Perimeter intrusion detection system

Abstract

The invention relates to a perimeter intrusion detection system, and the system comprises an optical fiber signal collection unit which is used for collecting and storing an optical fiber vibration signal; the optical fiber signal processing and model training unit is used for converting the optical fiber vibration signal into a spectrogram through fast Fourier transform so as to train a frequency spectrum classification model based on a deep learning algorithm; the scene image acquisition unit is used for acquiring scene images in which people exist under normal illumination and dark light conditions; the scene image labeling unit is used for labeling the collected scene image; the personnel detection model training unit is used for training a personnel detection model based on a deep learning algorithm by using the collected and labeled scene image and is used for detecting personnel in the scene image; and the integrated detection unit is used for integrating the frequency spectrum classification model and the personnel detection model, and the perimeter intrusion detection system judges that personnel intrusion occurs only when the frequency spectrum classification model judges that the behavior of a person exists and the personnel detection model detects the person.

Description

Technical Field

[0001] The present invention relates to the field of deep learning algorithms and security monitoring technology, and in particular to a perimeter intrusion detection system that uses an algorithm model based on both optical fiber signal data and scene image data, such as deep learning and convolutional neural network model training. Background Art

[0002] In most cases, the fiber optic perimeter intrusion detection system can be used in many places with high security requirements, such as oil depots, cultural and museum venues, etc. Traditional fiber optic perimeter intrusion detection is based on the signal fluctuations generated by the vibration of the optical fiber to make intrusion judgments. The problem brought by this technology is that when the environment is not relatively simple, there are many external reasons that can cause the vibration of the optical fiber, resulting in false alarms.

[0003] In the prior art, the human intrusion detection technology based on video image shooting has been applied to many scenarios, but this technology also has many defects. For example, false alarms of intrusion may be generated at night or due to insufficient light.

[0004] CN118334806A discloses a perimeter intrusion detection method based on a distributed optical fiber hanging net buried composite strategy. First, the distributed optical fiber is laid in two parts: hanging net and buried. The short-time energy characteristics are extracted from the raw data collected by the distributed optical fiber, wherein the characteristic data of the buried part is judged by a threshold value, and the suspicious area data within a period of time is cached; the characteristic data of the hanging net part is screened by the early warning line, and then the final suspicious area of ​​the hanging net part is determined by the LSTM classification model. The two parts of the suspicious area are compounded according to the composite strategy to obtain the accurate intrusion location. The present invention adopts distributed optical fiber sensing unit detection, which can be laid over long distances in different terrains, and can greatly reduce the cost of perimeter security maintenance. In response to human intrusion activities, the present invention collects intrusion information from two parts, the ground and the fence, and adopts different strategies, which solves the problem of interference of extreme weather such as wind and snow on intrusion detection, and improves the real-time and reliability of detection.

[0005] CN118334806A aims at the defects of the existing single-point signal pattern recognition technology, fully considers the interference of severe weather conditions such as strong winds on detection, and can be applied to a variety of complex environments.

[0006] However, the above-mentioned various prior art solutions still have many problems and disadvantages, including at least the following:

[0007] (1) At present, perimeter intrusion detection systems based on optical fiber signals generally make judgments based on optical fiber vibration signals. Although some systems combine machine learning methods to judge intrusions, they are difficult to avoid false alarms in complex environments such as hail, strong winds and heavy rain. However, these environments will also cause optical fibers to generate vibration signals; and

[0008] (2) Perimeter intrusion detection systems based on video image algorithms generally detect people in images. This system is often greatly affected by light and illumination. At night or under certain lighting conditions, it may cause false detection of people, causing the system to make incorrect judgments.

[0009] Therefore, the present invention aims to overcome or at least alleviate the defects of the above-mentioned two detection technologies, and makes improvements based on the existing technology, and proposes a perimeter intrusion detection system and technology based on an algorithm model of optical fiber signals and scene image data, which is applied to occasions with high security requirements and high intrusion detection accuracy, so as to ensure high reliability, accuracy and high adaptability of intrusion detection, and achieve more technical advantages.

[0010] The information included in this background section of the present specification, including any references cited herein and any description or discussion thereof, is included for technical reference purposes only and is not to be construed as subject matter that will limit the scope of the present invention. Summary of the invention

[0011] The present invention is proposed in view of the above and other more concepts.

[0012] According to the main inventive concept of the present invention, the following technical problems are mainly solved: (1) In order to reduce the false alarm of the optical fiber vibration signal, the present invention proposes to convert the optical fiber vibration signal into a spectrum image of the signal, and use the CNN network to classify the spectrum image for intrusion judgment; (2) In order to reduce the problem of false alarm of the video image algorithm under lighting conditions, the present invention collects a large number of pictures under low light, dark light and normal light conditions and adds them to the training of the target detection algorithm, and uses such a trained model to detect human intrusion; (3) In order to further improve the accuracy of perimeter intrusion detection, the present invention combines the optical fiber vibration signal judgment module (based on the deep convolutional neural network model) and the video image detection module (based on the deep convolutional neural network model) to comprehensively judge the intrusion situation, thereby greatly improving the accuracy and reliability of intrusion detection and effectively reducing false alarms of intrusion.

[0013] The present invention can be applied to places with very high security requirements, such as oil depots, museums, art galleries and other places and occasions with long perimeters. The traditional method of using human patrols consumes a lot of manpower on the one hand, and on the other hand, it is impossible to detect illegal intrusions in time. The present invention proposes a perimeter intrusion detection system based on an algorithm model of both optical fiber signals and scene image data to solve the above and other problems.

[0014] More specifically, according to an aspect of the present invention, there is provided a perimeter intrusion detection system, comprising: an optical fiber signal acquisition unit, for acquiring and storing optical fiber vibration signals at predetermined time intervals; an optical fiber signal processing and model training unit, which is connected to the optical fiber signal acquisition unit, for converting the acquired optical fiber vibration signals into a frequency spectrum through a fast Fourier transform, and using the frequency spectrum to train a frequency spectrum classification model based on a deep learning algorithm, wherein the frequency spectrum classification model is used to classify the frequency spectrum to distinguish between human behavior signals and non-human behavior signals; a scene image acquisition unit, for acquiring corresponding scene images with human presence under normal lighting conditions and under dark lighting conditions; and a scene image annotation unit, connected to the scene image acquisition unit, used to annotate the acquired scene images and annotate the positions of people in the scene images; a personnel detection model training unit, connected to the scene image annotation unit, used to train a personnel detection model based on a deep learning algorithm using the acquired and annotated scene images, wherein the personnel detection model is used to detect people in the scene images; an integrated detection unit, connected to the optical fiber signal processing and model training unit and the personnel detection model training unit, used to integrate a spectrum classification model and a personnel detection model, wherein only when the spectrum classification model determines that it is a human behavior and the personnel detection model detects a person, the perimeter intrusion detection system determines that a human intrusion has occurred.

[0015] According to one embodiment, the optical fiber signal acquisition unit is configured to classify the optical fiber vibration signal into a human behavior signal and a non-human behavior signal according to different working conditions for model training.

[0016] According to one embodiment, when the optical fiber signal processing and model training unit converts the optical fiber vibration signal into a spectrum diagram, the fast Fourier transform formula adopted is: image=fft(signal), wherein signal represents the collected optical fiber vibration signal, image represents the corresponding spectrum image after conversion, and fft represents fast Fourier transform; and the spectrum classification model is trained based on the spectrum diagram until the model converges.

[0017] According to one embodiment, the spectrum classification model is a convolutional neural network model, wherein the convolution layer of the convolutional neural network model extracts local features by sliding the convolution kernel on the spectrum image, the pooling layer performs downsampling to reduce the size of the feature map, the fully connected layer integrates the features and maps them to the classification space, and the output layer uses the Softmax function to calculate the probability of human behavior signals and non-human behavior signals, and outputs the probability distribution. The final signal category is the one with the highest probability.

[0018] According to one embodiment, the spectrum classification model is a convolutional neural network model based on the Resnet50 architecture.

[0019] According to one embodiment, the personnel detection model is YOLOv8, and when the personnel detection model training unit is training the personnel detection model YOLOv8, when the loss of the personnel detection model YOLOv8 basically no longer changes, it is determined that the model has converged, and the training of the personnel detection model is completed at this time.

[0020] According to one embodiment, the spectrum classification model determines whether it is a human behavior signal under the following conditions: label = SpectrumModel (image spectrum ), label∈{0,1}, where SpectrumModel represents the spectrum classification model, image spectrum represents the converted spectrum graph, label represents the classification result of the spectrum classification model, wherein label=0 represents that the signal corresponding to the spectrum graph represents human behavior; label=1 represents that the signal corresponding to the spectrum graph represents non-human behavior.

[0021] According to one embodiment, the determination condition of whether a person is detected by the person detection model is: 1 ,y 1 ,x 2 ,y 2 ,score=PersonDetectModel(frame), where PersonDetectModel represents the person detection model, frame represents the image obtained from the camera deployed on the perimeter, x 1 ,y 1 ,x 2 ,y 2 , score respectively represents the upper left and lower right coordinates of the person in the image and the score of the detection result, wherein, when score>0.5, the person detection model determines that a person is detected.

[0022] According to an embodiment, the integrated detection unit is configured to notify a security manager of human intrusion information when the perimeter intrusion detection system determines that human intrusion has occurred.

[0023] According to an embodiment, the scene image labeling unit is configured to label the person in the scene image as person when labeling the position of the person in the scene image.

[0024] More embodiments of the present invention can also achieve other advantageous technical effects that are not listed one by one. These other technical effects may be partially described below and are predictable and understandable to those skilled in the art after reading the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The above features and advantages of these embodiments and other features and advantages and the manner in which they are achieved will become more apparent and embodiments of the present invention may be better understood by referring to the following description taken in conjunction with the accompanying drawings, in which:

[0026] Figure 1 It is a schematic system deployment diagram of a perimeter intrusion detection system based on an algorithm model of optical fiber signals and scene image data for perimeter intrusion detection constructed and deployed according to an embodiment of the present invention, schematically showing the process of training and deploying the perimeter intrusion detection system based on an algorithm model of optical fiber signals and scene image data of this embodiment.

[0027] Figure 2 is based on Figure 1 The schematic diagram of the perimeter intrusion detection system based on the algorithm model of the optical fiber signal and the scene image data of the illustrated embodiment schematically shows the general workflow and system model of the system according to the embodiment. DETAILED DESCRIPTION

[0028] In the following description of the drawings and specific embodiments, the details of one or more embodiments of the present invention will be described. Other features, objects and advantages of the present invention can be clearly seen from these descriptions, drawings and claims.

[0029] It should be understood that the illustrated and described embodiments are not limited in application to the details of the construction and arrangement of the components set forth in the following description or illustrated in the accompanying drawings. The illustrated embodiments may be other embodiments and may be implemented or executed in various ways. Each example is provided by explaining the disclosed embodiments rather than limiting them. In fact, it will be apparent to those skilled in the art that various modifications and variations may be made to the embodiments of the present invention without departing from the scope or essence of the present invention. For example, a feature illustrated or described as part of one embodiment may be used together with another embodiment to still produce another embodiment. Therefore, the present invention discloses such modifications and variations that fall within the scope of the appended claims and their equivalent elements.

[0030] Likewise, it is understood that the phrases and terms used herein are for descriptive purposes and should not be considered restrictive. The use of "include," "comprise," or "have" and variations thereof herein is intended to be open-ended to include the items listed thereafter and their equivalents and additional items.

[0031] The present invention will be described in more detail below with reference to specific embodiments of the present invention and the accompanying drawings.

[0032] Figure 1 It is a schematic system deployment diagram of a perimeter intrusion detection system based on an algorithm model of optical fiber signals and scene image data for perimeter intrusion detection constructed and deployed according to an embodiment of the present invention, schematically showing the process of training and deploying the perimeter intrusion detection system based on an algorithm model of optical fiber signals and scene image data of this embodiment. Figure 2 is based on Figure 1 The schematic diagram of the perimeter intrusion detection system based on the algorithm model of the optical fiber signal and the scene image data of the illustrated embodiment schematically shows the general workflow and system model of the system according to the embodiment.

[0033] like Figure 1-2 As shown schematically, schematic training and deployment of a perimeter intrusion detection system based on an algorithm model of optical fiber signals and scene image data according to an embodiment of the present invention, as well as a general workflow and system model of the perimeter intrusion detection system based on an algorithm model of optical fiber signals and scene image data, are schematically illustrated for perimeter intrusion detection.

[0034] Combined with Figure 1-2 As schematically shown, an exemplary implementation of a perimeter intrusion detection system based on an algorithm model of optical fiber signals and scene image data according to an embodiment of the present invention is described in detail below.

[0035] 1. Fiber Optic Signal Acquisition

[0036] 1.1. Equipment installation and preparation

[0037] Install railings with optical fibers or other suitable fiber optic sensor devices in the areas where perimeter monitoring is required, ensuring that the optical fibers can sense vibrations caused by possible intrusions.

[0038] Connect the optical fiber signal acquisition device and set the acquisition parameters, including the time interval for collecting a period of data, such as an interval of 5-30 seconds, for example, every 10 seconds.

[0039] 1.2. Data collection classification

[0040] We started to collect a large number of fiber optic vibration signals under different situations / working conditions, including normal interference-free state, simulated human intrusion (such as arranging personnel to climb railings, hammering railings, etc.), and different environmental interference (such as strong wind and hail weather simulation). Figure 1 .

[0041] For example, data is stored according to a segment of data collected every 10 seconds (10s). The collected data categories can be divided into: human behavior signals, that is, behavior signals generated by humans (for example, signals generated by humans climbing a railing with optical fibers, humans hammering / hitting a railing with optical fibers, etc.); and non-human behavior signals, for example, signals caused by the environment (for example, signals generated by optical fiber vibration caused by strong winds, hail, etc., etc.).

[0042] During the acquisition process, the signal is initially marked as a human behavior signal or a non-human behavior signal according to its source. For example, when an obvious regular impact signal is detected and there is no known environmental interference source around, it is marked as a possible human behavior signal. If it is consistent with environmental factors (such as vibration characteristics caused by strong wind), it is marked as a non-human behavior signal.

[0043] 2. Fiber Optic Signal Processing and Model Training

[0044] 2.1. Spectrum graph conversion

[0045] The collected and classified optical fiber vibration signals are input into the optical fiber signal processing and model training unit, and each vibration signal is converted into a corresponding spectrum diagram using the fast Fourier transform formula. An exemplary transformation formula is as follows:

[0046] image=fft(signal)

[0047] Wherein, signal represents the collected vibration signal, image represents the corresponding spectrum image after conversion, and fft represents fast Fourier transform. Based on these spectrum images, the present invention is used to train a spectrum classification model based on a deep learning algorithm, which is generally a convolutional neural network (CNN) model.

[0048] 2.2. Spectrum classification model construction and training

[0049] A spectrum classification model is constructed, i.e., a convolutional neural network (CNN) model based on, for example, the Resnet50 architecture.

[0050] Convolutional Neural Network (CNN) is a deep learning model specifically used for visual tasks such as image recognition and classification. It extracts and classifies features through a series of convolutional layers, pooling layers, and fully connected layers. The core components of this model generally include: (1) Convolutional layer: This is the core of CNN. It slides the convolution kernel on the image to identify local features in the image and generate feature maps; (2) Pooling layer: Also known as downsampling layer, it is used to reduce the size of the feature map and reduce computational complexity while retaining important features. The most common ones are maximum pooling and average pooling; (3) Fully connected layer: Similar to traditional neural networks, each node is connected to all nodes in the previous layer. It is used to integrate the features extracted by the convolutional layer and the pooling layer and output the final classification result; (4) Output layer: Usually uses the Softmax function for multi-classification tasks and outputs the probability distribution of all categories. The workflow of a convolutional neural network (CNN) model generally includes: the input layer inputs image data; the convolution layer extracts local features of the image and gradually extracts higher-level features through multiple convolution layers; the pooling layer downsamples the feature map to reduce the data dimension; the activation function introduces nonlinear characteristics so that the model can learn complex features; the fully connected layer integrates the extracted features and maps them to the classification space; the output layer usually uses the Softmax function for multi-classification tasks and outputs the probability distribution of all categories.

[0051] Regarding the CNN model as a spectrum classification model, we adopt the common Resnet50, that is, a deep learning convolutional neural network based on the residual network architecture. ResNet effectively solves these problems through residual blocks and jump connections, allowing the network to be trained deeper while maintaining good performance. In addition, ResNet50 performs well in many computer vision tasks, especially in image classification, object detection and segmentation.

[0052] The converted spectrogram is used as input data, and human behavior signals and non-human behavior signals are used as two types of output labels to train the CNN model.

[0053] During the training process, the loss function of the model is monitored. As the number of training rounds increases, when the loss basically does not change (for example, the loss change for several consecutive rounds is less than the set threshold, such as 0.01), it means that the CNN model has converged. At this time, the CNN model training is stopped and the training of the spectrum classification model (SpectrumModel) is completed. Figure 1 .

[0054] 3. Scene Image Collection and Annotation

[0055] Image acquisition

[0056] Install image / video capture devices such as cameras in specific perimeter areas to capture images of scenes with people under normal lighting and dim lighting (including low illumination, darkness, etc.). Ensure that the captured images cover different lighting scenes, as well as different viewing angles, distances, and activity states of people.

[0057] Image Annotation

[0058] Manually annotate the collected scene images or use automated annotation tools (if the accuracy meets the requirements). The annotation content includes the upper left and lower right coordinates of the person in the image, and the person category in the image is marked as person. Figure 1 .

[0059] 4. Personnel Detection Model Training

[0060] 4.1 Construction and training of the target detection model YOLOv8 model

[0061] Construct, for example, a target detection model YOLOv8 as a personnel detection model. Use the collected corresponding scene images with people under normal lighting and dark lighting conditions to train a model for detecting people. An example of the present invention uses the open source target detection model YOLOv8, which is also a target detection algorithm based on a convolutional neural network (CNN). When the loss of the YOLOv8 model is basically no longer changed after training, it means that the YOLOv8 model has converged and the YOLOv8 model has been trained. The target detection model YOLOv8 is an advanced deep learning convolutional neural network model, which is part of the YOLO (You Only Look Once) series. It is the latest version of the YOLO series open sourced by Ultralytics in January 2023, and is designed to provide state-of-the-art object detection performance; YOLOv8 supports image classification, object detection, and instance segmentation tasks, and can run on a variety of hardware platforms, including CPUs and GPUs.

[0062] Input the labeled scene images into the YOLOv8 model for training and set appropriate hyperparameters, such as learning rate, batch size, etc.

[0063] During the training process, the loss function of the model is also monitored. When the loss basically does not change (for example, the loss change for several consecutive rounds is less than the set threshold), the training is stopped. At this time, the YOLOv8 model converges and the training of the person detection model (PersonDetectModel) is completed. Figure 1 .

[0064] 5. Perimeter Intrusion Detection System Integration and Detection

[0065] Model Ensemble

[0066] The trained spectrum classification model (SpectrumModel) and personnel detection model (PersonDetectModel) are integrated into the perimeter intrusion detection system of the present invention. Figure 1-2 .

[0067] 5.2. General detection work of intrusion detection in perimeter intrusion detection system

[0068] When someone approaches or tries to invade the perimeter, the optical fiber generates vibration and other signals. The optical fiber signal acquisition unit collects the optical fiber vibration signal and converts it into a spectrum graph through the system. The spectrum graph is input into the spectrum classification model (SpectrumModel), and the output of the spectrum classification model is used to determine whether human behavior is detected.

[0069] According to an example, the determination is made according to the output of the spectrum classification model (SpectrumModel) as follows:

[0070] label=SpectrumModel(image spectrum ), label∈{0,1},

[0071] Among them, image spectrum It represents the converted spectrum image, and label represents the result of model classification. label = 0 means that the signal represents human behavior; label = 1 means that the signal represents non-human behavior.

[0072] At the same time, the camera installed in the corresponding area captures the image, inputs the image into the person detection model (PersonDetectModel), obtains the coordinates and detection score of the person in the image, and determines whether the person is detected according to the set score threshold.

[0073] According to an example, the output of the person detection model (PersonDetectModel) is analyzed and determined as follows:

[0074] x 1 ,y 1 ,x 2 ,y 2 ,score=PersonDetectModel(frame),

[0075] Among them, x 1 ,y 1 ,x 2 ,y 2,score represents the upper left and lower right coordinates of the person in the image and the score of the detection result; generally speaking, when score>0.5, the detection result is basically considered valid, indicating that the person detection model detects the person.

[0076] Finally, for example, the integrated detection unit of the perimeter intrusion detection system of the present invention combines the results of the two models, namely, the spectrum classification model (SpectrumModel) and the personnel detection model (PersonDetectModel), to make a judgment. That is, only when the spectrum classification model (SpectrumModel) determines that it is a human behavior and the personnel detection model (PersonDetectModel) detects a person, will the perimeter intrusion detection system of the present invention determine that a human intrusion has occurred, triggering the alarm mechanism to notify the security management personnel. See Figure 2 .

[0077] 6. Perimeter Intrusion Detection System Maintenance and Optimization

[0078] 6.1. Regular inspection

[0079] Regularly check the operating status of fiber optic sensors, cameras, acquisition equipment, and models to ensure normal data collection and processing.

[0080] 6.2. Model Update

[0081] According to the actual detection situation, new data is collected to update and optimize the spectrum classification model and the personnel detection model to improve the accuracy and reliability of detection. For example, if a misjudgment occurs in a special environment, more data in that environment is collected to adjust the model training.

[0082] The foregoing description of several embodiments of the present invention is presented for illustrative purposes. The foregoing description is not intended to be exhaustive, nor is it intended to limit the present invention to the precise steps and / or forms disclosed, and it is apparent that many modifications and variations may be made in light of the above teachings. The scope of the present invention and all equivalents are intended to be defined by the appended claims.

Claims

1. A perimeter intrusion detection system, characterized in that: The perimeter intrusion detection system comprises: An optical fiber signal acquisition unit, used to acquire and store optical fiber vibration signals at predetermined time intervals; an optical fiber signal processing and model training unit, which is connected to the optical fiber signal acquisition unit and is used to convert the acquired optical fiber vibration signal into a frequency spectrum through a fast Fourier transform, and use the frequency spectrum to train a frequency spectrum classification model based on a deep learning algorithm, wherein the frequency spectrum classification model is used to classify the frequency spectrum to distinguish between human behavior signals and non-human behavior signals; A scene image acquisition unit, used to acquire corresponding scene images with people under normal lighting conditions and dark lighting conditions; A scene image annotation unit, connected to the scene image acquisition unit, for annotating the acquired scene image and annotating the position of the person in the scene image; a personnel detection model training unit, connected to the scene image annotation unit, for training a personnel detection model based on a deep learning algorithm using the collected and annotated scene images, wherein the personnel detection model is used to detect personnel in the scene images; and an integrated detection unit, which is connected to the optical fiber signal processing and model training unit and the personnel detection model training unit, and is used to integrate the spectrum classification model and the personnel detection model, The perimeter intrusion detection system determines that a human intrusion has occurred only when the spectrum classification model determines that it is a human behavior and the human detection model detects a human.

2. The perimeter intrusion detection system according to claim 1, characterized in that: The optical fiber signal acquisition unit is configured to classify the optical fiber vibration signal into a human behavior signal and a non-human behavior signal according to different working conditions for model training.

3. The perimeter intrusion detection system according to claim 1 or 2, characterized in that: When the optical fiber signal processing and model training unit converts the optical fiber vibration signal into a frequency spectrum, the fast Fourier transform formula used is: image=fft(signal), Among them, signal represents the collected optical fiber vibration signal, image represents the corresponding spectrum image after conversion, and fft represents fast Fourier transform; and The spectrum classification model is trained based on the spectrum graph until the model converges.

4. The perimeter intrusion detection system according to claim 3, characterized in that: The spectrum classification model is a convolutional neural network model, wherein the convolution layer of the convolutional neural network model extracts local features by sliding the convolution kernel on the spectrum image, the pooling layer performs downsampling to reduce the size of the feature map, the fully connected layer integrates the features and maps them to the classification space, and the output layer uses the Softmax function to calculate the probability of human behavior signals and non-human behavior signals, outputs the probability distribution, and the final signal category is the one with the highest probability.

5. The perimeter intrusion detection system according to claim 1, 2 or 4, characterized in that: The spectrum classification model is a convolutional neural network model based on the Resnet50 architecture.

6. The perimeter intrusion detection system according to claim 1, 2 or 4, characterized in that: The personnel detection model is YOLOv8, and when the personnel detection model training unit is training the personnel detection model YOLOv8, when the loss of the personnel detection model YOLOv8 basically no longer changes, it is determined that the model has converged, and the training of the personnel detection model is completed at this time.

7. The perimeter intrusion detection system according to claim 1, 2 or 4, characterized in that: The spectrum classification model determines whether it is a human behavior signal based on the following conditions: label-SpectrumModel(image spectrum ),label∈{0,1}, Among them, SpectrumModel represents the spectrum classification model, image spectrum represents the converted spectrum graph, label represents the classification result of the spectrum classification model, wherein label=0 represents that the signal corresponding to the spectrum graph represents human behavior; label=1 represents that the signal corresponding to the spectrum graph represents non-human behavior.

8. The perimeter intrusion detection system according to claim 1, 2 or 4, characterized in that: The determination condition of whether a person is detected by the person detection model is: x1,y1,x2,y2,score=PersonDetectModel(frame), Among them, PersonDetectModel represents the person detection model, frame represents the image obtained from the camera deployed on the perimeter, x1, y1, x2, y2, score respectively represent the upper left and lower right coordinates of the person in the image and the score of the detection result, wherein, when score>0.5, the person detection model determines that a person is detected.

9. The perimeter intrusion detection system according to claim 1, 2 or 4, characterized in that: The integrated detection unit is configured to notify the security management personnel of the human intrusion information when the perimeter intrusion detection system determines that a human intrusion has occurred.

10. The perimeter intrusion detection system according to claim 8, characterized in that: The scene image annotation unit is configured to annotate the person in the scene image as person when annotating the position of the person in the scene image.

Images