A vibration-optoelectronic fusion type unattended perimeter alarm system

Abstract

The present invention relates to a vibration-optical fusion type unattended perimeter alarm system, belonging to the technical field of detection and alarm, and solving the problems that the existing video monitoring system lacks real-time early warning function and cannot achieve full coverage. The present invention includes at least one unattended perimeter alarm device, which includes an optoelectronic control module, an image acquisition device, a communication antenna, a GPS antenna, a signal processing module, a power supply module, a sound alarm module, a flashing light alarm module, a GPS positioning module, an Internet of Things communication module, and a vibration sensor. The vibration sensor and the image acquisition device respectively output vibration data and image data to the signal processing module. The signal processing module uses a vibration-optical fusion strategy to determine the final intrusion early warning result, and issues control commands to the sound and flashing light alarm modules respectively according to the intrusion early warning result, and uploads the intrusion early warning result, image, and location information to the cloud server. The present invention combines multiple functions, is suitable for complex environments, and has high early warning accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of detection and alarm, and in particular to a vibration-optoelectronic fusion type unattended perimeter alarm system. Background Art

[0002] At present, the perimeter security system mainly relies on the video surveillance system. Although the video surveillance system has played its due role, some problems and deficiencies have also emerged. For example, there is a lack of relatively professional duty personnel, the video surveillance system is not regularly viewed by anyone, and the system cannot exert its maximum use efficiency; the video data volume is huge, and the burden of manual review is heavy. At the same time, with the continuous growth of the number of video surveillance, at all levels, it is necessary to store and retrieve a large amount of video surveillance data. The workload is huge, and it is impossible to view and retrieve it in the first time only by manpower. The application efficiency of the system is not high, and there is also a lack of real-time warning function. The existing video surveillance system cannot effectively realize the real-time warning of suspicious targets in key areas, and can only reverse-check clues through the system after the case (event) occurs, which cannot meet the actual work needs. The video surveillance system cannot achieve full coverage either. Affected by natural conditions, there are no infrastructure such as electricity and network in some areas. It is difficult and costly to build a video surveillance system, and there are control blind spots and dead corners.

[0003] In summary, the existing video surveillance system cannot meet the various requirements of perimeter protection. Developing low-power intelligent perimeter warning technology is of great significance for promoting the practical application of the perimeter security system, providing scientific and technological means for establishing a digital and modern perimeter security system, ensuring perimeter security, preventing illegal elements from invading, and ensuring perimeter stability. Summary of the Invention

[0004] In order to solve the problems of the existing video surveillance system, such as the lack of real-time warning function and the inability to achieve full coverage, the present invention provides a vibration-optoelectronic fusion type unattended perimeter alarm system, which can collect and analyze the vibration signals generated by moving targets, cooperate with optoelectronic information to reduce the false alarm rate, and has the advantages of diversified communication means and the ability to achieve all-weather prevention and control.

[0005] In order to solve the above problems, the present invention adopts the following technical solutions:

[0006] A vibration-optoelectronic fusion type unattended perimeter alarm system includes at least one unattended perimeter alarm device. The unattended perimeter alarm device includes a multi-sensor load unit, a main control communication processing unit, and a signal acquisition unit. The multi-sensor load unit includes an optoelectronic control module, an image acquisition device, a communication antenna, and a GPS antenna. The main control communication processing unit includes a signal processing module, a power supply module, a sound alarm module, a flashing light alarm module, a GPS positioning module, and an Internet of Things communication module. The signal acquisition unit includes a vibration sensor;

[0007] The vibration sensor performs real-time vibration perception. When the vibration sensor detects a target, it outputs vibration data to the signal processing module. The optoelectronic control module is synchronously triggered and controls the image acquisition device to start. The image acquisition device transmits the acquired image data to the signal processing module. The signal processing module uses a vibration-optoelectronic fusion strategy to determine the final intrusion warning result, and issues control commands to the sound alarm module and the flashing light alarm module respectively according to the intrusion warning result, causing the sound alarm module to emit an alarm sound and the flashing light alarm module to emit light. At the same time, the signal processing module also uploads the intrusion warning result and the captured image to the cloud server through the Internet of Things communication module and the communication antenna. The GPS positioning module synchronously uploads the current location information to the cloud server through the GPS antenna. The vibration-optoelectronic fusion strategy refers to combining the information of vibration and image in two dimensions using a multimodal information fusion method, extracting features with high discrimination using a lightweight edge neural network, forming a high-dimensional space vector, and identifying and classifying the target.

[0008] Compared with the prior art, the present invention has the following beneficial effects:

[0009] (1) The vibration-optoelectronic fusion type unattended perimeter alarm system of the present invention adopts a multi-functional design, integrating functions of detection, evidence collection (photographing), and warning (warning light + sound);

[0010] (2) The hardware components of the present invention have good environmental adaptability and can work in indoor, outdoor, and outdoor environments with humidity, high temperature / low temperature;

[0011] (3) The hardware components of the present invention have a long battery life. Using passive seismic vibration detection technology, the longest battery life after one deployment can reach 6 months;

[0012] (4) The present invention adopts a unique multi-source sensing fusion technology, combining sensing technologies such as seismic vibration, infrared detection, and target recognition, effectively reducing the false alarm rate and improving the alarm accuracy;

[0013] (5) The present invention has the function of the Internet of Things and can realize the push of alarm information;

[0014] (6) The present invention has an automatic positioning function, supports multiple positioning such as GPS and Beidou, and can automatically report the device location information. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is the hardware principle architecture diagram of a single unattended perimeter alarm device;

[0017] Figure 2 It is the networking diagram of the unattended perimeter alarm device;

[0018] Figure 3 It is the schematic diagram of the ground multi-motion target recognition model.

[0019] Description of the reference numerals:

[0020] 10. Multi-sensor payload unit; 11. Photoelectric control module; 12. Image acquisition device; 13. Communication antenna; 14. GPS antenna;

[0021] 20. Main control communication processing unit; 21. Signal processing module; 22. Power supply module; 23. Sound alarm module; 24. Flashing light alarm module; 25. GPS positioning module; 26. Control button module; 27. Internet of Things communication module;

[0022] 30. Signal acquisition unit; 31. Vibration sensor. Specific implementation manners

[0023] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0024] See Figures 1 - 3 , this embodiment provides a vibration-optoelectronic fusion type unattended perimeter alarm system, which includes at least one unattended perimeter alarm device. As Figure 1 shown, the unattended perimeter alarm device includes a multi-sensor payload unit 10, a main control communication processing unit 20, and a signal acquisition unit 30. The multi-sensor payload unit 10 includes a photoelectric control module 11 and an image acquisition device 12 connected to the photoelectric control module 11. The image acquisition device 12 can be a high-definition camera or a COMS camera, etc., all of which have functions such as infrared detection and photographing. The multi-sensor payload unit 10 also includes a communication antenna 13 and a GPS antenna 14.

[0025] The main control communication processing unit 20 includes a signal processing module 21, a power module 22, a sound alarm module 23, a flashing light alarm module 24, a GPS positioning module 25 and an Internet of Things communication module 27. The GPS positioning module 25 is connected to the GPS antenna 14, and the GPS positioning module 25 can support GPS and Beidou multiple positioning. The Internet of Things communication module 27 is connected to the communication antenna 13. The power module 22 is respectively connected to the signal processing module 21 and the photoelectric control module 11, and is used to power the multi-sensor load unit 10 and the main control communication processing unit 20. Optionally, the power module 22 is provided with power by the battery compartment. The Internet of Things communication module 27 can adopt any one of the 2G module, 4G module, NB-IoT module, LoRa module, Wi-Fi module and BlueTooth module, or any combination of several of them.

[0026] The signal acquisition unit 30 includes a vibration sensor 31, and the vibration sensor 31 is connected to the signal processing module 21. Optionally, the vibration sensor 31 adopts a dynamic vibration sensor, which is a passive sensor and can convert the vibration of the shallow surface into an electrical signal by cutting the magnetic flux line through the internal coil without the need for additional power input, and transmit it to the signal processing module 21 after internal filtering and amplification.

[0027] The main control communication processing unit 20 also includes a control button module 26, which is connected to the signal processing module 21. The control button module 26 is used to select the working mode of the alarm system (warning mode, sentry mode, duty mode). Each time the control button module 26 is triggered, it outputs a corresponding switching signal to the signal processing module 21, and the signal processing module 21 controls the different working states of the sound alarm module 23 and the flashing light alarm module 24.

[0028] The three working modes in this embodiment are as follows:

[0029] (1) Warning mode:

[0030] a) The flashing light of the flashing light alarm module 24 keeps flashing slowly. After an intrusion warning occurs, i.e. after the signal processing module 21 determines the intrusion warning result, the signal processing module 21 controls the flashing light alarm module 24 to flash faster from far to near according to the target distance. After the intrusion warning is lifted, the flashing light alarm module 24 is turned off;

[0031] b) After determining the intrusion warning result, the signal processing module 21 controls the sound alarm module 23 to sound an alarm and speed up the alarm from far to near according to the target distance.

[0032] (2) Sentinel mode:

[0033] The flashing light alarm module 24 is closed by default. After the intrusion warning result is determined, the flashing light of the flashing light alarm module 24 starts to flash and the flashing frequency increases from far to near according to the target distance. The sound alarm module 23 sounds an alarm. After the intrusion warning is lifted, the flashing light alarm module 24 is closed again.

[0034] (3) On-duty mode:

[0035] The flashing light alarm module 24 is turned off by default and will not be turned on even after the intrusion warning result is determined.

[0036] After selecting the working mode (warning mode, sentry mode, duty mode) through the control button module 26, the photoelectric control module 11 sends a working signal to the image acquisition device 12, completes the parameter configuration, and enters the standby state. The signal processing module 21 sends a working signal to the vibration sensor 31 for real-time vibration perception. When a person or vehicle intrudes and passes by, the vibration sensor 31 detects the target, and the vibration sensor 31 collects the vibration data generated by the target on the shallow surface in real time and transmits it to the signal processing module 21. The photoelectric control module 11 is triggered synchronously, and the image acquisition device 12 is controlled to start, infrared detection begins, images are taken, and the collected image data is transmitted to the signal processing module 21.

[0037] Based on the received vibration data and image data, the signal processing module 21 adopts the vibration photoelectric fusion strategy to determine the final intrusion warning result, and sends control commands to the sound alarm module 23 and the flashing light alarm module 24 respectively according to the intrusion warning result, so that the sound alarm module 23 emits an alarm sound and the flashing light alarm module 24 emits a light. At the same time, the signal processing module 21 also uploads the intrusion warning result and the captured image to the cloud server through the Internet of Things communication module 27 and the communication antenna 13, and the GPS positioning module 25 uploads the current location information to the cloud server synchronously through the GPS antenna 14. Among them, the vibration photoelectric fusion strategy refers to combining the dual-dimensional information of vibration (microseism) and image (i.e., vibration data and image data) by using a multimodal information fusion method, using a lightweight edge-end neural network to extract features with high discrimination, forming a high-dimensional space vector, and using the high-dimensional space vector to identify and classify the target to reduce the false alarm rate and false alarm rate.

[0038] The multimodal information fusion method in this embodiment specifically includes the following steps:

[0039] Firstly, the feature vectors of the original vibration data and image data are trained by edge neural network and fine-grained target recognition technology respectively, and the target classification results and probability distribution of vibration data and image data are obtained by lightweight artificial neural network respectively;

[0040] Secondly, at the decision-making level, a weighted fusion method is adopted to obtain the final target classification result and its probability. The formula for the weighted fusion method is as follows:

[0041] (1)

[0042] where, represents the probability weighted normalization value of the target being the th class result; is the weighted value of the vibration data, and the normalized vibration energy ratio is adopted; is the probability that the vibration data identifies the target as the th class, is the probability that the image data identifies the target as the th class;

[0043] The calculation formula of

[0044] (2)

[0045] where, is the number of points of all vibration data, represents the amplitude of the th vibration data, is the number of points of effective vibration data, represents the amplitude of the th effective vibration data, represents the sum of the energies of the total vibration data, represents the sum of the energies of the effective vibration data. Among them, the effective vibration data refers to the vibration data obtained by only capturing the vibration wave generated by the target and discarding the vibration wave caused by environmental noise in the collected vibration wave.

[0046] When the signal processing module 21 performs target classification and recognition on the original vibration data, it first adopts the method of modal decomposition to perform noise reduction processing on the vibration data collected by the vibration sensor 31. The process of noise reduction processing specifically includes the following steps:

[0047] Use the variational mode decomposition algorithm (VMD) to decompose the vibration signal into modes, as shown in formula (3):

[0048] (3)

[0049] Among the modes obtained after decomposition, contains the most effective components of the target vibration signal, contains the least effective components of the target vibration signal.

[0050] Next, for the modes obtained after decomposition, the modes are divided into information-dominated modes, noise-dominated modes, and pure-noise modes. For the modes dominated by noise components, wavelet threshold denoising is used to process the modes to suppress the noise components; for the pure-noise modes, the entire mode is discarded. The information-dominated modes in each mode are the first to the th mode before the th mode, and the noise-dominated modes are the th mode and several subsequent modes; the first modes are retained, and wavelet threshold denoising is used to process the th mode to the th mode. The processed modes are denoted as , , ……, . The Euclidean distance is used to determine the pure-noise modes in the decomposed modes. When the Euclidean distance satisfies specific judgment conditions, the decomposed modes can be considered pure-noise modes. The judgment conditions are shown in formula (4):

[0051] (4)

[0052] where is the Euclidean distance between the original vibration signal and the th mode , ; is the Euclidean distance between the original vibration signal and the th mode ; is the scaling factor, which can be adjusted according to the degree of restriction on the noise components. For example, is set to 1.1. The mode satisfying the above formula (4) is determined as the demarcation point between the pure-noise modes and the noise-dominated modes. The modes after are considered pure-noise modes, and the pure-noise modes are discarded. Finally, all the modes after denoising are obtained , , ……, . Using the modes , , ……, after denoising, a denoised signal is constructed, as shown in formula (5):

[0053] (5)

[0054] where The remaining modal quantity after noise reduction processing.

[0055] After obtaining the denoised signal the denoised signal is input into a variety of ground moving target recognition models for target recognition, and the target classification results and their probability distributions of the vibration data are obtained.

[0056] In order to achieve fast recognition of ground moving targets, in this embodiment, a lightweight pedestrian and vehicle classification model is built by training feature vectors using an edge-side neural network, and then a variety of vehicle classification models are constructed. The lightweight pedestrian and vehicle classification model and a variety of vehicle classification models are combined into a variety of ground moving target recognition models, and the accuracy rate of the variety of ground moving target recognition models is verified using a variety of moving target sample data and verification data. The samples with classification errors are analyzed, and the variety of ground moving target recognition models are improved by combining machine learning algorithms. The model parameters are adjusted to seek the optimal solution of the variety of ground moving target recognition models, the target classification accuracy rate is improved, and fast recognition of a variety of targets at a relatively long distance with high accuracy is achieved.

[0057] The process of building a lightweight variety of vehicle classification models includes: analyzing the characteristics of vibration signals generated by different vehicles driving in the same environment, and extracting and screening the characteristic values of pedestrian vibration signals and vehicle vibration signals; training feature vectors by combining an edge-side neural network and fine-grained target recognition technology, using the edge-side neural network to process feature data, combining fine-grained target recognition technology to process image feature data, and using the information of local areas to capture minute differences to achieve target recognition. Select a suitable algorithm according to the training accuracy rate and algorithm stability to build a lightweight variety of vehicle classification models. The process of building a lightweight pedestrian and vehicle classification model is similar to that of a lightweight variety of vehicle classification models, and will not be elaborated here.

[0058] Optionally, the target recognition process can also extract the characteristic values of pedestrian vibration signals and vehicle vibration signals by analyzing the differences of different moving target signals, the time-domain characteristics and frequency-domain distributions of different target vibration signals generated, for example Figure 3 as shown, use an LSTM network to extract the time-domain waveform characteristics of the signal to obtain time-domain characteristics; at the same time, use the short-time Fourier transform to characterize the time-frequency characteristics of the signal to obtain a time-frequency diagram, and then use a CNN to extract the characteristics of the low-frequency part of the signal to obtain time-frequency domain characteristics; combine the two feature extraction networks to build a ground moving target recognition model; since the time-domain waveform and frequency-domain energy information of the signal contain characteristics that can characterize the target category, combine a CNN and an LSTM network to build a target recognition model, extract features from the time-domain signal and time-frequency domain of the target, perform feature combination, and finally perform Softmax classification through the recognition layer to obtain the recognition result.

[0059] When the signal processing module 21 performs target classification and recognition on the original image data, it can be implemented using existing image target recognition algorithms, such as the YOLOv5 algorithm, etc., which will not be elaborated here.

[0060] To further improve the imaging quality of the image acquisition device 12, the image acquisition device 12 can perform fill-light shooting by integrating the climate compensation algorithm and the ambient environment data collected by the vibration sensor.

[0061] When the alarm system is working, the vibration sensor 31 performs vibration perception in real time. After the vibration sensor 31 detects a target, it is amplified and filtered through the low-pass filter of the signal conditioning circuit, and then through the calculation and analysis of the signal processing module 21, it can accurately and quickly identify various moving targets such as people (walking, running, jumping), vehicles, etc., and trigger the sound alarm module 23 and the flash alarm module 24. At the same time when the vibration sensor 31 detects a target, the image acquisition device 12 synchronously performs real-time image shooting.

[0062] In this embodiment, through a variety of feature selections, combined with the edge-side neural network and fine-grained target recognition technology, a lightweight ground multi-moving target recognition model is built, and the model is optimized by combining machine learning to achieve the recognition of ground multi-moving targets, and then judge whether the target is a person, a vehicle, or others. Turn on the sound warning and the light flashing warning, and provide a trigger signal to the multi-sensor load unit 10 to take pictures of the ground moving targets, providing a basis for subsequent tracking.

[0063] The alarm system of this embodiment can prevent and control all-weather, collect and analyze the vibration signals generated by moving targets, cooperate with the multi-sensor load unit 10 and the signal acquisition unit 30 to reduce the false alarm rate, and the communication means are diversified.

[0064] Furthermore, the cloud server displays the received intrusion warning results and images, monitors the occurrence of illegal intrusion behaviors in the defense area in real time, and gives an alarm reminder. At the same time, the cloud server is also used to respond to the user's status query and parameter configuration requests for the multi-sensor load unit 10 and the main control communication processing unit 20. The monitoring personnel can perform status query and parameter configuration on any vibration-optical fusion type unattended alarm device for intelligent management and control. Reduce the work intensity of the management department by means of technology and ensure the perimeter security.

[0065] The unattended perimeter alarm devices in this embodiment can be set to multiple, as Figure 2 shown, and multiple unattended perimeter alarm devices respectively transmit data to the cloud server directionally through their respective Internet of Things communication modules and communication antennas, and the cloud server uploads the data to the control center or the mobile phone APP for data interaction.

[0066] A vibration-optoelectronic fusion type unattended perimeter alarm system proposed by the present invention is applicable to a variety of complex climate environments, uses a variety of wireless communication modules to complete data interaction, and reports the early warning status to the cloud server in real time, realizing an intelligent perimeter security system, reducing the work intensity of the management department by means of technology, improving the work efficiency of the management department, and ensuring perimeter security.

[0067] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0068] The above-described embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it cannot be understood as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent should be subject to the appended claims.

Claims

1. A vibration photoelectric fusion type unmanned perimeter alarm system, characterized in that: The invention comprises at least one unmanned perimeter alarm device, wherein the unmanned perimeter alarm device comprises a multi-sensor load unit (10), a main control communication processing unit (20) and a signal acquisition unit (30), wherein the multi-sensor load unit (10) comprises a photoelectric control module (11), an image acquisition device (12), a communication antenna (13) and a GPS antenna (14), the main control communication processing unit (20) comprises a signal processing module (21), a power module (22), a sound alarm module (23), a flashing light alarm module (24), a GPS positioning module (25) and an Internet of Things communication module (27), and the signal acquisition unit (30) comprises a vibration sensor (31); The vibration sensor (31) senses vibration in real time. When the vibration sensor (31) detects a target, it outputs vibration data to the signal processing module (21). The photoelectric control module (11) is synchronously triggered and controls the image acquisition device (12) to start. The image acquisition device (12) transmits the acquired image data to the signal processing module (21). The signal processing module (21) uses a vibration photoelectric fusion strategy to determine the final intrusion warning result, and sends control commands to the sound alarm module (23) and the flashing light alarm module (24) according to the intrusion warning result, so that the sound alarm module (23) emits an alarm sound and the flashing light alarm module (24) emits a light. At the same time, the signal processing module (21) also sends a control command to the sound alarm module (23) and the flashing light alarm module (24) through the physical The network communication module (27) and the communication antenna (13) upload the intrusion warning results and the captured images to the cloud server, and the GPS positioning module (25) synchronously uploads the current location information to the cloud server through the GPS antenna (14). The cloud server displays the received intrusion warning results and images, and is also used to respond to user status inquiries and parameter configuration requests for the multi-sensor load unit (10) and the main control communication processing unit (20), wherein the vibration photoelectric fusion strategy refers to combining the dual-dimensional information of vibration and image by using a multi-modal information fusion method, and using a lightweight edge-end neural network to extract features with high discrimination, forming a high-dimensional space vector, and identifying and classifying the target; The multimodal information fusion method comprises the following steps: The edge neural network and fine-grained target recognition technology are used to train feature vectors for the original vibration data and image data, and a lightweight artificial neural network is used to obtain the target classification results and probability distribution of the vibration data and the target classification results and probability distribution of the image data respectively; The weighted fusion method is used at the decision level to obtain the final target classification result and its probability. The formula of the weighted fusion method is as follows: Among them, P i represents the probability weighted normalized value of the target being the i-th type result; W1 is the weighted value of the vibration data, using the normalized vibration energy ratio; P 1i is the probability that the vibration data identifies the target as the i-th category, P 2i is the probability that the image data recognizes the target as the i-th category; the calculation formula of W1 is as follows: Where N is the total number of vibration data points, x n represents the amplitude of the nth vibration data, M is the number of valid vibration data points, y m represents the amplitude of the mth valid vibration data, Represents the sum of the energy of the total vibration data, Represents the sum of the energy of valid vibration data; When the signal processing module (21) performs target classification and recognition on the original vibration data, it first uses a modal decomposition method to perform noise reduction processing on the vibration data collected by the vibration sensor (31). The noise reduction processing process includes the following steps: The vibration signal s(t) is decomposed into K modes using the variational mode decomposition algorithm. The formula is as follows: The K modes obtained after decomposition are divided into information-dominated modes, noise-dominated modes and pure noise modes, where the information-dominated modes are from the 1st mode to the m-1th mode, the noise-dominated modes start from the mth mode, and the dividing point u between the noise-dominated mode and the pure noise mode is p (t) is determined by the following formula: Among them, D j is the original vibration signal and the jth mode u j (t), 1≤j≤K; D i is the original vibration signal and the i-th mode u i (t) Euclidean distance between them; β is the scaling factor; u p The modes after (t) are considered as pure noise modes. After retaining the information-dominant modes and discarding the pure noise modes, the noise-dominant modes are processed by wavelet threshold denoising to obtain all the modes after denoising u1'(t), u'2(t), ..., u' p (t); Using the noise-reduced modes u1'(t), u'2(t), ..., u' p (t) constructs the denoised signal s'(t), the formula is as follows: Where p is the number of remaining modes after denoising; The denoised signal s'(t) is input into a ground multiple motion target recognition model for target recognition, and the target classification result of the vibration data and its probability distribution are obtained, wherein the ground multiple motion target recognition model is obtained by combining a lightweight human-vehicle classification model and a multiple vehicle classification model, and the accuracy of the ground multiple motion target recognition model is verified using multiple motion target sample data and verification data, and the misclassified samples are analyzed, and the ground multiple motion target recognition model is improved in combination with a machine learning algorithm, and the model parameters are adjusted to seek the optimal solution of the ground multiple motion target recognition model.

2. The vibration photoelectric fusion type unattended perimeter alarm system according to claim 1 is characterized in that: The main control communication processing unit (20) also includes a control button module (26) connected to the signal processing module (21), and the control button module (26) is used to select a working mode of the alarm system.

3. The vibration photoelectric fusion type unmanned perimeter alarm system according to claim 2 is characterized in that: The working mode includes a warning mode, and the warning mode is: a) the flashing light of the flashing light alarm module (24) remains in a slow flashing state. After the intrusion warning result is determined, the flashing frequency is increased from far to near according to the target distance. After the intrusion warning is lifted, the flashing light alarm module (24) is turned off; b) After determining the intrusion warning result, the sound alarm module (23) emits an alarm sound and speeds up the alarm sound from far to near according to the target distance.

4. The vibration photoelectric fusion type unattended perimeter alarm system according to claim 2 is characterized in that: The working mode includes a sentinel mode, and the sentinel mode is: The flashing light alarm module (24) is closed by default. After the intrusion warning result is determined, the flashing light of the flashing light alarm module (24) starts to flash and the flashing frequency increases from far to near according to the target distance. The sound alarm module (23) emits an alarm sound. After the intrusion warning is lifted, the flashing light alarm module (24) is closed again.

5. The vibration photoelectric fusion type unmanned perimeter alarm system according to claim 2 is characterized in that: The working mode includes a duty mode, and the duty mode is: The flashing light alarm module (24) is closed by default and will not be turned on even after the intrusion warning result is confirmed.

6. The vibration photoelectric fusion type unmanned perimeter alarm system according to claim 1 is characterized in that: The Internet of Things communication module (27) adopts any one of a 2G module, a 4G module, a NB-IoT module, a LoRa module, a Wi-Fi module and a BlueTooth module, or a combination of any of the above.

Images