CN114584924A - Intelligent unattended sensor system and target identification method - Google Patents
Intelligent unattended sensor system and target identification method Download PDFInfo
- Publication number
- CN114584924A CN114584924A CN202210191616.6A CN202210191616A CN114584924A CN 114584924 A CN114584924 A CN 114584924A CN 202210191616 A CN202210191616 A CN 202210191616A CN 114584924 A CN114584924 A CN 114584924A
- Authority
- CN
- China
- Prior art keywords
- signal
- unit
- processing unit
- model
- communication module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000004891 communication Methods 0.000 claims abstract description 72
- 238000012545 processing Methods 0.000 claims abstract description 52
- 238000001514 detection method Methods 0.000 claims abstract description 31
- 238000006243 chemical reaction Methods 0.000 claims abstract description 10
- 239000000203 mixture Substances 0.000 claims description 24
- 238000012544 monitoring process Methods 0.000 claims description 21
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 19
- 238000009432 framing Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 11
- 238000004458 analytical method Methods 0.000 claims description 7
- QVFWZNCVPCJQOP-UHFFFAOYSA-N chloralodol Chemical compound CC(O)(C)CC(C)OC(O)C(Cl)(Cl)Cl QVFWZNCVPCJQOP-UHFFFAOYSA-N 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 5
- FBOUIAKEJMZPQG-AWNIVKPZSA-N (1E)-1-(2,4-dichlorophenyl)-4,4-dimethyl-2-(1,2,4-triazol-1-yl)pent-1-en-3-ol Chemical compound C1=NC=NN1/C(C(O)C(C)(C)C)=C/C1=CC=C(Cl)C=C1Cl FBOUIAKEJMZPQG-AWNIVKPZSA-N 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- 230000035939 shock Effects 0.000 claims 1
- 230000010354 integration Effects 0.000 abstract description 4
- 230000008901 benefit Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 22
- 238000005516 engineering process Methods 0.000 description 18
- 238000009826 distribution Methods 0.000 description 15
- 239000011159 matrix material Substances 0.000 description 11
- 238000004364 calculation method Methods 0.000 description 10
- 238000000605 extraction Methods 0.000 description 9
- 230000007123 defense Effects 0.000 description 7
- 239000013598 vector Substances 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000010295 mobile communication Methods 0.000 description 6
- 239000000523 sample Substances 0.000 description 6
- 238000009434 installation Methods 0.000 description 5
- 241001465754 Metazoa Species 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000006855 networking Effects 0.000 description 4
- 239000004575 stone Substances 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 238000012935 Averaging Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 230000005389 magnetism Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000001338 self-assembly Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000009412 basement excavation Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 229910052744 lithium Inorganic materials 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000013307 optical fiber Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 239000004576 sand Substances 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 229920000642 polymer Polymers 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000013138 pruning Methods 0.000 description 1
- 230000005616 pyroelectricity Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000002945 steepest descent method Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/021—Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/06—Optimizing the usage of the radio link, e.g. header compression, information sizing, discarding information
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W72/00—Local resource management
- H04W72/50—Allocation or scheduling criteria for wireless resources
- H04W72/54—Allocation or scheduling criteria for wireless resources based on quality criteria
- H04W72/542—Allocation or scheduling criteria for wireless resources based on quality criteria using measured or perceived quality
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W84/00—Network topologies
- H04W84/18—Self-organising networks, e.g. ad-hoc networks or sensor networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Quality & Reliability (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses an intelligent unattended sensor system, which comprises a power supply conversion unit, and a seismic detection unit, a signal processing unit, a data processing unit and a communication unit which are sequentially connected; the earthquake detection unit is used for converting the earthquake signals into electric signals, the signal processing unit is used for converting the electric signals into digital signals, and the data processing unit is used for analyzing and processing the digital signals to obtain target identification results. The invention also discloses a target identification method based on the system, which comprises the following steps: s1, the seismic wave detection unit is used for converting the vibration signal into an electric signal; s2, the signal processing unit is used for converting the electric signal into a digital signal; s3, the data processing unit is used for processing the digital signal to obtain an identification result; and S4, the communication unit is used for outputting the target recognition result. The invention has the advantages of flexible layout, good concealment, high recognition rate, multiple recognition types, strong anti-interference, good integration, all-weather work and the like.
Description
Technical Field
The invention mainly relates to the technical field of target detection, in particular to an intelligent unattended sensor system and a target identification method.
Background
With the development of informatization technology, important key area security early warning is continuously developed towards the direction of combining networking, digitalization, intellectualization and multiple technical means. From the last century of 'civil air defense combination defense' to 'technical defense', the development goes through the first generation of security early warning technology, to the second generation of early warning technology which takes signal driving as the characteristic, and is gradually transiting to the third generation of early warning technology which takes target driving as the characteristic at present.
The second generation of signal driving type security early warning technology is mainly represented by the traditional technologies of vibrating optical fibers, leaking cables, infrared correlation, microwave correlation, radars, videos and the like, and the technologies are respectively long in length and have a certain promotion effect on improvement of security level. However, these conventional techniques have the frequent occurrence of alarm leakage and false alarm, and cannot realize all-weather, all-day-long and three-dimensional real-time monitoring and protection. In recent years, advanced and reliable digital processing technology has been completely used in alarm processing platforms of security early warning systems and centers, and the technology is gradually mature. The frontmost devices of the system, such as the detectors, have not made any greater technical breakthrough, but rather the stability of the system is improved by signal post-processing techniques or by a combination of several detection techniques or by the addition of a single technique. Therefore, the technical lag of the source of the alarm system is a weak link causing the problems of false and missed alarm of the whole system, so that the problem of false and missed alarm of the detector is solved well, and the method is the fundamental point for reducing the false and missed alarm rate of the whole alarm system. The third generation of security early warning technology is based on advanced passive seismic wave detection means, a vibration sensor obtains target detection data, and processing equipment processes and fuses the detection data to realize detection and identification of an invasion target.
At present, the development and application of the intelligent vibration ground sensor system in China are not systematic and are still in a lower level. Most of the existing research achievements are still in an experimental stage or an algorithm research stage, and have no usability, and some achievements only have the problems of single target recognition capability or poor anti-interference capability and the like, cannot adapt to various complex application occasions, and have poor practicability and applicability.
At present, the guard and defense of regions such as military bases, key infrastructures, frontiers and the like mainly depend on personnel patrol, surveillance cameras, vibration optical cables and other reconnaissance and monitoring systems, and in recent years, unmanned aerial vehicles are also gradually used for executing region patrol and reconnaissance and monitoring tasks. However, the above method and system still have a series of problems in warning and protection of critical facilities, border sea defense and other areas:
1) technical means such as electronic fences, infrared correlation, vibration optical cables and the like which are commonly used at present are easy to interfere, the false alarm rate is high, wind sand, vehicles, animals, rolling stones/falling stones can trigger alarm, and target types (small animals, people, vehicles, falling stones and the like) cannot be distinguished;
2) common equipment such as videos, radars and the like has limited visible distance, is easily influenced by rain, snow, weather and the like, is easily shielded under complex landforms, and is not easy to find hidden/disguised targets and underground excavated targets;
3) the unmanned aerial vehicle continuously moves in the monitoring area, although local monitoring information can be continuously updated through autonomous cooperation, uninterrupted monitoring on an interest point is difficult to maintain, meanwhile, the endurance of the unmanned aerial vehicle is limited, continuous and uninterrupted monitoring capability cannot be provided, the number of the unmanned aerial vehicles is increased, and the problem of complex multi-machine cooperative control is solved;
4) in mountainous areas/wasteland areas, border areas and other areas far away from cities and towns, the terrains and the geomorphology of the areas are complex, infrastructure guarantee does not exist, the arrangement and the installation of systems such as radars, videos and vibrating optical fibers are difficult, the self-protection capability is poor, targets are striking, the systems are easy to find, avoid and damage, and the cooperative sensing among the devices is difficult to realize due to the fact that power supply, communication networks and the like cannot be covered.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: aiming at the problems in the prior art, the invention provides an intelligent unattended sensor system which is flexible in arrangement, good in concealment, high in recognition rate, multiple in recognition type, strong in anti-interference and good in integration, and correspondingly provides a target recognition method with high recognition precision.
In order to solve the technical problems, the technical scheme provided by the invention is as follows:
an intelligent unattended sensor system comprises a seismic wave detection unit, a signal processing unit, a data processing unit, a communication unit and a power supply conversion unit; the seismic wave detection unit, the signal processing unit, the data processing unit and the communication unit are sequentially connected; the power supply conversion unit is connected with each unit and is used for providing power supply for each unit; the earthquake detection unit is used for converting a vibration signal generated by target intrusion into an electric signal, the signal processing unit is used for converting the electric signal into a digital signal, the data processing unit is used for carrying out analysis processing on the digital signal to obtain a target identification result, and the communication unit is used for outputting the target identification result to external equipment.
Preferably, the communication unit comprises a wired communication module or a wireless communication module, and the wired communication module comprises a CAN communication module or/and an RS485 communication module; the wireless communication module comprises one or more of a Lora communication module, a Zigbee communication module, a 4G communication module or a 5G communication module.
Preferably, the seismic wave detection unit comprises a plurality of vibration sensors, and the plurality of vibration sensors form a sensor self-assembly communication network system to realize cooperative sensing and real-time monitoring of the plurality of vibration sensors in a deployment area; and a Zigbee ad hoc network is adopted among the vibration sensors to complete interconnection among the vibration sensors.
Preferably, the seismic detecting unit is a moving coil magnetoelectric vibration sensor.
The invention also discloses a target identification method based on the intelligent unattended sensor system, which comprises the following steps:
s1, the seismic wave detection unit is used for converting the vibration signal generated by the target intrusion into an electric signal;
s2, the signal processing unit is used for converting the electric signal into a digital signal;
s3, the data processing unit is used for analyzing and processing the digital signal to obtain a target recognition result;
and S4, the communication unit is used for outputting the target recognition result to an external device.
Preferably, in step S3, the specific process of analyzing and processing the digital signal to obtain the target recognition result is as follows:
s3.1, noise reduction and framing are carried out on the digital signals;
s3.2, simultaneously extracting a plurality of time domain features of the framed data;
s3.3, taking the plurality of time domain characteristics obtained in the step S3.2 as training samples, and respectively training the Gaussian mixture model and the xgboost model to obtain the trained Gaussian mixture model and the trained xgboost model;
s3.4, respectively identifying the signals through the trained Gaussian mixture model and the trained xgboost model, and outputting corresponding identification results;
and S3.5, synthesizing the recognition results output by the Gaussian mixture model and the xgboost model by adopting a decision algorithm, and outputting a final recognition result.
Preferably, in step S3.2, four time domain features are extracted by using a method of extracting signal features, specifically: extracting the peak-to-peak value of the framed data at a fixed time interval to serve as a first time domain feature; extracting the kurtosis of the framed data at a fixed time interval to serve as a second time domain feature; calculating the sum of the maximum n point values of each frame of dataiAs a third time domain feature; will sumiSum of n point values which is maximum from the previous framei-1Is taken as the fourth time-domain feature.
Preferably, the extraction process of the four time domain features is as follows: with x1Points are a time window, x2The point is the sliding step length, so that the h-dimension peak-to-peak characteristic F _ ppv is obtainedi＝[p1,p2,..,ph]；
To (n)1+n2) Window signal of s in x3Points are a time window, x4The point is the sliding step length, and the kurtosis characteristic F _ kurtosis of w dimension is obtained by framingi＝[k1,k2,...,kw]Sum of the maximum n point values per frame data sumiAnd its maximum n point values and sum with the previous framei-1Is characteristic of the ratio ofi＝Sumi/Sumi-1。
Preferably, in step S3.4, for each class prediction probability given by the GMM model and the xgboost model, a threshold shifting method is adopted, and the recognition result is considered to be credible if the probability is greater than a preset value.
Preferably, the specific process of step S3.5 is:
adopting scoring logics of a Gaussian mixture model and an xgboost model game: at the time of first decision, the output of one model is randomly selected from the two models as the total output, and the real label y is comparediIf the xgboost model is correct and the GMM is incorrect, then the xgboost model adds 1 point, and the next predicted total output is determined by the xgboost model output, or vice versa; if the two models predict correctly or incorrectly, the scores are not added, and the total output of the next round of prediction is determined by the party with more scores at present.
Compared with the prior art, the invention has the advantages that:
the invention is suitable for military/civil security and protection fields such as edge sea defense, key infrastructure protection, military operation, pipelines/wells, barracks and the like; the method can be independent of the foundation facility guarantee of commercial power, public network and the like, and has the characteristics of flexible layout, good concealment, high recognition rate, multiple recognition types, strong interference resistance, good integration, all-weather work and the like. The invention can expand various sensors such as sound, magnetism, pyroelectric, video, radar, unmanned aerial vehicle and the like, fuse various information sources from different sensors, and comprehensively analyze to obtain stable and reliable information for correctly understanding the surrounding environment, so that the system has stronger fault tolerance.
The invention adopts two design forms of wire and wireless, wherein the wire form adopts a wire power supply/communication mode, and the wireless form adopts a battery/wireless ad hoc network/4G/5G communication mode, thus realizing the installation and use in urban environment and also realizing the installation and use in facilities and security regions without basic power supply and the like. When the intelligent vibration sensor system is arranged in a city or other areas depending on mobile communication, a ground sensor network can be constructed by adopting mobile communication network resources, and when the sensor system is arranged in the field and other application scenes depending on no communication, the ground sensor network can be constructed by an ad hoc network communication module which is independently developed, so that 24-hour uninterrupted monitoring on 'interest points' in all weather is realized, and the cooperative sensing is realized through the ad hoc network.
According to the method, the complex target classification is realized through a deep learning algorithm, the identification accuracy can be improved by increasing the sample amount, and the identification performance superior to that of the traditional machine learning method is obtained; different application scenes are divided, targeted samples are collected for training, and a training feature library which is most matched with the actual environment is selected to match the recognition target when the training feature library is used. The GMM and the xgboost model are trained through a large number of samples to obtain the trained model, and the model identification has universality; as long as different time domains of the target signal are different, the method can obtain better identification performance; by extracting the time domain characteristic quantity, the input data of the model is reduced, the calculation quantity in the operation process is greatly reduced, the neural network calculation can be realized on a microprocessor with limited calculation resources, and the algorithm can be deployed in an embedded system; the output of the whole model is comprehensively judged through mutual game of the two model results, and the method is more accurate than the traditional voting method and the averaging method.
Drawings
FIG. 1 is a schematic diagram of the operation of an embodiment of the system of the present invention.
Fig. 2 is a block diagram of the system of the present invention according to the first embodiment.
Fig. 3 is a block diagram of the system of the present invention in the second embodiment.
Fig. 4 is a schematic diagram of the ad hoc network communication network according to the second embodiment of the system of the present invention.
Fig. 5 is a block diagram of the intelligent vibration sensor in the second embodiment of the system of the present invention.
Fig. 6 is a block diagram of a wireless receiving device according to a second embodiment of the system of the present invention.
FIG. 7 is a flow chart of a method of the present invention in an embodiment.
Detailed Description
The invention is further described below with reference to the figures and the specific embodiments of the description.
The first embodiment is as follows:
as shown in fig. 2, the intelligent unattended sensor system according to the embodiment of the present invention includes a seismic detection unit, a signal processing unit, a data processing unit, a communication unit, and a power conversion unit; the earthquake wave detection unit, the signal processing unit, the data processing unit and the communication unit are connected in sequence; the power conversion unit is connected with each unit and used for providing power for each unit; the earthquake detection unit is used for converting a vibration signal generated by target intrusion into an electric signal, the signal processing unit is used for converting the electric signal into a digital signal, the data processing unit is used for analyzing and processing the digital signal to obtain a target identification result, and the communication unit is used for outputting the target identification result to external equipment. Wherein above-mentioned communication unit is wired communication module, including CAN communication module or/and RS485 communication module, will discern categorised result output to monitoring platform respectively through above two kinds of communication modules and carry out the result and show.
In a specific embodiment, the seismic wave detection unit comprises a plurality of vibration sensors and a probe, and a plurality of moving coil magnetoelectric vibration sensors form a sensor self-assembly communication network system to realize cooperative sensing and real-time monitoring of the plurality of vibration sensors in a deployment area; the vibration sensor adopts a moving coil magnetoelectric vibration sensor, the vibration monitoring range is 14-500Hz, the amplitude is 0db-80db, the personnel detection range is within 25 meters, the vehicle detection range is within 50 meters of the wheeled vehicle and within 100 meters of the tracked vehicle; the probe is used for picking up the vibration signal and transmitting the vibration signal to the vibration sensor.
In an embodiment, the signal processing unit mainly includes an AD processor and a signal processing circuit, and is of a 24-bit ADs131M04 type, and is configured to sample and convert an electrical signal generated by vibration into a digital signal with high stability and process the digital signal. The data processing unit mainly comprises a microcontroller, wherein the microcontroller adopts an STM32F722RET6 model and is used for acquiring the digital signals output by the signal processing unit and simultaneously carrying out target recognition algorithm processing on the digital signals to obtain the information of the intrusion target recognition classification result.
In one embodiment, the power conversion unit mainly includes an external/on-board voltage conversion module for converting an external input voltage into a voltage suitable for each unit of the intelligent vibration sensor system, thereby ensuring the reliability of the system operation.
The invention is suitable for military/civil security and protection fields such as edge sea defense, key infrastructure protection, military operation, pipelines/wells, barracks and the like; the method can be independent of the foundation facility guarantee such as commercial power, public network and the like, and has the characteristics of flexible layout, good concealment, high recognition rate, multiple recognition types, strong interference resistance, good integration, all-weather work and the like. The invention can expand various sensors such as sound, magnetism, pyroelectricity, video, radar, unmanned aerial vehicle and the like, fuse various information sources from different sensors, and comprehensively analyze to obtain stable and reliable information for correctly understanding the surrounding environment, so that the system has stronger fault tolerance.
Example two:
the difference between this embodiment and the first embodiment is only the communication method. The first embodiment adopts a wired communication mode, and is suitable for urban environment; the embodiment of the invention adopts a wireless communication mode as shown in fig. 3, and is suitable for large-area environments such as complex severe regions or borders. Specifically, when a mobile communication region depends on, a mobile communication network can be adopted to transmit the identification and classification result information to a monitoring platform for early warning prompt; when the application scene without communication support is carried out, the information of the identification and classification result can be transmitted to the monitoring platform through the ad hoc communication network for early warning prompt.
Wherein the ad hoc communication network is schematically illustrated in fig. 4. When a plurality of vibration sensors are used simultaneously, a sensor self-assembly communication network system is formed, and multi-sensor cooperative sensing and real-time monitoring of a deployment area are achieved. The sensor layer adopts a Zigbee ad hoc network technology to complete interconnection among sensors in a small range, the sensors and the wireless receiving equipment adopt a Lora networking technology to realize the receiving and aggregation of sensor data within the range of 2-5 kilometers, and the sensor information is reported to the monitoring platform after being gathered. In order to realize the receiving and gathering of the sensor information in a larger range, repeater equipment can be designed, the receiving and sending of the equipment adopt the Lora networking technology, and the receiving and gathering of the sensor data in a range larger than 5 kilometers are realized.
The ad hoc network of the above embodiment has the following features:
1) robustness: after the system is laid, automatic networking can be realized without control, and routing can be automatically switched after partial nodes fail, so that data communication is not disconnected, and the survivability is strong.
2) Safety: the channel transmission encryption is realized by utilizing a grouping-based sensor security strategy and adopting a multi-row matrix key distribution technology and a Gaussian distribution technology, and the self-communication module is provided with a self-destruction program which can be automatically started when an illegal access is encountered or a self-destruction command is received.
As shown in fig. 5 and 6, specifically, the communication unit is a wireless communication module and mainly includes Lora/Zigbee/4G/5G wireless communication modules, the Lora/Zigbee wireless communication module transmits the identification and classification result to the wireless receiving device through the external antenna unit, so that the identification and classification signals of not less than 1000 intelligent vibration sensors can be received, and the 4G/5G wireless communication module directly outputs the identification and classification result to the monitoring platform for result display.
In one embodiment, the voltage conversion unit is used for converting the battery supply voltage into the applicable voltage of each unit of the intelligent vibration sensor. The battery power supply unit mainly comprises a charging management circuit and a polymer lithium battery, and the standby time is more than 30 days. The battery power supply unit may further include an energy collection module, such as a solar collection module, for collecting solar energy to power the vibration sensor to extend the standby time. Also, a higher capacity lithium battery may be selected or the number of batteries may be increased to extend the standby time.
In a specific embodiment, the data processing unit mainly comprises a microcontroller, the microcontroller adopts an STM32F103C8T7 model low-power-consumption processor for receiving and processing the identification and classification information of the plurality of sensors, and simultaneously, the identification and classification results are output to the monitoring platform through the Bluetooth communication module for result display.
The wireless version of the intelligent vibration sensor system can realize cooperative sensing and remote transmission among the sensors through the ad hoc network communication network; the wireless version of the intelligent vibration sensor system also comprises a solar energy acquisition module which is used for acquiring solar energy to supply power to the vibration sensor so as to prolong the standby time; the wireless version of the intelligent vibration sensor system can automatically realize the switching between the sleep mode and the working mode so as to prolong the standby time; the wireless version of the intelligent vibration sensor system adopts a sensor security strategy based on grouping, utilizes a multi-row matrix key distribution technology and a Gaussian distribution technology to realize channel transmission encryption, and the self-communication module is provided with a self-destruction program which can be automatically started when illegal access is encountered or a self-destruction command is received.
The device is buried underground, and the target is identified through the vibration information generated by the target, so that the device is not influenced by the occlusion of complex landforms and the hiding and disguising of the target, can detect and identify the underground excavation target, has good concealment, and is easy to avoid artificial damage.
The invention adopts two design forms of wire and wireless, wherein the wire form adopts a wire power supply/communication mode, and the wireless form adopts a battery/wireless ad hoc network/4G/5G communication mode, thus realizing the installation and use in urban environment and also realizing the installation and use in facilities and security regions without basic power supply and the like. When the intelligent vibration sensor system is arranged in a city or other areas depending on mobile communication, a ground sensor network can be constructed by adopting mobile communication network resources, and when the sensor system is arranged in the field and other application scenes depending on no communication, the ground sensor network can be constructed by an ad-hoc network communication module which is independently developed, so that 24-hour all-weather uninterrupted monitoring on 'interest points' and cooperative sensing through the ad-hoc network are realized.
The design of the invention fully considers the expansion of internal and external interfaces, and various sensors such as internal expandable sound, magnetism, pyroelectric and the like and external sensor nodes (video, radar, unmanned aerial vehicle and the like) form a sensor network through wired/wireless communication links, so that the multi-source detection and cooperative identification and positioning are carried out on the target, and the concealment and disguise of the target are effectively cracked. The difference and complementarity of the performance of various sensors are fully utilized, various information sources from different sensors are fused, and stable and reliable information for correctly understanding the surrounding environment is obtained through comprehensive analysis, so that the system has stronger fault tolerance, and the speed of system information processing and the correctness of decision are improved.
According to the invention, the vibration sensors are arranged at certain intervals on the periphery of the control area, and in the detection range of the sensors (the circumferential range taking the arrangement points of the sensors as the circle center), once personnel and vehicles pass through, fall rocks or underground excavation activities, the sensor system immediately transmits the identification and classification result information to the monitoring platform in a wireless mode or a wired mode for early warning and prompting, and the arrangement intervals of the sensors are determined according to the actual use environment.
As shown in fig. 1 and 7, an embodiment of the present invention further provides a target identification method based on the intelligent unattended sensor system, which includes the steps of:
s1, the seismic wave detection unit is used for converting the vibration signal generated by the target intrusion into an electric signal;
s2, the signal processing unit is used for converting the electric signal into a digital signal;
s3, the data processing unit is used for analyzing and processing the digital signal to obtain a target recognition result;
and S4, the communication unit is used for outputting the target recognition result to the external equipment.
In an embodiment, in step S3, the specific process of analyzing and processing the digital signal to obtain the target recognition result includes:
s3.1, noise reduction and framing are carried out on the digital signals;
s3.2, simultaneously extracting a plurality of time domain features of the framed data;
s3.3, taking the plurality of time domain features obtained in the step S3.2 as training samples, and respectively training the Gaussian mixture model and the xgboost model to obtain the trained Gaussian mixture model and the trained xgboost model;
s3.4, respectively identifying the signals through the trained Gaussian mixture model and the trained xgboost model, and outputting corresponding identification results;
and S3.5, synthesizing the recognition results output by the Gaussian mixture model and the xgboost model by adopting a decision algorithm, and outputting a final recognition result.
According to the method, through a deep learning algorithm, the classification of complex targets is realized, and the identification accuracy can be improved by increasing the sample amount, so that the identification performance superior to that of the traditional machine learning method is obtained; different application scenes are divided, targeted samples are collected for training, and a training feature library which is most matched with the actual environment is selected to match the recognition target when the training feature library is used.
The GMM and the xgboost model are trained through a large number of samples to obtain the trained model, and the model identification has universality; as long as different time domains of the target signal are different, the method can obtain better identification performance; by extracting the time domain characteristic quantity, the input data of the model is reduced, the calculation quantity in the operation process is greatly reduced, the neural network calculation can be realized on a microprocessor with limited calculation resources, and the algorithm can be deployed in an embedded system; the output of the whole model is comprehensively judged through mutual game of the two model results, and the method is more accurate than the traditional voting method and the averaging method.
In a specific embodiment, in step S3.2, four time domain features are extracted by using a method for extracting signal features, specifically: extracting the peak-to-peak value of the framed data at a fixed time interval to serve as a first time domain feature; extracting the kurtosis of the framed data at a fixed time interval to serve as a second time domain feature; calculating the sum of the maximum n point values of each frame of dataiAs a third time domain feature; will su miSum of n point values which is maximum from the previous framei-1Is taken as the fourth time-domain feature.
The extraction process of the four time domain features is as follows: with x1Points are a time window, x2The point is the sliding step length, so that the h-dimension peak-to-peak characteristic F _ ppv is obtainedi＝[p1,p2,..,ph]；
To (n)1+n2) Window signal of s in x3Points are a time window, x4The point is the sliding step length, and the kurtosis characteristic F _ kurtosis of w dimension is obtained by framingi＝[k1,k2,...,kw]Sum of the maximum n point values per frame data sumiAnd its maximum n point values and sum with the previous framei-1Is characteristic of the ratio ofi＝Sumi/Sumi-1。
In a specific embodiment, in step S3.4, for each class prediction probability given by the GMM model and the xgboost model, a threshold shifting method is adopted, and the recognition result is considered to be authentic only when the probability is greater than a preset value, so as to improve the accuracy of signal recognition.
In a specific embodiment, the specific process of step S3.5 is:
adopting scoring logics of a Gaussian mixture model and an xgboost model game: at the time of the first decision, the output of one model is randomly selected from the two models as the total output, and the real label y is comparediIf the xgboost model is correct and the GMM is incorrect, then the xgboost model adds 1 point, and the next predicted total output is determined by the xgboost model output, or vice versa; if the two models predict correctly or incorrectly, the scores are not added, and the total output of the next round of prediction is determined by the party with more scores at present. The output of the whole model is comprehensively judged through mutual game of the two model results, and the method is more accurate than the traditional voting method and the averaging method.
The following is a more complete description of the analysis of the above signals to obtain the target recognition result according to a complete embodiment:
pretreatment: and preprocessing the acquired vibration signals, including noise reduction, band-pass filtering and framing. Aiming at that the vibration signal of the target object is mainly a low-frequency signal, a wavelet self-adaptive threshold denoising algorithm is used firstly, a wavelet threshold function is determined in a self-adaptive mode according to the characteristic of noise under wavelet transformation, the peak signal-to-noise ratio of the denoised signal is used as a performance measurement index, and then band-pass filtering is carried out to filter most of high-frequency noise; and a dynamic framing method for segmenting the weakest signal segment is adopted, the position of the weakest position at the tail part of the signal is automatically calculated, framing is carried out, and the complete signal is retained to the maximum extent.
Specifically, the noise reduction process is as follows: the wavelet transformation is called a mathematical microscope for signal analysis, can highlight local characteristics of signals in a time domain and a frequency domain, and has low entropy, decorrelation, multi-resolution and base selection flexibility. The method adopts gradually-refined time-frequency sampling step length for high-frequency components, and can highlight the details of the object. A self-adaptive threshold denoising algorithm is provided based on different properties of signals and noise under wavelet transformation, a wavelet threshold function is self-adaptively determined according to the characteristics of the noise under the wavelet transformation, and the peak signal-to-noise ratio of denoised signals is used as a performance measurement index.
Let X be ═ X0,x1,x2,…,xN-1]As observed values of the noisy signal, namely:
xi＝si+ni，i＝0,1,2,…,N-1 (1)
xiis the signal value, s, of a certain moment of the noisy signal xiIs the true value of the signal at time i, niIs independently distributed white Gaussian noise, and the invention aims to find out the estimated value of the signal S according to the observed value X
Is the true value s of the signal at time iiAn estimated value of
The minimum Mean Square Error (MSE) with S is minimal, replacing the mathematical expectation with the mean value:
applying discretization on the basis of wavelet coefficients of various scalesInverse wave transform (IDWT) to obtain an estimate of signal s
In the above method, the basis of wavelet transform and inverse wavelet transform generally adopts orthogonal wavelet basis, and at this time, according to Parseval formula, it can be obtained:
where j is the scale of the wavelet decomposition, cdj,kFor the kth wavelet coefficient at wavelet decomposition scale j,
is cdj,kAn estimate of (d).
The traditional soft threshold function is generally a piecewise function, and even if the piecewise function is continuous, the derivative of the piecewise function is not necessarily continuous, which cannot meet the adaptive iteration requirement of the threshold, so the invention adopts the following threshold function, the function has infinite continuous derivative, and the adaptive iteration operation can be carried out to dynamically seek the optimal threshold:
wherein x is a noisy signal and t is a threshold.
The algorithm for denoising the adaptive filter adopts a minimum mean square error algorithm, namely an LMS algorithm. The LMS algorithm is based on the steepest descent method in the optimization method, i.e. the threshold t (k +1) at the next time should be equal to the threshold t (k) at the present time plus a gradient Δ t (k) proportional to a negative mean square error function, i.e.:
tj(k+1)＝tj(k)-μjΔtj(k)，j＝1,2,...,J (5)
in the formula tjIs a wavelet coefficient threshold value u at the scale jjIs an adaptive iteration step size, Δ t, with a scale of jjIs based on the wavelet coefficient of j time and the scale thereofThe gradient of the estimated minimum mean square error, i.e.:
wherein,
is the minimum mean square error of the wavelet coefficients and their estimates at the scale j.
Thus, the best estimation value of the wavelet coefficient of each scale can be obtained, and the specific steps are as follows:
1) discrete Wavelet Transform (DWT) is applied to the observed signal X to obtain wavelet coefficients cd of each scalejk(j＝1,2,…,J)；
2) Processing the wavelet coefficient of each scale by using the threshold function provided by the invention to obtain the estimation value of the wavelet coefficient of each scale, and then obtaining delta t according to the formula (6)j(k) Then t isj(k+1)＝tj(k)-ujΔtj(k)。
3) Inverse Discrete Wavelet Transform (IDWT) is performed on the estimated values of the wavelet coefficients of each scale to obtain estimated values of the signal
Output of
4) Sampling the next observation signal X, and repeating the steps 1) -4).
Specifically, the framing process is as follows: framing currently employs a method of segmentation at the weakest signal segment. Data analysis has found that when the signal is not complete, such as with a truncated step signal, the calculation of the feature quantity is affected. In order to reduce this situation and improve the accuracy of feature quantity calculation, dynamic framing is adopted. The position of the weakest part of the tail part of the signal is automatically calculated and is divided into frames, so that the complete signal can be kept to the maximum extent.
Wherein the characteristic extraction process comprises the following steps: and extracting static characteristics and dynamic characteristics of different typical scene vibration signals. The vibration signal data is collected in urban traffic roads, rivers, fields, parks, forests and other environments and comprises vibration signals of normal walking, running, wheel rolling and the like of people in various high-low noise, rain and stone falling scenes, time domain feature extraction is carried out on the preprocessed vibration training data, and related peak value and kurtosis information are extracted. The peak-to-peak value is the maximum value-the minimum value of the signal at the same time point, the waveform of the signal can be drawn by the peak-to-peak value, and the effect of the peak-to-peak value is obvious, so that the influence caused by environmental noise can be reduced.
Wherein the peak-to-peak value is calculated as follows: xp-p＝max{x(n)}-min{x(n)}
In the formula, min { } represents a minimum function. The peak value and the peak value of the signal give out the range of the amplitude change of the signal, and can be used as the basis for testing the measuring range and the dynamic range of the acquisition device. The signal acquisition frequency is 1000HZ, but taking into account the amount of computation, and making the signal smoother, the addition of adjacent points in the signal averages to 500 HZ.
The wavelet packet analysis noise reduction of the first stage is carried out to obtain the estimation value of the observation signal
In the feature extraction stage, firstly, signals are framed, and n is adopted in consideration of the performance of a chip and the real-time property required by detection1s (actual length 2048) time window is used for the noise-reduced signal
Framing is performed and to further improve the signal-to-noise ratio at each n1N is taken before s frame2s is taken as a reference segment, and n is taken as1s as the detection segment. In the specific calculation of (n)1+n2) s (actual length 8047) signal, one (n)1+n2) The s signal is apparently not corresponding to a peak-to-peak value, and is then again for this (n)1+n2) The s signal is framed, x1 points are taken as a time window, x2 points are taken as sliding steps, and 320-dimensional peak-to-peak characteristic F _ ppv [ p ] is obtained1,p2,..,p320]. And extracting kurtosis characteristics after the peak-to-peak characteristics are taken.
The kurtosis is also called as kurtosis, whether the vibration signal obeys Gaussian distribution or not can be detected by the kurtosis in time domain analysis, the kurtosis is sensitive to impact components in the vibration signal, and the steepness of a discrete vibration signal probability quality function can be reflected.
Calculating the kurtosis:
in the formula, E { } represents the mathematical expectation, E { }2 represents the square of the mathematical expectation, μxIs the average value of a certain segment of signal x, which has n signal points, x (i) is the signal point at a certain time, KxI.e., kurtosis.
When Kx is 3, x (n) is a gaussian signal; when Kx >3, x (n) is an ultra-Gaussian signal, the probability mass function of the signal is more concentrated near the mean value of the signal, and the shape of the signal is steeper than that of the Gaussian distribution; when Kx <3, it is a sub-gaussian signal whose probability mass function shape is flatter than a gaussian distribution. The super-gaussian signal and the sub-gaussian signal are collectively referred to as a non-gaussian signal. Kx is a dimensionless parameter and is sensitive to impact components in the signal, and the stronger the impact, the larger the impact is.
In the same way, pair (n)1+n2) s signal, with a window size of x3, framed by a sliding step of x4, yields a kurtosis feature F _ kurtosis [ k ] of 78 dimensions1,k2,...,k78]；
The two feature vectors are spliced to obtain a 398-dimensional combined feature F _ mix [ F _ ppv; f _ kurtosis ];
then move n backward3s proceeds to the next stage (n)1+n2) s signal feature extraction, repeating the above featuresAnd (4) performing a feature extraction process.
Model training: and establishing a model of training data and establishing a template library of the vibration signal. And combining the extracted characteristic parameters by adopting a typical model xgboost of the gradient lifting tree and a Gaussian Mixture Model (GMM), and comprehensively judging the results of the two by using a decision algorithm.
Wherein Xgboost is one of boosting algorithms:
given n samples, the dataset of m features D { (x)i,yi)}(|D|＝n,xi∈Rm,yiE.g. R), D belongs to from RmAnd mapping to R, and adopting the result of K times of iteration as an output result by the lifting tree model. For xiOutput of (2)
The expression is as follows:
where Φ is an abstract function mapping relationship, F ═ F (x) wq(x)}(q:Rm->T,w∈RT) Representing a set of the lifting tree structure space, fkFor one mapping in F, the meaning of each variable is as follows:
q-table tree structure, which can map samples to corresponding leaf nodes;
w is the leaf node weight;
t represents the number of leaf nodes of the lifting tree;
each fkAll correspond to independent tree structures;
the objective function of Xgboost is as follows:
the above formula is divided into two parts, the first part
As a loss function, the second part
Is a regular term; y isiIn order to be a real label, the label,
to predict value, fkExpressing the structure of the kth tree, wherein omega is a regular term and the expression is as follows:
wherein gamma and lambda are constant coefficients, the importance degrees of the front part and the rear part in the control formula are the importance degrees, and w is the leaf node weight of each tree.
In xgboost, the regularization term includes two parts: one part is to increase the number T of the leaf nodes of the tree, which is used for controlling the complexity of the tree and achieving the effect of pruning; the other part is the sum of the squares of the weights w of the leaf nodes of each tree. The regularization term may allow the learner to avoid over-fitting as much as possible. With the above objective function, each sub-tree of xgboost tends to learn a simpler tree. In addition, when the regular term parameter is 0, the target expression is degraded into a traditional gradient lifting tree model.
Then, 398-dimensional F _ mix characteristics are obtained in the previous characteristic extraction, and are put into an xgboost model for training to obtain a trained xgboost model. Of course, normalization is performed before the data is put into the model.
GMM can use this distribution to characterize the target, since both target audio and vibration signal features form a specific distribution in the feature space. The Gaussian mixture model is used for approximating and simulating target feature distribution by using linear combination of a plurality of Gaussian distributions, and the audio which is most likely to generate the test signal feature and the target corresponding to the vibration distribution model are used as recognition results.
The gaussian mixture model is essentially a multi-dimensional probability density function, which can be used to represent the probability density function of the vibration signal feature vector. Clustering the feature vectors, regarding each class as a multi-dimensional Gaussian distribution function, then solving the mean value, covariance matrix and mixed weight of each class, and taking the mean value, covariance matrix and mixed weight as the training template of the target. And finally substituting the observation sequence into the template of each target to find the maximum posterior probability, namely the corresponding identified target.
The probability density function of an M-order Gaussian mixture model is obtained by weighted summation of M Gaussian probability density functions, as shown in the following formula:
where M is the order of the mixture model, X is a D-dimensional random vector, wi(i ═ 1, 2.., M) is a mixing weight, and the following condition is satisfied:
bi(X) (i ═ 1, 2.. times, M) are sub-distributions, each sub-distribution being a joint gaussian probability distribution in D dimension, which can be expressed as:
wherein muiIs a mean vector, ΣiIs a covariance matrix. The complete Gaussian mixture model is described by three parameters of a parameter mean vector, a covariance matrix and a mixing weight. A model λ can thus be represented as a triplet as follows:
λ＝{wi,μi,∑i},i＝1,2,...,M
covariance matrix sigma in formulaiThe normal matrix or the diagonal matrix can be taken. Since the diagonal matrix is simple to calculate and has good performance, the diagonal matrix is taken as follows:
formula (middle)
Is the variance of the k-dimension component of the feature vector corresponding to the i-th component of the GMM.
For a vibration signal of length T, time sequence X ═ X (X)1,x2,...,xT) Its GMM probability can be written as:
or expressed logarithmically:
the class with the maximum likelihood probability obtained in the N classes according to the Bayesian theorem is the recognition result:
then, 398-dimensional F _ mix characteristics are obtained in the characteristic extraction, and are put into an xgboost model for training to obtain a trained GMM model.
And (3) decision algorithm: and (4) extracting test data characteristics, and predicting by using the two models trained in the step (3), wherein two prediction results generated by the prediction need to be comprehensively judged, so that the decision algorithm is provided.
Output of Xgboost
The output of the GMM is i, the output values of the two models may or may not be consistent, and the following scoring logic is used for decision:
A. output prediction of xgboost
B. Outputting a prediction i of the GMM;
C. at the time of first decision, the output of one model is randomly selected as the total output from the two models, and the real label y is comparediIf xgboost is correct and GMM is incorrect, xgboost +1 score and the next predicted total output is determined by xgboost and vice versa; if both models predict correctly or incorrectly, no score is added, and the total output of the next round of prediction is determined by the party who scores more at present.
The passive seismic wave detection method is adopted, based on advanced technologies such as pattern recognition, edge calculation, deep learning and ultra-low power consumption design, the passive seismic wave detection method is buried underground, targeted sample training and recognition can be carried out through a deep learning algorithm, and various signals such as single person, multiple persons, small animals, wheeled vehicles and tracked vehicles can be effectively distinguished, so that the target types are distinguished; external interference such as sand blown by the wind, rockfall, sleet can be automatically filtered, different types of targets such as people, vehicles and animals can be selectively identified and filtered according to different corresponding scenes, filtering interference is achieved, and false alarm is reduced.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention.
Claims (10)
1. An intelligent unattended sensor system is characterized by comprising a seismic wave detection unit, a signal processing unit, a data processing unit, a communication unit and a power supply conversion unit; the seismic wave detection unit, the signal processing unit, the data processing unit and the communication unit are sequentially connected; the power supply conversion unit is connected with each unit and is used for providing power supply for each unit; the earthquake detection unit is used for converting a vibration signal generated by target intrusion into an electric signal, the signal processing unit is used for converting the electric signal into a digital signal, the data processing unit is used for carrying out analysis processing on the digital signal to obtain a target identification result, and the communication unit is used for outputting the target identification result to external equipment.
2. The intelligent unattended sensor system according to claim 1, wherein the communication unit comprises a wired communication module or a wireless communication module, and the wired communication module comprises a CAN communication module or/and an RS485 communication module; the wireless communication module comprises one or more of a Lora communication module, a Zigbee communication module, a 4G communication module or a 5G communication module.
3. The intelligent unattended sensor system according to claim 2, wherein the seismic demodulation unit comprises a plurality of vibration sensors, the plurality of vibration sensors form a sensor ad hoc communication network system, and cooperative sensing and real-time monitoring of the plurality of vibration sensors in a deployment area are achieved; and a Zigbee ad hoc network is adopted among the vibration sensors to complete interconnection among the vibration sensors.
4. The intelligent unattended sensor system according to claim 3, wherein the seismic demodulation unit is a moving coil magnetoelectric shock sensor.
5. An object identification method based on the intelligent unattended sensor system according to any one of claims 1-4, characterized by comprising the steps of:
s1, the seismic wave detection unit is used for converting the vibration signal generated by the target intrusion into an electric signal;
s2, the signal processing unit is used for converting the electric signal into a digital signal;
s3, the data processing unit is used for analyzing and processing the digital signal to obtain a target recognition result;
and S4, the communication unit is used for outputting the target recognition result to an external device.
6. The method for identifying an object according to claim 5, wherein in step S3, the specific process of analyzing and processing the digital signal to obtain the object identification result is as follows:
s3.1, noise reduction and framing are carried out on the digital signals;
s3.2, simultaneously extracting a plurality of time domain features of the framed data;
s3.3, taking the plurality of time domain characteristics obtained in the step S3.2 as training samples, and respectively training the Gaussian mixture model and the xgboost model to obtain the trained Gaussian mixture model and the trained xgboost model;
s3.4, respectively identifying the signals through the trained Gaussian mixture model and the trained xgboost model, and outputting corresponding identification results;
and S3.5, synthesizing the recognition results output by the Gaussian mixture model and the xgboost model by adopting a decision algorithm, and outputting a final recognition result.
7. The method of claim 6, wherein in step S3.2, four time domain features are extracted by using a method of extracting signal features, specifically: extracting the peak-to-peak value of the framed data at a fixed time interval to serve as a first time domain feature; extracting the kurtosis of the framed data at a fixed time interval to serve as a second time domain feature; calculating the sum of the maximum n point values of each frame of dataiAs a third time domain feature; will su miSum of n point values which is maximum from the previous framei-1Is taken as the fourth time-domain feature.
8. The method of claim 7, wherein the four time domain features are extracted by the following steps: with x1Points are a time window, x2The point is the sliding step length, so that the h-dimension peak-to-peak characteristic F _ ppv is obtainedi＝[p1,p2,..,ph]；
To (n)1+n2) Window signal of s in x3Points are a time window, x4The point is the sliding step length, and the kurtosis characteristic F _ kurtosis of w dimension is obtained by framingi＝[k1,k2,...,kw]Sum of the maximum n point values per frame data sumiAnd its maximum n point values and sum with the previous framei-1Is characteristic of the ratio ofi＝Sumi/Sumi-1。
9. The method for identifying multi-class objects for an unattended sensor system according to any one of claims 5 to 8, wherein in step S3.4, for each class prediction probability given by the GMM model and the xgboost model, a threshold shift method is adopted, and the identification result is considered to be credible if the probability is greater than a preset value.
10. The method for identifying multi-class objects of an unattended sensor system according to any one of claims 5-8, wherein the specific process of step S3.5 is as follows:
adopting scoring logics of a Gaussian mixture model and an xgboost model game: at the time of first decision, the output of one model is randomly selected from the two models as the total output, and the real label y is comparediIf the xgboost model is correct and the GMM is incorrect, then the xgboost model adds 1 point, and the next predicted total output is determined by the xgboost model output, or vice versa; if both models predict correctly or incorrectly, no point is added, and the total output of the next round of prediction is determined by the party with more scores at present.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210191616.6A CN114584924B (en) | 2022-02-28 | 2022-02-28 | Intelligent unattended sensor system and target identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210191616.6A CN114584924B (en) | 2022-02-28 | 2022-02-28 | Intelligent unattended sensor system and target identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114584924A true CN114584924A (en) | 2022-06-03 |
| CN114584924B CN114584924B (en) | 2024-05-31 |
Family
ID=81775591
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210191616.6A Active CN114584924B (en) | 2022-02-28 | 2022-02-28 | Intelligent unattended sensor system and target identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114584924B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115619777A (en) * | 2022-12-05 | 2023-01-17 | 中国民航大学 | Method and system for detecting ice and snow state of road surface and computer readable storage medium |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102799899A (en) * | 2012-06-29 | 2012-11-28 | 北京理工大学 | Special audio event layered and generalized identification method based on SVM (Support Vector Machine) and GMM (Gaussian Mixture Model) |
| CN103544791A (en) * | 2012-07-10 | 2014-01-29 | 中国矿业大学(北京) | Underground invasion monitoring system on basis of seismic waves |
| US20160178739A1 (en) * | 2014-12-19 | 2016-06-23 | Hyundai Mobis Co., Ltd. | Radar system for vehicle and operating method thereof |
| US20190095781A1 (en) * | 2017-09-23 | 2019-03-28 | Nanoprecise Sci Corp. | System and method for automated fault diagnosis and prognosis for rotating equipment |
| CN109699012A (en) * | 2019-01-18 | 2019-04-30 | 深圳斐锐科技有限公司 | Unattended ground transaucer system and its communication means and self-organized network communication module |
| CN112102955A (en) * | 2020-09-07 | 2020-12-18 | 武汉科瓴智能科技有限公司 | Patient disease prediction control system and method based on Gaussian mixture model |
| US20210182603A1 (en) * | 2019-12-05 | 2021-06-17 | Indian Institute Of Technology Delhi | Person identification and imposter detection using footfall generated seismic signals |
| WO2021128510A1 (en) * | 2019-12-27 | 2021-07-01 | 江苏科技大学 | Bearing defect identification method based on sdae and improved gwo-svm |
| CN113283331A (en) * | 2021-05-20 | 2021-08-20 | 长沙融创智胜电子科技有限公司 | Multi-class target identification method and system for unattended sensor system |
| CN113781728A (en) * | 2021-08-02 | 2021-12-10 | 盐城市湛安智感科技有限公司 | Vibration sensing system and method based on group intelligent optimization |
| US20220252714A1 (en) * | 2019-12-05 | 2022-08-11 | Mitsubishi Electric Corporation | Radar signal processing device, radar sensor system, and signal processing method |
-
2022
- 2022-02-28 CN CN202210191616.6A patent/CN114584924B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102799899A (en) * | 2012-06-29 | 2012-11-28 | 北京理工大学 | Special audio event layered and generalized identification method based on SVM (Support Vector Machine) and GMM (Gaussian Mixture Model) |
| CN103544791A (en) * | 2012-07-10 | 2014-01-29 | 中国矿业大学(北京) | Underground invasion monitoring system on basis of seismic waves |
| US20160178739A1 (en) * | 2014-12-19 | 2016-06-23 | Hyundai Mobis Co., Ltd. | Radar system for vehicle and operating method thereof |
| US20190095781A1 (en) * | 2017-09-23 | 2019-03-28 | Nanoprecise Sci Corp. | System and method for automated fault diagnosis and prognosis for rotating equipment |
| CN109699012A (en) * | 2019-01-18 | 2019-04-30 | 深圳斐锐科技有限公司 | Unattended ground transaucer system and its communication means and self-organized network communication module |
| US20210182603A1 (en) * | 2019-12-05 | 2021-06-17 | Indian Institute Of Technology Delhi | Person identification and imposter detection using footfall generated seismic signals |
| US20220252714A1 (en) * | 2019-12-05 | 2022-08-11 | Mitsubishi Electric Corporation | Radar signal processing device, radar sensor system, and signal processing method |
| WO2021128510A1 (en) * | 2019-12-27 | 2021-07-01 | 江苏科技大学 | Bearing defect identification method based on sdae and improved gwo-svm |
| CN112102955A (en) * | 2020-09-07 | 2020-12-18 | 武汉科瓴智能科技有限公司 | Patient disease prediction control system and method based on Gaussian mixture model |
| CN113283331A (en) * | 2021-05-20 | 2021-08-20 | 长沙融创智胜电子科技有限公司 | Multi-class target identification method and system for unattended sensor system |
| CN113781728A (en) * | 2021-08-02 | 2021-12-10 | 盐城市湛安智感科技有限公司 | Vibration sensing system and method based on group intelligent optimization |
Non-Patent Citations (3)
| Title |
|---|
| HONGYU CHEN 等: "Fault Diagnosis of Uninterruptible Power System Based on Gaussian Mixed Model and XGBoost", 2020 15TH INTERNATIONAL CONFERENCE ON COMPUTER SCIENCE & EDUCATION(ICCSE), 22 September 2020 (2020-09-22) * |
| LINGLING NI 等: "Streamflow forecasting using extreme gradient boosting model coupled with Gaussian mixture model", JOURNAL OF HYDROLOGY 586 (2020) 124901, 30 March 2020 (2020-03-30) * |
| 刘夏;陈磊;刘亢;李苑辉;陈明锐;: "基于XGBoost和GM模型的三亚市旅游客流量预测", 科技与经济, no. 06 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115619777A (en) * | 2022-12-05 | 2023-01-17 | 中国民航大学 | Method and system for detecting ice and snow state of road surface and computer readable storage medium |
| CN115619777B (en) * | 2022-12-05 | 2023-03-17 | 中国民航大学 | Method and system for detecting ice and snow state of road surface and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114584924B (en) | 2024-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108226892B (en) | Deep learning-based radar signal recovery method in complex noise environment | |
| US9619999B2 (en) | Sensor event assessor input/output controller | |
| CN114419825B (en) | High-speed rail perimeter intrusion monitoring device and method based on millimeter wave radar and camera | |
| CN101556724B (en) | Safety management system of optical fiber perimeter and pattern recognition method thereof | |
| CN103176164B (en) | Based on the Multi-Target Passive Tracking method of wireless sensor network | |
| CN104266894B (en) | A kind of mine microquake signal preliminary wave moment extracting method based on correlation analysis | |
| CN103810717B (en) | A kind of human body behavioral value method and device | |
| McErlean et al. | Distributed detection and tracking in sensor networks | |
| US10165228B2 (en) | Sensor event assessor training and integration | |
| CN115761537B (en) | Power transmission line foreign matter intrusion identification method oriented to dynamic feature supplementing mechanism | |
| CN112261719B (en) | Area positioning method combining SLAM technology with deep learning | |
| CN112541480B (en) | Online identification method and system for tunnel foreign matter invasion event | |
| CN114584924B (en) | Intelligent unattended sensor system and target identification method | |
| US8405524B2 (en) | Seismic method for vehicle detection and vehicle weight classification | |
| CN115060184B (en) | Optical fiber perimeter intrusion detection method and system based on recursion diagram | |
| CN116453278A (en) | Intrusion monitoring method combining deep learning intelligent detection and optical fiber vibration sensing | |
| CN102938876B (en) | Passive target tracking method based on received signal strength of wireless sensor network | |
| Wu et al. | Improved generalization in signal identification with unsupervised spiking neuron networks for fiber-optic distributed acoustic sensor | |
| CN117452496A (en) | Target detection and identification method for seismoacoustic sensor | |
| Lacombe et al. | Seismic detection algorithm and sensor deployment recommendations for perimeter security | |
| Xu et al. | Vehicle classification under different feature sets with a single anisotropic magnetoresistive sensor | |
| CN108333575B (en) | Gaussian prior and interval constraint based time delay filtering method for mobile sound source | |
| CN112698665A (en) | Unmanned aerial vehicle detection positioning method | |
| Zetterqvist et al. | Elephant DOA Estimation using a Geophone Network | |
| Siewert et al. | Acoustic, Seismic, and Visual Camera Sensor Fusion Experiments for Large Animal Detection and Tracking with Scalability |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Xu Yan Inventor after: Dong Zhi Inventor after: Zhou Chunlei Inventor before: Dong Zhi Inventor before: Xu Yan Inventor before: Zhou Chunlei |