A smart security alarm system and method based on wireless networking

Abstract

This invention discloses an intelligent security alarm system and method based on wireless networking, relating to the fields of IoT and intelligent security technology. It includes a security data acquisition module, a security feature extraction module, a security anomaly detection module, a security alarm module, and an intelligent security management module. The security data acquisition module collects human infrared, environmental sound, and three-dimensional vibration data, extracts security features, detects movement and sound anomalies, constructs an alarm model, performs tiered alarms, and disseminates information wirelessly, displaying the distribution of security events in a heat map. This invention achieves wireless acquisition and intelligent fusion of multimodal security data, significantly improving the accuracy and real-time performance of anomaly detection. Through tiered alarms and heat map display, it intuitively reflects the distribution of security events, facilitating rapid response and decision-making, reducing false alarm rates, simplifying system deployment and maintenance, and enhancing the intelligence and practicality of the security system.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things and smart security technology, specifically to a smart security alarm system and method based on wireless networking. Background Technology

[0002] A traditional security alarm system is a security network composed of various sensors, function keys, detectors, actuators, and other components. This system is often considered the brain of a home or specific location security system. Functionally, traditional security alarm systems are mainly used to achieve security goals such as fire prevention, theft prevention, gas leak alarms, and emergency assistance. It relies on intelligent control network technology and is managed and controlled by a microcomputer. It automatically issues an alarm when accidents such as burglary, theft, fire spread, gas leaks, or emergency assistance occur. In terms of system composition, a traditional security alarm system includes key components such as front-end detectors, intermediate transmission sections, and alarm hosts. These components work together to detect environmental changes in real time, collect and process environmental data, and once an anomaly is detected, it issues an alarm through the alarm device to remind relevant personnel to take countermeasures. In summary, with its reliable security protection functions and flexible application scenarios, the traditional security alarm system has become an indispensable security guarantee for various industries.

[0003] To address the high false alarm rate caused by insufficient multimodal sensing fusion in traditional security alarm systems, existing technologies employ multiple sensors operating in parallel, achieving security protection through independent signal detection and analysis. However, this approach still results in isolated information between sensors and a lack of effective fusion, leading to frequent false alarms when facing complex and ever-changing security threats, thus affecting the system's stability and reliability. Furthermore, the complex deployment and poor scalability of wired networks in traditional security alarm systems remain unresolved. Wired networks not only require cumbersome wiring but also present significant challenges for subsequent expansion and adjustments once the layout is complete, limiting the flexibility and adaptability of security systems and failing to meet the diverse and dynamic security protection needs of modern environments. Therefore, to address these issues, a wireless networking-based intelligent security alarm system and method are proposed. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent security alarm system and method based on wireless networking to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: Firstly, an intelligent security alarm system based on wireless networking, comprising a security data acquisition module, a security feature extraction module, a security anomaly detection module, a security alarm module, and an intelligent security management module;

[0006] The security data acquisition module collects and preprocesses human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data.

[0007] The security feature extraction module extracts features from the preprocessed human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data to construct a security feature sequence list.

[0008] The security anomaly detection module is used to perform abnormal movement detection and abnormal sound recognition by combining the security feature sequence list;

[0009] The security alarm module, based on the analysis results of the security feature sequence list and the security anomaly detection module, constructs a security alarm model and outputs the alarm probability.

[0010] The intelligent security management module performs hierarchical alarms based on the analysis results of the security alarm model, publishes security alarm information using wireless networking technology, and presents the distribution of regional security events in the form of a heat map.

[0011] A further improvement to the technical solution of the present invention is that the security anomaly detection module includes a motion detection unit and a sound recognition unit:

[0012] The motion detection unit constructs an abnormal motion detection model based on the extracted infrared intensity change rate features and outputs the probability of abnormal motion.

[0013] The sound recognition unit constructs an abnormal sound recognition model based on the energy proportion characteristics of the audio frequency band and the characteristics of the first three Mel-frequency cepstral coefficients, and outputs the probability of abnormal voiceprints.

[0014] A further improvement to the technical solution of this invention lies in the following: the acquisition and preprocessing process of human body movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data in the security data acquisition module includes:

[0015] Pyroelectric sensor nodes are installed at building entrances, corridors, and key perimeter areas using a self-organizing network protocol to collect infrared signal intensity data of human movement. The spacing between the pyroelectric sensor nodes does not exceed 8 meters, the vertical installation height is 1.2-1.5 meters, and the tilt angle is between 15 and 30 degrees. A mesh topology is automatically constructed through a dynamic routing algorithm.

[0016] Four-unit microphone arrays are deployed diagonally in the building's interior space and under the outdoor eaves to collect ambient sound waveform data. The four-unit microphone arrays are spaced 1.5 meters apart and 2.2 to 2.5 meters above the ground. They are directly connected to the gateway via a star topology. The four-unit microphone array nodes use a carrier sense multiple access (CMA) collision avoidance mechanism to compete for the channel.

[0017] Triaxial piezoelectric accelerometers are deployed on the door and window frames and safe bases of the building. The axes of the triaxial piezoelectric accelerometers are perpendicular to the monitoring plane, fixed with epoxy resin, and connected through a hybrid networking architecture to collect three-dimensional vibration waveform data.

[0018] Preprocessing is performed on the collected human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data;

[0019] When the rate of change of infrared signal intensity of human movement exceeds 0.5 microwatts per square centimeter per second, historical data recording is activated, a 10 Hz time series is generated, the arithmetic mean of five consecutive sampling points is calculated, the current point data is replaced, and instantaneous interference pulses with a duration of less than 0.5 seconds are eliminated.

[0020] Calculate the signal energy of the ambient sound waveform data within each 10-millisecond window. When the energy exceeds three times the silent baseline for three consecutive windows, it is determined as the start point of the effective segment. When the energy falls back below the baseline for more than 50 milliseconds, it is determined as the end point. Divide the sampling point values ​​within the effective segment by the maximum absolute value of the original data to normalize the maximum absolute value to 1.

[0021] The triaxial piezoelectric accelerometer node calibrates the local clock based on a precise time protocol, ensuring that the deviation between the three-dimensional vibration waveform data and the infrared signal timestamp is less than 5 milliseconds. Based on the reading of the built-in temperature sensor, the triaxial acceleration zero-point offset is adjusted according to a preset temperature drift coefficient.

[0022] A further improvement to the technical solution of this invention lies in that: in the security feature extraction module, the process of constructing the security feature sequence table includes:

[0023] The infrared intensity change rate sequence is obtained by dividing the difference between two adjacent infrared signal intensity sampling points of human movement by the sampling interval;

[0024] A short-time Fourier transform is performed on the normalized effective audio segments. A window is divided every 25 milliseconds, with adjacent windows overlapping by 12.5 milliseconds. The sum of squared amplitudes of the frequency points in the 500-2000 Hz band is divided by the sum of squared amplitudes of the frequency points in the 0-8000 Hz band to obtain the proportion of audio frequency band energy within the calculation window.

[0025] The effective audio segment is passed through a triangular filter bank with 40 Mel scale distributions. The output of the triangular filter is the logarithm of the signal energy of the corresponding frequency band. The discrete cosine transform is performed on the logarithm of the signal energy of the corresponding frequency band output by the 40 triangular filters, and the first three Mel cepstral coefficients are taken as characteristic components.

[0026] Calculate the total vibration energy integral E of the three-dimensional vibration waveform data within a 1-second window, and apply dynamic baseline normalization to the total energy integral to obtain the normalized vibration energy. The calculation process is as follows:

[0027]

[0028] Among them, a x (i), a y (i) and a z (i) represents the acceleration of the i-th sampling point on the x-axis, y-axis and z-axis respectively, Δt is the sampling interval time, N is the number of sampling points, μ is the mean vibration energy in the past hour, and σ is the standard deviation of vibration energy in the past hour.

[0029] The extracted infrared intensity change rate sequence, audio frequency band energy proportion, first three Mel-frequency cepstral coefficients, and normalized vibration energy are integrated to form a security feature sequence table.

[0030] A further improvement to the technical solution of this invention lies in the following: the process of constructing an abnormal movement detection model and outputting the abnormal movement probability in the movement detection unit includes:

[0031] Based on the isolated forest algorithm, decision trees are trained using historical normal infrared intensity change rate sequence data. The number and depth of decision trees are pre-set in the edge gateway of the intelligent security alarm system. Each decision tree selects the mean and variance of the normal infrared intensity change rate of the window as the splitting feature. The gateway receives the latest window mean and variance and inputs them into the isolated forest model to construct an abnormal movement detection model.

[0032] Traverse the decision tree and calculate the baseline path length c(n) and the average path length L of the sample in the decision tree. p After standardization, it is transformed into the abnormal movement probability P1, and its calculation process is as follows:

[0033]

[0034] Where n is the number of samples in the training dataset, γ is the Euler-Marcheroni constant, and P1∈[0,1].

[0035] A further improvement to the technical solution of this invention lies in the fact that the process of constructing an abnormal sound recognition model and outputting the probability of abnormal voiceprints in the sound recognition unit includes:

[0036] Max-min normalization was applied to the energy proportion characteristics of the 500-2000 Hz audio band, and standard deviation normalization was applied to the first three order Mel cepstral coefficients to construct a 4-dimensional voiceprint feature vector containing the audio band energy proportion, first-order Mel coefficient, second-order Mel coefficient, and third-order Mel coefficient.

[0037] Based on the support vector machine classification model architecture, a radial basis kernel function is selected to map the 4-dimensional voiceprint feature vector to a high-dimensional space. The historical environmental sound waveform dataset is input into the support vector machine classification model architecture to solve for the optimal hyperplane. The sample point closest to the hyperplane is selected as the support vector β, and the trained hyperplane parameters are pre-set to the edge gateway of the intelligent security alarm system to build an abnormal sound recognition model.

[0038] Input the 4-dimensional voiceprint feature vector into the abnormal sound recognition model, calculate its distance d with the hyperplane, and use the Sigmoid function to map this distance to the abnormal voiceprint probability P2. The calculation process is as follows:

[0039]

[0040] Where, α i For support vector weights, y i For the support vector labels, normal = 1, abnormal = -1, K is the result of the radial basis kernel function calculation, b is the bias term of the classification hyperplane, P2∈[0,1], the upper and lower limits of the threshold range of abnormal voiceprint probability are set, and a hierarchical transmission strategy is adopted for data transmission.

[0041] A further improvement to the technical solution of this invention lies in the following: the process of constructing a security alarm model and outputting the alarm probability in the security alarm module includes:

[0042] The security alarm module receives abnormal movement probability, abnormal soundprint probability, and normalized vibration energy features, and integrates them into a joint feature vector. A security alarm model is constructed using a logistic regression model architecture, and the alarm probability P is calculated. The calculation process is as follows:

[0043]

[0044] Where w1, w2, w3, w4 and w5 are the weight parameters of each input feature, and b is the bias term;

[0045] Based on historical alarm event data and normal environment data, the weight parameters of each input feature of logistic regression are optimized with the goal of minimizing cross-entropy loss. The trained weight vector and bias term are solidified in the edge gateway flash memory and differential parameter update packets are received monthly through a low-power wide-area communication channel.

[0046] The latest collected abnormal movement probability, abnormal soundprint probability, and normalized vibration energy characteristics are input into the security alarm model to obtain the current alarm probability P.

[0047] A further improvement to the technical solution of this invention lies in the following: the process of hierarchical alarm based on the analysis results of the security alarm model in the intelligent security management module includes:

[0048] Based on the analysis results of the security alarm model, different alarm levels are divided into high-risk alarm level, medium-risk alarm level and low-risk alarm level, and corresponding alarm probability threshold ranges are matched for each alarm level.

[0049] If the alarm probability is within the alarm probability threshold range of the high-risk alarm level, it is determined to be an intrusion event, triggering the audible and visual alarm and pushing the real-time location coordinates to the cloud. If the alarm probability is within the alarm probability threshold range of the medium-risk alarm level, the local buzzer will sound intermittently and the gateway LED will flash. If the alarm probability is within the alarm probability threshold range of the low-risk alarm level, an event log will be stored. The event log includes a timestamp, sensor location, and alarm probability. The environmental baseline is statistically analyzed every 24 hours, and the moving average and standard deviation of the alarm probability over the past 24 hours are calculated to adaptively adjust the alarm probability threshold.

[0050] A further improvement to the technical solution of this invention lies in the following: the process of using wireless networking technology to publish security alarm information and presenting the distribution of regional security events in the form of a heat map in the intelligent security management module includes:

[0051] Each sensor node is deployed through a self-organizing network protocol, star topology, and hybrid networking architecture. Alarm event data, including location coordinates, timestamps, and alarm probabilities, is transmitted to the edge gateway. High-risk alarms are transmitted directly to the cloud through a low-power wide-area communication channel with a latency of no more than 200ms. Medium-risk alarms are transmitted to the gateway for aggregation through a multi-hop transmission via the self-organizing network protocol, and forward error correction coding is enabled. Low-risk alarms are stored locally for 24 hours and then uploaded to the cloud in batches.

[0052] The intelligent security management module calibrates the timestamps of each sensor node based on a precise time protocol. The location coordinates of the sensor nodes are pre-stored in a cloud database. Discrete alarm events are mapped to a building planar grid to calculate the event density D(x,y) of each grid point. The calculation process is as follows:

[0053]

[0054] Where N is the total number of alarm events in the current time period, and P i Let K be the alarm probability of the i-th event, K be the kernel function, and h be the bandwidth parameter that is automatically adjusted according to the sparsity of the event spatial distribution. If the event spacing exceeds 5 meters, then h = 2 meters; if the event spacing is within 2 meters, then h = 1 meter.

[0055] The event density of each grid point is normalized to the [0,1] interval. An upper and lower limit of the normalized density threshold is set. If the event density of a grid point exceeds the upper limit of the normalized density threshold, it is judged as high density and a dark red to purple RGB gradient is used. If the event density of a grid point is within the range of the upper and lower limits of the normalized density threshold, it is judged as medium density and an orange to red RGB gradient is used. If the event density of a grid point is lower than the lower limit of the normalized density threshold, it is judged as low density and a light green to yellow RGB gradient is used. Based on WebGL technology, the grid density value is rendered as a heat map layer and superimposed on the building floor plan. The client pulls the latest density matrix from the cloud every 10 seconds.

[0056] The intelligent security management module selects the optimal data transmission path based on signal strength and node power using a dynamic routing algorithm. It combines edge computing nodes to perform local density pre-calculation and data compression, adopts a dual-path concurrent approach of low-power wide-area communication and self-organizing network, and implements energy efficiency management strategies simultaneously. It dynamically adjusts the sensor sampling rate and node sleep cycle based on event density, and achieves network self-healing based on a heartbeat detection mechanism.

[0057] Secondly, a method for intelligent security alarm based on wireless networking, implemented based on an intelligent security alarm system based on wireless networking as described in any one of claims 1-9, includes the following steps:

[0058] S1. Collect and preprocess human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data;

[0059] S2. Extract features from the preprocessed human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data to construct a security feature sequence list;

[0060] S3. Based on the security feature sequence list, construct an abnormal movement detection model and output the abnormal movement probability;

[0061] S4. Based on the security feature sequence list, construct an abnormal sound recognition model and output the probability of abnormal voiceprints;

[0062] S5. Based on the analysis results of the security feature sequence list and the security anomaly detection module, construct a security alarm model and output the alarm probability;

[0063] S6. Based on the analysis results of the security alarm model, hierarchical alarms are generated, and security alarm information is disseminated using wireless networking technology. The distribution of regional security events is presented in the form of a heat map.

[0064] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0065] 1. This invention provides an intelligent security alarm system and method based on wireless networking. Through multimodal data acquisition and feature extraction, it effectively improves the accuracy and comprehensiveness of security monitoring, can more accurately identify abnormal behavior and environmental changes, and reduces the probability of false alarms and missed alarms.

[0066] 2. This invention provides an intelligent security alarm system and method based on wireless networking. By adopting wireless networking technology, the complexity of system deployment and maintenance is simplified, the system's flexibility and scalability are enhanced, and it can adapt to security needs of different scales and environments.

[0067] 3. This invention provides an intelligent security alarm system and method based on wireless networking. Through the intelligent security management module, it realizes hierarchical alarm and heat map display, providing security personnel with an intuitive and efficient monitoring means, which helps to quickly respond to and handle security incidents, and improves the overall security efficiency and security. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0069] Figure 1 This is a block diagram of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1, such as Figure 1 As shown, the present invention provides an intelligent security alarm system based on wireless networking, including a security data acquisition module, a security feature extraction module, a security anomaly detection module, a security alarm module, and an intelligent security management module;

[0072] The security data acquisition module collects and preprocesses infrared signal intensity data of human movement, environmental sound waveform data, and three-dimensional vibration waveform data. Pyroelectric sensor nodes are installed at building entrances, corridors, and key perimeter areas using a self-organizing network protocol to collect infrared signal intensity data of human movement. The spacing between the pyroelectric sensor nodes does not exceed 8 meters, the vertical installation height is 1.2-1.5 meters, and the tilt angle is 15-30 degrees. A mesh topology is automatically constructed using a dynamic routing algorithm. Four-unit microphone arrays are deployed diagonally in the building's interior space and under the eaves to collect environmental sound waveform data. The spacing between the four microphone arrays is 1.5 meters, and the distance from the ground is 2.2-2.5 meters. They are directly connected to the gateway via a star topology. The four-unit microphone array nodes use a carrier sense multiple access (CMA) collision avoidance mechanism to compete for channel contention. Triaxial piezoelectric accelerometers are deployed on the door and window frames and the base of safes in the building. The axes of the triaxial piezoelectric accelerometers are perpendicular to the monitoring plane, fixed with epoxy resin, and connected through a hybrid networking architecture to collect data. The system collects three-dimensional vibration waveform data and preprocesses the acquired human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data. When the rate of change of human movement infrared signal intensity exceeds 0.5 microwatts per square centimeter per second, historical data recording is activated, a 10 Hz time series is generated, the arithmetic mean of five consecutive sampling points is calculated, the current point data is replaced, and instantaneous interference pulses with a duration of less than 0.5 seconds are eliminated. The signal energy of the environmental sound waveform data within each 10-millisecond window is calculated. When the energy of three consecutive windows exceeds three times the silent baseline, it is determined as the start point of the effective segment. When the energy falls back below the baseline for more than 50 milliseconds, it is determined as the end point. The sampling point values ​​within the effective segment are divided by the maximum absolute value of the original data to normalize the maximum absolute value to 1. The three-axis piezoelectric accelerometer node calibrates the local clock based on a precise time protocol to ensure that the deviation between the three-dimensional vibration waveform data and the infrared signal timestamp is less than 5 milliseconds. Based on the reading of the built-in temperature sensor, the zero-point offset of the three-axis acceleration is adjusted according to the preset temperature drift coefficient.

[0073] The security feature extraction module extracts features from preprocessed human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data, constructing a security feature sequence list. The difference between two adjacent human movement infrared signal intensity sampling points is divided by the sampling interval to obtain the infrared intensity change rate sequence. A short-time Fourier transform is performed on the normalized effective audio segments, dividing the data into windows every 25 milliseconds with an overlap of 12.5 milliseconds between adjacent windows. The sum of squared amplitudes of frequency points in the 500-2000 Hz band is divided by the sum of squared amplitudes of frequency points in the 0-8000 Hz band to obtain the energy proportion of the audio frequency band within the calculation window. The effective audio segments are passed through a triangular filter bank with 40 Mel scale distributions. The output of the triangular filters is the logarithm of the corresponding frequency band signal energy. A discrete cosine transform is performed on the logarithms of the corresponding frequency band signal energy outputs of the 40 triangular filters, and the first three Mel cepstral coefficients are taken as feature components. The total vibration energy integral E of the three-dimensional vibration waveform data within a 1-second window is calculated. Dynamic baseline normalization is applied to the total energy integral to obtain the normalized vibration energy. The calculation process is as follows:

[0074]

[0075] Among them, a x (i), a y (i) and a z (i) represents the acceleration of the i-th sampling point on the x-axis, y-axis and z-axis respectively, Δt is the sampling interval time, N is the number of sampling points, μ is the mean vibration energy in the past hour, σ is the standard deviation of vibration energy in the past hour. The extracted infrared intensity change rate sequence, audio frequency band energy ratio, first three Mel-frequency cepstral coefficients and normalized vibration energy are integrated to form a security feature sequence table.

[0076] The security anomaly detection module, combined with a security feature sequence list, performs abnormal movement detection and abnormal sound recognition. It includes a movement detection unit and a sound recognition unit. The movement detection unit is based on the isolated forest algorithm, training decision trees using historical normal infrared intensity change rate sequence data. The number and depth of the decision trees are pre-set in the edge gateway of the intelligent security alarm system. Each decision tree selects the mean and variance of the normal infrared intensity change rate of a window as splitting features. The gateway receives the latest window mean and variance, inputs them into the isolated forest model, constructs the abnormal movement detection model, traverses the decision trees, and calculates the baseline path length c(n) and the average path length L of the sample in the decision tree. p After standardization, it is transformed into the abnormal movement probability P1, and its calculation process is as follows:

[0077]

[0078] Where n is the number of samples in the training dataset, γ is the Euler-Marcheroni constant, P1∈[0,1], the sound recognition unit performs maximum-minimum normalization on the energy proportion feature of the 500-2000 Hz audio frequency band, and standard deviation normalization on the first three orders of Mel-frequency cepstral coefficients, constructing a 4-dimensional voiceprint feature vector containing the audio frequency band energy proportion, first-order Mel-frequency coefficient, second-order Mel-frequency coefficient, and third-order Mel-frequency coefficient. Based on the support vector machine classification model architecture, the radial basis kernel function is selected to map the 4-dimensional voiceprint feature vector to a high-dimensional space. The historical environmental sound waveform dataset is input to the support vector machine classification model architecture to solve for the optimal hyperplane. The sample point closest to the hyperplane is selected as the support vector β, and the trained hyperplane parameters are preset to the edge gateway of the intelligent security alarm system to construct an abnormal sound recognition model. The 4-dimensional voiceprint feature vector is input to the abnormal sound recognition model, and the distance d between it and the hyperplane is calculated. The Sigmoid function is used to map this distance to the abnormal voiceprint probability P2. The calculation process is as follows:

[0079]

[0080]

[0081] Where, α i For support vector weights, y i For support vector labels, normal = 1, abnormal = -1, K is the result of radial basis kernel function calculation, b is the bias term of classification hyperplane, P2∈[0,1], set the upper and lower limits of the threshold range of abnormal voiceprint probability, and use a hierarchical transmission strategy for data transmission;

[0082] The security alarm module, based on the analysis results of the security feature sequence list and the security anomaly detection module, constructs a security alarm model and outputs the alarm probability. The security alarm module receives the probability of abnormal movement, the probability of abnormal sound patterns, and normalized vibration energy features, integrating them into a joint feature vector. A security alarm model is constructed using a logistic regression model architecture, and the alarm probability P is calculated. The calculation process is as follows:

[0083]

[0084] Where w1, w2, w3, w4, and w5 are the weight parameters of each input feature, and b is the bias term. Based on historical alarm event data and normal environment data, the weight parameters of each input feature of the logistic regression are optimized with the goal of minimizing cross-entropy loss. The trained weight vector and bias term are solidified in the edge gateway flash memory. The differential parameter update packet is received monthly through the low-power wide-area communication channel. The latest collected abnormal movement probability, abnormal soundprint probability, and normalized vibration energy feature are input into the security alarm model to obtain the current alarm probability P.

[0085] The intelligent security management module performs tiered alarms based on the analysis results of the security alarm model. It uses wireless networking technology to publish security alarm information and presents the distribution of regional security events in the form of a heat map. Different alarm levels are defined according to the analysis results: high-risk, medium-risk, and low-risk alarm levels. Each alarm level is matched with a corresponding alarm probability threshold range. If the alarm probability falls within the high-risk alarm level's threshold range, it is determined to be an intrusion event, triggering an audible and visual alarm and pushing real-time location coordinates to the cloud. If the alarm probability falls within the medium-risk alarm level's threshold range, a local buzzer sounds intermittently and the gateway LED flashes. If the alarm probability falls within the low-risk alarm level's threshold range, an event log is stored, containing a timestamp, sensor location, and alarm probability. The system calculates the alarm probability rate by establishing a baseline environmental data every 24 hours, calculating the moving average and standard deviation of the alarm probability over the past 24 hours, and adaptively adjusting the alarm probability threshold. Sensor nodes are deployed using self-organizing network protocols, star topologies, and hybrid network architectures. Alarm event data, including location coordinates, timestamps, and alarm probabilities, is transmitted to the edge gateway. High-risk alarms are directly transmitted to the cloud via a low-power wide-area communication channel with a latency of no more than 200ms. Medium-risk alarms are transmitted via a self-organizing network protocol through multiple hops to the gateway for aggregation, with forward error correction coding enabled. Low-risk alarms are stored locally for 24 hours and then uploaded to the cloud in batches. The intelligent security management module calibrates the timestamps of each sensor node based on a precise time protocol. Sensor node location coordinates are pre-stored in a cloud database. Discrete alarm events are mapped to a building planar grid to calculate the event density D(x,y) of each grid point. The calculation process is as follows:

[0086]

[0087] Where N is the total number of alarm events in the current time period, and P iLet K be the alarm probability of the i-th event, K be the kernel function, and h be the bandwidth parameter automatically adjusted according to the sparsity of the event spatial distribution. If the event spacing exceeds 5 meters, then h = 2 meters; if the event spacing is within 2 meters, then h = 1 meter. The event density of each grid point is normalized to the [0,1] interval. A normalized density threshold is set with upper and lower limits. If the event density of a grid point exceeds the upper limit of the normalized density threshold, it is considered high density and an RGB gradient from dark red to purple is used. If the event density of a grid point is within the range of the upper and lower limits of the normalized density threshold, it is considered medium density and an RGB gradient from orange to red is used. If the event density of a grid point is lower than the normalized density threshold, it is considered medium density. If the density threshold is lowered, it is considered low density. An RGB gradient from light green to yellow is used. Based on WebGL technology, the grid density values ​​are rendered as a heat map and overlaid on the building floor plan. The client pulls the latest density matrix from the cloud every 10 seconds. The intelligent security management module selects the optimal data transmission path based on signal strength and node power consumption through a dynamic routing algorithm. It combines edge computing nodes to perform local density pre-calculation and data compression. It adopts a dual-path concurrent approach of low-power wide-area communication and self-organizing network to implement energy efficiency management strategies simultaneously. It dynamically adjusts the sensor sampling rate and node sleep cycle based on event density and realizes network self-healing based on a heartbeat detection mechanism.

[0088] Example 2, as Figure 1 As shown, based on Embodiment 1, the present invention provides a technical solution: an intelligent security alarm method based on wireless networking, implemented based on the aforementioned intelligent security alarm system based on wireless networking, comprising the following steps:

[0089] S1. Collect and preprocess human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data;

[0090] S2. Extract features from the preprocessed human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data to construct a security feature sequence list;

[0091] S3. Based on the security feature sequence list, construct an abnormal movement detection model and output the abnormal movement probability;

[0092] S4. Based on the security feature sequence list, construct an abnormal sound recognition model and output the probability of abnormal voiceprints;

[0093] S5. Based on the analysis results of the security feature sequence list and the security anomaly detection module, construct a security alarm model and output the alarm probability;

[0094] S6. Based on the analysis results of the security alarm model, hierarchical alarms are generated, and security alarm information is disseminated using wireless networking technology. The distribution of regional security events is presented in the form of a heat map.

[0095] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart security alarm system based on wireless networking, characterized in that: It includes a security data acquisition module, a security feature extraction module, a security anomaly detection module, a security alarm module, and an intelligent security management module; The security data acquisition module collects and preprocesses human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data. The security feature extraction module extracts features from the preprocessed human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data. It integrates the extracted infrared intensity change rate sequence, audio frequency band energy ratio, first three Mel-frequency cepstral coefficients, and normalized vibration energy to construct a security feature sequence table. The security anomaly detection module is used to perform abnormal movement detection and abnormal sound recognition by combining the security feature sequence list; The security alarm module, based on the analysis results of the security feature sequence list and the security anomaly detection module, constructs a security alarm model and outputs the alarm probability. The intelligent security management module performs hierarchical alarms based on the analysis results of the security alarm model, publishes security alarm information using wireless networking technology, and presents the distribution of regional security events in the form of a heat map. The security anomaly detection module includes a motion detection unit and a sound recognition unit: The motion detection unit constructs an abnormal motion detection model based on the extracted infrared intensity change rate features and outputs the probability of abnormal motion. The sound recognition unit constructs an abnormal sound recognition model based on the energy proportion characteristics of the audio frequency band and the characteristics of the first three Mel-frequency cepstral coefficients, and outputs the probability of abnormal voiceprints. The security alarm module receives abnormal movement probability, abnormal soundprint probability, and normalized vibration energy features, and integrates them into a joint feature vector. A security alarm model is constructed using a logistic regression model architecture to calculate the alarm probability P. The calculation process is as follows: ； in, and Here, b represents the weight parameters for each input feature, and b is the bias term. For abnormal movement probability, For abnormal voiceprint probability and This represents the normalized vibrational energy characteristics.

2. The intelligent security alarm system based on wireless networking according to claim 1, characterized in that: The security data acquisition module includes the following processes for acquiring and preprocessing human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data: Pyroelectric sensor nodes are installed at building entrances, corridors, and key perimeter areas using a self-organizing network protocol to collect infrared signal intensity data of human movement. A four-unit microphone array was deployed diagonally in the building's interior space and under the outdoor eaves to collect ambient sound waveform data. Triaxial piezoelectric accelerometers are deployed on the door and window frames and the base of safes in the building to collect three-dimensional vibration waveform data; The collected infrared signal intensity data of human movement, environmental sound waveform data, and three-dimensional vibration waveform data are preprocessed. Specifically, when the rate of change of the infrared signal intensity of human movement exceeds 0.5 microwatts per square centimeter per second, historical data recording is activated to generate a 10 Hz time series. The environmental sound waveform data is normalized to obtain the normalized effective audio segments. The three-dimensional vibration waveform data is time-calibrated, and the zero-point offset of the triaxial acceleration is adjusted according to the preset temperature drift coefficient.

3. The intelligent security alarm system based on wireless networking according to claim 2, characterized in that: The process of constructing the security feature sequence table in the security feature extraction module includes: Based on the preprocessed human movement infrared signal intensity data, the infrared intensity change rate is obtained by dividing the difference between two adjacent human movement infrared signal intensity sampling points by the sampling interval, and combined into an infrared intensity change rate sequence. A short-time Fourier transform is performed on the normalized effective audio segments to divide the calculation window, thereby obtaining the proportion of audio frequency band energy within the calculation window. Specifically, a window is divided every 25 milliseconds, with adjacent windows overlapping by 12.5 milliseconds. The sum of squares of the frequency point amplitudes in the 500-2000 Hz band is divided by the sum of squares of the frequency point amplitudes in the 0-8000 Hz band to obtain the proportion of audio frequency band energy within the calculation window. The effective audio segment is passed through a triangular filter bank with 40 Mel scale distributions. The output of the triangular filter is the logarithm of the signal energy of the corresponding frequency band. The discrete cosine transform is performed on the logarithm of the signal energy of the corresponding frequency band output by the 40 triangular filters, and the first three Mel cepstral coefficients are taken as characteristic components. Calculate the total vibration energy integral of three-dimensional vibration waveform data within a 1-second window. Dynamic baseline normalization is applied to the total energy integral to obtain the normalized vibrational energy. ; The extracted infrared intensity change rate sequence, audio frequency band energy proportion, first three Mel-frequency cepstral coefficients, and normalized vibration energy are integrated to form a security feature sequence table.

4. The intelligent security alarm system based on wireless networking according to claim 3, characterized in that: In the motion detection unit, the process of constructing an abnormal motion detection model and outputting the abnormal motion probability includes: Based on the isolated forest algorithm, decision trees are trained using historical normal infrared intensity change rate sequence data. The number and depth of decision trees are pre-set in the edge gateway of the intelligent security alarm system. Each decision tree selects the mean and variance of the normal infrared intensity change rate of the window as the splitting feature. The gateway receives the latest window mean and variance and inputs them into the isolated forest model to construct an abnormal movement detection model. Traverse the decision tree and calculate the baseline path length. and the average path length of the sample in the decision tree After standardization, it is converted into abnormal movement probability. .

5. The intelligent security alarm system based on wireless networking according to claim 4, characterized in that: In the sound recognition unit, the process of constructing an abnormal sound recognition model and outputting the probability of abnormal voiceprints includes: Max-min normalization was applied to the energy proportion characteristics of the 500-2000 Hz audio band, and standard deviation normalization was applied to the first three order Mel cepstral coefficients to construct a 4-dimensional voiceprint feature vector containing the audio band energy proportion, first-order Mel coefficient, second-order Mel coefficient, and third-order Mel coefficient. Based on the support vector machine (SVM) classification model architecture, a radial basis function (RBF) kernel is selected to map the 4D voiceprint feature vector to a high-dimensional space. The historical environmental sound waveform dataset is input into the SVM classification model architecture to solve for the optimal hyperplane. The sample points closest to the hyperplane are selected as support vectors. The trained hyperplane parameters are then pre-set to the edge gateway of the intelligent security alarm system to construct an abnormal sound recognition model. Input a 4D voiceprint feature vector into the abnormal sound recognition model and calculate its distance to the hyperplane. The Sigmoid function is used to map this distance to the probability of abnormal voiceprints. Set upper and lower limits for the threshold range of abnormal voiceprint probability, and adopt a hierarchical transmission strategy for data transmission.

6. The intelligent security alarm system based on wireless networking according to claim 5, characterized in that: In the security alarm module, the process of constructing the security alarm model and outputting the alarm probability includes: The security alarm module receives abnormal movement probability, abnormal sound pattern probability, and normalized vibration energy features, integrates them into a joint feature vector, and uses a logistic regression model architecture to construct a security alarm model and calculate the alarm probability P. Based on historical alarm event data and normal environment data, the weight parameters of each input feature of logistic regression are optimized with the goal of minimizing cross-entropy loss. The trained weight vector and bias term are solidified in the edge gateway flash memory and differential parameter update packets are received monthly through a low-power wide-area communication channel. The latest collected abnormal movement probability, abnormal soundprint probability, and normalized vibration energy characteristics are input into the security alarm model to obtain the current alarm probability P.

7. The intelligent security alarm system based on wireless networking according to claim 6, characterized in that: The intelligent security management module includes the following process for hierarchical alarm based on the analysis results of the security alarm model: Based on the analysis results of the security alarm model, different alarm levels are divided into high-risk alarm level, medium-risk alarm level and low-risk alarm level, and corresponding alarm probability threshold ranges are matched for each alarm level. If the alarm probability is within the alarm probability threshold range of the high-risk alarm level, it is determined to be an intrusion event, triggering the audible and visual alarm and pushing the real-time location coordinates to the cloud. If the alarm probability is within the alarm probability threshold range of the medium-risk alarm level, the local buzzer will sound intermittently and the gateway LED will flash. If the alarm probability is within the alarm probability threshold range of the low-risk alarm level, an event log will be stored. The event log includes a timestamp, sensor location, and alarm probability. The environmental baseline is statistically analyzed every 24 hours, and the moving average and standard deviation of the alarm probability over the past 24 hours are calculated to adaptively adjust the alarm probability threshold.

8. The intelligent security alarm system based on wireless networking according to claim 7, characterized in that: The intelligent security management module uses wireless networking technology to publish security alarm information and presents the distribution of regional security events in the form of a heat map, including the following processes: Each sensor node is deployed through a self-organizing network protocol, star topology, and hybrid networking architecture. Alarm event data, including location coordinates, timestamps, and alarm probabilities, is transmitted to the edge gateway. High-risk alarms are transmitted directly to the cloud through a low-power wide-area communication channel with a latency of no more than 200ms. Medium-risk alarms are transmitted to the gateway via a multi-hop transmission through the self-organizing network protocol and forward error correction coding is enabled. Low-risk alarms are stored locally for 24 hours and then uploaded to the cloud in batches. The intelligent security management module calibrates the timestamps of each sensor node based on a precise time protocol. The sensor node location coordinates are pre-stored in a cloud database, and discrete alarm events are mapped to a building planar grid to calculate the event density of each grid point. ; The event density of each grid point is normalized to the [0,1] interval. An upper and lower limit of the normalized density threshold is set. If the event density of a grid point exceeds the upper limit of the normalized density threshold, it is judged as high density and a dark red to purple RGB gradient is used. If the event density of a grid point is within the range of the upper and lower limits of the normalized density threshold, it is judged as medium density and an orange to red RGB gradient is used. If the event density of a grid point is lower than the lower limit of the normalized density threshold, it is judged as low density and a light green to yellow RGB gradient is used. Based on WebGL technology, the grid density value is rendered as a heat map layer and superimposed on the building floor plan. The client pulls the latest density matrix from the cloud every 10 seconds. The intelligent security management module selects the optimal data transmission path based on signal strength and node power using a dynamic routing algorithm. It combines edge computing nodes to perform local density pre-calculation and data compression, adopts a dual-path concurrent approach of low-power wide-area communication and self-organizing network, and implements energy efficiency management strategies simultaneously. It dynamically adjusts the sensor sampling rate and node sleep cycle based on event density, and achieves network self-healing based on a heartbeat detection mechanism.

9. A method for intelligent security alarm based on wireless networking, implemented based on an intelligent security alarm system based on wireless networking as described in any one of claims 1-8, characterized in that: Includes the following steps: S1. Collect and preprocess human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data; S2. Extract features from the preprocessed human movement infrared signal intensity data, environmental sound waveform data, and three-dimensional vibration waveform data to construct a security feature sequence list; S3. Based on the security feature sequence list, construct an abnormal movement detection model and output the abnormal movement probability; S4. Based on the security feature sequence list, construct an abnormal sound recognition model and output the probability of abnormal voiceprints; S5. Based on the analysis results of the security feature sequence list and the security anomaly detection module, construct a security alarm model and output the alarm probability; S6. Based on the analysis results of the security alarm model, hierarchical alarms are generated, and security alarm information is disseminated using wireless networking technology. The distribution of regional security events is presented in the form of a heat map.

Images