CN109724703A - Temperature correction method under complex scene based on pattern-recognition - Google Patents
Temperature correction method under complex scene based on pattern-recognition Download PDFInfo
- Publication number
- CN109724703A CN109724703A CN201811635832.5A CN201811635832A CN109724703A CN 109724703 A CN109724703 A CN 109724703A CN 201811635832 A CN201811635832 A CN 201811635832A CN 109724703 A CN109724703 A CN 109724703A
- Authority
- CN
- China
- Prior art keywords
- target object
- scene
- infrared image
- temperature
- emissivity
- 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
Landscapes
- Radiation Pyrometers (AREA)
Abstract
The present invention proposes a kind of temperature correction method under the complex scene based on pattern-recognition, comprising: according to thermometric scene, statistics appears in all target objects under the thermometric scene;The corresponding emissivity of the target object is obtained, and is stored;Infrared image target object identification learning model is built, it can title and its emissivity corresponding to the different objects under automatic identification scene;According to infrared image target object identification learning model, the target object classification occurred under scene to be detected is identified;Obtain the corresponding emissivity of target object classification occurred under the scene to be detected;Its temperature is corrected according to the corresponding emissivity of target object classification occurred under the scene to be detected.It is can be realized under complex scene by means of the present invention, the temperature of the object is exported according to the emissivity real time correction of different objects.
Description
Technical field
The present invention relates to infrared temperature detection technique field, in particular under a kind of complex scene based on pattern-recognition
Temperature correction method.
Background technique
With the development of infrared technique, infrared measurement of temperature has a wide range of applications space at present, as a kind of contactless
Thermometry has many advantages, such as that temperature-measuring range is wide, accuracy is high, observation in real time can be achieved.But the precision of infrared measurement of temperature by
The influence of many extraneous factors, mainly has: object is at a distance from temperature measuring equipment, the emissivity on testee surface, extraneous ring
Border (atmospheric transmissivity, atmospheric emission rate, ambient temperature and ambient humidity) etc..Emissivity is the weight for describing object thermal radiation property
Parameter is wanted, the numerical value of material surface emissivity by virtue is absolute black body unit plane under the radiant power and same temperature of material unit area
The ratio between long-pending radiant power, degree of the characterization actual object radianting capacity close to black body radiation.In radiant thermometric technology, transmitting
Rate is to obtain the necessary known unique parameters of target temperature, and influence the bottleneck of current radiation temperature measurement accuracy.Therefore into
During row object temperature measures, it is necessary to which the emission ratio of hard objectives object measures if the emissivity of object is lower
The radiation temperature arrived will generate larger difference with actual temperature.Many sides have been proposed to the measurement of object emission rate at present
Method, the emissivity of common materials has had more accurately that measured value is as reference substantially in life, when fixation measuring is single
When object, the result of infrared measurement of temperature can be corrected according to the known emissivity of the object.
Currently, infrared measurement of temperature be frequently used in target it is single in the case where, such as electric system application in, through survey
Examination, the emissivity of power equipment is generally 0.85-0.95, so the infrared radiation thermometer emissivity to electric power thermometric is generally fixed to
0.95.But infrared measurement of temperature is applied in complex scene, what which occurred is no longer fixed single object, works as thermometric
When having the target object of multiple types in scene, such as possible someone, animal, automobile, due to the slin emissivity of different materials
Difference can not set the emissivity suitable for all target objects of a standard to temp measuring system.
Summary of the invention
In view of this, the invention proposes a kind of temperature correction methods under complex scene based on pattern-recognition, mainly
By identifying the classification of the target object occurred under the scene, enough emissivity according to different objects, real time correction output should
The temperature of object improves the precision of infrared measurement of temperature under complex scene.
A kind of temperature correction method under complex scene based on pattern-recognition, comprising:
According to thermometric scene, statistics appears in all target objects under the thermometric scene;
The corresponding emissivity of the target object is obtained, and is stored;
The infrared image sample of the thermometric scene is collected, and is marked, handles, sorts out;
Target object feature in the infrared image sample is extracted, the convolutional Neural net based on spatial pyramid pond is built
The infrared image target object identification learning model of network carries out Classification and Identification to the target object in infrared image;
According to the infrared image target object identification learning model, the target object class occurred under scene to be detected is identified
Not;
Obtain the corresponding emissivity of target object classification occurred under the scene to be detected;
Its temperature is corrected according to the corresponding emissivity of target object classification occurred under the scene to be detected.
It is further, described to obtain the corresponding emissivity of the target object, specifically: by inquiry or according in scene
Target object measure.
Further, the infrared image sample of the thermometric scene is collected at the place, and is marked, is handled, sorting out,
The processing includes: the background information inhibited in infrared image, the detailed information of prominent target object.
Further, described to extract target object feature in the infrared image sample, it builds based on spatial pyramid pond
The infrared image target object identification learning model of the convolutional neural networks of change, specifically:
By in whole infrared image sample input convolutional neural networks, disposable feature extraction is carried out, characteristic spectrum is obtained;
Each candidate frame region is determined in characteristic spectrum;
Pyramid space pond is used to each candidate frame, extracts the feature vector of regular length;
Feature vector Classification and Identification is carried out using algorithm of support vector machine.
It is further, described that each candidate frame region is determined in characteristic spectrum, specifically: it is real by following transformational relation
It is existing:
(x, y)=(S × x ', S × y ');Wherein (x ', y ') indicates the coordinate points on characteristic spectrum, and (x, y) indicates input
Whole infrared image sample on point, S is all not long products in convolutional neural networks.
Further, according to the corresponding emissivity of target object classification occurred under the scene to be detected to its temperature into
Row correction, specifically:
The true temperature calculation formula of target object is obtained according to infrared radiation temperature law are as follows:
Wherein ε is the actual transmission rate of target object, and ε ' is the emissivity of infrared radiation thermometer setting, and being set as 1, T1 is mesh
The true temperature of object is marked, T2 is the temperature of infrared radiation thermometer measurement.
It is an advantage of the present invention that the target object classification that a certain complex scene occurs is identified by deep learning, and
It is positioned, different emissivity is corresponded to according to different classifications, temperature correction is carried out to it.SPP-net first is by space gold word
The thought of tower is added in convolutional neural networks, realizes the multiple dimensioned input of data, solves convolutional neural networks in data
The problem of loss of data or distortion are caused when pretreatment;And SPP-net only obtains an original image feature extraction of progress whole
The characteristic spectrum of picture saves a large amount of calculating time.The present invention selects SPP-net to carry out the identification of target object and determine
Position, precision and calculating speed as a result have all obtained a degree of guarantee.Secondly, based on each mesh occurred under real-time scene
Object is marked, accurate temperature output is carried out there is also many difficulties by infrared measurement of temperature, because of the precision of infrared measurement of temperature and very much
Factor has relationship, and the present invention accurately finds out the target object classification occurred in scene and position by pattern-recognition, according to target
Classification carry out emissivity correction, to output more accurate temperature.
The present invention proposes a kind of temperature correction method under the complex scene based on pattern-recognition, comprising: according to thermometric field
Scape, statistics appear in all target objects under the thermometric scene;The corresponding emissivity of the target object is obtained, is gone forward side by side
Row storage;Build infrared image target object identification learning model, can different objects institute under automatic identification scene it is right
The title and its emissivity answered;According to infrared image target object identification learning model, the mesh occurred under scene to be detected is identified
Mark object category;Obtain the corresponding emissivity of target object classification occurred under the scene to be detected;According to described to be detected
The corresponding emissivity of target object classification occurred under scene is corrected its temperature.It can be realized by means of the present invention
Under complex scene, the temperature of the object is exported according to the emissivity real time correction of different objects.
Detailed description of the invention
It, below will be to embodiment or the prior art in order to illustrate more clearly of the present invention or technical solution in the prior art
Attached drawing needed in description is briefly described, it should be apparent that, the accompanying drawings in the following description is only in the present invention
The some embodiments recorded for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the temperature correction embodiment of the method flow chart under a kind of complex scene based on pattern-recognition of the present invention;
Fig. 2 is that the present invention is based on the infrared image target object identification learnings of the convolutional neural networks in spatial pyramid pond
Model builds schematic diagram.
Specific embodiment
Technical solution in embodiment in order to enable those skilled in the art to better understand the present invention, and make of the invention
Above objects, features, and advantages can be more obvious and easy to understand, makees with reference to the accompanying drawing to technical solution in the present invention further detailed
Thin explanation.
The present invention proposes a kind of temperature correction method under the complex scene based on pattern-recognition, as shown in Figure 1, comprising:
S101: according to thermometric scene, statistics appears in all target objects under the thermometric scene;
S102: the corresponding emissivity of the target object is obtained, and is stored;
Above-mentioned steps are mainly to obtain the emissivity of each target object;
S103: the infrared image sample of the thermometric scene is collected, and is marked, handles, sorts out;
S104: target object feature in the infrared image sample is extracted, the convolution based on spatial pyramid pond is built
The infrared image target object identification learning model of neural network carries out Classification and Identification to the target object in infrared image;
S105: according to the infrared image target object identification learning model, the target occurred under scene to be detected is identified
Object category;
Above-mentioned steps are built mainly for infrared image target object identification learning model, extract target object feature into
Row study makes it possible to the different objects under automatic identification scene and corresponds to classification and corresponding emissivity;
S106: the corresponding emissivity of target object classification occurred under the scene to be detected is obtained;
S107: school is carried out to its temperature according to the corresponding emissivity of target object classification occurred under the scene to be detected
Just.
It is further, described to obtain the corresponding emissivity of the target object, specifically: by inquiry or according in scene
Target object measure.
It is main that method is measured according to the target object in scene are as follows:
The emissivity of infrared radiation thermometer or thermal infrared imager is adjusted to 1;Measured object is remained into temperature constant state;Use standard
Contact type thermometric indicator measure current standard value temperature;Current measured value temperature is measured with infrared radiation thermometer or thermal infrared imager
Degree (notices that test point is as consistent as possible);Bring two groups of data into formula: emissivity=Shi Cezhi standard value calculate
Emissivity out.
Further, the infrared image sample of the thermometric scene is collected at the place, and is marked, is handled, sorting out,
The processing includes: the background information inhibited in infrared image, the detailed information of prominent target object.
Sample is mainly in the training process divided into training sample, test sample and verification sample by the classification to sample
This, for the training to model, improves its accuracy rate.
Further, described to extract target object feature in the infrared image sample, it builds based on spatial pyramid pond
The infrared image target object identification learning model of the convolutional neural networks (SPP-net) of change, specifically:
By in whole infrared image sample input convolutional neural networks, disposable feature extraction is carried out, characteristic spectrum is obtained;
Each candidate frame region is determined in characteristic spectrum;
Pyramid space pond is used to each candidate frame, extracts the feature vector of regular length;
Feature vector Classification and Identification is carried out using algorithm of support vector machine.
Fig. 2 is the infrared image target object identification learning model of the convolutional neural networks based on spatial pyramid pond
Build flow diagram.
The reason of selecting SPP-net: the thought of spatial pyramid is added in convolutional neural networks by SPP-net first,
The multiple dimensioned input for realizing data solves convolutional neural networks and causes loss of data or distortion in data prediction
Problem;Secondly SPP-net only carries out the extraction of a characteristic spectrum to original image, saves a large amount of calculating time.
Since traditional Target Recognition Algorithms are to carry out convolution to the multiple regions of picture to obtain multiple features, rolling up
The location information in corresponding original image region is obtained in long-pending process, but SSP-net is by carrying out once-through operation to picture
The characteristic spectrum of whole picture is extracted, therefore we need to find corresponding region on original image;
Therefore further, it is described that each candidate frame region is determined in characteristic spectrum, specifically: pass through following transformational relation
It realizes:
(x, y)=(S × x ', S × y ');Wherein (x ', y ') indicates the coordinate points on characteristic spectrum, and (x, y) indicates input
Whole infrared image sample on point, S is all not long products in convolutional neural networks.
Further, according to the corresponding emissivity of target object classification occurred under the scene to be detected to its temperature into
Row correction, specifically:
The true temperature calculation formula of target object is obtained according to infrared radiation temperature law are as follows:
Wherein ε is the actual transmission rate of target object, and ε ' is the emissivity of infrared radiation thermometer setting, due to target object kind
Class number is greater than one, therefore the emissivity of infrared radiation thermometer is set as the true temperature that 1, T1 is target object, and T2 is infrared
The temperature of temperature measurer measurement.
It is an advantage of the present invention that the target object classification that a certain complex scene occurs is identified by deep learning, and
It is positioned, different emissivity is corresponded to according to different classifications, temperature correction is carried out to it.SPP-net first is by space gold word
The thought of tower is added in convolutional neural networks, realizes the multiple dimensioned input of data, solves convolutional neural networks in data
The problem of loss of data or distortion are caused when pretreatment;And SPP-net only obtains an original image feature extraction of progress whole
The characteristic spectrum of picture saves a large amount of calculating time.The present invention selects SPP-net to carry out the identification of target object and determine
Position, precision and calculating speed as a result have all obtained a degree of guarantee.Secondly, based on each mesh occurred under real-time scene
Object is marked, accurate temperature output is carried out there is also many difficulties by infrared measurement of temperature, because of the precision of infrared measurement of temperature and very much
Factor has relationship, and the present invention accurately finds out the target object classification occurred in scene and position by pattern-recognition, according to target
Classification carry out emissivity correction, to output more accurate temperature.
The present invention proposes a kind of temperature correction method under the complex scene based on pattern-recognition, comprising: according to thermometric field
Scape, statistics appear in all target objects under the thermometric scene;The corresponding emissivity of the target object is obtained, is gone forward side by side
Row storage;Build infrared image target object identification learning model, can different objects institute under automatic identification scene it is right
The title and its emissivity answered;According to infrared image target object identification learning model, the mesh occurred under scene to be detected is identified
Mark object category;Obtain the corresponding emissivity of target object classification occurred under the scene to be detected;According to described to be detected
The corresponding emissivity of target object classification occurred under scene is corrected its temperature.It can be realized by means of the present invention
Under complex scene, the temperature of the object is exported according to the emissivity real time correction of different objects.
Although depicting the present invention by embodiment, it will be appreciated by the skilled addressee that the present invention there are many deformation and
Variation is without departing from spirit of the invention, it is desirable to which the attached claims include these deformations and change without departing from of the invention
Spirit.
Claims (6)
1. a kind of temperature correction method under complex scene based on pattern-recognition characterized by comprising
According to thermometric scene, statistics appears in all target objects under the thermometric scene;
The corresponding emissivity of the target object is obtained, and is stored;
The infrared image sample of the thermometric scene is collected, and is marked, handles, sorts out;
Target object feature in the infrared image sample is extracted, the convolutional neural networks based on spatial pyramid pond are built
Infrared image target object identification learning model carries out Classification and Identification to the target object in infrared image;
According to the infrared image target object identification learning model, the target object classification occurred under scene to be detected is identified;
Obtain the corresponding emissivity of target object classification occurred under the scene to be detected;
Its temperature is corrected according to the corresponding emissivity of target object classification occurred under the scene to be detected.
2. the method as described in claim 1, which is characterized in that it is described to obtain the corresponding emissivity of the target object, specifically
Are as follows: it is measured by inquiry or according to the target object in scene.
3. the method as described in claim 1, which is characterized in that the infrared image sample of the thermometric scene is collected at the place,
And in being marked, handle, sorting out, the processing includes: the background information inhibited in infrared image, protrudes the thin of target object
Save information.
4. the method as described in claim 1, which is characterized in that described to extract target object spy in the infrared image sample
Sign, builds the infrared image target object identification learning model of the convolutional neural networks based on spatial pyramid pond, specifically:
By in whole infrared image sample input convolutional neural networks, disposable feature extraction is carried out, characteristic spectrum is obtained;
Each candidate frame region is determined in characteristic spectrum;
Pyramid space pond is used to each candidate frame, extracts the feature vector of regular length;
Feature vector Classification and Identification is carried out using algorithm of support vector machine.
5. method as claimed in claim 4, which is characterized in that it is described that each candidate frame region is determined in characteristic spectrum, specifically
Are as follows: it is realized by following transformational relation:
(x, y)=(S × x ', S × y ');Wherein (x ', y ') indicates the coordinate points on characteristic spectrum, and (x, y) indicates the whole of input
The point on infrared image sample is opened, S is all not long products in convolutional neural networks.
6. the method as described in claim 1, which is characterized in that according to the target object classification occurred under the scene to be detected
Corresponding emissivity is corrected its temperature, specifically:
The true temperature calculation formula of target object is obtained according to infrared radiation temperature law are as follows:
Wherein ε is the actual transmission rate of target object, and ε ' is the emissivity of infrared radiation thermometer setting, and being set as 1, T1 is object
The true temperature of body, T2 are the temperature of infrared radiation thermometer measurement.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811635832.5A CN109724703A (en) | 2018-12-29 | 2018-12-29 | Temperature correction method under complex scene based on pattern-recognition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811635832.5A CN109724703A (en) | 2018-12-29 | 2018-12-29 | Temperature correction method under complex scene based on pattern-recognition |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109724703A true CN109724703A (en) | 2019-05-07 |
Family
ID=66296731
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811635832.5A Pending CN109724703A (en) | 2018-12-29 | 2018-12-29 | Temperature correction method under complex scene based on pattern-recognition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109724703A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110455413A (en) * | 2019-07-09 | 2019-11-15 | 中国科学院西安光学精密机械研究所 | A kind of medium-and-large-sized livestock farm body-temperature monitoring device and monitoring method |
CN111412992A (en) * | 2020-04-03 | 2020-07-14 | 华为技术有限公司 | Method for measuring temperature and electronic equipment |
CN112710400A (en) * | 2020-12-23 | 2021-04-27 | Oppo(重庆)智能科技有限公司 | Temperature measuring method and device, storage medium and electronic equipment |
CN113405671A (en) * | 2021-06-18 | 2021-09-17 | 成都信息工程大学 | Infrared earth surface temperature observation system and observation method thereof |
CN113610695A (en) * | 2021-05-07 | 2021-11-05 | 浙江兆晟科技股份有限公司 | Infrared telescope full-frame imaging output method and system |
CN114001825A (en) * | 2020-07-14 | 2022-02-01 | 华为技术有限公司 | Body temperature testing method, electronic device and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105917354A (en) * | 2014-10-09 | 2016-08-31 | 微软技术许可有限责任公司 | Spatial pyramid pooling networks for image processing |
CN107798336A (en) * | 2017-09-18 | 2018-03-13 | 广东电网有限责任公司东莞供电局 | A kind of infrared measurement of temperature image component recognition methods |
-
2018
- 2018-12-29 CN CN201811635832.5A patent/CN109724703A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105917354A (en) * | 2014-10-09 | 2016-08-31 | 微软技术许可有限责任公司 | Spatial pyramid pooling networks for image processing |
CN107798336A (en) * | 2017-09-18 | 2018-03-13 | 广东电网有限责任公司东莞供电局 | A kind of infrared measurement of temperature image component recognition methods |
Non-Patent Citations (3)
Title |
---|
姬彪: ""红外测温原理及其影响因素"", 《全国第十四届红外加热暨红外医学发展研讨会论文及论文摘要集》 * |
李东阳: ""基于多尺度的深度卷积网络的场景识别"", 《万方学位论文》 * |
薛生虎 等: ""辐射测温在钢铁工业中的应用及发射率对测量的影响"", 《计量学报》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110455413A (en) * | 2019-07-09 | 2019-11-15 | 中国科学院西安光学精密机械研究所 | A kind of medium-and-large-sized livestock farm body-temperature monitoring device and monitoring method |
CN111412992A (en) * | 2020-04-03 | 2020-07-14 | 华为技术有限公司 | Method for measuring temperature and electronic equipment |
CN114001825A (en) * | 2020-07-14 | 2022-02-01 | 华为技术有限公司 | Body temperature testing method, electronic device and storage medium |
CN112710400A (en) * | 2020-12-23 | 2021-04-27 | Oppo(重庆)智能科技有限公司 | Temperature measuring method and device, storage medium and electronic equipment |
CN113610695A (en) * | 2021-05-07 | 2021-11-05 | 浙江兆晟科技股份有限公司 | Infrared telescope full-frame imaging output method and system |
CN113405671A (en) * | 2021-06-18 | 2021-09-17 | 成都信息工程大学 | Infrared earth surface temperature observation system and observation method thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109724703A (en) | Temperature correction method under complex scene based on pattern-recognition | |
KR102578769B1 (en) | Thermographic image processing using neural networks to identify corrosion under insulation (CUI) | |
CN103196564B (en) | Infrared thermal imaging temperature measuring method by correcting surface emissivity through image segmentation | |
CN103439342B (en) | Based on the Infrared Non-destructive Testing method of thermal map temporal aspect | |
Xia et al. | A hybrid method based on extreme learning machine and k-nearest neighbor for cloud classification of ground-based visible cloud image | |
CN102800096B (en) | Robustness estimation algorithm of camera parameter | |
CN105352998B (en) | The independent element number of impulse eddy current thermal-induced imagery determines method | |
CN105509899A (en) | Method and system for infrared image temperature measurement calibration | |
CN109798981A (en) | Temperature determining method, temperature measuring equipment, storage medium | |
EP0456412B1 (en) | Temperature measuring method using infrared sensors and processor | |
CN116879297B (en) | Soil moisture collaborative inversion method, device, equipment and medium | |
CN113255181A (en) | Heat transfer inverse problem identification method and device based on deep learning | |
Wang et al. | An energy-efficient classification system for peach ripeness using YOLOv4 and flexible piezoelectric sensor | |
Han et al. | Defect detection: Defect classification and localization for additive manufacturing using deep learning method | |
CN110160661A (en) | A kind of measurement method and device of the body surface temperature based on visible radiograph | |
Chen et al. | An intelligent vision recognition method based on deep learning for pointer meters | |
CN113706478A (en) | Visual measurement method for forging temperature of nuclear forging | |
CN116385243B (en) | Intelligent garbage classification collection and transportation management method and system | |
CN112053341A (en) | Surface temperature measurement method of alloy material based on light reflectivity and artificial intelligence | |
CN116878669A (en) | Temperature compensation method based on short wave infrared temperature measurement, fire monitoring method and system | |
CN113553708B (en) | Method and device for tracing key influence factors of simulation model | |
CN115082697A (en) | Bimetallic strip temperature measurement method based on OpenCV image edge recognition | |
CN110111277B (en) | Planar thermal image repairing method and device | |
CN110895701B (en) | Forest fire online identification method and device based on CN and FHOG | |
Li et al. | A quantitative analysis method for pipeline defect inspection based on infrared thermal imaging |
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: 20190507 |