CN115471966A - Self-learning intrusion alarm method, system, medium and equipment based on vibration optical fiber detection - Google Patents
Self-learning intrusion alarm method, system, medium and equipment based on vibration optical fiber detection Download PDFInfo
- Publication number
- CN115471966A CN115471966A CN202210921545.0A CN202210921545A CN115471966A CN 115471966 A CN115471966 A CN 115471966A CN 202210921545 A CN202210921545 A CN 202210921545A CN 115471966 A CN115471966 A CN 115471966A
- Authority
- CN
- China
- Prior art keywords
- optical fiber
- time
- alarm
- vibration
- host
- 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.)
- Pending
Links
- 239000013307 optical fiber Substances 0.000 title claims abstract description 82
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000001514 detection method Methods 0.000 title claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 75
- 238000012216 screening Methods 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 16
- 238000013507 mapping Methods 0.000 claims description 13
- 238000004458 analytical method Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 6
- 230000003321 amplification Effects 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 5
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 5
- 238000001228 spectrum Methods 0.000 claims description 5
- 230000001419 dependent effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 239000000835 fiber Substances 0.000 claims 4
- 230000006870 function Effects 0.000 description 19
- 238000004422 calculation algorithm Methods 0.000 description 7
- 230000004913 activation Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000003909 pattern recognition Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000009194 climbing Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012567 pattern recognition method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B13/00—Burglar, theft or intruder alarms
- G08B13/02—Mechanical actuation
- G08B13/12—Mechanical actuation by the breaking or disturbance of stretched cords or wires
- G08B13/122—Mechanical actuation by the breaking or disturbance of stretched cords or wires for a perimeter fence
- G08B13/124—Mechanical actuation by the breaking or disturbance of stretched cords or wires for a perimeter fence with the breaking or disturbance being optically detected, e.g. optical fibers in the perimeter fence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B13/00—Burglar, theft or intruder alarms
- G08B13/18—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength
- G08B13/189—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems
- G08B13/194—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems
- G08B13/196—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems using television cameras
Abstract
The invention provides a self-learning intrusion alarm method, a system, a medium and equipment based on vibration optical fiber detection, which comprise the following steps: step 1: obtaining alarm information based on the vibration optical fiber detection module; step 2: acquiring picture information of a camera corresponding to the alarm position point and automatically judging whether the coverage area has sign information of a person or not by the camera; and step 3: carrying out positive/negative sample folder classification on the received time-frequency training images; and 4, step 4: carrying out sample equilibrium screening on positive/negative samples; and 5: judging whether the model training program is running or not, and if not, then according to the updated positive/negative sample; and 6: and forwarding the generated model to the corresponding DVS vibration optical fiber host, setting a moment to update the model, and restarting the corresponding DVS vibration optical fiber host. The invention automatically sends the trained model to the corresponding DVS host according to the corresponding host IP, thereby greatly reducing the whole engineering operation cost.
Description
Technical Field
The invention relates to the technical field of optical fiber vibration sensing and the technical field of computer vision fusion, in particular to a self-learning intrusion alarm method, a system, a medium and equipment based on vibration optical fiber detection.
Background
Machine intelligence, conventional machine learning or deep learning, often learns a mapping relationship between inputs and outputs. An algorithm model based on the optical fiber vibration sensing signal pattern recognition system needs to acquire a large amount of data and carry out manual labeling in the processes of initial construction and subsequent iteration. In the research field, a few methods are proposed to alleviate the burdensome data annotation task. However, in engineering practice, especially in large-scale artificial intelligence practical application (large number of algorithm models), these methods are either difficult to land or rely on complicated and fragmented manual labor, lack of a platform and effective tool support of a system, and have great improvement space in efficiency and effect.
Patent document CN111951505B (application number: CN 202010866358.8) discloses a fence vibration intrusion positioning and pattern recognition method based on a distributed optical fiber system, which is implemented by constructing a pattern recognition classifier, training and testing the pattern recognition classifier by using a data training set, positioning an intrusion point by using the position of a sensor collecting a vibration signal when the distributed optical fiber system collects the vibration signal, and recognizing a specific intrusion pattern of the vibration signal by using the pattern recognition classifier.
In order to solve the problems of fast construction and continuous iterative optimization from 0 to 1 of a large number of algorithm models in engineering practice, an artificial intelligence algorithm iteration platform is constructed. The platform aims to open a self-learning closed loop of a data-sample-model-optical fiber sensing system, and provides powerful support for construction of a perimeter intrusion linear alarm recognition system in links of data analysis, sample marking, model training, model performance evaluation, deployment and the like so as to accelerate a product iteration cycle. Particularly, in practical engineering application, due to the limitation of scenes and time, a large number of negative samples are lacked at the initial stage of model training, the samples cannot be acquired through manual supervision type acquisition and cannot be manufactured in advance manually, and the negative samples can be acquired and perfected only by long-time running of equipment in corresponding scenes.
The AI self-learning optical fiber intelligent platform system can effectively discriminate, distinguish and reprocess the transmitted data information and then continuously nourish the machine, thereby achieving the closed loop function of autonomous, controllable and cyclic learning, which is the intellectualization of the optical fiber sensing intelligent identification system.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a self-learning intrusion alarm method, a system, a medium and equipment based on vibration optical fiber detection.
The self-learning intrusion alarm method based on vibration optical fiber detection provided by the invention comprises the following steps:
step 1: obtaining alarm information based on the vibration optical fiber detection module, wherein the alarm information comprises alarm time, an IP address of a corresponding vibration optical fiber host, an optical fiber alarm position point and a time-frequency image for training;
and 2, step: inquiring a preset camera number-vibration optical fiber alarm range mapping table according to the optical fiber alarm position point, acquiring picture information of the camera corresponding to the alarm position point and automatically judging whether the coverage area has the mark information of a person by the camera;
and 3, step 3: carrying out positive/negative sample folder classification on the received time-frequency training image according to the sign information of whether a person exists in the alarm information;
and 4, step 4: carrying out sample equilibrium screening on positive/negative samples;
and 5: judging whether the model training program is running or not, if not, calling python codes through C + + according to the updated positive/negative samples to finish automatic training of the EfficientNet network model and generate a C + + callable form;
step 6: and sending the C + + callable model and the corresponding IP information to a vibration optical fiber detection module, forwarding the generated model to a corresponding DVS vibration optical fiber host according to the IP address, setting a time to update the model, and restarting the corresponding DVS vibration optical fiber host, thereby completing model updating.
Preferably, all the vibration optical fiber hosts are connected to the same switch and connected to the server host, the DVS vibration optical fiber software operates the alarm between the host and the server host to transmit information in JSON format, and the content of the information includes { Time: the time at which the alarm occurred; IP: indicating an IP address of the corresponding vibration host; pos: the meter mark is used for mapping to obtain a corresponding camera and acquiring specific information of the corresponding camera; picture: the time-frequency graph training sample data generated by the corresponding vibration host is stored in the short-time Fourier transform system, wherein the obtained two-dimensional time-frequency graph is obtained by performing short-time Fourier transform on a signal value which is subjected to 50Hz low-frequency high-pass filtering at an alarm position point by taking 0.3 second as a unit by the corresponding alarm host, a one-dimensional signal is converted into two-dimensional time-frequency graph data, and the calculation process of the short-time Fourier transform is shown as a formula 1:
wherein x (t) represents a signal at a time point t; h (τ -t) represents an analysis window function centered at the time point t; x (τ) h (τ -t) represents a slice of the signal around the analysis time point t; for a given time t, STFT (t, f) is the spectrum at that time instant; τ represents the integral dependent variable time; j represents the complex imaginary identifier; f denotes the transformed frequency.
Preferably, an EfficientNet network structure is built, a mixed dimension amplification method is utilized, the network width, the network depth and the image resolution ratio are balanced through a mixed coefficient phi to obtain higher precision, and the specific calculation process is shown as a formula 2:
wherein d represents the network depth; w represents the network width; r represents the image resolution; α, β, and γ are constants that are artificially set to affect the depth of the network, the width of the network, and the resolution of the image, respectively.
Preferably, the step 4 comprises: and storing the number of the training samples stored last time, copying all samples in a small number of sample folders into a positive/negative sample folder when the number of the two types of samples is increased by the same sample amount compared with the number of the training samples in the previous time, copying the newly increased samples for the other type of samples, and then randomly copying the rest samples into the negative/positive sample folder until the number of the samples is the same as that of the positive/negative sample folder.
The self-learning intrusion alarm system based on vibration optical fiber detection provided by the invention comprises:
a module M1: obtaining alarm information based on a vibration optical fiber detection module, wherein the alarm information comprises alarm time, an IP address of a corresponding vibration optical fiber host, an optical fiber alarm position point and a time-frequency image used for training;
a module M2: inquiring a preset camera number-vibration optical fiber alarm range mapping table according to the optical fiber alarm position point, and acquiring picture information of the camera corresponding to the alarm position point and mark information of whether a person exists in the coverage area range or not by the camera;
a module M3: carrying out positive/negative sample folder classification on the received time-frequency training image according to the sign information of whether a person exists in the alarm information;
a module M4: carrying out sample equilibrium screening on positive/negative samples;
a module M5: judging whether the model training program is running or not, if not, calling python codes through C + + according to the updated positive/negative samples to finish automatic training of the EfficientNet network model and generate a C + + callable form;
a module M6: and sending the C + + callable model and the corresponding IP information to a vibration optical fiber detection module, forwarding the generated model to a corresponding DVS vibration optical fiber host according to the IP address, setting a time to update the model, and restarting the corresponding DVS vibration optical fiber host, thereby completing model updating.
Preferably, all the vibrating optical fiber hosts are connected to the same switch and connected with the server host, the alarm between the DVS vibrating optical fiber software operating host and the server host carries out information transmission through a JSON format, and the content of the information comprises (Time: the time at which the alarm occurred; IP: indicating an IP address of the corresponding vibration host; pos: the meter mark is used for mapping to obtain a corresponding camera and acquiring specific information of the corresponding camera; picture: the time-frequency graph training sample data generated by the corresponding vibration host is stored in the time-frequency graph training sample data, wherein the acquired two-dimensional time-frequency graph is obtained by performing short-time Fourier transform on a signal value which is subjected to 50Hz low-frequency high-pass filtering processing at an alarm position point of the corresponding alarm host in a unit of 0.3 second, and converting a one-dimensional signal into two-dimensional time-frequency graph data, and the calculation process of the short-time Fourier transform is shown as a formula 1:
wherein x (t) represents a signal at a time point t; h (τ -t) represents an analysis window function centered at the time point t; x (τ) h (τ -t) represents a slice of the signal around the analysis time point t; for a given time t, STFT (t, f) is the spectrum at that time instant; τ represents the integral dependent variable time; j represents the complex imaginary identifier; f denotes the transformed frequency.
Preferably, the EfficientNet network structure is built by utilizing a mixed dimension amplification method, the network width, the network depth and the image resolution ratio are balanced through a mixing coefficient phi to obtain higher precision, and the specific calculation process is shown as a formula 2:
wherein d represents the network depth; w represents the network width; r represents the image resolution; α, β, and γ are constants that are artificially set to affect the network depth, network width, and image resolution, respectively.
Preferably, the module M4 comprises: and storing the number of the training samples stored last time, copying all samples in a small number of sample folders into a positive/negative sample folder when the number of the two types of samples is increased by the same sample amount compared with the number of the training samples in the previous time, copying the newly increased samples for the other type of samples, and then randomly copying the rest samples into the negative/positive sample folder until the number of the samples is the same as that of the positive/negative sample folder.
According to the present invention, a computer-readable storage medium is provided, in which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the method.
The self-learning intrusion alarm device based on vibration optical fiber detection provided by the invention comprises: a controller;
the controller comprises the computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the self-learning intrusion alert method based on vibrating optical fiber detection; or the controller comprises the self-learning intrusion alarm system based on vibration optical fiber detection.
Compared with the prior art, the invention has the following beneficial effects:
(1) In the perimeter intrusion linear alarm recognition system, under the influence of large difference formed by the application environment of a product, various intrusion data generated in different scenes have large difference, so that training models in cross scenes cannot be used in a crossed manner, and meanwhile, due to the limitation of scenes and time, a large number of training samples are lacked in the initial stage of model training, and the samples cannot be acquired through manual supervision type acquisition and cannot be manufactured in advance manually;
(2) In the actual engineering application, the phenomenon of unbalanced training positive and negative samples easily occurs, the unbalanced samples often cause the performance of a machine learning algorithm to be worse, and finally the model obtained by training is biased to predict the type with more training data in the actual application, so that the prediction effect is poor, therefore, the invention adopts a sample balancing strategy of 'getting new and randomly compensating difference', not only ensures that the newly added samples are fully utilized, but also considers part of old samples to a certain extent, and the sample balancing screening strategy has better effect on the model obtained by final training in the actual scene application after verification;
(3) Because the classification and identification effects of the initialized model of the perimeter intrusion linear alarm identification system are not obvious enough, the system forms the self-learning functions of automatic model training and automatic operation by automatically supplementing reliable labeled data, thereby opening the self-learning closed loop of the data-sample-model-optical fiber sensing system, and enabling the whole system to continuously self nourish in the operation process, thereby achieving the closed loop function of autonomous, controllable and cyclic learning;
(4) The multiple vibration optical fiber hosts are simultaneously connected to the same switch and connected with the training server host, the model self-learning function of the detection hosts of the multiple vibration optical fibers is realized through a single training server host in a polling mode, and finally, the trained models are automatically sent to the corresponding DVS hosts according to the corresponding host IP, so that the whole engineering operation cost is greatly reduced.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a flow chart of classification model self-learning and overall system data;
FIG. 2 is a diagram of a system self-learning closed loop framework;
FIG. 3a is a diagram of a negative example training sample, and FIG. 3b is a diagram of a positive example training sample;
fig. 4a is a graph of accuracy of the EfficientNetB4 network training process, and fig. 4b is a graph of loss value of the network training process.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will aid those skilled in the art in further understanding the present invention, but are not intended to limit the invention in any manner. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.
The embodiment is as follows:
the detailed flow of the system is shown in fig. 1, and the system framework is shown in fig. 2.
The system mainly comprises two modules: the vibration optical fiber detection module and the training server module, and the whole self-learning system algorithm comprises the following steps:
step 1: the existing vibration optical fiber detection module generates alarm information, and the generated vibration alarm information comprises: alarming time, an IP address corresponding to the vibration optical fiber host, a corresponding optical fiber intrusion alarming position point and intrusion alarming picture data used for training;
step 2: inquiring a mapping table of camera number-vibration optical fiber alarm range according to the optical fiber alarm position point, acquiring picture information of a camera corresponding to the alarm position point and automatically judging whether the coverage area has the mark information of a person by the camera;
and 3, step 3: sending the acquired alarm information (including alarm time, alarm host IP, existence of a person mark and an alarm time-frequency graph) to a training server module;
and 4, step 4: carrying out folder classification on the received time-frequency training images according to the flag bit information of whether people exist in the alarm information or not;
and 5: the number of the training samples stored last time is saved, when the number of the two types of samples is increased by N samples compared with the number of the training samples in the previous time, all samples in a small number of sample folders are copied into a positive/negative sample folder, the newly increased samples are copied for the other type of samples, then the rest samples are randomly copied into the negative/positive sample folder until the number of the samples is the same as that of the samples in the positive/negative sample folder, so that the balance of the positive and negative samples is ensured, and the attention to the newly added samples is taken into consideration at the same time, wherein the positive and negative training samples are shown in fig. 3a and fig. 3 b.
And 6: judging whether the model training program is running or not, if not, calling python codes by C + + according to the updated positive and negative samples to finish automatic training of the EfficientNet network model and generate a C + + callable form;
and 7: and sending the C + + callable model and the corresponding IP information to the vibration optical fiber detection module, forwarding the generated model to the corresponding DVS vibration optical fiber host according to the IP address, wherein the optical fiber host can set a better opportunity to update the model, and restart the corresponding DVS vibration optical fiber host to complete model updating.
Example 2:
example 2 is a preferred example of example 1.
The invention provides a self-learning intrusion alarm method based on vibration optical fiber detection, which comprises the following steps:
step 1: for a DVS vibration optical fiber software operation host, the DVS vibration optical fiber software operation host is composed of N vibration optical fiber software operation hosts supervised by a current platform, wherein each vibration host corresponds to one IP, such as (127.0.0.1), (127.0.0.2) \8230; (127.0.0.n), all vibration hosts are connected to the same switch and connected with a server host, an alarm between the DVS vibration optical fiber software operation host and the server host performs information transmission through a JSON format, and the content of the information includes (Time: the time at which the alarm occurred; IP: indicating an IP address of the corresponding vibration host; pos: the meter mark is used for mapping to obtain a corresponding camera and acquiring specific information of the corresponding camera; picture: the time-frequency graph training sample data generated by the corresponding vibration host is stored in the time-frequency graph training sample data, wherein the acquired two-dimensional time-frequency graph is obtained by performing short-time Fourier transform (STFT) on a signal value which is subjected to 50Hz low-frequency high-pass filtering processing at an alarm position point by taking 0.3 second (1024 sampling points) as a unit through the corresponding alarm host, a one-dimensional signal is converted into two-dimensional time-frequency graph data, and the calculation process of the short-time Fourier transform is shown as a formula 1:
where x (t) denotes the signal at time t, h (τ -t) denotes the analysis window function centered at time t, and x (τ) h (τ -t) denotes a slice of the signal around analysis time t. For a given time t, STFT (t, f) can be considered as the spectrum at that instant.
Step 2: the system comprises a vibration optical fiber detection module, a plurality of cameras, a human body detection function, an alarm position point in an alarm protocol, a camera ID preset in advance and a vibration optical fiber alarm range mapping table, wherein each camera has a human body detection function, when alarm information generated by a DVS vibration optical fiber software operation host is received, the corresponding camera is found through the camera ID preset in advance and the vibration optical fiber alarm range mapping table, whether a pedestrian exists in the corresponding camera is confirmed, when the pedestrian exists in the current range is detected, the current training sample is judged to be a positive sample, otherwise, the current sample is judged to be a negative sample, a new alarm information is formed and is forwarded to a training server host, and the alarm protocol which is sent to the training server module by the vibration optical fiber detection module is specifically formed in a form of { Time: xx/xx/xx/xx, IP:127.0.0.x, isPerson; the added IsPerson mark represents whether a pedestrian exists in the current visible light camera, wherein '0' represents that the pedestrian does not exist, and '1' represents that the pedestrian exists.
And step 3: the method comprises the steps that information sent by a vibration optical fiber detection module is received, isPerson flag bit information is extracted, the fact that a sample stored in a Picture is a positive sample or a negative sample is determined and is respectively placed under two established folders (temporary positive sample training data and temporary negative sample training data), the two folders correspond to an upper-level directory and are folders named by IP addresses, the sample data on a DVS vibration optical fiber software operation host corresponding to the folder is shown, the sample training folder only represents a temporary file, and finally a trained Picture is stored in the positive and negative sample folders extracted through a balance strategy;
and 4, step 4: the sample balance screening strategy has great significance on the final model effect, the sample imbalance means that the number of samples corresponding to each class is different and the difference is large in the classification problem, the unbalanced samples often cause the performance of a machine learning algorithm to be poor, for example, in a positive sample and a negative sample, the ratio of a prediction pair to a prediction error is consistent, but the accuracy rate is different due to different sample ratios, and the accuracy rate and the recall rate of the positive sample and the negative sample cannot be truly reflected. The positive and negative sample balance strategy in the invention is as follows: the folder with a small number of samples in the temporary positive and negative sample folders is taken as a reference, and a strategy of 'getting new and randomly compensating the difference' is adopted for the other folder, namely, all samples which are more than those obtained in the previous training are extracted, and the rest difference samples are randomly sampled in the rest data, so that the newly added samples are fully utilized, the weight of the newly added samples is increased, and meanwhile, part of old samples are reserved, and the method has extremely important significance in model training by using the model which is trained in the early stage as a pre-training model.
And 5: and judging whether the model training program is running or not, if not, calling python codes by C + + according to the updated positive and negative samples to finish automatic training of the EfficientNet network model and generate a C + + callable form. The method for building the EfficientNet network structure utilizes a mixed dimension amplification method, the network width, the network depth and the image resolution ratio are balanced through a mixed coefficient phi to obtain higher precision, and the specific calculation process is shown in a formula 2.
Table 1 shows that the backbone network structure in the EfficientNet B4 network structure is constructed on the basis of the MBConv module in the mnset (modified network based on MobileNet), the (3 × 3 convolutional layer + batch normalization once + ReLU activation function) structure in the MBConv module is decomposed into the (1 × 1 convolutional layer + batch normalization layer + ReLU activation function) + (3 × 3 convolutional layer + batch normalization layer + ReLU activation function) structure, and because the Swish activation function has the characteristics of no upper bound, no lower bound, non-monotonous and smooth, the activating function in the backbone network structure is superior to the conventional ReLU activation function in accuracy, so in the EfficientNet B4 network structure used in the present invention, the ReLU function in the backbone network is replaced with the Swish function.
TABLE 1 EfficientNet B4 backbone network architecture
And (4) training an EfficientNet B4 classification network model. The model training environment used by the present invention is configured as follows:
operating the system: win10 bit
Processor (CPU): intel (R) Xeon (R) E5-2680 v4 CPU @2.40GHz fourteen-core twenty-eight thread
Memory (RAM): 16.0GB
Graphics card (GPU): NVIDIA GeForce RTX 3090
A programming environment: python + OpenCV + pyTorch
The model training set samples are divided into two types, respectively: positive sample data (including data generated by behaviors such as strong wind interference, carton interference, branch interference and cart passing interference) and positive sample data (including data generated by human intrusion behaviors such as climbing intrusion, knocking intrusion and drilling net intrusion). The method is characterized in that an EfficientNet B4 structure is adopted, a piecewise attenuation function is used for setting the learning rate, an Auto model search method is used for automatically adjusting and comparing the initial learning rate and other super parameters, the accuracy rate change and the training loss change in the training process are shown in a figure 4a and a figure 4B, and as the training times are increased, the model training loss value approaches to 0, the accuracy rate approaches to 1, and the network model shows a gradual convergence trend in the whole training process.
Step 6: the model trained for the training server host is expressed in { IP:127.0.0.X, model.
Those skilled in the art will appreciate that, in addition to implementing the systems, apparatus, and various modules thereof provided by the present invention in purely computer readable program code, the same procedures can be implemented entirely by logically programming method steps such that the systems, apparatus, and various modules thereof are provided in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system, the apparatus, and the modules thereof provided by the present invention may be considered as a hardware component, and the modules included in the system, the apparatus, and the modules for implementing various programs may also be considered as structures in the hardware component; modules for performing various functions may also be considered to be both software programs for performing the methods and structures within hardware components.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.
Claims (10)
1. A self-learning intrusion alarm method based on vibration optical fiber detection is characterized by comprising the following steps:
step 1: obtaining alarm information based on the vibration optical fiber detection module, wherein the alarm information comprises alarm time, an IP address of a corresponding vibration optical fiber host, an optical fiber alarm position point and a time-frequency image for training;
step 2: inquiring a preset camera number-vibration optical fiber alarm range mapping table according to the optical fiber alarm position point, and acquiring picture information of the camera corresponding to the alarm position point and mark information of whether a person exists in the coverage area range or not by the camera;
and 3, step 3: carrying out positive/negative sample folder classification on the received time-frequency training image according to the sign information of whether a person exists in the alarm information;
and 4, step 4: carrying out sample equilibrium screening on positive/negative samples;
and 5: judging whether the model training program is running or not, if not, calling python codes through C + + according to the updated positive/negative samples to finish automatic training of the EfficientNet network model and generate a C + + callable form;
step 6: and sending the C + + callable model and the corresponding IP information to a vibration optical fiber detection module, forwarding the generated model to a corresponding DVS vibration optical fiber host according to the IP address, setting a time to update the model, and restarting the corresponding DVS vibration optical fiber host, thereby completing model updating.
2. The self-learning intrusion alarm method based on vibration fiber detection according to claim 1, wherein all vibration fiber hosts are connected to the same switch and to the server host, the DVS vibration fiber software runs the alarm between the host and the server host to transmit information in JSON format, and the content of the information includes { Time: the time when the alarm occurred; IP: indicating an IP address of the corresponding vibration host; pos: the meter mark is used for mapping to obtain a corresponding camera and acquiring specific information of the corresponding camera; picture: the time-frequency graph training sample data generated by the corresponding vibration host is stored in the time-frequency graph training sample data, wherein the acquired two-dimensional time-frequency graph is obtained by performing short-time Fourier transform on a signal value which is subjected to 50Hz low-frequency high-pass filtering processing at an alarm position point of the corresponding alarm host in a unit of 0.3 second, and converting a one-dimensional signal into two-dimensional time-frequency graph data, and the calculation process of the short-time Fourier transform is shown as a formula 1:
wherein x (t) represents a signal at a time point t; h (τ -t) represents an analysis window function centered at the time point t; x (τ) h (τ -t) represents a slice of the signal around the analysis time point t; for a given time t, STFT (t, f) is the spectrum at that time instant; τ represents the integral dependent variable time; j represents the complex imaginary identifier; f denotes the transformed frequency.
3. The self-learning intrusion alarm method based on vibration optical fiber detection according to claim 1, wherein an EfficientNet network structure is built, a mixed dimension amplification method is utilized, the network width, the network depth and the image resolution ratio are balanced by a mixed coefficient phi, the scaling ratios of three dimensions are used for obtaining higher precision, and the specific calculation process is shown as a formula 2:
wherein d represents the network depth; w represents the network width; r represents the image resolution; α, β, and γ are constants that are artificially set to affect the depth of the network, the width of the network, and the resolution of the image, respectively.
4. The self-learning intrusion alarm method based on the vibrating optical fiber detection according to claim 1, wherein the step 4 comprises: and storing the number of the training samples stored last time, copying all samples in a small number of sample folders into positive/negative sample folders when the number of the two types of samples is increased by the same sample amount compared with the number of the training samples in the previous time, copying the newly increased samples for the other type of samples, and then randomly copying the rest samples into the negative/positive sample folders until the number of the samples in the same positive/negative sample folders is the same.
5. A self-learning intrusion alarm system based on vibration optical fiber detection is characterized by comprising:
a module M1: obtaining alarm information based on a vibration optical fiber detection module, wherein the alarm information comprises alarm time, an IP address of a corresponding vibration optical fiber host, an optical fiber alarm position point and a time-frequency image used for training;
a module M2: inquiring a preset camera number-vibration optical fiber alarm range mapping table according to the optical fiber alarm position point, and acquiring picture information of the camera corresponding to the alarm position point and mark information of whether a person exists in the coverage area range or not by the camera;
a module M3: carrying out positive/negative sample folder classification on the received time-frequency training image according to the sign information of whether a person exists in the alarm information;
a module M4: carrying out sample equilibrium screening on positive/negative samples;
a module M5: judging whether the model training program is running or not, if not, calling python codes through C + + according to the updated positive/negative sample to finish automatic training of the efficientNet network model and generate a C + + callable form;
a module M6: and sending the C + + callable model and the corresponding IP information to a vibration optical fiber detection module, forwarding the generated model to a corresponding DVS vibration optical fiber host according to the IP address, setting a time to update the model, and restarting the corresponding DVS vibration optical fiber host, thereby completing model updating.
6. The self-learning intrusion alarm system based on vibration fiber detection according to claim 5, wherein all vibration fiber hosts are connected to the same switch and to the server host, the DVS vibration fiber software runs the alarm between the host and the server host for information transmission in JSON format, and the content of the information includes { Time: the time when the alarm occurred; IP: indicating an IP address of the corresponding vibration host; pos: the meter mark is used for mapping to obtain a corresponding camera and acquiring specific information of the corresponding camera; picture: the time-frequency graph training sample data generated by the corresponding vibration host is stored in the time-frequency graph training sample data, wherein the acquired two-dimensional time-frequency graph is obtained by performing short-time Fourier transform on a signal value which is subjected to 50Hz low-frequency high-pass filtering processing at an alarm position point of the corresponding alarm host in a unit of 0.3 second, and converting a one-dimensional signal into two-dimensional time-frequency graph data, and the calculation process of the short-time Fourier transform is shown as a formula 1:
wherein x (t) represents a signal at a time point t; h (τ -t) represents an analysis window function centered at the time point t; x (τ) h (τ -t) represents a slice of the signal around the analysis time point t; for a given time t, STFT (t, f) is the spectrum at that time instant; τ represents the integral dependent variable time; j represents the complex imaginary identifier; f denotes the transformed frequency.
7. The self-learning intrusion alarm system based on vibration optical fiber detection according to claim 5, wherein an EfficientNet network structure is built, a mixed dimension amplification method is utilized, the network width, the network depth and the image resolution ratio are balanced by a mixed coefficient phi, the scaling ratios of three dimensions are used for obtaining higher precision, and the specific calculation process is shown as a formula 2:
wherein d represents the network depth; w represents the network width; r represents the image resolution; α, β, and γ are constants that are artificially set to affect the depth of the network, the width of the network, and the resolution of the image, respectively.
8. The self-learning intrusion alarm system based on vibrating optical fiber detection according to claim 5, wherein the module M4 comprises: and storing the number of the training samples stored last time, copying all samples in a small number of sample folders into a positive/negative sample folder when the number of the two types of samples is increased by the same sample amount compared with the number of the training samples in the previous time, copying the newly increased samples for the other type of samples, and then randomly copying the rest samples into the negative/positive sample folder until the number of the samples is the same as that of the positive/negative sample folder.
9. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 4.
10. A self-learning intrusion alarm device based on vibrating optical fiber detection, comprising: a controller;
the controller comprising a computer readable storage medium of claim 9 having a computer program stored thereon, which computer program when executed by a processor implements the steps of the self-learning intrusion alert method based on vibrating optical fiber detection of any one of claims 1 to 4; alternatively, the controller comprises a self-learning intrusion alert system based on vibrating optical fibre detection according to any one of claims 5 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210921545.0A CN115471966A (en) | 2022-08-02 | 2022-08-02 | Self-learning intrusion alarm method, system, medium and equipment based on vibration optical fiber detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210921545.0A CN115471966A (en) | 2022-08-02 | 2022-08-02 | Self-learning intrusion alarm method, system, medium and equipment based on vibration optical fiber detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115471966A true CN115471966A (en) | 2022-12-13 |
Family
ID=84365884
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210921545.0A Pending CN115471966A (en) | 2022-08-02 | 2022-08-02 | Self-learning intrusion alarm method, system, medium and equipment based on vibration optical fiber detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115471966A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543631A (en) * | 2018-11-28 | 2019-03-29 | 公安部沈阳消防研究所 | A kind of fire image detection alarm method based on machine learning |
| CN110555717A (en) * | 2019-07-29 | 2019-12-10 | 华南理工大学 | method for mining potential purchased goods and categories of users based on user behavior characteristics |
| CN111145475A (en) * | 2019-12-12 | 2020-05-12 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Intrusion alarm system, method and medium based on vibration optical fiber and deep learning |
| CN112950279A (en) * | 2021-03-19 | 2021-06-11 | 猎居网络科技(上海)有限公司 | Accurate marketing strategy model construction method and device based on machine learning |
| CN113628170A (en) * | 2021-07-16 | 2021-11-09 | 武汉大学 | Laser line extraction method and system based on deep learning |
| CN114742096A (en) * | 2022-03-22 | 2022-07-12 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Intrusion alarm method and system based on vibration optical fiber detection and complete action extraction |
-
2022
- 2022-08-02 CN CN202210921545.0A patent/CN115471966A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543631A (en) * | 2018-11-28 | 2019-03-29 | 公安部沈阳消防研究所 | A kind of fire image detection alarm method based on machine learning |
| CN110555717A (en) * | 2019-07-29 | 2019-12-10 | 华南理工大学 | method for mining potential purchased goods and categories of users based on user behavior characteristics |
| CN111145475A (en) * | 2019-12-12 | 2020-05-12 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Intrusion alarm system, method and medium based on vibration optical fiber and deep learning |
| CN112950279A (en) * | 2021-03-19 | 2021-06-11 | 猎居网络科技(上海)有限公司 | Accurate marketing strategy model construction method and device based on machine learning |
| CN113628170A (en) * | 2021-07-16 | 2021-11-09 | 武汉大学 | Laser line extraction method and system based on deep learning |
| CN114742096A (en) * | 2022-03-22 | 2022-07-12 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Intrusion alarm method and system based on vibration optical fiber detection and complete action extraction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6926335B2 (en) | Variable rotation object detection in deep learning | |
| CN106971152B (en) | Method for detecting bird nest in power transmission line based on aerial images | |
| JP7266668B2 (en) | Video object fast detection method, apparatus, server and storage medium | |
| CN101271515A (en) | Image detection device capable of recognizing multi-angle objective | |
| Yang et al. | Deep learning‐based bolt loosening detection for wind turbine towers | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN105654066A (en) | Vehicle identification method and device | |
| CN111611889B (en) | Miniature insect pest recognition device in farmland based on improved convolutional neural network | |
| CN112668696A (en) | Unmanned aerial vehicle power grid inspection method and system based on embedded deep learning | |
| CN107194396A (en) | Method for early warning is recognized based on the specific architecture against regulations in land resources video monitoring system | |
| CN111652835A (en) | Method for detecting insulator loss of power transmission line based on deep learning and clustering | |
| CN111008576B (en) | Pedestrian detection and model training method, device and readable storage medium | |
| CN107180245A (en) | A kind of indoor and outdoor scene recognition method and device | |
| CN110378837A (en) | Object detection method, device and storage medium based on fish-eye camera | |
| CN112907614A (en) | Yoov 5-segnet insulator string contour extraction method based on depth feature fusion | |
| CN111444816A (en) | Multi-scale dense pedestrian detection method based on fast RCNN | |
| CN114724140A (en) | Strawberry maturity detection method and device based on YOLO V3 | |
| Gopal et al. | Tiny object detection: Comparative study using single stage CNN object detectors | |
| CN110618129A (en) | Automatic power grid wire clamp detection and defect identification method and device | |
| CN113313110A (en) | License plate type recognition model construction and license plate type recognition method | |
| CN112926415A (en) | Pedestrian avoiding system and pedestrian monitoring method | |
| CN115471966A (en) | Self-learning intrusion alarm method, system, medium and equipment based on vibration optical fiber detection | |
| CN112101313A (en) | Machine room robot inspection method and system | |
| CN110334703B (en) | Ship detection and identification method in day and night image | |
| CN117152513A (en) | Vehicle boundary positioning method for night scene |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20221213 |