number 200 in the global layer marker. A layer marker "Layer 3" is added to the feature derivation and transmission unit of interaction layer three, corresponding to the start number 3 and end number 300. These layer markers clarify the layer position of each feature derivation and transmission unit in the interaction process.

[0056] Step S123: Based on the hierarchical marker, perform preliminary sorting on all the feature derivation and transmission units to form a preliminary hierarchical sequence, wherein the preliminary hierarchical sequence is arranged in the order of interaction level one, interaction level two, and interaction level three.

[0057] Based on the hierarchical labels "Level 1", "Level 2", and "Level 3", the three feature derivation and propagation units are initially sorted to obtain the preliminary hierarchical sequence: Interactive Level 1 Feature Derivation and Propagation Unit, Interactive Level 2 Feature Derivation and Propagation Unit, and Interactive Level 3 Feature Derivation and Propagation Unit. This preliminary hierarchical sequence is arranged in a progressive order of the interactive levels.

[0058] Step S124: Calculate the transmission connection parameters between two adjacent feature-derived transmission units. The transmission connection parameters include the overlap parameter of the derived features of the preceding and following units and the continuity parameter of the transmission path. The overlap parameter records the inheritance ratio data of the preceding and following derived features. Obtain the interval difference parameter of the time sequence marker. The interval difference parameter independently records the time interval data of the preceding and following units.

[0059] Calculate the transmission connection parameters between the feature derivation and transmission units of interaction level one and interaction level two. The overlap parameter is obtained by calculating the proportion of identical feature data in the derivative features of interaction level one and interaction level two, such as the overlap ratio of fingerprint texture derivative features of interaction level one and fingerprint-related parts in the composite physiological features of interaction level two. The continuity parameter of the transmission path analyzes whether the transmission path from interaction level one to interaction level two is smooth and whether there are any path interruptions or abnormal jumps. The time stamp interval difference parameter is the difference between the end timestamp of the feature derivation and transmission unit of interaction level one and the start timestamp of the feature derivation and transmission unit of interaction level two, i.e., t5. Similarly, calculate the transmission connection parameters and the time stamp interval difference parameter t6 between the feature derivation and transmission units of interaction level two and interaction level three.

[0060] Step S125: Adjust the feature association strength parameter of adjacent units according to the transmission connection parameter. The association strength parameter is used to represent the association relationship between the derived features of the previous unit and the derived basis of the next unit. The adjustment of the association strength parameter is based on the comprehensive judgment result of the overlap parameter and the continuity parameter.

[0061] For adjacent units in interaction levels one and two, if the overlap parameter is high and the propagation path coherence parameter is good, it indicates that the derived features of the preceding unit are closely related to the derived basis of the following unit. In this case, the feature association strength parameter should be increased. If the overlap parameter is low or the coherence parameter is poor, the association strength parameter should be appropriately decreased. Taking into account the values ​​of the overlap parameter and the coherence parameter, the adjusted association strength parameter is calculated using a preset algorithm (such as weighted average) to accurately reflect the degree of association between the derived features of preceding and following units.

[0062] Step S126: Extract the core derived features in each feature derivation and transmission unit. The core derived features are key feature combinations that can uniquely identify identity attributes. The key feature combinations include stable features and key derived features that are retained after multi-level transmission.

[0063] Core derived features are extracted from the feature derivation and transmission unit of interaction level one, such as the stable combination of key fingerprint texture feature points after processing. The core derived features of interaction level two are the combination of key parameters from the composite physiological features fusing fingerprints and voiceprints; these parameters have high identity distinguishability. The core derived features of interaction level three are the stable pattern feature combination of head posture and voice command coordinated response. These core derived features remain relatively stable during transmission at different levels and play a crucial role in identity recognition.

[0064] Step S127: Establish a transmission association diagram of core derived features of adjacent units. The transmission association diagram records the transmission path data, transmission intensity parameters, and derivation change mode data of the core features of the previous unit to the core features of the next unit.

[0065] For example, step S1271: Extract all core derived features in the previous feature derivation and transmission unit, and record the attribute type data, data range, transmission start point identification data, and functional positioning data of each core derived feature in the feature derivation and transmission unit. The functional positioning data records the effect of the feature on identity recognition.

[0066] Taking the previous unit as an interaction level-1 feature derivation and transmission unit as an example, we extract its core derived features, such as the combination of key feature points of fingerprint texture. The attribute type data of this feature is recorded as physiological feature, the data range is the feature value of a specific area of ​​the fingerprint image, the transmission start point identification data is "level 1-feature 1", and the functional positioning data is "basic identity recognition feature, used for preliminary identity verification".

[0067] Step S1272: Extract all core derived features in the next feature-derived transmission unit and record their attribute type data, data range, transmission endpoint identifier data, and functional positioning data.

[0068] The next unit is the interaction level two feature derivation and transmission unit, whose core derivation feature is a combination of composite physiological features that fuse fingerprints and voiceprints. The recorded attribute type data is composite physiological features, the data range is the value range of the fused feature vector, the transmission endpoint identifier data is "level 2 - feature 1", and the functional positioning data is "core identity recognition feature, improving the accuracy of identity verification".

[0069] Step S1273: Compare the attribute type data and data range of the core derived features of the preceding and following units, and filter out feature pairs with the same attribute type data and overlapping data ranges.

[0070] We compared the fingerprint texture key feature point combination (attribute type: physiological feature, data range: A1-A2) at interaction level one with the composite physiological feature combination (attribute type: composite physiological feature, data range: B1-B2) at interaction level two. Since the attribute types are not completely identical, but the composite physiological feature contains fingerprint-related attributes and their data ranges overlap (e.g., the value ranges of the fingerprint feature parts overlap), this pair of features was selected as the feature pair.

[0071] Step S1274: Analyze the data changes of each feature pair, calculate the overlap ratio and difference ratio of the feature data of the previous unit and the feature data of the next unit. The overlap ratio records the amount of inherited data of the feature, and the difference ratio records the amount of derived data of the feature.

[0072] For the selected feature pairs, calculate the overlap ratio between the fingerprint texture key feature point data of interaction level one and the fingerprint-related data in the composite physiological features of interaction level two, i.e., the proportion of overlapping data to the total amount of feature data in the previous unit. The difference ratio is the proportion of the part of the feature data in the subsequent unit that is different from the feature data in the previous unit (such as voiceprint-related data) to the total amount of feature data in the subsequent unit.

[0073] Step S1275: Set the transmission path type data for feature pairs based on the overlap ratio data. When the overlap ratio data exceeds the preset ratio value, it is a direct transmission path. When the overlap ratio data does not exceed the preset ratio value, it is an indirect transmission path. The direct transmission path has no intermediate feature conversion step, while the indirect transmission path includes a feature conversion step.

[0074] If the overlap ratio exceeds a preset value (e.g., 70%), the transmission path type is set to direct transmission path, indicating that the features of the previous unit are directly transmitted to the next unit without intermediate conversion. If the overlap ratio does not exceed the preset value, it is an indirect transmission path, which requires a feature conversion step, such as fusing fingerprint and voiceprint features. In this example, assuming the overlap ratio is 60%, which does not exceed the preset ratio, the transmission path type is indirect transmission path.

[0075] Step S1276: Set the derived change method data according to the transmission path type data and the difference ratio data. The derived change method data corresponding to the direct transmission path includes data expansion operation, accuracy improvement operation, and dimension increase operation. The derived change method data corresponding to the indirect transmission path includes attribute conversion operation, feature fusion operation, and pattern extraction operation.

[0076] Since the transmission path type is an indirect transmission path, and considering the difference ratio data, the derived change method data is set as a feature fusion operation. That is, the fingerprint texture features of the previous unit are fused with the voiceprint frequency features collected in the next unit to form a composite physiological feature. At the same time, the fusion weight and method are determined according to the magnitude of the difference ratio data to ensure that the derived features can retain the key information of the original features while incorporating new feature data.

[0077] Step S1277: Calculate the transmission strength value based on the degree of association between the features of the preceding and following units. The degree of association is evaluated by the overlap ratio data of the feature pairs and the functional positioning similarity data, respectively. The overlap ratio data is used to measure the association at the data level, and the functional positioning similarity data is used to measure the association at the semantic level.

[0078] The overlap ratio data (60%) and the functional positioning similarity data (e.g., the functional positioning of interaction level one is basic identity recognition, and the functional positioning of interaction level two is core identity recognition, and the similarity between the two is high, so it is set to 80%) are assigned weights (e.g., each accounting for 50%), and the degree of association is calculated by weighted summation: (60%×50%+80%×50%)=70%. Then the degree of association is converted into a transmission strength value (e.g., between 0 and 1, 70% corresponds to 0.7).

[0079] Step S1278: Integrate the transmission start point identifier data, transmission end point identifier data, transmission path type data, derived change mode data, and transmission intensity value of each feature pair to form a single transmission association entry. Each transmission association entry corresponds to a set of transmission relationship data of feature pairs.

[0080] Integrating the above data, the following related entries are formed: the transmission start point identifier data "Level 1 - Feature 1", the transmission end point identifier data "Level 2 - Feature 1", the transmission path type data "Indirect transmission path", the derived change method data "Feature fusion operation", and the transmission intensity value "0.7".

[0081] Step S1279: Arrange all the transit-related items in the hierarchical order of the preceding and following units. The transit starting point feature is located on the left side of the diagram, and the transit ending point feature is located on the right side of the diagram. Connect them with lines to represent the transit path. The line thickness corresponds to the numerical value of the transit intensity. Also, add the derived change mode label to the diagram and record the feature change details corresponding to each transit path to form a transit-related diagram.

[0082] In the transmission relationship diagram, the core derived feature of interaction level one, "Level 1 - Feature 1," is placed on the left, and the core derived feature of interaction level two, "Level 2 - Feature 1," is placed on the right. The two are connected by a line, the thickness of which is set according to the transmission strength value of 0.7; the larger the value, the thicker the line. Next to the line, a label indicating the derivative change method, "Feature Fusion Operation (Fingerprint + Voiceprint)," is added to record the details of feature changes. Through this layout and labeling, a clear transmission relationship diagram is formed.

[0083] Step S128: Based on the transmission connection parameters, feature association strength parameters, and interval difference parameters of the transmission association graph and time sequence markers, adjacent feature derivation transmission units are chained together to form unit connection modules. Each unit connection module contains two adjacent units, transmission association information data, and time sequence interval data.

[0084] By combining the transmission connection parameters (overlap and coherence parameters), feature association strength parameters, transmission association graph, and time-series marker interval difference parameter t5 of interaction level one and interaction level two, these two feature-derived transmission units are linked in a chain. During the connection process, the transmission path and mode are determined by the transmission association graph, the tightness of the connection is determined by the feature association strength parameter, and the time-series marker interval difference parameter t5 records the time interval between the two. The above information is integrated into the unit connection module, which includes the feature-derived transmission units of interaction level one and interaction level two, transmission association information (transmission path, strength, change mode, etc.), and time-series interval data t5.

[0085] Step S129: Connect all unit connection modules in series according to the order of the preliminary hierarchical sequence to form a preliminary time sequence chain. The preliminary time sequence chain records the hierarchical progressive relationship data of feature derivation and transmission.

[0086] The unit connection modules between interaction level 1 and interaction level 2, and between interaction level 2 and interaction level 3, are continuously connected in series according to the preliminary sequence of the levels (interaction level 1, interaction level 2, interaction level 3). The preliminary sequence chain after series connection includes, in sequence, the feature derivation and transmission unit of interaction level 1, the unit connection module (level 1 and level 2), the feature derivation and transmission unit of interaction level 2, the unit connection module (level 2 and level 3), and the feature derivation and transmission unit of interaction level 3, recording the progressive relationship of feature derivation and transmission among the three levels.

[0087] Step S1210: Optimize the structure of the preliminary timing chain, adjust the transmission association parameters of each unit connection module, and generate a feature-derived transmission timing chain.

[0088] The transmission correlation parameters of each unit connection module in the preliminary time sequence chain are analyzed, such as transmission strength parameters and correlation strength parameters. For connection modules with low transmission strength or weak correlation strength, the transmission correlation parameters are adjusted appropriately according to the overall feature transmission logic and requirements, such as increasing transmission strength or optimizing transmission paths. After structural optimization, the final feature-derived transmission time sequence chain is generated, which can more accurately and smoothly reflect the dynamic evolution trajectory of features as the interaction level progresses.

[0089] Step S130: Retrieve the preset identity benchmark chain library, which stores the initial benchmark chain corresponding to the known identity. Through a self-learning evolution model combined with the current interaction scenario parameters and historical interaction data, the initial benchmark chain is dynamically updated to generate a scenario-adaptive benchmark chain.

[0090] The system retrieves a pre-defined identity baseline chain library from the database. This library stores initial baseline chains for multiple known users (such as the phone owner and authorized users). For the current interaction subject, and considering the interaction scenario parameters of the current smartphone (such as ambient light and device battery level) and the user's historical interaction data (such as previous unlocking methods and application usage habits), the system adjusts the initial baseline chain using a self-learning evolution model to better adapt it to the current scenario, thus generating a scenario-adaptive baseline chain.

[0091] Step S131: Retrieve the preset identity benchmark chain library. The preset identity benchmark chain library stores multiple initial benchmark chains corresponding to known identities. Each initial benchmark chain is the feature-derived transmission benchmark trajectory data formed by the known identity through multiple interactions in a standard scenario, including core benchmark features, transmission path data, and benchmark strength parameters at each level.

[0092] In the pre-defined identity baseline chain library, each known identity (such as User A and User B) corresponds to an initial baseline chain. Taking User A as an example, their initial baseline chain is feature-derived transmission baseline trajectory data formed through multiple smartphone unlocking and application operation interactions under standard scenarios (such as normal indoor lighting and a fully charged device). This initial baseline chain includes core baseline features at the first interaction level (such as User A's standard fingerprint texture features), transmission path data (the transmission path from fingerprint features to composite features), and baseline strength parameters (standard strength values ​​for each transmission stage).

[0093] Step S132: Collect scene parameters of the current human-computer interaction scenario. The scene parameters include four types of environmental data and device-related data that directly affect feature collection: ambient lighting parameters, device operating parameters, network transmission parameters, and interactive interface parameters.

[0094] In the current interaction scenario, ambient lighting parameters are collected by the phone's light sensor, including data such as light intensity (e.g., LUX value range) and light uniformity. Device operating parameters include the phone's current battery percentage, CPU usage, and memory usage. Network transmission parameters include the type of network the phone is currently connected to (e.g., Wi-Fi, 4G), network latency, and signal strength. Interaction interface parameters include screen brightness, the layout of the currently displayed application interface, and font size. These parameters directly affect the quality and accuracy of feature acquisition; for example, insufficient lighting may affect fingerprint acquisition.

[0095] Step S133: Retrieve the historical interaction data corresponding to the known identity. The historical interaction data includes feature derivation and transmission record data, identity verification result data, and feature adjustment record data in similar scenarios in the past.

[0096] Retrieve user A's historical interaction data in similar scenarios to the current one (e.g., both indoors, using email applications). This includes feature derivation and transmission records such as fingerprint, voiceprint, and gesture features; authentication result data recording past successes and failures of authentication and their reasons; and feature adjustment records showing past adjustments to feature parameters due to changes in the environment, such as adjusting fingerprint collection thresholds in low-light conditions.

[0097] Step S134: Construct a self-learning evolution model, which includes a scene adaptation layer, a history fusion layer, and a feature update layer. The scene adaptation layer performs feature influence analysis of scene parameters, the history fusion layer performs historical data change extraction, and the feature update layer performs dynamic adjustment of baseline features.

[0098] The scene adaptation layer of the self-learning evolution model receives current scene parameters and analyzes the influence of different scene parameters on the core benchmark features at each level, such as the impact of light intensity on fingerprint image clarity. The history fusion layer processes the retrieved historical interaction data, extracting the patterns and trends of feature derivation and transmission, such as the changes in user A's operating rhythm under different battery levels. The feature update layer calculates the update magnitude and direction of each feature parameter in the initial benchmark chain based on the outputs of the scene adaptation layer and the history fusion layer, realizing the dynamic adjustment of benchmark features.

[0099] Step S135: Input the scene parameters into the scene adaptation layer of the self-learning evolution model, analyze the influence of scene parameters on the baseline features of each level, calculate the scene influence coefficient of each baseline feature, and record the modification requirements of scene parameters on feature data.

[0100] Scene parameters such as ambient light and device operating parameters are input into the scene adaptation layer. For example, if the current ambient light intensity is low, the scene adaptation layer analyzes and concludes that this lighting condition will reduce the contrast of the fingerprint image captured by the fingerprint acquisition module, thus affecting the accuracy of fingerprint features. For the core baseline feature (fingerprint texture feature) of interaction level one, a scene influence coefficient is calculated. A positive scene influence coefficient indicates that the contrast of fingerprint features needs to be enhanced or the feature extraction threshold adjusted to adapt to low-light scenarios. Similarly, the impact of low device battery on operation rhythm features is analyzed, and the corresponding scene influence coefficient is calculated.

[0101] Step S136: Input the historical interaction data into the historical fusion layer of the self-learning evolution model, extract the change data of feature derivation and transmission during the historical interaction process, including the fluctuation data of feature transmission intensity and the stable data of derived features, and generate historical change parameters.

[0102] After historical interaction data is input into the historical fusion layer, the layer extracts, through statistical analysis, the feature transmission intensity fluctuation data of user A under similar low-light scenarios, such as the variation range of the transmission intensity from fingerprint features to voiceprint features under different lighting conditions. Simultaneously, it extracts stable data of derived features, such as the stable frequency range of voiceprint features under various scenarios. Based on this data, historical variation parameters are generated to reflect the variation patterns of features in historical scenarios.

[0103] Step S137: Based on the scene influence coefficient and historical change parameters, the update magnitude data and update direction data of each core benchmark feature in the initial benchmark chain are calculated through the feature update layer of the self-learning evolution model. The update magnitude data is positively correlated with the scene influence coefficient and historical change parameters.

[0104] Step S1371: Traverse each level in the initial baseline chain, extract the core baseline features, corresponding transmission path parameters, transmission strength parameters, and feature attribute description data for each level. The feature attribute description data includes the feature data range, collection dimension data, and stable value range.

[0105] Traverse the interaction levels one, two, and three of user A's initial baseline chain. Extract the core baseline features (fingerprint texture features), transmission path parameters (path information of fingerprint features transmitted to level two), transmission strength parameters (standard transmission strength values), and feature attribute description data (fingerprint feature data range, acquisition dimension is two-dimensional image, stable numerical range) of interaction level one. Similarly, extract the corresponding data for interaction levels two and three.

[0106] Step S1372: Classify the scene influence coefficients according to the feature acquisition dimension, and divide them into three categories: illumination influence coefficient, device influence coefficient, and network influence coefficient. Each category corresponds to the influence of a type of scene parameter on the feature.

[0107] The scene influence coefficients are categorized as follows: the illumination influence coefficient corresponds to the impact of ambient lighting parameters on features, such as the impact on visual features like fingerprints and faces; the device influence coefficient corresponds to the impact of device operating parameters, such as the impact of battery power on operating rhythm; and the network influence coefficient corresponds to the impact of network transmission parameters, such as the impact of network latency on response feedback features.

[0108] Step S1373: Classify the historical change parameters according to feature type into three categories: physiological feature change parameters, behavioral feature change parameters, and response feature change parameters. Each category corresponds to the historical change data of one type of feature.

[0109] Historical change parameters are classified into physiological characteristic change parameters (such as historical changes in physiological characteristics such as fingerprints and voiceprints), behavioral characteristic change parameters (such as changes in behavioral characteristics such as operating rhythm and interaction interval), and response characteristic change parameters (such as changes in response characteristics such as feedback delay and response content).

[0110] Step S1374: For each core benchmark feature, extract the corresponding scene influence coefficient classification and historical change parameter classification, and calculate the comprehensive influence parameter through the feature update layer of the self-learning evolution model. The comprehensive influence parameter is the weighted sum of the scene influence coefficient and the historical change parameter.

[0111] Taking the core baseline feature of fingerprint texture at interaction level one as an example, we extract its corresponding illumination influence coefficient and physiological feature change parameter. We set the weight of the scene influence coefficient to 0.6 and the weight of the historical change parameter to 0.4, and then summed the two to obtain the comprehensive influence parameter. This comprehensive influence parameter reflects the combined impact of scene and historical factors on the core baseline feature.

[0112] Step S1375: Based on the comprehensive influence parameter and the data range in the feature attribute description data, set the update direction data of the core benchmark feature. When the comprehensive influence parameter is positive, the update direction data is the positive expansion of the data range. When the comprehensive influence parameter is negative, the update direction data is the reverse adjustment of the data range. The updated feature conforms to physiological logic and behavioral logic.

[0113] If the comprehensive influence parameter of the core baseline features of the fingerprint texture is positive, the update direction data is set to positively expand the data range based on the data range in its feature attribute description data (such as the range of the number of fingerprint feature points), that is, increase the upper limit of the number of feature points to accommodate the possible loss of feature points under low light conditions. At the same time, it is ensured that the expanded number of feature points conforms to normal physiological logic and does not exceed the reasonable range of human fingerprint features.

[0114] Step S1376: Based on the magnitude of the comprehensive influence parameter and the stable value range in the feature attribute description data, set the update amplitude data of the core benchmark feature. The larger the value of the comprehensive influence parameter, the larger the update amplitude data, and the updated feature data is within the stable value range.

[0115] The larger the value of the comprehensive influence parameter, the greater the impact of the scenario and historical factors on the feature, and the larger the update magnitude data should be. For example, when the comprehensive influence parameter is 0.8 (maximum value is 1), the update magnitude data is set to 20% of the stable value range in the feature attribute description data, ensuring that the updated feature data is still within the stable value range and maintaining the stability and reliability of the feature.

[0116] Step S1377: For the transmission path parameters, adjust the data transmission delay parameters of the transmission path based on the network transmission parameters in the scenario influence coefficient; adjust the node connection method data of the transmission path based on the path change data in the historical change parameters.

[0117] If the network latency in the current network transmission parameters is high, the data transmission latency parameter of the transmission path should be appropriately increased according to the scenario impact coefficient to match the actual network conditions. At the same time, referring to the path change data in the historical parameters, such as the adjustment records of the connection method of the transmission path nodes in the past under high-latency networks, the node connection method of the transmission path should be optimized, such as by adding intermediate buffer nodes.

[0118] Step S1378: For the transmission strength parameter, based on the strength fluctuation data in the historical change parameters and the equipment response parameters in the scene influence coefficient, adjust the numerical range of the transmission strength, and record the correlation data of the adjusted transmission strength features.

[0119] Based on the fluctuation range of transmission intensity in historical change parameters and the device response parameters in the current scenario influence coefficient (such as slow device response speed), the lower limit of the numerical range of transmission intensity is appropriately reduced. This allows for a certain degree of reduction in transmission intensity when the device response is slow, but it can still reflect the closeness of the correlation between features.

[0120] Step S1379: According to the set update amplitude data and update direction data, adjust the core reference features, transmission path parameters, and transmission intensity parameters of each level one by one, record the adjustment process data and the numerical change data before and after the adjustment of each parameter, and form parameter adjustment record data.

[0121] Based on the update magnitude and direction data, the data range of the core baseline features of the fingerprint texture at interaction level one is forward expanded. The transmission delay and node connection method of the transmission path parameters are adjusted, and the numerical range of the transmission strength parameters is adjusted. Simultaneously, the parameter values ​​before adjustment, the algorithms and parameters used during adjustment, and the parameter values ​​after adjustment are recorded to form parameter adjustment record data, facilitating subsequent traceability and verification.

[0122] Step S138: According to the updated amplitude data and updated direction data, dynamically adjust the core reference features, transmission path parameters and transmission intensity parameters of each level in the initial reference chain.

[0123] Based on the update magnitude data and update direction data calculated in step S137, the core baseline features (such as fingerprints, voiceprints, and gesture features), transmission path parameters (such as path nodes and transmission delays), and transmission strength parameters (such as transmission strength values ​​between each level) of the interaction levels one, two, and three in the initial baseline chain of user A are dynamically adjusted to make the initial baseline chain more in line with the current scenario and the latest interaction habits of users.

[0124] Step S139: Calculate the hierarchical coherence parameters of the adjusted baseline chain. The hierarchical coherence parameters are calculated by the transmission path overlap parameter and the feature association strength parameter. Record the transmission smoothness data and the derived logic consistency data between the baseline features of each level.

[0125] Calculate the overlap parameter of adjacent level transmission paths in the adjusted baseline chain, i.e., the overlap ratio between the adjusted transmission path and the original path, and the correlation strength parameter between features at each level. Calculate the hierarchical coherence parameter using a weighted average or similar method. A higher hierarchical coherence parameter value indicates smoother transmission between baseline features at each level and more consistent derivation logic.

[0126] Step S1310: Perform secondary optimization on the adjusted baseline chain based on the hierarchical coherence parameters, fine-tune the transmission path parameters and transmission strength parameters, and generate a scene-adaptive baseline chain.

[0127] If the hierarchical coherence parameter does not reach the preset threshold, it indicates that the adjusted baseline chain is insufficient in terms of transmission smoothness or logical consistency. At this time, based on the specific value of the hierarchical coherence parameter, the transmission path parameter (such as further optimizing node connections) and transmission strength parameter (such as adjusting the strength value) are fine-tuned until the hierarchical coherence parameter reaches the preset threshold, generating the final scene-adaptive baseline chain.

[0128] Step S140: Extract the cross-dimensional core matching features of the feature derivation and transmission time-series chain and the scene adaptation benchmark chain, perform bidirectional trajectory alignment processing based on the dynamic weight allocation strategy, and generate trajectory alignment results. The dynamic weight allocation strategy adjusts the matching weight values ​​in real time according to the importance of the features.

[0129] Core matching features that are present and stable across the physiological, behavioral, and response dimensions are extracted from the feature derivation and propagation time-series chain and the scene adaptation baseline chain. Then, according to a dynamic weight allocation strategy, different matching weights are assigned to these features, and bidirectional trajectory alignment is performed to make the feature trajectories of the two chains overlap as much as possible, generating trajectory alignment results for subsequent identity recognition.

[0130] Step S141: Extract the core derived features, transmission path parameters, transmission intensity parameters, and timing markers of each level in the feature derivation and transmission time sequence chain to form a feature set to be aligned. The feature set to be aligned is arranged in hierarchical order and contains complete feature derivation and transmission information data.

[0131] From the feature derivation and transmission time sequence chain, extract the core derived features of interaction level one (such as the fingerprint texture features of the current interactive subject), transmission path parameters (fingerprint feature transmission path), transmission strength parameters (current transmission strength value), and time sequence markers (timestamps of level one); similarly, extract the corresponding data for interaction levels two and three. Arrange the above data in the order of interaction levels one, two, and three to form a feature set to be aligned. This feature set to be aligned contains complete information on the feature derivation and transmission of the current interactive subject.

[0132] Step S142: Extract the core reference features, transmission path parameters, transmission intensity parameters, and temporal markers of each level in the scene adaptation reference chain to form a reference feature set. The hierarchical division of the reference feature set is consistent with that of the feature set to be aligned.

[0133] From the scene adaptation baseline chain (such as User A's scene adaptation baseline chain), extract the core baseline features (User A's adjusted fingerprint texture baseline features), transmission path parameters (adjusted transmission path), transmission strength parameters (baseline transmission strength value), and time stamps (time stamps of baseline layer one) for interaction layer one. The corresponding data for interaction layers two and three are extracted in the same way. Arrange the above data in hierarchical order to form a baseline feature set, whose hierarchical division is consistent with the feature set to be aligned, facilitating feature comparison.

[0134] Step S143: Identify cross-dimensional core matching features. Cross-dimensional core matching features are features that exist in all three dimensions: physiological perception dimension, behavioral pattern dimension, and response feedback dimension, and whose data fluctuation range is less than the preset fluctuation threshold. These include fixed attribute data in core physiological features, repetitive pattern data in core behavioral features, and stable feedback data in core response features.

[0135] Step S1431: Traverse all features in the feature set to be aligned and the baseline feature set, and filter out common features that exist simultaneously in the three dimensions of physiological perception, behavioral pattern, and response feedback. The common features include unified attribute data of identity in different perception dimensions.

[0136] It iterates through all features in the feature set to be aligned and the baseline feature set, such as fingerprint features (physiological dimension), operation rhythm features (behavioral dimension), and feedback delay features (response dimension). Features that exist in all three dimensions are selected, such as user A's fingerprint features (physiological), specific operation rhythm patterns (behavioral), and fixed feedback delays to specific instructions (response). These features together constitute unified attribute data of identity across different dimensions.

[0137] Step S1432: Perform stability analysis on each common feature, calculate stability parameters based on the data fluctuation range of the feature in each level. When the data fluctuation range is less than the preset fluctuation threshold, the value of the stability parameter increases by a preset ratio. The stability parameter is calculated from the invariant data of the feature during the interaction process.

[0138] Taking user A's operational rhythm characteristics as an example, we analyze the fluctuation range of their operational rhythm data at interaction levels one, two, and three. If this fluctuation range is less than a preset fluctuation threshold (e.g., ±5%), the stability parameter value increases (e.g., from 0.5 to 0.7). The stability parameter is obtained by calculating the proportion of invariant data for the feature at each level; the more invariant data, the larger the stability parameter.

[0139] Step S1433: Perform identity differentiation analysis on each common feature. By comparing the same common feature data of different known identities, calculate the differentiation parameter. When the difference in feature data between different identities is greater than the preset difference threshold, the value of the differentiation parameter increases by a preset ratio. The differentiation parameter is calculated from the relevant data of the feature to differentiate different identities.

[0140] The operation rhythm characteristics of User A are compared with those of other known users (such as User B), and the data differences are calculated. If the difference exceeds a preset difference threshold (e.g., 30%), the differentiation parameter value is increased (e.g., from 0.6 to 0.8). The differentiation parameter is calculated by statistically analyzing the percentage of times this feature can distinguish User A from other users, and other relevant data.

[0141] Step S1434: Perform scene adaptation analysis on each common feature, and calculate the adaptation parameters in combination with the current interaction scene parameters. When the deviation between the feature data and the scene parameters is less than the preset compatibility threshold, the adaptation parameter value increases by a preset ratio. The adaptation parameters are calculated from the data content of the feature in the current scene.

[0142] Analyze the user A's operational rhythm characteristics under the current scenario parameters (e.g., low battery) and calculate the deviation value from the standard operational rhythm under this scenario. If the deviation value is less than a preset compatibility threshold (e.g., 10%), the adaptation parameter value is increased (e.g., from 0.5 to 0.6). The adaptation parameter is calculated based on the degree of matching between the feature data and the scenario parameters.

[0143] Step S1435: Filter out common features whose stable parameters, distinguishing parameters, and adaptability parameters all exceed preset values, and set them as cross-dimensional core matching features. The stable parameters, distinguishing parameters, and adaptability parameters all meet the preset value requirements.

[0144] The preset values ​​for the stability parameter, differentiation parameter, and adaptation parameter are set to 0.6, 0.7, and 0.5, respectively. Common features where all three parameters exceed their respective preset values ​​are selected, such as user A's fingerprint texture features, specific operation rhythm pattern features, and fixed feedback delay features. These features are then set as cross-dimensional core matching features.

[0145] Step S144: Construct a dynamic weight allocation model. The dynamic weight allocation model calculates the matching weight value based on three dimensions: stable data of matching features, identity differentiation data, and scene adaptation data. When the fluctuation range corresponding to the stable data is less than a preset fluctuation threshold, the matching weight value increases according to the preset weight adjustment ratio. When the difference data corresponding to the identity differentiation data is greater than a preset difference threshold, the matching weight value increases according to the preset weight adjustment ratio. When the deviation value corresponding to the scene adaptation data is less than a preset compatibility threshold, the matching weight value increases according to the preset weight adjustment ratio.

[0146] Step S1441: Construct a dynamic weight allocation model, which includes an input layer, a feature processing layer, a weight calculation layer, and an output layer. The input layer receives stable parameters, distinguishing parameters, and adaptation parameters. The feature processing layer performs parameter standardization processing. The weight calculation layer calculates the matching weight values ​​through a weighted summation algorithm. The output layer outputs the final weight values.

[0147] The input layer of the dynamic weight allocation model receives stable parameters, discriminative parameters, and adaptation parameters of the cross-dimensional core matching features. The feature processing layer standardizes these parameters, transforming them into values ​​between 0 and 1. The weight calculation layer performs a weighted sum of the standardized parameters, with the stable parameter, discriminative parameter, and adaptation parameter having weights of 0.3, 0.4, and 0.3, respectively. The output layer outputs the calculated matching weight values.

[0148] Step S1442: Assign different model weight values ​​to the stable parameters, distinguishing parameters, and adaptation parameters of the dynamic weight allocation model. Input the stable parameters, distinguishing parameters, and adaptation parameters of each cross-dimensional core matching feature into the dynamic weight allocation model. After the standardization operation of the feature processing layer, the initial value of the matching weight is obtained through the weighted summation algorithm of the weight calculation layer.

[0149] The model weights with stable parameters are set to 0.3, the discrimination parameter to 0.4, and the adaptation parameter to 0.3. Taking user A's fingerprint texture features as an example, its stable parameter is 0.7, the discrimination parameter is 0.8, and the adaptation parameter is 0.6. After being input into the model, the feature processing layer normalizes it to 0.7, 0.8, and 0.6. The weight calculation layer obtains the initial matching weight value of 0.71 by weighted summation (0.7×0.3+0.8×0.4+0.6×0.3)=0.71.

[0150] Step S1443: Normalize the initial value of the matching weight, and bind the normalized matching weight value with the corresponding cross-dimensional core matching feature, various parameter values, and model calculation process data to form a weight allocation table.

[0151] The initial values ​​of the matching weights for all cross-dimensional core matching features are normalized so that their sum equals 1. For example, the initial matching weight for fingerprint texture features is 0.71, for operation rhythm pattern features it is 0.65, and for fixed feedback delay features it is 0.58; after normalization, these values ​​are 0.38, 0.35, and 0.27, respectively. These normalized matching weight values ​​are then bound to the corresponding features, parameters, and calculation process data to form a weight allocation table.

[0152] Step S145: Using the dynamic weight allocation model, assign corresponding matching weight values ​​to each cross-dimensional core matching feature and generate a weight allocation table. The weight allocation table contains the weight values ​​and calculation process data of each cross-dimensional core matching feature.

[0153] Following step S144, the dynamic weight allocation model calculates and assigns matching weight values ​​to each cross-dimensional core matching feature, generating a weight allocation table that includes the weight values ​​of each feature, stable parameters, distinguishing parameters, adaptation parameters, and calculation processes (such as standardization methods and weighted summation formulas).

[0154] Step S146: Establish a bidirectional trajectory coordinate system. The horizontal axis of the coordinate system is the hierarchical time-series axis of the feature set to be aligned, and the vertical axis is the hierarchical time-series axis of the baseline feature set. Each node in the coordinate system corresponds to a level of cross-dimensional core matching feature combination data.

[0155] Establish a two-dimensional bidirectional trajectory coordinate system. The horizontal axis represents the hierarchical temporal sequence of the feature set to be aligned, with the temporal markers of interaction levels one, two, and three from left to right. The vertical axis represents the hierarchical temporal sequence of the baseline feature set, with the temporal markers of the corresponding levels from bottom to top. Each node in the coordinate system (e.g., (Level 1, Level 1)) corresponds to the cross-dimensional core matching feature combination data of the feature set to be aligned and the baseline feature set at that level.

[0156] Step S147: Map the cross-dimensional core matching features of the feature set to be aligned and the reference feature set to the bidirectional trajectory coordinate system to form the feature trajectory to be aligned and the reference feature trajectory.

[0157] The cross-dimensional core matching features in the feature set to be aligned (such as the fingerprint of the current interactive subject, operation rhythm, and feedback delay features) are mapped to the horizontal axis of the coordinate system in hierarchical time sequence to form the feature trajectory to be aligned. Similarly, the cross-dimensional core matching features in the baseline feature set (such as the baseline fingerprint of user A, operation rhythm, and feedback delay features) are mapped to the vertical axis in hierarchical time sequence to form the baseline feature trajectory.

[0158] Step S148: Calculate the relative deviation between the cross-dimensional core matching features of the feature trajectory to be aligned and the cross-dimensional core matching features of the baseline feature trajectory at the same level node, and obtain the weighted difference value by combining the matching weight value.

[0159] At the (Level 1, Level 1) node in the coordinate system, calculate the relative deviation between the cross-dimensional core matching feature (current fingerprint feature) of the interaction level 1 in the feature set to be aligned and the cross-dimensional core matching feature (user A's baseline fingerprint feature) of the interaction level 1 in the baseline feature set, such as the Euclidean distance of the feature vectors. Then multiply this relative deviation value by the matching weight value of the feature in the weight allocation table to obtain the weighted difference value. Perform the same calculation for each level node.

[0160] Step S149: Adjust the transmission path parameters and timing markers of the feature set to be aligned according to the weighted difference value, so as to keep the node deviation between the feature trajectory to be aligned and the reference feature trajectory within the minimum range.

[0161] If the weighted difference value of a node at a certain level is large, it indicates that the feature to be aligned deviates significantly from the reference feature at that level. In this case, adjust the transmission path parameters (such as optimizing the feature transmission path to reduce transmission loss) and time stamps (such as fine-tuning the timestamp to make the feature occurrence time closer to the reference feature) of that level in the feature set to be aligned to reduce the weighted difference value and minimize the node deviation between the feature trajectory to be aligned and the reference feature trajectory.

[0162] Step S1410: Repeat the mapping operation, difference calculation operation, and adjustment operation until the sum of the weighted difference values ​​of all level nodes reaches the minimum. Extract the matching parameters of each level. The matching parameters include three types of information data: weighted difference value, matching weight value, and trajectory overlap data. Generate trajectory alignment results.

[0163] The process of mapping the features to be aligned and the baseline features to the coordinate system, calculating the weighted difference value, and adjusting the parameters of the features to be aligned is repeated multiple times. After each adjustment, the sum of the weighted difference values ​​of all level nodes is recalculated, and the adjustment stops when the sum reaches the minimum value. At this point, the weighted difference values ​​of each level, the matching weight values, and the trajectory overlap data (such as the proportion of overlapping nodes and the feature similarity of the overlapping parts) are extracted, and this information is integrated to generate the trajectory alignment result.

[0164] Step S150: Generate an identity recognition conclusion based on the trajectory alignment result. The identity recognition conclusion includes the identity attribution information of the interactive subject, feature derivation and transmission matching data, and weight distribution data.

[0165] Based on the comprehensive alignment score and matching parameters at each level in the trajectory alignment results, it is determined whether the current interactive subject matches the preset identity (such as user A), and an identity recognition conclusion containing identity attribution, matching data and weight distribution is generated.

[0166] Step S151: Analyze the core parameters in the trajectory alignment result, including the weighted difference value of each level, the matching weight value of the cross-dimensional core matching feature, the overlap data between the feature trajectory to be aligned and the benchmark feature trajectory, and the transmission path adaptation data.

[0167] The trajectory alignment results are analyzed to extract the weighted difference values ​​of each level (interaction levels one, two, and three), such as 0.12 for level one, 0.08 for level two, and 0.15 for level three; the matching weight values ​​of cross-dimensional core matching features (such as fingerprint feature 0.38, operation rhythm feature 0.35, and feedback delay feature 0.27); the overlap data between the feature trajectory to be aligned and the baseline feature trajectory (such as the overlap node ratio of 80%); and the transmission path adaptation data (such as the transmission path overlap degree of 75%).

[0168] Step S152: Retrieve the preset identity recognition evaluation standard, which includes the qualified values ​​of each core parameter, weight allocation ratio data, and comprehensive evaluation method data. The qualified values ​​are statistically derived based on historical identity verification sample data.

[0169] In the preset identity recognition evaluation criteria, the passing value for each level of weighted difference is set to 0.2 (less than 0.2 is considered passing), the passing value for trajectory overlap data is 70% (greater than 70% is considered passing), and the passing value for transmission path adaptation data is 65% (greater than 65% is considered passing). The weight allocation ratio is 40% for level-weighted difference, 30% for trajectory overlap data, and 30% for transmission path adaptation data. The comprehensive evaluation method compares the weighted sum with the passing score (e.g., 0.8).

[0170] Step S153: Compare the weighted difference value of each level with the corresponding qualified value in the identity recognition assessment standard, count the number of levels with a weighted difference value lower than the qualified value, and calculate the qualified ratio data of each level.

[0171] The weighted difference values ​​for each level were compared with the passing value of 0.2. The values ​​for Level 1 were 0.12 (pass), Level 2 were 0.08 (pass), and Level 3 were 0.15 (pass). All three levels were passing, and the passing rate for each level was 100%.

[0172] Step S154: Compare the trajectory overlap data and transmission path adaptation data with the corresponding qualified values ​​in the identity recognition evaluation standard to determine the qualified status of the two indicators.

[0173] If 80% of the trajectory overlap data is greater than the acceptable value of 70%, it is considered acceptable; if 75% of the transmission path adaptation data is greater than the acceptable value of 65%, it is considered acceptable.

[0174] Step S155: Based on the weight allocation ratio data in the identity recognition evaluation standard, convert the hierarchical qualification ratio data, trajectory overlap data qualification status, and transmission path adaptation data qualification status into quantitative scoring data.

[0175] A 100% pass rate for the hierarchical data corresponds to a quantitative score of 1.0, which multiplied by a weight of 40% equals 0.4; a pass rate for trajectory overlap data corresponds to a quantitative score of 1.0, which multiplied by a weight of 30% equals 0.3; and a pass rate for transmission path adaptation data corresponds to a quantitative score of 1.0, which multiplied by a weight of 30% equals 0.3.

[0176] Step S156: Using the comprehensive evaluation method data in the identity recognition evaluation standard, the various quantitative score data are weighted and calculated to obtain the comprehensive alignment score data. If the comprehensive alignment score data is higher than the qualified score value in the identity recognition evaluation standard, the known identity information to which the scene adaptation benchmark chain belongs is extracted, including unique identity identification data, feature derivation and transmission benchmark model data, and historical interaction feature data.

[0177] The overall alignment score is 0.4 + 0.3 + 0.3 = 1.0, which is higher than the passing score of 0.8. At this point, the known identity information of the scene adaptation baseline chain is extracted, namely, the unique identifier data of user A (such as user ID: 001), the feature derivation and transfer baseline model data (the baseline feature derivation and transfer model of user A), and the historical interaction feature data (the past interaction feature records of user A).

[0178] Step S157: Integrate known identity information, comprehensive alignment scoring data, specific values ​​of various core parameters, and weight allocation table to form an identity verification file. The identity verification file is used to record the core logical data and data basis data for identity matching.

[0179] The system integrates user A's identity information, comprehensive alignment score of 1.0, weighted difference values ​​at each level (0.12, 0.08, 0.15), trajectory overlap data of 80%, transmission path adaptation data of 75%, and weight allocation table (fingerprint 0.38, operation rhythm 0.35, feedback delay 0.27) to form an identity verification document, which records in detail the logic and basis of identity matching.

[0180] Step S158: If the comprehensive alignment score data is lower than the qualified score value in the identity recognition evaluation standard, analyze the level where the weighted difference value is greater than the set difference value, the corresponding cross-dimensional core matching features, set the feature dimension data and interaction level data that need to be collected, and generate supplementary interaction suggestion data.

[0181] Assuming the overall alignment score is 0.7, which is lower than the passing score of 0.8, analysis reveals that the weighted difference value for level three is 0.25, exceeding the set difference value of 0.2. The corresponding cross-dimensional core matching feature is the feedback delay feature. Therefore, it is necessary to supplement the data with more data from the response feedback dimension, such as adding an interaction level four (e.g., feedback after opening an email), and generating supplementary interaction suggestion data, such as "Please check your email to supplement the feedback feature data."

[0182] Step S159: Generate the final identity recognition conclusion based on the identity confirmation document or supplementary interaction suggestion data. The identity recognition conclusion includes the identity attribution information of the interaction subject, matching basis data, and specific operation guidance data for supplementary interaction.

[0183] Based on the identity verification document, the identity recognition conclusion is generated: "The interaction subject is identified as User A (ID: 001), the matching criterion is a comprehensive alignment score of 1.0, the weighted difference values ​​at each level are all below the qualified value, the trajectory overlap is 80%, and the transmission path adaptability is 75%." If supplementary interaction is required, the conclusion includes specific operation instructions for the supplementary interaction, such as "Please perform an email check to supplement the feedback feature data. Operation instructions: Click on the inbox in the email application and browse the email list." Figure 2 This diagram illustrates the hardware structure of a human-computer interaction identity intelligent recognition system 100 based on multimodal perception, as provided in an embodiment of the present invention. Figure 2 As shown, the human-computer interaction intelligent identification system 100 based on multimodal perception may include a processor 1a, a machine-readable storage medium 1b, a bus 1c, and a communication unit 1d.

[0184] Machine-readable storage medium 1b can store data and / or instructions. In some embodiments, machine-readable storage medium 1b can store data acquired from an external terminal. In some embodiments, machine-readable storage medium 1b can store data and / or instructions used by the multimodal perception-based human-computer interaction intelligent identification system 100 to execute or use in order to complete the exemplary methods described in this invention. In a specific implementation, one or more processors 1a execute computer-executable instructions stored in machine-readable storage medium 1b, enabling processor 1a to execute the multimodal perception-based human-computer interaction intelligent identification method as described in the above method embodiments. Processor 1a, machine-readable storage medium 1b, and communication unit 1d are connected via bus 1c, and processor 1a can be used to control the transmission and reception actions of communication unit 1d. The specific implementation process of processor 1a can be found in the various method embodiments executed by the multimodal perception-based human-computer interaction intelligent identification system 100 described above, and their implementation principles and technical effects are similar, so they will not be repeated here.

[0185] Furthermore, embodiments of the present invention also provide a readable storage medium containing computer-executable instructions. When the processor executes the computer-executable instructions, the above-described human-computer interaction identity intelligent recognition method based on multimodal perception is implemented.

[0186] It should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof. Similarly, it should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof.

Claims

1. A human-computer interaction identity intelligent recognition method based on multimodal perception, characterized in that, The method includes: Capture the hierarchical sequence of behavioral signals of the interactive subject during human-computer interaction, trigger multi-dimensional perception feature derivation and transmission based on the gradient features of the hierarchical sequence of behavioral signals, and generate feature derivation and transmission units. The feature derivation and transmission units include physiological perception features, behavioral pattern features, response feedback features, derivation data and transmission relationship data that progress with the interaction level. According to the progressive order of the behavior signal hierarchy sequence, the feature derivation and transmission units are chained together to form a feature derivation and transmission time sequence chain. The feature derivation and transmission time sequence chain records the dynamic evolution trajectory data of feature derivation relationship data and transmission path data as the interaction level progresses. The preset identity benchmark chain library is retrieved. The preset identity benchmark chain library stores the initial benchmark chain corresponding to the known identity. The initial benchmark chain is dynamically updated by combining the current interaction scenario parameters and historical interaction data through a self-learning evolution model to generate a scenario-adaptive benchmark chain. Extract the cross-dimensional core matching features of the feature derivation and transmission time-series chain and the scene adaptation benchmark chain, perform bidirectional trajectory alignment processing based on the dynamic weight allocation strategy, and generate trajectory alignment results. The dynamic weight allocation strategy adjusts the matching weight values ​​in real time according to the importance of the features. An identity recognition conclusion is generated based on the trajectory alignment result. The identity recognition conclusion includes the identity attribution information of the interactive subject, feature derivation and transmission matching data, and weight distribution data.

2. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 1, characterized in that, The process involves capturing a hierarchical sequence of behavioral signals from the interacting subject during human-computer interaction. Based on the gradient features of this hierarchical sequence, multi-dimensional perceptual feature derivation and transmission are triggered to generate... Feature derivation and transfer unit, including: The system continuously captures the hierarchical sequence of behavioral signals generated by the interactive subject in the human-computer interaction interface. The hierarchical sequence of behavioral signals includes the hierarchical electrical signal sequence formed by touch operation, the hierarchical sound wave signal sequence formed by voice interaction, and the hierarchical visual signal sequence formed by posture and movement. Each signal level corresponds to a level of behavioral complexity of the interactive subject. The behavioral signal hierarchy sequence is split into three levels according to the signal complexity classification standard: interaction level one signal, interaction level two signal, and interaction level three signal. Each level signal contains corresponding signal strength parameters, change rate parameters, and duration period parameters. Extract the gradient features of each level of signal. The gradient features include signal complexity gradient, intensity change gradient, and period change gradient. The gradient features record the data of the progressive process of interactive behavior. Based on the gradient features, the feature derivation direction corresponding to each level signal is set. The feature derivation direction points to the new feature type that needs to be derived from the existing perceptual features at that level. The interaction level one signal corresponds to the initial feature derivation, and the interaction level two signal and the interaction level three signal correspond to the composite feature derivation. For each feature derivation direction, the corresponding multi-dimensional perception and acquisition module is called. Each perception and acquisition module is responsible for acquiring one initial feature. Multiple perception and acquisition modules are started synchronously based on the continuous period of hierarchical signals to acquire initial physiological features, initial behavioral features, and initial response features. According to the preset feature transfer rules, the initial features collected at the current level are transferred and fused with the features derived from the previous level to generate the derived features of the current level. The transfer and fusion process includes feature attribute superposition operation, data range expansion operation, and correlation relationship strengthening operation. The transmission strength parameter is calculated during the feature transmission process. The transmission strength parameter is derived based on the degree of correlation between features at different levels. The degree of correlation is determined by the overlap ratio of feature data and the attribute similarity parameter, respectively. The overlap ratio is used to measure the inheritance ratio at the data level, and the attribute similarity parameter is used to measure the similarity at the semantic level. The initial features, derived features, transmission intensity parameters, and time sequence markers of the current level are bound together to form feature transmission binding data. The time sequence markers record the hierarchical order data and time node data generated by the features. The feature transfer binding data is structurally standardized, and the standardized feature transfer binding data is integrated to generate a feature derivative transfer unit containing initial features, derived features, transfer path data, transfer intensity parameters, and temporal information. Each feature derivative transfer unit corresponds to a feature derivative transfer process data of an interaction level.

3. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 1, characterized in that, The step of chaining the feature derivation and transmission units in the progressive order of the behavioral signal hierarchy to form a feature derivation and transmission time sequence includes: Extract the global hierarchical marker of the behavior signal hierarchy sequence. The global hierarchical marker includes the start number, end number, hierarchical progression interval time data, and hierarchical association type data of each hierarchical signal. The hierarchical association type data records the behavioral logical relationship data between the preceding and following hierarchical levels. Add a corresponding hierarchical marker to each feature derivation and transmission unit. The hierarchical marker is consistent with the hierarchical signal number in the global hierarchical marker, and marks the interaction hierarchical position data corresponding to each feature derivation and transmission unit. Based on the hierarchical markers, all the feature derivation and transmission units are initially sorted to form a preliminary hierarchical sequence, which is arranged in the order of interaction level one, interaction level two, and interaction level three. Calculate the transmission connection parameters between two adjacent feature-derived transmission units. The transmission connection parameters include the overlap parameter of the derived features of the preceding and following units and the coherence parameter of the transmission path. The overlap parameter records the inheritance ratio data of the preceding and following derived features. Obtain the interval difference parameter of the time sequence marker. The interval difference parameter independently records the time interval data of the preceding and following units. The feature association strength parameter of adjacent units is adjusted according to the transmission connection parameter. The association strength parameter is used to represent the association relationship between the derived features of the previous unit and the derived basis of the next unit. The adjustment of the association strength parameter is based on the comprehensive judgment result of the overlap parameter and the coherence parameter. Extract the core derived features from each feature derivation and transmission unit. The core derived features are key feature combinations that can uniquely identify identity attributes. The key feature combinations include stable features and key derived features that are retained after multi-level transmission. Establish a transmission association diagram of core derived features of adjacent units. The transmission association diagram records the transmission path data, transmission intensity parameters, and derivation change mode data of the core features of the previous unit to the core features of the next unit. Based on the aforementioned transmission connection parameters, feature association strength parameters, and interval difference parameters of the transmission association graph and time series markers, adjacent feature derivation transmission units are chained together to form unit connection modules. Each unit connection module contains two adjacent units, transmission association information data, and time series interval data. Following the order of the preliminary hierarchical sequence, all unit connection modules are connected in series to form a preliminary time sequence chain. The preliminary time sequence chain records the hierarchical progressive relationship data of feature derivation and transmission. The initial timing chain is structurally optimized by adjusting the transmission correlation parameters of each unit connection module to generate a feature-derived transmission timing chain.

4. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 1, characterized in that, The process of retrieving a preset identity benchmark chain library and dynamically updating the initial benchmark chain using a self-learning evolution model combined with current interaction scenario parameters and historical interaction data to generate a scenario-adaptive benchmark chain includes: Retrieve a preset identity benchmark chain library. The preset identity benchmark chain library stores multiple initial benchmark chains corresponding to known identities. Each initial benchmark chain is the feature-derived transmission benchmark trajectory data formed by the known identity through multiple interactions in a standard scenario. It includes core benchmark features, transmission path data, and benchmark strength parameters at each level. Collect scene parameters of the current human-computer interaction scenario. The scene parameters include four types of environmental data and device-related data that directly affect feature collection: ambient lighting parameters, device operating parameters, network transmission parameters, and interactive interface parameters. Retrieve the historical interaction data corresponding to the known identity. The historical interaction data includes feature derivation and transmission record data, identity verification result data, and feature adjustment record data in similar scenarios in the past. A self-learning evolution model is constructed, which includes a scene adaptation layer, a history fusion layer, and a feature update layer. The scene adaptation layer performs feature influence analysis of scene parameters, the history fusion layer performs change extraction of historical data, and the feature update layer performs dynamic adjustment of baseline features. The scene parameters are input into the scene adaptation layer of the self-learning evolution model. The influence of the scene parameters on the baseline features at each level is analyzed. The scene influence coefficient of each baseline feature is calculated. The scene influence coefficient records the modification requirements of the scene parameters on the feature data. The historical interaction data is input into the historical fusion layer of the self-learning evolution model to extract the change data of feature derivation and transmission during the historical interaction process, including the fluctuation data of feature transmission intensity and the stable data of derived features, and to generate historical change parameters. Based on the scene influence coefficient and historical change parameters, the update magnitude data and update direction data of each core benchmark feature in the initial benchmark chain are calculated through the feature update layer of the self-learning evolution model. The update magnitude data is positively correlated with the scene influence coefficient and historical change parameters. Based on the updated amplitude data and updated direction data, the core reference features, transmission path parameters, and transmission strength parameters of each level in the initial reference chain are dynamically adjusted; The hierarchical coherence parameters of the adjusted baseline chain are calculated. The hierarchical coherence parameters are obtained by the transmission path overlap parameter and the feature association strength parameter. Data on the smoothness of transmission between baseline features at each level and data on the consistency of derived logic are recorded. The adjusted baseline chain is further optimized based on the hierarchical coherence parameters, and the transmission path parameters and transmission strength parameters are fine-tuned to generate a scene-adaptive baseline chain.

5. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 4, characterized in that, Based on the scenario influence coefficient and historical change parameters, the feature update layer of the self-learning evolution model calculates the update magnitude and update direction data of each core benchmark feature in the initial benchmark chain. According to the update magnitude and update direction data, the core benchmark features, transmission path parameters, and transmission strength parameters at each level of the initial benchmark chain are dynamically adjusted, including: Traverse each level in the initial baseline chain and extract the core baseline features, corresponding transmission path parameters, transmission strength parameters, and feature attribute description data for each level. The feature attribute description data includes the feature data range, collection dimension data, and stable value range. The scene influence coefficients are classified according to the feature acquisition dimension into three categories: illumination influence coefficient, device influence coefficient, and network influence coefficient. Each category corresponds to the influence of a type of scene parameter on the feature. The historical change parameters are classified according to feature type into three categories: physiological feature change parameters, behavioral feature change parameters, and response feature change parameters. Each category corresponds to the historical change data of one type of feature. For each core benchmark feature, the corresponding scene impact coefficient classification and historical change parameter classification are extracted. The comprehensive impact parameter is calculated by the feature update layer of the self-learning evolution model. The comprehensive impact parameter is the weighted sum of the scene impact coefficient and the historical change parameter. Based on the comprehensive influence parameter and the data range in the feature attribute description data, the update direction data of the core benchmark feature is set. When the comprehensive influence parameter is positive, the update direction data is the positive expansion of the data range. When the comprehensive influence parameter is negative, the update direction data is the reverse adjustment of the data range. The updated feature conforms to physiological logic and behavioral logic. Based on the magnitude of the comprehensive influence parameter and the stable value range in the feature attribute description data, the update amplitude data of the core benchmark feature is set. The larger the value of the comprehensive influence parameter, the larger the update amplitude data, and the updated feature data is within the stable value range. For the transmission path parameters, adjust the data transmission delay parameters of the transmission path based on the network transmission parameters in the scenario influence coefficient; adjust the node connection method data of the transmission path based on the path change data in the historical change parameters. For the transmission strength parameter, based on the intensity fluctuation data in the historical change parameters and the equipment response parameters in the scene influence coefficient, the numerical range of the transmission strength is adjusted, and the adjusted transmission strength records the data on the closeness of the correlation between features. According to the set update amplitude data and update direction data, the core benchmark features, transmission path parameters, and transmission intensity parameters of each level are adjusted one by one. The adjustment process data and the numerical change data before and after the adjustment of each parameter are recorded to form parameter adjustment record data.

6. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 1, characterized in that, The step of extracting the cross-dimensional core matching features of the feature-derived transit time-series chain and the scene adaptation baseline chain, performing bidirectional trajectory alignment processing based on a dynamic weight allocation strategy, and generating trajectory alignment results includes: Extract the core derived features, transmission path parameters, transmission intensity parameters, and timing markers of each level in the feature derivation and transmission time-series chain to form a feature set to be aligned. The feature set to be aligned is arranged in hierarchical order and contains complete feature derivation and transmission information data. Extract the core benchmark features, transmission path parameters, transmission intensity parameters, and temporal markers of each level in the scene adaptation benchmark chain to form a benchmark feature set. The benchmark feature set is consistent with the hierarchical division of the feature set to be aligned. Identify cross-dimensional core matching features. Cross-dimensional core matching features are features that exist in all three dimensions: physiological perception, behavioral pattern, and response feedback, and whose data fluctuation range is less than a preset fluctuation threshold. These include fixed attribute data in core physiological features, repetitive pattern data in core behavioral features, and stable feedback data in core response features. A dynamic weight allocation model is constructed. The dynamic weight allocation model calculates the matching weight value based on three dimensions: stable data of matching features, identity differentiation data, and scene adaptation data. When the fluctuation range of the stable data is less than a preset fluctuation threshold, the matching weight value increases according to the preset weight adjustment ratio. When the difference data of the identity differentiation data is greater than a preset difference threshold, the matching weight value increases according to the preset weight adjustment ratio. When the deviation value of the scene adaptation data is less than a preset compatibility threshold, the matching weight value increases according to the preset weight adjustment ratio. The dynamic weight allocation model assigns a corresponding matching weight value to each cross-dimensional core matching feature, generating a weight allocation table. The weight allocation table contains the weight value and calculation process data for each cross-dimensional core matching feature. Establish a bidirectional trajectory coordinate system. The horizontal axis of the coordinate system is the hierarchical temporal axis of the feature set to be aligned, and the vertical axis is the hierarchical temporal axis of the baseline feature set. Each node in the coordinate system corresponds to a level of cross-dimensional core matching feature combination data. The cross-dimensional core matching features of the feature set to be aligned and the reference feature set are mapped to the bidirectional trajectory coordinate system to form the feature trajectory to be aligned and the reference feature trajectory. Calculate the relative deviation between the cross-dimensional core matching features of the feature trajectory to be aligned and the cross-dimensional core matching features of the benchmark feature trajectory at the same level node, and obtain the weighted difference value by combining the matching weight value. The transmission path parameters and timing markers of the feature set to be aligned are adjusted according to the weighted difference value to keep the node deviation between the feature trajectory to be aligned and the reference feature trajectory within the minimum range. Repeat the mapping operation, difference calculation operation, and adjustment operation until the sum of the weighted difference values ​​of all level nodes reaches the minimum. Extract the matching parameters of each level. The matching parameters include three types of information data: weighted difference value, matching weight value, and trajectory overlap data. Generate trajectory alignment results.

7. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 6, characterized in that, The process involves identifying cross-dimensional core matching features, constructing a dynamic weight allocation model, and assigning corresponding matching weight values ​​to each cross-dimensional core matching feature using this model, thereby generating a weight allocation table, including: Traverse all features in the feature set to be aligned and the baseline feature set, and filter out common features that exist simultaneously in the three dimensions of physiological perception, behavioral pattern and response feedback. The common features include unified attribute data of identity in different perception dimensions. For each common feature, a stability analysis is performed. Based on the data fluctuation range of the feature in each level, a stability parameter is calculated. When the data fluctuation range is less than the preset fluctuation threshold, the value of the stability parameter increases by a preset ratio. The stability parameter is calculated from the invariant data of the feature during the interaction process. For each common feature, an identity differentiation analysis is performed. By comparing the data of the same common feature of different known identities, a differentiation parameter is calculated. When the difference in feature data between different identities is greater than a preset difference threshold, the value of the differentiation parameter increases by a preset ratio. The differentiation parameter is calculated from the relevant data of the feature to differentiate different identities. For each common feature, a scene adaptation analysis is performed. Combined with the current interaction scene parameters, the adaptation parameters are calculated. When the deviation between the feature data and the scene parameters is less than the preset compatibility threshold, the adaptation parameter value is increased by a preset ratio. The adaptation parameters are calculated from the data content of the feature in the current scene. Common features that exceed preset values ​​in terms of stable parameters, distinguishing parameters, and adaptability parameters are selected and set as cross-dimensional core matching features, while the stable parameters, distinguishing parameters, and adaptability parameters all meet the preset value requirements. A dynamic weight allocation model is constructed, which includes an input layer, a feature processing layer, a weight calculation layer, and an output layer. The input layer receives stable parameters, distinguishing parameters, and adaptation parameters. The feature processing layer performs parameter standardization processing. The weight calculation layer calculates the matching weight values ​​through a weighted summation algorithm. The output layer outputs the final weight values. Different model weight values ​​are assigned to the stable parameters, distinguishing parameters, and adaptation parameters of the dynamic weight allocation model. The stable parameters, distinguishing parameters, and adaptation parameters of each cross-dimensional core matching feature are input into the dynamic weight allocation model. After the standardization operation of the feature processing layer, the initial value of the matching weight is obtained by the weighted summation algorithm of the weight calculation layer. The initial values ​​of the matching weights are normalized, and the normalized matching weight values ​​are bound to the corresponding cross-dimensional core matching features, various parameter values, and model calculation process data to form a weight allocation table.

8. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 1, characterized in that, The step of generating an identity recognition conclusion based on the trajectory alignment result includes: The core parameters in the trajectory alignment result are analyzed, including the weighted difference values ​​of each level, the matching weight values ​​of cross-dimensional core matching features, the overlap data between the feature trajectory to be aligned and the benchmark feature trajectory, and the transmission path adaptation data. Retrieve the preset identity recognition evaluation standard, which includes the qualified values ​​of each core parameter, weight allocation ratio data, and comprehensive evaluation method data. The qualified values ​​are statistically derived based on historical identity verification sample data. The weighted difference value of each level is compared with the corresponding qualified value in the identity recognition assessment standard. The number of levels with weighted difference values ​​lower than the qualified value is counted, and the qualified ratio data of each level is calculated. The trajectory overlap data and transmission path adaptation data are compared with the corresponding qualified values ​​in the identity recognition evaluation standard to determine the qualified status of the two indicators. Based on the weight allocation ratio data in the identity recognition evaluation standard, the hierarchical qualification ratio data, trajectory overlap data qualification status, and transmission path adaptation data qualification status are transformed into quantitative scoring data. By using the comprehensive evaluation method data in the identity recognition evaluation standard, the various quantitative score data are weighted and calculated to obtain the comprehensive alignment score data. If the comprehensive alignment score data is higher than the qualified score value in the identity recognition evaluation standard, the known identity information to which the scenario adaptation benchmark chain belongs is extracted, including unique identity identification data, feature derivation and transmission benchmark model data, and historical interaction feature data. Integrate known identity information, comprehensive alignment scoring data, specific values ​​of various core parameters, and weight allocation table to form an identity verification file. The identity verification file is used to record the core logical data and data basis data for identity matching. If the comprehensive alignment score data is lower than the qualified score value in the identity recognition evaluation standard, analyze the level where the weighted difference value is greater than the set difference value, the corresponding cross-dimensional core matching features, set the feature dimension data and interaction level data that need to be collected, and generate supplementary interaction suggestion data. Based on the identity verification document or supplementary interaction suggestion data, a final identity recognition conclusion is generated. The identity recognition conclusion includes the identity attribution information of the interaction subject, matching basis data, and specific operation guidance data for supplementary interaction.

9. The human-computer interaction identity intelligent recognition method based on multimodal perception according to claim 2, characterized in that, For each feature derivation direction, the corresponding multi-dimensional sensing and acquisition module is invoked. Each sensing and acquisition module is responsible for acquiring one initial feature. Multiple sensing and acquisition modules are synchronously started based on the continuous period of hierarchical signals to acquire initial physiological features, initial behavioral features, and initial response features, including: Based on the characteristic derivation direction of each level of signal, the initial feature types to be collected are set. The initial feature types include fingerprint texture data, facial contour data, voiceprint frequency data, and palm vein data in the initial physiological features; operation rhythm data, movement amplitude data, and interaction interval data in the initial behavioral features; and feedback delay data, response content data, and operation accuracy data in the initial response features. Each initial feature type has a dedicated perception and acquisition module: fingerprint texture data corresponds to the fingerprint acquisition module, facial contour data corresponds to the visual acquisition module, voiceprint frequency data corresponds to the audio acquisition module, and operation rhythm data corresponds to the touch acquisition module. Each module has corresponding acquisition functions and technical parameters. Retrieve the device operating parameters for each sensing and acquisition module, including acquisition accuracy parameters, signal response speed parameters, data transmission rate parameters, and operating voltage range parameters; Based on the duration of the signal at the current level, calculate the acquisition duration of each sensing and acquisition module; Based on the signal response speed parameters of the sensing and acquisition modules, the start-up advance time of each sensing and acquisition module is calculated; According to the calculated start-up advance time, start-up instructions are sent to the corresponding sensing and acquisition modules in sequence. The start-up instructions include acquisition duration data, acquisition frequency data, and data transmission format data. After all sensing and acquisition modules are started, the operating status of each sensing and acquisition module is monitored in real time. The device status feedback signal is used to determine whether the module is in normal acquisition state. If an abnormality occurs, a restart command is sent immediately. Based on the intensity change rate parameter of the hierarchical signal, the acquisition frequency of the sensing acquisition module is adjusted. When the intensity change rate parameter exceeds the preset range, the acquisition frequency value is increased, and when the intensity change rate parameter is within the preset range, the initial acquisition frequency is maintained. Each sensing and acquisition module collects corresponding initial feature data in real time according to the set acquisition parameters. The fingerprint acquisition module collects image data of fingerprint texture, the visual acquisition module collects dynamic image data of facial contour, the audio acquisition module collects sound wave data of voiceprint frequency, and the palm vein acquisition module collects palm vein data. It receives initial feature data transmitted from each sensing and acquisition module in real time, adds acquisition timestamp data and module identification data to each data block, associates it with the corresponding hierarchical signal and acquisition module, and completes the initial feature acquisition operation.

10. A human-computer interaction intelligent identity recognition system based on multimodal perception, characterized in that, The system includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to run the programs, instructions, or code in the memory to implement the human-computer interaction intelligent identity recognition method based on multimodal perception as described in any one of claims 1-9.

Images