CN109919071B - Flame identification method based on infrared multi-feature combined technology - Google Patents
Flame identification method based on infrared multi-feature combined technology Download PDFInfo
- Publication number
- CN109919071B CN109919071B CN201910149529.2A CN201910149529A CN109919071B CN 109919071 B CN109919071 B CN 109919071B CN 201910149529 A CN201910149529 A CN 201910149529A CN 109919071 B CN109919071 B CN 109919071B
- Authority
- CN
- China
- Prior art keywords
- flame
- feature
- area
- temperature
- region
- 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.)
- Active
Links
Abstract
The invention discloses a flame identification method based on an infrared multi-feature combined technology, which can monitor objects in a field of view all weather by an infrared thermal imager, avoid the influence caused by insufficient illumination and interferents, can find a suspected flame area by utilizing a temperature threshold and a temperature sudden change coefficient, extract the temperature feature of flame, the motion feature, the contour feature and the texture feature of the flame from the suspected flame area, and then input the features into a trained BP neural network, thereby realizing multi-feature combined flame identification aiming at an indoor space. The method can more intuitively analyze the collection characteristics and the temperature information of the flame by monitoring whether the flame is generated or not through infrared rays, and simultaneously, the method of the neural network is utilized to combine the identification of various characteristics together, thereby improving the stability and the identification rate in the system.
Description
Technical Field
The invention relates to the field of flame identification, and particularly provides a flame identification method based on an infrared multi-feature combined technology.
Background
With the continuous and high-speed development of Chinese economy, the influence of fire on the life and property safety of people is greater and greater, and the requirement for flame identification is higher and higher. In the prior art, a high-definition camera is mostly used for monitoring flame, the purpose of flame identification is achieved by identifying flame characteristics, and the interference of a background and other objects is caused while temperature information is lost.
The existing method has at least the following disadvantages:
1. by adopting the visible light camera, although a high-definition flame image can be shot, the most important temperature information of the flame cannot be acquired;
2. the method only aims at identifying the flame with single characteristics, so that the identification result cannot be accurately and efficiently obtained, and especially aims at solving the problems that the shot flame tends to be white and cannot utilize the textural characteristics of the flame due to over-bright flame and over-high temperature;
3. the existing method for removing the flame background can not accurately remove objects with shapes similar to the flame and is interfered by illumination and field environment.
Therefore, the development of a new flame identification method is becoming an urgent problem to be solved.
Disclosure of Invention
In view of this, the present invention provides a flame identification method based on an infrared multi-feature combination technique, so as to solve the problems in the prior art that temperature information of flames cannot be utilized, texture information of flames with too high temperature or brighter flame is difficult to be utilized, and flame backgrounds with shapes similar to the flames are difficult to be removed.
The technical scheme provided by the invention is as follows: a flame identification method based on infrared multi-feature combined technology comprises the following steps:
s1: acquiring an infrared image and temperature information of a monitored area by using an infrared camera;
s2: searching a suspected flame area S and determining the lowest temperature T of the suspected flame area SminCarrying out high-pass filtering on the infrared image to filter out a low-temperature background and further obtain a suspected flame area S0Wherein, the suspected flame area S in the infrared image is that the temperature is greater than the temperature threshold value T0Or the temperature becomes large to a temperature shock coefficient K0Multiple and K0A set of more than twice pixel points;
s3: in the suspected flame region S0On the basis of the area S, 5-8 pixels are outwards expanded to obtain the area S1So as to obtain more edge information and temperature information during post-processing;
s4: in the region S1And extracting the motion characteristic, the contour characteristic and the texture characteristic of the flame, and inputting the characteristic information into the trained BP neural network to realize the identification of the flame.
More preferably, in S2, the temperature threshold T0And coefficient of temperature shock K0Are all preset values, temperature threshold T0Is 3 times of the average ambient temperature and has a temperature shock coefficient K0Is 4.
More preferably, in S3, the pseudo flame region S is0On the basis of the above-mentioned three-dimensional image data, 5 pixels are outward expanded to obtain region S1。
Further preferably, in S4, the motion feature of the flame is a geometric feature of a flame motion trajectory, and the feature of a flame centroid motion trajectory is used as a flame motion feature value, where the flame motion feature extraction step is as follows:
a. region S extraction using Sobel operator1Obtaining a flame connected domain Q according to the edge information;
b. calculating the position of the mass center of the flame by using a mass center formula;
c. combining the position of the mass center of the flame of each frame of image to obtain the motion trail of the mass center of the flame;
wherein, the centroid formula is as follows:
wherein Q represents a flame communication region, NQRepresents the number of pixels in the flame connected region Q, (x)i,yi) Are coordinates of the center of mass.
Further preferably, in S4, the contour feature of the flame includes an outer contour feature of the flame and a circularity feature of the flame.
Further preferably, the extracting step of the contour characteristic value of the flame is as follows: obtaining a suspected flame area S by using temperature information during flame combustion1Then, for S1Carrying out median filtering to reduce noise of the infrared image to obtain a region S2Then, the Sobel operator is used to pair the region S2Performing edge extraction to obtain the profile S of the flame3And taking the profile information of the flame as the external profile characteristic value of the flame.
More preferably, the circularity is a ratio of a square of a perimeter of a boundary of the region to an area thereof, and the profile S of the flame is obtained3Then, calculateS3Area and S of3A circumference of (2), wherein S3Has an area of region S3Number of inner pixels, S3The perimeter of the circle is the accumulated sum of the distances between adjacent pixel points, and the calculation formula of the circularity is as follows:
wherein S iskIs the area of the flame, i.e. S3Area of (L)kIs the circumference of the flame, i.e. S3The circumference of (a).
More preferably, in S4, the flame texture feature extraction step includes:
for region S1Performing wavelet analysis, and grouping the regions S using second-order wavelet filter1And processing, analyzing the size of the spatial change rate of the pixel value, decomposing the region by using a second-order filter bank to obtain three sub-image components of a diagonal (HH), a vertical (HL) and a horizontal (LH), and taking the square sum of the three components as the texture characteristic value of the flame.
Further preferably, the training process of the BP neural network is as follows: a BP neural network is adopted, combustion flames of 6 flammable materials, namely wood, cotton cloth, gasoline, paper, alcohol and cloth, are selected as training signals of the BP neural network in indoor windless, breezy and strong wind environments, and simultaneously, objects which are high in temperature and similar to the flames in the environment are selected as training signals of the neural network.
Preferably, the number of hidden layer nodes of the BP neural network is 20, the hidden layer neurons adopt a tansig function, and the output layer adopts a logsig function.
According to the flame identification method based on the infrared multi-feature combination technology, the thermal infrared imager can be used for monitoring objects in a field of view all weather, the influence caused by insufficient illumination and interferents is avoided, the temperature information of a monitoring area can be analyzed in real time, whether flames exist or not can be analyzed more visually, and meanwhile, the neural network can be used for identifying the flames more quickly and more accurately.
Detailed Description
The invention will be further explained with reference to specific embodiments, without limiting the invention.
The invention provides a flame identification method based on an infrared multi-feature combined technology, which comprises the following steps:
s1: acquiring an infrared image and temperature information of a monitored area by using an infrared camera;
s2: searching a suspected flame area S and determining the lowest temperature T of the suspected flame area SminCarrying out high-pass filtering on the infrared image to filter out a low-temperature background and further obtain a suspected flame area S0Wherein, the suspected flame area S in the infrared image is that the temperature is greater than the temperature threshold value T0Or the temperature becomes large to a temperature shock coefficient K0Multiple and K0A set of more than twice pixel points;
wherein the temperature threshold value T0And coefficient of temperature shock K0Are all preset values, temperature threshold T0Preferably 3 times of the average ambient temperature, and a temperature shock coefficient K0Is 4;
s3: in the suspected flame region S0On the basis of the area S, 5-8 pixels are outwards expanded to obtain the area S1So as to obtain more edge information and temperature information during post-processing;
s4: in the region S1Extracting the motion characteristic, contour characteristic and texture characteristic of the flame, and inputting the characteristic information into the trained BP neural network to realize the identification of the flame;
the flame motion characteristic is that the geometric characteristic of the flame motion track is utilized, the characteristic of the flame centroid motion track is used as the flame motion characteristic value, and the extraction steps of the flame motion characteristic are as follows:
a. region S extraction using Sobel operator1Obtaining a flame connected domain Q according to the edge information;
b. calculating the position of the mass center of the flame by using a mass center formula;
c. combining the position of the mass center of the flame of each frame of image to obtain the motion trail of the mass center of the flame;
wherein, the centroid formula is as follows:
wherein Q represents a flame communication region, NQRepresents the number of pixels in the flame connected region Q, (x)i,yi) Are coordinates of the center of mass.
The invention collects infrared images, eliminates the interference of low-temperature objects in the environment due to wind and other environments, and completes the identification of flame by combining the motion rule of the flame centroid.
The flame contour characteristics comprise the outer contour characteristics of the flame and the circularity characteristics of the flame, and the extraction steps of the flame contour characteristic value are as follows:
the existing flame profile extraction method mainly extracts flame from background by utilizing the color characteristics of the flame profile, and the method adopts an infrared camera and utilizes the temperature information of the flame during combustion to obtain a suspected flame area S1Then, for S1Carrying out median filtering to reduce noise of the infrared image to obtain a region S2Then, the Sobel operator is used to pair the region S2Performing edge extraction to obtain the profile S of the flame3Taking the profile information of the flame as an external profile characteristic value of the flame;
the invention adopts the circularity information of the flame as the circularity characteristic of the flame, the circularity is the ratio of the square of the perimeter of the regional boundary to the area of the regional boundary, and the profile S of the flame is obtained3Then, calculate S3Area and S of3A circumference of (2), wherein S3Has an area of region S3Number of inner pixels, S3The perimeter of the circle is the accumulated sum of the distances between adjacent pixel points, and the calculation formula of the circularity is as follows:
wherein S iskIs the area of the flame, i.e. S3Area of (L)kThe circumference of the flame is S3The circumference of (a).
The extraction steps of the textural features of the flame are as follows:
for region S1Performing wavelet analysis, and grouping the regions S using second-order wavelet filter1And processing, analyzing the size of the spatial change rate of the pixel value, decomposing the region by using a second-order filter bank to obtain three sub-image components of a diagonal (HH), a vertical (HL) and a horizontal (LH), and taking the square sum of the three components as the texture characteristic value of the flame.
And then inputting the obtained flame motion characteristics, contour characteristics and texture characteristics into a trained BP neural network.
The training process of the BP neural network is as follows: a BP neural network is adopted, combustion flames of 6 flammable materials including wood, cotton cloth, gasoline, paper, alcohol and cloth are selected as training signals of the BP neural network respectively in indoor windless, breezy and strong wind environments, and meanwhile, objects which are high in temperature and similar to the flames in the environment are selected as training signals of the neural network, such as a swaying infrared spotlight and a hot air blower, wherein the number of nodes of a hidden layer of the BP neural network is 20, neurons of the hidden layer adopt a tansig function, and an output layer adopts a logsig function.
According to the flame identification method based on the infrared multi-feature combination technology, all-weather monitoring can be carried out on objects in a visual field range through the thermal infrared imager, the influences caused by insufficient illumination and interferents are avoided, the temperature information of a monitoring area can be analyzed in real time, whether flames exist or not can be analyzed more visually, and meanwhile, the flames can be identified more quickly and more accurately by utilizing the neural network.
The embodiments of the present invention have been written in a progressive manner with emphasis placed on the differences between the various embodiments, and similar elements may be found in relation to each other.
While the embodiments of the present invention have been described in detail, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.
Claims (10)
1. The flame identification method based on the infrared multi-feature combined technology is characterized by comprising the following steps of:
s1: acquiring an infrared image and temperature information of a monitored area by using an infrared camera;
s2: searching a suspected flame area S and determining the lowest temperature T of the suspected flame area SminCarrying out high-pass filtering on the infrared image to filter out a low-temperature background and further obtain a suspected flame area S0Wherein, the suspected flame area S in the infrared image is that the temperature is greater than the temperature threshold value T0Or the temperature becomes large to a temperature shock coefficient K0Multiple and K0A set of more than twice pixel points;
s3: in the suspected flame region S0On the basis of the area S, 5-8 pixels are outwards expanded to obtain the area S1So as to obtain more edge information and temperature information during post-processing;
s4: in the region S1And extracting the motion characteristic, the contour characteristic and the texture characteristic of the flame, and inputting the characteristic information into the trained BP neural network to realize the identification of the flame.
2. The flame identification method based on the infrared multi-feature combined technology as claimed in claim 1, wherein: at S2, temperature threshold T0And coefficient of temperature shock K0Are all preset values, temperature threshold T0Is 3 times of the average ambient temperature and has a temperature shock coefficient K0Is 4.
3. The flame identification method based on the infrared multi-feature combined technology as claimed in claim 1, wherein: in S3, in the pseudo flame area S0On the basis of the above-mentioned three-dimensional image data, 5 pixels are outward expanded to obtain region S1。
4. The flame identification method based on the infrared multi-feature combined technology as claimed in claim 1, wherein: in S4, the motion feature of the flame is a geometric feature of a flame motion trajectory, and the feature of a flame centroid motion trajectory is used as a flame motion feature value, wherein the extraction steps of the flame motion feature are as follows:
a. region S extraction using Sobel operator1Obtaining a flame connected domain Q according to the edge information;
b. calculating the position of the mass center of the flame by using a mass center formula;
c. combining the position of the mass center of the flame of each frame of image to obtain the motion trail of the mass center of the flame;
wherein, the centroid formula is as follows:
wherein Q represents a flame communication region, NQRepresents the number of pixels in the flame connected region Q, (x)i,yi) Are coordinates of the center of mass.
5. The flame identification method based on the infrared multi-feature combined technology as claimed in claim 1, wherein: in S4, the profile features of the flame include an outer profile feature of the flame and a circularity feature of the flame.
6. The method for identifying flames based on the infrared multi-feature combined technology according to claim 5, wherein: the method comprises the following steps of: obtaining a suspected flame area S by using temperature information during flame combustion1Then, for S1Carrying out median filtering to reduce noise of the infrared image to obtain a region S2Then, the Sobel operator is used to pair the region S2Performing edge extraction to obtain the profile S of the flame3And taking the profile information of the flame as the external profile characteristic value of the flame.
7. The method for identifying flames based on the infrared multi-feature combined technology as claimed in claim 6, wherein: the circularity is the ratio of the square of the perimeter of the zone boundary to its area, giving the profile S of the flame3Then, calculate S3Area and S of3A circumference of (2), wherein S3Has an area of region S3Number of inner pixels, S3The perimeter of the circle is the accumulated sum of the distances between adjacent pixel points, and the calculation formula of the circularity is as follows:
wherein S iskIs the area of the flame, i.e. S3Area of (L)kIs the circumference of the flame, i.e. S3The circumference of (a).
8. The flame identification method based on the infrared multi-feature combined technology as claimed in claim 1, wherein: in S4, the flame texture feature extraction step includes:
for region S1Performing wavelet analysis, and grouping the regions S using second-order wavelet filter1And processing, analyzing the size of the spatial change rate of the pixel value, decomposing the region by using a second-order filter bank to obtain three sub-image components of a diagonal (HH), a vertical (HL) and a horizontal (LH), and taking the square sum of the three components as the texture characteristic value of the flame.
9. The flame identification method based on the infrared multi-feature combined technology as claimed in claim 1, wherein: the training process of the BP neural network is as follows: a BP neural network is adopted, combustion flames of 6 flammable materials, namely wood, cotton cloth, gasoline, paper, alcohol and cloth, are selected as training signals of the BP neural network in indoor windless, breezy and strong wind environments, and simultaneously, objects which are high in temperature and similar to the flames in the environment are selected as training signals of the neural network.
10. A flame identification method based on infrared multi-feature combined technology according to claim 9, characterized in that: the number of hidden layer nodes of the BP neural network is 20, the hidden layer neurons adopt tansig functions, and the output layer adopts logsig functions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910149529.2A CN109919071B (en) | 2019-02-28 | 2019-02-28 | Flame identification method based on infrared multi-feature combined technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910149529.2A CN109919071B (en) | 2019-02-28 | 2019-02-28 | Flame identification method based on infrared multi-feature combined technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109919071A CN109919071A (en) | 2019-06-21 |
| CN109919071B true CN109919071B (en) | 2021-05-04 |
Family
ID=66962679
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910149529.2A Active CN109919071B (en) | 2019-02-28 | 2019-02-28 | Flame identification method based on infrared multi-feature combined technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919071B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111579080B (en) * | 2020-04-30 | 2021-09-14 | 沈阳天眼智云信息科技有限公司 | Self-calibration method of infrared thermal image body temperature monitor |
| CN111739250B (en) * | 2020-07-01 | 2022-02-15 | 广东工业大学 | Fire detection method and system combining image processing technology and infrared sensor |
| CN112347942A (en) * | 2020-11-09 | 2021-02-09 | 深圳英飞拓科技股份有限公司 | Flame identification method and device |
| CN114241701A (en) * | 2021-11-24 | 2022-03-25 | 中国民用航空飞行学院 | Multi-parameter fusion detection method based on BP neural network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107798811A (en) * | 2017-10-26 | 2018-03-13 | 上海腾盛智能安全科技股份有限公司 | A kind of tunnel fire monitoring device, monitoring system and monitoring method |
| EP3301640A1 (en) * | 2016-09-30 | 2018-04-04 | JENOPTIK Advanced Systems GmbH | Method and thermal camera for the contactless measurement of a temperature or for observing quickly moving ir scenes with a thermal ir imager |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116746B (en) * | 2013-03-08 | 2016-08-03 | 中国科学技术大学 | A kind of video flame detection method based on multiple features fusion technology |
| CN103886344B (en) * | 2014-04-14 | 2017-07-07 | 西安科技大学 | A kind of Image Fire Flame recognition methods |
| CN103971114B (en) * | 2014-04-23 | 2018-03-23 | 天津航天中为数据系统科技有限公司 | Forest fire detection method based on air remote sensing |
| CN104732244B (en) * | 2015-04-15 | 2018-01-23 | 大连交通大学 | The Classifying Method in Remote Sensing Image integrated based on wavelet transformation, how tactful PSO and SVM |
| CN104834920A (en) * | 2015-05-25 | 2015-08-12 | 成都通甲优博科技有限责任公司 | Intelligent forest fire recognition method and device based on multispectral image of unmanned plane |
| ITUA20162954A1 (en) * | 2016-04-28 | 2017-10-28 | St Microelectronics Srl | SEMICONDUCTOR DEVICE FOR ULTRAVIOLET AND INFRARED RADIATION DETECTION AND ITS MANUFACTURING METHOD |
| CN108564760A (en) * | 2018-06-06 | 2018-09-21 | 广西防城港核电有限公司 | Fire detection device under nuclear power station extreme environmental conditions and detection method |
| CN108921215A (en) * | 2018-06-29 | 2018-11-30 | 重庆邮电大学 | A kind of Smoke Detection based on local extremum Symbiotic Model and energy spectrometer |
-
2019
- 2019-02-28 CN CN201910149529.2A patent/CN109919071B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3301640A1 (en) * | 2016-09-30 | 2018-04-04 | JENOPTIK Advanced Systems GmbH | Method and thermal camera for the contactless measurement of a temperature or for observing quickly moving ir scenes with a thermal ir imager |
| CN107798811A (en) * | 2017-10-26 | 2018-03-13 | 上海腾盛智能安全科技股份有限公司 | A kind of tunnel fire monitoring device, monitoring system and monitoring method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109919071A (en) | 2019-06-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919071B (en) | Flame identification method based on infrared multi-feature combined technology | |
| CN107609470B (en) | Method for detecting early smoke of field fire by video | |
| CN107705288B (en) | Infrared video detection method for dangerous gas leakage under strong interference of pseudo-target motion | |
| CN104303193B (en) | Target classification based on cluster | |
| CN102819735B (en) | Flame detection method based on video frame image | |
| WO2018023916A1 (en) | Shadow removing method for color image and application | |
| CN101944267B (en) | Smoke and fire detection device based on videos | |
| CN102567722B (en) | Early-stage smoke detection method based on codebook model and multiple features | |
| CN106600625A (en) | Image processing method and device for detecting small-sized living thing | |
| CN110874592B (en) | Forest fire smoke image detection method based on total bounded variation | |
| CN101493980A (en) | Rapid video flame detection method based on multi-characteristic fusion | |
| CN108038867A (en) | Fire defector and localization method based on multiple features fusion and stereoscopic vision | |
| CN105046218B (en) | A kind of multiple features traffic video smog detection method based on serial parallel processing | |
| CN104853151A (en) | Large-space fire monitoring system based on video image | |
| CN111951250B (en) | Fire detection method based on image | |
| CN103903020B (en) | A kind of fire image recognition methods and device based on CodeBook | |
| CN108090397A (en) | Pedestrian detecting system based on infrared image | |
| CN110503637B (en) | Road crack automatic detection method based on convolutional neural network | |
| CN109711256A (en) | A kind of low latitude complex background unmanned plane target detection method | |
| CN101930540A (en) | Video-based multi-feature fusion flame detecting device and method | |
| CN109727202A (en) | Gas infrared video Enhancement Method, device and storage medium | |
| CN112132870B (en) | Early smoke detection method for forest fire | |
| CN114677406A (en) | Method for identifying petroleum flame from video stream by using accumulated frame difference and color statistics | |
| CN112950589A (en) | Dark channel prior defogging algorithm of multi-scale convolution neural network | |
| CN110197465B (en) | Foggy image enhancement method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |