Danger early warning method and system for child smart watch

Abstract

The invention discloses a danger early warning method and system for a child smart watch. The method comprises the steps that S1, the smart watch collects surrounding environment data in real time; s2, based on the environmental data, analyzing and identifying potential danger types and levels by using a pre-trained danger prediction model; s3, when a specific danger is identified, triggering corresponding early warning information and pushing the early warning information to associated equipment; s4, adjusting the parameters of the danger prediction model according to the received feedback so as to improve the recognition accuracy; according to the danger early warning method and system for the child smart watch, the problems that the accuracy of an early warning system in the prior art is influenced, sound and image analysis models adopted by most products are relatively fixed, the early warning system is difficult to quickly adapt to continuously changing environments, and the risk of malicious cracking or disguise attack is faced, so that the early warning system is possibly invalid, and the early warning effect is poor are solved. And the generalization ability of an algorithm model in a variable environment cannot be improved while the stability and effectiveness of an early warning system are guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular to a danger warning method and system for children's smart watches. Background Art

[0002] In the prior art, in order to solve the problem of children's safety, a danger warning method for children's smart watches has appeared on the market. This method analyzes the sound and image signals in the environment and intelligently detects changes in the surrounding environment in order to prevent accidents. For example, when the watch senses abnormal sounds such as vehicle horns or abnormal noises, it will activate an emergency alarm and send a warning signal to the preset safety monitoring terminal, thereby providing additional safety protection for children.

[0003] However, this technology still has certain limitations in practical applications. First, the complex and changeable noise environment can easily interfere with the intelligent algorithm's capture and screening of effective warning signals, which affects the accuracy of the warning system. Secondly, the sound and image analysis models used by most products on the market are relatively fixed and difficult to quickly adapt to the changing environment. In addition, these models are also at risk of being maliciously cracked or disguised, which may cause the warning system to fail. Therefore, how to ensure the stability and effectiveness of the warning system while improving the generalization ability of the algorithm model in a changing environment has become an important challenge in current technical research. Summary of the invention

[0004] The purpose of the present invention is to provide a danger warning method and system for children's smart watches, so as to solve the problem of how to improve the generalization ability of the algorithm model in a changing environment while ensuring the stability and effectiveness of the warning system.

[0005] To achieve the above object, the present invention provides the following technical solution: a danger warning method for a children's smart watch, the method comprising:

[0006] S1: Smartwatch collects surrounding environment data in real time;

[0007] S2: Based on environmental data, use the pre-trained hazard prediction model to analyze and identify potential hazard types and levels;

[0008] S3: When a specific danger is identified, the corresponding warning information is triggered and pushed to the associated devices;

[0009] S4: Adjust the hazard prediction model parameters according to the received feedback to improve the recognition accuracy.

[0010] Preferably, the step S1 specifically includes:

[0011] Start the sensor module to obtain image data;

[0012] Start the sound collection module to receive audio signals;

[0013] Read geographic location and time tag information;

[0014] The sensor module transmits synchronously to the data processing unit;

[0015] All the acquired raw data are encrypted and uploaded to the cloud server for preliminary analysis.

[0016] Preferably, the step S2 specifically includes:

[0017] Get the noise level L in the environment;

[0018] Get the difference between the ambient temperature T and the preset temperature T0 |T-T0|;

[0019] Scoring based on the safety index S of the current location;

[0020] An additional weighting factor W is recorded when the ambient lighting level E is above a predetermined safety value.

[0021] Preferably, the step S3 specifically includes:

[0022] Constructing alarm message content according to the recognition result;

[0023] Customize push rules to filter the list of target associated devices to be pushed;

[0024] Add security verification mechanism to push messages to ensure privacy protection;

[0025] Select an instant messaging channel to send warning message text and sound prompts;

[0026] Determine whether additional emergency procedures need to be initiated and responded to;

[0027] The target associated device list to be pushed by the custom push rule is sorted by the following formula to prioritize the associated devices:

[0028] P d =β1×E c +β2×L d -β3×T d ,

[0029] Among them, P d Represents the priority of the associated device, E c is the physical distance between the device and the child, L d is the average response time of the device in the push history, T d is the power state of the device, β1, β2, and β3 are adjustment coefficients;

[0030] The security verification mechanism is added to the push message to ensure privacy protection. The warning message is encrypted and processed. The receiving device needs to pass the identity authentication before decrypting the message. Only the verified device can successfully decrypt and read the message content.

[0031] The selecting of an instant messaging channel to send a warning message text and a sound prompt comprises:

[0032] The optimal communication channel is selected using the following formula:

[0033] C s =argmin(γ1×T c +γ2×R b -γ3×P r ),

[0034] Among them, T c is the transmission delay of the channel, R b is the bandwidth utilization, P r is the transmission success rate of the channel, and γ1, γ2, and γ3 are the weight coefficients of channel selection.

[0035] Preferably, the step S4 specifically includes:

[0036] Collect feedback data packages from users, including hazard handling records and personal assessments;

[0037] Verify the validity of feedback information to prevent malicious interference;

[0038] Analyze the error recognition patterns and causes of effective feedback extraction;

[0039] Update or adjust correlation coefficients in risk prediction algorithms;

[0040] Verify the effectiveness and accuracy of the improved algorithm on the test set and deploy it.

[0041] Preferably, the step of scoring based on the safety index S of the current location specifically includes:

[0042] Calculate the distance D from the known dangerous area based on the location information;

[0043] If D is less than the safety distance threshold DT, the danger level increases by α level;

[0044] If D is greater than or equal to the safety distance threshold DT, the danger level decreases by β levels;

[0045] The smaller the danger zone distance, the higher S is.

[0046] Preferably, the step of: the smaller the distance to the dangerous area, the higher S is;

[0047] For each distance D, apply the formula S = max(5 × α × (D / DT) 2 ,1);

[0048] For each measurement result S, find the mean;

[0049] If the average result is less than 10%, it is considered normal and the alert level is not raised;

[0050] On the contrary, if the result is higher than 30%, it will be considered a high-risk area and an emergency warning will be issued.

[0051] Preferably, the step of applying the formula S=max(5×α×(D / DT)2,1) to each distance D comprises:

[0052] Record the calculated result R for evaluation;

[0053] If S is in the range of [5α,1], the calculation process is confirmed to be correct;

[0054] This calculation process is performed for each time period t and the results are compared to monitor changes;

[0055] Update the model parameter γ based on the changes in results for all time periods.

[0056] Preferably, the step of updating the model parameter γ based on the result changes of all time periods includes:

[0057] Perform normalization processing N(SD) on the risk score SD of all time periods to obtain NSD;

[0058] Calculate the mean value of the change of the normalized score NSD, mean△(NSD), and compare it with a preset threshold threshold. If (mean△(NSD)≥threshold), then γ=γ+1;

[0059] Adjust the specific value of γ according to whether the conditions are met to enhance dynamic adaptability;

[0060] The final adjusted new prediction parameter combination Λ includes {γ, D, S}.

[0061] A danger warning system for a children's smart watch, used to implement the steps of a danger warning method for a children's smart watch, the system comprising:

[0062] Data acquisition module, used for smart watches to collect surrounding environment data in real time;

[0063] The analysis and identification module uses pre-trained hazard prediction models to analyze and identify potential hazard types and levels based on environmental data;

[0064] Warning push module: when the analysis and identification module identifies a specific danger, the corresponding warning information is triggered and pushed to the associated equipment;

[0065] The model adjustment module is used to adjust the risk prediction model parameters according to the received feedback information to improve the recognition accuracy.

[0066] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0067] The danger warning method and system for children's smart watches collects surrounding environment data in real time through smart watches, uses a pre-trained danger prediction model to analyze and identify potential danger types and levels, triggers corresponding warning information to be pushed to associated devices when a specific danger is identified, adjusts the danger prediction model parameters according to the received feedback to improve the recognition accuracy, can more accurately identify the potential danger types and levels in the surrounding environment, reduces false alarms and missed alarms caused by complex noise environments, thereby significantly improving the accuracy of warnings, and has more stable performance in different environments. It can quickly adapt to environmental changes and ensure effective monitoring and evaluation of multiple danger sources in complex environments, thereby providing more comprehensive child safety protection, can timely trigger warning information and push it to associated devices, ensure that parents or guardians can receive danger signals and take corresponding measures in the first place, further improve children's safety protection, and solve the problems that the accuracy of the existing warning system is affected, the sound and image analysis models used by most products are relatively fixed, and it is difficult to quickly adapt to the changing environment, and there is a risk of being maliciously cracked or disguised attacks, which may cause the warning system to fail, and it is impossible to improve the generalization ability of the algorithm model in a changing environment while ensuring the stability and effectiveness of the warning system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 is a flow chart of the method of the present invention;

[0069] Figure 2 It is a schematic diagram of module connection of the present invention. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] like Figure 1 As shown, a danger warning method for a children's smart watch, the method comprising:

[0072] S1: Smartwatch collects surrounding environment data in real time;

[0073] S2: Based on environmental data, use the pre-trained hazard prediction model to analyze and identify potential hazard types and levels;

[0074] S3: When a specific danger is identified, the corresponding warning information is triggered and pushed to the associated devices;

[0075] S4: Adjust the hazard prediction model parameters according to the received feedback to improve the recognition accuracy.

[0076] First, as a daily accessory for the wearer, children's smart watches will continuously and in real time capture information about the surrounding environment during normal operation, including but not limited to audio, vision, and various signals provided by sensors, such as images from the built-in camera, sound clips recorded by the microphone, and changes in movement and behavior monitored by devices such as accelerometers. The collection of environmental data is not limited to fresh content captured in real time, but also includes information that has been saved before, such as recordings of the last few minutes. This wide range of data acquisition helps the smart watch system build a more complete view of the world around it.

[0077] For example, if a child is playing outdoors, the neighborhood he is in may be noisy and crowded. The watch will continuously detect the volume intensity and sound characteristics (whether it is a scream or a warning sound such as a car horn) through the built-in microphone, and obtain the distribution of people and cars on the street through the front camera; at the same time, temperature, pressure and other sensory data will be uploaded. Such a comprehensive data foundation is conducive to the next step of risk determination and subsequent actions.

[0078] Next, once the data is collected, it is analyzed. This process involves a pre-trained model to predict and assess the risks of the captured data, and extract indicators about the likelihood and severity of possible problems. The design concept of this module is to draw on a lot of past case experience and use relevant theoretical guidance to effectively deal with real-world situations. The model itself is the result of a complex algorithmic network. It can infer the size of potential risks based on the various information received and decide the next course of action: send out an alarm signal, remind attention, or choose to ignore and continue to observe subsequent changes. In this scenario, we assume that a good training library has been established and the model has been trained.

[0079] Suppose in a corner of a park, a smartwatch notices a child standing alone without an adult. From the visual image data, it is known that a suspicious stranger is approaching and the facial expression shows emotional signals such as tension or confusion; and the sound characteristics heard indicate that there may be an emergency call or other abnormal noise. Based on the above comprehensive considerations (such as the stranger's pace of approaching the child in a short period of time), after judging the pre-set threshold, the risk assessment program will eventually mark the entire situation as a medium-level suspicious contact risk; and then initiate the early warning response plan. It is worth noting here that even in a relatively calm park environment, there are many variables that need to be taken into consideration, which puts more pressure on the system recognition mechanism and requires continuous optimization to adapt to changes in different environmental conditions.

[0080] If a safety hazard is subsequently identified, the watch will immediately activate the early warning response strategy to send a warning notification to the predetermined contact device, which may be the child's father's mobile phone or the school office fixed landline. These can be clearly specified in the settings to ensure that timely and effective external assistance is implemented. In addition, in an emergency, in addition to displaying text instructions, physical means such as ringing and vibration can be used to enhance the prompt effect, making it easier for the target group to notice the existence of the warning information.

[0081] In the previous example, in order for parents to quickly know the situation and take countermeasures, after determining that the warning line has been reached, the smart terminal worn on the child's wrist will immediately initiate a message transmission request to send detailed information such as text, pictures, and location directly to the designated recipient; and take corresponding auxiliary measures based on the current situation, such as cyclically sounding alarms to attract the attention of nearby pedestrians to assist in prevention or turning on the flash mode to attract the attention of others and look for possible help.

[0082] The last key part is how to optimize and update the existing security feedback. Since the external situation is ever-changing, even if a lot of work is done in advance, it is difficult to ensure that every link is 100% perfect. Therefore, it is necessary to regularly adjust and improve the original mechanism to make it more in line with actual combat needs. This requires smart watches to have a certain degree of self-learning ability.

[0083] Suppose the child finally successfully escapes from the risk event mentioned above and provides his own feedback and suggestions when asked by the parents for the specific reasons afterwards, for example, he thinks that the sound warning used at that time was too sharp and may cause the child to panic, which has a counter-effect not as good as expected; or he proposes whether the accuracy of location sharing can be further improved to make the help-seeking efficiency more efficient. The system should actively collect and summarize these user feedbacks, and then integrate them as a new input source into the next round of danger interpretation process for retraining, so that it can face similar situations more calmly in the future without obvious loopholes. The practice of improving overall accuracy through continuous correction and optimization can not only make this special-purpose bracelet closer to people's ideal demands, but also greatly enhance children's own ability to cope with accidents.

[0084] In summary, it can be seen that such a design process covers the important stages of the entire hazard warning process and considers various possibilities as carefully as possible. Through this method, we can effectively protect the life, health and safety of children wearing such equipment in various environments, and bring peace of mind and convenience to their parents and caregivers.

[0085] The step S1 specifically includes:

[0086] Start the sensor module to obtain image data;

[0087] Start the sound collection module to receive audio signals;

[0088] Read geographic location and time tag information;

[0089] The sensor module transmits synchronously to the data processing unit;

[0090] All the acquired raw data are encrypted and uploaded to the cloud server for preliminary analysis.

[0091] This implementation method acquires image data and audio signals respectively by starting the sensor module and the sound acquisition module. These data are read together with the device's geographic location and time tag information to ensure the spatiotemporal synchronization of the data. Subsequently, the sensor module will synchronously transmit the acquired data to the data processing unit for preliminary processing and integration. All raw data are uploaded to the cloud server for further analysis after encryption, which not only improves the security of the data, but also ensures the integrity and accuracy of the data, enables efficient data acquisition and processing, and ensures the spatiotemporal consistency and security of the data. By synchronously acquiring multiple data types and uploading them to the cloud with encryption, the system can provide more comprehensive and accurate analysis results. In addition, the encryption mechanism enhances the privacy protection of the data and reduces the risk of data leakage during transmission.

[0092] Next, the steps described specifically refer to further refining a danger warning method for children's smart watches to facilitate understanding of each step of the operation process in actual applications.

[0093] The first step is to start the sensor module to obtain image data. This means turning on the camera or image sensing element installed in the smart watch to capture the surrounding environment. For example, when the watch worn by a child is set to turn on this mode at any time, the sensor will continuously collect images of the surrounding environment; when a specific trigger condition or a preset time interval is detected, a snapshot or short video will be captured and saved for subsequent analysis.

[0094] The next step is to start the sound collection module to receive audio signals. This involves starting a microphone or other voice input device to capture the surrounding audio stream. For example, when a child passes through a noisy area or encounters a potential threatening situation, the sound recording is automatically triggered, and the surrounding sounds are recorded through the microphone as reference data for early warning.

[0095] The third step is to read the geographic location and time tag information. The exact geographic location of the child's smart watch is determined through GPS or other positioning technologies, and the timestamp of the recorded image and audio data is marked, so that it can be understood where and when the data was obtained. For example, when a child passes a certain location after school and encounters harassment, the locator on the watch not only provides accurate latitude and longitude coordinate information but also shows the specific time period (such as 5:15 pm).

[0096] The following steps are to make the data collected by the sensor be uniformly transmitted to the processor or CPU through the internal channel of the hardware. This is called the sensor module. The data is synchronously transmitted to the data processing unit, that is, the internal control system of the watch receives the above-mentioned various raw information, providing the necessary basic data support for the subsequent comprehensive analysis and uploading of the obtained data. For example, the images captured by the camera, the recording files and the GPS coordinates are sent to the processing unit at the exact moment to wait for the next processing instruction to be executed.

[0097] The last step is to ensure that all valid content received, recorded, and processed is transmitted to the server in a secure manner, that is, all the original data obtained is encrypted to avoid illegal intrusions and privacy leaks, and uploaded to a cloud server specially established for preliminary analysis for background monitoring and management and later deep algorithm analysis. For example, the audio and video files that have been encrypted and compressed are then securely transmitted to a secure area in the cloud via mobile networks or Wi-Fi for the first rapid identification, to determine possible risk factors, and to provide timely feedback to children and their parents and guardians so that they know how to respond to emergencies.

[0098] The above steps specifically explain the main implementation processes and scenarios in the operation of the smart children's watch danger warning solution, from the initial triggering of the collection action to the final sending of data to the cloud for analysis.

[0099] Next, the steps described mainly include the following steps and their examples:

[0100] First, the environmental data is formatted to meet the standards for the hazard prediction model input. This usually involves converting the collected data into a specific form or performing certain preprocessing tasks on the data, such as removing inconsistent data, filling missing values, or unifying the data format so that it can be accepted by the hazard prediction model. For example, information such as GPS coordinates, volume levels, or temperatures collected in the environment of a children's smart watch may initially be in various forms. We need to write corresponding programs or scripts to convert these raw information in various formats into the specific structure required by the model.

[0101] Loading a pre-trained hazard prediction model is a subsequent step. This stage involves calling previously developed software modules or function libraries to reference pre-trained models that can effectively identify potential risk scenarios, such as support vector machines (SVMs), decision trees, neural networks and other machine learning models. In a practical case, suppose we already have a hazard assessment model that has been verified to be effective in real-world use cases. Now we want to import it into the current environment for further operation. This action can be achieved by reading a pre-stored parameter file after training or by calling a cloud service.

[0102] It is also very important to extract the key feature vectors from the environmental data. In this step, we need to use algorithms to automatically select the most critical elements for subsequent risk assessment. These elements may be values ​​extracted from the above-mentioned formatted GPS location, surrounding noise intensity and other signals, or more complex representation methods, such as frequency patterns. For example, when we consider the change of GPS coordinates as a key feature of possible danger to children, we calculate the speed and frequency of children's movement in a specific area over a period of time. In this way, we can more specifically identify those iconic movement features that indicate that children are at potential risk, so as to more effectively predict risks.

[0103] Next, the key feature vector is used to calculate the probability of potential danger through the loaded danger prediction model. This involves using the numerical value or vector obtained in the previous step as input to the trained prediction engine, which automatically calculates the specific probability of danger under the environmental factors. In the scenario discussed in the above example, we will obtain the probability of a child facing different types of risks at a certain moment, such as the risk of being lost or approaching a high-risk location.

[0104] The final output is based on the calculated risk value and its corresponding warning level. This result will be used to directly provide an alarm level or display it to the user of the children's smart watch, informing them of the current risks and the corresponding urgency and severity, helping parents and children to take reasonable actions in a timely manner. In our scenario, for example, if the system determines that the child is approaching a marked dangerous area and moving quickly, it will send a high-alert level notification to the child's parents or designated person in charge, telling them that the situation is very urgent and that appropriate actions must be taken to ensure the safety of the child. Through this mechanism, we have completed the entire logical process of danger warning and successfully achieved a comprehensive understanding of its functions.

[0105] Next, the steps described are specifically the implementation process of further detailing and specifying the claims of the original children's smart watch danger warning method. The specific meaning of each sub-step is explained below, and a hypothetical example is provided to illustrate it.

[0106] First, the alarm message content is constructed based on the recognition results. Here, it means constructing specific warning information content based on the results of previous environmental detection or behavior analysis. For example, suppose the smartwatch is equipped with face recognition technology and geolocation services, and has identified that an unknown person is approaching a child at the scene, and the location is in an unknown area near a school. Then the task at this stage is to convert these key data into warning messages that are easy to understand and act on.

[0107] Secondly, customizing push rules to filter the list of target associated devices for push means developing a push mechanism that uses certain logic to decide to whom the warning should be pushed. For example, in the above example, the watch can determine whether to send the alarm to the child's parents, school administrators, and other bound guardian contacts based on the preset logic; and it may adjust the push strategy based on the degree of urgency, so that the notification can accurately reach the party with the most decision-making ability at the first time, so as to make a quick and effective response.

[0108] Next, adding a security verification mechanism to push messages to ensure privacy protection means enhancing the security of the message delivery process to prevent sensitive data from being leaked or improperly used, such as using two-factor authentication or encrypted communication to ensure that only the authorized recipient can view the specific content. In this scenario, all sent secure messages will be end-to-end encrypted, and when transmitted to a designated mobile phone, watch or other connected terminal, the recipient must undergo password verification before reading the message, which helps to strengthen the overall protection level of user information and personal data.

[0109] Then, when it comes to choosing an instant messaging channel to send warning message text and sound prompts, it means selecting a communication method suitable for the current situation to send the previously prepared information. Generally speaking, there will be multiple alternative options for users to choose from. Back to the case, considering the different available devices of different guardians in emergency situations, the push notification of the warning system can be sent through multiple channels by selecting the simultaneous sending of mobile phone software notifications and watch vibration alarms.

[0110] The last step is to confirm whether it is necessary to start additional emergency procedures and respond, which means judging whether it is necessary to activate additional emergency response measures to deal with the possibility of more serious situations, and automatically executing subsequent actions according to preset instructions. Continuing with the above hypothetical scenario discussion, when a stranger appears and gradually approaches the child, once it is determined that there is an actual threat (for example, it is recognized from the surveillance image that the suspicious person is approaching the child or trying to take him away by force), the children's smart watch will trigger a high-level alarm program, not only issuing a regular warning, but also automatically starting the camera to record the video for subsequent tracking, and immediately dialing the 24-hour duty phone number of the public security department. These actions will be automatically performed under the guidance of the pre-set emergency handling rules without the need for manual operation by the user, playing an important role in protecting the safety of children.

[0111] Through these step-by-step practical demonstrations, it is not difficult to see that the entire children's smart watch danger warning method can effectively and quickly transmit key warnings and take corresponding preventive measures to deal with complex and changing situations.

[0112] Next, the steps are described in detail:

[0113] Collecting feedback data packets from the user end means collecting the behavior data, operation data and any operation records of children's smart watch users when encountering abnormal conditions into the background database. To illustrate with a real case, suppose a parent discovers a safety issue that has not been correctly warned by the system through a children's smart watch, and then reports it to the customer service, and the background can record this data. This feedback data will include the record of the watch's handling when the danger occurs and the user's personal opinion and evaluation of the situation.

[0114] The purpose of feedback validity verification is to confirm that the information received is authentic and reliable and is sent out for normal feedback needs, rather than false alarms caused by abnormal use. For example, customer service may receive multiple repeated feedback reports from users about a specific dangerous situation. In the verification stage, it can be confirmed that the situation mentioned by the user is indeed true and it is ruled out that it is a false alarm created by humans.

[0115] Then, we will analyze the valid information content and find out the deficiencies in the recognition process. Continuing with the above example, after the verification phase is confirmed, we will analyze in detail why the smartwatch did not accurately recognize the accident and find out the specific factors that caused this situation.

[0116] On this basis, the relevant prediction coefficients of the danger warning can be corrected and improved, such as modifying the parameter thresholds related to the recognition of similar scenarios to the incident, to improve the recognition efficiency in similar situations in the future. The above example shows that after understanding the root cause of the false alarm, the weight of the recognition of this special scene during intelligent analysis can be readjusted to make the judgment more accurate and reliable.

[0117] Finally, the modified danger warning algorithm is put into the preset environment for repeated verification to ensure that the updated version can meet a wider range of usage needs before it is pushed. For example, after the above coefficients are updated, it is necessary to implement a series of tests in a virtual simulation environment to ensure that the algorithm has sufficient accuracy and reliability in actual operation before it can be deployed to all children's smart watches for formal use.

[0118] Next, the steps to identify potential hazard types and levels based on the environmental data using a pre-trained hazard prediction model analysis are described, first understanding the specific operations of its components and how to implement them. These steps take into account different parameters in the environment for a comprehensive assessment:

[0119] Get the noise level L in the environment;

[0120] Get the difference between the ambient temperature T and the preset temperature T0 |T-T0|;

[0121] Scoring based on the safety index S of the current location;

[0122] An additional weighting factor W is recorded when the ambient lighting level E is above a predetermined safety value.

[0123] The first step is to obtain information about the noise level L in the environment. This means collecting surrounding noise or using the built-in sensor of a smart watch to monitor the level of surrounding sound. For example, the watch collects the current background noise of 70dB.

[0124] The second step involves monitoring the gap between the current environment and the ideal safety conditions. This is done by calculating the difference between the actual ambient temperature T and an ideal, defined reference point preset temperature T0, |T-T0|. If the preset ideal safe room temperature is 25°C and the detected temperature is 28°C, the temperature gap is 3°C (a positive number means the temperature is too high, which may indicate a fire or other heat-related potential risk).

[0125] The third step is to use the location information to calculate the inherent safety of the place and quantify the rating, called the safety index S, which can be obtained by analyzing historical or real-time accident rates, public security conditions and other factors provided by geolocation services. Assume that the S score is low in a high-risk area (for example, only 10 out of 1 to 100), while it is better to get a high score in a safer place (for example, the safety score in a park may be 95).

[0126] The last step takes the visibility environment into account: if the measured lighting level E is far below the safety baseline requirement, an additional weight factor W will be introduced to reflect the additional hazards that may occur under dim light conditions. This variable can be used to further improve the accuracy of warning signals and the classification of emergency levels.

[0127] The above measurement results together constitute the input to the pre-trained prediction algorithm for comprehensive assessment of the potential threat type and level. As an example, a child wearing this device in a relatively unfamiliar area, where the ambient noise is 80 decibels, 4 degrees Celsius higher than the expected indoor comfort temperature difference (27℃ vs 23℃), and is located in an area with a low evaluation value (S is only 13 points), plus it is dark at night and only illuminated by street lights (low E level causes higher w value), all of which will be aggregated and processed by the danger prediction program, so as to output the corresponding risk level to notify parents.

[0128] The step S3 specifically includes:

[0129] Constructing alarm message content according to the recognition result;

[0130] Customize push rules to filter the list of target associated devices to be pushed;

[0131] Add security verification mechanism to push messages to ensure privacy protection;

[0132] Select an instant messaging channel to send warning message text and sound prompts;

[0133] Determine whether additional emergency procedures need to be initiated and responded to;

[0134] Construct the alarm message content based on the recognition results: The system first constructs the alarm message content based on the results provided by the recognition module (such as image or audio analysis, sensor data). These contents usually include a detailed description of the warning information, such as the detected abnormal situation, a summary of relevant data (such as location, time, recognized objects or events), etc. This step ensures that the alarm message can accurately reflect the recognition results, which is convenient for subsequent processing and response. Custom push rules filter the list of target associated devices to push: The system uses custom push rules to filter the most suitable associated devices for message push. The filtering rules may include parameters such as the physical distance of the device, response time, and device status to ensure that the message can be delivered to devices that can respond quickly. This step improves the response speed and efficiency of the system by optimizing device selection. Add a security verification mechanism to the push message to ensure privacy protection: Before the message is pushed, the system will encrypt the message content and require the receiving device to authenticate. This step ensures that only authenticated devices can decrypt and read the message content, preventing the message from being intercepted or leaked during transmission, thereby protecting the user's privacy and data security. Select the instant messaging channel to send the warning message text and sound prompt: The system selects the optimal communication channel to send the warning message text and sound prompt based on factors such as the transmission delay, bandwidth utilization, and transmission success rate of the instant messaging channel. This step further improves the effectiveness of the early warning system by optimizing the communication channel selection to ensure that the message can be transmitted to the target device in a timely and reliable manner. Confirm whether additional emergency procedures need to be initiated and respond: After completing the above steps, the system will check whether the current warning situation requires the initiation of additional emergency procedures (such as starting an emergency broadcast, sounding an alarm, or notifying relevant emergency departments). If necessary, the system will respond immediately and execute the corresponding emergency procedures to ensure a quick and effective response to emergencies. Through the above steps, this implementation method can achieve efficient and reliable early warning information processing and transmission. From constructing the alarm message to confirming whether to start the emergency procedure, the system can respond quickly to ensure the timeliness and accuracy of the warning information. The application of custom push rules and security verification mechanisms further enhances the security and adaptability of the system. In addition, selecting the optimal communication channels and flexibly initiating emergency procedures allow the system to operate efficiently in different scenarios.

[0135] In actual applications, the way to construct the content of the alarm message can be adjusted according to different recognition results. For example, multimedia elements (pictures, video clips) can be added to enrich the expression of the alarm information. Push rules can also be customized according to the application scenario, such as adding geo-fence rules or time sensitivity settings. In terms of security verification mechanism, more advanced encryption methods or multi-factor verification can be considered to meet higher security requirements. When selecting a communication channel, the weight coefficient can be adjusted according to the specific usage scenario to ensure that the most reliable communication method is selected in special circumstances (such as network congestion or emergency).

[0136] The list of target associated devices for push filtering by custom push rules uses the following formula to prioritize the associated devices:

[0137] P d =β1×E c +β2×L d -β3×T d ,

[0138] Among them, P d Represents the priority of the associated device, E c is the physical distance between the device and the child, L d is the average response time of the device in the push history, T d is the power state of the device, β1, β2, and β3 are adjustment coefficients;

[0139] Through this priority sorting formula, the system can more intelligently and effectively select suitable devices for warning message push. This method can ensure that the warning message is delivered first to the device that is closest, responds fastest and has sufficient power, thereby significantly improving the response speed and reliability of the system. The introduction of the adjustment coefficient also enables the system to flexibly adapt to different application scenarios, enhancing the adaptability and practicality of the system. In actual applications, the various parameters and adjustment coefficients in the formula can be modified according to different needs. For example, other parameters, such as the stability or security of the device's network connection, can be introduced to further optimize device selection. In addition, the formula form can also be adjusted according to actual needs, such as using a linear weighted approach to balance various factors, or by introducing more dynamic adjustment mechanisms to cope with different application scenarios.

[0140] Add a security verification mechanism to push messages to ensure privacy protection. Warning messages are encrypted. The receiving device must pass identity authentication before decrypting the message. Only authenticated devices can successfully decrypt and read the message content.

[0141] Select instant messaging channels to send warning message text and sound prompts including:

[0142] The optimal communication channel is selected using the following formula:

[0143] C s =argmin(γ1×T c +γ2×R b -γ3×P r ),

[0144] Among them, T c is the transmission delay of the channel, R b is the bandwidth utilization, P r is the transmission success rate of the channel, and γ1, γ2, and γ3 are the weight coefficients of channel selection.

[0145] This security verification mechanism ensures the security of warning message transmission by performing encryption processing and identity authentication during message transmission. Encryption processing prevents messages from being stolen or tampered with during transmission, while identity authentication further ensures that only authorized devices can decrypt and access the message content. This dual protection mechanism effectively protects the security of the system and the privacy of users. Through this implementation, the system can effectively prevent unauthorized devices from accessing sensitive warning messages and avoid potential privacy leakage risks. At the same time, due to the use of encryption processing and identity authentication, the system can ensure the integrity and confidentiality of the message content, thereby improving the security of the overall system and user trust. In actual applications, the encryption algorithm can select different types and strengths according to system requirements, such as using asymmetric encryption algorithms in high-security scenarios and symmetric encryption algorithms in performance-sensitive scenarios. The identity authentication method can also be flexibly adjusted to add two-factor authentication or biometric-based verification to further enhance security.

[0146] The selection mechanism comprehensively evaluates the advantages and disadvantages of each channel by considering the transmission delay, bandwidth utilization and transmission success rate of the communication channel, and dynamically selects the optimal communication channel for message transmission according to the actual situation. The weight coefficients γ1, γ2, and γ3 are used to adjust the importance of each factor, so that the system can flexibly select the best communication channel according to different application scenarios. This implementation method can significantly improve the transmission efficiency and success rate of warning messages by optimizing the selection of communication channels. Selecting a communication channel with low delay, high bandwidth utilization and high success rate can ensure the timely and reliable transmission of warning messages, thereby improving the response speed of the system and user experience. In practical applications, the weight coefficients γ1, γ2, and γ3 can be adjusted according to specific needs. For example, in an emergency, the weight of the transmission success rate Pr can be increased, and the channel with the highest transmission reliability can be given priority. In non-emergency situations, it may be more inclined to select the channel with the lowest delay. In addition, more evaluation factors, such as communication cost or network load, can be introduced to further optimize the selection strategy of the communication channel.

[0147] Step S4 specifically includes:

[0148] Collect feedback data packages from users, including hazard handling records and personal assessments;

[0149] Verify the validity of feedback information to prevent malicious interference;

[0150] Analyze the error recognition patterns and causes of effective feedback extraction;

[0151] Update or adjust correlation coefficients in risk prediction algorithms;

[0152] Verify the effectiveness and accuracy of the improved algorithm on the test set and deploy it.

[0153] The user-side interface collects feedback data packets after handling dangerous situations. These data packets usually include the user's operation records during the dangerous handling process, as well as the user's personal evaluation of the system prompts. These data provide a basis for subsequent analysis and algorithm adjustment. The collected feedback information will first be verified for validity. The purpose of this step is to prevent users from maliciously providing false or disruptive feedback and ensure the authenticity and reliability of the feedback data. Validity verification may include checking data consistency, analyzing user behavior patterns, and cross-verifying user identity information. The verified valid feedback will be further analyzed. The system will extract the possible error patterns and their causes in the hazard identification process through data mining and pattern recognition technology. This step helps to identify weaknesses or deficiencies in the algorithm and provide a basis for subsequent improvements. According to the analysis results of the error recognition pattern, the system will update or adjust the correlation coefficient in the hazard prediction algorithm. These adjustments may involve the weight of the risk factor, threshold setting, or optimization of pattern recognition parameters, aiming to improve the accuracy and robustness of the algorithm. After adjusting the algorithm, the system will use a predefined test set to verify the improved algorithm. By comparing the test results, the system can evaluate the effectiveness and accuracy of the algorithm improvement. After verification, the improved algorithm will be deployed in the actual system to improve the overall risk prediction capability. This implementation method can effectively identify and correct potential problems in the risk prediction algorithm by systematically collecting and analyzing user feedback. This feedback-driven algorithm optimization process helps to continuously improve the accuracy and response speed of the system, thereby enhancing the user experience and the overall security of the system. In addition, by verifying the improved algorithm, the system can ensure the reliability and effectiveness of the new algorithm and reduce the false positive rate and missed positive rate.

[0154] In actual applications, the way feedback data packets are collected can be adjusted according to the characteristics of different user groups, such as collecting feedback through mobile applications, online platforms, or smart devices. The validity verification mechanism can also be strengthened according to security requirements, adding multiple levels of verification steps, such as adding manual review or using machine learning models for anomaly detection. For algorithm adjustment, more advanced optimization methods can be introduced, such as adaptive algorithm optimization based on neural networks, to further improve the intelligence level of the algorithm. Finally, algorithm verification can introduce more diverse test sets, including data under different scenarios or boundary conditions, to ensure the robustness of the algorithm in various practical applications.

[0155] Next, the safety index S based on the current location is described:

[0156] The safety index S based on the current location is scored as follows:

[0157] Calculate the distance D from the known dangerous area based on the location information;

[0158] If D is less than the safety distance threshold DT, the danger level increases by α level;

[0159] If D is greater than or equal to the safety distance threshold DT, the danger level decreases by β levels;

[0160] The smaller the danger zone distance, the higher S is.

[0161] The process first calculates the distance D between the location and the known dangerous area based on the location information obtained by the children's smart watch. This step is mainly to evaluate the potential danger of the current area. The next step is to set a certain safety distance threshold DT. When the calculated distance D is lower than this safety threshold, it means that it is close to or in a relatively high-risk area. Therefore, the overall safety level is increased by α levels, where α is a positive integer to reflect a higher degree of danger; conversely, if the distance D is equal to or exceeds the threshold, it is considered that the child is currently in a relatively safe place, and the safety level is lowered by β levels to reflect this situation. β is a positive integer indicating a reduction to a lower risk state. Finally, based on the above two judgments, the safety index S tends to increase as the distance D decreases.

[0162] To demonstrate how this process is applied, an example is given in a hypothetical scenario: a child wearing a smartwatch is in a city park, and the safety distance threshold DT = 200 meters is set. If the watch confirms through GPS positioning that the location is 180 meters away from the nearest pond, that is, the distance D = 180 meters, then according to regulations, the danger level set in the watch will be increased by one level (i.e., increase α level) to prompt parents to pay attention; assuming that the child then walks out of the vicinity of the pond and the distance increases to 250 meters, the safety index in the watch setting will be reduced, and the danger level will drop by one level (i.e., reduce β level), indicating that it is now in a lower risk state. Such processing logic can provide timely feedback to guardians to help them understand the safety status of their children.

[0163] Next, the steps and examples of how S is higher as the distance to the danger zone decreases are described:

[0164] For each distance D, apply the formula S = max(5 × α × (D / DT) 2 ,1);

[0165] For each measurement result S, find the mean;

[0166] If the average result is less than 10%, it is considered normal and the alert level is not raised;

[0167] On the contrary, if the result is higher than 30%, it will be considered a high-risk area and an emergency warning will be issued.

[0168] For this process, the first step is to determine a formula to calculate the safety status S of a certain point. Assuming that there are multiple different monitoring points, and for the straight-line distance (D) between the location of each monitoring point and the dangerous area, the calculation formula used is S = max (5 × α × (D / DT) 2 ,1), where α and DT are predefined parameters or variables, α is used to adjust the value range of S, and DT is used as the distance threshold. For example, if α is 2 and DT is 40 meters, if the distance from a monitoring point to the nearest potential danger zone is 20 meters, then we can get: S = max(10×(20 / 40) 2 ,1)=max(10×0.25,1)=2.5.

[0169] Secondly, after obtaining the S values ​​of all locations, the data needs to be averaged. Assume that after the above calculation, the specific values ​​of the five monitoring points are 3.5, 2, 5, 3 and 2.5, and the average result is 3. It is worth noting here that if the monitoring area contains more measurement points, the same operation can be performed to obtain the average S value and analyze and judge it.

[0170] The state adjustment of the alarm threshold is determined again based on the average value. If the calculated average value does not reach 10% of the total reference value (assuming the total reference value is 100), it indicates that the environment is relatively safe, and it is determined to be normal and no alarm is triggered; on the contrary, when this value reaches 30% of the total (for example, the reference value is still set at 100, then more than 30 is in the warning state), it means that the current environment is very dangerous and action should be taken as soon as possible, such as starting a higher level of reminder measures to notify people around to be alert, just in case.

[0171] Next, we describe the application of formula S for each distance D. 3D =max(5×α×(D / DT) 2 ,1) In this part, it should be noted that this is to implement a calculation step in a danger assessment algorithm for a safety warning system for children's smart watches. The S mentioned here is the threat level or signal level determined according to the change in the distance between the children's smart watch and the dangerous area.

[0172] Record the calculated result R for evaluation;

[0173] If S is in the range of [5α,1], the calculation process is confirmed to be correct;

[0174] This calculation process is performed for each time period t and the results are compared to monitor changes;

[0175] Update the model parameter γ based on the changes in results for all time periods.

[0176] The various elements in the formula are explained as follows: S represents the safety assessment score, α represents the weight factor; D is the current distance to the danger zone; DT is a preset basic reference distance or threshold distance. The core of the entire calculation is to calculate the safety assessment score based on the dynamic distance and the distance change over the reference time, and to ensure that the score does not drop too low through the maximum function. Now let's explain this process step by step:

[0177] Record the calculation result R: After the current calculation formula is given, the first thing to do is to make a record of the actual output R of S, which is convenient for subsequent comparison, verification and subsequent processing. This step is to ensure the traceability and debuggability of the entire system.

[0178] Verify the validity of the results: Then, we need to confirm whether the obtained S is within the specified interval [5α, 1]. This is because within this range, the system can be highly responsive to truly threatening situations while avoiding invalid alarms in irrelevant situations, thus ensuring the rationality and stability of the system.

[0179] Continuous time period monitoring and data collection: Then in actual operation, the change of distance D at different time points is detected and evaluated in real time, the formula calculation of S is executed at each time point t, and the data obtained each time is stored, so as to establish a complete sequence of behavioral data of children's watches in different time periods.

[0180] Tune model parameters based on results: Finally, conduct an overall analysis based on all evaluation data results. If the overall performance of the model is poor, the probability of misjudgment is high, or there is a significant deviation from expectations, the value of parameter γ needs to be adjusted. By reviewing and analyzing past calculation history to guide the logical design of adjusting model parameters, the early warning mechanism can be closer to perfection.

[0181] For example, a child's smartwatch is 10 meters away from the preset potential risk location. At this time, α is set to 0.5, and DT is set to 20m (which means that if the distance to the area exceeds 20m, it is considered very far). Then the current D / DT = 0.5. Therefore, 5α×(0.5) 2 =5×0.5×0.25=0.625 is less than 1. According to max(5×α×(D / DT)2,1), the result R here will be set to 1. Assuming that this time is 8 o'clock in the morning, this R will be recorded to confirm that it is between the allowed [0.5 (because α=0.5, 5×0.5=2.5, that is, 2.5×0.25=0.625 is less than or equal to 2.5, which meets [5α,1]),1]; at 6 o'clock in the evening, when D becomes 2 meters, R changes to max(5×0.5×(2 / 20) 2 ,1)=1, and it is still valid; finally, based on the change of the result R in these two time periods, the change of the child’s position is analyzed to decide whether γ needs to be adjusted, such as reducing γ to make the watch more sensitive.

[0182] Next, the process of updating the model parameter γ based on the changes in results in all time periods and its specific examples are described.

[0183] Perform normalization processing N(SD) on the risk score SD of all time periods to obtain NSD;

[0184] Calculate the mean value of the change of the normalized score NSD, mean△(NSD), and compare it with a preset threshold threshold. If (mean△(NSD)≥threshold), then γ=γ+1;

[0185] Adjust the specific value of γ according to whether the conditions are met to enhance dynamic adaptability;

[0186] The final adjusted new prediction parameter combination Λ includes {γ, D, S}.

[0187] For this step, it mainly makes an iterative improvement to the existing model parameters to ensure that the system can adapt to new or changing situations. First, we need to normalize the danger score (SD) of the children's smartwatch during all data collection time periods to obtain NSD. This is a consistency operation performed to ensure the comparability of data in all time periods, facilitating subsequent analysis and parameter adjustment. For example, if the original score range is very wide (e.g., between 0 and 200), then this range difference will affect the effectiveness of the analysis; by mapping it to the interval from 0 to 1, such a problem is eliminated.

[0188] The standard score obtained after normalizing the danger score SD (N(SD)) is NSD. Subsequently, based on the fluctuation of the normalized score NSD over a period of time, it is decided whether to modify the model parameter γ. Here, it involves evaluating whether the change amount of the mean of NSD (meanΔ(NSD)) is greater than or equal to a preset threshold (threshold). If the condition holds (i.e., the change reaches a certain degree), then the model parameter γ is updated to the next value (γ + 1).

[0189] For example, assume we find that NSD has increased by an average of 0.3 units, and the preset threshold is 0.2. Then, because 0.3 > 0.2 (meeting the condition if(meanΔ(NSD) ≥ threshold)), this means that there has been a meaningful change in the level of danger in the environment, and the system believes it is necessary to raise the vigilance level. Therefore, γ changes from the original 1 to γ + 1. If the condition is not met (such as meanDelta < threshold), then γ remains unchanged.

[0190] Finally, after the above adjustments, the final new prediction parameter γ is updated as a part of the adjusted parameter combination Λ. The Λ parameter combination also includes elements such as the distance parameter D and the danger score threshold parameter S to ensure that the most appropriate danger prediction can still be provided in the latest situation. Overall, this algorithm helps to ensure that the system can make timely adjustments according to the real-time changes in the environment and danger situations.

[0191] The present invention aims to provide an efficient danger warning method, especially for applications on smartwatches designed for children, to ensure that children can be continuously and effectively monitored in a complex environment, accurately identify various potential hazard sources, and take corresponding warning measures in a timely manner. The method is specifically implemented through the following steps: First, the smartwatch is built-in with a variety of environmental sensors and a data processing module, enabling it to continuously and real-time collect data from the surrounding environment. This step ensures the effective acquisition of environmental information.

[0192] The collected data is then passed to the hazard prediction model pre-installed in the device. The model is trained by learning from a large number of existing cases and can effectively analyze and determine the potential safety issues currently faced, and then assess the hazard level. This mechanism overcomes the problems of false positives and false negatives in complex scenarios, greatly improves the accuracy of hazard assessment and classification, reduces unnecessary warnings, and avoids missing important safety hazard prompts.

[0193] When the model detects and confirms a certain level of potential danger signals, the specific warning content will be quickly sent to the associated related devices such as the parent's mobile phone, in order to inform the guardian immediately. This fast and targeted communication strategy allows parents or relevant persons in charge to intervene in possible situations at the first time, building a more solid barrier for the safety of children's life and property; in addition, based on whether the corresponding dangerous information actually exists, the prediction algorithm can be reversely optimized and improved, so that the accuracy of the entire warning system is continuously improved, and the possibility of misjudgment is gradually reduced.

[0194] This danger warning mechanism can better cope with the ever-changing and complex outdoor scenes in reality, and effectively solves the problem of accurate warning of smart watches under complex conditions. In general, the smart watch collects surrounding environmental data in real time → uses the pre-trained danger prediction model based on the environmental data to intelligently identify and evaluate the level → quickly sends an early warning to the associated terminal receiving end after the danger occurs, and adjusts and improves the system based on the actual effectiveness feedback. These steps are closely combined to form a complete solution, so that different types of threats can be responded to in a timely and correct manner, and continuous iteration is carried out to ensure the improvement of the algorithm effect, truly realizing the implementation of the technical goals.

[0195] A danger warning system for a children's smart watch is also provided, which is used to implement the steps of the danger warning method for a children's smart watch, and the system includes:

[0196] Data acquisition module, used for smart watches to collect surrounding environment data in real time;

[0197] The analysis and identification module uses pre-trained hazard prediction models to analyze and identify potential hazard types and levels based on environmental data;

[0198] Warning push module: when the analysis and identification module identifies a specific danger, the corresponding warning information is triggered and pushed to the associated equipment;

[0199] The model adjustment module is used to adjust the risk prediction model parameters according to the received feedback information to improve the recognition accuracy.

[0200] The core working principle of the system is to collect environmental data in real time through smart watches and analyze and identify them using pre-trained danger prediction models. The data acquisition module is responsible for obtaining environmental data around the watch, such as sound, light, temperature, location, etc. Based on these environmental data, the analysis and identification module detects possible dangers through the danger prediction model and determines the type and level of danger. When the warning push module detects danger, it immediately pushes the warning information to the device associated with the smart watch, such as the parent's mobile phone or monitoring platform, to ensure timely measures. The model adjustment module adjusts the parameters of the danger prediction model according to user feedback to improve the recognition accuracy and reliability of the system. The danger warning system can monitor the environment in which children are located in real time, identify potential dangers in a timely manner, and provide effective warning information, which greatly improves the safety of children. By continuously adjusting and optimizing the danger prediction model, the recognition accuracy of the system can be gradually improved, reducing the possibility of false alarms and missed alarms, and ensuring the timeliness and accuracy of the warning information. In addition, the system can be linked with a variety of devices to achieve all-round child safety monitoring, and improve the attention and protection capabilities of families and society for children's safety.

[0201] In this embodiment, the data acquisition module can collect more types of data, such as air quality, humidity, ultraviolet intensity, etc., according to different environmental requirements to meet more complex environmental monitoring needs. In addition, the danger prediction model of the analysis and identification module can adopt different machine learning algorithms, such as deep learning, support vector machine, etc., to improve the analysis ability and accuracy of the model. The push method of the early warning push module can also be diversified, such as through text messages, APP notifications, and even voice reminders to ensure the timeliness of information transmission. The model adjustment module can flexibly adjust the model parameters according to the feedback information of different users, or use adaptive algorithms to automatically optimize the model to adapt to different environments and individual needs.

[0202] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A danger warning method for a children's smart watch, characterized in that: The method comprises: S1: Smartwatch collects surrounding environment data in real time; S2: Based on environmental data, use the pre-trained hazard prediction model to analyze and identify potential hazard types and levels; S3: When a specific danger is identified, the corresponding warning information is triggered and pushed to the associated devices; S4: Adjust the hazard prediction model parameters according to the received feedback to improve the recognition accuracy.

2. A danger warning method for a children's smart watch according to claim 1, characterized in that: The step S1 specifically includes: Start the sensor module to obtain image data; Start the sound collection module to receive audio signals; Read geographic location and time tag information; The sensor module transmits synchronously to the data processing unit; All the acquired raw data are encrypted and uploaded to the cloud server for preliminary analysis.

3. A danger warning method for a children's smart watch according to claim 1, characterized in that: The step S2 specifically includes: Get the noise level L in the environment; Get the difference between the ambient temperature T and the preset temperature T0 |T-T0|; Scoring based on the safety index S of the current location; An additional weighting factor W is recorded when the ambient lighting level E is above a predetermined safety value.

4. A danger warning method for a children's smart watch according to claim 1, characterized in that: The step S3 specifically includes: Constructing alarm message content according to the recognition result; Customize push rules to filter the list of target associated devices to be pushed; Add security verification mechanism to push messages to ensure privacy protection; Select an instant messaging channel to send warning message text and sound prompts; Determine whether additional emergency procedures need to be initiated and responded to; The target associated device list to be pushed by the custom push rule is sorted by the following formula to prioritize the associated devices: P d =β1×E c +β2×L d -β3×T d , Among them, P d Represents the priority of the associated device, E c is the physical distance between the device and the child, L d is the average response time of the device in the push history, T d is the power state of the device, β1, β2, and β3 are adjustment coefficients; The security verification mechanism is added to the push message to ensure privacy protection. The warning message is encrypted and processed. The receiving device needs to pass the identity authentication before decrypting the message. Only the verified device can successfully decrypt and read the message content. The selecting of an instant messaging channel to send a warning message text and a sound prompt comprises: The optimal communication channel is selected using the following formula: C s =argmin(γ1×T c +γ2×R b -γ3×P r ), Among them, T c is the transmission delay of the channel, R b is the bandwidth utilization, P r is the transmission success rate of the channel, and γ1, γ2, and γ3 are the weight coefficients of channel selection.

5. The danger warning method for a children's smart watch according to claim 1, characterized in that: The step S4 specifically includes: Collect feedback data packages from users, including hazard handling records and personal assessments; Verify the validity of feedback information to prevent malicious interference; Analyze the error recognition patterns and causes of effective feedback extraction; Update or adjust correlation coefficients in risk prediction algorithms; Verify the effectiveness and accuracy of the improved algorithm on the test set and deploy it.

6. A danger warning method for a children's smart watch according to claim 3, characterized in that: The step of scoring based on the safety index S of the current location specifically includes: Calculate the distance D from the known dangerous area based on the location information; If D is less than the safety distance threshold DT, the danger level increases by α level; If D is greater than or equal to the safety distance threshold DT, the danger level decreases by β levels; The smaller the danger zone distance, the higher S is.

7. A danger warning method for a children's smart watch according to claim 6, characterized in that: The steps of: the smaller the distance to the dangerous area, the higher S is; For each distance D, apply the formula S = max(5 × α × (D / DT) 2 ,1); For each measurement result S, find the mean; If the average result is less than 10%, it is considered normal and the alert level is not raised; On the contrary, if the result is higher than 30%, it will be considered a high-risk area and an emergency warning will be issued.

8. A danger warning method for a children's smart watch according to claim 7, characterized in that: The step of applying the formula S=max(5×α×(D / DT)2,1) to each distance D comprises: Record the calculated result R for evaluation; If S is in the range of [5α,1], the calculation process is confirmed to be correct; This calculation process is performed for each time period t and the results are compared to monitor changes; Update the model parameter γ based on the changes in results for all time periods.

9. A danger warning method for a children's smart watch according to claim 8, characterized in that: The step of updating the model parameter γ based on the result changes of all time periods includes: Perform normalization processing NSD on the risk score SD of all time periods to obtain NSD; Calculate the mean value of the change of the normalized score NSD, mean△(NSD), and compare it with a preset threshold threshold. If (mean△(NSD)≥threshold), then γ=γ+1; Adjust the specific value of γ according to whether the conditions are met to enhance dynamic adaptability; The final adjusted new prediction parameter combination Λ includes {γ, D, S}.

10. A danger warning system for a children's smart watch, used to implement the steps of a danger warning method for a children's smart watch as described in any one of claims 1 to 9, characterized in that: The system comprises: Data acquisition module, used for smart watches to collect surrounding environment data in real time; The analysis and identification module uses pre-trained hazard prediction models to analyze and identify potential hazard types and levels based on environmental data; Warning push module: when the analysis and identification module identifies a specific danger, the corresponding warning information is triggered and pushed to the associated equipment; The model adjustment module is used to adjust the risk prediction model parameters according to the received feedback information to improve the recognition accuracy.

Images