CN110179466A - Breathing detection system after calamity based on intelligent terminal - Google Patents
Breathing detection system after calamity based on intelligent terminal Download PDFInfo
- Publication number
- CN110179466A CN110179466A CN201910476112.7A CN201910476112A CN110179466A CN 110179466 A CN110179466 A CN 110179466A CN 201910476112 A CN201910476112 A CN 201910476112A CN 110179466 A CN110179466 A CN 110179466A
- Authority
- CN
- China
- Prior art keywords
- breathing
- calamity
- layer
- detection
- frame
- 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
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Physiology (AREA)
- Pulmonology (AREA)
- Pathology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention belongs to sensor network, mobile computing and voice processing technology field, a kind of specifically breathing detection system after calamity based on intelligent terminal.System of the invention includes trigger module and detection module；The trigger module is used for the automatic triggering system after calamity and starts to work, and the detection module is used to detect the breath state of the trapped person of terminal surrounding.The present invention provides trapped person's breathing detections after calamity, according to current technology, corresponding signal transmission or detection means can be set according to the type of terminal device, to provide a kind of possible search and rescue reference direction to rescue worker, rather than the search of blindness, the probability that trapped person is rescued can be improved under specific circumstances.
Description
Technical field
The invention belongs to sensor network, mobile computing and voice processing technology field, specifically one kind is based on
Breathing detection system after the calamity of intelligent terminal.
Background technique
For unbroken natural calamity often to the society of the mankind, economy causes heavy losses.All can in annual worldwide
The natural calamities such as various earthquakes, flood, cyclone, mud-rock flow occur；Government, scientific research institution etc. are also being dedicated to always pair
The prevention of disaster, the improvement after rescue and calamity.However, accurate forecast disaster times is so the task of a great challenge.?
2000-2012 results in global 1,200,000 people dead, and 2,900,000,000 people are influenced by disaster is different degrees of, and cause 1.7 trillion
The economic loss of dollar.Although the mankind enter the information age from the industrial age, still lacks and contain natural disaster
Ability, or even the ability of prediction disaster is also very limited, therefore the efficiency of Post disaster relief work is to reduce people's life and wealth
Produce the key of loss.According to rescue 72 hours laws of gold, the effective disaster area information that obtains is the key that rescue.It is ruined after calamity
The Base communication facility that going out property is destroyed can not support, and existing scheme is by building temporary base communications facility, or benefit
Local small range networking is carried out with the near-field communication of mobile phone.At this time, if it is possible to which the respiration information for obtaining trapped person will
It is of crucial importance to rescue work.Currently, carrying out the technology of breathing detection to trapped person after calamity using smart phone not yet.
Summary of the invention
The object of the present invention is in view of the above-mentioned problems, provide breathing detection system after a kind of calamity based on intelligent terminal.
The technical scheme is that
Breathing detection system after calamity based on intelligent terminal, which is characterized in that the system comprises trigger modules and detection
Module；The trigger module is used for the automatic triggering system after calamity and starts to work, and the detection module is for detecting terminal surrounding
Trapped person breath state.
Further, refer to after the calamity after occurring after earthquake disaster, then the triggering mode of the touch block is, according to monitoring
To P-wave signal or trigger after receiving corresponding push signal.
The foundation of above scheme is that seismic wave is divided into two kinds, and P wave and S wave, after earthquake occurs, P wave is spread first, main
Small vibration is caused, it is the main reason for damaging that rear S wave, which reaches, within 10 seconds or so.Now to the monitoring precision of P wave
It is higher, after monitoring P wave, one can be pushed to the terminal device of devastated and special is disappeared by way of push
Breath is to trigger system.
Further, the vital sign of the detection module detection is breathing, specifically:
It is identified from voice signal according to the voice signal that intelligent terminal acquires by trained Recognition with Recurrent Neural Network
Breathing out；
The Recognition with Recurrent Neural Network includes 1 layer of input layer, 5 layers of hidden layer and 1 layer of output layer, and enabling input is x, hlIndicate the
L hidden layer, then corresponding h0It indicates input x, enables and not recycling for 4 layers before hidden layer, for each time t, preceding 4 layers of result meter
It calculates are as follows:
Wherein, g (x)=min { max { 0, x }, 20 } is the linear unit activating function of limitation of truncation, Wl,blFor l layers of power
Value matrix and biasing；
The 5th layer of hidden layer is circulation layer:
Output layer uses sigmod activation primitive
Loss function is as follows using the form of cross entropy, whereinIndicate the prediction result of neural network
The training method of the Recognition with Recurrent Neural Network is to enable X={ (x1,y1),(x2,y2),…,(xN,yN) it is training set,
WhereinFor a certain frame in one section of voice signal, yi={ 0,1 } be the frame label, 0
It indicates to breathe no more in the frame, 1 indicates there is breathing in the frame；Assuming that X is one section of recording for 10 seconds comprising breathing, sample frequency
For 44100Hz, 40 milliseconds are enabled as a frame, then X is a training set comprising 250 samples, and the dimension of each sample is 1764, i.e.,
N=250, n=1764；The target of Recognition with Recurrent Neural Network is that the voice signal inputted to one section exports a string of 0-1 sequences, for sentencing
Break and whether contains breathing in this section of voice.
Trapped person's breathing detection can be set according to current technology according to terminal after aspects of which providing calamity
Corresponding signal transmission or detection means is arranged in standby type, to provide a kind of possible search and rescue reference to rescue worker
Direction, rather than the search of blindness can improve the probability that trapped person is rescued under specific circumstances.
Such as breathing detection system after above-mentioned calamity is mounted on mobile phone, then can trigger module trigger calamity after breathing detection
The colleague of system triggers the signal sending system being correspondingly arranged, and sends signal to the external world with the frequency of certain agreement, makes extraneous energy
Enough receive and identify the signal, for judge whether there is indicator of trapped personnel and indicator of trapped personnel whether also survival provide it is a kind of according to
According to.Certainly, other intelligent terminals can also install breathing detection system after calamity, and be designed accordingly according to respective feature
Signalling formula.
Fig. 2 describes the structure of entire depth Recognition with Recurrent Neural Network (DRNN).
Beneficial effects of the present invention are, by the breathing of indicator of trapped personnel after real-time monitoring calamity, can determine that trapped person is
No survival, the signal issued by detection terminal, can provide guidance for rescue worker.
Detailed description of the invention
Fig. 1 is post-processing schematic diagram；
Fig. 2 is Recognition with Recurrent Neural Network structure chart；
Fig. 3 is field experiment experimental situation；
Fig. 4 test data collection process；
Specific embodiment
Bright technical solution of the present invention is described in detail with reference to the accompanying drawings and examples.
System trigger:
Seismic wave is divided into two kinds, P wave and S wave.After earthquake occurs, P wave is spread first, mainly causes small vibration,
Rear S wave reaches within 10 seconds or so, is the main reason for damaging.It is now higher to the monitoring precision of P wave, when monitoring P wave
Later, a special message can be pushed to trigger iSurvivor to the mobile phone of devastated by way of push.
Breathing detection:
The core of this module is one large-scale Recognition with Recurrent Neural Network of training to identify to one section of voice signal
[5,6,7,8,9,10,11], identify wherein whether the breathing containing someone.Enable X={ (x1,y1),(x2,y2),…,(xN,
yN) it is training set, whereinFor a certain frame in one section of voice signal, yi=0,
It 1 } is the label of the frame, 0 indicates to breathe no more in the frame, and 1 indicates there is breathing in the frame.Assuming that X is one section comprising breathing
Recording (sample frequency 44100Hz) in 10 seconds enables 40 milliseconds as a frame, then X is a training set comprising 250 samples, each
The dimension of sample is 1764.(N=250, n=1764).The target of RNN is that the voice signal inputted to one section exports a string of 0-1
Sequence, for judging whether contain breathing in this section of voice.
The present invention selects one 6 layers of RNN model, and first layer is input layer, and wherein 2-5 is hidden layer, and the 6th layer is output
Layer.For input an x, hlIndicate first of hidden layer, then corresponding h0Indicate input x, first 4 layers do not recycle.Therefore, for each
Time t, preceding 4 layers of result calculate as follows:
Wherein g (x)=min { max { 0, x }, 20 } 0, x } it is limitation linear unit (ReLu) activation primitive being truncated, Wl,bl
Weight matrix and biasing for l layers.Layer 5 is circulation layer:
Layer 6 uses sigmod activation primitive (whereinInput is mapped between 0-1 as output,
Last loss function uses the form of cross entropy:
Gradient is calculated using the reverse conduction (BPTT) about the time.The present invention is using the learning rate of a constant entire
In training process.
First subgraph for scheming .1 is the continuous 1000 frame i.e. spirogram of 40 second datas picked out at random from test set,
The longitudinal axis is the corresponding label value of each frame.There it can be seen that being the label misplayed, second subgraph inside green dotted-line ellipse
It is after DRNN identification as a result, there it can be seen that normal breath data is continuous and in periodically, and the label beaten
Or wrong point as a result, it is all intermittent or comes up suddenly.Therefore, post-processing operation is added in the present invention, both removes identification knot
It is less than the respiration case of 10 frames in fruit, frame number inside second subgraph Green dotted-line ellipse for scheming .1 is both less than to the view of 10 frames
For mistake point, gives and correct.Understand from intuitive, the time once exhaled is less than 0.5s, is the probability very little of eupnea.
Embodiment
Experiment for DRNN, the present invention first train extensive RNN model of the invention on a cluster, then training
Good model parameter imports iSurvivor.
Breathing detection experiment:
Simulation test: figure .4 is test data collection process, before volunteer is sitting in desk, four mobile phones putting in front from
The distance of table edge is respectively 0.2m, 0.4m, 0.6m, 0.8m.Because distance is remoter, signal-to-noise ratio is lower, labels and is just more difficult to.
Start to acquire so the present invention enables four mobile phones synchronize, in this way as their label, need to only be beaten for nearest mobile phone
Upper label.The present invention acquires the breath data of 10 people, everyone is 2 minutes, and tagged as test set use for its
To examine the effect of DRNN of the invention.Meanwhile the present invention chooses two models of DNN and DBN as a comparison.
Experimental result such as table 1, wherein BA is discrimination of the model to breathing frame, and NBA is identification of the model to apnea frame
Rate, WAR are total discrimination:
Table 1
As can be seen that DRNN is compared to DNN and DBN, WAR highest；Reason is their BA value all relatively, but
Differ larger in NBA value, the NBA of DRNN is more than 99%, if each frame is regarded as a sample, positive example sample and negative example
The ratio of sample is more than 1:4.Last column TTR of table .1 is the respiration rate that identifies than upper true respiration rate, from
In it can also be seen that the TTR index of DRNN is also highest, because DRNN can consider the breathing frame of history compared to DBN and DNN.
On-the-spot test: breathing is identified finally, the present invention has carried out present system in the Wenchuan County town Ying Xiu earthquake relics
The experiment of rate.Fig. 3 is experiment scene, and experimenter hides in ruins, places 4 mobile phones respectively up and down around it, due to
Experimental situation is dangerous and experiment difficulty is larger, and the present invention has only carried out test in 1 minute, and experimenter oneself is allowed to record its breathing
Number, respiration rate be 19 times, last four mobile phones know respectively respiration rate out be 15 times, 15 times, 11 times, 6 times, respectively
It is 0.48m, 0.7m, 1.1m, 1.7m with a distance from experimenter.After distance is more than 1m, recognition accuracy declines brighter
It is aobvious.
Claims (3)
1. breathing detection system after the calamity based on intelligent terminal, which is characterized in that the system comprises trigger modules and detection mould
Block；The trigger module is used for the automatic triggering system after calamity and starts to work, and the detection module is for detecting terminal surrounding
Breathing detection.
2. breathing detection system after the calamity according to claim 1 based on intelligent terminal, which is characterized in that be after the calamity
After occurring after finger earthquake disaster, then the triggering mode of the trigger module is, according to the P-wave signal monitored or receives
It is triggered after corresponding push signal.
3. breathing detection system after the calamity according to claim 2 based on intelligent terminal, which is characterized in that the detection mould
The vital sign of block detection is breathing, specifically:
It identifies and exhales from voice signal by trained Recognition with Recurrent Neural Network according to the voice signal that intelligent terminal acquires
Sound absorption；
The Recognition with Recurrent Neural Network includes 1 layer of input layer, 5 layers of hidden layer and 1 layer of output layer, and enabling input is x, hlIndicate first it is hidden
Containing layer, then corresponding h0It indicates input x, enables and not recycling for 4 layers before hidden layer, for each time t, preceding 4 layers of result is calculated are as follows:
Wherein, g (x)=min { max { 0, x }, 20 } is the linear unit activating function of limitation of truncation, Wl,blFor l layers of weight square
Battle array and biasing；
The 5th layer of hidden layer is circulation layer:
Output layer uses sigmod activation primitive,
Loss function is as follows using the form of cross entropy, whereinIndicate the prediction result of neural network
The training method of the Recognition with Recurrent Neural Network is to enable X={ (x1,y1),(x2,y2),…,(xN,yN) it is training set, whereinFor a certain frame in one section of voice signal, yi={ 0,1 } is the label of the frame, and 0 indicates
It breathes no more in the frame, 1 indicates there is breathing in the frame, and T is total sampling time；Assuming that X is one section of recording for 10 seconds comprising breathing
Sound, sample frequency 44100Hz enable 40 milliseconds as a frame, then X is a training set comprising 250 samples, the dimension of each sample
Degree is 1764, i.e. N=250, n=1764；The target of Recognition with Recurrent Neural Network is that the voice signal inputted to one section exports a string of 0-1
Sequence, for judging whether contain breathing in this section of voice.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910476112.7A CN110179466A (en) | 2019-06-03 | 2019-06-03 | Breathing detection system after calamity based on intelligent terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910476112.7A CN110179466A (en) | 2019-06-03 | 2019-06-03 | Breathing detection system after calamity based on intelligent terminal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110179466A true CN110179466A (en) | 2019-08-30 |
Family
ID=67719776
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910476112.7A Pending CN110179466A (en) | 2019-06-03 | 2019-06-03 | Breathing detection system after calamity based on intelligent terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110179466A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108573713A (en) * | 2017-03-09 | 2018-09-25 | 株式会社东芝 | Speech recognition equipment, audio recognition method and storage medium |
| CN108648748A (en) * | 2018-03-30 | 2018-10-12 | 沈阳工业大学 | Acoustic events detection method under hospital noise environment |
| CN109036410A (en) * | 2018-08-30 | 2018-12-18 | Oppo广东移动通信有限公司 | Audio recognition method, device, storage medium and terminal |
| CN109065055A (en) * | 2018-09-13 | 2018-12-21 | 三星电子（中国）研发中心 | Method, storage medium and the device of AR content are generated based on sound |
| CN109471391A (en) * | 2018-11-09 | 2019-03-15 | 高波 | A kind of system and its prompting device and operating method for improving Post disaster relief efficiency |
-
2019
- 2019-06-03 CN CN201910476112.7A patent/CN110179466A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108573713A (en) * | 2017-03-09 | 2018-09-25 | 株式会社东芝 | Speech recognition equipment, audio recognition method and storage medium |
| CN108648748A (en) * | 2018-03-30 | 2018-10-12 | 沈阳工业大学 | Acoustic events detection method under hospital noise environment |
| CN109036410A (en) * | 2018-08-30 | 2018-12-18 | Oppo广东移动通信有限公司 | Audio recognition method, device, storage medium and terminal |
| CN109065055A (en) * | 2018-09-13 | 2018-12-21 | 三星电子（中国）研发中心 | Method, storage medium and the device of AR content are generated based on sound |
| CN109471391A (en) * | 2018-11-09 | 2019-03-15 | 高波 | A kind of system and its prompting device and operating method for improving Post disaster relief efficiency |
Non-Patent Citations (1)
| Title |
|---|
| 郝志峰: "《数据科学与数学建模》", 31 January 2019, 华中科技大学出版社 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111626162B (en) | Water rescue system based on space-time big data analysis and drowning warning condition prediction method | |
| CN110133610A (en) | ULTRA-WIDEBAND RADAR action identification method based on time-varying distance-Doppler figure | |
| CN109065046A (en) | Method, apparatus, electronic equipment and the computer readable storage medium that voice wakes up | |
| CN106251860B (en) | Unsupervised novelty audio event detection method and system for security field | |
| CN105574489B (en) | Based on the cascade violence group behavior detection method of level | |
| CN112148997B (en) | Training method and device for multi-modal countermeasure model for disaster event detection | |
| CN110516138A (en) | A kind of food safety affair early warning system threatening information bank based on multi-source self refresh | |
| CN106205606A (en) | A kind of dynamic positioning and monitoring method based on speech recognition and system | |
| CN109376613A (en) | Video brainpower watch and control system based on big data and depth learning technology | |
| CN110197332A (en) | A kind of overall control of social public security evaluation method | |
| CN101968848A (en) | Video monitoring method and system and video monitoring alarm system | |
| CN108983690A (en) | Building Indoor Environment intelligent monitor system | |
| CN106197557B (en) | A kind of wall vibration detecting system and its detection method | |
| CN109002746A (en) | 3D solid fire identification method and system | |
| Davani et al. | Reporting the unreported: Event extraction for analyzing the local representation of hate crimes | |
| CN111191498A (en) | Behavior recognition method and related product | |
| Ning et al. | Fall detection algorithm based on gradient boosting decision tree | |
| CN110930632A (en) | Early warning system based on artificial intelligence | |
| CN114241557A (en) | Image recognition method, device and equipment, intelligent door lock and medium | |
| CN110179466A (en) | Breathing detection system after calamity based on intelligent terminal | |
| CN109509329B (en) | Drowning alarm method based on wearable device and wearable device | |
| CN116912770A (en) | Public place smoking detection method based on improved YOLOv8 | |
| CN116392122A (en) | Mental stress level judging method for diver escape training based on micro-expressions | |
| CN115311601A (en) | Fire detection analysis method based on video analysis technology | |
| Gowrishankar et al. | IoT based Smart ID Card for Working Woman Safety |
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 |
Application publication date: 20190830 |
|
| RJ01 | Rejection of invention patent application after publication |