CN117235590A - Light-weight optical fiber perimeter protection algorithm based on Markov transition field - Google Patents
Light-weight optical fiber perimeter protection algorithm based on Markov transition field Download PDFInfo
- Publication number
- CN117235590A CN117235590A CN202311032469.9A CN202311032469A CN117235590A CN 117235590 A CN117235590 A CN 117235590A CN 202311032469 A CN202311032469 A CN 202311032469A CN 117235590 A CN117235590 A CN 117235590A
- Authority
- CN
- China
- Prior art keywords
- signal
- lightweight
- optical fiber
- markov transition
- time sequence
- 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 31
- 230000007704 transition Effects 0.000 title claims abstract description 28
- 238000012545 processing Methods 0.000 claims abstract description 14
- 230000006399 behavior Effects 0.000 claims abstract description 12
- 238000013135 deep learning Methods 0.000 claims abstract description 12
- 238000006243 chemical reaction Methods 0.000 claims abstract description 10
- 230000008859 change Effects 0.000 claims abstract description 7
- 230000007123 defense Effects 0.000 claims abstract description 7
- 238000001914 filtration Methods 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 20
- 238000010586 diagram Methods 0.000 claims description 15
- 238000000034 method Methods 0.000 claims description 12
- 239000000835 fiber Substances 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000013508 migration Methods 0.000 claims description 3
- 230000005012 migration Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012856 packing Methods 0.000 claims description 2
- 230000000737 periodic effect Effects 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 abstract description 2
- 230000009545 invasion Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000009194 climbing Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of manufacturing of optical fiber perimeter protection equipment, in particular to a lightweight optical fiber perimeter protection algorithm based on a Markov transition field, which is used for pre-emphasis processing of optical fiber signals of a perimeter defense area acquired by an optical fiber vibration sensing system, and improving a high-frequency part of an intrusion signal so as to inhibit low-frequency noise in the signals; converting the pre-emphasis signal by adopting a Markov transition field, stateing the value of the one-dimensional time sequence signal, calculating the conversion probability of the time sequence signal value between every two time sequence values, and obtaining the change characteristics of the time sequence signal value; the lightweight deep learning algorithm model is constructed, a depth separable module is used, smaller precision is sacrificed, the calculated amount is greatly reduced, and the model is suitable for lightweight deployment of the edge end; inputting the feature map converted by the Markov transition field into a lightweight model for prediction to obtain the behavior type of the intrusion signal, and realizing the filtering of the interference event and the identification and alarm of the intrusion behavior.
Description
Technical field:
the invention relates to the technical field of manufacturing of optical fiber perimeter protection equipment, in particular to a lightweight optical fiber perimeter protection algorithm based on a Markov transition field.
The background technology is as follows:
invasion actions such as personnel climbing and artificial damage often occur in perimeter protection scenes such as industrial park fences, school perimeter, expressway/railway fence and the like, and the traditional manual inspection method is limited by time and cost and cannot quickly and accurately discover the invasion actions; the distributed optical fiber perimeter intrusion detection technology can realize long-distance all-weather detection, is matched with the real-time identification of a deep learning algorithm model, and is widely applied to perimeter protection scenes. Because of the factors of multiple interference factors of the perimeter environment, long monitoring distance and the like, and the characteristics of sensitive signals, high resolution and large data volume of the optical fiber, the perimeter protection algorithm has the problems of multiple false alarms, multiple occupied computing resources and the like, high-power equipment support is required for the construction of the signal filtering, feature extraction and recognition algorithm, and a host computer or a server with higher performance is required for the perimeter protection algorithm based on deep learning to perform computation.
Traditional perimeter protection algorithms often extract only a single or a few features of vibration signals to perform threshold judgment, and have the disadvantages of more debugging parameters, complex debugging and more false positives. Or the recognition is performed by adopting a deep learning mode, frequency domain information such as kurtosis and spectrum information is extracted as characteristics, time correlation characteristics among vibration signals are difficult to extract, meanwhile, the model is complex in calculation and large in calculation amount, and the light-weight deployment requirement on the existing low-calculation-force edge equipment is often difficult to meet.
The invention comprises the following steps:
aiming at the defects and shortcomings in the prior art, the invention provides a lightweight optical fiber perimeter protection algorithm based on Markov transition fields in unattended stations, valve chambers, central control rooms and machine rooms, which optimizes the signal filtering process, fully considers the computing capacity of the edge end and uses an identification model suitable for edge end deployment, so that the algorithm can be deployed in various scenes with limited computing power.
The invention is achieved by the following measures:
a lightweight fiber optic perimeter protection algorithm based on markov transition fields, comprising the steps of:
step 1: pre-emphasis processing is carried out on the optical fiber signals of the perimeter defense area acquired by the optical fiber vibration sensing system, and the high-frequency part of the intrusion signal is promoted so as to inhibit low-frequency noise in the signals;
step 2: converting the pre-emphasis signal by adopting a Markov transition field, stateing the value of a one-dimensional time sequence signal, calculating the conversion probability of the time sequence signal value between every two time sequence values, acquiring the change characteristics of the time sequence signal value, and constructing a conversion probability matrix, thereby converting the one-dimensional time sequence signal into a two-dimensional image as a signal characteristic diagram;
step 3: the lightweight deep learning algorithm model is constructed, a depth separable module is used, smaller precision is sacrificed, the calculated amount is greatly reduced, and the model is suitable for lightweight deployment of the edge end;
step 4: inputting the feature map converted by the Markov transition field into a lightweight model for prediction to obtain the behavior type of the intrusion signal, and realizing the filtering of the interference event and the identification and alarm of the intrusion behavior.
In the deployment environment of the perimeter defense area in the step 1, the original optical fiber signals acquired by the optical fiber vibration sensing system are expressed in a form comprising a time domain information matrix: d (D) t×f ＝(d ij ) t×f (i=1, 2,) T, j=1, 2, f), where T represents time, f represents sampling rate, first for the first periodic fiber signal T n ＝(d ij ) t×f (t=1, n=f), the time period length can be selected according to actual conditions, and downsampling processing is carried out on multi-second data; processing the optical fiber signal by adopting a pre-emphasis method to obtain a vibration signal X with enhanced high-frequency information n ＝x n -αx n-1 ，x n ∈X n N=1, 2, f, and the same processing mode is carried out on the data of each subsequent period.
In the step 2 of the invention, the X obtained in the step 1 n Markov transition field conversion is carried out, and X: { X is defined 1 ,x 2 ,...,x n Dividing it into Q regions, arbitrary x i Can be mapped to q i Above, the element following probability of the signal point in region j being in region i can be obtained: t is t i,j ＝P(x i ∈q i |x t-1 ∈q i ) Obtaining a matrix T Q×Q ＝(t i,j ) 1≤i,j≤Q Further deriving a migration probability matrix M, M from the matrix i,j ＝P(q i →q j ) And the matrix probability M is a two-dimensional characteristic diagram obtained by carrying out Markov transition field transformation on the one-dimensional vibration signal.
In the step 3 of the invention, the constructed lightweight deep learning algorithm model network structure consists of three parts of a two-dimensional convolution module, a lightweight convolution module and a classification module:
wherein, conv2d composed of 3*3 convolution kernels comprises a two-dimensional convolution module, the padding is 1, and stride=1; the lightweight convolution module comprises 3 depth separable convolution modules, wherein the depth separable convolution modules consist of a Depthwise depth convolution module and Conv2d with a convolution kernel size of 1*1, a packing of 1 and a stride=1; the Depthwise depth convolution module carries out convolution feature extraction by using Conv2d with the size of 3*3 for each channel of the feature map, and finally carries out channel fusion by using Conv2d with the size of 1*1 to realize the change of the channel number, and simultaneously adds input and output by using shortcut; the classification module firstly carries out Dropout operation, and then uses Linear to carry out mapping of classification events to obtain event types predicted by the model.
Compared with the prior art, the optical fiber vibration signal is processed rapidly and effectively by adopting a pre-emphasis mode, low-frequency noise is restrained, and high-frequency intrusion characteristic signals are supplemented; the fiber vibration signal is converted by adopting a Markov transition field mode, so that the time sequence characteristics of the fiber vibration signal are effectively extracted; a lightweight identification model is built, the extracted features are learned, and intrusion events can be identified rapidly and accurately; based on the above, the lightweight algorithm can be deployed on low-computation-force edge equipment to realize rapid and accurate perimeter intrusion event detection.
Description of the drawings:
FIG. 1 is a test confusion matrix for the algorithm model of the present invention.
Fig. 2 is a flow chart of the present invention.
FIG. 3 is a schematic diagram of raw data in an embodiment of the present invention.
Fig. 4 is a schematic diagram of pre-emphasis processed data in an embodiment of the invention.
FIG. 5 is a schematic diagram of different intrusion behavior characteristics according to an embodiment of the present invention.
Fig. 6 is a diagram of a lightweight deep learning network structure in accordance with the present invention.
The specific embodiment is as follows:
the invention will be further described with reference to the drawings and examples.
Based on the characteristics of the optical fiber vibration signals, the computing capacity of edge equipment is considered, and a lightweight perimeter protection algorithm suitable for edge end deployment is provided. The algorithm adopts a pre-emphasis method to process the fiber vibration, inhibit low-frequency noise signals and supplement high-frequency intrusion behavior characteristic signals; converting the pre-emphasized one-dimensional vibration signal by using a Markov transition field to obtain a two-dimensional characteristic diagram based on signal time continuity; and training the feature map by using a lightweight deep learning model, and judging an optical fiber perimeter intrusion vibration signal by using the model obtained by training.
The whole implementation flow of the invention is shown in fig. 2, and specifically comprises the following steps:
the method comprises the steps that firstly, optical fiber signals of a perimeter defense area collected by an optical fiber vibration sensing system are subjected to pre-emphasis processing, the high-frequency part of an intrusion signal is promoted, and low-frequency noise in the signal is restrained;
converting the pre-emphasis signal by adopting a Markov transition field, stateing the value of the one-dimensional time sequence signal, calculating the conversion probability of the time sequence signal value between every two time sequence values, acquiring the change characteristics of the time sequence signal value, and constructing a conversion probability matrix, thereby converting the one-dimensional time sequence signal into a two-dimensional image as a signal characteristic diagram;
thirdly, constructing a lightweight deep learning algorithm model, and using a depth separable module, wherein smaller precision is sacrificed, the calculated amount is greatly reduced, and the model is suitable for lightweight deployment of edge ends;
and fourthly, inputting the feature map converted by the Markov transition field into a lightweight model for prediction to obtain the behavior type of the intrusion signal, and realizing the filtering of the interference event and the identification and alarm of the intrusion behavior.
Example 1:
in the deployment environment of the perimeter defense area, the original optical fiber signals acquired by the optical fiber vibration sensing system are expressed in a form containing a time domain information matrix: d (D) t×f ＝(d ij ) t×f ，
(i=1, 2,) t, j=1, 2,., f, where t represents time and f represents sampling rate, as shown in fig. 3;
first for the first period of the optical fiber signal T n ＝(d ij ) t×f (t=1, n=f), the time period length can be selected according to actual conditions, and downsampling processing is carried out on multi-second data; processing the optical fiber signal by adopting a pre-emphasis method to obtain a vibration signal X with enhanced high-frequency information n ＝x n -αx n-1 ，x n ∈X n N=1, 2,..f. Carrying out the same processing mode on the data of each period; the processed data is shown in figure 4;
for X n Markov transition field conversion is carried out, and X: { X is defined 1 ,x 2 ,...,x n Dividing it into Q regions, arbitrary x i Can be mapped to q i Above, the element following probability of the signal point in region j being in region i can be obtained: t is t i,j ＝P(x i ∈q i |x t-1 ∈q i ) Obtaining a matrix T Q×Q ＝(t i,j ) 1≤i,j≤Q Further deriving a migration probability matrix M, M from the matrix i,j ＝P(q i →q j ) Moment of coupleThe matrix probability M is a two-dimensional characteristic diagram obtained by converting a one-dimensional vibration signal through a Markov transition field, as shown in fig. 5, the left side of the diagram in fig. 5 is a characteristic diagram of impact damage invasion behavior, and the right side of the diagram is a characteristic diagram of human crossing invasion behavior;
the method comprises the steps of constructing a light-weight deep learning algorithm model network structure, wherein the network structure is shown in fig. 6, and the light-weight network mainly comprises three parts of a two-dimensional convolution module, a light-weight convolution module and a classification module: 1. the two-dimensional convolution module consists of a 3*3 convolution kernel, conv2d and padding are 1, and stride=1; 2. the lightweight convolution module comprises 3 depth separable convolution modules, wherein the depth separable convolution modules are composed of a Depthwise depth convolution module and Conv2d with a convolution kernel size of 1*1, a padding of 1 and stride=1. The Depthwise depth convolution module carries out convolution feature extraction by using Conv2d with the size of 3*3 for each channel of the feature map, and finally carries out channel fusion by using Conv2d with the size of 1*1 to realize the change of the channel number, and simultaneously adds input and output by using shortcut; 3. the classification module firstly carries out Dropout operation, and then uses Linear to carry out mapping of classification events to obtain event types predicted by the model.
With the algorithm of the invention, the total time of data for one time period is about 350 milliseconds, wherein the pre-emphasis processing takes about 5 milliseconds, and the Markov transition time is about 345 milliseconds; model inference predicts an average time of around 77 milliseconds and a model size of 9MB. In contrast to Resnet18, resnet18 takes on average 91 milliseconds or so and the model size is 43MB. The performance of the recognition model is tested, 200 samples are selected for each event to test the recognition accuracy of the model, the experimental result shows that the recognition accuracy of the model is 96.88%, and the test confusion matrix result is shown in fig. 1. The intrusion event represented by the category label "0" is impact damage, the intrusion event represented by the label "1" is a person-representative flap, the intrusion event represented by the label "2" is a person-representative flap, and the event represented by the label "3" is interference.
The total time consumption of the algorithm for judging the intrusion event in one time period is about 427 milliseconds, the recognition accuracy is 96.88 percent, the algorithm can be rapidly deployed to the edge end, and the perimeter intrusion event can be rapidly and accurately recognized.
The method comprises the steps of firstly, adopting pre-emphasis treatment with small calculated amount for an original signal, inhibiting fundamental frequency amplitude and supplementing a high-frequency signal; the signal after pre-emphasis treatment is subjected to feature extraction by adopting a Markov transition field, so that a feature extraction mode is optimized, and the time continuous feature of the optical fiber vibration signal is fully ensured; a lightweight model is built, continuous characteristics of optical fiber vibration signals are focused on learning, model parameters are few, prediction time is short, alarm accuracy is high, and the algorithm is more suitable for edge end deployment. The lightweight algorithm provided by the invention can accurately identify intrusion events and interference without needing to carry out a large amount of parameter adjustment; the filtering and characteristic calculation process is optimized, a lightweight deep learning model is built, the problem that high-power equipment is required by the existing deep learning algorithm is solved, and the method is more suitable for edge end deployment.
Claims (4)
1. A lightweight fiber optic perimeter protection algorithm based on markov transition fields, comprising the steps of:
step 1: pre-emphasis processing is carried out on the optical fiber signals of the perimeter defense area acquired by the optical fiber vibration sensing system, and the high-frequency part of the intrusion signal is promoted so as to inhibit low-frequency noise in the signals;
step 2: converting the pre-emphasis signal by adopting a Markov transition field, stateing the value of a one-dimensional time sequence signal, calculating the conversion probability of the time sequence signal value between every two time sequence values, acquiring the change characteristics of the time sequence signal value, and constructing a conversion probability matrix, thereby converting the one-dimensional time sequence signal into a two-dimensional image as a signal characteristic diagram;
step 3: the lightweight deep learning algorithm model is constructed, a depth separable module is used, smaller precision is sacrificed, the calculated amount is greatly reduced, and the model is suitable for lightweight deployment of the edge end;
step 4: inputting the feature map converted by the Markov transition field into a lightweight model for prediction to obtain the behavior type of the intrusion signal, and realizing the filtering of the interference event and the identification and alarm of the intrusion behavior.
2. The lightweight fiber optic perimeter protection algorithm based on markov transition fields of claim 1, wherein in the deployment environment of the perimeter defense area in step 1, the original fiber optic signal collected by the fiber optic vibration sensing system is represented in a form including a time domain information matrix: d (D) t×f ＝(d ij ) t×f (i=1, 2,) T, j=1, 2, f), where T represents time, f represents sampling rate, first for the first periodic fiber signal T n ＝(d ij ) t×f (t=1, n=f), the time period length can be selected according to actual conditions, and downsampling processing is carried out on multi-second data; processing the optical fiber signal by adopting a pre-emphasis method to obtain a vibration signal X with enhanced high-frequency information n ＝x n -αx n-1 ，x n ∈X n N=1, 2, f, and the same processing mode is carried out on the data of each subsequent period.
3. The lightweight fiber optic perimeter protection algorithm based on Markov transitions of claim 1, wherein in step 2, for X obtained in step 1 n Markov transition field conversion is carried out, and X: { X is defined 1 ,x 2 ,...,x n Dividing it into Q regions, arbitrary x i Can be mapped to q i Above, the element following probability of the signal point in region j being in region i can be obtained: t is t i,j ＝P(x i ∈q i |x t-1 ∈q i ) Obtaining a matrix T Q×Q ＝(t i,j ) 1≤i,j≤Q Further deriving a migration probability matrix M, M from the matrix i,j ＝P(q i →q j ) And the matrix probability M is a two-dimensional characteristic diagram obtained by carrying out Markov transition field transformation on the one-dimensional vibration signal.
4. The light-weight optical fiber perimeter protection algorithm based on the Markov transition field of claim 1, wherein in the step 3, the constructed light-weight deep learning algorithm model network structure is composed of three major parts of a two-dimensional convolution module, a light-weight convolution module and a classification module:
wherein, conv2d composed of 3*3 convolution kernels comprises a two-dimensional convolution module, the padding is 1, and stride=1; the lightweight convolution module comprises 3 depth separable convolution modules, wherein the depth separable convolution modules consist of a Depthwise depth convolution module and Conv2d with a convolution kernel size of 1*1, a packing of 1 and a stride=1; the Depthwise depth convolution module carries out convolution feature extraction by using Conv2d with the size of 3*3 for each channel of the feature map, and finally carries out channel fusion by using Conv2d with the size of 1*1 to realize the change of the channel number, and simultaneously adds input and output by using shortcut; the classification module firstly carries out Dropout operation, and then uses Linear to carry out mapping of classification events to obtain event types predicted by the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311032469.9A CN117235590A (en) | 2023-08-16 | 2023-08-16 | Light-weight optical fiber perimeter protection algorithm based on Markov transition field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311032469.9A CN117235590A (en) | 2023-08-16 | 2023-08-16 | Light-weight optical fiber perimeter protection algorithm based on Markov transition field |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117235590A true CN117235590A (en) | 2023-12-15 |
Family
ID=89081569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311032469.9A Pending CN117235590A (en) | 2023-08-16 | 2023-08-16 | Light-weight optical fiber perimeter protection algorithm based on Markov transition field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117235590A (en) |
-
2023
- 2023-08-16 CN CN202311032469.9A patent/CN117235590A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109766992B (en) | Industrial control abnormity detection and attack classification method based on deep learning | |
| CN114755745B (en) | Hail weather identification and classification method based on multi-channel depth residual shrinkage network | |
| CN112562255B (en) | Intelligent image detection method for cable channel smoke and fire conditions in low-light-level environment | |
| CN110390308B (en) | Video behavior identification method based on space-time confrontation generation network | |
| CN111103976A (en) | Gesture recognition method and device and electronic equipment | |
| CN112507861A (en) | Pedestrian detection method based on multilayer convolution feature fusion | |
| CN116579616B (en) | Risk identification method based on deep learning | |
| CN111738044A (en) | Campus violence assessment method based on deep learning behavior recognition | |
| CN114898470A (en) | Fall behavior detection method and system based on improved YOLOv5 | |
| CN111695413A (en) | Signal first arrival pickup method and device combining U-Net and Temporal encoding | |
| CN116363748A (en) | Power grid field operation integrated management and control method based on infrared-visible light image fusion | |
| CN112308087A (en) | Integrated imaging identification system and method based on dynamic vision sensor | |
| CN116956222A (en) | Multi-complexity behavior recognition system and method based on self-adaptive feature extraction | |
| CN113707175B (en) | Acoustic event detection system based on feature decomposition classifier and adaptive post-processing | |
| CN111476102A (en) | Safety protection method, central control equipment and computer storage medium | |
| CN111242028A (en) | Remote sensing image ground object segmentation method based on U-Net | |
| CN110599460A (en) | Underground pipe network detection and evaluation cloud system based on hybrid convolutional neural network | |
| CN117235590A (en) | Light-weight optical fiber perimeter protection algorithm based on Markov transition field | |
| Arshad et al. | Anomalous Situations Recognition in Surveillance Images Using Deep Learning | |
| CN110458219B (en) | phi-OTDR vibration signal identification algorithm based on STFT-CNN-RVFL | |
| CN113936300A (en) | Construction site personnel identification method, readable storage medium and electronic device | |
| CN112101244A (en) | Method for predicting optical fiber vibration signal danger level based on deep learning | |
| CN117808650B (en) | Precipitation prediction method based on Transform-Flownet and R-FPN | |
| CN113516242B (en) | Self-attention mechanism-based through-wall radar human body action recognition method | |
| CN117152668B (en) | Intelligent logistics implementation method, device and equipment based on Internet of things |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |