CN111783564A - Method for rapidly detecting wearing safety of respiratory tract protection equipment - Google Patents
Method for rapidly detecting wearing safety of respiratory tract protection equipment Download PDFInfo
- Publication number
- CN111783564A CN111783564A CN202010542666.5A CN202010542666A CN111783564A CN 111783564 A CN111783564 A CN 111783564A CN 202010542666 A CN202010542666 A CN 202010542666A CN 111783564 A CN111783564 A CN 111783564A
- Authority
- CN
- China
- Prior art keywords
- training
- model
- images
- rapidly detecting
- network
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/162—Detection; Localisation; Normalisation using pixel segmentation or colour matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/143—Sensing or illuminating at different wavelengths
Abstract
The invention provides a method for rapidly detecting the wearing safety of respiratory tract protective equipment, which comprises the following steps: acquiring original data of a human face wearing respiratory protection equipment, preprocessing the original data and dividing an image into three data sets, namely a training set, a testing set and a verification set; constructing a deep learning model and setting model parameters; training the deep learning model by using the image of the training set, and stopping training when the network converges; and calling the trained model, and applying the model in an actual detection scene. The detection method can be used without using refrigeration type infrared thermal imager high-precision infrared thermal imaging hardware equipment, and can be used for collecting face data by using a common Fiier infrared thermal imager, through innovation of an algorithm, whether respiratory tract protection equipment leaks can be subjected to express nondestructive detection, leakage points can also be pointed out clearly, the detection accuracy is high up to 90%, mAP reaches 85%, the system response is fast, and the robustness is good.
Description
Technical Field
The invention belongs to the technical field of air tightness detection of protective equipment, and particularly relates to a method for rapidly detecting wearing safety of respiratory tract protective equipment.
Background
All workers entering a biological safety detection laboratory (BSL-1/2/3) need certain protective measures, wherein respiratory tract protective equipment is an important ring, the higher the laboratory is, the higher the requirement on the tightness of the respiratory tract protective equipment worn by the laboratory is, and the phenomenon of air leakage cannot occur after the workers wear the respiratory tract protective equipment. Based on this reason, some gauze mask fit degree tester began to appear in the market, and the principle of these instruments, the concentration that needs the inside and outside gas dust of detection gauze mask directly judges gauze mask fit effect basically, but this type of method can't accomplish to accomplish harmless, contactless, the short-term test to the respiratory protection device of wearer. In view of the above disadvantages, researchers at home and abroad research a detection method for detecting whether the mask leaks air by using an infrared thermal imaging temperature measurement technology, but only the usability of the method is mentioned under an ideal condition, and in practical project application, a plurality of limiting factors still exist,
1. in order to improve the judgment rate, if a high-precision infrared thermal imager can be the best choice, but the existing high-resolution infrared thermal imager, namely a refrigeration type infrared thermal imager, is immature in domestic technology, foreign products belong to products forbidden for the China trade, and excellent high-resolution refrigeration type infrared thermal imagers cannot be purchased from normal market ways in China;
2. the external environment temperature has certain influence on detection, because the infrared thermal imager detects whether the respiratory tract protection device is safely worn, the infrared thermal imager analyzes the temperature data of the protection device and the edge of the human face, and when the environment temperature is higher or the temperature is closer to the human body temperature, the temperature information is difficult to accurately extract and analyze;
3. leaked gas can be covered by the body temperature of a human body, and a detector with poor resolution ratio cannot easily detect temperature difference change, so that a leakage point cannot be detected;
4. the gas tends to form a thermal barrier across the entire face, thereby rendering the leak point undetectable.
Disclosure of Invention
The invention provides a method for rapidly detecting the wearing safety of respiratory tract protective equipment, which aims to solve the problem that the safety of the respiratory tract protective equipment cannot be effectively detected through hardware equipment at home at present.
The invention provides a method for rapidly detecting the wearing safety of respiratory tract protective equipment, which comprises the following steps:
s1, collecting original data of the respiratory protection equipment worn by the human face, wherein the original data comprise images in upper edge leakage, lower edge leakage, left side leakage, right side leakage and normal leakage-free states, preprocessing the images, and dividing the images into three data sets, namely a training set, a testing set and a verification set;
s2, constructing a deep learning model and setting model parameters;
s3, training the deep learning model by using the image of the training set, and stopping training when the network converges;
and S4, calling the trained model and applying the model in an actual detection scene.
Preferably, the raw data in step S1 is acquired by using a feier infrared thermal imager.
Preferably, the preprocessing of the image in step S1 refers to manually deleting the blurred or obstructed image of the respiratory tract protection equipment and classifying the remaining images.
Preferably, the number of images in the training set, the test set and the verification set in step S1 is according to 7: 1.5: 1.5.
Preferably, the deep learning model selected in step S2 is a Resnet model + SSD model.
Preferably, the step S2 of setting the model parameters includes determining a learning rate, selecting an initial learning rate of 0.01, performing step iterative training, selecting an SGD + momentum optimizer, performing L2 regularization, and performing gradient centering.
Preferably, in step S3, the method for training the deep learning model using the images of the training set includes training whether respiratory tract protection equipment leaks using a Resnet network, where the size of the image input in the network is 299 × 299 pixels, after training, using the Resnet network as a reference network of the SSD network, then using the images for marking the leaking position to train the SSD network, and after training, integrating the networks.
The invention has the beneficial effects that: the detection method can be used without using refrigeration type infrared thermal imager high-precision infrared thermal imaging hardware equipment, and can be used for collecting face data by using a common Fiier infrared thermal imager, through innovation of an algorithm, whether respiratory tract protection equipment leaks can be quickly and nondestructively detected, leakage points can also be clearly pointed out, the detection leakage accuracy rate is up to 90%, mAP reaches 85%, the system response is fast, and the robustness is good.
Drawings
FIG. 1 is a flow chart of the detection method of the present invention;
FIG. 2 is an image to be detected according to an embodiment;
FIG. 3 is a diagram illustrating an output result of an image to be detected according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and should not be construed as limiting the scope of the invention.
Examples
Referring to fig. 1, the method for rapidly detecting the wearing safety of the respiratory tract protection equipment of the embodiment includes the following steps:
s1, collecting original data of the respiratory protection equipment worn by the human face, wherein the original data comprise images in upper edge leakage, lower edge leakage, left side leakage, right side leakage and normal leakage-free states, preprocessing the images, and dividing the images into three data sets, namely a training set, a testing set and a verification set;
specifically, this embodiment uses the feier infrared thermal imager to gather raw data, including normal data and various data that may reveal, will reveal and divide into four kinds of situations, reveal for the upper limb respectively, reveal on the lower limb, reveal on the left side, reveal on the right side, gather a large amount of images and gather a large amount of normal images simultaneously to these four kinds of circumstances of revealing.
Preprocessing the image refers to manually deleting the blurred or blocked image of the respiratory protection equipment and classifying the rest of the images. The elimination of those blurred or occluded images ensures that the image data is sharp and that subsequent classification is accurate.
Data in the sorted data set is scrambled, and then the data in the sorted data set is sorted according to the following sequence of 7: the training set, test set, and validation set for each class are divided by a ratio of 1.5: 1.5.
S2, constructing a deep learning model, wherein the deep learning model selected in the embodiment is a Resnet model + an SSD model, and setting model parameters;
during experiments, several common models are selected, namely a Resnet model, a Googlenet model and a Vgg model. And training the model by using the same data set and training strategy to obtain an accuracy result. And comprehensively considering the model parameters and the result accuracy, and finally selecting the Resnet model with the highest accuracy. The Resnet model has a deeper network layer number, so that data characteristics can be well extracted, and the Resnet network adopts an identity mapping method to solve the problems of gradient disappearance, gradient explosion and network degradation of a deep network.
Setting model parameters sequentially comprises:
1. determining a learning rate, selecting an initial learning rate of 0.01,
2. the training mode adopts step iterative training,
3. the image training parameters are migrated into the network as network initial parameters in a migration learning mode,
4. an SGD + momentum optimizer is selected for updating parameters,
5. an L2 regularization term is added to improve the network generalization capability. In the model training, the addition of an L2 regularization term can constrain, adjust or reduce the coefficient estimation towards the zero direction, and reduce the model complexity and the instability degree in the learning process, wherein the formula is as shown in formula (1):
in equation (1): p is equal to 2, wiRepresenting model parameters, in the model, applying L2 regularization by changing the loss function of the model, the loss function used by the model is a cross-entropy loss function, whose formula is as follows (2):
wherein xiWhich represents a label or a tag of the article,
representing the model predicted values. The Resnet model loss function after adding L2 regularization is as follows:
the lambda represents an adjustment factor which determines the punishment degree of the model complexity, and the model complexity is determined by a weight coefficient, so that the weight coefficient needs to be reduced to minimize the loss function, and the anti-interference capability of the model can be improved.
6. Gradient centering is adopted, and gradient reaches zero mean value by centering gradient vectors, so that the method is directly executed on the gradient.
S3, training the deep learning model by using the image of the training set, and stopping training when the network converges; the method for training the deep learning model by using the image of the training set comprises the steps of firstly training classified data by using a Resnet network, using the image size input in the network as 299 x 299 pixels, after training is finished, using the Resnet network as a reference network of an SSD network, then training the SSD network by using the image marked with a leakage position, using a training mode used in the process as iterative training, then selecting an SGD + momentum optimizer to update parameters, and after training is finished, integrating the network.
And S4, calling the trained model and applying the model in an actual detection scene. Fig. 2 shows an image to be detected of the input deep learning model, fig. 3 shows a detection result, the result is displayed as air leakage on the left side, the air leakage position is marked, and the position of a small square frame in the image is a leakage point.
Claims (7)
1. A method for rapidly detecting the wearing safety of respiratory tract protective equipment is characterized by comprising the following steps:
s1, collecting original data of the respiratory protection equipment worn by the human face, wherein the original data comprise images in upper edge leakage, lower edge leakage, left side leakage, right side leakage and normal leakage-free states, preprocessing the images, and dividing the images into three data sets, namely a training set, a testing set and a verification set;
s2, constructing a deep learning model and setting model parameters;
s3, training the deep learning model by using the image of the training set, and stopping training when the network converges;
and S4, calling the trained model and applying the model in an actual detection scene.
2. The method for rapidly detecting the wearing safety of respiratory protective equipment according to claim 1, wherein: the raw data of step S1 is collected using a feier infrared thermal imager.
3. The method for rapidly detecting the wearing safety of respiratory protective equipment according to claim 1, wherein: the preprocessing of the image in step S1 refers to manually deleting the blurred or blocked image of the respiratory tract protection equipment, and classifying the remaining images.
4. The method for rapidly detecting the wearing safety of respiratory protective equipment according to claim 1, wherein: the number of images in the training set, the test set, and the verification set in step S1 is according to 7: 1.5: 1.5.
5. The method for rapidly detecting the wearing safety of respiratory protective equipment according to claim 1, wherein: the deep learning model selected in step S2 is the Resnet model + SSD model.
6. The method for rapidly detecting the wearing safety of respiratory protection equipment according to claim 5, wherein the method comprises the following steps: step S2, setting model parameters includes determining learning rate, selecting 0.01 initial learning rate, adopting step iterative training, selecting SGD + momentum optimizer, adopting L2 regularization, and adopting gradient centralization.
7. The method for rapidly detecting the wearing safety of the respiratory tract protective equipment according to claim 6, wherein the method for training the deep learning model by using the images of the training set is to use a Resnet network to train whether the respiratory tract protective equipment leaks, the size of the images input into the network is 299 x 299 pixels, after the training is completed, the Resnet network is used as a reference network of the SSD network, then the SSD network is trained by using the images for marking the leakage positions, and after the training is completed, the networks are integrated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010542666.5A CN111783564A (en) | 2020-06-15 | 2020-06-15 | Method for rapidly detecting wearing safety of respiratory tract protection equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010542666.5A CN111783564A (en) | 2020-06-15 | 2020-06-15 | Method for rapidly detecting wearing safety of respiratory tract protection equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111783564A true CN111783564A (en) | 2020-10-16 |
Family
ID=72756636
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010542666.5A Pending CN111783564A (en) | 2020-06-15 | 2020-06-15 | Method for rapidly detecting wearing safety of respiratory tract protection equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111783564A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112101328A (en) * | 2020-11-19 | 2020-12-18 | 四川新网银行股份有限公司 | Method for identifying and processing label noise in deep learning |
| CN114091131A (en) * | 2021-12-08 | 2022-02-25 | 中国矿业大学 | Optimal scheme decision system for cloth mask support design based on cloud platform |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020195107A1 (en) * | 2001-05-18 | 2002-12-26 | Smaldone Gerald C. | Face masks for use in pressurized drug delivery systems |
| US20120314080A1 (en) * | 2011-06-10 | 2012-12-13 | Lee Yeu Yong | Gas leakage detecting system and method |
| CN103575475A (en) * | 2012-08-09 | 2014-02-12 | 北汽福田汽车股份有限公司 | Vehicle sealing performance detecting device and method |
| JP2014181930A (en) * | 2013-03-18 | 2014-09-29 | Sumco Corp | Human head type air flow testing device, and method for performing quality inspection of dust-proof implement using the device |
| KR20150145951A (en) * | 2014-06-20 | 2015-12-31 | (주)야긴스텍 | Sensing system for gas leakage |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| DE102016226152A1 (en) * | 2016-12-23 | 2018-06-28 | Volkswagen Aktiengesellschaft | Method and system for leak testing a container |
| CN110243542A (en) * | 2019-05-31 | 2019-09-17 | 国网上海市电力公司 | A kind of SF6Gas leaks imaging detection method and detection device |
| CN110487487A (en) * | 2019-09-25 | 2019-11-22 | 云南电网有限责任公司电力科学研究院 | A kind of transformer respiratory system leak hunting method based on infrared imagery technique |
| CN110728223A (en) * | 2019-10-08 | 2020-01-24 | 济南东朔微电子有限公司 | Helmet wearing identification method based on deep learning |
| CN111062429A (en) * | 2019-12-12 | 2020-04-24 | 上海点泽智能科技有限公司 | Chef cap and mask wearing detection method based on deep learning |
| CN111094956A (en) * | 2017-09-22 | 2020-05-01 | 沙特阿拉伯石油公司 | Processing the thermographic image with a neural network to identify Corrosion Under Insulation (CUI) |
-
2020
- 2020-06-15 CN CN202010542666.5A patent/CN111783564A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020195107A1 (en) * | 2001-05-18 | 2002-12-26 | Smaldone Gerald C. | Face masks for use in pressurized drug delivery systems |
| US20120314080A1 (en) * | 2011-06-10 | 2012-12-13 | Lee Yeu Yong | Gas leakage detecting system and method |
| CN103575475A (en) * | 2012-08-09 | 2014-02-12 | 北汽福田汽车股份有限公司 | Vehicle sealing performance detecting device and method |
| JP2014181930A (en) * | 2013-03-18 | 2014-09-29 | Sumco Corp | Human head type air flow testing device, and method for performing quality inspection of dust-proof implement using the device |
| KR20150145951A (en) * | 2014-06-20 | 2015-12-31 | (주)야긴스텍 | Sensing system for gas leakage |
| DE102016226152A1 (en) * | 2016-12-23 | 2018-06-28 | Volkswagen Aktiengesellschaft | Method and system for leak testing a container |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| CN111094956A (en) * | 2017-09-22 | 2020-05-01 | 沙特阿拉伯石油公司 | Processing the thermographic image with a neural network to identify Corrosion Under Insulation (CUI) |
| CN110243542A (en) * | 2019-05-31 | 2019-09-17 | 国网上海市电力公司 | A kind of SF6Gas leaks imaging detection method and detection device |
| CN110487487A (en) * | 2019-09-25 | 2019-11-22 | 云南电网有限责任公司电力科学研究院 | A kind of transformer respiratory system leak hunting method based on infrared imagery technique |
| CN110728223A (en) * | 2019-10-08 | 2020-01-24 | 济南东朔微电子有限公司 | Helmet wearing identification method based on deep learning |
| CN111062429A (en) * | 2019-12-12 | 2020-04-24 | 上海点泽智能科技有限公司 | Chef cap and mask wearing detection method based on deep learning |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112101328A (en) * | 2020-11-19 | 2020-12-18 | 四川新网银行股份有限公司 | Method for identifying and processing label noise in deep learning |
| CN114091131A (en) * | 2021-12-08 | 2022-02-25 | 中国矿业大学 | Optimal scheme decision system for cloth mask support design based on cloud platform |
| CN114091131B (en) * | 2021-12-08 | 2023-09-29 | 中国矿业大学 | Cloud platform-based cloth mask support design optimal scheme decision system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108898047B (en) | Pedestrian detection method and system based on blocking and shielding perception | |
| CN109800648A (en) | Face datection recognition methods and device based on the correction of face key point | |
| CN111860160B (en) | Method for detecting wearing of mask indoors | |
| CN110458059A (en) | A kind of gesture identification method based on computer vision and identification device | |
| CN108961675A (en) | Fall detection method based on convolutional neural networks | |
| CN111783564A (en) | Method for rapidly detecting wearing safety of respiratory tract protection equipment | |
| CN108229390A (en) | Rapid pedestrian detection method based on deep learning | |
| CN106875380B (en) | A kind of heterogeneous image change detection method based on unsupervised deep neural network | |
| CN107121140B (en) | A kind of location acquiring method based on Multiple Source Sensor | |
| CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
| CN108564085A (en) | A kind of method of automatic reading pointer type instrument reading | |
| CN108986140A (en) | Target scale adaptive tracking method based on correlation filtering and color detection | |
| CN104134059B (en) | Keep the bad image detecting method under the mixing deformation model of colouring information | |
| CN106530271B (en) | A kind of infrared image conspicuousness detection method | |
| CN107451999A (en) | foreign matter detecting method and device based on image recognition | |
| CN105975925B (en) | Partial occlusion pedestrian detection method based on joint-detection model | |
| CN110232379A (en) | A kind of vehicle attitude detection method and system | |
| CN107392885A (en) | A kind of method for detecting infrared puniness target of view-based access control model contrast mechanism | |
| CN108492298A (en) | Based on the multispectral image change detecting method for generating confrontation network | |
| CN106156758B (en) | A kind of tidal saltmarsh method in SAR seashore image | |
| CN110503623A (en) | A method of Bird's Nest defect on the identification transmission line of electricity based on convolutional neural networks | |
| CN109635634A (en) | A kind of pedestrian based on stochastic linear interpolation identifies data enhancement methods again | |
| CN110378232A (en) | The examination hall examinee position rapid detection method of improved SSD dual network | |
| CN107016694A (en) | A kind of SF based on infrared video6Gas Leakage Detection method | |
| CN110334760A (en) | A kind of optical component damage detecting method and system based on resUnet |
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 | ||
| CB02 | Change of applicant information |
Address after: No.1 XueYue Road, Chengyang District, Qingdao City, Shandong Province 266108 Applicant after: Qingdao Zhongrui Intelligent Instrument Co.,Ltd. Address before: 266000 No. 1 Xueyue Road, Chengyang District, Qingdao City, Shandong Province Applicant before: QINGDAO ZHONGRUI INTELLIGENT INSTRUMENT Co.,Ltd. |
|
| CB02 | Change of applicant information |