Intrusion monitoring identification and event positioning device and method based on DAS system

Abstract

The invention belongs to the technical field of signal processing, and more specifically relates to an intrusion monitoring identification and event positioning device and method based on a DAS system. According to the invention, mode identification and accurate positioning are carried out on various vibration sources around by using sensing optical cables laid in a large range based on an optical fiber distributed sound wave vibration sensing technology. According to the method, respective advantages of a direct detection method and a coherent detection method are exerted, the vibration position can be quickly demodulated by using the direct detection method, the dynamic process of vibration change can be obtained by using a digital coherent method, and then vibration event identification is realized. The method has the advantages of being wide in monitoring area range, high in recognition accuracy, high in real-time performance, high in stability and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and more specifically, relates to an intrusion monitoring and identification and event locating device and method based on a DAS system. Background Art

[0002] Distributed acoustic vibration sensing technology based on optical fiber plays an important role in the security monitoring of national and urban infrastructure, such as optical cables, cables, long-distance oil and gas pipelines, bridges, tunnels and regional perimeters. When there is a disturbance event in the outside world and it acts on the nearby sensing optical cable, the phase of the backscattered light signal in the sensing optical cable is modulated, and the detection of external disturbance events is achieved by detecting the backscattered light signal and combining the phase demodulation technology. Traditional detection methods include direct detection method and coherent detection method. The direct detection scheme usually directly relies on the trajectory of the backscattering intensity changing with time. It has the advantages of simple structure, easy implementation, and relatively mature vibration demodulation algorithm, but its detection signal-to-noise ratio is low. Unlike direct detection, the coherent detection scheme records the dynamic process of vibration events by demodulating the amplitude and phase changes of Rayleigh scattered light. Coherent detection linearly measures external vibration events by extracting the phase of the coherent signal, which not only makes up for the deficiency of direct detection that can only detect the intensity of backscattered light, but also expands the modulatable dimension of optical fiber sensing. The coherent detection method has a high detection signal-to-noise ratio, but the high computational complexity of the digital coherent process leads to a slow demodulation speed. In digital coherence, calculating the location of a vibration event requires demodulating the amplitude of the collected fiber global signal, and then judging the location of the vibration event based on the amplitude change. This makes data analysis more difficult, and the high computational complexity leads to a slow demodulation speed, affecting the real-time nature of the alarm. On the other hand, due to The distributed characteristics, high sensitivity and interference vibrations often exist in the detection environment make it difficult to monitor long-distance vibrations. This is accompanied by a high false alarm rate. Therefore, the machine learning method can be combined to improve the recognition accuracy of the φ-OTDR system and thus improve the alarm accuracy.

[0003] In response to this series of problems, an algorithm for accurate identification of intrusion monitoring and rapid event location based on the DAS system was studied. First, the two detection methods of direct detection and coherent detection were integrated to give full play to their respective advantages. The external vibration position was rapidly demodulated by direct detection. On the basis of demodulating the vibration position, the light intensity direct detection method of φ-OTDR was combined with the digital coherent detection technology. The dynamic process of vibration change was obtained by using the digital coherent method to achieve rapid demodulation of the beat frequency signal phase. Then, the time-frequency two-dimensional features of the demodulated signal were extracted. In order to further enhance the time-frequency two-dimensional feature extraction capability of the signal, a time-space two-dimensional sound field image enhancement technology of the LSTM network based on the channel attention mechanism was proposed. The generalization ability of the model was improved by the data enhancement method, and two attention mechanisms were added to improve the model's ability to extract target features, ultimately achieving high-precision pattern recognition and rapid location of external disturbance events.

[0004] When there is a disturbance event in the outside world and it acts on the nearby sensing optical cable, the phase of the backscattered light signal in the sensing optical cable is modulated, and the detection of the external disturbance event is achieved by detecting the backscattered light signal and combining it with phase demodulation technology. Traditional detection methods include direct detection method and coherent detection method. The direct detection method has the advantages of simple structure, easy implementation, and relatively mature vibration demodulation algorithm, but its detection signal-to-noise ratio is low. The coherent detection method has a high detection signal-to-noise ratio, but the high computational complexity of the digital coherence process leads to a slow demodulation speed. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides an intrusion monitoring identification and event positioning device and method based on a DAS system, and uses a wide range of sensor cables to perform pattern recognition and precise positioning of various vibration sources around the surrounding area based on optical fiber distributed acoustic vibration sensing technology (DAS). The present invention gives play to the respective advantages of direct detection method and coherent detection method, and can use the digital coherent method to obtain the dynamic process of vibration change while using the direct detection method to achieve rapid demodulation of vibration position, thereby realizing vibration event recognition. However, due to the complex environment, the accuracy of the system in identifying external disturbance events is low. In view of this problem, the present invention further proposes a spatiotemporal two-dimensional sound field image enhancement technology based on an LSTM network of a channel attention mechanism, analyzes the spatiotemporal two-dimensional features of the target signal, improves the generalization ability of the model through a data enhancement method, and adds two attention mechanisms to improve the model's ability to extract target features, and finally realizes high-precision pattern recognition and rapid positioning of external disturbance events. The present invention can be applied to the fields of national and urban infrastructure, communication light, cable, long-distance oil and gas pipelines, and security monitoring of regional perimeters, and has the characteristics of wide monitoring area, high recognition accuracy, high real-time performance, and high stability.

[0006] The technical solution of the present invention is: an intrusion monitoring accurate identification and event rapid positioning device based on a DAS system, comprising a narrow linewidth light source, a first coupler, an acousto-optic modulation module, a driver, a pulse EDFA, a circulator, a sensing optical fiber, a 1×2 coupler, a 2×2 coupler, a balanced detector, a photoelectric detector, a data acquisition card and a host computer, the output end of the narrow linewidth light source is connected to the input end of the first coupler, the output end of the first coupler includes a first end and a second end, the output end of the driver is connected to the power supply end of the acousto-optic modulation module, and the first end of the first coupler is connected to the input end of the acousto-optic modulation module, and the second end of the first coupler is connected to one end of the 2×2 coupler; the pulse EDFA The input end is connected to the output end of the acousto-optic modulation module; the circulator includes a first port, a second port and a third port; the first port is connected to the output end of the pulse EDFA, the second port is sent to the sensing optical fiber, and the third port is connected to the input end of the 1×2 coupler; the 1×2 coupler includes two output ports, one output port and the second end of the first coupler are respectively connected to the two input ends of the 2×2 coupler, and the other output port is connected to the output end of the photodetector; the two output ends of the 2×2 coupler are respectively connected to the two input ends of the balanced detector; the input end of the data acquisition card is connected to the output end of the balanced detector and the output end of the photodetector; the input end of the host computer is connected to the output end of the data acquisition card.

[0007] According to the above-mentioned device for accurate identification of intrusion monitoring and rapid location of events based on a DAS system, it is characterized in that the first coupler is a 90:10 coupler.

[0008] The present invention also discloses a method for accurate identification of intrusion monitoring and rapid event location based on a DAS system, which is characterized by comprising the following steps:

[0009] Step 1: The acoustic vibration dynamic time domain signal of the external event in this area is obtained by fusion of direct detection and coherent detection. The output electrical signal of the photodetector is expressed as:

[0010]

[0011] Among them, ρ is the photoelectric conversion rate, A0 is the initial optical power, α is the optical attenuation rate, l is the location of the optical fiber, and σ is the comprehensive reflection coefficient of the optical fiber;

[0012] The output electrical signal of the balanced detector is expressed as:

[0013]

[0014] Among them, A s (t) is the amplitude of the backscattered Rayleigh light signal, A L(t) is the amplitude of the intrinsic light signal, Δθ(t) is the phase difference between the intrinsic light and the signal light, and Δω is the angle difference between the intrinsic light and the signal light;

[0015] When a disturbance event occurs around the optical fiber, the intensity of the backward Rayleigh scattered light signal in Formula 1 is solved by direct detection to obtain the approximate location area of ​​the external disturbance event, and the phase information Δθ(t) of the beat frequency signal in the approximate location area of ​​the event is solved to obtain the phase change information of the optical signal caused by the external disturbance vibration signal;

[0016] Step 2: dividing the acoustic vibration dynamic time domain signal in the spatial domain, and arranging the segmented data in spatial order to obtain a space-time two-dimensional acoustic vibration graph of the external event, and then segmenting the space-time two-dimensional acoustic vibration graph along the time axis to obtain a multi-frame space-time two-dimensional acoustic vibration graph;

[0017] Step 3: Build a long short-term memory network as a time dimension feature extractor, and extract the time series features in the multi-frame acoustic wave vibration space-time two-dimensional map, stack the extracted multi-frame time features by channel, obtain the image channel weighted information through the channel attention mechanism, multiply the channel weighted information with the original feature map, process it channel by channel, and enhance the important feature channels of the image target;

[0018] Step 4: Send the enhanced multi-channel acoustic wave vibration space-time two-dimensional image to the convolution layer for image feature extraction, and select a bunch of target feature candidate frames in the image through convolution operation; train a multi-dimensional support vector machine to classify the target feature candidate frames into foreground and background, and introduce a branch candidate region generation neural network to correct the candidate frame using border regression;

[0019] Step 5: In the generation of image candidate frames, increase the weight parameters of the foreground candidate frames and increase the loss function of the background candidate frames to improve the network's ability to capture foreground target features. Compare all candidate frames, remove redundant candidate frames, and retain the image candidate frames with the best performance for pattern recognition.

[0020] According to the method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system as described above, it is characterized in that: in step 2, the specific process of dividing the spatial domain is: dividing the acoustic vibration dynamic time domain signal in the spatial domain according to the spatial resolution of the system.

[0021] According to the method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system as described above, the method is characterized in that: the specific process of step four is: the initial features of the target feature candidate frame are designed based on artificial experience, and multiple initial target candidate frames are generated in combination with the regional candidate frame network; then a set of instances that may be highly overlapping is predicted for each candidate frame, and a confidence score and fine-tuned location information is provided for it; finally, the Set NMS post-processing method is combined to suppress repeated results from different candidate frames, and the target candidate frame is feature-matched with the real target to eliminate potential false positives.

[0022] According to the method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system as described above, it is characterized in that the neural network in step four uses a neural network generated by multiple candidate regions.

[0023] According to the method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system as described above, it is characterized in that the loss function in step five adopts the EMD Loss function.

[0024] In general, the above technical scheme conceived by the present invention can achieve the following beneficial effects: (1) The present invention utilizes the integration of direct detection and coherent detection to give full play to their respective advantages, and realizes rapid demodulation of the external vibration position by direct detection. On the basis of demodulating the vibration position, the φ-OTDR light intensity direct detection method is combined with the digital coherent detection technology, and the dynamic process of vibration change is obtained by the digital coherent method, so as to realize rapid demodulation of the beat frequency signal phase. Finally, high-precision detection and rapid positioning of external disturbance events are realized.

[0025] (2) The present invention performs time-frequency analysis on the demodulated signal and, combined with the distributed characteristics of optical fiber sensing, generates a vibration space-time image that can better describe different vibration events from different dimensions. In order to further improve the ability to extract space-time features, a data enhancement method is used to enhance the signal space-time data.

[0026] (3) The present invention provides a spatiotemporal two-dimensional sound field image enhancement technology based on an LSTM network with a channel attention mechanism. The spatiotemporal two-dimensional sound field image of the signal is framed in the time dimension using time framing processing, and the multi-frame spatiotemporal two-dimensional sound field image is subjected to temporal feature extraction in combination with an LSTM network. The generalization ability of the model is improved by using data enhancement methods, and two attention mechanisms are added to improve the model's ability to extract spatiotemporal two-dimensional features of target features, thereby further improving the event recognition accuracy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is the installation diagram of the DAS system-based intrusion monitoring, precise identification and event rapid location system.

[0028] Figure 2 This is a flow chart of the spatiotemporal two-dimensional sound field image enhancement technology based on the LSTM network with channel attention mechanism.

[0029] Explanation of the reference numerals: narrow linewidth light source 2, first coupler 3, acousto-optic modulation module 4, driver 5, pulse EDFA 6, circulator 7, sensing optical fiber 8, 1×2 coupler 9, 2×2 coupler 10, balanced detector 11, photodetector 12, data acquisition card 13, host computer 14. DETAILED DESCRIPTION

[0030] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0031] like Figure 1 As shown, an intrusion monitoring accurate identification and event rapid location device based on a DAS system of the present invention comprises: a narrow line width light source 2, a first coupler 3, an acousto-optic modulation module 4, a driver 5, a pulse EDFA 6, a circulator 7, a sensing optical fiber 8, a 1×2 coupler 9, a 2×2 coupler 10, a balanced detector 11, a photodetector 12, a data acquisition card 13 and a host computer 14. The first coupler 3 is preferably a 90:10 coupler, that is, the first end and the second end account for 90%:10% of the input end optical signal respectively, and the proportion of the output signal light can be changed according to the design. The output end of the narrow linewidth light source 2 is connected to the input end of the first coupler 3, the output end of the first coupler 3 includes a first end accounting for 90% of the input optical signal and a second end accounting for 10% of the input optical signal, the first end of the first coupler 3 is connected to the input end of the acousto-optic modulation module 4, and the second end of the first coupler 3 is connected to one end of the 2×2 coupler 10; the output end of the driver 5 is connected to the power supply end of the acousto-optic modulation module 4, and the input end of the pulse EDFA 6 is connected to the output end of the acousto-optic modulation module 4; the circulator 7 includes a first port, a second port and a third port; wherein the first port is connected to the output end of the pulse EDFA 6, the second port is connected to the output end of the acousto-optic modulation module 4, and the third port is connected to the output end of the acousto-optic modulation module 4. The second port is sent to the sensing optical fiber 8, and the third port is connected to the input end of the 1×2 coupler 9; the 1×2 coupler 9 includes two output ports, one output port and the second end of the first coupler 3 accounting for 10% of the input optical signal are respectively connected to the two input ends of the 2×2 coupler 10, and the other output port is connected to the output end of the photodetector 12; the two output ends of the 2×2 coupler 10 are respectively connected to the two input ends of the balanced detector 11; the input end of the data acquisition card 13 is connected to the output end of the balanced detector 11 and the output end of the photodetector 12; the input end of the host computer 14 is connected to the output end of the data acquisition card 13.

[0032] When the rapid positioning and precise identification device of the present invention is working, the narrow linewidth light source 2 provides an optical signal; the first coupler 3 divides the input end optical signal into two optical signals with a ratio of 90:10; the acousto-optic modulation module 4 modulates the narrow linewidth light source 2 into a pulse detection optical signal; the driver 5 provides drive for the acousto-optic modulation module 4; the pulse EDFA 6 optically amplifies the pulse detection optical signal; the circulator 7 sends the detection optical signal sent into the first port to the second port for output, and the second port is sent into the sensing optical fiber 8, and backward Rayleigh light scattering occurs in the sensing optical fiber 8, and the Rayleigh scattered light backscattered enters the second port of the circulator 7 and is sent to the third port of the circulator 7, and the light emitted from the third port of the circulator is sent to the 1×2 coupler. 9 divides the backscattered light signal into two, wherein one light signal and the second end of the first coupler 3 accounting for 10% of the input light signal enter the two input ends of the 2×2 coupler 10 in sequence for coherence to form a coherent light beat frequency signal; the two signal lights are outputted by the 2×2 coupler 10 and enter the balanced detector BPD (11) to convert the light signal into an electrical signal; and the other backscattered light signal enters the photoelectric detector 12 for optical amplification and converts the light signal into an electrical signal, and the data of the two electrical signals are collected by the data acquisition card 13; the data is processed by the host computer 14 to obtain the approximate position of the external disturbance event and the acoustic vibration signal information near the approximate position, which is used for subsequent further accurate identification.

[0033] The host computer 14 of the present invention includes an external disturbance event rapid positioning module and an event precise identification module. The external disturbance event rapid positioning module integrates the external sound wave signal dynamic rapid demodulation method of direct detection and coherent detection, uses the direct detection method to perform intensity demodulation on multiple backscattering curves, determines the approximate location area of ​​the external disturbance event, combines the OTDR technology to lock the optical fiber area, and then demodulates the phase of the backscattered light signal of the optical fiber area through the coherent detection method, and quickly extracts the dynamic signal of the sound wave vibration at the approximate location area of ​​the external disturbance event.

[0034] like Figure 2 As shown, the present invention also discloses a method for accurate identification of intrusion monitoring and rapid event location based on a DAS system, which adopts a spatiotemporal two-dimensional sound field image enhancement technology based on an LSTM network with a channel attention mechanism; the method comprises the following steps:

[0035] Step 1: The acoustic vibration dynamic time domain signal of the external event in the section is obtained by a demodulation technology that combines direct detection and coherent detection. The photoelectric detector 12 directly detects the intensity of the backward Rayleigh scattered light signal, and its expression is shown in Formula 1; the balanced detector 11 detects the beat frequency signal after the backward Rayleigh scattered light signal beats with the intrinsic light signal, and its expression is shown in Formula 2;

[0036] The output electrical signal of the photodetector can be expressed as:

[0037]

[0038] Among them, ρ is the photoelectric conversion rate, A0 is the initial optical power, α is the optical attenuation rate, l is the location point of the optical fiber, and σ is the comprehensive reflection coefficient of the optical fiber.

[0039] The output electrical signal of the balanced detector can be expressed as:

[0040]

[0041] Among them, A s (t) is the amplitude of the backscattered Rayleigh light signal, A L (t) is the amplitude of the intrinsic light signal, Δθ(t) is the phase difference between the intrinsic light and the signal light, and Δω is the angle difference between the intrinsic light and the signal light.

[0042] When a disturbance event occurs around the optical fiber, the intensity of the backward Rayleigh scattered light signal in Formula 1 is solved by direct detection to obtain the approximate location area of ​​the external disturbance event. Then, the phase information Δθ(t) of the beat frequency signal in this area is solved to obtain the phase change information of the optical signal caused by the external disturbance vibration signal.

[0043] Step 2: Divide the sound wave vibration dynamic time domain signal in the spatial domain, and arrange the segmented data in spatial order to obtain a space-time two-dimensional graph of the sound wave vibration of the external event, and then segment the space-time two-dimensional graph of the sound wave vibration along the time axis to obtain multiple frames of the space-time two-dimensional graph of the sound wave vibration.

[0044] The specific process of spatial domain division is: according to the spatial resolution of the system, that is, the repetition frequency of the light pulses emitted by the system light source, the dynamic time domain signal of the acoustic vibration is divided in the spatial domain. The interval of spatial division corresponds to the spatial resolution of the system. The spatial resolution of the system can be adjusted according to the actual application needs.

[0045] Step 3: Build a long short-term memory network (LSTM) as a time dimension feature extractor, and extract the time series features in the multi-frame space-time two-dimensional image of acoustic wave vibration. Stack the extracted multi-frame time features through channels, obtain the image channel weighted information through the channel attention mechanism, multiply the channel weighted information with the original feature map, and process it channel by channel to enhance the important feature channels of the image target.

[0046] Step 4: Send the enhanced multi-channel acoustic wave vibration space-time two-dimensional image to the convolution layer for image feature extraction, and select a bunch of target feature candidate frames in the image through convolution operation. Then train a multi-dimensional support vector machine to classify the target feature candidate frames into foreground and background. In addition, introduce a branch candidate region generation neural network to use border regression to correct the candidate frame and obtain a more accurate image candidate frame.

[0047] In this step, the initial features of the target feature candidate box are designed based on artificial experience, and multiple initial target candidate boxes are generated in combination with the regional candidate box network; then a set of possibly highly overlapping instances is predicted for each candidate box, providing it with a confidence score and fine-tuned location information, and finally combined with the Set NMS post-processing method to suppress repeated results from different candidate boxes, while adding optional fine-tuning comparisons, and comparing the features of the target candidate box with the real target to eliminate potential false positives and obtain image candidate boxes with more accurate information.

[0048] The neural network in this step is preferably a neural network generated by multiple candidate regions. The neural network generated by multiple candidate regions can better capture the nonlinear characteristics in the hidden space and has a stronger expression ability. It can better adapt to complex input data and get results at a faster speed. It can also reduce the possibility of overfitting and help prevent overfitting. At the same time, the additional cost of adding a prediction branch is actually negligible, but the improvement is significant.

[0049] Step 5: In the generation of image candidate frames, increase the weight parameters of the foreground candidate frames and increase the loss function of the background candidate frames to improve the network's ability to capture foreground target features. Compare all candidate frames, remove redundant candidate frames, and retain the image candidate frames with the best performance for pattern recognition.

[0050] The loss function of this step can use the EMD Loss loss function, including classification loss and candidate box regression loss. Add the weight parameters of the learnable foreground candidate boxes. During the model training process, the parameters are continuously fine-tuned by calculating the probability distribution difference between the candidate boxes and the true target each time, so that the candidate boxes are constantly close to the true target. For large background candidate boxes, supervise the learning of candidate box instance set prediction.

[0051] The device and method of the present invention utilize the fusion of direct detection and coherent detection. The event recognition accuracy of external disturbance events is as high as over 90%, the response to external disturbance events is less than 0.5s, and the positioning accuracy of disturbance events is 5 meters. It has the characteristics of wide monitoring area, high recognition accuracy, high real-time performance, high stability, etc.

Claims

1. A DAS-based intrusion monitoring accurate identification and event rapid location device, comprising a narrow line width light source, a first coupler, an acousto-optic modulation module, a driver, a pulse EDFA, a circulator, a sensing optical fiber, a 1×2 coupler, a 2×2 coupler, a balanced detector, a photodetector, a data acquisition card and a host computer, wherein the output end of the narrow line width light source is connected to the input end of the first coupler, the output end of the first coupler includes a first end and a second end, and the output end of the driver is connected to the power end of the acousto-optic modulation module, characterized in that: The first end of the first coupler is connected to the input end of the acousto-optic modulation module, and the second end of the first coupler is connected to one end of the 2×2 coupler; the input end of the pulse EDFA is connected to the output end of the acousto-optic modulation module; the circulator includes a first port, a second port and a third port; wherein the first port is connected to the output end of the pulse EDFA, the second port is sent to the sensing optical fiber, and the third port is connected to the input end of the 1×2 coupler; the 1×2 coupler includes two output ports, one output port and the second end of the first coupler are respectively connected to the two input ends of the 2×2 coupler, and the other output port is connected to the output end of the photoelectric detector; the two output ends of the 2×2 coupler are respectively connected to the two input ends of the balanced detector; the input end of the data acquisition card is connected to the output end of the balanced detector and the output end of the photoelectric detector; the input end of the host computer is connected to the output end of the data acquisition card.

2. According to claim 1, a DAS system-based intrusion monitoring accurate identification and event rapid location device is characterized by: The first coupler is a 90:10 coupler.

3. A method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system, characterized by: The following steps are involved: Step 1: The acoustic vibration dynamic time domain signal of the external event in this area is obtained by fusion of direct detection and coherent detection. The output electrical signal of the photodetector is expressed as: Among them, ρ is the photoelectric conversion rate, A0 is the initial optical power, α is the optical attenuation rate, l is the location of the optical fiber, and σ is the comprehensive reflection coefficient of the optical fiber; The output electrical signal of the balanced detector is expressed as: Among them, A s (t) is the amplitude of the backscattered Rayleigh light signal, A L (t) is the amplitude of the intrinsic light signal, Δθ(t) is the phase difference between the intrinsic light and the signal light, and Δω is the angle difference between the intrinsic light and the signal light; When a disturbance event occurs around the optical fiber, the intensity of the backward Rayleigh scattered light signal in Formula 1 is solved by direct detection to obtain the approximate location area of ​​the external disturbance event, and the phase information Δθ(t) of the beat frequency signal in the approximate location area of ​​the event is solved to obtain the phase change information of the optical signal caused by the external disturbance vibration signal; Step 2: dividing the acoustic vibration dynamic time domain signal in the spatial domain, and arranging the segmented data in spatial order to obtain a space-time two-dimensional acoustic vibration graph of the external event, and then segmenting the space-time two-dimensional acoustic vibration graph along the time axis to obtain a multi-frame space-time two-dimensional acoustic vibration graph; Step 3: Build a long short-term memory network as a time dimension feature extractor, and extract the time series features in the multi-frame acoustic wave vibration space-time two-dimensional map, stack the extracted multi-frame time features by channel, obtain the image channel weighted information through the channel attention mechanism, multiply the channel weighted information with the original feature map, process it channel by channel, and enhance the important feature channels of the image target; Step 4: Send the enhanced multi-channel acoustic wave vibration space-time two-dimensional image to the convolution layer for image feature extraction, and select a bunch of target feature candidate frames in the image through convolution operation; train a multi-dimensional support vector machine to classify the target feature candidate frames into foreground and background, and introduce a branch candidate region generation neural network to correct the candidate frame using border regression; Step 5: In the generation of image candidate frames, increase the weight parameters of the foreground candidate frames and increase the loss function of the background candidate frames to improve the network's ability to capture foreground target features. Compare all candidate frames, remove redundant candidate frames, and retain the image candidate frames with the best performance for pattern recognition.

4. According to claim 3, a method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system is characterized by: In step 2, the specific process of dividing the space domain is: dividing the acoustic wave vibration dynamic time domain signal in the space domain according to the spatial resolution of the system.

5. According to claim 3, a method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system is characterized by: The specific process of step 4 is as follows: the initial features of the target feature candidate box are designed based on artificial experience, and multiple initial target candidate boxes are generated in combination with the regional candidate box network; then a set of possibly highly overlapping instances is predicted for each candidate box, providing it with a confidence score and fine-tuned location information, and finally combined with the Set NMS post-processing method to suppress duplicate results from different candidate boxes, perform feature comparison between the target candidate box and the real target, and eliminate potential false positives.

6. The method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system according to claim 3 is characterized by: The neural network in step 4 uses a neural network generated by multiple candidate regions.

7. The method for accurate identification of intrusion monitoring and rapid location of events based on a DAS system according to claim 3 is characterized by: The loss function in step 5 adopts the EMD Loss loss function.

Images