Method and device for controlling a car fragrance

Abstract

The application provides a control method and device for vehicle-mounted fragrance, and relates to the technical field of vehicle control. The method comprises the following steps: obtaining a scene label of a destination, wherein the scene label of the destination is from a preset N standardized label; predicting, according to the scene label of the destination and a current state of a vehicle, a plurality of future scene labels corresponding to each position of the vehicle within a future time length through a scene recognition model; in the case that a target scene label exists in the plurality of future scene labels, determining a current remaining distance between the vehicle and a target scene corresponding to the target scene label, and determining a release starting point distance of the vehicle-mounted fragrance according to the target scene label, wherein the target scene label is a scene label of a known scene in the N standardized labels; if the current remaining distance is not greater than the release starting point distance, determining a release parameter of the vehicle-mounted fragrance according to the target scene label, and controlling the vehicle-mounted fragrance to work according to the release parameter.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, and in particular to a method and device for controlling in-vehicle fragrance. Background Technology

[0002] With the iterative upgrades of smart cockpit technology, in-car fragrance systems have become a core component for enhancing driving comfort and creating a personalized cabin atmosphere. They can adjust the cabin odor environment according to changes in the driving scenario, thereby alleviating driver fatigue and improving the overall passenger experience. However, most existing in-car fragrance devices still rely on manual operation, such as the driver manually selecting the type of fragrance or adjusting the release intensity. While a few control methods can achieve scene linkage with simple signals like vehicle speed and music type, they fail to deeply integrate with in-vehicle map data. This results in fragrance control being unable to automatically adapt to map-related information such as destination attributes, route scenarios, and real-time road conditions during navigation. This independent control mode not only makes the in-car fragrance device less intelligent and unable to fully realize its comfort value, but also increases the driver's manual operation burden during driving, easily distracting driving attention and affecting driving safety. Summary of the Invention

[0003] This invention provides a method and apparatus for controlling in-vehicle fragrance devices, which solves the technical problem in related technologies where in-vehicle fragrance devices are controlled independently by the driver, resulting in a lack of integration with vehicle map data and thus a low level of intelligence in fragrance release.

[0004] In a first aspect, embodiments of the present invention provide a method for controlling a vehicle-mounted fragrance, applied to a vehicle equipped with a vehicle-mounted fragrance, the method comprising: Obtain the scene tags of the destination, which are derived from N preset standardized tags; Based on the scene label of the destination and the current state of the vehicle, a scene recognition model is used to predict multiple future scene labels corresponding to the various locations of the vehicle in the future within a certain time period. If a target scene label exists among the multiple future scene labels, determine the current remaining distance between the vehicle and the target scene corresponding to the target scene label, and determine the scene label of a known scene among the N standardized labels that is the distance from the target scene label to the release starting point of the in-vehicle fragrance. If the current remaining distance is not greater than the release starting distance, the release parameters of the car fragrance are determined according to the target scene label, and the car fragrance is controlled to work according to the release parameters.

[0005] Preferably, obtaining the scene tag of the destination includes: obtaining destination information, the destination information including the name of the destination, the navigation code of the destination, and the address information of the destination; searching for whether there is a standardized tag corresponding to the navigation code of the destination in N standardized tags in the navigation database; if there is, then using the standardized tag corresponding to the navigation code of the destination as the scene tag of the destination; if there is no, then matching the name of the destination with N regular expressions corresponding to the N standardized tags respectively; if the match is successful, then using the standardized tag corresponding to the successfully matched regular expression as the scene tag of the destination; if the match is unsuccessful, then analyzing the address information of the destination to obtain the scene tag of the destination.

[0006] Preferably, the step of predicting multiple future scene labels corresponding to the vehicle's various locations within a future time period using a scene recognition model based on the destination scene label and the vehicle's current state includes: collecting the vehicle's current state and the environmental state of the vehicle's current environment according to a preset sampling step size, and combining them with the destination scene label to form a temporal feature vector corresponding to each sampling step size; after forming P temporal feature vectors, inputting the P temporal feature vectors into the scene recognition model; and predicting the future scene labels corresponding to the vehicle's location at the times corresponding to Q sampling steps in the future using the scene recognition model to obtain Q future scene labels, where the future time period is the total duration of the Q sampling steps.

[0007] Preferably, before determining the current remaining distance between the vehicle and the target scene corresponding to the target scene label, the method further includes: if there is a scene label for any known scene among the Q future scene labels, determining the target scene closest to the current position of the vehicle from the scene labels of the known scenes, and using the standardized label corresponding to the target scene as the target scene label; if there is no scene label for any known scene among the Q future scene labels, then determining that there is no target scene label among the Q future scene labels.

[0008] Preferably, the method further includes: if there is no target scene label among the Q future scene labels, performing the step of collecting the current state of the vehicle and the current state of the environment in which the vehicle is located according to a preset sampling step size, and combining the scene label of the destination to form a temporal feature vector corresponding to each sampling step size.

[0009] Preferably, determining the release starting distance for activating the in-vehicle fragrance based on the target scene label includes: obtaining the basic release distance and scene correction coefficient corresponding to the target scene label; determining the release starting distance based on the basic release distance, the scene correction coefficient, and the current vehicle speed, wherein the release starting distance is the distance between the vehicle and the target scene corresponding to the target scene label when the in-vehicle fragrance is activated.

[0010] Preferably, determining the release parameters of the in-vehicle fragrance based on the target scene label includes: obtaining the current cabin environment status of the vehicle; and determining the release parameters corresponding to the target scene label based on the current cabin environment status of the vehicle using a fragrance release model, wherein the release parameters include release concentration and / or release type.

[0011] Preferably, after controlling the in-vehicle fragrance to operate according to the release parameters, the method further includes: collecting user feedback; uploading the user feedback and the vehicle status during the operation of the in-vehicle fragrance to the cloud, so that the cloud updates the fragrance release model based on the user feedback and the vehicle status during the operation of the in-vehicle fragrance.

[0012] Secondly, embodiments of the present invention provide a control device for a car fragrance system, the device comprising: The acquisition module is used to acquire the scene tags of the destination, wherein the scene tags of the destination are derived from N preset standardized tags; The prediction module is used to predict multiple future scene labels corresponding to the various locations of the vehicle within a future time period based on the scene labels of the destination and the current state of the vehicle, using a scene recognition model. The determination module is used to determine the current remaining distance between the vehicle and the target scene corresponding to the target scene label when a target scene label exists among the multiple future scene labels, and to determine the release starting distance of the in-vehicle fragrance based on the target scene label, wherein the target scene label is the scene label of a known scene among the N standardized labels; The determining module is further configured to, if the current remaining distance is not greater than the release starting point distance, determine the release parameters of the in-vehicle fragrance based on the target scene label, and control the in-vehicle fragrance to operate according to the release parameters.

[0013] Thirdly, embodiments of the present invention provide a vehicle, including: processor; A memory for storing instructions to be executed by the processor, wherein the processor is configured to execute the instructions to implement the method as described in the first aspect.

[0014] Fourthly, embodiments of the present invention provide a storage medium that, when instructions in the storage medium are executed by a processor of an electronic device, enables the vehicle to perform the method as described in the first aspect.

[0015] Fifthly, embodiments of the present invention provide a computer program product, the program product comprising a computer program, the computer program being executed by a processor as described in the first aspect.

[0016] According to an embodiment of the present invention, a method for controlling a car fragrance is provided, applied to a vehicle. The vehicle is equipped with a car fragrance. The method acquires a destination scene tag, which is derived from N preset standardized tags. Based on the destination scene tag and the current state of the vehicle, a scene recognition model predicts multiple future scene tags corresponding to the vehicle's locations within a future timeframe. If a target scene tag exists among the multiple future scene tags, the method determines the current remaining distance between the vehicle and the target scene corresponding to the target scene tag, and determines the release starting distance for activating the car fragrance based on the target scene tag, where the target scene tag is a scene tag of a known scene from the N standardized tags. If the current remaining distance is not greater than the release starting distance, the method determines the release parameters of the car fragrance based on the target scene tag and controls the car fragrance to operate according to the release parameters. Using destination scene labels and the vehicle's current state as input, the system predicts future scene labels corresponding to various vehicle positions during the journey. When a target scene label corresponding to a known scene exists, based on the pre-bound release start distance, and when the current remaining distance between the vehicle and the target scene is no greater than the release start distance, the system determines the fragrance release parameters adapted to that target scene label. This achieves precise pre-coupling of fragrance release with future driving scenarios in the spatial dimension. Compared to existing technologies where fragrance devices are independently controlled by the driver and cannot be deeply integrated with vehicle map data, resulting in release behavior deviating from actual scenario requirements, this system uses map navigation data to drive fragrance decisions, enabling in-vehicle fragrances to predict and adjust according to changes in the journey scenario. This improves the scene adaptability of the cabin odor environment and significantly enhances the intelligence level of in-vehicle fragrance control. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1The flowchart of a method for controlling a car fragrance provided in some embodiments of the present invention is shown; Figure 2 The structure of a control device for a car fragrance provided in some embodiments of the present invention is shown; Figure 3 The structure of a vehicle fragrance control system provided in some embodiments of the present invention is shown; Figure 4 The diagram shows a structural block diagram of a vehicle provided in some embodiments of this application. Detailed Implementation

[0019] As described in the background section, with the iterative upgrades of smart cockpit technology, in-car fragrance systems have become one of the core components for enhancing driving comfort and creating a personalized cabin atmosphere. They can adjust the cabin odor environment according to changes in the driving scenario, thereby alleviating driver fatigue and improving the overall passenger experience. However, most existing in-car fragrance devices still rely on manual operation, such as the driver manually selecting the type of fragrance or adjusting the release intensity. While a few control methods can achieve scenario linkage with simple signals such as vehicle speed and music type, they fail to deeply integrate with in-vehicle map data. This results in fragrance control being unable to automatically adapt to map-related information such as destination attributes, route scenarios, and real-time road conditions during navigation. This independent control mode not only makes the in-car fragrance device less intelligent and unable to fully realize its comfort value, but also increases the driver's manual operation burden during driving, easily distracting driving attention and affecting driving safety.

[0020] This invention provides a method for controlling in-vehicle fragrance. Taking destination scene labels and the vehicle's current state as input, it predicts future scene labels corresponding to various vehicle locations during the journey. When a target scene label corresponding to a known scene exists, based on a pre-bound release start distance, and when the current remaining distance between the vehicle and the target scene is not greater than the release start distance, it determines fragrance release parameters adapted to the target scene label. This achieves precise pre-coupling of fragrance release with future driving scenarios in a spatial dimension. Compared to existing technologies where fragrance devices are independently controlled by the driver and cannot be deeply integrated with vehicle map data, resulting in release behavior deviating from actual scenario requirements, this method uses map navigation data to drive fragrance decisions, enabling in-vehicle fragrance to predict and adjust according to changes in the travel scenario. This improves the scene adaptability of the cabin odor environment and significantly enhances the intelligence level of in-vehicle fragrance control.

[0021] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0022] In this embodiment of the invention, the vehicle fragrance control method provided by this embodiment of the invention can be applied to a vehicle, wherein the vehicle is equipped with a vehicle fragrance and is controlled by the vehicle's electronic control unit (ECU).

[0023] In an embodiment of the present invention, Figure 1 The flowchart illustrates a method for controlling a car fragrance system according to some embodiments of the present invention. For example... Figure 1 As shown, the in-vehicle fragrance control method provided in this embodiment of the invention includes steps 110 to 140.

[0024] Step 110: Obtain the scene label of the destination, which comes from N preset standardized labels.

[0025] In this embodiment of the invention, the destination refers to the end point of the vehicle's current journey. The scene label is used to characterize the scene type to which the destination belongs. This label must be selected from a preset set of N standardized labels, where N is a positive integer greater than 1. The standardized labels are a set of standardized labels formed after standardizing various scenes, such as scenic spots, commercial areas, medical areas, office areas, residential areas, and congested road sections, according to unified naming rules, classification codes, and semantic constraints. Among the N standardized labels, there are (N-1) scene labels for known scenes, corresponding to specific scene types that can be predefined and identified. For example, the standardized label for a scenic spot scene could be "natural scenic spot," the standardized label for a commercial area scene could be "commercial district," the standardized label for a medical area scene could be "medical facility," the standardized label for an office area scene could be "office," the standardized label for a residential area scene could be "residence," and the standardized label for a congested road section scene could be "transportation hub." The N standardized labels also include a label for an unknown scene, "other," used as a fallback label when no known scene type can be matched, ensuring that a valid scene label is output under any input condition.

[0026] In this embodiment of the invention, the scene tag of the destination can be directly obtained through the classification of the destination by the navigation software. For example, when the destination is "xx park", the scene tag of the destination can be directly determined as "natural scenic spot" according to the classification in the navigation software.

[0027] Step 120: Based on the scene label of the destination and the current state of the vehicle, predict multiple future scene labels corresponding to the various locations of the vehicle in the future time period using a scene recognition model.

[0028] In this embodiment of the invention, the current state of the vehicle specifically includes information such as the current vehicle speed and the remaining distance between the vehicle's current location and its destination. The scene recognition model is a predictive model used to infer future state changes based on the current input state, such as a Long Short-Term Memory (LSTM) network model. The future scene label refers to the standardized label output by the scene recognition model corresponding to the scene type of the vehicle's location at a future time.

[0029] In this embodiment of the invention, the destination scene label and the vehicle's current state are input into the scene recognition model. The scene recognition model identifies the temporal feature correlation between the two and then generates scene type prediction results for various locations along the vehicle's future journey based on these features. Finally, it outputs multiple future scene labels, constructing a future scene label sequence to provide a scene-based basis for the subsequent early triggering and parameter adjustment of the fragrance system. Furthermore, the output of the scene recognition model can include not only multiple future scene labels but also the confidence score corresponding to each future scene label.

[0030] Step 130: If a target scene label exists among the multiple future scene labels, determine the current remaining distance between the vehicle and the target scene corresponding to the target scene label, and determine the release starting distance of the in-vehicle fragrance based on the target scene label. The target scene label is the scene label of a known scene among the N standardized labels.

[0031] In this embodiment of the invention, the target scene label can be any known scene label from N standardized labels. For example, if multiple future scene labels sequentially include "other natural scenic spots and other residential places", the target scene label can be "natural scenic spots" or "residential places". Each target scene label can be pre-bound with a corresponding release start distance. The release start distance can be the critical distance at which the in-vehicle fragrance should be activated when the vehicle's current location is still a certain distance from the physical area indicated by the target scene (i.e., the target scene). The purpose is to ensure that when the vehicle enters the scene area, the driver and passengers can promptly experience the appropriate fragrance effect. When the multiple future scene labels output in step 120 include a target scene label, the current remaining distance between the vehicle's current location and the target scene corresponding to the target scene label must first be determined. At the same time, the pre-bound release start distance is retrieved based on the target scene label. Then, by comparing the current remaining distance with the release start distance, it is determined whether the vehicle has entered the fragrance release preparation area of ​​the scene, so as to decide whether to perform fragrance activation and parameter adjustment operations.

[0032] Step 140: If the current remaining distance is not greater than the release starting point distance, then determine the release parameters of the car fragrance according to the target scene label, and control the car fragrance to work according to the release parameters.

[0033] In this embodiment of the invention, if the current remaining distance is not greater than the release starting distance, i.e., when the vehicle enters the target scene corresponding to the target scene label, the release parameters of the in-vehicle fragrance are determined according to the target scene label. The release parameters are a set of data defining the specific output state of the in-vehicle fragrance, including activation parameters indicating whether the fragrance is turned on, the type of fragrance release (i.e., the specific fragrance type), and the release concentration (i.e., the release intensity level). The release parameters can be determined by pre-binding each target scene label with its corresponding release parameters and obtaining them directly through a lookup table; or by using a neural network model to dynamically infer them based on the target scene labels. After determining the release parameters, a control command is sent to the in-vehicle fragrance actuator, causing it to start working according to the activation state, fragrance type, and concentration included in the release parameters, thereby automatically adjusting the cabin odor to match the scene before the vehicle arrives at the target scene.

[0034] In this embodiment of the invention, the destination scene label and the vehicle's current state are used as inputs to predict the future scene labels corresponding to each position of the vehicle during the journey. When a target scene label that meets the filtering conditions exists, based on the pre-bound release start distance, and when the current remaining distance between the vehicle and the target scene is not greater than the release start distance, the fragrance release parameters adapted to the target scene label are determined. This achieves precise pre-coupling of fragrance release with future driving scenarios in the spatial dimension. Compared to the shortcomings of existing technologies where fragrance devices are independently controlled by the driver and cannot be deeply integrated with vehicle map data, resulting in release behavior deviating from actual scenario requirements, this invention uses map navigation data to drive fragrance decisions, enabling in-vehicle fragrances to predict and adjust according to changes in the journey scenario. This not only improves the scenario adaptability of the cabin odor environment but also significantly enhances the intelligence level of in-vehicle fragrance control.

[0035] In this embodiment of the invention, to make the destination scene label obtained in step 110 more accurate, the method provided by this embodiment includes a three-layer matching principle based on regular expressions. Specifically, one implementation method for obtaining the destination scene label in step 110 may include: obtaining destination information, the destination information including the destination name, the destination navigation code, and the destination address information; searching for whether there is a standardized label corresponding to the navigation code of the destination among N standardized labels in the navigation database; if there is, using the standardized label corresponding to the navigation code of the destination as the scene label of the destination; if there is no, matching the destination name with the N regular expressions corresponding to the N standardized labels respectively; if the match is successful, using the standardized label corresponding to the successfully matched regular expression as the scene label of the destination; if the match is unsuccessful, analyzing the address information of the destination to obtain the scene label of the destination.

[0036] In this embodiment of the invention, destination information can be obtained from the vehicle navigation system. The destination information includes the destination name, the destination's navigation code, and the destination's address information. The destination name can be a Point of Interest (POI) name, and the destination's navigation code can be the category code corresponding to the POI name in the vehicle navigation system. After obtaining the destination information, data preprocessing can be performed first. The POI names are cleaned by removing special symbols, numerical suffixes, and branch identifiers. For example, if the POI name of the destination is "XX Park (South Gate)," after cleaning, it becomes "XX Park". Then, a first-level matching is performed, searching among N standardized tags in the navigation database for a standardized tag corresponding to the destination's navigation code (e.g., navigation code "110100" corresponds to the standardized tag "Natural Scenic Area"). If it exists, the standardized tag corresponding to the destination's navigation code is used as the destination's scene tag; if it does not exist, a second-level matching is performed, matching the destination name using N regular expressions corresponding to the N standardized tags.

[0037] In this embodiment of the invention, the regular expression corresponding to each standardized tag can be exemplified as follows: "Natural scenic area" = (park | forest park | wetland park | botanical garden | zoo | scenic area | scenic spot | mountain | lake | sea | island | canyon | waterfall | grassland | forest); "Medical facilities" = (hospitals, health centers, clinics, outpatient departments, maternal and child health care centers, children's hospitals, disease control centers, physical examination centers); "Commercial district" = shopping mall | shopping center | plaza | pedestrian street | supermarket | convenience store | restaurant | hotel | KTV | cinema | amusement park.

[0038] In this embodiment of the invention, if the point of interest name of the destination contains any word on the right side of the regular expression, the scene label of the destination can be determined as the standardized label on the left side of the expression. If any one of the N regular expressions matches successfully, the standardized label corresponding to the successfully matched regular expression is taken as the scene label of the destination; if the match fails, a three-level matching is performed, analyzing the address information of the destination and combining the administrative division and road type in the address information for auxiliary determination to obtain the scene label of the destination. During the matching process, when multiple keywords match different candidate scenes at the same time, a weighted vote can be performed based on the keyword weight of each matching item (such as the relevance of the keyword to the scene, the location of occurrence, specificity, etc.), and finally the scene label with the highest weight score is output. The confidence level of the current matching result is calculated based on the matching level (level 1, 2, or 3) and keyword matching degree (such as the number of matched keywords and the typicality of keywords). If the confidence level is not lower than 0.9, the label is directly adopted. If the confidence level is between 0.7 and 0.9, the label is corrected by combining user historical preference data (e.g., prioritizing frequently visited scene types). If the confidence level is lower than 0.7, it is marked as "other" scene to avoid unreliable mismatches. By constructing a hierarchical POI thesaurus and regular expression matching rules, millisecond-level parsing of POI names and category labels returned by the vehicle map is achieved, outputting standardized scene labels. In scenarios where the computing power of the vehicle terminal is limited, it has both high real-time performance and accuracy.

[0039] In this embodiment of the invention, the scene recognition model can be a Convolutional Neural Network-Long Short-Term Memory (CNN-LSM) model. Step 120, which involves predicting multiple future scene labels corresponding to the vehicle's various locations within a future timeframe based on the destination scene label and the vehicle's current state, may include: collecting the vehicle's current state and the environmental state of the vehicle's current environment according to a preset sampling step size, and combining this with the destination scene label to construct a temporal feature vector corresponding to each sampling step size; inputting the P temporal feature vectors into the scene recognition model after constructing P temporal feature vectors; and predicting the future scene labels corresponding to the vehicle's location at the times corresponding to Q sampling steps in the future, thereby obtaining Q future scene labels, where the future timeframe is the total duration of the Q sampling steps.

[0040] In this embodiment of the invention, the preset sampling step size can be any duration such as one minute or half a minute. The temporal feature vector corresponding to each sampling step size can include multi-dimensional features, and each feature can be processed by one-hot encoding to obtain the one-hot code corresponding to each feature. Specifically, each temporal feature vector includes the destination scene label, the vehicle's current state including the current road type (which can be the road type collected at the last collection moment in the sampling step, including highway / urban expressway / urban road / rural road), the vehicle's current state including the real-time congestion level (which can be the congestion level collected at the last collection moment in the sampling step, represented by 0-3, where 0 represents smooth traffic and 3 represents severe congestion), the vehicle's current state including the remaining distance (in km, which can be the remaining distance between the vehicle's current location and the destination collected at the last collection moment in the sampling step, and normalized), the vehicle's current state including the current speed (in km / h, which can be the vehicle's current speed collected at the last collection moment in the sampling step, and normalized), and the vehicle's current state including the estimated arrival time (in min, which can be the data collected at the last collection moment in the sampling step, and normalized). In addition, the temporal feature vector may also include the environmental state of the current environment of the vehicle, such as time period labels (e.g., morning peak / off-peak / evening peak / nighttime), weather type (e.g., sunny / rainy / snowy / foggy), and cabin state, such as in-vehicle temperature (in °C, and normalized), number of people in the vehicle (1-5, and normalized), air conditioning status (on / off), and music style label.

[0041] In this embodiment of the invention, after constructing P temporal feature vectors, the P temporal feature vectors are input into the scene recognition model, where P can be any positive integer, such as 5 or 10. The scene recognition model is run every five or ten minutes.

[0042] In this embodiment of the invention, the scene recognition model predicts the future scene label corresponding to the vehicle's location at each of the next Q sampling steps, resulting in Q future scene labels. The future duration is the total duration of the Q sampling steps, where Q can be any positive integer not less than P, such as 10 or 20. Correspondingly, the future duration can be 10 minutes or 20 minutes. After inputting P temporal feature vectors into the scene recognition model, spatial features are extracted by a convolutional neural network (CNN) layer (3×3 kernel size, 64 output channels, ReLU activation function) of the scene recognition model class, and then dimensionality is reduced by a pooling layer (2×2 pooling size) to obtain a feature sequence. The feature sequence is then input into a bidirectional Long Short-Term Memory (LSTM) network layer (128 hidden units) to capture temporal dependencies. This is followed by a dropout layer (0.2 dropout rate) to prevent overfitting. A fully connected layer (64 units, ReLU activation function) further maps the sequence. Finally, the output layer outputs the probability distribution of the scene category for each of the next Q sampling steps (e.g., 10 minutes in the future, 1 minute per step), along with the confidence score for each predicted label. The model uses cross-entropy as the loss function. To meet the requirements for vehicle-mounted deployment, the scene recognition model is compressed to 12MB using INT8 quantization, with a single-step inference latency of no more than 15ms. This ensures real-time output of future scene label sequences and their confidence scores while maintaining prediction accuracy.

[0043] In this embodiment of the invention, before determining the current remaining distance between the vehicle and the target scene corresponding to the target scene label in step 130, the method further includes: if there is a scene label for any known scene among the Q future scene labels, determining the target scene closest to the current position of the vehicle from the scene labels of the known scenes, and using the standardized label corresponding to the target scene as the target scene label; if there is no scene label for any known scene among the Q future scene labels, then determining that there is no target scene label among the Q future scene labels.

[0044] In this embodiment of the invention, in order to make the control of the in-vehicle fragrance more matched to the journey, if there is a scene label of any known scene among the Q future scene labels, the target scene closest to the current position of the vehicle can be determined from the scene labels of the known scene, and the standardized label corresponding to the target scene is used as the target scene label, so that the vehicle releases different fragrances in sequence according to the order of passing through the scenes, making the release of the in-vehicle fragrance more intelligent.

[0045] In this embodiment of the invention, the specified scene label can be a standardized label with an actual scene type, that is, a scene label other than "other" among N standardized labels. If no target scene label exists among the Q future scene labels, the following steps are performed: collecting the current state of the vehicle and the current state of the environment in which the vehicle is located according to a preset sampling step size, and combining this with the scene label of the destination to construct a temporal feature vector corresponding to each sampling step size; and periodically inputting P temporal feature vectors corresponding to each sampling step size into the scene recognition model.

[0046] In this embodiment of the invention, when it is determined that a target scene label exists among Q future scene labels, step 130, which involves determining the release starting distance for activating the in-vehicle fragrance based on the target scene label, includes: obtaining the basic release distance and scene correction coefficient corresponding to the target scene label; determining the release starting distance based on the basic release distance, the scene correction coefficient, and the current vehicle speed, wherein the release starting distance is the distance between the vehicle and the target scene corresponding to the target scene label when the in-vehicle fragrance is activated.

[0047] In this embodiment of the invention, the release starting point distance of the car fragrance can be determined by the following formula: D = D0 + k * v + c Where: D is the release starting point distance (i.e. how many kilometers away from the destination when the fragrance starts to be released, in km), D0 is the base release distance corresponding to the target scene label, k is the vehicle speed correction coefficient, v is the current average vehicle speed (km / h), and c is the scene correction coefficient.

[0048] In this embodiment of the invention, Table 1 below shows the correspondence between the parameters in the formula for calculating the scene label and the starting point of the release distance.

[0049] Table 1. Correspondence between scene labels and parameters Scene tags <![CDATA[Basic release distance D0 (km)]]> Vehicle speed correction factor k Scene correction coefficient c Natural scenic area 1.0 0.005 0.2 Business district 0.5 0.003 0.1 medical facilities 0.3 0.002 0.0 Office space 0.8 0.004 0.15 place of residence 0.6 0.003 0.1 Transportation hub 0.2 0.001 -0.1 In this embodiment of the invention, the release starting distance can be corrected in multiple dimensions based on real-time dynamic factors. For example, the current average vehicle speed is recalculated every minute, and the release starting distance is updated accordingly. If the road congestion level ahead reaches level 2 or above, the release starting distance is increased by 0.3 kilometers; if the air conditioning fan speed is greater than level 3, the release starting distance is increased by 0.2 kilometers. At the same time, the basic release distance is personalized based on the user's historical adjustment records, with an adjustment range of ±0.2 kilometers. Through the above comprehensive correction mechanism, it can dynamically adapt to actual driving conditions and individual user differences, ensuring the accuracy and adaptability of the fragrance release timing. A release starting distance algorithm based on linear regression is adopted. By analyzing the user's perception of fragrance concentration in different scenarios, a mathematical model is established between the remaining distance, current vehicle speed, and the optimal release starting point, thereby theoretically determining the ideal release starting point so that the fragrance concentration in the cabin reaches the optimal level when the vehicle arrives at the target scenario.

[0050] In this embodiment of the invention, step 140, which involves determining the release parameters of the in-vehicle fragrance based on the target scene label, includes: obtaining the current cabin environment state of the vehicle; and determining the release parameters corresponding to the target scene label based on the current cabin environment state of the vehicle using a fragrance release model, wherein the release parameters include the release concentration.

[0051] In this embodiment of the invention, the current cabin environment state of the vehicle may include multiple continuous variables such as in-vehicle temperature, in-vehicle humidity, number of occupants, current vehicle speed, air conditioning fan speed, and current fragrance release duration, and each variable is normalized. Then, the current cabin environment state of the vehicle is used as input to the fragrance release model, which determines the release parameters corresponding to the target scene label. The fragrance release model can be a Deep Q-Network (DQN) algorithm, modeling the fragrance release process as a Markov Decision Process (MDP). The cabin environment state parameters constitute the state space S, and the fragrance release concentration constitutes the discrete action space A (including five intensity levels: off, ultra-low 20%, low 40%, medium 60%, and high 80%). A reward function R is constructed using user comfort feedback. The reward function R can be expressed as follows: R=w1*R_comfort+w2*R_feedback+w3*R_safety Where R is the reward function; R_comfort is the environmental comfort reward, calculated based on the human comfort range of temperature and humidity, with higher rewards for being closer to the comfort range; R_feedback is the user explicit feedback reward, with negative rewards for users manually adjusting the fragrance and positive rewards for users not adjusting it; R_safety is the safety reward, which avoids excessively strong fragrances causing driver discomfort when the vehicle speed is >60km / h; w1, w2, and w3 are weighting coefficients, with w1=0.5, w2=0.3, and w3=0.2.

[0052] In this embodiment of the invention, the DQN algorithm can adopt a dual DQN structure, including a current Q-network and a target Q-network. Both DQN networks are 3-layer fully connected networks (64 units in the input layer, 128 units in the hidden layer, and 5 units in the output layer). An experience replay pool with a capacity of 10,000 is set up, and 64 samples are randomly sampled from the replay pool for training each time. The target Q-network adopts a soft update method with an update coefficient τ=0.001. The action selection from the action space follows an ε-greedy strategy (initial ε=1.0, decaying by 0.001 every 1000 training steps, and finally stabilizing at 0.1). The fragrance release model can output the Q value corresponding to each action at each decision time according to the current cabin environment state, and select the action with the largest Q value as the current optimal release intensity. This, together with the fragrance type corresponding to the target scene label, constitutes a complete release parameter, realizing personalized dynamic fragrance adjustment.

[0053] After controlling the in-vehicle fragrance to operate according to the release parameters in step 140, the method further includes: collecting user feedback; uploading the user feedback and the vehicle status during the operation of the in-vehicle fragrance to the cloud, so that the cloud updates the fragrance release model based on the user feedback and the vehicle status during the operation of the in-vehicle fragrance.

[0054] In this embodiment of the invention, during vehicle operation, cabin status data can be collected periodically, for example every 30 seconds, and an action selection can be performed. When the user manually adjusts the fragrance (increases, decreases, or turns it off), the current experience is immediately recorded as user feedback, triggering an online model update. Simultaneously, after the vehicle is turned off each day, the user feedback and the vehicle status during the fragrance's operation are uploaded to the cloud. The cloud then performs a global model update and distributes the updated model to the vehicle's infotainment system via Over-The-Air (OTA) technology, achieving continuous optimization of the overall user experience.

[0055] Specifically, after the in-car fragrance system completes its action output, it binds the current state vector S, the executed action A, the user feedback label (positive / negative), and the corresponding scene label ID, and writes them to the local data log in real time. The user feedback label definition includes positive feedback if the user does not manually operate, and negative feedback if the user manually increases / decreases / turns off the fragrance. For a single scene label (such as a scenic area or a congested road), all historical actions and feedback under that label are statistically analyzed. If the negative feedback ratio for that scene is >20%, the base fragrance intensity and target concentration bound to that scene are automatically adjusted. If similar feedback occurs in multiple scenes (such as generally lowering the concentration in a high-temperature scene), the default fragrance configuration table corresponding to the scene label is updated uniformly. The cloud aggregates the full feedback dataset daily as incremental training samples, which are input into the matching process of obtaining the scene label for the destination in step 110 to optimize keyword weights and conflict judgment rules, and correct scene labels corresponding to easily confused POIs. Incremental training samples can also be input into the navigation scene recognition model to adjust the weights of temporal features, correct the influence coefficients of related features such as congestion, temperature, and people, and optimize the scene prediction results for future durations, thereby improving scene matching accuracy from the source. Furthermore, the cloud can aggregate feedback from multiple vehicle users, complete global training of the DQN model, output a new fragrance control strategy, update the mapping relationship between scene labels and release parameters, and update scene coefficients such as D0, k, and c in the linear regression release timing algorithm. After the update is completed, the updated algorithm is remotely deployed to the vehicle's infotainment system via OTA, completing end-to-end rule iteration and achieving continuous optimization of scene recognition accuracy, in-vehicle fragrance release parameters, and release timing.

[0056] Figure 2 The diagram shows a structural diagram of a control device for a car fragrance system provided in some embodiments of the present invention, such as... Figure 2 As shown, the vehicle fragrance control device provided in this embodiment of the invention includes: The acquisition module 210 is used to acquire the scene tags of the destination, wherein the scene tags of the destination are derived from a preset set of N standardized tags; The prediction module 220 is used to predict multiple future scene labels corresponding to the various locations of the vehicle within a future time period based on the scene labels of the destination and the current state of the vehicle through a scene recognition model. The determination module 230 is used to determine the current remaining distance between the vehicle and the target scene corresponding to the target scene label when a target scene label exists among the multiple future scene labels, and to determine the release starting distance of the in-vehicle fragrance based on the target scene label, wherein the target scene label is the scene label of a known scene among the N standardized labels; The determining module 230 is further configured to, if the current remaining distance is not greater than the release starting distance, determine the release parameters of the in-vehicle fragrance according to the target scene label, and control the in-vehicle fragrance to work according to the release parameters.

[0057] It should be noted that the embodiments of the vehicle fragrance control device in this specification and the embodiments of the vehicle fragrance control method in this specification are based on the same inventive concept. Therefore, the specific implementation of this embodiment can be referred to the corresponding embodiments of the vehicle fragrance control method mentioned above, and the repeated parts will not be described again.

[0058] Figure 3 The diagram illustrates the structure of a vehicle fragrance control system according to some embodiments of the present invention. For example... Figure 3 As shown, the vehicle fragrance control system provided in this embodiment of the invention includes the following perception layer, decision layer and execution layer. Each layer includes multiple components, and the components communicate with each other through a vehicle bus (CAN / LIN / Ethernet).

[0059] Specifically, the perception layer includes a positioning module and an in-vehicle map navigation module. The positioning module contains GPS / BeiDou units and an inertial navigation unit, providing the map navigation module with high-precision and high-reliability real-time positioning information. The in-vehicle map navigation module is the core data source of the system. It is responsible for route planning and real-time collection of multi-dimensional map data related to the trip, including but not limited to: destination attributes, identifying destination types (such as scenic spots, shopping malls, hospitals, service areas) through POI (Point of Interest) information; route scenarios, identifying road types (highways, tunnels, urban roads) and real-time traffic conditions (congested, smooth); dynamic information: remaining distance between the vehicle and the destination, estimated time of arrival (ETA), etc.

[0060] The decision-making level is equivalent to... Figure 2The in-vehicle fragrance control device shown includes a scene recognition module, a data storage module, an interaction module, and a cockpit domain controller. The scene recognition module receives raw data from the map module and uses a built-in map data parsing algorithm and navigation scene recognition model to deeply mine and fuse the data, outputting structured scene labels such as "on the way to a natural scenic area," "approaching a congested section," and "approaching a shopping mall." The data storage module stores the linkage strategy library, user preference settings, historical data logs, etc., and supports remote updates and optimizations via over-the-air (OTA) technology. The interaction module includes a voice interaction unit and a central control screen. It receives manual commands and preference settings from the user and provides feedback on the current operating status of the fragrance system. The cockpit domain controller receives the scene recognition results and, by integrating the "map data-fragrance control" linkage strategy library in the data storage module with the user's personalized preferences, generates precise fragrance control commands (including fragrance type, release intensity, and release timing).

[0061] The execution layer includes a fragrance execution module, comprising fragrance containers holding various fragrances, a precise release device, and a valve for adjustable concentration. It is responsible for parsing and executing instructions from the domain controller to ensure precise fragrance release.

[0062] Based on the above description and as follows Figure 3 The control system for the in-vehicle fragrance system shown in this embodiment of the invention includes the following steps: Step 1 (System Initialization): After the vehicle is powered on, initialize each module and load the linkage strategy library and user preference data.

[0063] Step 2 (Data Acquisition): The user sets the navigation destination, and the vehicle's map navigation module begins planning the route and continuously collecting multi-dimensional map data. For example, the destination POI is determined to be "XX National Forest Park".

[0064] Step 3 (Scene Recognition): The scene recognition module analyzes and merges the collected data to identify the current and future navigation scenes. For example, if the destination POI is analyzed as "Forest Park", the scene is determined to be a "Natural Scenic Area Scene"; at the same time, if a red congested road section is detected ahead of the path, it is determined that we are about to enter a "Congested Road Section Scene".

[0065] Step 4 (Command Generation): The cockpit domain controller receives the scene recognition results, queries the linkage strategy library, and generates specific fragrance control commands based on user preferences.

[0066] Step 5 (Command Issuance and Execution): The domain controller sends the command to the fragrance execution module via the vehicle bus. The execution module then completes the fragrance switching, concentration adjustment, and release. For example, when the fragrance type is switched to "Forest Scent" or the user-preset "Pine Scent," release is triggered when the remaining distance is 1 kilometer. Combining data from inside and outside the vehicle sensors (such as temperature 28℃, 3 people in the vehicle), the intensity is set to medium through a reinforcement learning model.

[0067] Step 6 (Status Feedback and Closed-Loop Control): The fragrance execution module feeds back its current operating status (e.g., "Switched to forest fragrance, medium concentration") to the domain controller. Simultaneously, the map module continuously monitors road condition changes. Once the scene changes (e.g., exiting congestion), a new round of recognition and decision-making is immediately triggered (repeating steps 2-6), achieving seamless dynamic adaptation between the scene and the fragrance.

[0068] Step 7 (End Control): When the vehicle reaches its destination or the user actively ends navigation, the system stops releasing the fragrance or switches to the default mode according to a preset strategy. For example, after arriving at the destination, the fragrance will continue to be released for 15 minutes, and then automatically stop or switch to the normal mode.

[0069] In this embodiment of the invention, in-vehicle map data is used as the core trigger source to achieve deep coupling between the fragrance system and the user's travel intentions, enabling fragrance control to leap from passive response to a new stage of intelligent proactive perception. Through the collaboration of multi-dimensional map data analysis and AI scene recognition models, the system can accurately match diverse travel scenarios such as scenic spots, congested roads, shopping malls, and hospitals, automatically adjusting the fragrance type, release intensity, and activation timing. This effectively solves the pain point of mismatch between scene and fragrance in traditional solutions, significantly improving driving comfort and emotional experience. Simultaneously, the fully automatic fragrance control completely eliminates the need for drivers to manually operate the fragrance system while driving, preventing driver distraction caused by fragrance adjustments and effectively ensuring driving safety. Furthermore, by combining reinforcement learning algorithms with user preference settings, the system can continuously learn and adapt to individual differences in needs, achieving a personalized intelligent olfactory experience, and enabling continuous evolution of the experience through OTA remote upgrade mechanisms. Furthermore, the present invention adopts a modular system design and standardized communication interface, making the solution easy to be ported and deployed across different vehicle models. In the future, it can easily access more sensory data (such as driver heart rate, fatigue, etc.) and actuators (such as ambient lighting, seat massage, etc.), thereby building a more comprehensive immersive intelligent cockpit experience.

[0070] Figure 4 This is a structural block diagram of a vehicle provided as an embodiment of this application. Figure 4As shown, the vehicle provided in this application embodiment includes a processor 410 and a memory 420, the memory being used to store instructions executable by the processor; wherein, the processor is configured to execute the instructions to implement any of the embodiments provided in the above method embodiments.

[0071] In an exemplary embodiment, the vehicle may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the vehicle fragrance control method provided in any of the above method embodiments.

[0072] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory including instructions, which can be executed by a processor of a device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the instructions in this non-transitory computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform the in-vehicle fragrance control method provided in any of the above-described method embodiments.

[0073] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the vehicle fragrance control method provided in any of the above-described method embodiments.

[0074] The above description does not provide detailed technical specifications regarding the structure of each layer. However, those skilled in the art should understand that layers and regions of desired shapes can be formed using various technical means. Furthermore, to form the same structure, those skilled in the art can also design methods that are not entirely identical to those described above. Additionally, although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be advantageously combined.

[0075] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0076] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for controlling a car fragrance, applied to a vehicle, wherein the vehicle is equipped with a car fragrance, characterized in that, The method includes: Obtain the scene tags of the destination, which are derived from N preset standardized tags; Based on the scene label of the destination and the current state of the vehicle, a scene recognition model is used to predict multiple future scene labels corresponding to the various locations of the vehicle in the future within a certain time period. If a target scene label exists among the multiple future scene labels, determine the current remaining distance between the vehicle and the target scene corresponding to the target scene label, and determine the release starting distance for activating the in-vehicle fragrance based on the target scene label. The target scene label is the scene label of a known scene among the N standardized labels. If the current remaining distance is not greater than the release starting distance, the release parameters of the car fragrance are determined according to the target scene label, and the car fragrance is controlled to work according to the release parameters.

2. The method as described in claim 1, characterized in that, The process of obtaining the scene tags for the destination includes: Obtain destination information, which includes the destination name, the destination navigation code, and the destination address information; Search among the N standardized labels in the navigation database for a standardized label that corresponds to the navigation code of the destination; If it exists, the standardized label corresponding to the navigation code of the destination will be used as the scene label of the destination; If it does not exist, the name of the destination is matched using the N regular expressions corresponding to the N standardized tags respectively; If a match is successful, the standardized tag corresponding to the matching regular expression will be used as the scene tag for the destination. If the match fails, the address information of the destination is analyzed to obtain the scene label of the destination.

3. The method as described in claim 1, characterized in that, The step of predicting multiple future scene labels corresponding to the vehicle's various locations within a future time period based on the destination scene label and the vehicle's current state using a scene recognition model includes: The current state of the vehicle and the environmental state of the environment in which the vehicle is currently located are collected according to a preset sampling step size, and combined with the scene label of the destination to form a temporal feature vector corresponding to each sampling step size. After constructing P temporal feature vectors, the P temporal feature vectors are input into the scene recognition model; The scene recognition model predicts the future scene label corresponding to the vehicle's location at each of the next Q sampling steps, thus obtaining Q future scene labels. The future duration is the total duration of the Q sampling steps.

4. The method as described in claim 3, characterized in that, Before determining the current remaining distance between the vehicle and the target scene corresponding to the target scene label, the method further includes: If any of the Q future scene labels contains a scene label for a known scene, determine the target scene closest to the current position of the vehicle from the scene labels of the known scene, and use the standardized label corresponding to the target scene as the target scene label; If none of the Q future scene labels contain a scene label for a known scene, then it is determined that the target scene label does not exist among the Q future scene labels.

5. The method as described in claim 4, characterized in that, The method further includes: If no target scene label is found among the Q future scene labels, the following steps are performed: collecting the current state of the vehicle and the current state of the environment in which the vehicle is located according to a preset sampling step size, and combining them with the scene label of the destination to form a temporal feature vector corresponding to each sampling step size.

6. The method as described in claim 1, characterized in that, The step of determining the release starting distance of the in-car fragrance based on the target scene label includes: Obtain the base release distance and scene correction coefficient corresponding to the target scene label; The release starting distance is determined based on the basic release distance, the scene correction coefficient, and the vehicle's current speed. The release starting distance is the distance between the vehicle and the target scene corresponding to the target scene label when the in-vehicle fragrance is activated.

7. The method as described in claim 1, characterized in that, Determining the release parameters of the in-vehicle fragrance based on the target scene label includes: Obtain the current cabin environment status of the vehicle; The release parameters corresponding to the target scene label are determined based on the current cabin environment of the vehicle using the fragrance release model.

8. The method as described in claim 7, characterized in that, After controlling the in-vehicle fragrance to operate according to the release parameters, the method further includes: Collect user feedback; The user feedback and the vehicle status during the operation of the in-car fragrance system are uploaded to the cloud so that the cloud can update the fragrance release model based on the user feedback and the vehicle status during the operation of the in-car fragrance system.

9. A control device for a car fragrance system, applied to a vehicle, wherein the vehicle is equipped with a car fragrance system, characterized in that, The device includes: The acquisition module is used to acquire the scene tags of the destination, wherein the scene tags of the destination are derived from N preset standardized tags; The prediction module is used to predict multiple future scene labels corresponding to the various locations of the vehicle within a future time period based on the scene labels of the destination and the current state of the vehicle, using a scene recognition model. The determination module is used to determine the current remaining distance between the vehicle and the target scene corresponding to the target scene label when a target scene label exists among the multiple future scene labels, and to determine the release starting distance of the in-vehicle fragrance based on the target scene label, wherein the target scene label is the scene label of a known scene among the N standardized labels; The determining module is further configured to, if the current remaining distance is not greater than the release starting point distance, determine the release parameters of the in-vehicle fragrance based on the target scene label, and control the in-vehicle fragrance to operate according to the release parameters.

10. A vehicle, characterized in that, The vehicles include: processor; A memory for storing instructions to be executed by the processor, wherein the processor is configured to execute the instructions to implement the method as described in any one of claims 1 to 8.

Images