Anti-theft and anti-intrusion detection method and device based on multi-sensor fusion

Abstract

The invention provides an anti-theft and anti-intrusion detection method and device based on multi-sensor fusion. Comprising the following steps: acquiring field acquisition data of a construction suspension field, performing denoising processing on the field acquisition data based on a median filter to obtain denoised field data, performing standardization processing on the denoised field data to obtain preprocessed field data, and performing feature extraction processing on the preprocessed field data based on a convolutional neural network and fast Fourier transform to obtain preprocessed field data; obtaining field local feature data, constructing a global relation model of the construction suspension field, and obtaining a global feature vector from the global relation model based on a graph neural network; and constructing a real-time intrusion detection model based on a time sequence convolutional network, inputting the global feature vector into the real-time intrusion detection model, outputting intrusion behavior confidence, and performing anti-theft detection on a construction suspension site. The problem that in the prior art, protection on a construction suspension site is prone to being disturbed by the environment, and consequently protection safety is low is solved.

Description

Technical Field

[0001] The present application relates to the technical field of anti-theft and anti-intrusion detection technology, and in particular to an anti-theft and anti-intrusion detection method, device, computer-readable storage medium and electronic device based on multi-sensor fusion. Background Art

[0002] Construction sites are complex environments, characterized by open spaces, complex personnel flows, and the presence of expensive construction materials and equipment. Therefore, ensuring safety and security during construction pauses, including the installation of anti-theft and anti-intrusion systems, is crucial. At these sites, these systems must adapt to complex, changing, and unattended environments. Traditional anti-theft systems typically rely on a single type of sensor, which is not only susceptible to environmental interference but also limited by inclement weather. This can lead to false alarms and missed alerts, significantly reducing the security of these systems. Summary of the Invention

[0003] The main purpose of this application is to provide an anti-theft and anti-intrusion detection method, device, computer-readable storage medium and electronic device based on multi-sensor fusion, so as to at least solve the problem that the protection of the construction suspension site in the existing technology is easily affected by environmental interference, resulting in low protection security.

[0004] In order to achieve the above-mentioned purpose, according to one aspect of the present application, an anti-theft and anti-intrusion detection method based on multi-sensor fusion is provided, including: obtaining on-site collected data of the construction suspension site, and denoising the on-site collected data based on a median filter to obtain denoised field data, and standardizing the denoised field data based on a Z-score normalization algorithm to obtain pre-processed field data, wherein the on-site collected data includes visual data, infrared sensor data and physical vibration data, and the on-site collected data is collected by multiple sensors; performing feature extraction processing on the pre-processed field data based on a convolutional neural network and a fast Fourier transform to obtain on-site local feature data, and constructing a global relationship model of the construction suspension site based on the on-site local feature data, and obtaining a global feature vector from the global relationship model based on a graph neural network; constructing a real-time intrusion detection model based on a temporal convolutional network, and inputting the global feature vector into the real-time intrusion detection model, outputting an intrusion behavior confidence, and using the intrusion behavior confidence to perform anti-theft detection on the construction suspension site.

[0005] Optionally, performing denoising on the field collected data based on a median filter to obtain denoised field data includes: using the output formula of the median filter: y i =median(x1,x2,…,x n), denoising the field data to obtain the denoised field data, where (x1, x2, ..., x n ) is the on-site collected data, y i The denoised field data; performing standardization on the denoised field data based on a Z-score standardization algorithm to obtain preprocessed field data, including: using the Z-score standardization algorithm: The denoised field data is standardized to obtain pre-processed field data Z i ,in, μ is the mean of the data collected on site, and σ is the standard deviation of the data collected on site.

[0006] Optionally, feature extraction processing is performed on the preprocessed field data based on a convolutional neural network and a fast Fourier transform to obtain field local feature data, including: performing Fourier transform processing on the preprocessed physical vibration data in the preprocessed field data based on the fast Fourier transform to obtain frequency domain analysis data of the preprocessed physical vibration data; determining the amplitude value of the frequency domain analysis data, and converting the amplitude value into a frequency domain feature vector; and performing feature extraction on the frequency domain feature vector based on the convolutional neural network to obtain physical vibration feature data in the field local feature data.

[0007] Optionally, performing feature extraction on the frequency domain feature vector based on the convolutional neural network to obtain physical vibration feature data in the on-site local feature data includes: determining a first calculation formula of a convolution layer of the convolutional neural network as: Input the frequency domain feature vector into the convolution layer to obtain the vibration frequency domain feature G i，j , where w j，k is the weight of the jth convolution kernel, b j is the corresponding bias term, K is the size of the convolution kernel, ReLU is the activation function, and f i+k-1 is the frequency domain feature vector; the pooling layer of the convolutional neural network is used to perform maximum pooling processing on the vibration frequency domain feature to obtain the pooled vibration frequency domain feature; based on the Softmax function, the pooled vibration frequency domain feature is converted into the physical vibration feature data through the fully connected layer of the convolutional neural network.

[0008] Optionally, in the process of using the intrusion behavior confidence to perform anti-theft detection on the construction suspension site, the method also includes: constructing a hierarchical warning rule, wherein the hierarchical warning rule includes no intrusion behavior, low-level intrusion warning, intermediate intrusion warning and high-level intrusion alarm, the low-level intrusion warning is a suspected intrusion behavior, which is recorded and silently notified to the user, the intermediate intrusion warning is a high-probability intrusion behavior, which triggers a local alarm and remote notification at the construction suspension site, and the high-level intrusion alarm determines that there is an intrusion behavior, triggers a full alarm, issues a siren on-site warning, saves the video of the construction suspension site and contacts the alarm center; according to the intrusion behavior confidence and the hierarchical warning rules, the construction suspension site is subjected to real-time warning processing.

[0009] Optionally, after constructing the real-time intrusion detection model based on the temporal convolutional network, the method further includes: training the real-time intrusion detection model using multiple sets of training data to obtain the trained real-time intrusion detection model, wherein each set of training data in the multiple sets of training data includes: a historical global feature vector obtained from historical on-site data collected at the construction suspension site and intrusion category label data corresponding to the historical global feature, obtained within a historical time period; in the process of training the real-time intrusion detection model using the multiple sets of training data, the method further includes: according to the loss function formula: Determine the loss value of the real-time intrusion detection model, and according to the loss value, use the Adam optimizer to update the model parameters of the real-time intrusion detection model based on minimizing the loss function, where F is the loss value, is the real intrusion category label data, with a value of 0 or 1, O1 is the probability that the real-time intrusion detection model predicts the label to be 0, and O2 is the probability that the real-time intrusion detection model predicts the label to be 1.

[0010] Optionally, performing feature extraction processing on the pre-processed field data based on a convolutional neural network and a fast Fourier transform includes: determining a second calculation formula of a convolutional layer of the convolutional neural network as:

[0011] The preprocessed visual data or preprocessed infrared sensor data in the preprocessed field data is input into the convolution layer to obtain a feature map corresponding to the preprocessed visual data or the preprocessed infrared sensor data, wherein X (l+1) [i, j, k] is the feature map, X (l) is the pre-processed visual data or the pre-processed infrared sensing data, W (l) is the convolution kernel, K h is the height of the convolution kernel, K w is the width of the convolution kernel, σ is the nonlinear activation function, is the bias of the k-th convolution kernel of the l-th layer, i represents the row index of the corresponding feature map, j represents the column index of the corresponding feature map, k represents the corresponding convolution kernel in the corresponding feature map, p represents the sliding index of the corresponding convolution kernel k in height, q represents the sliding index of the corresponding convolution kernel k in width, and m represents the channel index of the corresponding feature map; the pooling layer of the convolutional neural network is used to perform maximum pooling processing on the feature map to obtain a maximum pooling feature map, and the maximum pooling feature map is subjected to dimensionality reduction processing to obtain visual feature data or infrared sensing feature data in the local feature data of the scene.

[0012] According to another aspect of the present application, an anti-theft and anti-intrusion detection device based on multi-sensor fusion is provided, comprising: a processing unit for acquiring on-site collected data of a construction suspension site, and performing denoising processing on the on-site collected data based on a median filter to obtain denoised on-site data, and performing standardization processing on the denoised on-site data based on a Z-score standardization algorithm to obtain pre-processed on-site data, wherein the on-site collected data includes visual data, infrared sensor data and physical vibration data, and the on-site collected data is collected by multiple sensors; a feature extraction unit for performing feature extraction processing on the pre-processed on-site data based on a convolutional neural network and a fast Fourier transform to obtain on-site local feature data, and constructing a global relationship model of the construction suspension site based on the on-site local feature data, and obtaining a global feature vector from the global relationship model based on a graph neural network; an anti-theft and anti-intrusion detection unit for constructing a real-time intrusion detection model based on a temporal convolutional network, and inputting the global feature vector into the real-time intrusion detection model, outputting an intrusion behavior confidence, and using the intrusion behavior confidence to perform anti-theft detection on the construction suspension site.

[0013] According to another aspect of the present application, a computer-readable storage medium is provided, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the anti-theft and anti-intrusion detection methods based on multi-sensor fusion.

[0014] According to another aspect of the present application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include means for executing any one of the anti-theft and anti-intrusion detection methods based on multi-sensor fusion.

[0015] The technical solution of the present application is applied to obtain on-site data collected at the construction suspension site, and denoise the on-site data based on a median filter to obtain denoised on-site data, and standardize the denoised on-site data based on a Z-score normalization algorithm to obtain pre-processed on-site data, wherein the on-site data includes visual data, infrared sensor data, and physical vibration data, and the on-site data is collected by multiple sensors; feature extraction is performed on the pre-processed on-site data based on a convolutional neural network and a fast Fourier transform to obtain local feature data of the site, and a global relationship model of the construction suspension site is constructed based on the local feature data of the site, and a global feature vector is obtained from the global relationship model based on a graph neural network; a real-time intrusion detection model is constructed based on a time-series convolutional network, and the global feature vector is input into the real-time intrusion detection model to output the intrusion behavior confidence, and the intrusion behavior confidence is used to perform anti-theft detection on the construction suspension site. By comprehensively collecting the visual data, sensor data, and physical vibration data of the construction suspension site and performing preprocessing, feature extraction, and other operations in sequence, noise and interference information in the data can be effectively removed and key features can be extracted. On this basis, a global relationship model and a real-time intrusion detection model are established, making the system's processing and analysis of on-site data more precise. This allows for more accurate identification of intrusions, reduces the probability of misjudgments and missed detections, and provides strong support for construction site safety management. This addresses the problem that existing technologies for protecting suspended construction sites are susceptible to environmental interference, resulting in lower security. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings that constitute part of this application are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a mobile terminal that performs an anti-theft and anti-intrusion detection method based on multi-sensor fusion according to an embodiment of the present application is shown;

[0018] Figure 2 A schematic flow chart of an anti-theft and anti-intrusion detection method based on multi-sensor fusion according to an embodiment of the present application is shown;

[0019] Figure 3 A schematic structural diagram of an anti-theft and anti-intrusion detection system based on multi-sensor fusion according to an embodiment of the present application is shown;

[0020] Figure 4 A schematic diagram of a flow chart of a specific anti-theft and anti-intrusion detection method based on multi-sensor fusion provided in accordance with an embodiment of the present application is shown;

[0021] Figure 5The figure shows a structural block diagram of an anti-theft and anti-intrusion detection device based on multi-sensor fusion provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0023] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] As introduced in the background technology, the existing technology for the protection of the construction suspension site is easily affected by environmental interference, resulting in low security of the protection. In order to solve the problem that the existing technology for the protection of the construction suspension site is easily affected by environmental interference, resulting in low security of the protection, the embodiments of the present application provide an anti-theft and anti-intrusion detection method, device, computer-readable storage medium and electronic device based on multi-sensor fusion.

[0026] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure diagram of a mobile terminal of an anti-theft and anti-intrusion detection method based on multi-sensor fusion according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1Only one is shown) a processor 102 (the processor 102 may include but is not limited to a microprocessor MCU or a programmable logic device FPGA and other processing devices) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input and output device 108 for communication functions. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0028] Memory 104 can be used to store computer programs, such as application software programs and modules, such as the computer program corresponding to the anti-theft and anti-intrusion detection method based on multi-sensor fusion in the embodiments of the present invention. Processor 102 executes the computer programs stored in memory 104 to perform various functional applications and data processing, thereby implementing the aforementioned method. Memory 104 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, memory 104 may further include memory remotely located from processor 102, which can be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. Transmission device 106 is used to receive or transmit data via a network. Specific examples of such networks may include a wireless network provided by the mobile terminal's telecommunications provider. In one example, transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0029] In this embodiment, a theft-prevention and anti-intrusion detection method based on multi-sensor fusion is provided, which runs on a mobile terminal, a computer terminal or a similar computing device. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Figure 2 It is a flowchart of an anti-theft and anti-intrusion detection method based on multi-sensor fusion according to an embodiment of the present application.

[0031] like Figure 2 As shown, the method includes the following steps:

[0032] Step S201: acquiring on-site data collected at the construction site where the construction was suspended, performing denoising processing on the on-site data using a median filter to obtain denoised on-site data, and performing normalisation processing on the denoised on-site data using a Z-score normalisation algorithm to obtain pre-processed on-site data, wherein the on-site data collected includes visual data, infrared sensor data and physical vibration data, and the on-site data collected is obtained by multiple sensors;

[0033] Specifically, several high-precision intelligent cameras are selected to collect visual data to provide multi-angle visual data, directly view the activities of intruders, conduct real-time monitoring and playback, and support automatic identification and differentiation of people, vehicles, animals, etc.; several infrared sensors are used to collect infrared sensor data, provide preliminary signals for intrusion detection, trigger the camera to record images or videos, and at the same time solve the blind spot problem of cameras in large-scale monitoring, and make up for the shortcomings of optical cameras under low-light conditions to realize the function of night monitoring; several vibration sensors are selected to collect physical vibration data, provide vibration signals caused by physical contact of intruders, such as cutting fences, knocking on doors and windows, etc., to help the system identify intrusion behaviors more accurately and activate the alarm device; the above sampling frequency is set to collect data every 0.1 second, and the time synchronization protocol NTP is used to ensure that the data of all collection units have a unified timestamp, and no subsequent time alignment and interpolation processing is required, which is more real-time and more accurate in identifying intrusion behaviors.

[0034] The Z-score normalization algorithm calculates the degree of deviation of each data point from the mean and converts it into a value under the standard normal distribution, which helps to eliminate dimensional differences between data and improve the effect of feature extraction.

[0035] Step S202: performing feature extraction processing on the pre-processed site data based on a convolutional neural network and a fast Fourier transform to obtain local feature data of the site, constructing a global relationship model of the construction suspension site based on the local feature data of the site, and obtaining a global feature vector from the global relationship model based on a graph neural network;

[0036] Specifically, convolutional neural networks (CNNs) excel at processing two-dimensional image data, automatically learning and extracting features from images. Fast Fourier transforms (FFTs) can convert time series data into frequency domain representations, making them particularly suitable for analyzing physical vibration data. Graph neural networks (GNNs) can process structured data, such as graphs or grids, making them ideal for analyzing correlations between sensors and building global relationship models.

[0037] Step S203: construct a real-time intrusion detection model based on a temporal convolutional network, input the global feature vector into the real-time intrusion detection model, output the intrusion behavior confidence, and use the intrusion behavior confidence to perform anti-theft detection on the construction suspension site.

[0038] Specifically, the Temporal Convolutional Network (TCN) can capture long-term and short-term dependencies in time series data, which is crucial for real-time intrusion detection. This enables more precise processing and analysis of field data, enabling more accurate identification of intrusions, reducing the probability of false positives and missed detections, and providing strong support for construction site safety management.

[0039] Through this embodiment, by applying the above steps S201, S202, and S203, by comprehensively collecting the visual data, sensor data, and physical vibration data of the construction suspension site, and sequentially performing preprocessing, feature extraction, and other operations, it is possible to effectively remove noise and interference information in the data and extract key features. On this basis, a global relationship model and a real-time intrusion detection model are established, making the system's processing and analysis of on-site data more accurate, thereby being able to more accurately identify intrusion behaviors, reduce the probability of misjudgment and missed judgment, and provide strong support for the safety management of the construction site. This solves the problem that the existing technology of protecting the construction suspension site is easily affected by environmental interference, resulting in low protection security.

[0040] In the specific implementation process, the above-mentioned field data is denoised based on the median filter to obtain denoised field data, including: using the output formula of the median filter: y i =median(x1,x2,…,x n ), denoising the above-mentioned field data is performed to obtain the above-mentioned denoised field data, where (x1, x2, ..., x n ) is the above-mentioned on-site collected data, y i The denoised field data is the above-mentioned denoised field data; the denoised field data is normalized based on the Z-score normalization algorithm to obtain preprocessed field data, including: using the above-mentioned Z-score normalization algorithm: The above denoised field data are standardized to obtain the pre-processed field data Z i ,in, μ is the mean of the above-mentioned on-site collected data, and σ is the standard deviation of the above-mentioned on-site collected data.

[0041] This method's median filter effectively removes random noise introduced by the external environment, such as physical vibrations caused by wind and passing vehicles. The Z-score normalization algorithm ensures fair treatment of each feature, even when data levels vary significantly between sensors, preventing certain features from dominating the subsequent feature extraction process due to their large magnitude. The combination of these two algorithms not only improves data quality but also simplifies subsequent feature extraction, allowing the model to focus on learning true intrusion patterns rather than superficial fluctuations in the data. This preprocessing method also includes, but is not limited to, Gaussian filtering, wavelet transforms, and normalization, the specific choice of which depends on the characteristics of the data and the application scenario.

[0042] Specifically, feature extraction processing is performed on the above-mentioned preprocessed field data based on a convolutional neural network and a fast Fourier transform to obtain field local feature data, including: based on the above-mentioned fast Fourier transform, Fourier transform processing is performed on the preprocessed physical vibration data in the above-mentioned preprocessed field data to obtain frequency domain analysis data of the above-mentioned preprocessed physical vibration data; the amplitude value of the above-mentioned frequency domain analysis data is determined, and the above-mentioned amplitude value is converted into a frequency domain feature vector; based on the above-mentioned convolutional neural network, feature extraction is performed on the above-mentioned frequency domain feature vector to obtain physical vibration feature data in the above-mentioned field local feature data.

[0043] The physical vibration data used in this method typically carries a large amount of time series information, and directly processing this raw data often makes it difficult to capture the key characteristics of intrusion behavior. Using the Fast Fourier Transform (FFT), we convert this time-domain information into the frequency domain, which helps identify abnormal activity at specific frequencies, such as low-frequency vibrations that may be caused by an unauthorized intruder. A convolutional neural network then performs deep learning on these frequency-domain feature vectors, automatically identifying which frequency components are associated with intrusion behavior. This process overcomes the complexity and subjectivity of vibration signal analysis in traditional methods, enabling automated and objective feature extraction, significantly improving the efficiency and accuracy of intrusion detection.

[0044] More specifically, extracting features from the frequency domain feature vector based on the convolutional neural network to obtain physical vibration feature data in the on-site local feature data includes: determining a first calculation formula of the convolutional layer of the convolutional neural network as: Input the above frequency domain feature vector into the above convolution layer to obtain the vibration frequency domain feature G i，j , where w j，k is the weight of the jth convolution kernel, b j is the corresponding bias term, K is the size of the convolution kernel, ReLU is the activation function, and f i+k-1is the above-mentioned frequency domain feature vector; the pooling layer of the above-mentioned convolutional neural network is used to perform maximum pooling processing on the above-mentioned vibration frequency domain features to obtain pooled vibration frequency domain features; based on the Softmax function, the above-mentioned pooled vibration frequency domain features are converted into the above-mentioned physical vibration feature data through the fully connected layer of the above-mentioned convolutional neural network.

[0045] This method uses the convolutional layer of a convolutional neural network to extract local features from the input data through a sliding window. The ReLU activation function increases the model's nonlinear expression capabilities, helping the model learn more complex features. The maximum pooling operation further enhances the model's ability to resist deformation, allowing it to maintain a high recognition rate when faced with intrusions of different angles and sizes. The Softmax function converts the feature vector into a probability distribution, which intuitively reflects the possibility of intrusion contained in the physical vibration data. The entire process is automatically completed through a deep learning framework, eliminating the need for manual feature design, greatly reducing the difficulty of model development while improving its generalization capabilities. In addition to physical vibration features, this process is also applicable to feature extraction from visual and infrared data, including but not limited to edge detection, texture analysis, temperature change recognition, etc., further enriching the model's input information.

[0046] Furthermore, in the process of using the above-mentioned intrusion behavior confidence to conduct anti-theft detection on the above-mentioned construction suspension site, the above-mentioned method also includes: constructing a hierarchical warning rule, wherein the above-mentioned hierarchical warning rule includes no intrusion behavior, low-level intrusion warning, intermediate intrusion warning and high-level intrusion alarm, the above-mentioned low-level intrusion warning is a suspected intrusion behavior, which is recorded and silently notified to the user, the above-mentioned intermediate intrusion warning is a high-probability intrusion behavior, which triggers a local alarm and remote notification at the above-mentioned construction suspension site, and the above-mentioned high-level intrusion alarm determines that there is an intrusion behavior, triggers a full alarm, issues a siren on-site warning, saves the video of the above-mentioned construction suspension site and contacts the alarm center; according to the above-mentioned intrusion behavior confidence and the above-mentioned hierarchical warning rules, the above-mentioned construction suspension site is subjected to real-time warning processing.

[0047] Specifically, based on the confidence level and predicted value of intrusion behavior identified by the optimized real-time intrusion detection model, early warning classification is performed, enabling appropriate response measures to be taken based on the degree of security risk. This scientific early warning classification mechanism rationally allocates security resources, enabling timely and intensive early warnings and emergency response measures for high-risk intrusions, while more moderate monitoring measures can be adopted for low-risk situations, significantly improving construction site safety while also enhancing safety management efficiency and resource utilization.

[0048] This method utilizes a hierarchical warning system designed to take into account the severity and urgency of intrusions. Different thresholds are set to differentiate warning levels. This strategy reduces false alarms and avoids unnecessary panic, while also ensuring a rapid response in truly urgent situations. For example, a low-level intrusion warning might be a minor vibration caused by an animal or natural phenomenon. The system records these events and discreetly notifies the user, facilitating subsequent analysis to determine whether sensitivity adjustment is necessary. A mid-level warning might be a person or vehicle passing by. In this case, the system triggers an on-site alarm and sends a notification to relevant personnel, prompting them to conduct an inspection. A high-level intrusion alarm, however, occurs when the system confirms an illegal intrusion, immediately triggering a full-scale alarm to ensure on-site safety. This warning mechanism not only improves the system's intelligence and adaptability, but also reduces the workload of monitoring personnel, allowing for more efficient resource allocation.

[0049] Furthermore, after constructing the real-time intrusion detection model based on the temporal convolutional network, the method further includes: training the real-time intrusion detection model with multiple sets of training data to obtain the trained real-time intrusion detection model, wherein each set of training data includes: a historical global feature vector obtained from historical on-site data collected at the construction suspension site and intrusion category label data corresponding to the historical global feature, obtained during a historical time period; in the process of training the real-time intrusion detection model with multiple sets of training data, the method further includes: according to the loss function formula: Determine the loss value of the real-time intrusion detection model, and according to the loss value, use the Adam optimizer to update the model parameters of the real-time intrusion detection model based on minimizing the loss function, where F is the loss value, is the real intrusion category label data, which takes a value of 0 or 1. O1 is the probability that the above real-time intrusion detection model predicts the label to be 0, and O2 is the probability that the above real-time intrusion detection model predicts the label to be 1.

[0050] The training of this method's real-time intrusion detection model is a core component of the entire system, determining its predictive capabilities and stability. Temporal convolutional networks (TCNs) can effectively process time series data and capture the temporal dynamics of intrusion behavior. By training with large amounts of labeled historical data, the model can learn typical patterns of intrusion behavior and quickly make judgments based on new data. This training process includes, but is not limited to, the use of supervised learning, semi-supervised learning, or reinforcement learning algorithms, depending on the amount and type of available data. Once trained, the model can monitor suspended construction sites in real time without human intervention, promptly identifying potential intrusion threats. This is crucial for ensuring construction site safety.

[0051] Specifically, the feature extraction process is performed on the pre-processed field data based on the convolutional neural network and the fast Fourier transform, including: determining the second calculation formula of the convolution layer of the convolutional neural network as:

[0052] The preprocessed visual data or preprocessed infrared sensor data in the preprocessed field data is input into the convolution layer to obtain a feature map corresponding to the preprocessed visual data or the preprocessed infrared sensor data, wherein X (l+1) [i, j, k] is the above feature map, X (l) is the pre-processed visual data or the pre-processed infrared sensor data, W (l) is the convolution kernel, K h is the height of the convolution kernel, K w is the width of the convolution kernel, σ is the nonlinear activation function, is the bias of the k-th convolution kernel in the l-th layer, i represents the row index of the corresponding feature map, j represents the column index of the corresponding feature map, k represents the corresponding convolution kernel in the corresponding feature map, p represents the sliding index of the corresponding convolution kernel k in height, q represents the sliding index of the corresponding convolution kernel k in width, and m represents the channel index of the corresponding feature map; the pooling layer of the above-mentioned convolutional neural network is used to perform maximum pooling processing on the above-mentioned feature map to obtain a maximum pooling feature map, and the above-mentioned maximum pooling feature map is subjected to dimensionality reduction processing to obtain visual feature data or infrared sensing feature data in the local feature data of the scene.

[0053] Optimizing model parameters in this method is a key step in improving model performance. The loss function reflects the gap between the model's predictions and the true labels. By minimizing the loss function, we can adjust the model parameters to better fit the training data. The Adam optimizer is an adaptive learning rate optimization algorithm that dynamically adjusts the learning rate based on historical gradient information, accelerating the convergence process while avoiding the risk of overfitting. This optimization process is not limited to the Adam algorithm, but also includes but is not limited to optimization algorithms such as SGD (stochastic gradient descent) and RMSprop. The specific choice depends on the complexity of the model and the characteristics of the data. Through continuous iterative optimization, the real-time intrusion detection model can gradually improve its ability to identify intrusion behaviors, ultimately achieving ideal detection results.

[0054] In addition, this embodiment also includes adaptive threshold adjustment. Traditional intrusion detection systems often use fixed thresholds to determine the alarm level. However, in actual applications, environmental factors at the construction site (such as weather, time of day, background noise, etc.) may change over time, resulting in the fixed threshold not being able to adapt to all situations, thus affecting the accuracy of detection. To this end, we propose an adaptive threshold adjustment mechanism that can dynamically adjust the confidence threshold for identifying intrusion behavior based on environmental changes. The specific implementation steps are as follows:

[0055] 1) Environmental perception: Use additional environmental sensors (such as temperature, humidity, and light sensors) to monitor the real-time environmental status of the construction site.

[0056] 2) Threshold learning: A model is trained based on a machine learning algorithm (such as support vector machine (SVM) or random forest) that can automatically adjust the confidence threshold for identifying infringements based on the environmental state.

[0057] 3) Real-time application: When the system detects environmental changes (such as night or bad weather), it immediately calls the environmental perception model to predict the new threshold and applies it to the early warning classification processing of the real-time intrusion detection model.

[0058] The adaptive threshold adjustment of this method can significantly reduce false alarms and missed alarms caused by environmental changes. The system can maintain stable performance in various environments and effectively detect intrusions even under harsh conditions.

[0059] This embodiment also includes a deep reinforcement learning decision-making strategy. This embodiment can further integrate deep reinforcement learning (DRL) to optimize alarm handling strategies and response speed. The DRL model can learn how to best respond in different situations, such as when to send an alarm, when to call in additional resources (such as police or drones), and how to most effectively utilize existing security resources. The specific implementation steps are as follows:

[0060] 1) State space definition: The current state of the construction site (including visual, sensor, vibration data and environmental factors) is used as the state space of the DRL model.

[0061] 2) Action space definition: Define a series of possible actions, such as "send a low-level alarm", "activate a local alarm", "call a high-level alarm", and "start an emergency response plan".

[0062] 3) Reward mechanism design: Define the reward / penalty mechanism based on the accuracy of the alarm, response speed, resource consumption, etc.

[0063] 4) DRL model training: In a simulated environment, use the Deep Q Network (DQN) or Policy Gradient algorithm to train the DRL model so that it learns to select the optimal action under various conditions.

[0064] 5) Real-time deployment: The trained DRL model is deployed in the real-time intrusion detection optimization module. Whenever a suspected intrusion behavior is detected, the DRL model will predict the best response strategy based on the current status and execute it.

[0065] This DRL model can make more informed alarm handling decisions based on comprehensive on-site conditions, avoiding wasted resources. By learning the optimal alarm response path, the system can take appropriate measures more quickly, improving the timeliness and effectiveness of alarms. Furthermore, DRL can automatically adjust response strategies to match available resources, ensuring maximum system security within limited resource conditions.

[0066] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the anti-theft and anti-intrusion detection method based on multi-sensor fusion of the present application will be described in detail below with reference to specific embodiments.

[0067] This embodiment relates to a specific anti-theft and anti-intrusion detection system based on multi-sensor fusion, such as Figure 3 As shown, the background center of the anti-theft and anti-intrusion detection system includes a data acquisition module, a data preprocessing module, a multi-source data feature extraction module, and a real-time intrusion detection optimization module;

[0068] Data acquisition module, used to collect visual data, sensor data and physical vibration data of the construction suspension site;

[0069] The data preprocessing module denoises the visual data, sensor data, and physical vibration data of the construction suspension site using a median filter and standardizes the denoised data using a Z-score standardization algorithm to obtain preprocessed visual data, preprocessed sensor data, and preprocessed physical vibration data;

[0070] The multi-source data feature extraction module extracts features from pre-processed visual data, pre-processed sensor data, and pre-processed physical vibration data based on convolutional neural networks and fast Fourier transforms, obtains local visual feature data, sensor local feature data, and physical vibration local feature data, establishes a global relationship model, and obtains a global feature vector based on a graph neural network.

[0071] The real-time intrusion detection optimization module establishes a real-time intrusion detection model based on the temporal convolutional network, optimizes the real-time intrusion detection model based on the Adam optimizer, obtains the confidence level of intrusion behavior identification and intrusion behavior prediction value, and performs early warning classification processing based on the above-mentioned confidence level of intrusion behavior identification and intrusion behavior prediction value.

[0072] It should be further explained that, in the specific implementation process, the specific process of collecting visual data, sensor data, and physical vibration data at the construction site includes:

[0073] The preset data acquisition device includes: an image data acquisition unit, an infrared sensor data acquisition unit and a vibration sensor data acquisition unit, and sets the same starting sampling time and sampling frequency;

[0074] The image data acquisition unit is used to collect picture data and video frame data of the area corresponding to the construction suspension site, which is recorded as visual data; the infrared sensor data acquisition unit is used to collect a two-dimensional thermal map of the area corresponding to the construction suspension site, which is recorded as sensor data; the vibration sensor data acquisition unit is used to collect vibration sensor data of the area corresponding to the construction suspension site, which is recorded as physical vibration data;

[0075] It should be further explained that the data acquisition device is powered by a portable photovoltaic power system, which features flexibility, portability, and independent operation, making it suitable for unmanned construction sites during construction pauses. The image data acquisition unit uses several high-precision intelligent cameras to provide multi-angle visual data, allowing direct viewing of intruder activities, real-time monitoring and playback, and automatic identification and differentiation of people, vehicles, animals, etc. The infrared sensing data acquisition unit is composed of several infrared sensors, which provide preliminary intrusion detection signals, triggering the camera to record images or videos. This solves the blind spot problem of cameras in large-scale monitoring and compensates for the shortcomings of optical cameras in low-light conditions, enabling nighttime monitoring. The vibration sensing data acquisition unit uses several vibration sensors to provide vibration signals caused by physical contact by intruders, such as cutting fences or knocking on doors and windows, helping the system to more accurately identify intrusions and activate the alarm. The sampling frequency is set to collect data every 0.1 second, using the time synchronization protocol NTP to ensure that the data of all acquisition units has a unified timestamp, eliminating the need for subsequent time alignment and interpolation processing, resulting in higher real-time performance and more accurate identification of intrusions.

[0076] It should be further explained that this embodiment relates to a specific anti-theft and anti-intrusion detection method based on multi-sensor fusion, such as Figure 4As shown in the figure, first, the visual data, sensor data and physical vibration data of the construction suspension site are collected, and then denoising and standardization processing are performed to generate preprocessed data. The preprocessed data is then subjected to feature extraction processing to establish a global relationship model and generate a global feature vector. Then, a real-time intrusion detection model is constructed and optimized. The global feature vector is input into the real-time intrusion detection model to obtain the confidence of identifying intrusion behavior and the intrusion behavior detection value. A hierarchical warning rule is constructed, and the severity of the intrusion behavior is calculated according to the hierarchical warning rule. The hierarchical warning rule includes no intrusion behavior (in line with the severity interval S1), low-level intrusion warning (in line with the severity interval S2), medium-level intrusion warning (in line with the severity interval S3) and high-level intrusion alarm (in line with the severity interval S4). Low-level intrusion warning is suspected intrusion behavior, which is recorded and silently notified to the user. Medium-level intrusion warning is a high-probability intrusion behavior, which triggers a local alarm and remote notification at the construction suspension site. High-level intrusion alarm determines that there is intrusion behavior, triggers a full alarm, sounds a siren on-site warning, saves the video of the construction suspension site and contacts the alarm center.

[0077] In the specific implementation process, the specific process of pre-processing the visual data, sensor data and physical vibration data of the construction suspension site includes:

[0078] Acquire visual data, sensor data, and physical vibration data from construction halt sites;

[0079] The noise in the corresponding data collected by the corresponding data acquisition unit is removed based on the median filter, and the denoised data is standardized based on the Z-score normalization algorithm to eliminate the differences between different dimensions.

[0080] The output formula of the above median filter is: i =median(x1,x2,…,x n ); where (x1, x2, ..., x n ) is a set of data points collected by the corresponding data collection unit;

[0081] Formulas for normalizing data include: Where n is the window size, μ is the mean of the data points collected by the corresponding data acquisition unit; σ is the standard deviation of the data points collected by the corresponding data acquisition unit, and Z i is the corresponding standardized data;

[0082] For example, the physical vibration data is preprocessed, and the specific process includes:

[0083] Set the window size w, set the window size according to the actual situation;

[0084] Starting from the first data point of the above physical vibration data, the sliding window traverses the entire sequence, and each point is replaced by the median of the data in the window. According to the output formula y of the above median filter i =median(x1,x2,…,x w ), obtaining a temperature change sequence in the filtered physical vibration data or sensor data;

[0085] According to the standardized formula: Obtain standardized physical vibration data or temperature change sequence in sensor data;

[0086] Preprocess the visual data and sensor data. The specific process includes:

[0087] According to each pixel in the visual data and sensor data, the window size is set to v×v;

[0088] Extract all pixel values ​​within the window, calculate the median, replace the original pixel values, and obtain the filtered visual data and the two-dimensional heat map in the sensor data;

[0089] It should be further explained that for edge pixels in visual data, edge filling can be used when the window exceeds the boundary.

[0090] Expand the two-dimensional matrix into a one-dimensional vector (row or column by row) and calculate the mean and standard deviation of the visual data and sensor data;

[0091] For each pixel value, according to the standardized formula: Calculate the global mean and standard deviation; obtain standardized visual data and sensor data;

[0092] The processed visual data, sensory data, and physical vibration data are respectively recorded as preprocessed visual data, preprocessed sensory data, and preprocessed physical vibration data.

[0093] It should be further explained that, in the specific implementation process, the specific process of extracting features from the pre-processed visual data, pre-processed sensor data, and pre-processed physical vibration data includes:

[0094] The specific process of feature extraction for preprocessed visual data includes:

[0095] Based on the convolutional neural network, the preprocessed visual data is extracted, and the preprocessed visual data is marked as I, and the preprocessed visual data I∈R H×W×C ;It should be further explained that the convolutional neural network consists of multiple layers of convolution operations;

[0096] Set up convolutional layers to preprocess visual data;

[0097] The input of the first layer of the convolutional layer of the preprocessed visual data is denoted as X (l) , the convolution kernel is recorded as W (l) , the calculation formula for obtaining the visual data output feature map corresponding to the preprocessed visual data I is:

[0098] in, And X (l+1) The visual data output feature map of the l+1th layer preprocessed visual data I; K h is the height of the convolution kernel; K w is the width of the convolution kernel; σ is the nonlinear activation function; is the bias of the k-th convolution kernel in the l-th layer, i represents the row index of the corresponding feature map, j represents the column index of the corresponding feature map, k represents the corresponding convolution kernel in the corresponding feature map, p represents the sliding index of the corresponding convolution kernel k in height, q represents the sliding index of the corresponding convolution kernel k in width, and m represents the channel index of the corresponding feature map.

[0099] Set up a pooling layer for preprocessing visual data. It should be further explained that by downsampling, based on the maximum pooling method, the resolution of the visual data output feature map is reduced to reduce the computational complexity while retaining key features.

[0100] The above maximum pooling formula is:

[0101]

[0102] According to the above preprocessed visual data convolution layer and preprocessed visual data pooling layer, after L layers of convolution, a low-resolution and high-dimensional visual data output feature map is obtained, which is recorded as X (L) and

[0103] Setting a pre-processed visual data fully connected layer; it should be further explained that the pre-processed visual data fully connected layer further reduces the dimension of the visual data output feature map of the high-dimensional features output by the pre-processed visual data convolution layer;

[0104] Output feature map X for the above visual data (L) Perform dimensionality reduction processing, including:

[0105] The output formula of the fully connected layer of the above preprocessed visual data is:

[0106] Z=Flatten(X (L) );in, Z is the visual one-dimensional vector;

[0107] According to the above visual one-dimensional vector Z, the visual feature one-dimensional vector z is obtained. The above visual feature one-dimensional vector z is:

[0108] z=σ(W fc ×Z+b fc ); where W fc is the weight matrix of the fully connected layer for preprocessing visual data; b fc is the corresponding bias term.

[0109] The specific process of feature extraction for preprocessed sensor data includes:

[0110] Based on the convolutional neural network, the pre-processed sensor data is extracted, and the pre-processed sensor data is marked as F, and the pre-processed sensor data F∈R H×W×C ;

[0111] Set up convolutional layers to pre-process sensor data;

[0112] The input of the lth layer of the above convolutional layer is denoted as Y (l) , the convolution kernel is recorded as V (l) , the calculation formula for obtaining the corresponding output feature map is:

[0113] in, And Y (l+1) K is the sensor data output feature map of the l+1th layer preprocessed sensor data F; h is the height of the convolution kernel; K w is the width of the convolution kernel; σ is the nonlinear activation function; is the bias of the k-th convolution kernel in the l-th layer.

[0114] Set up a pooling layer for preprocessing sensor data;

[0115] According to the above pre-processed sensor data pooling layer, the maximum pooling formula is obtained as follows:

[0116]

[0117] According to the above-mentioned pre-processed sensor data convolution layer and pre-processed sensor data pooling layer, after L layers of convolution, a low-resolution and high-dimensional sensor data output feature map is obtained, which is recorded as Y (L) and

[0118] Set up a fully connected layer for preprocessing sensor data;

[0119] Output feature map Y for the above sensor data (L) Perform dimensionality reduction processing, including:

[0120] The output formula of the fully connected layer of the above preprocessed sensor data is:

[0121] S=Flatten(Y(L) );in, S is the sensing one-dimensional vector;

[0122] According to the above-mentioned one-dimensional sensing vector S, the one-dimensional sensing feature vector s is obtained. The above-mentioned one-dimensional sensing feature vector s is:

[0123] s=σ(W fs ×S+b fc ); where W fs is the weight matrix of the fully connected layer for preprocessing sensor data; b fs is the corresponding bias term.

[0124] The specific process of feature extraction for preprocessed physical vibration data includes:

[0125] The pre-processed physical vibration data is recorded as x(t) = {x1, x2, ..., x n};

[0126] Based on the fast Fourier transform, the pre-processed physical vibration data x(t) is subjected to fast Fourier transform to obtain the frequency domain analysis formula of the pre-processed physical vibration data:

[0127] X(f)=FFT(x(t));

[0128] According to the above pre-processed physical vibration data frequency domain analysis X(f), the corresponding amplitude value |X(f)| is obtained. The above amplitude value According to the above |X(f)|, the vibration frequency domain eigenvector F is obtained freq , the above vibration frequency domain eigenvector F freq for:

[0129] F freq ={|X(f1)|, |X(f2)|,...,|X(f m )|}; where f is the frequency,

[0130] Based on the convolutional neural network, the vibration frequency domain feature vector F freq Perform feature extraction, including:

[0131] A pre-processing physical vibration data convolution layer is set; it should be further explained that the pre-processing physical vibration data convolution layer is used to process the vibration frequency domain feature vector F freq Perform frequency domain feature extraction;

[0132] According to the above preprocessed physical vibration data convolution layer, the corresponding vibration frequency domain features are obtained Among them, w j，k is the weight of the jth convolution kernel; b jis the corresponding bias term; K is the size of the convolution kernel; ReLU is the activation function;

[0133] Set up a pre-processed physical vibration data pooling layer; it should be further explained that the above-mentioned pre-processed physical vibration data pooling layer is used to reduce the vibration frequency domain feature G i，j characteristic dimensions;

[0134] Based on maximum pooling, the vibration data output feature p is obtained j , the above vibration data output feature p i for:

[0135] p i =max(G i，j+u ); where u is the pooling window size;

[0136] Set up a pre-processed physical vibration data fully connected layer; it should be further explained that the pre-processed physical vibration data fully connected layer is used to output the vibration data feature p i Convert to probability distribution;

[0137] Based on the Softmax function, the vibration data output feature p i Converted into probability distribution, the formula is:

[0138] y=Softmax(w p ×p i +b p ); where y is the corresponding vibration data output feature p i The vibration probability vector of w p is the weight matrix; b p is the bias vector.

[0139] It should be further explained that, in the specific implementation process, the specific process of establishing the global relationship model includes:

[0140] Obtain the visual feature one-dimensional vector z, the sensor feature one-dimensional vector s and the vibration probability vector y; It should be further explained that the above visual feature one-dimensional vector z∈R d , sensor feature one-dimensional vector s∈R d And the vibration probability vector y∈R d ;

[0141] According to the above visual feature one-dimensional vector z, sensor feature one-dimensional vector s and vibration probability vector y, an initial feature matrix is ​​generated And H (0) ∈R N×d ; Where N is the number of corresponding sensors and d is the dimension of the corresponding feature vector.

[0142] Construct a feature graph model, specifically including:

[0143] Construct a graph G = (V, E), where V is the set of sensor nodes (including cameras, infrared sensors, and vibration sensors) of size N; E represents the relationship between corresponding sensors;

[0144] Define the adjacency matrix A∈R N×N , indicating the connection relationship between the corresponding sensors; setting the weight A ab , represents the relationship strength between sensors a and b; it should be further explained that the above weight A ab Set based on physical location, relevance, domain knowledge, etc.

[0145] Define the self-loop matrix O∈R N×N ; According to the above adjacency matrix A and self-loop matrix O, obtain the self-loop adjacency matrix It needs to be further explained that Indicates the connection of the sensor node itself; represents the original weight when a≠b;

[0146] Based on the graph neural network, feature fusion is performed on the one-dimensional visual feature vector z, the one-dimensional sensor feature vector s, and the vibration probability vector y, specifically including:

[0147] Normalize the adjacency matrix A to obtain the normalized adjacency matrix A * , the above normalized adjacency matrix A * for:

[0148] in, is the degree matrix; it needs to be further explained that the degree matrix for:

[0149] For the above initial feature matrix H (0) And the normalized adjacency matrix A * Perform the first layer of graph convolution operation. The specific formula is:

[0150] H (1) =ReLU(A * H (0) W (0) +b (0) ); where W (0) is the weight matrix and W (0) ∈R d×d* ;d * is the hidden layer dimension; b (0) is the corresponding bias vector;

[0151] After performing three-layer graph convolution according to the above operations, the final feature matrix H is obtained (L),It needs to be further explained that the higher order fusion features mentioned above ,help capture a wider range of relational information while avoiding ,oversmoothing caused by too many layers; at this point the feature vector of each ,sensor node already contains the global information aggregated from other ,sensor nodes.

[0152] Based on global average pooling, the final feature matrix H is obtained (L) The global eigenvector h fusion , the above global eigenvector h fusion for:

[0153] Wherein, N is the number of corresponding sensors; represents the feature vector of the corresponding sensor node;

[0154] It should be further explained that, in the specific implementation process, the specific process of establishing a real-time intrusion detection model includes:

[0155] According to the set sampling time and sampling frequency, several groups of global feature vectors h are obtained. fusion ;

[0156] For several sets of global eigenvectors h fusion The groups are numbered and recorded as g = 1, 2, 3, ..., c; c is a natural number;

[0157] The global eigenvector h of the cs group fusion As sample data, and s is a natural number less than c, and using the above sample data, the mean of the sample data is obtained, which is recorded as the sample set;

[0158] The remaining sets of historical alarm information and historical deviation analysis reports are used as test sets;

[0159] According to the above sample set and test set, a training sample set is formed;

[0160] Based on the temporal convolutional network, a standard intrusion detection model is constructed;

[0161] The training sample set is input into the standard intrusion detection model to train the standard intrusion detection model. The standard intrusion detection model after training is recorded as the real-time intrusion detection model Z TCN , the above real-time intrusion detection model Z TCN is: Z TCN =TCN(h fusion );

[0162] According to the above real-time intrusion detection model Z TCN , obtain the corresponding global feature Z g ; The above global feature Z g for:

[0163] The above global feature Z g Input into the fully convolutional neural network model to obtain the category two-dimensional vector The specific process includes:

[0164] Set the above fully convolutional neural network model to K layers;

[0165] The above global feature Z g As the initial input global feature, and recorded as H (0) ;

[0166] According to the above fully convolutional neural network model, the K layers and the initial input global feature H (0) , obtain the output global feature H (k) ;

[0167] The above output global feature H (k) for:

[0168] H (k) =ReLU(W (k) H (k-1) +b (k) ), k=1, 2, ..., K-1; where W (k) is the corresponding weight; b (k) is the corresponding bias term; H (k-1) is the corresponding input global feature;

[0169] Set the intrusion behavior categories to be identified, including normal categories and intrusion categories;

[0170] At the Kth layer, based on the Softmax activation function, we obtain the category two-dimensional vector The above two-dimensional vector for:

[0171] Among them, O1 and O2 represent the probability that the input belongs to the normal category and the intrusion category respectively.

[0172] It should be further explained that, in the specific implementation process, the specific process of optimizing the above-mentioned real-time intrusion detection model includes:

[0173] The neural network is trained based on the corresponding intrusion behavior. The cross-entropy loss function is used to measure the difference between the predicted category and the true label. The Adam optimizer is selected for optimization. The specific process includes:

[0174] The above two-dimensional vector Input into the neural network to obtain the category prediction value;

[0175] The error between the above category prediction value and the true label is calculated through the loss function, and the formula is:

[0176] Among them, F is the loss value; is the true label, which takes a value of 0 or 1; O1 is the probability that the category predicted by the real-time intrusion detection model is 0; O2 is the probability that the category predicted by the real-time intrusion detection model is 1;

[0177] The parameters of the real-time intrusion detection model are updated according to the Adam optimizer. Based on minimizing the loss function, the update formula is:

[0178] in, is the learning rate; θ is the model parameter of the real-time intrusion detection model.

[0179] It should be further explained that, in the specific implementation process, the specific process of early warning classification includes:

[0180] According to the optimized real-time intrusion detection model, the confidence γ of the identified intrusion behavior and the intrusion behavior prediction value δ are obtained;

[0181] According to the above-mentioned intrusion behavior confidence γ and intrusion behavior prediction value δ, the intrusion behavior severity ε is obtained; the above-mentioned intrusion behavior severity ε is:

[0182] ε=v1×γ+v2×δ; where v1 and v2 are corresponding weights;

[0183] If the severity of the intrusion behavior ε∈[0, 50%], there is no intrusion behavior;

[0184] If the severity of the intrusion behavior ε∈(50%, 70%], it is suspected to be an intrusion behavior and is identified as a low-level warning. The system records the data and silently notifies;

[0185] If the severity of the intrusion behavior ε∈(70%, 90%], the intrusion behavior with a higher probability is identified as a medium-level warning, triggering a local alarm and remote notification;

[0186] If the severity of the intrusion behavior ε∈(90%, 100%), the high-confidence intrusion behavior is identified as a high-level alarm, triggering a full alarm, issuing a siren on-site warning, saving the on-site video and contacting the alarm center.

[0187] Compared with the prior art, the beneficial effects of this embodiment include:

[0188] 1. By comprehensively collecting visual, sensory, and physical vibration data from suspended construction sites, and then performing preprocessing and feature extraction, we can effectively remove noise and interference from the data and extract key features. Based on this, we establish a global relationship model and a real-time intrusion detection model, making the system's processing and analysis of on-site data more precise. This allows for more accurate identification of intrusions, reduces the probability of false positives and missed detections, and provides strong support for construction site safety management.

[0189] 2. Based on the confidence level and predicted value of intrusion behavior identified by the optimized real-time intrusion detection model, early warning classification is performed, enabling appropriate response measures to be taken according to the degree of security risk. This scientific early warning classification mechanism can rationally allocate security resources. For high-risk intrusion behaviors, high-intensity early warnings can be issued in a timely manner and emergency response measures can be taken. For low-risk situations, more moderate monitoring methods can be adopted, greatly improving the safety of the construction site, while also enhancing the efficiency of safety management and resource utilization.

[0190] The embodiment of the present application also provides an anti-theft and anti-intrusion detection device based on multi-sensor fusion. It should be noted that the anti-theft and anti-intrusion detection device based on multi-sensor fusion of the embodiment of the present application can be used to execute the anti-theft and anti-intrusion detection method based on multi-sensor fusion provided by the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0191] The following introduces the anti-theft and anti-intrusion detection device based on multi-sensor fusion provided in an embodiment of the present application.

[0192] Figure 5 Schematic diagram of an anti-theft and anti-intrusion detection device based on multi-sensor fusion according to an embodiment of the present application.

[0193] like Figure 5 As shown, the device includes:

[0194] Processing unit 51 is configured to obtain field data collected at the construction site, perform denoising on the field data using a median filter to obtain denoised field data, and perform normalization on the denoised field data using a Z-score normalization algorithm to obtain preprocessed field data, wherein the field data includes visual data, infrared sensor data, and physical vibration data, and is collected by multiple sensors.

[0195] a feature extraction unit 52 for performing feature extraction processing on the pre-processed site data based on a convolutional neural network and a fast Fourier transform to obtain local site feature data, constructing a global relationship model of the construction suspension site based on the local site feature data, and obtaining a global feature vector from the global relationship model based on a graph neural network;

[0196] The anti-theft and anti-intrusion detection unit 53 is used to build a real-time intrusion detection model based on a temporal convolutional network, input the above-mentioned global feature vector into the above-mentioned real-time intrusion detection model, output the intrusion behavior confidence, and use the above-mentioned intrusion behavior confidence to perform anti-theft detection on the above-mentioned construction suspension site.

[0197] In this embodiment, a processing unit is used to obtain on-site collected data of the construction suspension site, and denoise the on-site collected data based on a median filter to obtain denoised field data, and standardize the denoised field data based on a Z-score standardization algorithm to obtain preprocessed field data, wherein the on-site collected data includes visual data, infrared sensor data and physical vibration data, and the on-site collected data is collected by multiple sensors; a feature extraction unit is used to perform feature extraction processing on the preprocessed field data based on a convolutional neural network and a fast Fourier transform to obtain local feature data of the site, and construct a global relationship model of the construction suspension site based on the local feature data of the site, and obtain a global feature vector from the global relationship model based on a graph neural network; an anti-theft and anti-intrusion detection unit is used to construct a real-time intrusion detection model based on a temporal convolutional network, and input the global feature vector into the real-time intrusion detection model, output the intrusion behavior confidence, and use the intrusion behavior confidence to perform anti-theft detection on the construction suspension site. By comprehensively collecting visual, sensory, and physical vibration data from suspended construction sites, and then performing preprocessing and feature extraction, the system effectively removes noise and interference from the data and extracts key features. Based on this, a global relationship model and a real-time intrusion detection model are established, enabling the system to more accurately process and analyze on-site data. This allows for more precise identification of intrusions, reduces the probability of false positives and missed detections, and provides strong support for construction site safety management. This addresses the issue of existing technologies that protect suspended construction sites, which are susceptible to environmental interference and result in lower security.

[0198] As an optional solution, the processing unit includes a denoising processing module and a standardization processing module; the denoising processing module is used to adopt the output formula of the median filter: y i =median(x1,x2,…,x n ), denoising the above-mentioned field data is performed to obtain the above-mentioned denoised field data, where (x1, x2, ..., x n ) is the above-mentioned on-site collected data, yi The denoised field data is as described above; the standardization processing module is used to adopt the above Z-score standardization algorithm: The above denoised field data are standardized to obtain the pre-processed field data Z i ,in, μ is the mean of the above-mentioned on-site collected data, and σ is the standard deviation of the above-mentioned on-site collected data.

[0199] An optional solution, the feature extraction unit includes a Fourier transform processing module, a first conversion module and a feature extraction module; the Fourier transform processing module is used to perform Fourier transform processing on the preprocessed physical vibration data in the above-mentioned preprocessed field data based on the above-mentioned fast Fourier transform, so as to obtain frequency domain analysis data of the above-mentioned preprocessed physical vibration data; the first conversion module is used to determine the amplitude value of the above-mentioned frequency domain analysis data, and convert the above-mentioned amplitude value into a frequency domain feature vector; the feature extraction module is used to perform feature extraction on the above-mentioned frequency domain feature vector based on the above-mentioned convolutional neural network, so as to obtain the physical vibration feature data in the above-mentioned field local feature data.

[0200] In an optional solution, the feature extraction unit further includes a first determination module, a first maximum pooling processing module, and a second conversion module; the first determination module is used to determine the first calculation formula of the convolutional layer of the above-mentioned convolutional neural network: Input the above frequency domain feature vector into the above convolution layer to obtain the vibration frequency domain feature G i，j , where w j，k is the weight of the jth convolution kernel, b j is the corresponding bias term, K is the size of the convolution kernel, ReLU is the activation function, and f i+k-1 is the above-mentioned frequency domain feature vector; the first maximum pooling processing module is used to use the pooling layer of the above-mentioned convolutional neural network to perform maximum pooling processing on the above-mentioned vibration frequency domain features to obtain pooled vibration frequency domain features; the second conversion module is used to convert the above-mentioned pooled vibration frequency domain features into the above-mentioned physical vibration feature data based on the Softmax function through the fully connected layer of the above-mentioned convolutional neural network.

[0201] An optional solution, the device also includes a construction unit and an early warning processing unit; the construction unit is used to construct a hierarchical early warning rule in the process of using the above-mentioned intrusion behavior confidence to perform anti-theft detection on the above-mentioned construction suspension site, wherein the above-mentioned hierarchical early warning rule includes no intrusion behavior, low-level intrusion warning, intermediate intrusion warning and high-level intrusion alarm, the above-mentioned low-level intrusion warning is a suspected intrusion behavior, which is recorded and silently notified to the user, the above-mentioned intermediate intrusion warning is a high-probability intrusion behavior, which triggers a local alarm and remote notification at the above-mentioned construction suspension site, and the above-mentioned high-level intrusion alarm determines that there is an intrusion behavior, triggers a full alarm, issues a siren on-site warning, saves the video of the above-mentioned construction suspension site and contacts the alarm center; the early warning processing unit is used to perform real-time early warning processing on the above-mentioned construction suspension site according to the above-mentioned intrusion behavior confidence and the above-mentioned hierarchical early warning rules.

[0202] In an optional solution, the device further includes a training unit and an updating unit; the training unit is used to train the real-time intrusion detection model using multiple sets of training data after constructing the real-time intrusion detection model based on the temporal convolutional network, thereby obtaining the trained real-time intrusion detection model, wherein each set of the multiple sets of training data includes: a historical global feature vector obtained from the historical on-site data collected at the construction suspension site and intrusion category label data corresponding to the historical global feature, obtained during a historical time period; the updating unit is used to, in the process of training the real-time intrusion detection model using the multiple sets of training data, according to the loss function formula: Determine the loss value of the real-time intrusion detection model, and according to the loss value, use the Adam optimizer to update the model parameters of the real-time intrusion detection model based on minimizing the loss function, where F is the loss value, is the real intrusion category label data, which takes a value of 0 or 1. O1 is the probability that the above real-time intrusion detection model predicts the label to be 0, and O2 is the probability that the above real-time intrusion detection model predicts the label to be 1.

[0203] In an optional solution, the feature extraction unit includes a second determination module and a second maximum pooling processing module; the second determination module is used to determine the second calculation formula of the convolution layer of the above convolutional neural network: The preprocessed visual data or preprocessed infrared sensor data in the preprocessed field data is input into the convolution layer to obtain a feature map corresponding to the preprocessed visual data or the preprocessed infrared sensor data, wherein X (l+1) [i, j, k] is the above feature map, X (l) is the pre-processed visual data or the pre-processed infrared sensor data, W (l) is the convolution kernel, K h is the height of the convolution kernel, Kw is the width of the convolution kernel, σ is the nonlinear activation function, is the bias of the k-th convolution kernel of the l-th layer, i represents the row index of the corresponding feature map, j represents the column index of the corresponding feature map, k represents the corresponding convolution kernel in the corresponding feature map, p represents the sliding index of the corresponding convolution kernel k in height, q represents the sliding index of the corresponding convolution kernel k in width, and m represents the channel index of the corresponding feature map; the second maximum pooling processing module is used to use the pooling layer of the above-mentioned convolutional neural network to perform maximum pooling processing on the above-mentioned feature map to obtain a maximum pooling feature map, and perform dimensionality reduction processing on the above-mentioned maximum pooling feature map to obtain visual feature data or infrared sensing feature data in the local feature data of the scene.

[0204] The multi-sensor fusion-based anti-theft and anti-intrusion detection device includes a processor and memory. The processing unit, feature extraction unit, and anti-theft and anti-intrusion detection unit are all stored as program units in the memory. The processor executes these program units stored in the memory to implement the corresponding functions. All of these modules are located in the same processor; alternatively, the modules can be located in different processors in any combination.

[0205] The processor includes a core, which retrieves the corresponding program unit from memory. One or more cores can be configured, and adjusting the core parameters addresses the issue of existing technologies that protect suspended construction sites from environmental interference, resulting in low security.

[0206] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0207] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program. When the program is run, the device where the computer-readable storage medium is located is controlled to execute the anti-theft and anti-intrusion detection method based on multi-sensor fusion.

[0208] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes the anti-theft and anti-intrusion detection method based on multi-sensor fusion when running.

[0209] An embodiment of the present invention provides an electronic device comprising a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements at least the steps of the aforementioned multi-sensor fusion-based anti-theft and anti-intrusion detection method. The device herein may be a server, a PC, a PAD, a mobile phone, or the like.

[0210] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes at least the steps of the above-mentioned anti-theft and anti-intrusion detection method based on multi-sensor fusion.

[0211] Obviously, those skilled in the art will appreciate that the various modules or steps of the present invention described above can be implemented using a general-purpose computing device, can be centralized on a single computing device, or can be distributed across a network of multiple computing devices. They can be implemented using program code executable by the computing device, and thus, can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described herein can be performed in a different order than that shown, or can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0212] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0213] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0214] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0215] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0216] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0217] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0218] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0219] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0220] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A theft and intrusion prevention detection method based on multi-sensor fusion, characterized in that: include: Acquiring field data collected at the construction suspension site, performing denoising processing on the field data based on a median filter to obtain denoised field data, and performing normalization processing on the denoised field data based on a Z-score normalization algorithm to obtain preprocessed field data, wherein the field data collected includes visual data, infrared sensor data, and physical vibration data, and the field data collected is obtained by multiple sensors; Performing feature extraction processing on the pre-processed site data based on a convolutional neural network and a fast Fourier transform to obtain local feature data of the site, constructing a global relationship model of the construction suspension site based on the local feature data of the site, and obtaining a global feature vector from the global relationship model based on a graph neural network; A real-time intrusion detection model is constructed based on a temporal convolutional network, and the global feature vector is input into the real-time intrusion detection model to output an intrusion behavior confidence. The intrusion behavior confidence is used to perform anti-theft detection on the construction suspension site.

2. The method according to claim 1, characterized in that De-noising the field-collected data based on a median filter to obtain de-noised field data, including: The output formula of the median filter is: i =median(x1,x2,…,x n ), denoising the field data to obtain the denoised field data, where (x1, x2, ..., x n ) is the on-site collected data, y i The denoised field data; The denoised field data is standardized based on a Z-score standardization algorithm to obtain pre-processed field data, including: using the Z-score standardization algorithm: The denoised field data is standardized to obtain pre-processed field data Z i ,in, μ is the mean of the data collected on site, and σ is the standard deviation of the data collected on site.

3. The method according to claim 1, characterized in that The pre-processed field data is subjected to feature extraction processing based on a convolutional neural network and a fast Fourier transform to obtain field local feature data, including: Based on the fast Fourier transform, the preprocessed physical vibration data in the preprocessed field data is subjected to Fourier transform processing to obtain frequency domain analysis data of the preprocessed physical vibration data; Determining the amplitude value of the frequency domain analysis data, and converting the amplitude value into a frequency domain feature vector; Feature extraction is performed on the frequency domain feature vector based on the convolutional neural network to obtain physical vibration feature data in the on-site local feature data.

4. The method according to claim 3, characterized in that Performing feature extraction on the frequency domain feature vector based on the convolutional neural network to obtain physical vibration feature data in the on-site local feature data includes: The first calculation formula for determining the convolutional layer of the convolutional neural network is: Input the frequency domain feature vector into the convolution layer to obtain the vibration frequency domain feature G i，j , where w j，k is the weight of the jth convolution kernel, b j is the corresponding bias term, K is the size of the convolution kernel, ReLU is the activation function, and f i+k-1 is the frequency domain eigenvector; Using the pooling layer of the convolutional neural network to perform maximum pooling on the vibration frequency domain features to obtain pooled vibration frequency domain features; The pooled vibration frequency domain features are converted into the physical vibration feature data through the fully connected layer of the convolutional neural network based on the Softmax function.

5. The method according to claim 1, wherein In the process of performing anti-theft detection on the construction suspension site using the intrusion behavior confidence level, the method further includes: Constructing hierarchical warning rules, wherein the hierarchical warning rules include no intrusion behavior, low-level intrusion warning, medium-level intrusion warning, and high-level intrusion alarm. The low-level intrusion warning is a suspected intrusion behavior, which is recorded and silently notified to the user. The medium-level intrusion warning is a high-probability intrusion behavior, which triggers a local alarm and remote notification at the construction suspension site. The high-level intrusion alarm determines that there is an intrusion behavior, triggers a full alarm, sounds a siren on-site warning, saves a video of the construction suspension site, and contacts the alarm center. According to the intrusion behavior confidence level and the graded warning rules, a real-time warning process is performed on the construction suspension site.

6. The method according to claim 1, characterized in that After building a real-time intrusion detection model based on a temporal convolutional network, the method further includes: The real-time intrusion detection model is trained using multiple sets of training data to obtain the trained real-time intrusion detection model, wherein each set of training data includes: a historical global feature vector obtained from historical on-site data collected at the construction suspension site and intrusion category label data corresponding to the historical global feature, obtained within a historical time period; In the process of training the real-time intrusion detection model using multiple sets of training data, the method further includes: According to the loss function formula: Determine the loss value of the real-time intrusion detection model, and according to the loss value, use the Adam optimizer to update the model parameters of the real-time intrusion detection model based on minimizing the loss function, where F is the loss value, is the real intrusion category label data, with a value of 0 or 1, O1 is the probability that the real-time intrusion detection model predicts the label to be 0, and O2 is the probability that the real-time intrusion detection model predicts the label to be 1.

7. The method according to claim 1, characterized in that The pre-processed field data is subjected to feature extraction processing based on a convolutional neural network and a fast Fourier transform, including: The second calculation formula for determining the convolutional layer of the convolutional neural network is: The preprocessed visual data or preprocessed infrared sensor data in the preprocessed field data is input into the convolution layer to obtain a feature map corresponding to the preprocessed visual data or the preprocessed infrared sensor data, wherein X (l+1) [i, j, k] is the feature map, X (l) is the pre-processed visual data or the pre-processed infrared sensing data, W (l) is the convolution kernel, K h is the height of the convolution kernel, K w is the width of the convolution kernel, σ is the nonlinear activation function, is the bias of the k-th convolution kernel in the l-th layer, i represents the row index of the corresponding feature map, j represents the column index of the corresponding feature map, k represents the corresponding convolution kernel in the corresponding feature map, p represents the sliding index of the corresponding convolution kernel k in height, q represents the sliding index of the corresponding convolution kernel k in width, and m represents the channel index of the corresponding feature map; The pooling layer of the convolutional neural network is used to perform maximum pooling processing on the feature map to obtain a maximum pooling feature map, and the maximum pooling feature map is subjected to dimensionality reduction processing to obtain visual feature data or infrared sensing feature data in the local feature data of the scene.

8. An anti-theft and anti-intrusion detection device based on multi-sensor fusion, characterized in that: include: a processing unit, configured to obtain field data collected at the construction suspension site, perform denoising on the field data based on a median filter to obtain denoised field data, and perform standardisation on the denoised field data based on a Z-score standardisation algorithm to obtain pre-processed field data, wherein the field data collected includes visual data, infrared sensor data and physical vibration data, and the field data collected is obtained by a plurality of sensors; a feature extraction unit, configured to perform feature extraction processing on the preprocessed site data based on a convolutional neural network and a fast Fourier transform to obtain local feature data of the site, construct a global relationship model of the construction suspension site based on the local feature data of the site, and obtain a global feature vector from the global relationship model based on a graph neural network; The anti-theft and anti-intrusion detection unit is used to build a real-time intrusion detection model based on a temporal convolutional network, input the global feature vector into the real-time intrusion detection model, output the intrusion behavior confidence, and use the intrusion behavior confidence to perform anti-theft detection on the construction suspension site.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the anti-theft and anti-intrusion detection method based on multi-sensor fusion according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing the anti-theft and anti-intrusion detection method based on multi-sensor fusion as described in any one of claims 1 to 7.

Images