Lighting parameter generation method based on behavior semantic analysis, lighting system and terminal

Abstract

The application provides a lighting parameter generation method based on behavior semantic analysis, a lighting system and a terminal. The method comprises: acquiring multi-modal behavior clues of a user in a lighting environment; constructing a behavior semantic graph based on the multi-modal behavior clues; inputting the behavior semantic graph into a pre-trained emotion state inference model, outputting a structured emotion state parameter through hierarchical emotion inference by the emotion state inference model; the structured emotion state parameter comprises an emotion type, an emotion intensity, an emotion persistence parameter and an emotion confidence; and generating a target lighting parameter matched with the user's emotion state and the lighting environment according to the structured emotion state parameter. Through accurate analysis of the user's behavior semantics, the application can generate a lighting parameter matched with the user's real emotion state and complex lighting environment, thereby improving the intelligent degree and context adaptability of lighting regulation.

Description

Technical Field

[0001] This application belongs to the field of intelligent lighting technology, and relates to a method for generating lighting parameters based on behavioral semantic analysis, a lighting system and a terminal. Background Technology

[0002] With the development of smart lighting and smart home technologies, lighting systems are evolving from passive control devices into intelligent systems that can deeply understand user status, adapt to changes in the spatial environment, and proactively generate appropriate lighting parameters.

[0003] In existing technologies, one type of solution uses physiological or quasi-physiological data such as millimeter-wave radar, infrared thermal imaging, acoustic arrays, heart rate, respiration, or body surface temperature to infer a user's health status, emotional state, or physiological load, and then performs home interactions such as lighting, temperature control, or reminders accordingly. While this type of solution can obtain relatively rich sensory data, it requires additional sensors, and some signals involve human physiological privacy, voice, or image information. Furthermore, long-term deployment costs are high, and user acceptance and privacy security are insufficient.

[0004] Another approach calculates cognitive load, spatial perception response patterns, or lighting needs based on spatial structure or the dynamic characteristics of tourists' movements in specific public spaces, and generates lighting control commands based on the results. This type of approach can control lighting without relying on physiological data, but it is geared towards tourist trajectories and spatial structures in cultural tourism or public spaces, and cannot directly solve the problem of low-intrusion inference of user emotional state parameters in smart home lighting. Summary of the Invention

[0005] This application provides a lighting parameter generation method, lighting system, and terminal based on behavioral semantic analysis, which solves the problem that existing lighting systems are unable to accurately infer the user's emotional state under conditions of low privacy risk and low intrusion, and generate lighting parameters that match the user's emotional state and lighting environment accordingly.

[0006] In a first aspect, this application provides a method for generating illumination parameters based on behavioral semantic analysis, comprising: acquiring multimodal behavioral cues of a user in a lighting environment; the multimodal behavioral cues include lighting interaction events and their corresponding contextual metadata, the contextual metadata including lighting temporal rhythm events, content type tag events with preset non-sensitive standards, and lighting device spatial topology events; constructing a behavioral semantic graph based on the multimodal behavioral cues; the behavioral semantic graph describes the semantic associations between the multimodal behavioral cues through multi-type nodes and multi-type relational edges, and characterizes the strength of different associations through relational edge weights; inputting the behavioral semantic graph into a pre-trained emotion state inference model, and outputting structured emotion state parameters by the emotion state inference model through hierarchical emotion inference; the structured emotion state parameters include emotion type, emotion intensity, emotion persistence parameters, and emotion confidence; and generating target illumination parameters that match the user's emotion state and the lighting environment based on the structured emotion state parameters.

[0007] In one implementation of the first aspect, constructing a behavioral semantic graph based on the multimodal behavioral cues includes: converting the multimodal behavioral cues into a sequence of event tuples with a unified data structure; constructing a user's personal behavioral baseline based on historical user behavior data; performing difference calculations on the event tuple sequence and the user's personal behavioral baseline at multiple time scales to obtain a behavioral offset feature vector characterizing the degree of deviation of user behavior from its historical habits; performing reliability assessment on the multimodal behavioral cues based on a preset reliability evaluation index to obtain a modal reliability vector; obtaining the temporal correlation between the content type tag event and the lighting interaction event, and generating a content interference masking matrix based on the temporal correlation; the content interference masking matrix is ​​used to reduce the misleading influence of irrelevant or interfering content on the inference of the user's true emotional state; performing gated fusion processing on the behavioral offset feature vector based on the modal reliability vector and the content interference masking matrix to obtain a fused feature vector; and determining the nodes and relational edges used to construct the behavioral semantic graph based on the fused feature vector.

[0008] In one implementation of the first aspect, the method of performing difference calculations on the event tuple sequence and the user's personal behavior baseline at multiple time scales to obtain a behavior offset feature vector representing the degree of deviation of the user's behavior from its historical habits includes: establishing a multi-scale time window; the multi-scale time window includes a short-term window, a medium-term window, and a long-term window, wherein the short-term window is used to represent the user's immediate emotional disturbance, the medium-term window is used to represent the user's daily state trend, and the long-term window is used to represent the user's long-term habit baseline; calculating the standardized difference values ​​between the statistical observation values ​​of the event tuple sequence and the user's personal behavior baseline in each scale time window; and combining the standardized difference values ​​corresponding to each scale time window to obtain the behavior offset feature vector.

[0009] In one implementation of the first aspect, the reliability assessment of the multimodal behavioral clues based on preset reliability evaluation indicators to obtain a modal reliability vector includes: defining multiple reliability evaluation indicators; the reliability evaluation indicators include data integrity indicators, time freshness indicators, data stability indicators, user identity confidence indicators, and modal conflict degree indicators, wherein the data integrity is defined as the ratio of the number of valid behavioral clues to the theoretical number of behavioral clues within a preset time window; the time freshness indicator is defined as the most recent sampling time of the behavioral clue; the data stability indicator is defined as the statistical variance of the behavioral clues within the preset time window; the user identity confidence indicator is determined based on user identity recognition results or device binding information; the modal conflict degree indicator is defined as... The modal reliability is defined as the consistency score between any modal behavior clue and other modal behavior clues in terms of temporal features, intensity features, or pattern features. Each modal behavior clue is traversed, and the data integrity index, time freshness index, data stability index, user identity confidence index, and modal conflict degree index corresponding to each modal behavior clue are obtained. The data integrity index, time freshness index, data stability index, user identity confidence index, and modal conflict degree index are normalized and weighted and summed according to preset weight coefficients to obtain the modal reliability corresponding to each modal behavior clue. The modal reliability vector is obtained by combining the modal reliability vectors corresponding to all modal behavior clues in modal order.

[0010] In one implementation of the first aspect, the behavioral semantic graph is input into a pre-trained emotion state inference model. The emotion state inference model infers structured emotion state parameters through hierarchical emotion inference, including: identifying user behavior goals and activity tendencies based on the behavioral semantic graph to obtain an activity intention vector; extracting intensity feature values ​​based on the fused feature vector and mapping them to the emotion intensity; inferring the user's emotion type probability distribution through conditional probability calculation based on the activity intention vector, the fused feature vector, and the content interference masking matrix, and determining the emotion type according to the emotion type probability distribution; evaluating the stability of the user's emotion state based on the dynamic change characteristics of the emotion type and the emotion intensity in the time domain to generate the emotion persistence parameter and the emotion confidence; and combining the emotion type, the emotion intensity, the emotion persistence parameter, and the emotion confidence into the structured emotion state parameters.

[0011] In one implementation of the first aspect, the stability assessment of the user's emotional state based on the dynamic change characteristics of the emotion type and the emotion intensity in the time domain, in order to generate the emotion persistence parameter and the emotion confidence score, includes: calculating the time change rate of the probability distribution of the emotion type and the time change rate of the emotion intensity between adjacent time windows; obtaining a historical emotion state transition matrix by statistically learning the transition patterns between emotion states based on the user's historical emotion data; obtaining the duration of the behavior deviation feature by statistically analyzing the duration of continuous deviations from the user's personal behavior baseline in the behavior deviation feature vector; and calculating the emotion persistence parameter and the emotion confidence score based on the time change rate of the probability distribution of the emotion type, the time change rate of the emotion intensity, the historical emotion state transition matrix, and the duration of the behavior deviation feature.

[0012] In one implementation of the first aspect, the method further includes: if the emotional confidence level is lower than a preset confidence threshold, or the modal conflict level index is higher than a preset conflict threshold, then a conservative output strategy is executed; the conservative output strategy includes at least one of outputting zone lighting suggestions, outputting local lighting suggestions, maintaining the current lighting state suggestions, or triggering user confirmation feedback.

[0013] In one implementation of the first aspect, generating target lighting parameters that match the user's emotional state and the surrounding lighting environment based on the structured emotional state parameters includes: acquiring the user identity confidence index, space occupancy weight, and device control weight for each user in a multi-user home lighting scenario, wherein the space occupancy weight is calculated based on the user's dwell time, activity intensity, or spatial location importance in a preset space area; the device control weight is determined based on user role permissions, historical control frequency, or preset priority rules; calculating the difference between the structured emotional state parameters of any two users and determining whether the difference exceeds a preset threshold; if the difference between the structured emotional state parameters of at least two users exceeds the preset threshold, it is determined that there is a multi-user emotional state conflict; weighting and aggregating the structured emotional state parameters of each user with an emotional state conflict based on the user identity confidence index, the space occupancy weight, and the device control weight to generate aggregated emotional state parameters; and generating target lighting parameters that match the multi-user home lighting scenario based on the aggregated emotional state parameters.

[0014] Secondly, this application provides a lighting system, comprising: a lighting device; a light parameter generation module, configured to execute the light parameter generation method based on behavioral semantic analysis as described in any of the preceding claims to generate target light parameters; and a lighting control module, connected to the light parameter generation module and the lighting device, configured to generate lighting control instructions based on the target light parameters and control the lighting device to perform lighting adjustment based on the lighting control instructions.

[0015] Thirdly, this application provides a terminal, comprising: a memory for storing a computer program; and a processor for executing the computer program stored in the memory, so that the terminal performs the method described in any of the above-mentioned embodiments.

[0016] As described above, the illumination parameter generation method, illumination system, and terminal based on behavioral semantic analysis described in this application have the following beneficial effects:

[0017] (1) It can effectively protect user privacy and security, achieve low-intrusion behavior semantic awareness, and reduce additional hardware costs and privacy risks;

[0018] (2) By constructing a multi-scale personal baseline difference mechanism, the interference of individual differences on emotion inference is effectively eliminated, and the personalized adaptation ability and judgment accuracy of emotion state recognition are significantly improved.

[0019] (3) By introducing a content interference shielding mechanism, the accuracy and semantic purity of sentiment inference in multimedia usage scenarios are significantly improved;

[0020] (4) In uncertain scenarios with low confidence or conflicting information from multiple sources, the conservative output strategy ensures that the mood lighting system will not produce abrupt, highly saturated, highly dynamic or unsuitable light environments in uncertain scenarios, thereby improving the stability and safety of long-term home use. Attached Figure Description

[0021] Figure 1 The flowchart shown is a method for generating illumination parameters based on behavioral semantic analysis in one embodiment of this application.

[0022] Figure 2 The diagram shown is a flowchart of the construction process of a behavioral semantic graph in one embodiment of this application.

[0023] Figure 3 The flowchart shown is a process for generating behavior offset feature vectors in one embodiment of this application.

[0024] Figure 4 The flowchart shown is a process for reliability assessment of multimodal behavioral cues in one embodiment of this application.

[0025] Figure 5 The diagram shows the workflow of an emotion state inference model in one embodiment of this application.

[0026] Figure 6 The flowchart shown is a process for assessing the stability of a user's emotional state in one embodiment of this application.

[0027] Figure 7 The flowchart shown is a process for generating target illumination parameters in one embodiment of this application.

[0028] Figure 8 The diagram shown is a structural schematic of a lighting system according to an embodiment of this application.

[0029] Figure 9 The diagram shown is a structural schematic of a terminal according to an embodiment of this application. Detailed Implementation

[0030] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0031] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0032] In practical applications of mood lighting in the home, users generate a massive number of lighting interaction events daily through terminals such as remote controls, smart panels, mobile apps, voice assistants, and audio-visual and gaming devices. Although these lighting interaction events differ from direct physiological signals or continuous spatial movement trajectories, their dynamic changes in dimensions such as temporal rhythm distribution, frequency of manual intervention, content selection preferences, scene switching logic, multi-device collaboration modes, and spatial migration characteristics contain rich behavioral semantic information, which can indirectly reflect the user's emotional state, such as fatigue, excitement, anxiety, focus, relaxation, or sleepiness.

[0033] However, current technologies have not fully explored the potential value of such lighting interaction events. Existing solutions often treat these lighting interaction events merely as discrete control commands or simple automated triggering conditions, lacking the ability to deeply analyze the underlying behavioral semantics. This results in a large amount of lighting interaction data that could be used to characterize users' emotional states being idle or inefficiently utilized, limiting the performance improvement of home smart lighting systems in terms of seamless and personalized emotional adaptation.

[0034] To address the shortcomings of existing technologies, this application provides a lighting parameter generation method, lighting system, and terminal based on behavioral semantic analysis. This method can accurately infer the user's emotional state under conditions of low privacy risk and low intrusion, and generate lighting parameters that match the user's emotional state and lighting environment accordingly, effectively realizing emotional and contextual adaptive lighting.

[0035] The technical solution provided in this application has broad scene adaptability and can be applied to a variety of application scenarios such as home mood lighting, health lighting, age-friendly lighting, audio-visual entertainment lighting, child care lighting, and personalized light environment control.

[0036] This application is preferably deployed in a home gateway or local edge computing node. Through localization and data anonymization, this solution fundamentally cuts off the transmission link and storage risks of raw privacy data. Compared with traditional smart lighting systems that rely on centralized cloud processing, this application significantly reduces the risks of data leakage, misuse, and compliance breaches while ensuring the integrity of the emotion inference function, especially meeting the highly sensitive requirements of users in the home environment.

[0037] The following will describe in detail the principle and implementation of a lighting parameter generation method, lighting system and terminal based on behavioral semantic analysis according to this embodiment, so that those skilled in the art can understand the lighting parameter generation method, lighting system and terminal based on behavioral semantic analysis according to this embodiment without creative effort.

[0038] Please see Figure 1 The above is a flowchart of a lighting parameter generation method based on behavioral semantic analysis in one embodiment of this application.

[0039] like Figure 1 As shown, this embodiment provides a method for generating illumination parameters based on behavioral semantic analysis, including the following steps S100 to S400.

[0040] In step S100, multimodal behavioral cues of the user in the lighting environment are obtained.

[0041] The lighting environment refers to the interactive space of a home's lighting environment, which is composed of lighting equipment, intelligent control terminals, and users. It encompasses the physical spaces and digital interfaces where users interact with the lighting environment and related intelligent devices in their daily lives. For example, the lighting environment can be not only a home living space, but also an office space, a commercial exhibition space, or the interior space of public infrastructure.

[0042] The multimodal behavioral cues include lighting interaction events and their corresponding contextual metadata, which include lighting time rhythm events, content type tag events with preset non-sensitive standards, and lighting device spatial topology events.

[0043] Specifically, the lighting interaction events include at least one of the following: turning the lighting device on, turning it off, adjusting brightness, adjusting color temperature, adjusting color, adjusting dynamic effects, switching scenes, reverting scenes, interrupting automatic mode, and manually covering.

[0044] The time rhythm events include the frequency and duration of lighting use in different time periods, reflecting differences in use between weekdays and non-weekdays, interactive changes at night or in the early morning, and shifts in usage rhythms over days, weeks, or months. For example, in the past three months, the average time the desk lamp in the study has been turned on has gradually shifted from 21:00 to 23:30, and the proportion of usage time on weekends has increased from 30% to 60%.

[0045] The content type tag events refer to content category tags such as audio, video, games, reading, sleep aids, sports, or social media linked to lighting. For example, when a smart speaker or mobile app launches a white noise / meditation program, the corresponding tag is: [Sleep Aid / Relaxation]. This tag can be provided by audio-visual devices, gaming terminals, mobile applications, or home gateways. It should be noted that the system only records content category tags and does not collect or store the actual content itself to ensure privacy and security.

[0046] The spatial topology events of the lighting equipment are used to describe the usage relationships of lighting equipment in different spaces, including states such as simultaneous use, alternating use, multi-space switching, multi-device collaboration, or isolated use of equipment. For example, if the dining room chandelier, the sideboard light strip, and the kitchen worktable lamp are adjusted to the preset "dining scene" parameters at almost the same time, the topology state is cross-space synchronous collaboration, indicating that the equipment in multiple physical spaces constitutes a logically integrated functional area.

[0047] In one embodiment of this application, the multimodal behavioral cues are presented in the form of an event stream, and the data source is strictly limited to the interaction logs, status logs, and context labels of the smart lighting system and associated IoT devices.

[0048] It should be noted that this application explicitly excludes the collection and reliance on physiological signals such as heart rate, respiration, body temperature, electroencephalogram (EEG), and electrodermal conductance (TEF), as well as biometric or privacy-sensitive data such as video images and voice content. This ensures user privacy and security from the source, achieves low-intrusion behavioral semantic perception, and reduces additional hardware costs and privacy risks.

[0049] In this implementation, by introducing multimodal behavioral cues, the semantic ambiguity problem inherent in a single signal source is effectively overcome, improving the accuracy and robustness of activity intent recognition and enabling the generated lighting parameters to more accurately match the user's actual needs. Furthermore, compared to existing solutions that rely on visual or physiological signals, this application significantly reduces the system's hardware costs and computational overhead while fully protecting user privacy and security.

[0050] In step S200, a behavioral semantic graph is constructed based on the multimodal behavioral cues.

[0051] The behavioral semantic graph is a heterogeneous semantic graph that takes user behavior events as its core and integrates time, space, device, content and user intervention information. It describes the semantic association between the multimodal behavioral cues through multi-type nodes and multi-type relation edges, and characterizes the strength of different associations through relation edge weights.

[0052] Specifically, the node types in the behavioral semantic graph include user nodes, time nodes, spatial nodes, device nodes, content nodes, and event nodes; the relationship edge types include temporal adjacency edges, spatial co-occurrence edges, device linkage edges, content association edges, and manual intervention edges; the relationship edge weights are determined based on the weight influence factors corresponding to the relationship edge type.

[0053] In one embodiment of this application, the weighting factors include event time interval, spatial adjacency, device linkage strength, user manual intervention strength, and content interference shielding coefficient.

[0054] Specifically, the weights of each type of relation edge in the behavioral semantic graph are determined as follows: the weight of the temporal adjacency edge is determined based on the event time interval; the shorter the time interval, the stronger the continuity of the behavior, and the higher the weight. The weight of the spatial co-occurrence edge is determined based on the spatial adjacency relationship; this weight reflects the degree of correlation between behaviors occurring in different spaces. The weight of the device linkage edge is determined based on the device linkage strength; this weight is used to characterize the degree of behavioral coupling when multiple devices interact collaboratively. The weight of the content association edge is modified by combining content relevance and the content interference shielding coefficient to filter out the interference of non-emotionally related noise content on semantic expression. The weight of the manual intervention edge is determined based on the intensity of user manual intervention; explicit adjustment operations are given higher semantic saliency weights to accurately capture the user's active emotional expression intention.

[0055] Please see Figure 2 The diagram shown is a flowchart of the construction process of a behavioral semantic graph in one embodiment of this application.

[0056] like Figure 2 As shown, constructing a behavioral semantic graph based on the multimodal behavioral cues includes the following steps S210 to S270.

[0057] In step S210, the multimodal behavioral cues are converted into a sequence of event tuples with a unified data structure.

[0058] Each event tuple in the event tuple sequence can be represented as:

[0059] x_i={t_i,u_i,r_i,d_i,a_i,p_i,c_i},

[0060] Where x_i is an event tuple, which is a set of identifiers; t_i is a collection time identifier, used to mark the timestamp of the occurrence of the behavior clue; u_i is a user identity confidence identifier, used to characterize the credibility of the user to whom the current behavior belongs; r_i is a spatial location identifier, used to indicate the physical spatial area where the behavior clue occurred; d_i is a lighting device identifier, used to uniquely identify the lighting device that triggered or associated with the behavior; a_i is a behavior action identifier, used to encode the specific interaction action type; p_i is a lighting parameter identifier, used to record the lighting state parameters related to the occurrence of the behavior; and c_i is a context category identifier, used to label the non-sensitive context label category to which the behavior clue belongs.

[0061] In this implementation, by converting the multimodal behavioral cues into a sequence of event tuples with a unified data structure, the format differences between heterogeneous data sources can be eliminated, providing a standardized input representation for subsequent semantic graph construction.

[0062] In step S220, a baseline of user personal behavior is constructed based on the user's historical behavior data.

[0063] The user's historical behavior data can be multimodal behavioral clues obtained from the user within a historical time window.

[0064] Based on users' historical behavior data, the construction of a user's personal behavior baseline includes: performing cluster analysis on users' historical behavior data to extract typical behavior patterns of users in normal scenarios; statistically analyzing the frequency and temporal distribution patterns of various behavioral actions, device linkages, and context categories to generate a normal behavior probability distribution model; and integrating the typical behavior patterns with the normal behavior probability distribution model to form a personal behavior baseline that represents the user's individual habit characteristics.

[0065] In this embodiment, the user's personal behavior baseline includes the mean and the standard deviation. The mean is used to characterize the average level of a certain behavioral characteristic of the user within a historical time window, and the standard deviation is used to characterize the degree of fluctuation of the behavioral characteristic relative to the average level.

[0066] In step S230, the event tuple sequence and the user's personal behavior baseline are differentially calculated at multiple time scales to obtain a behavior offset feature vector that characterizes the degree of deviation of the user's behavior from its historical habits.

[0067] Please see Figure 3 The above is a flowchart showing the generation process of the behavior offset feature vector in one embodiment of this application.

[0068] like Figure 3As shown, the process of performing differential calculations on the event tuple sequence and the user's personal behavior baseline at multiple time scales to obtain a behavior offset feature vector that characterizes the degree of deviation of the user's behavior from its historical habits includes the following steps S231 to S233.

[0069] In step S231, a multi-scale time window is established.

[0070] The multi-scale time window includes a short-term window, a medium-term window, and a long-term window, wherein the short-term window is used to characterize the user's immediate emotional disturbances, the medium-term window is used to characterize the user's daily state trends, and the long-term window is used to characterize the user's long-term habit baseline.

[0071] For example, a short-term window W_s, a medium-term window W_m, and a long-term window W_l can be set respectively. The value of W_s ranges from 1 to 10 minutes and is used to capture the instantaneous behavioral disturbances caused by sudden emotional stimuli, reflecting the instantaneous fluctuation characteristics of emotional state. The value of W_m ranges from 30 to 180 minutes and is used to depict the evolution trend of the user's behavioral state within a single day, representing the continuous or transitional changes of emotion on a multi-hour scale. The value of W_l can be 7 days, 14 days, or 30 days and is used to anchor the user's long-term behavioral habit baseline.

[0072] In step S232, the standardized difference between the statistical observations of the event tuple sequence within each time window and the user's personal behavior baseline is calculated.

[0073] For example, the standardized difference between the statistical observation f_k(t) of the k-th modal behavioral cue within the current time window and the mean μ_k and standard deviation σ_k used to characterize the baseline of the user's personal behavior can be calculated using the following formula:

[0074] Δf_k(t)=(f_k(t)-μ_k) / (σ_k+ε);

[0075] Here, ε is a smoothing term to prevent the denominator from being zero, which is used to ensure the stability of numerical calculations.

[0076] In this embodiment, the statistical observations may be the frequency of nighttime light activation, the ratio of manual operation events to automatic control events within a window, the probability of scene rollback conditions, or the usage ratio of isolated devices.

[0077] In step S233, the standardized difference values ​​corresponding to each scale time window are combined to obtain the behavior offset feature vector.

[0078] The user behavior offset feature vector includes at least one of the following: time rhythm offset, intervention disturbance, scene hesitation, device spatial consistency, and automatic control deviation.

[0079] Specifically, the temporal rhythm offset is used to characterize the degree of deviation of the user's current behavior temporal distribution from historical habitual rhythms, reflecting the impact of emotional fluctuations on behavioral rhythm; the intervention disturbance is used to characterize the degree of abnormality of the frequency and intensity of the user's manual intervention operations relative to the baseline, capturing the significance of proactive emotional expression; the scene hesitation is used to characterize the degree of hesitation and repetition of the user during space switching or device selection, reflecting the emotional state implied by decision uncertainty; the device spatial consistency is used to characterize the degree of matching between the current device usage mode and spatial location relative to historical norms, identifying non-habitual space-device associations generated by emotion-driven factors; the automatic control deviation is used to characterize the degree of user coverage or modification of automatic lighting control strategies, reflecting the intensity of conflict between subjective will and system presets.

[0080] This application constructs a multi-scale personal baseline difference mechanism, which anchors the user's own behavioral norms at three levels: short-term, medium-term, and long-term. This ensures that the behavioral offset feature vector only responds to relative changes within the individual, thereby effectively eliminating the interference of individual differences on emotion inference and significantly improving the personalized adaptation capability and judgment accuracy of emotion state recognition.

[0081] In step S240, the reliability of the multimodal behavioral cues is evaluated based on preset reliability evaluation indicators to obtain a modal reliability vector.

[0082] Please see Figure 4 The above is a flowchart of a reliability assessment of multimodal behavioral cues in one embodiment of this application.

[0083] like Figure 4 As shown, the reliability assessment of the multimodal behavioral cues based on the preset reliability evaluation index to obtain the modal reliability vector includes the following steps S241 to S244.

[0084] In step S241, multiple reliability evaluation metrics are defined.

[0085] The reliability evaluation metrics include data integrity metrics, time freshness metrics, data stability metrics, user identity confidence metrics, and modal conflict metrics. Specifically, data integrity is defined as the ratio of the number of valid behavioral cues to the theoretical number of behavioral cues within a preset time window; time freshness metrics are defined as the most recent sampling time of a behavioral cue; data stability metrics are defined as the statistical variance of behavioral cues within a preset time window; user identity confidence metrics are determined based on user identity recognition results or device binding information; and modal conflict metrics are defined as the consistency score between any modal behavioral cue and other modal behavioral cues in terms of temporal characteristics, intensity characteristics, or pattern characteristics.

[0086] In step S242, each modal behavior clue is traversed, and the data integrity index, the time freshness index, the data stability index, the user identity confidence index, and the modal conflict degree index corresponding to each modal behavior clue are obtained respectively.

[0087] In step S243, the data integrity index, the time freshness index, the data stability index, the user identity confidence index, and the modal conflict degree index are normalized respectively, and then weighted and summed according to preset weight coefficients to obtain the modal reliability corresponding to each modal behavior clue.

[0088] In step S244, the modal reliability corresponding to all modal behavior cues is combined in modal order to obtain the modal reliability vector.

[0089] This application introduces modal reliability, which allows for dynamic adjustment of the contribution weight of each modal behavioral cue in the emotional state inference process based on the real-time data quality of each modality.

[0090] For example, when a behavioral cue has missing data, abnormal values, or semantic conflicts with other modal cues, the system adaptively reduces the fusion weight of the cue based on the modality reliability vector, thereby suppressing the negative impact of low-quality data on the sentiment inference results.

[0091] Through the aforementioned dynamic weighting strategy, the emotion inference process can effectively avoid misjudgments caused by single-modal failure or noise interference, and significantly improve the robustness and inference reliability of the system in complex uncontrolled environments.

[0092] In step S250, the temporal correlation between the content type tag event and the lighting interaction event is obtained, and a content interference masking matrix is ​​generated based on the temporal correlation.

[0093] While users may be watching movies, playing games, or moving to music, such content may drive changes in lighting, but this does not necessarily reflect the user's true emotional state. If changes in lighting parameters are primarily triggered by media content and the user has not manually intervened, reverted to a previous scene, or engaged in abnormal spatial transitions, then the weighting of this type of interactive content label in the emotion inference results should be reduced.

[0094] Based on this need, this application introduces a content interference shielding matrix to reduce the misleading influence of irrelevant or interfering content on the inference of users' true emotional state.

[0095] In one embodiment of this application, the step of generating the content interference masking matrix includes: taking the time of the lighting interaction event as the anchor point, extracting an analysis window of a preset duration before and after the event, retrieving content type tag events within the window, and extracting related features such as time offset, overlap duration, and trigger order. When the time difference between the lighting parameter change and the content tag trigger time is less than a threshold, the parameter value matches the content lighting effect template, and there is no manual intervention or scene rollback in the subsequent observation window, the change is determined to be content-driven. For content-driven events, a suppression coefficient with a value range of [0,1] is generated by combining the time offset, the preset interference level of the content type, and the user's historical adjustment sensitivity through a weighted attenuation function. The suppression coefficients corresponding to all content tags in the current window are arranged according to the category dimension to generate a diagonal content interference masking matrix, which is used to perform weighted attenuation of the original content tag features during subsequent sentiment inference.

[0096] In this implementation, by introducing a content interference shielding mechanism, the system can effectively decouple the atmosphere rendering effect of media content from the actual outward manifestation of the user's inner emotions, avoid misjudging the automated light response induced by the climax of the plot, the tense rhythm of the game, or the beat of the music as the user's own emotional fluctuations, and significantly improve the accuracy and semantic purity of emotion inference in multimedia usage scenarios.

[0097] In step S260, based on the modal reliability vector and the content interference masking matrix, the behavior offset feature vector is subjected to gated fusion processing to obtain a fused feature vector.

[0098] In one embodiment of this application, the behavior offset feature vector is gated and fused using the following formula based on the modal reliability vector and the content interference masking matrix:

[0099] F_t = Gate(B_t, R_t, M_t);

[0100] Where Gate() represents the element-wise weighted gating function, F_t is the fused feature vector, B_t is the behavior offset feature vector, R_t is the modal reliability vector, and M_t is the content interference masking matrix.

[0101] In this implementation, through the aforementioned gated fusion processing, the system achieves adaptive purification and integration of multi-source heterogeneous cues at the feature level. On one hand, the modal reliability vector filters out unreliable feature components caused by device offlineness, signal loss, or conflicts from the data quality dimension; on the other hand, the content interference masking matrix suppresses non-emotional pseudo-feature components caused by media content linkage from the semantic purity dimension. These two elements work synergistically on the behavioral offset feature vector, ensuring that the final output fused feature vector retains only high-confidence behavioral change information that truly reflects the user's internal emotional state. This provides a high-quality, low-noise input representation for the subsequent emotion state inference model, significantly improving the robustness and accuracy of emotion inference results in complex family environments.

[0102] In step S270, based on the fused feature vector, the nodes and relation edges used to construct the behavioral semantic graph are determined.

[0103] In one embodiment of this application, determining the nodes and relational edges for constructing the behavioral semantic graph based on the fused feature vector includes: performing semantic parsing and entity extraction on the fused feature vector to identify semantic units such as behavioral subjects, behavioral actions, interactive objects, environmental contexts, and emotional representations contained therein, and instantiating each semantic unit as a node in the semantic graph; determining the relational edges between nodes based on the temporal co-occurrence relationship, causal relationship, and spatial proximity relationship between each semantic unit in the fused feature vector; and performing topological assembly of the determined set of nodes and set of relational edges to generate the user behavior semantic graph corresponding to the current time window.

[0104] Compared to directly using the original feature vectors for sentiment inference, behavioral semantic graphs not only retain high-confidence behavior-emotion association information, but also capture the contextual dependencies and logical constraints between multiple cues through explicit relational modeling. This helps to enhance the sentiment inference model's ability to resolve ambiguous scenarios and perceive long-term behavioral patterns.

[0105] In step S300, the behavioral semantic graph is input into a pre-trained emotion state inference model, which then outputs structured emotion state parameters through hierarchical emotion inference.

[0106] When training the emotion state inference model, content type labels can be introduced as contextual confounding variables to achieve deconfounding learning of content-induced pseudo-emotion signals.

[0107] Specifically, when content type labels represent media content with strong emotional induction, such as horror, thriller, sadness, or comedy, deconfounding constraints are applied to facial expression features, behavioral bias features, and lighting interaction features during training. This constraint, through adversarial training or causal intervention mechanisms, forces the model to strip away components highly correlated with content type labels from the aforementioned multimodal features, retaining only residual feature representations that are independent of the media content narrative and truly reflect the user's internal emotional state.

[0108] The structured emotional state parameters include emotion type, emotion intensity, emotion duration parameter, and emotion confidence score. The emotion type represents the semantic category of the user's current dominant emotion, such as calm, pleasure, excitement, anxiety, sadness, anger, and confusion. The emotion intensity represents the activation level of the current emotion type, such as the gradient change between mild pleasure and strong excitement. The emotion duration parameter represents the stability of the current emotional state over time and its expected duration. The emotion confidence score represents the reliability of the inference results of the current emotion type and emotion intensity.

[0109] Please see Figure 5 The diagram shows the workflow of an emotion state inference model in one embodiment of this application.

[0110] like Figure 5 As shown, the behavioral semantic graph is input into a pre-trained emotion state inference model, and the emotion state inference model outputs structured emotion state parameters through hierarchical emotion inference, including the following steps S310 to S350.

[0111] In step S310, based on the behavioral semantic graph, user behavior goals and activity tendencies are identified to obtain an activity intent vector.

[0112] Specifically, the activity intention vector includes at least one intention probability among work, entertainment, rest, sleep, social, and care.

[0113] In step S320, intensity feature values ​​are extracted based on the fused feature vector and mapped to the emotion intensity.

[0114] Specifically, the intensity feature value can be a dimensional component related to physiological arousal, movement amplitude, interaction frequency, and environmental response sensitivity. After nonlinear transformation and normalization, a scalarized emotion intensity value is generated.

[0115] In step S330, based on the activity intent vector, the fusion feature vector, and the content interference masking matrix, the user's emotion type probability distribution is inferred through conditional probability calculation, and the emotion type is determined according to the emotion type probability distribution.

[0116] Specifically, the probability distribution of the emotion type can be estimated using the following conditional probability formula:

[0117] P(e|F_t, I_t, M_t);

[0118] Where F_t represents the fused feature vector, I_t represents the activity intention vector, M_t represents the content interference masking matrix, and e represents the candidate emotion type label, whose value traverses all discrete categories in the predefined emotion ontology set, such as calm, joy, excitement, anxiety, sadness, anger, confusion, etc.

[0119] When determining the emotion type, the candidate emotion type label that maximizes the conditional probability can be selected as the deterministic emotion type output for the current time window.

[0120] In step S340, based on the dynamic change characteristics of the emotion type and the emotion intensity in the time domain, the stability of the user's emotional state is evaluated to generate the emotion persistence parameter and the emotion confidence level.

[0121] Please see Figure 6 The above is a flowchart of a user's emotional state stability assessment in one embodiment of this application.

[0122] like Figure 6 As shown, based on the dynamic change characteristics of the emotion type and the emotion intensity in the time domain, the stability assessment of the user's emotional state is performed to generate the emotion persistence parameter and the emotion confidence level, including the following steps S341 to S344.

[0123] In step S341, the time change rate of the probability distribution of the emotion type and the time change rate of the emotion intensity are calculated between adjacent time windows.

[0124] In step S342, based on the user's historical emotion data, the historical emotion state transition matrix is ​​obtained by statistically learning the transition patterns between emotion states.

[0125] In step S343, the duration of the behavior deviation feature is obtained by statistically analyzing the duration of continuous deviation from the user's personal behavior baseline in the behavior deviation feature vector.

[0126] In step S344, the emotion persistence parameter and the emotion confidence level are calculated based on the time change rate of the emotion type probability distribution, the time change rate of the emotion intensity, the historical emotion state transition matrix, and the duration of the behavioral offset feature.

[0127] In step S350, the emotion type, the emotion intensity, the emotion persistence parameter, and the emotion confidence level are combined into the structured emotion state parameter.

[0128] The structured emotional state parameters are represented as follows:

[0129] E_t={P(e), E_t, D_t, C_t},

[0130] Where E_t is the structured emotional state parameter, P(e) represents the probability distribution of the emotion type, E_t represents the emotion intensity, D_t represents the emotion persistence parameter, and C_t represents the emotion confidence level.

[0131] It should be noted that the emotion state inference model does not directly map the original behavioral features to emotion states. Instead, it employs a phased, hierarchical inference strategy: first, it outputs an activity intent vector; then, using this vector as a conditional prior, it combines a behavioral offset feature vector with a content interference masking matrix to jointly infer the user's emotion type probability distribution. The reason for this design is that user behavior representations are significantly context-dependent; the same behavioral pattern may correspond to drastically different emotional semantics under different activity intents. If intent recognition is skipped and emotion classification is performed directly, the model is highly susceptible to misjudging content-induced behavior as genuine negative emotions. Through pre-emptive activity intent recognition, the system establishes a reliable contextual anchor for subsequent emotion inference; the content interference masking matrix, based on this anchor, further strips away pseudo-emotional signals strongly correlated with the current media content, ensuring that the output probability distribution truly reflects the user's internal emotional state after detaching from content narrative interference.

[0132] In step S400, target lighting parameters that match the user's emotional state and the surrounding lighting environment are generated based on the structured emotional state parameters.

[0133] Please see Figure 7 The above is a flowchart illustrating the generation of target illumination parameters in one embodiment of this application.

[0134] like Figure 7 As shown, generating target lighting parameters that match the user's emotional state and the surrounding lighting environment based on the structured emotional state parameters includes the following steps S401 to S405.

[0135] In step S401, the user identity confidence index, space occupancy weight, and device control weight of each user in the multi-user home lighting scenario are obtained.

[0136] Specifically, the user identity confidence index can be generated based on at least one of the following: terminal login information, lighting control entry trigger records, device control ownership, space occupancy sensor information, and historical behavior patterns. The space occupancy weight is calculated based on the user's dwell time in a preset space area, activity intensity, or spatial location importance; the device control weight is determined based on user role permissions, historical control frequency, or preset priority rules.

[0137] In step S402, the difference between the structured emotional state parameters of any two users is calculated, and it is determined whether the difference exceeds a preset threshold.

[0138] In step S403, if the difference between the structured emotional state parameters of at least two users exceeds a preset threshold, it is determined that there is a multi-user emotional state conflict.

[0139] In step S404, based on the user identity confidence index, the space occupancy weight, and the device control weight, the structured emotional state parameters of each user with conflicting emotional states are weighted and aggregated to generate aggregated emotional state parameters.

[0140] If the degree of difference does not exceed a preset threshold, then it is determined that there is no emotional conflict.

[0141] In step S405, target lighting parameters matching the multi-user home lighting scene are generated based on the aggregated emotional state parameters.

[0142] For example, in a multi-user family living room scenario, User A is in an audio-visual entertainment state and is controlling the main light, exhibiting an emotional state of "immersion / mild excitement," while User B is in a reading area subtly adjusting the brightness of a floor lamp, exhibiting an emotional state of "focus / calm." The emotional conflict between the two users significantly exceeds a high-order threshold, which the system classifies as a high-level antagonistic conflict. At this point, the system can automatically switch to a neutral lighting mode: such as an output color temperature of 3500–4000K, illuminance of 200–250 lux, and basic lighting parameters with no dynamic changes. In neutral lighting mode, neither user is favored. Alternatively, the current lighting state can be maintained until the conflict subsides or the user explicitly intervenes. This strategy prevents discomfort caused by forcibly merging opposing emotions, demonstrating the system's restraint and safety in socially sensitive situations.

[0143] In this implementation, weighted aggregation is not a simple numerical average, but a multi-dimensional decision-making process that embeds spatial topology, device permissions, and behavioral intentions. This enables lighting adjustment in multi-user coexistence scenarios, significantly improving the usability and emotional inclusivity of the smart lighting system in real-life living environments.

[0144] In one embodiment of this application, the illumination parameter generation method based on behavioral semantic analysis further includes: if the emotion confidence is lower than a preset confidence threshold, or the modal conflict level index is higher than a preset conflict threshold, then a conservative output strategy is executed; the conservative output strategy includes at least one of outputting zone lighting suggestions, outputting local lighting suggestions, maintaining the current lighting state suggestions, or triggering user confirmation feedback.

[0145] In uncertain scenarios with low confidence or conflicting information from multiple sources, the system can ensure that the mood lighting system will not produce abrupt, highly saturated, highly dynamic or unsuitable light environments in uncertain scenarios through a conservative output strategy, thereby improving the stability and safety of long-term home use.

[0146] It should be noted that the scope of protection of the illumination parameter generation method based on behavioral semantic analysis described in this application is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principles of this application is included within the scope of protection of this application.

[0147] Please see Figure 8 The image shown is a schematic diagram of the structure of a lighting system according to an embodiment of this application.

[0148] like Figure 8 As shown, this application provides a lighting system, including:

[0149] Lighting equipment;

[0150] The illumination parameter generation module is used to execute the illumination parameter generation method based on behavioral semantic analysis as described in any of the above items to generate target illumination parameters;

[0151] The lighting control module is connected to the illumination parameter generation module and the lighting device, and is used to generate lighting control commands based on the target illumination parameters, and control the lighting device to perform lighting adjustment based on the lighting control commands.

[0152] It should be noted that the structure and principle of the illumination parameter generation module described in this embodiment correspond one-to-one with the steps in the illumination parameter generation method based on behavioral semantic analysis described above, so they will not be repeated here.

[0153] The present application will be further described below with reference to embodiments. It should be understood that the following embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit the scope of protection of the present application. All equivalent substitutions, combinations, modifications or improvements made under the technical concept of the present application should be included within the scope of protection of the present application.

[0154] In a typical residential living room scenario, users control multiple smart lights through various entry points such as smart panels, mobile apps, and voice assistants. Within a short window of the last 5 minutes, the system detected that the user increased the brightness three times, decreased the color temperature twice, and eventually reverted to the previously preset "relaxation" scene. Meanwhile, in the long-term behavioral baseline of the last 7 days, the user typically directly activated the fixed movie-watching scene during the same time period and rarely made any manual interventions.

[0155] The system converts the aforementioned original lighting interaction events into a unified sequence of event tuples, and calculates a high intervention disturbance amount and scene hesitation index accordingly. Since the current media content type tag indicates that no strongly emotionally resonant video or game content is playing, the content interference shielding coefficient is at a low level, suggesting that this manual intervention behavior is unlikely to be induced by external content, and its behavioral semantics have a high inherent emotional orientation. Furthermore, behavioral semantic graph analysis shows a high-weight connection relationship between the user node, living room space node, main light device node, and scene rollback event node, corroborating that this series of operations has clear contextual consistency and intentional coherence.

[0156] Based on this, the emotion state inference model first outputs an activity intention vector, determining that the probability of the current dominant activity intention being the "rest / relaxation" category is significantly higher than other candidate intentions. Subsequently, combining this intention prior with the aforementioned high intervention perturbation and low content interference masking coefficient, it infers that the user still exhibits a mild irritability or fatigue emotional shift even under the "rest / relaxation" intention. The final output structured emotion state parameters are: emotion type is fatigue or mild irritability, emotion intensity is low to medium, emotion duration is moderate, and confidence level is higher than the preset confidence threshold.

[0157] Finally, target lighting parameters are generated based on structured emotional state parameters: the overall color temperature is lowered to the 2700–3000K warm white light range, the illuminance is maintained at a gentle level of 150–200 lux, the dynamic change rate is reduced to the lowest level to avoid visual stimulation, and LEDs directly shining into the user's eyes are automatically turned off or anti-glare shielding strategies are activated. This lighting scheme respects the user's explicit intention of "rest / relaxation" while specifically alleviating their implicit irritability or fatigue, achieving a closed-loop response from the perception of behavioral disturbances to emotionally appropriate lighting.

[0158] Please see Figure 9 The image shown is a schematic diagram of the structure of a terminal according to an embodiment of this application.

[0159] like Figure 9 As shown, this application embodiment provides a terminal, including:

[0160] The memory is used to store computer programs;

[0161] A processor, the processor being configured to execute a computer program stored in the memory, so as to cause the terminal to perform any of the methods described above.

[0162] Preferably, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0163] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or modules or units may be electrical, mechanical, or other forms.

[0164] The modules / units described as separate components may or may not be physically separate. The components shown as modules / units may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules / units can be selected to achieve the objectives of the embodiments of this application, depending on actual needs. For example, the functional modules / units in the various embodiments of this application may be integrated into one processing module, or each module / unit may exist physically separately, or two or more modules / units may be integrated into one module / unit.

[0165] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0166] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0167] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for generating illumination parameters based on behavioral semantic analysis, characterized in that, include: Obtain multimodal behavioral cues of users in the lighting environment; The multimodal behavioral cues include lighting interaction events and their corresponding contextual metadata, which include lighting temporal rhythm events, content type tag events with preset non-sensitive standards, and lighting device spatial topology events. Based on the multimodal behavioral cues, a behavioral semantic graph is constructed; the behavioral semantic graph describes the semantic associations between the multimodal behavioral cues through multi-type nodes and multi-type relation edges, and the strength of different associations is represented by the relation edge weights; The behavioral semantic graph is input into a pre-trained emotion state inference model, which infers emotions through hierarchical emotion inference and outputs structured emotion state parameters. The structured emotion state parameters include emotion type, emotion intensity, emotion persistence parameters, and emotion confidence. Based on the structured emotional state parameters, target lighting parameters that match the user's emotional state and the surrounding lighting environment are generated. The construction of the behavioral semantic graph based on the multimodal behavioral cues includes: The multimodal behavioral cues are converted into a sequence of event tuples with a unified data structure; Based on users' historical behavior data, construct a baseline for individual user behavior; The event tuple sequence and the user's personal behavior baseline are differentially calculated at multiple time scales to obtain a behavior offset feature vector that characterizes the degree of deviation of the user's behavior from its historical habits. Based on preset reliability evaluation indicators, the reliability of the multimodal behavioral cues is assessed to obtain a modal reliability vector; The temporal correlation between the content type tag event and the lighting interaction event is obtained, and a content interference masking matrix is ​​generated based on the temporal correlation. The content interference masking matrix is ​​used to reduce the misleading influence of irrelevant or interfering content on the inference of the user's true emotional state. Based on the modal reliability vector and the content interference shielding matrix, the behavior offset feature vector is subjected to gated fusion processing to obtain a fused feature vector. Based on the fused feature vector, the nodes and relation edges used to construct the behavioral semantic graph are determined; The behavioral semantic graph is input into a pre-trained emotion state inference model, which then performs hierarchical emotion inference and outputs structured emotion state parameters, including: Based on the aforementioned behavioral semantic graph, user behavioral goals and activity tendencies are identified to obtain an activity intent vector; Based on the fused feature vector, intensity feature values ​​are extracted and mapped to the emotion intensity; Based on the activity intent vector, the fusion feature vector, and the content interference masking matrix, the user's emotion type probability distribution is inferred through conditional probability calculation, and the emotion type is determined according to the emotion type probability distribution. Based on the dynamic change characteristics of the emotion type and the emotion intensity in the time domain, the stability of the user's emotional state is evaluated to generate the emotion persistence parameter and the emotion confidence level. The emotion type, the emotion intensity, the emotion persistence parameter, and the emotion confidence level are combined to form the structured emotion state parameter; Based on the structured emotional state parameters, generating target lighting parameters that match the user's emotional state and the surrounding lighting environment includes: The system acquires the user identity confidence index, space occupancy weight, and device control weight for each user in a multi-user home lighting scenario. The space occupancy weight is calculated based on the user's dwell time, activity intensity, or spatial location importance in a preset space area. The device control weight is determined based on user role permissions, historical control frequency, or preset priority rules. Calculate the difference between the structured emotional state parameters of any two users, and determine whether the difference exceeds a preset threshold; If the difference between the structured emotional state parameters of at least two users exceeds a preset threshold, it is determined that there is a multi-user emotional state conflict. Based on the user identity confidence index, the space occupancy weight, and the device control weight, the structured emotional state parameters of each user with conflicting emotional states are weighted and aggregated to generate aggregated emotional state parameters. Based on the aggregated emotional state parameters, target lighting parameters that match the multi-user home lighting scene are generated.

2. The method according to claim 1, characterized in that, The event tuple sequence is differentially calculated with the user's personal behavior baseline across multiple time scales to obtain a behavior offset feature vector characterizing the degree of deviation of user behavior from its historical habits, including: Establish a multi-scale time window; the multi-scale time window includes a short-time window, a medium-time window and a long-time window, wherein the short-time window is used to characterize the user's immediate emotional disturbance, the medium-time window is used to characterize the user's daily state trend, and the long-time window is used to characterize the user's long-term habit baseline; Calculate the standardized difference between the statistical observations of the event tuple sequence within each time window and the user's personal behavior baseline; The standardized difference values ​​corresponding to each time window at each scale are combined to obtain the behavior offset feature vector.

3. The method according to claim 1, characterized in that, Based on preset reliability evaluation indicators, the reliability of the multimodal behavioral cues is assessed to obtain a modal reliability vector, including: Several reliability evaluation metrics are defined, including data integrity, time freshness, data stability, user identity confidence, and modal conflict level. Data integrity is defined as the ratio of the number of valid behavioral cues to the theoretical number of behavioral cues within a preset time window. Time freshness is defined as the most recent sampling time of a behavioral cue. Data stability is defined as the statistical variance of behavioral cues within a preset time window. User identity confidence is determined based on user identification results or device binding information. Modal conflict level is defined as the consistency score between any modal behavioral cue and other modal behavioral cues in terms of temporal characteristics, intensity characteristics, or pattern characteristics. Traverse each modal behavior clue and obtain the data integrity index, the time freshness index, the data stability index, the user identity confidence index, and the modal conflict degree index corresponding to each modal behavior clue; The data integrity index, the time freshness index, the data stability index, the user identity confidence index, and the modal conflict degree index are normalized respectively, and then weighted and summed according to preset weight coefficients to obtain the modal reliability corresponding to each modal behavior clue. The modal reliability vector is obtained by combining the modal reliability corresponding to all modal behavior cues in modal order.

4. The method according to claim 1, characterized in that, Based on the dynamic changes of the emotion type and the emotion intensity in the time domain, a stability assessment of the user's emotional state is performed to generate the emotion persistence parameter and the emotion confidence level, including: Calculate the rate of change of the probability distribution of the emotion type and the rate of change of the emotion intensity between adjacent time windows; Based on users' historical emotion data, the transition patterns between emotion states are statistically learned to obtain a historical emotion state transition matrix; The duration of the behavior deviation feature is obtained by statistically analyzing the duration of continuous deviations from the user's personal behavior baseline in the behavior deviation feature vector. Based on the time rate of change of the probability distribution of the emotion type, the time rate of change of the emotion intensity, the historical emotion state transition matrix, and the duration of the behavioral shift feature, the emotion persistence parameter and the emotion confidence level are calculated.

5. The method according to claim 3, characterized in that, Also includes: If the emotional confidence level is lower than the preset confidence threshold, or the modal conflict level index is higher than the preset conflict threshold, then a conservative output strategy is executed. The conservative output strategy includes at least one of the following: outputting zone lighting suggestions, outputting local lighting suggestions, maintaining the current lighting state, or triggering user confirmation feedback.

6. A lighting system, characterized in that, include: Lighting equipment; The illumination parameter generation module is used to execute the illumination parameter generation method based on behavioral semantic analysis as described in any one of claims 1 to 5, so as to generate target illumination parameters; The lighting control module is connected to the illumination parameter generation module and the lighting device, and is used to generate lighting control commands based on the target illumination parameters, and control the lighting device to perform lighting adjustment based on the lighting control commands.

7. A terminal, characterized in that, include: The memory is used to store computer programs; A processor for executing a computer program stored in the memory to cause the terminal to perform the method of any one of claims 1 to 5.

Images