Automatic unlocking method and system of intelligent lock

Abstract

The invention relates to the technical field of intelligent locks, in particular to an automatic unlocking method and system of an intelligent lock. The starting method comprises the following steps: acquiring biological electric field data of a user; constructing a user emotion model based on the biological electric field data; acquiring environment electromagnetic field intensity data; evaluating an environmental safety coefficient according to the environmental electromagnetic field intensity data; fusing the user emotion model and the environment safety coefficient to generate unlocking probability distribution; dynamically adjusting the unlocking difficulty of the intelligent lock based on the unlocking probability distribution; when the intelligent lock is touched or awakened, unlocking operation is executed according to the current unlocking difficulty. According to the automatic unlocking method and system of the intelligent lock, the biological electric field data of the user and the environmental electromagnetic field intensity data are collected, and the technical problem that a traditional intelligent lock is insufficient in capacity of automatically sensing the user state and the environmental safety is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart locks, and in particular to a method and system for automatically opening a smart lock. Background Art

[0002] With the rapid development of smart home technology, smart locks have become an important part of modern home security systems. Traditional smart locks mainly rely on passwords, fingerprints or mobile phone apps for identity authentication and unlocking operations. However, these methods often require users to take the initiative, which may cause inconvenience in some cases, such as when users are carrying heavy objects with both hands. In addition, traditional smart lock systems usually do not consider the user's emotional state and the safety of the surrounding environment, which may lead to misoperation or safety hazards in special circumstances.

[0003] Based on the above background, the current technical problem facing the smart home field is: how to develop a smart lock system that can automatically sense the user status and environmental security, and intelligently adjust the difficulty of unlocking accordingly. Summary of the invention

[0004] The present invention provides a smart lock automatic opening method and system to solve the technical problem that the traditional smart lock system has insufficient ability to automatically sense the user status.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] On the one hand, a method for automatically opening a smart lock is provided, the steps of the opening method comprising:

[0007] Obtaining the user's bioelectric field data;

[0008] Building a user emotion model based on the bioelectric field data;

[0009] Collect environmental electromagnetic field strength data;

[0010] Evaluate the environmental safety factor according to the environmental electromagnetic field strength data;

[0011] The user emotion model and the environmental safety factor are integrated to generate a probability distribution of unlocking;

[0012] Dynamically adjusting the difficulty of opening the smart lock based on the unlocking probability distribution;

[0013] When the smart lock is touched or awakened, the unlocking operation is performed according to the current opening difficulty.

[0014] On the other hand, a smart lock automatic opening system is provided, the opening system comprising:

[0015] A bioelectric field acquisition module is used to obtain the user's bioelectric field data;

[0016] An emotion modeling module, used to construct a user emotion model based on the bioelectric field data;

[0017] Electromagnetic field detection module, used to collect environmental electromagnetic field strength data;

[0018] A safety assessment module, used to assess an environmental safety factor according to the environmental electromagnetic field strength data;

[0019] A probability distribution generation module, used to integrate the user emotion model and the environmental safety factor to generate an unlocking probability distribution;

[0020] A difficulty adjustment module, used to dynamically adjust the opening difficulty of the smart lock based on the unlocking probability distribution;

[0021] The execution module performs the unlocking operation according to the current opening difficulty when the smart lock is touched or awakened.

[0022] The beneficial effects of the present invention are:

[0023] The automatic opening method and system of the smart lock provided by the present invention collects the user's bioelectric field data and environmental electromagnetic field strength data, constructs a user emotional model and evaluates the environmental safety factor, thereby generating an unlocking probability distribution and dynamically adjusting the opening difficulty of the smart lock, effectively solving the technical problem that the traditional smart lock has insufficient ability to automatically perceive the user status and environmental safety. The method can realize automatic perception and intelligent adjustment of the smart lock, and unlocking can be completed without the user's active operation. At the same time, by considering the user's emotional state and environmental safety, the system can ensure safety to the greatest extent while ensuring convenience. In addition, the present invention also introduces a resonant frequency detection mechanism of the bioelectromagnetic field and the environmental electromagnetic field, as well as a long-term dynamic update mechanism of the user's emotional model, which further improves the performance of the smart lock. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The process of the automatic opening method of the smart lock in one embodiment of the present invention is as follows Figure 1 ;

[0025] Figure 2 This is a scene diagram of a method for automatically opening a smart lock in one embodiment of the present invention;

[0026] Figure 3 This is a distribution diagram of unlocking probabilities of a smart lock under different security risks in one embodiment of the present invention;

[0027] Figure 4 A diagram of a dynamic adjustment mechanism for opening difficulty of a smart lock in one embodiment of the present invention;

[0028] Figure 5It is a time series diagram of the electromagnetic field strength of the environment in which the smart lock is located in one embodiment of the present invention;

[0029] Figure 6 A flowchart of constructing a user emotion model in one embodiment of the present invention;

[0030] Figure 7 A flow chart of evaluating an environmental safety factor in one embodiment of the present invention;

[0031] Figure 8 A flowchart of calculating an environmental safety factor in one embodiment of the present invention;

[0032] Fig. 9 A flow chart of generating unlock probability distribution in one embodiment of the present invention;

[0033] Fig.10 A flowchart of adjusting the difficulty of opening a smart lock in one embodiment of the present invention;

[0034] Fig.11 The process of the automatic opening method of the smart lock in one embodiment of the present invention is as follows Figure 2 ;

[0035] Fig.12 This is a block diagram of a smart lock automatic opening system in one embodiment of the present invention;

[0036] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0037] 100. Bioelectric field acquisition module; 200. Emotion modeling module; 300. Electromagnetic field detection module; 400. Safety assessment module; 500. Probability distribution generation module; 600. Difficulty adjustment module; 700. Execution module. DETAILED DESCRIPTION

[0038] The present invention provides the following preferred embodiments:

[0039] Embodiment 1

[0040] In order to solve the problem that the traditional smart lock system cannot automatically sense the user status and environmental safety, this embodiment proposes a smart lock automatic opening method. The method obtains the user's bioelectric field data, builds a user emotion model, and evaluates the environmental safety factor in combination with the environmental electromagnetic field strength data, thereby achieving dynamic adjustment of the difficulty of opening the smart lock.

[0041] like Figures 1 to 5 As shown, the steps of the smart lock opening method include:

[0042] S101. Obtaining bioelectric field data of a user.

[0043] S102. Construct a user emotion model based on bioelectric field data.

[0044] S103, collecting environmental electromagnetic field strength data.

[0045] S104. Evaluate the environmental safety factor based on the environmental electromagnetic field strength data.

[0046] S105: Integrate the user emotion model and the environmental safety factor to generate an unlocking probability distribution.

[0047] S106. Dynamically adjust the opening difficulty of the smart lock based on the unlocking probability distribution.

[0048] S107. When the smart lock is touched or awakened, an unlocking operation is performed according to the current opening difficulty.

[0049] Furthermore, the user's bioelectric field data can be obtained through a non-invasive bioelectric field sensor. The sensor can be integrated into the housing or door handle of the smart lock. When the user approaches or touches the smart lock, the sensor can capture the user's bioelectric field signal. These signals may include but are not limited to bioelectric characteristics such as skin potential, electrocardiogram waveform or brain wave frequency.

[0050] It should be understood that building a user emotion model based on bioelectric field data is a complex process. This embodiment uses machine learning algorithms, such as support vector machines (SVM) or deep neural networks (DNN), to extract features and perform pattern recognition on the acquired bioelectric field data. By analyzing the time-frequency characteristics and nonlinear dynamic characteristics of the features, a multidimensional emotional state space can be constructed to accurately locate the user's current emotional state.

[0051] Furthermore, the collection of environmental electromagnetic field strength data can be achieved through an electromagnetic field strength sensor integrated in the smart lock system. The sensor can monitor the changes in the electromagnetic field strength of the surrounding environment in real time, including natural electromagnetic fields and man-made electromagnetic interference. The evaluation of the environmental safety factor is based on the statistical characteristics of these electromagnetic field data and the results of spectrum analysis.

[0052] It is understandable that the process of integrating the user emotion model and the environmental safety factor to generate the unlocking probability distribution can use methods such as Bayesian reasoning or fuzzy logic to map the user's emotional state and the environmental safety assessment results to a continuous probability distribution. The unlocking probability distribution reflects the possibility of performing the unlocking operation under the current conditions.

[0053] Furthermore, the difficulty of opening the smart lock is dynamically adjusted based on the probability distribution of unlocking to achieve adaptive security control. This embodiment adopts a dynamic threshold adjustment algorithm to adjust the verification steps or verification strength required for unlocking in real time according to the characteristics of the probability distribution (such as peak value, variance, etc.). For example, when the probability distribution presents a high peak value and low variance, the system may reduce the difficulty of opening, otherwise it will increase the difficulty.

[0054] It is important to understand that when the smart lock is touched or awakened, the system will perform the unlocking operation based on the current unlocking difficulty. This may include adjusting the required biometric matching accuracy, increasing or decreasing the verification steps, or changing the verification time window. This dynamic adjustment mechanism ensures that the system can maintain an appropriate balance between security and convenience in different situations.

[0055] The benefit of this embodiment is that by comprehensively considering the user's emotional state and environmental safety factors, the smart lock opening process is automated and intelligent. This not only improves the efficiency of unlocking, but also enhances the security and adaptability of the smart lock system. At the same time, the non-intrusive and real-time nature of the method enables it to continuously optimize the performance of the smart lock without affecting the normal use of the user.

[0056] Embodiment 2

[0057] In order to improve the accuracy and reliability of the automatic opening method of the smart lock, this embodiment further refines the collection content of the bioelectric field data, including the user's skin potential, electrocardiogram waveform and brain wave frequency. These bioelectric signals can more comprehensively reflect the user's physiological and psychological state, thereby providing a rich data foundation for building an accurate user emotion model.

[0058] Furthermore, the measurement of the user's skin potential can be achieved by integrating a microelectrode array on the surface of the smart lock. These electrodes can be made of conductive polymer materials with good biocompatibility and conductivity. When the user touches the smart lock, the electrodes can capture the slight potential changes on the skin surface, which are closely related to the user's emotional state and stress level. The skin potential signal is usually in the frequency range of 0.01-1Hz and the amplitude is between 0.1-10mV. The influence of environmental noise needs to be considered during the signal acquisition process, and low-pass filters and differential amplifiers can be used to improve the signal quality.

[0059] It should be understood that regarding the acquisition of the ECG waveform, this embodiment uses a capacitive coupling type non-contact ECG sensor, which can be embedded in the handle or front panel of the smart lock. This sensor can capture the weak electrical signals generated by the electrical activity of the heart through the capacitive coupling effect without the user directly contacting the electrodes. The frequency range of the ECG signal is usually between 0.05-150Hz, and the amplitude is about 0.1-5mV. In order to extract a valid ECG waveform, the system uses an adaptive filtering algorithm and independent component analysis (ICA) technology to eliminate motion artifacts and power supply interference.

[0060] Furthermore, the measurement of brain wave frequency takes into account the limitations of traditional electroencephalogram (EEG) equipment, and this embodiment adopts a new type of non-invasive prefrontal EEG sensing technology. This technology uses a special electrode array installed on the top or side of the smart lock to infer brain electrical activity by measuring the changes in the electric field in the prefrontal region. Brain wave signals are usually divided into different frequency bands, including delta waves (0.5-4Hz), theta waves (4-8Hz), alpha waves (8-13Hz), beta waves (13-30Hz) and gamma waves (30-100Hz). Each frequency band is associated with a specific brain state and cognitive process.

[0061] It is understandable that the combined analysis of these three types of bioelectric field data can provide more comprehensive information about the user's status. For example, sudden changes in skin potential may indicate emotional fluctuations, heart rate variability in the electrocardiogram can reflect stress levels, and changes in brain wave frequency may indicate attention status or cognitive load. By fusing these data, the system is able to build a multi-dimensional model of the user's emotional and physiological status.

[0062] Furthermore, in order to process these complex bioelectric signals, this embodiment uses signal processing technology and machine learning algorithms. First, the original signal is analyzed in time and frequency using wavelet transform to extract feature vectors. Then, these features are classified and pattern recognized using deep learning networks, such as long short-term memory networks (LSTM) or convolutional neural networks (CNN). This allows real-time and accurate recognition of the user's emotional state.

[0063] It should be understood that privacy protection is an issue that cannot be ignored during the collection and analysis of bioelectric field data. This embodiment uses data anonymization and local processing technology to ensure that the user's bioelectric information will not be leaked or abused. All data processing and analysis are completed on the local chip of the smart lock, and only the final emotion model results will be used for subsequent unlocking decisions.

[0064] The benefit of this embodiment is that by collecting and analyzing a variety of bioelectric field data, the accuracy of the user's emotional model is significantly improved. This multimodal bioelectric signal fusion method can not only capture subtle changes in the user's emotional state, but also adapt to individual differences under different users and environmental conditions. At the same time, the non-invasive data collection method ensures the convenience of user use, allowing the smart lock system to continuously optimize its performance and safety without affecting the normal use of the user.

[0065] Embodiment 3

[0066] In order to solve the problem of insufficient accuracy of traditional emotion recognition methods when processing complex bioelectric field data, this embodiment further optimizes the method of building a user emotion model based on bioelectric field data. This method achieves accurate modeling and recognition of user emotions by extracting time-frequency features, analyzing data complexity, building a multidimensional emotion state space, and locating the user's current emotion state.

[0067] like Figure 6 As shown, the steps of constructing a user emotion model based on bioelectric field data include:

[0068] S301, extracting time-frequency characteristics of bioelectric field data.

[0069] S302. Analyze the complexity of bioelectric field data using nonlinear dynamics methods.

[0070] S303. Construct a multidimensional emotional state space based on time-frequency characteristics and complexity analysis results.

[0071] S304: Locate the user's current emotional state in the multi-dimensional emotional state space.

[0072] Furthermore, the time-frequency feature extraction of bioelectric field data is the basis for building a user emotion model. This embodiment uses wavelet packet transform (WPT) technology to decompose the original bioelectric field signal. WPT can provide high resolution in both time and frequency domains, and is suitable for analyzing non-stationary signals. Specifically, using the Daubechies wavelet basis function, the signal is decomposed into 5 levels, and each level produces 2^5=32 sub-bands. Statistical features such as energy, entropy, and standard deviation are extracted from these sub-bands to form an initial feature vector.

[0073] It should be understood that after the time-frequency features are extracted, feature selection is required to reduce computational complexity and improve model performance. This embodiment uses a recursive feature elimination (RFE) algorithm, combined with a support vector machine (SVM) classifier, to select the most recognizable feature subset from the initial feature set. This process determines the optimal number of features through cross-validation to ensure that the selected features can fully characterize the signal characteristics without causing overfitting of the model.

[0074] Furthermore, the use of nonlinear dynamics methods to analyze the complexity of bioelectric field data is an innovative point of this embodiment. The multiscale entropy (MSE) analysis method is used here. MSE can evaluate the complexity of signals at different time scales, and is suitable for capturing long-range correlations and multiscale dynamic characteristics in bioelectric field data. In specific implementation, the original signal is first coarse-grained to generate time series of 20 different scales. Then the sample entropy (Sample Entropy) at each scale is calculated to obtain a curve of entropy value changing with scale. This curve reflects the complexity changes of the signal at different time scales, providing an additional dimension for the identification of emotional states.

[0075] It is understandable that a multidimensional emotional state space is constructed based on time-frequency features and complexity analysis results. This embodiment adopts a method combining principal component analysis (PCA) and t-distributed random neighbor embedding (t-SNE). First, PCA is used to perform preliminary dimensionality reduction on high-dimensional features, retaining the principal components that explain 95% of the variance. Then, the features after PCA dimensionality reduction are input into the t-SNE algorithm to further map the data to a three-dimensional space. t-SNE can maintain the local structure of the data and is suitable for visualizing high-dimensional data. The three-dimensional emotional state space constructed in this way can intuitively represent the relationship and distribution between different emotional states.

[0076] Furthermore, the current emotional state of the user is located in the multidimensional emotional state space to realize real-time emotion recognition. This embodiment adopts a Gaussian Mixture Model (GMM) to perform probabilistic modeling of the emotional state space. First, the GMM is trained using the Expectation-Maximization (EM) algorithm to determine the optimal number of Gaussian components. Then, for the newly input bioelectric field data, the features are extracted and mapped to the emotional state space through the aforementioned steps, and the posterior probability of belonging to each Gaussian component is calculated. Finally, the user's current emotional state is determined based on the distribution of the posterior probability.

[0077] It should be understood that in order to improve the accuracy of emotion recognition, this embodiment also introduces the processing of time series information. The Long Short-Term Memory (LSTM) network is used to model the continuous sequence of emotional states to capture the temporal dynamic characteristics of emotional changes. The input of the LSTM network is the probability distribution of emotional states obtained in the previous steps, and the output is the estimate of the emotional state after time series smoothing. This method can effectively reduce the impact of instantaneous noise and provide more stable and coherent emotion recognition results.

[0078] The benefit of this embodiment is that by combining time-frequency analysis, nonlinear dynamics and deep learning technology, a comprehensive analysis and accurate modeling of bioelectric field data is achieved. This multi-angle and multi-scale analysis method can capture the subtle differences in emotional expression and adapt to emotional changes under different individuals and environmental conditions. At the same time, the constructed multi-dimensional emotional state space provides an intuitive tool for the visualization and interpretation of emotions, which helps to improve the interpretability of the system and user trust.

[0079] Embodiment 4

[0080] In order to solve the problem that the traditional emotion model construction method lacks a unified mathematical expression, this embodiment constructs a formula for the user emotion model. The formula achieves accurate description and construction of the multi-dimensional emotion state space by integrating time-frequency features, complexity indicators and nonlinear mapping functions.

[0081] Specifically, the construction formula of the user emotion model is:

[0082] EM=f(TF,CD,λ)

[0083] Among them, EM is the emotion model, TF is the time-frequency feature vector, CD is the complexity index, λ is the nonlinear mapping function, and f is the multidimensional state space construction function. This formula provides a unified framework that integrates different types of features and analysis methods into one model.

[0084] It should be understood that the time-frequency feature vector TF is the basic input of the model. This embodiment uses the wavelet packet transform (WPT) combined with the empirical mode decomposition (EMD) method to extract time-frequency features. Specifically, the bioelectric field signal is first decomposed by WPT to obtain sub-signals in different frequency bands. Then, EMD is applied to each sub-signal to extract its intrinsic mode function (IMF). Finally, the energy, instantaneous frequency, Hilbert spectrum and other features of each IMF are calculated to form a TF vector. This method can effectively capture the time-varying characteristics and frequency distribution of the signal.

[0085] Furthermore, the calculation of the complexity index CD adopts the Multiscale Permutation Entropy (MPE) method. MPE can evaluate the complexity and irregularity of time series at different time scales. During the calculation process, the original signal is first coarsened to generate multiple time series at different scales. Then, the permutation entropy is calculated for each scale sequence. Finally, a curve reflecting the complexity change of the signal at different time scales is obtained, and the characteristic parameters of the curve (such as the area under the curve, slope, etc.) constitute the CD index.

[0086] It should be understood that this embodiment uses a deep autoencoder to achieve nonlinear mapping. The encoder part of the autoencoder consists of a multi-layer fully connected neural network, which is used to map high-dimensional input features to a low-dimensional latent space. The decoder part attempts to reconstruct the original input from the latent space. By minimizing the reconstruction error, the autoencoder learns a nonlinear representation of the data. The λ function actually corresponds to the trained encoder part, which can capture the complex nonlinear relationship between features.

[0087] Furthermore, a multidimensional state space function f is constructed. This embodiment uses a variational autoencoder (VAE) to implement the construction function f. VAE can not only learn the low-dimensional representation of data, but also generate latent variables that conform to a specific distribution. In the specific implementation, the encoder part of VAE outputs mean and variance parameters to define the Gaussian distribution in the latent space. The decoder samples from this distribution and reconstructs the input data. By minimizing the reconstruction error and KL divergence, VAE learns a continuous and structured emotional state space.

[0088] It should be understood that in order to improve the generalization ability of the model, this embodiment introduces an adversarial training mechanism. Specifically, a discriminator network is added to the VAE to form an Adversarial Variational Autoencoder (AVAE) structure. The discriminator attempts to distinguish the distribution of real data and generated data, while the generator (ie, the decoder of VAE) strives to generate samples that can deceive the discriminator. This adversarial process helps to generate more realistic and diverse emotional state representations.

[0089] Furthermore, in order to process the temporal dynamic characteristics of emotional states, this embodiment integrates a recurrent neural network (RNN) unit on the basis of AVAE. Specifically, an LSTM layer is introduced in the latent space to model the temporal dependency between continuous emotional states. In this way, the model can not only capture the emotional state at a single time point, but also consider the evolution of emotions over time, providing a more coherent and stable emotional model output.

[0090] It is understandable that the training process of the model is carried out in an end-to-end manner. First, the bioelectric field dataset with emotion annotations is used for pre-training to optimize the parameters of each component. Then, through the online learning mechanism, the model can continuously adapt to the emotional expression pattern of a specific user. The training objective function includes the weighted sum of reconstruction error, KL divergence, adversarial loss, and cross entropy loss for emotion classification. By balancing these loss terms, the model is able to simultaneously learn the low-dimensional representation of the data, the generation ability, and the emotion recognition ability.

[0091] The benefit of this embodiment is that, by providing a unified mathematical framework, it achieves accurate description and construction of complex emotion models. This method not only integrates multiple technologies such as time-frequency analysis, complexity evaluation and nonlinear mapping, but also introduces generative models and adversarial training mechanisms to improve the expressiveness and generalization of emotion models. At the same time, the interpretability and visualization capabilities of the model are also enhanced, providing a more reliable and transparent theoretical basis for the emotion perception function of the smart lock system.

[0092] Embodiment 5

[0093] In order to solve the problem of insufficient accuracy of traditional environmental safety assessment methods when dealing with complex electromagnetic environments, this embodiment further optimizes the method of evaluating environmental safety factors based on environmental electromagnetic field strength data. This method achieves accurate analysis and safety assessment of the electromagnetic environment around the smart lock by calculating entropy values, analyzing spectrum characteristics, detecting abnormal interference patterns, and comprehensively evaluating environmental safety factors.

[0094] like Figure 7 As shown, the steps of evaluating the environmental safety factor based on the environmental electromagnetic field strength data include:

[0095] S501. Calculate the entropy value of the ambient electromagnetic field strength.

[0096] S502: Analyze the frequency spectrum characteristics of the electromagnetic field intensity.

[0097] S503: Detect abnormal electromagnetic interference patterns.

[0098] S504. Evaluate the environmental safety factor based on the entropy value, spectrum characteristics and abnormal interference pattern.

[0099] Furthermore, the calculation of the entropy value of the ambient electromagnetic field strength is the basis for evaluating the complexity of the environment. This embodiment adopts the multiscale sample entropy (MSE) method to calculate the entropy value. First, the sliding window technology is used to coarse-grained the original electromagnetic field strength data to generate time series of different time scales. Then, the sample entropy is calculated for each scale sequence. In the specific calculation process, the embedding dimension m=2 is set, and the similarity threshold r is 0.15 times the standard deviation of the original sequence. By calculating the sample entropy of 20 different time scales, a curve showing the change of entropy value with scale is obtained. This curve reflects the complexity changes of the electromagnetic environment at different time scales, and provides an important basis for subsequent safety assessments.

[0100] It should be understood that the spectrum feature analysis is used to identify the characteristics of the electromagnetic environment. This embodiment uses wavelet packet transform (WPT) combined with adaptive spectrum estimation technology to analyze the spectrum characteristics of the electromagnetic field intensity. First, the signal is decomposed into 6 layers of WPT using the Daubechies wavelet basis function to obtain sub-signals of 64 frequency bands. Then, a multi-window spectrum estimation method, such as the Thomson multi-window method, is applied to each sub-signal to obtain a high-resolution and low-variance power spectrum density estimate. Feature parameters are extracted from the obtained power spectrum, including the energy distribution of each frequency band, the amplitude and phase information of the main frequency components, the number and position of spectrum peaks, etc. These spectrum features can reflect the frequency composition and dynamic characteristics of the electromagnetic environment.

[0101] Furthermore, the detection of abnormal electromagnetic interference patterns is used to ensure environmental safety. Specifically, a hybrid architecture of a long short-term memory network (LSTM) and a convolutional neural network (CNN) can be used. The long short-term memory network (LSTM) is used to capture the temporal dependency of electromagnetic signals, while the convolutional neural network (CNN) is used to extract local spatial features. The input of the network is the electromagnetic field intensity data in the sliding window and the spectrum characteristics obtained in the previous steps. By training the model on normal electromagnetic environment data, it learns the distribution of normal patterns. In the detection stage, the reconstruction error between the input data and the model prediction is calculated. When the error exceeds the preset threshold, it is determined to be an abnormal interference pattern. In order to improve the reliability of detection, integrated learning technology can also be introduced to combine the outputs of multiple basic models for decision making.

[0102] It is understandable that evaluating the environmental safety factor based on entropy, spectrum characteristics and abnormal interference patterns is a comprehensive process. This embodiment uses a fuzzy inference system (FIS) to implement this process. First, the various indicators (entropy curve characteristics, spectrum parameters, anomaly detection scores) obtained in the above steps are used as input variables of FIS. Then, a fuzzy rule set is designed, such as: if the entropy value is high and the spectrum is stable and there is no abnormal interference, the environmental safety factor is high. Through the Mamdani reasoning method and the center of gravity method, the environmental safety factor between 0 and 1 is finally obtained. In order to optimize the performance of FIS, the adaptive neuro-fuzzy inference system (ANFIS) technology is adopted to adjust the membership function parameters and rule weights through historical data training.

[0103] Furthermore, in order to improve the timeliness and adaptability of environmental safety assessment, this embodiment can also introduce an online learning mechanism. Specifically, an incremental learning algorithm, such as Online Stochastic Gradient Descent, is used to continuously update the parameters of the LSTM-CNN anomaly detection model and the ANFIS system. In this way, the system can continuously adapt to emerging electromagnetic environment patterns and interference types to maintain the accuracy of the assessment.

[0104] It should be understood that the assessment of environmental safety factors also needs to consider spatiotemporal correlation. This embodiment can adopt a spatiotemporal attention mechanism to capture changes in electromagnetic fields at different time and space scales. In specific implementation, a self-attention layer is added to the LSTM-CNN model, allowing the model to dynamically adjust the attention to different time steps and sensor positions when processing sequence data. This mechanism can effectively identify abnormal patterns in local and global ranges and improve the accuracy of environmental safety assessment.

[0105] The benefit of this embodiment is that, through multi-angle and multi-scale electromagnetic environment analysis, combined with signal processing and machine learning technology, an accurate assessment of environmental safety is achieved. This method can not only capture the complex dynamic characteristics of the electromagnetic environment, but also adapt to various unknown interference patterns. At the same time, the introduction of online learning and spatiotemporal attention mechanism enables the system to have the ability of continuous learning and self-adaptation, and can maintain high accuracy and reliability in complex and changing electromagnetic environments.

[0106] Embodiment 6

[0107] In order to solve the problem that the traditional environmental safety factor calculation method lacks accurate quantification and comprehensive consideration, this embodiment further optimizes the calculation steps and formula of the environmental safety factor. This method achieves accurate quantitative evaluation of the environmental safety status by comprehensively considering the entropy value of the electromagnetic field intensity, the spectrum abnormality and the abnormal interference degree.

[0108] like Figure 8 As shown, the calculation steps of the environmental safety factor include:

[0109] S601. Measure the environmental electromagnetic field strength.

[0110] S602: Calculate the entropy value and maximum entropy value of the electromagnetic field intensity.

[0111] S603: Analyze the frequency spectra of different frequency bands and calculate the abnormality of each frequency band.

[0112] S604: Allocate a weight to each frequency band.

[0113] S605: Calculate the abnormal interference degree and the maximum abnormal interference degree.

[0114] The calculation formula of environmental safety factor is:

[0115]

[0116] Among them, ESF is the environmental safety factor, H is the entropy value of the electromagnetic field intensity, Hmax is the maximum entropy value, wi is the weight of the i-th frequency band, Si is the spectrum anomaly of the i-th frequency band, AD is the abnormal interference degree, and ADmax is the maximum abnormal interference degree.

[0117] Furthermore, this embodiment uses a high-precision magnetic field sensor array to measure the ambient electromagnetic field strength, and specifically uses a magnetic sensor based on the tunnel magnetoresistance (TMR) effect, model TMR2305. This sensor has high sensitivity and wide bandwidth characteristics, and can accurately capture magnetic field changes in the range of 0.1Hz to 10kHz. The sensor array is arranged in a regular hexagon, with a total of 7 measurement points to cover the three-dimensional space around the smart lock. Each measurement point simultaneously records the magnetic field strength in the x, y, and z directions, and the sampling rate is set to 20kHz to ensure that high-frequency transient changes are captured.

[0118] It should be understood that the calculation of the entropy value of the electromagnetic field intensity adopts the modified sample entropy (MSE) algorithm. Compared with the traditional sample entropy, MSE has better statistical stability. During the calculation process, the original data is first normalized, and then the embedding dimension m=3 and the tolerance r=0.15×standard deviation are set. Through the sliding window technology, the MSE values ​​of different time scales (1-20) are calculated to obtain an entropy curve. The final entropy value H is taken as the integral average of this curve. The maximum entropy value Hmax is obtained by performing the same processing on the white noise signal as the theoretical upper limit of the entropy value.

[0119] Furthermore, spectrum analysis and abnormality calculation are used to evaluate the stability of the electromagnetic environment. This embodiment uses wavelet packet transform (WPT) combined with adaptive spectrum estimation technology to perform spectrum analysis. Specifically, the signal is decomposed into 7 layers of WPT using Daubechies 8 wavelet basis functions to obtain 128 frequency bands. For each frequency band, the Burg autoregressive model is applied to estimate the power spectrum density, and the model order is automatically selected by the Akaike Information Criterion (AIC). The calculation of the spectrum abnormality Si is based on the Kullback-Leibler divergence of the current power spectrum and the historical normal state power spectrum. In order to improve the computational efficiency, an approximate calculation method is used to convert the KL divergence into the logarithmic expectation of the ratio of the two power spectra.

[0120] It is understandable that the allocation of frequency band weights directly affects the accuracy of the environmental safety factor. This embodiment proposes a weight allocation algorithm. First, the weights of each frequency band are initialized based on prior knowledge. Then, the weights are continuously adjusted through an online learning mechanism. Specifically, the exponentially weighted moving average (EWMA) method is used to update the importance score of each frequency band, and the score is determined by the anomaly detection accuracy of the frequency band and its contribution to the overall safety status. The weight wi is finally normalized by the softmax function to ensure that the sum of all weights is 1.

[0121] Furthermore, the calculation of abnormal interference degree AD incorporates a variety of anomaly detection technologies. This embodiment combines two methods, Isolation Forest and Autoencoder. Isolation forest is used to detect global anomalies, while autoencoder focuses on capturing anomalies in local patterns. In specific implementation, electromagnetic field intensity data and spectrum features are used as inputs, and anomaly scores are obtained through these two models respectively. The final AD value is obtained by weighted averaging these two scores, and the weights are determined by cross-validation. The maximum abnormal interference degree ADmax is determined by historical data analysis and expert experience, representing the maximum acceptable abnormal level of the system.

[0122] It should be understood that the calculation formula of the environmental safety factor ESF comprehensively considers three aspects: entropy value, spectrum anomaly, and abnormal interference. The exponential term in the formula is designed to amplify small security threats, making the system more sensitive to potential risks. The coefficients α and β are used to adjust the degree of influence of entropy value and abnormal interference on the final safety factor, and their values ​​are optimized through a large amount of experimental data and machine learning methods. Specifically, the Bayesian optimization algorithm is used to determine the optimal α and β values ​​with the goal of maximizing the correlation between ESF and the manually labeled safety level.

[0123] Furthermore, in order to improve the timeliness of ESF calculation, this embodiment can also use sliding window technology and outlier processing mechanism. ESF calculation is based on a sliding window of the latest N data points, and the window size N is dynamically adjusted through an adaptive algorithm to balance real-time and stability. For outliers in the window, a modified Z-score method is used for detection and processing to prevent transient interference from causing excessive impact on ESF.

[0124] The benefit of this embodiment is that, through the refined electromagnetic environment analysis and comprehensive evaluation method, the environmental safety factor is accurately quantified. This method not only takes into account the overall complexity of the electromagnetic field (through entropy value), but also pays attention to the abnormal conditions of specific frequency bands and the overall abnormal interference level, providing a comprehensive and detailed safety assessment framework. At the same time, the adaptive weight allocation and online learning mechanism enable the system to continuously adapt to the changing electromagnetic environment and maintain the accuracy and timeliness of the assessment.

[0125] Embodiment 7

[0126] In order to solve the problem that the traditional smart lock opening decision method is not accurate enough when dealing with complex emotions and environmental factors, this embodiment further optimizes the method of integrating the user emotion model and the environmental safety factor to generate the unlocking probability distribution. This method achieves accurate quantification and probabilistic expression of the unlocking decision through mapping the emotional state to the credibility, calculating the weighted geometric mean, and constructing the probability density function.

[0127] like Fig. 9 As shown in FIG. 1 , the steps of integrating the user emotion model and the environmental safety factor to generate the unlocking probability distribution include:

[0128] S701: Map the user's current emotional state to emotional credibility.

[0129] S702: Calculate the weighted geometric mean of the emotional credibility and the environmental safety factor.

[0130] S703: Construct an unlocking probability density function based on the weighted geometric mean.

[0131] Furthermore, the mapping of the user's current emotional state to emotional credibility is the first step in the fusion process. This embodiment uses a Gaussian Process Regression (GPR) model to achieve this mapping. The input of the GPR model is a multidimensional emotional state vector output by the user's emotional model, and the output is an emotional credibility value in the range of [0,1]. In the specific implementation, a radial basis function (RBF) kernel is used as the covariance function of GPR, and its length scale and signal variance parameters are optimized by maximizing the marginal likelihood. In order to process high-dimensional emotional state space, an automatic relevance determination (ARD) mechanism is also introduced to automatically learn the importance of each emotional dimension. The training data of the model comes from a large number of users' historical emotional data and corresponding unlocking results, which are obtained through manual annotation and semi-supervised learning methods.

[0132] It should be understood that the GPR model not only provides a point estimate of the credibility of emotions, but also gives the uncertainty of the estimate. This uncertainty information plays an important role in the subsequent fusion process. In order to improve the generalization ability of the model, this embodiment also adopts ensemble learning technology, specifically using sparse GPR model integration with random feature expansion, which effectively solves the problem of large-scale data processing and real-time reasoning.

[0133] Furthermore, the fusion of emotional credibility and environmental safety factor adopts the weighted geometric mean method. This method can better reflect the mutual influence of the two factors compared with the arithmetic mean. The specific calculation formula is:

[0134]

[0135] Among them, F is the fusion result, EC is the emotional credibility, ESF is the environmental safety factor, w1 and w2 are the weights of emotional credibility and environmental safety factor respectively. The weight is determined by an adaptive mechanism and dynamically adjusted based on the current environmental conditions and user status. Specifically, the Multi-Armed Bandit algorithm is used to regard the weight selection as an exploration-exploitation problem, and the weight allocation strategy is continuously optimized through the Thompson sampling method.

[0136] It is understandable that in order to consider the uncertainty of the estimation of emotional credibility and environmental safety factor, this embodiment introduces the Dempster-Shafer evidence theory framework. The uncertainty of emotional credibility and the measurement error of environmental safety factor output by the GPR model are converted into basic probability assignment (BPA), and then the two evidence sources are fused through the Dempster combination rule. This method can better handle uncertainty and conflicting information and improve the reliability of the fusion results.

[0137] Furthermore, an unlocking probability density function is constructed based on the fusion result. This embodiment uses a non-parametric kernel density estimation (KDE) method to construct a probability density function. Specifically, a Gaussian kernel function with adaptive bandwidth is used, and the bandwidth parameter is automatically determined by the Silverman rule. In order to deal with the possible multi-peak distribution of the fusion result, a mixed kernel density estimation (Mixture KDE) technique is also introduced, and the expectation maximization (EM) algorithm is used to automatically determine the optimal number and parameters of kernel functions.

[0138] It should be understood that the construction of the probability density function not only takes into account the current fusion results, but also utilizes historical data. This embodiment designs an online update mechanism that dynamically updates the parameters of the kernel density estimate using the exponentially weighted moving average (EWMA) method. This method can smooth short-term fluctuations while maintaining sensitivity to new patterns. In order to prevent the probability distribution from being too concentrated or dispersed, a maximum entropy regularization term is also introduced to ensure that the probability density function maintains maximum uncertainty while satisfying known constraints.

[0139] Furthermore, in order to improve the accuracy of unlocking decisions, this embodiment introduces a Bayesian decision framework based on the probability density function. Specifically, the loss function of the two decision options of unlocking and not unlocking is defined, and the final decision is made by minimizing the expected loss. The design of the loss function takes into account the different effects of false unlocking and rejection of normal unlocking requests, and dynamically adjusts the loss function parameters through user feedback and system log analysis.

[0140] It is understandable that in order to handle extreme situations and emergencies, this embodiment also designs an anomaly detection and processing module. The Isolation Forest algorithm is used to monitor the anomalies of the fusion results and probability distribution in real time. When an anomaly is detected, the backup decision logic is triggered, such as adding additional identity authentication steps or starting a manual intervention mechanism.

[0141] The benefit of this embodiment is that through sophisticated probability models and decision theory, accurate fusion and quantification of user emotions and environmental safety factors are achieved. This method not only takes into account the uncertainty of each factor, but also introduces adaptive learning and exception handling mechanisms, which can make robust unlocking decisions in complex and changing scenarios. At the same time, the probabilistic expression provides a rich information basis for subsequent risk assessment and decision optimization. This method significantly improves the performance of the smart lock system in balancing security and convenience, and lays a theoretical and technical foundation for realizing a smarter and safer access control system.

[0142] Embodiment 8

[0143] In order to solve the problem that the traditional method of constructing the probability density function of unlocking is not flexible enough when dealing with complex emotions and environmental factors, this embodiment further refines the method of constructing the probability density function of unlocking. This method realizes accurate modeling of the probability distribution of unlocking through the steps of calculating the emotional credibility, utilizing the environmental safety factor, determining the adjustment parameters, calculating the μ value, and selecting the standard deviation parameters.

[0144] Specifically, the steps of constructing the unlocking probability density function based on the weighted geometric mean include: calculating the emotional credibility, obtaining the environmental safety factor, determining the adjustment parameter, calculating the μ value, and selecting an appropriate standard deviation parameter.

[0145] The calculation formula of the unlock probability density function is:

[0146]

[0147] The calculation formula of μ value is:

[0148]

[0149] Among them, σ is the standard deviation parameter, EC is the emotional credibility, ESF is the environmental safety factor, and α and β are adjustment parameters.

[0150] Furthermore, the calculation of emotional credibility adopts the method of improved Hidden Markov Model (HMM) combined with Particle Filter. Hidden Markov Model HMM is used to model the transition of emotional state, and the observation values ​​include the user's physiological signals (such as heart rate, skin electricity) and behavioral characteristics (such as operation frequency and strength). In order to deal with the nonlinear dynamic characteristics of emotional state, the particle filter algorithm is introduced. In the specific implementation, the auxiliary particle filter (APF) with adaptive resampling strategy is used, and the number of particles is dynamically adjusted to balance the computational efficiency and estimation accuracy. The emotional credibility is finally obtained by integrating the distribution of emotional state after filtering.

[0151] It should be understood that the use of the environmental safety factor not only takes into account its numerical value, but also its time series characteristics. This embodiment introduces a long short-term memory network (LSTM) to capture the time dependency of the environmental safety factor. The input of the LSTM network is the sequence of environmental safety factors in the past N time steps, and the output is a comprehensive assessment of the environmental safety level at the current moment. The network structure adopts a bidirectional LSTM with attention mechanism to better capture long-term dependencies. In this way, the system is able to take into account historical trends and sudden changes in environmental safety.

[0152] Furthermore, the Bayesian Optimization method is used to determine the adjustment parameters α and β. The optimization goal is to maximize the weighted sum of the accuracy of the unlocking decision and the security of the system. Bayesian optimization uses Gaussian process as a surrogate model and adopts expected improvement (EI) as the acquisition function. In order to deal with multi-objective optimization problems, the concept of Pareto efficiency frontier is introduced, and the Hypervolume indicator is used as the optimization criterion. During the optimization process, historical unlocking data and security event logs are used as training samples, and the performance of each set of parameters is evaluated through cross-validation.

[0153] It is understandable that the calculation of the μ value not only depends on the emotional credibility and environmental safety factor, but also takes into account the user's historical behavior pattern. This embodiment designs an adaptive mechanism based on reinforcement learning to dynamically adjust the calculation formula of the μ value. Specifically, a deep Q-network (DQN) is used to map the current state (including emotional credibility, environmental safety factor, time, location, etc.) to the optimal μ value adjustment strategy. In order to improve learning efficiency, Prioritized Experience Replay and Double Q-learning techniques are used.

[0154] Furthermore, the adaptive Metropolis-Hastings algorithm is used to select the standard deviation parameter σ. The algorithm can automatically adjust the σ value according to the current system state to balance exploration and utilization. In the specific implementation, an objective function based on the unlocking success rate and security risk is defined, and the Metropolis-Hastings algorithm is used to perform a random walk in the parameter space of σ, and the acceptance probability is based on the degree of improvement of the objective function. In order to improve the convergence speed of the algorithm, the simulated annealing mechanism is introduced to gradually reduce the probability of accepting a new σ value over time.

[0155] It should be understood that the calculation of the unlocking probability density function not only takes into account the various parameters in the formula, but also introduces the influence of external environmental factors. This embodiment can also introduce a multilayer perceptron (MLP) network to model the modulation effect of external factors on the probability density function. The input of MLP includes environmental characteristics such as the current time, weather conditions, and surrounding pedestrian density, and the output is the modulation coefficient of the probability density function. This design enables the system to adapt to different external environments and improve the situational adaptability of unlocking decisions.

[0156] It is understandable that in order to improve the time continuity and stability of the probability density function, the present embodiment adopts the Kalman filter technology. The parameters of the probability density function are regarded as state variables, and the Kalman filter is used to estimate and smooth them in real time. This method can effectively suppress the drastic fluctuations in the probability distribution caused by the input data noise, and provide a more stable and reliable basis for unlocking decision-making.

[0157] The benefit of this embodiment is that, through probability modeling and machine learning technology, the accurate construction and dynamic adjustment of the unlocking probability density function are achieved. This method not only takes into account the complex interaction of emotions and environmental factors, but also introduces mechanisms such as adaptive learning, multimodal modeling and time series smoothing, which can provide stable and accurate unlocking probability estimation in various complex scenarios. At the same time, by integrating multiple advanced algorithms, the system has strong environmental adaptability and self-learning ability, and can continuously optimize its decision model.

[0158] Embodiment 9

[0159] In order to solve the problem of insufficient flexibility of traditional smart lock opening methods in dynamically adjusting the opening difficulty and handling complex bio-electromagnetic field interactions, this embodiment further optimizes the method of dynamically adjusting the opening difficulty of smart locks based on the probability distribution of unlocking, and introduces bio-electromagnetic field resonance detection and user emotion model update mechanism. This method realizes refined control and personalized adaptation of the smart lock opening mechanism through probability distribution feature analysis, dynamic difficulty adjustment, bio-electromagnetic field detection and long-term trend analysis.

[0160] like Fig.10 As shown, the steps of dynamically adjusting the opening difficulty of the smart lock based on the unlocking probability distribution include:

[0161] S901. Calculate the peak value and variance of the unlock probability distribution.

[0162] S902. Determine the basic opening difficulty according to the peak value.

[0163] S903. Use variance to adjust the sensitivity of difficulty changes.

[0164] like Fig.11 As shown, the opening method also includes the following steps:

[0165] S11. Detecting the resonance frequency between the user's biological electromagnetic field and the environmental electromagnetic field.

[0166] S12. When a specific resonant frequency is detected, the opening difficulty is temporarily reduced.

[0167] S13. Analyze the long-term change trend of the user's bioelectric field data.

[0168] S14. Dynamically update the baseline of the user emotion model according to the changing trend.

[0169] Furthermore, the peak and variance calculations of the unlock probability distribution use an improved kernel density estimation (KDE) method. In the specific implementation, a Gaussian kernel function with adaptive bandwidth is used, and the bandwidth parameter is automatically selected by the Sheather-Jones method. In order to improve the computational efficiency, the KDE algorithm accelerated by the fast Fourier transform (FFT) is used. The peak detection uses the persistence diagram technology, which can effectively identify the main peaks in the multi-peak distribution. The variance calculation adopts a robust statistical method, using the median absolute deviation (MAD) as a measure of the degree of dispersion to reduce the impact of outliers.

[0170] It should be understood that the determination of the basic opening difficulty depends not only on the peak value of the probability distribution, but also on the position of the peak value and the overall shape of the distribution. This embodiment designs a reasoning system based on fuzzy logic, which takes the peak height, position and skewness of the distribution as input variables and outputs the basic opening difficulty. The fuzzy rule base is jointly constructed by expert knowledge and machine learning methods, and the Mamdani reasoning model is used for reasoning. In order to improve the adaptability of the system, an online adaptive mechanism is introduced, which can adjust the fuzzy rules in real time according to user feedback and system performance.

[0171] Furthermore, the adjustment of the sensitivity of difficulty changes uses variance information and a nonlinear mapping function. Specifically, the sigmoid function is used to map the variance value to the sensitivity adjustment coefficient. The parameters of the sigmoid function are dynamically optimized through the particle swarm optimization (PSO) algorithm, and the objective function considers the balance between system response speed and stability. In order to deal with the uncertainty of variance estimation, the interval fuzzy set theory is introduced, and the sensitivity adjustment coefficient is represented by interval fuzzy numbers, thereby providing a more robust adjustment mechanism.

[0172] It is understandable that the resonance frequency detection of the user's bio-electromagnetic field and the environmental electromagnetic field adopts multi-channel synchronous acquisition and real-time spectrum analysis technology. Specifically, a high-precision SQUID magnetometer array is used, model Tristan7Channel, and the sampling rate is set to 10kHz. The spectrum analysis uses short-time Fourier transform (STFT) combined with wavelet packet decomposition (WPD) method, which can capture the resonance phenomenon of wideband and narrowband at the same time. The identification of the resonance frequency uses an improved peak detection algorithm combined with phase consistency analysis to improve the detection accuracy and noise resistance.

[0173] Furthermore, the temporary difficulty reduction mechanism after a specific resonant frequency is detected adopts a dynamic threshold method. The threshold is set based on the statistical analysis of historical data and the prediction of the machine learning model. In the specific implementation, the importance score of the resonant frequency is predicted using a long short-term memory network (LSTM), and then the magnitude of the difficulty reduction is dynamically adjusted through an adaptive threshold algorithm. In order to prevent frequent difficulty adjustments from affecting the user experience, a cooling time mechanism is introduced, and the Markov Decision Process (MDP) is used to model and optimize the dynamic adjustment strategy of the cooling time.

[0174] It is important to understand that the long-term trend analysis of the user's bioelectric field data uses time series decomposition and nonlinear trend extraction techniques. First, the time series is decomposed into trend, seasonal, and residual components using the seasonal and trend decomposition (STL) method. Then the local regression smoothing (LOESS) method is applied to the trend component to extract nonlinear trends. In order to deal with possible structural breaks, the Bayesian online change point detection algorithm is introduced to identify the mutation points of the trend in real time.

[0175] Furthermore, the dynamic update of the user emotion model baseline is based on the extracted long-term trend, using incremental learning and forgetting mechanisms. In the specific implementation, the Online Passive-Aggressive algorithm is used in combination with an exponentially decaying forgetting function, which can adapt to the slow changes in user emotion patterns while retaining long-term trend information. In order to process multi-dimensional emotion features, tensor decomposition technology is introduced, and the Tucker decomposition method is used to approximate multi-dimensional emotion data, thereby reducing the complexity of update calculations.

[0176] The benefit of this embodiment is that through signal processing, machine learning and control theory methods, the intelligent lock opening mechanism is highly intelligent and personalized. This method not only takes into account the statistical characteristics of the probability distribution of unlocking, but also introduces bio-electromagnetic field resonance detection and long-term emotional trend analysis, which can more comprehensively evaluate the user status and environmental conditions. At the same time, through mechanisms such as dynamic thresholds, incremental learning and anomaly detection, the system has strong adaptive capabilities and security guarantees.

[0177] Embodiment 10

[0178] In order to solve the problem of low integration and insensitive response of traditional smart lock systems when integrating bioelectric field information, environmental electromagnetic field data and user emotional state for intelligent opening control, this embodiment proposes a highly integrated smart lock automatic opening system. Through the collaborative work of multiple functional modules, the system realizes full-process intelligent control from bioelectric field data collection, emotional modeling, environmental safety assessment to unlocking probability distribution generation and dynamic difficulty adjustment.

[0179] like Fig.12 As shown, the smart lock automatic opening system includes:

[0180] The bioelectric field acquisition module 100 is used to acquire the bioelectric field data of the user.

[0181] The emotion modeling module 200 is used to construct a user emotion model based on bioelectric field data.

[0182] The electromagnetic field detection module 300 is used to collect environmental electromagnetic field intensity data.

[0183] The safety assessment module 400 is used to assess the environmental safety factor according to the environmental electromagnetic field strength data.

[0184] The probability distribution generation module 500 is used to integrate the user emotion model and the environmental safety factor to generate the unlocking probability distribution.

[0185] The difficulty adjustment module 600 is used to dynamically adjust the opening difficulty of the smart lock based on the unlocking probability distribution.

[0186] Execution module 700, when the smart lock is touched or awakened, performs an unlocking operation according to the current opening difficulty.

[0187] Furthermore, the bioelectric field acquisition module 100 uses a highly sensitive capacitively coupled electrode array, in conjunction with a low-noise amplifier and a high-precision analog-to-digital converter. Specifically, dry electrode technology is used, and the electrode material is Ag / AgCl, which has good biocompatibility and signal stability. The signal conditioning circuit uses the ADS1299 chip from Texas Instruments, which has 24-bit resolution and a built-in programmable gain amplifier. In order to improve the spatial resolution, the electrode array adopts a 64-channel design and is arranged around the touch area of ​​the smart lock. The signal sampling rate is set to 1kHz, which can capture high-frequency bioelectric signal components.

[0188] It should be understood that the emotion modeling module 200 is based on a deep learning architecture and adopts a hybrid model of a convolutional neural network (CNN) and a long short-term memory network (LSTM). CNN is used to extract the spatial features of bioelectric field data, and LSTM is used to capture time series features. The model training adopts a transfer learning strategy, first pre-training on a large-scale public emotion dataset, and then fine-tuning on user-specific data. In order to deal with individual differences, a meta-learning framework is introduced, and the model-independent meta-learning (MAML) algorithm is used to improve the rapid adaptability of the model. The output of the emotional state adopts a multi-label classification method, which can simultaneously characterize multi-dimensional emotional features.

[0189] Furthermore, the electromagnetic field detection module 300 uses a broadband electromagnetic field sensor array, covering from extremely low frequency to microwave frequency bands. Specifically, Narda's NBM-550 model broadband electromagnetic field meter is used, equipped with EF0691 probe (frequency range 100kHz-6GHz) and HP-3550 probe (frequency range 27MHz-3.5GHz). In order to improve the spatial resolution, multiple sensors are arranged around the smart lock, and a three-dimensional electromagnetic field intensity distribution map is generated through an interpolation algorithm. Data acquisition adopts a real-time stream processing architecture, using Apache Kafka as the message middleware to ensure high throughput and low latency data transmission.

[0190] It is understandable that the safety assessment module 400 adopts a hybrid method based on fuzzy inference system and machine learning. First, the electromagnetic field strength data is preliminarily evaluated using a Mamdani-type fuzzy inference system, and the fuzzy rule base is constructed based on electromagnetic field safety standards and expert knowledge. Then, the fuzzy inference results are further optimized using a random forest algorithm, taking into account contextual information such as time and location. In order to deal with uncertainty, the Dempster-Shafer evidence theory is introduced to fuse the safety assessment results from different sources to obtain the final environmental safety factor.

[0191] Furthermore, the probability distribution generation module 500 adopts a hybrid architecture of a variational autoencoder (VAE) and a generative adversarial network (GAN). VAE is used to learn the potential representation of the user emotion model and the environmental safety factor, and GAN is used to generate high-quality unlocking probability distribution. An adaptive learning rate adjustment algorithm, such as the Adam optimizer, is used during training to improve convergence speed and stability. The generated probability distribution is compared with the true distribution to ensure the generation quality.

[0192] It should be understood that the difficulty adjustment module 600 is based on a reinforcement learning framework and uses a deep Q network (DQN) algorithm. The state space includes information such as the current unlock probability distribution, historical unlock records, and system status, and the action space is defined as a series of discrete difficulty adjustment options.

[0193] Furthermore, the execution module 700 adopts a modular design, including a touch sensing unit, an identity authentication unit and a lock core control unit. The touch sensing unit uses a capacitive touch sensor array and adopts a PSoC 4 chip, which has high sensitivity and low power consumption. The identity authentication unit integrates a variety of biometric technologies, including fingerprint recognition (using Synaptics' Natural ID fingerprint sensor) and facial recognition (using Intel RealSense D435i depth camera). The lock core control unit is driven by a stepper motor and uses ST Microelectronics' L6470 chip, which can achieve precise torque control and position feedback.

[0194] It is understandable that in order to improve the reliability and security of the system, this embodiment also designs a multi-layer security architecture. At the hardware level, a secure element (Secure Element) is used to store sensitive data and keys, and NXP's A7002 chip is used to support AES-256 encryption and ECC digital signatures. At the software level, a distributed identity authentication system based on blockchain is implemented, using the Hyperledger Fabric framework to ensure the immutability and traceability of identity information. At the same time, a behavioral anomaly detection algorithm is introduced, which monitors user operation behavior in real time based on the hidden Markov model (HMM) and triggers additional security measures when anomalies are detected.

[0195] The benefit of this embodiment is that by integrating sensing technology, deep learning algorithms and security architecture, a highly intelligent and personalized automatic opening system is realized. The system can not only accurately capture the user's bioelectric field information and environmental electromagnetic field data, but also achieve accurate emotion modeling and security assessment. The dynamic difficulty adjustment mechanism and multimodal execution module 700 further improve the adaptability of the system. At the same time, the multi-layer security design ensures that the system maintains a high level of security while providing convenience. This fully integrated solution sets a new standard for intelligent access control systems and provides an innovative technical path for the future development of smart home and IoT security.

Claims

1. A method for automatically opening a smart lock, characterized in that: The steps of the opening method include: Obtaining the user's bioelectric field data; Building a user emotion model based on the bioelectric field data; Collect environmental electromagnetic field strength data; Evaluate the environmental safety factor according to the environmental electromagnetic field strength data; The user emotion model and the environmental safety factor are integrated to generate a probability distribution of unlocking; Dynamically adjusting the difficulty of opening the smart lock based on the unlocking probability distribution; When the smart lock is touched or awakened, the unlocking operation is performed according to the current opening difficulty.

2. The method for automatically opening a smart lock according to claim 1, characterized in that: The bioelectric field data includes at least one of a user's skin potential, an electrocardiogram waveform, and a brain wave frequency.

3. The method for automatically opening a smart lock according to claim 1, characterized in that: The step of constructing a user emotion model based on bioelectric field data includes: Extract the time-frequency characteristics of bioelectric field data; Analyze the complexity of bioelectric field data using nonlinear dynamics methods; Based on the time-frequency characteristics and complexity analysis results, a multidimensional emotional state space is constructed; The user's current emotional state is located in the multi-dimensional emotional state space.

4. The method for automatically opening a smart lock according to claim 3, characterized in that: The construction formula of the user emotion model is: EM=f(TF,CD,λ) Among them, EM is the emotion model, TF is the time-frequency feature vector, CD is the complexity index, λ is the nonlinear mapping function, and f is the multidimensional state space construction function.

5. The method for automatically opening a smart lock according to claim 1, characterized in that: The step of evaluating the environmental safety factor according to the environmental electromagnetic field strength data comprises: Calculate the entropy value of the ambient electromagnetic field strength; Analyze the spectrum characteristics of electromagnetic field strength; Detect abnormal electromagnetic interference patterns; An environmental safety factor is evaluated based on the entropy value, the spectrum characteristics and the abnormal interference pattern.

6. The method for automatically opening a smart lock according to claim 5, characterized in that: The calculation steps of the environmental safety factor include: Measure the strength of the ambient electromagnetic field; Calculate the entropy value and maximum entropy value of electromagnetic field intensity; Analyze the spectrum of different frequency bands and calculate the abnormality of each frequency band; Assign weights to each frequency band; Calculate abnormal interference degree and maximum abnormal interference degree; The calculation formula of the environmental safety factor is: Among them, ESF is the environmental safety factor, H is the entropy value of the electromagnetic field intensity, Hmax is the maximum entropy value, wi is the weight of the i-th frequency band, Si is the spectrum anomaly of the i-th frequency band, AD is the abnormal interference degree, and ADmax is the maximum abnormal interference degree.

7. The method for automatically opening a smart lock according to claim 3, characterized in that: The step of integrating the user emotion model and the environmental safety factor to generate the unlocking probability distribution includes: Mapping the user's current emotional state to emotional credibility; Calculating a weighted geometric mean of the emotional credibility and the environmental safety factor; An unlock probability density function is constructed based on the weighted geometric mean.

8. The method for automatically opening a smart lock according to claim 7, characterized in that: The step of constructing the unlock probability density function based on the weighted geometric mean comprises: Calculate emotional credibility; Obtaining the environmental safety factor; Determine adjustment parameters; Calculate μ value; Choose an appropriate standard deviation parameter; The calculation formula of the unlocking probability density function is: in, σ is the standard deviation parameter, EC is the emotional credibility, ESF is the environmental safety factor, and α and β are adjustment parameters.

9. The method for automatically opening a smart lock according to claim 1, characterized in that: The step of dynamically adjusting the opening difficulty of the smart lock based on the unlocking probability distribution includes: Calculate the peak and variance of the unlock probability distribution; Determine the basic opening difficulty based on the peak value; Using variance to adjust the sensitivity of difficulty changes; The opening method further comprises the following steps: Detect the resonance frequency between the user's bio-electromagnetic field and the environmental electromagnetic field; When a specific resonant frequency is detected, the opening difficulty is temporarily reduced; Analyze the long-term trend of users’ bioelectric field data; Dynamically update the baseline of the user sentiment model based on changing trends.

10. A smart lock automatic opening system, characterized in that: The opening system comprises: A bioelectric field acquisition module is used to obtain the user's bioelectric field data; An emotion modeling module, used to construct a user emotion model based on the bioelectric field data; Electromagnetic field detection module, used to collect environmental electromagnetic field strength data; A safety assessment module, used to assess an environmental safety factor based on the environmental electromagnetic field strength data; A probability distribution generation module, used to integrate the user emotion model and the environmental safety factor to generate an unlocking probability distribution; A difficulty adjustment module, used to dynamically adjust the opening difficulty of the smart lock based on the unlocking probability distribution; The execution module is used to perform the unlocking operation according to the current opening difficulty when the user touches the smart lock.

Images