CN115100456A - Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide - Google Patents

Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide Download PDF

Info

Publication number
CN115100456A
CN115100456A CN202210587131.9A CN202210587131A CN115100456A CN 115100456 A CN115100456 A CN 115100456A CN 202210587131 A CN202210587131 A CN 202210587131A CN 115100456 A CN115100456 A CN 115100456A
Authority
CN
China
Prior art keywords
feature
matrix
feature vector
hydrogen peroxide
neural 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.)
Granted
Application number
CN202210587131.9A
Other languages
Chinese (zh)
Other versions
CN115100456B (en
Inventor
黄斌斌
罗霜
林金华
丘贵龙
袁瑞明
赖志林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fujian Tianfu Electronic Materials Co ltd
Original Assignee
Fujian Tianfu Electronic Materials Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Fujian Tianfu Electronic Materials Co ltd filed Critical Fujian Tianfu Electronic Materials Co ltd
Priority to CN202210587131.9A priority Critical patent/CN115100456B/en
Priority to PCT/CN2022/116515 priority patent/WO2023226228A1/en
Publication of CN115100456A publication Critical patent/CN115100456A/en
Application granted granted Critical
Publication of CN115100456B publication Critical patent/CN115100456B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • 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
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
    • G08B21/02—Alarms for ensuring the safety of persons
    • G08B21/12—Alarms for ensuring the safety of persons responsive to undesired emission of substances, e.g. pollution alarms
    • G08B21/14—Toxic gas alarms

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Multimedia (AREA)
  • Toxicology (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Environmental & Geological Engineering (AREA)
  • Business, Economics & Management (AREA)
  • Emergency Management (AREA)
  • Emergency Alarm Devices (AREA)

Abstract

The application relates to the field of intelligent detection of electronic grade gas, and particularly discloses an intelligent toxic and harmful gas alarm system and an alarm method for preparing electronic grade hydrogen peroxide, wherein a convolutional neural network model is used for extracting high-dimensional implicit associated features between hydrogen peroxide concentrations at various positions in a place to be monitored and topological features of a plurality of gas chromatographs so as to generate a feature vector and a feature matrix, in order to fuse feature information of the two, the spatial correlation of hydrogen peroxide distribution is enhanced based on the topological features so as to improve the classification accuracy, further a class feature vector is generated through class representation of spatial mapping of the feature vector to the feature matrix so as to realize simultaneous projection of the feature vector to a high-dimensional feature space of the feature matrix under a specific label class probability, and the alignment of the feature distribution of the feature vector in the high-dimensional feature space is improved, and then sensitivity and accuracy of early warning classification are improved.

Description

Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide
Technical Field
The invention relates to the field of intelligent detection of electronic-grade gas, in particular to an intelligent toxic and harmful gas alarm system and an alarm method for preparing electronic-grade hydrogen peroxide.
Background
The hydrogen peroxide (H2O 2) is produced by an electrolysis method, an isopropanol method and an anthraquinone method, and the anthraquinone method has general development in recent years due to the simple and easily obtained raw materials and the great energy consumption saving. However, various dangerous and harmful factors exist in the production process of the anthraquinone dioxide method, and very serious consequences can be caused once an accident occurs.
Hydrogen peroxide has a certain toxicity, which is mainly caused by its active oxidation, such as chemical burns to eyes, mucous membranes and skin, and fires to ordinary clothes. The hydrogen peroxide can be inhaled through respiratory tract, absorbed by skin contact and swallowed to cause poisoning. However, the vapor pressure of the device is small, and the volatility is low, so that whether the content of hydrogen peroxide gas in the space exceeds a preset standard or not is difficult to accurately monitor by a conventional monitoring means to generate intelligent early warning, and the safety of related personnel is ensured.
Therefore, an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide is expected.
In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like.
In addition, deep learning and neural networks also exhibit a level close to or even exceeding that of humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like.
In recent years, deep learning and development of a neural network provide a new idea and scheme for warning toxic and harmful gases in the preparation process of electronic-grade hydrogen peroxide.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems.
The embodiment of the application provides an intelligent toxic and harmful gas alarm system and an alarm method thereof for preparing electronic-grade hydrogen peroxide, wherein a convolutional neural network model is used for extracting high-dimensional implicit correlation characteristics among hydrogen peroxide concentrations at various positions in a place to be monitored and topological characteristics of a plurality of gas chromatographs so as to generate a characteristic vector and a characteristic matrix, in order to fuse characteristic information of the high-dimensional implicit correlation characteristics and the topological characteristics of the hydrogen peroxide concentrations at various positions in the place to be monitored, the spatial correlation of hydrogen peroxide distribution is strengthened based on the topological characteristics so as to improve the classification accuracy, component characteristic vectors are further generated through class characteristics of space mapping of the characteristic vectors to the characteristic matrix so as to realize the simultaneous projection of the characteristic vectors to the high-dimensional characteristic space of the characteristic matrix under a specific label class probability, and the alignment of the dimensions of the characteristic vectors in the high-dimensional characteristic space is improved, and then improve the sensitivity and accuracy of early warning classification.
According to one aspect of the application, an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide is provided, which comprises: an air data acquisition unit for acquiring a plurality of gas chromatograms acquired by a plurality of gas chromatographs disposed at a plurality of locations of a site to be monitored; a spatial encoding unit, configured to pass each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first feature maps; the channel correlation coding unit is used for cascading the plurality of first feature maps along the channel dimension and then obtaining a first feature vector through a second convolution neural network using a three-dimensional convolution kernel; a topology information extraction unit, configured to pass topology matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first feature matrix, where a feature value of each position at a non-diagonal position in the topology matrices is a distance between two corresponding gas chromatographs, and a feature value of each position at a diagonal position in the topology matrices is zero; a feature fusion unit, configured to fuse the first feature vector and the first feature matrix to obtain a classified feature vector based on a class characterization of a spatial mapping of the first feature vector to the first feature matrix, where the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a second norm of a feature vector obtained by dividing a feature vector obtained by multiplying the first feature matrix by the first feature vector; and the alarm result generating unit is used for enabling the classified characteristic vectors to pass through a classifier to obtain a classification result, and the classification result is used for representing whether an alarm prompt is generated or not.
In the above-mentioned intelligent poisonous and harmful gas alarm system for preparation of electronic grade hydrogen peroxide, the spatial coding unit is further configured to: each layer of the first convolutional neural network performs input data in forward transmission of the layer respectively: performing convolution processing on the input data based on a two-dimensional convolution kernel to generate a convolution feature map; pooling the convolved feature map to generate a pooled feature map; performing activation processing on the pooled feature map to generate an activated feature map; calculating the quotient of the eigenvalue mean of the eigenvalue matrix corresponding to each channel in the activation characteristic diagram and the sum of the eigenvalue mean of the eigenvalue matrix corresponding to all channels as the weighting coefficient of the eigenvalue matrix corresponding to each channel; and weighting the feature matrix of each channel by using the weighting coefficient of each channel in the activation feature map to generate the first feature map.
In the above intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide, the channel association coding unit is further configured to: processing the feature maps after the plurality of first feature maps are cascaded along the channel dimension by using a second convolutional neural network of the three-dimensional convolutional kernel according to the following formula to generate the first feature vector; wherein the formula is:
Figure 701330DEST_PATH_IMAGE001
wherein the content of the first and second substances,
Figure 758279DEST_PATH_IMAGE002
、
Figure 682373DEST_PATH_IMAGE003
and
Figure 690518DEST_PATH_IMAGE004
respectively representing the length, width and height of the three-dimensional convolution kernel,mis shown as
Figure 600836DEST_PATH_IMAGE005
The number of the layer characteristic maps is,
Figure 595513DEST_PATH_IMAGE006
is and
Figure 97033DEST_PATH_IMAGE005
first of the layermA convolution kernel connected with the feature map,
Figure 61315DEST_PATH_IMAGE007
in order to be biased,
Figure 899958DEST_PATH_IMAGE008
representing the activation function.
In the above intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide, the topology information extraction unit is further configured to: performing convolution processing, mean pooling along channel dimensions, and activation processing on input data in forward pass of layers using layers of the third convolutional neural network to generate the first feature matrix from a last layer of the third convolutional neural network, wherein an input of the first layer of the third convolutional neural network is a topology matrix of the plurality of gas chromatographs.
In the above intelligent toxic and harmful gas alarm system for preparing electronic grade hydrogen peroxide, the feature fusion unit is further configured to: based on the class characterization of the spatial mapping of the first feature vector to the first feature matrix, fusing the first feature vector and the first feature matrix to obtain the classified feature vector in the following formula; wherein the formula is:
Figure 400341DEST_PATH_IMAGE009
wherein, the first and the second end of the pipe are connected with each other,
Figure 167702DEST_PATH_IMAGE010
representing the first feature vector in a first set of features,
Figure 651904DEST_PATH_IMAGE011
representing the first feature matrix in a first order,
Figure 668139DEST_PATH_IMAGE012
representing the two-norm of the feature vector.
In the above-mentioned intelligent poisonous and harmful gas alarm system for preparation of electronic grade hydrogen peroxide, the alarm result generating unit is further configured to: processing the classification feature vector using the classifier to obtain the classification result with the following formula:
Figure 554187DEST_PATH_IMAGE013
wherein, in the step (A),
Figure 633395DEST_PATH_IMAGE014
to
Figure 463948DEST_PATH_IMAGE015
In order to be a weight matrix, the weight matrix,
Figure 519759DEST_PATH_IMAGE016
to
Figure 758849DEST_PATH_IMAGE017
In order to be a vector of the offset,
Figure 366547DEST_PATH_IMAGE018
the classified feature vector is obtained.
According to another aspect of the application, an alarm method of an intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation comprises the following steps: acquiring a plurality of gas chromatograms acquired by a plurality of gas chromatographs deployed at a plurality of positions of a site to be monitored; passing each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first feature maps; cascading the plurality of first feature maps along a channel dimension and then obtaining a first feature vector through a second convolutional neural network using a three-dimensional convolutional kernel; enabling the topological matrixes of the plurality of gas chromatographs to pass through a third convolutional neural network to obtain a first characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix are distances between the two corresponding gas chromatographs, and the characteristic values of the positions on diagonal positions in the topological matrix are zero; based on class characterization of spatial mapping of the first feature vector to the first feature matrix, fusing the first feature vector and the first feature matrix to obtain a classified feature vector, wherein the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a second norm of a feature vector obtained by dividing the feature vector obtained by multiplying the first feature matrix by the first feature vector; and enabling the classified feature vectors to pass through a classifier to obtain a classification result, wherein the classification result is used for representing whether an alarm prompt is generated or not.
In the above alarm method for an intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation, the step of passing each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first characteristic maps includes: each layer of the first convolutional neural network respectively performs the following operations on input data in the forward transmission of the layer: performing convolution processing on the input data based on a two-dimensional convolution kernel to generate a convolution feature map; pooling the convolved feature map to generate a pooled feature map; performing activation processing on the pooled feature map to generate an activated feature map; calculating the quotient of the eigenvalue mean of the eigenvalue matrix corresponding to each channel in the activation characteristic diagram and the sum of the eigenvalue mean of the eigenvalue matrix corresponding to all channels as the weighting coefficient of the eigenvalue matrix corresponding to each channel; and weighting the feature matrix of each channel by using the weighting coefficient of each channel in the activation feature map to generate the first feature map.
In the above alarm method for an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide, after cascading the plurality of first feature maps along a channel dimension, a first feature vector is obtained by using a second convolutional neural network of a three-dimensional convolutional kernel, and the method includes: processing the feature maps after the plurality of first feature maps are cascaded along the channel dimension by using a second convolutional neural network of the three-dimensional convolutional kernel according to the following formula to generate the first feature vector; wherein the formula is:
Figure 530069DEST_PATH_IMAGE019
wherein the content of the first and second substances,
Figure 920730DEST_PATH_IMAGE020
、
Figure 374845DEST_PATH_IMAGE021
and
Figure 527347DEST_PATH_IMAGE022
respectively representing the length, width and height of the three-dimensional convolution kernel,mis shown as
Figure 473437DEST_PATH_IMAGE023
The number of the layer characteristic maps is,
Figure 649814DEST_PATH_IMAGE024
is and
Figure 630540DEST_PATH_IMAGE023
first of a layermA convolution kernel connected to each of the feature maps,
Figure 219522DEST_PATH_IMAGE025
in order to be biased,
Figure 511963DEST_PATH_IMAGE026
representing an activation function.
In the above alarm method for an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide, the step of passing the topological matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first feature matrix includes: performing convolution processing, mean pooling along channel dimensions, and activation processing on input data in a forward pass of layers using layers of the third convolutional neural network to generate the first feature matrix from a last layer of the third convolutional neural network, wherein an input of the first layer of the third convolutional neural network is a topological matrix of the plurality of gas chromatographs.
In the above alarm method for an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide, based on the class representation of the spatial mapping from the first eigenvector to the first characteristic matrix, fusing the first eigenvector and the first characteristic matrix to obtain a classified eigenvector, including: based on the class characterization of the spatial mapping of the first feature vector to the first feature matrix, fusing the first feature vector and the first feature matrix to obtain the classified feature vector in the following formula; wherein the formula is:
Figure 683575DEST_PATH_IMAGE009
wherein the content of the first and second substances,
Figure 784386DEST_PATH_IMAGE010
representing the first feature vector in a first set of features,
Figure 544269DEST_PATH_IMAGE011
representing the first feature matrix in a first order,
Figure 58427DEST_PATH_IMAGE027
representing the two-norm of the feature vector.
In the above alarm method for an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide, the step of passing the classification feature vector through a classifier to obtain a classification result includes: processing the classification feature vector using the classifier to obtain the classification result with a formula:
Figure 797844DEST_PATH_IMAGE028
wherein, in the step (A),
Figure 723468DEST_PATH_IMAGE029
to
Figure 686876DEST_PATH_IMAGE030
In order to be a weight matrix, the weight matrix,
Figure 796652DEST_PATH_IMAGE031
to
Figure 841225DEST_PATH_IMAGE032
In order to be a vector of the offset,
Figure 323153DEST_PATH_IMAGE033
the classified feature vector is obtained.
Compared with the prior art, the intelligent toxic and harmful gas alarm system and the alarm method thereof for preparing electronic-grade hydrogen peroxide, provided by the application, use a convolutional neural network model to extract high-dimensional implicit correlation characteristics between hydrogen peroxide concentrations at various positions in a place to be monitored and topological characteristics of a plurality of gas chromatographs so as to generate a characteristic vector and a characteristic matrix, strengthen the spatial correlation of hydrogen peroxide distribution based on the topological characteristics in order to fuse characteristic information of the high-dimensional implicit correlation characteristics and improve classification accuracy, further generate component characteristic vectors through class characteristics of spatial mapping of the characteristic vectors to the characteristic matrix so as to realize simultaneous projection of the characteristic vectors to a high-dimensional characteristic space of the characteristic matrix under a specific label class probability and improve the alignment of the dimensions of the characteristic distribution of the characteristic vectors in the high-dimensional characteristic space, and then sensitivity and accuracy of early warning classification are improved.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments thereof with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is an application scene diagram of an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide according to an embodiment of the application.
Fig. 2 is a block diagram of an intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to an embodiment of the application.
Fig. 3 is a flowchart of an alarm method of the intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to the embodiment of the application.
Fig. 4 is a schematic structural diagram of an alarm method of an intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are merely some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.
Overview of a scene
As mentioned above, the hydrogen peroxide (H2O 2) production includes electrolytic process, isopropyl alcohol process and anthraquinone process, and the anthraquinone process has been developed widely in recent years due to its simple and easily available raw materials and its large energy saving. However, various dangerous and harmful factors exist in the production process of the anthraquinone dioxide method, and very serious consequences can be caused once an accident occurs.
Hydrogen peroxide has a certain toxicity, which is mainly caused by its active oxidation, such as chemical burns to eyes, mucous membranes and skin, and fires of ordinary clothes. The hydrogen peroxide can be inhaled through respiratory tract, absorbed by skin contact and swallowed to cause poisoning. However, the vapor pressure of the device is small, and the volatility is low, so that whether the content of hydrogen peroxide gas in the space exceeds a preset standard or not is difficult to accurately monitor by a conventional monitoring means to generate intelligent early warning, and the safety of related personnel is ensured.
Therefore, an intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide is expected.
In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. In addition, deep learning and neural networks also exhibit a level close to or even exceeding that of humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like.
In recent years, deep learning and development of a neural network provide a new idea and scheme for warning toxic and harmful gases in the preparation process of electronic-grade hydrogen peroxide.
The convolutional neural network has excellent performance in the aspect of extracting high-dimensional features of images, so that the convolutional neural network can be used for extracting high-dimensional hidden features in a gas chromatogram of air of a place to be monitored, and a classifier is used for classifying to determine whether an alarm signal needs to be generated.
However, the vapor pressure of hydrogen peroxide is small, the volatility is low, and the distribution of the volatilized hydrogen peroxide at each position in the to-be-monitored place is uneven, so that in order to improve the sensitivity and accuracy of the alarm, in the technical scheme of the application, gas chromatographs are deployed at a plurality of positions in the to-be-monitored place so as to judge whether to carry out intelligent early warning or not based on the global air analysis of the to-be-monitored place.
Specifically, in the technical solution of the present application, a plurality of gas chromatograms acquired by a plurality of gas chromatographs deployed at a plurality of locations of a site to be monitored are first acquired.
It should be understood that the gas chromatogram can show the air components and the ratio of each component in the air, but due to the characteristics of small vapor pressure and low volatility of hydrogen peroxide, the hydrogen peroxide existing in the air is weak in image representation of the gas chromatogram.
Therefore, in the embodiment of the present application, a convolutional neural network model using a channel attention mechanism is adopted to focus on capturing a high-dimensional feature representation of hydrogen peroxide in the gas chromatogram to obtain first feature maps corresponding to respective positions.
The hydrogen peroxide gas has the characteristic of spatial distribution uniformity in a place to be monitored, namely the hydrogen peroxide concentration at each position is correlated, for example, the hydrogen peroxide concentration is high at a place close to hydrogen peroxide preparation equipment, and the hydrogen peroxide concentration is low at a place far away from the hydrogen peroxide preparation equipment.
Therefore, the first feature map of each location is further encoded using a convolutional neural network model of a three-dimensional convolutional kernel to extract high-dimensional implicit associations between hydrogen peroxide of each location in the site to be monitored to obtain a first feature vector. However, the spatial correlation of the hydrogen peroxide distribution is not fully utilized by only using the three-dimensional convolutional neural network to extract the high-dimensional implicit correlation. Therefore, the topology information of the plurality of gas chromatographs is further taken into consideration.
Specifically, a convolutional neural network model is used for coding topological matrixes of the plurality of gas chromatographs so as to extract spatial high-dimensional implicit association features among the plurality of gas chromatographs to obtain a first feature matrix, wherein feature values of positions on non-diagonal positions in the topological matrixes are distances between two corresponding gas chromatographs, and feature values of positions on diagonal positions in the topological matrixes are zero.
Furthermore, early warning classification can be performed by fusing the first feature vector and the first feature matrix to obtain a classification feature vector comprising spatial topology information of the hydrogen peroxide at each position and composition related information of the hydrogen peroxide at each position.
However, in the process of mapping the first feature vector into the feature space where the first feature matrix is located to obtain the classified feature vector, since the first feature vector is subjected to channel dimension cascade of the feature map based on the channel attention mechanism, joint correlation in channel dimension is generated in the process of performing cross-channel feature extraction by the three-dimensional convolution kernel, and the first feature matrix is subjected to global pooling in channel dimension, so that the dimension alignment needs to be ensured as much as possible when mapping is performed.
Specifically, the calculation process of the classification feature vector is as follows:
Figure 283894DEST_PATH_IMAGE009
Figure 382431DEST_PATH_IMAGE034
representing the two-norm of the feature vector.
That is, the classification feature vector realizes simultaneous projection of the feature vector to a high-dimensional feature space of the feature matrix under a specific label class probability through class representation of spatial mapping of the feature vector to the feature matrix, so as to be adapted to joint correlation of the feature vector in a channel dimension, and improve the alignment of the feature distribution of the feature vector in the dimension of the high-dimensional feature space. Therefore, the sensitivity and the accuracy of early warning classification are improved.
Based on this, this application has proposed an intelligent poisonous and harmful gas alarm system for preparation of electron level hydrogen peroxide, and it includes: an air data acquisition unit for acquiring a plurality of gas chromatograms acquired by a plurality of gas chromatographs disposed at a plurality of locations of a site to be monitored; a spatial encoding unit, configured to pass each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first feature maps; the channel correlation coding unit is used for cascading the plurality of first feature maps along the channel dimension and then obtaining a first feature vector through a second convolutional neural network using a three-dimensional convolutional kernel; a topology information extraction unit, configured to pass topology matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first feature matrix, where a feature value of each position at a non-diagonal position in the topology matrices is a distance between two corresponding gas chromatographs, and a feature value of each position at a diagonal position in the topology matrices is zero; a feature fusion unit, configured to fuse the first feature vector and the first feature matrix to obtain a classified feature vector based on a class characterization of a spatial mapping of the first feature vector to the first feature matrix, where the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a second norm of a feature vector obtained by dividing a feature vector obtained by multiplying the first feature matrix by the first feature vector; and the alarm result generating unit is used for enabling the classified characteristic vectors to pass through a classifier to obtain a classification result, and the classification result is used for representing whether an alarm prompt is generated or not.
Fig. 1 is a diagram illustrating an application scenario of an intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to an embodiment of the present application. As shown in fig. 1, in this application scenario, first, a plurality of gas chromatograms are acquired by a plurality of gas chromatographs (e.g., T as illustrated in fig. 1) deployed at a plurality of locations of a site to be monitored (e.g., P as illustrated in fig. 1).
Then, the obtained gas chromatograms are input into a server (for example, a server S as illustrated in fig. 1) deployed with an intelligent toxic and harmful gas alarm algorithm for electronic-grade hydrogen peroxide preparation, wherein the server can process the gas chromatograms by using the intelligent toxic and harmful gas alarm algorithm for electronic-grade hydrogen peroxide preparation to obtain a classification result for representing whether an alarm prompt is generated.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary System
Fig. 2 illustrates a block diagram of an intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide production according to an embodiment of the present application.
As shown in fig. 2, an intelligent toxic and harmful gas alarm system 200 for electronic-grade hydrogen peroxide preparation according to an embodiment of the present application includes: an air data acquisition unit 210 for acquiring a plurality of gas chromatograms acquired by a plurality of gas chromatographs disposed at a plurality of positions of a site to be monitored; a spatial encoding unit 220, configured to pass each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first feature maps; a channel correlation encoding unit 230, configured to cascade the plurality of first feature maps along a channel dimension, and obtain a first feature vector through a second convolutional neural network using a three-dimensional convolutional kernel; a topology information extraction unit 240, configured to pass the topology matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first feature matrix, where a feature value of each position at a non-diagonal position in the topology matrices is a distance between two corresponding gas chromatographs, and a feature value of each position at a diagonal position in the topology matrices is zero; a feature fusion unit 250, configured to fuse the first feature vector and the first feature matrix to obtain a classified feature vector based on a class characterization of a spatial mapping of the first feature vector to the first feature matrix, where the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a two-norm of a feature vector obtained by multiplying the first feature matrix by the first feature vector divided by a feature vector obtained by multiplying the first feature matrix by the first feature vector; and an alarm result generating unit 260, configured to pass the classification feature vector through a classifier to obtain a classification result, where the classification result is used to characterize whether an alarm prompt is generated.
Specifically, in the embodiment of the present application, the air data acquiring unit 210 and the spatial encoding unit 220 are configured to acquire a plurality of gas chromatograms acquired by a plurality of gas chromatographs deployed at a plurality of locations of a site to be monitored, and pass each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first feature maps.
As described above, since the convolutional neural network has excellent performance in extracting high-dimensional features of an image, the convolutional neural network can be used to extract high-dimensional implicit features in a gas chromatogram of air of the site to be monitored, and a classifier is used to perform classification to determine whether an alarm signal needs to be generated. However, considering that the vapor pressure of hydrogen peroxide is small, the volatility is low, and the distribution of the volatilized hydrogen peroxide at each position in the to-be-monitored place is uneven, in order to improve the sensitivity and accuracy of the alarm, in the technical scheme of the application, gas chromatographs are deployed at a plurality of positions in the to-be-monitored place so as to judge whether to perform intelligent early warning based on the global air analysis of the to-be-monitored place.
That is, specifically, in the technical solution of the present application, a plurality of gas chromatograms are first collected by a plurality of gas chromatographs disposed at a plurality of positions of the site to be monitored.
It should be understood that the gas chromatogram can show the air components and the ratio of each component in the air, but due to the characteristics of small vapor pressure and low volatility of hydrogen peroxide, the hydrogen peroxide existing in the air is weak in image representation of the gas chromatogram.
Therefore, in the technical solution of the present application, a convolutional neural network model of a channel attention mechanism is further adopted to focus on capturing a high-dimensional feature representation of hydrogen peroxide in the gas chromatogram to obtain first feature maps corresponding to the respective positions.
More specifically, in an embodiment of the present application, the spatial encoding unit is further configured to: each layer of the first convolutional neural network respectively performs the following operations on input data in the forward transmission of the layer: firstly, performing convolution processing on the input data based on a two-dimensional convolution kernel to generate a convolution characteristic diagram; then, performing pooling processing on the convolution feature map to generate a pooled feature map; then, activating the pooled feature map to generate an activated feature map; then, calculating the quotient of the eigenvalue mean value of the eigenvalue matrix corresponding to each channel in the activation characteristic diagram and the sum of the eigenvalue mean values of the eigenvalue matrices corresponding to all channels as the weighting coefficient of the eigenvalue matrix corresponding to each channel; and finally, weighting the feature matrix of each channel by using the weighting coefficient of each channel in the activation feature map to generate the first feature map.
Specifically, in this embodiment of the present application, the channel correlation encoding unit 230 is configured to obtain the first feature vector by using a second convolutional neural network of a three-dimensional convolutional kernel after cascading the plurality of first feature maps along the channel dimension.
It should be understood that, because the hydrogen peroxide gas has the characteristic of spatial distribution uniformity in the place to be monitored, that is, the hydrogen peroxide concentrations at various positions are correlated, for example, the hydrogen peroxide concentration is high at the place close to the hydrogen peroxide preparation equipment, and the hydrogen peroxide concentration is low at the place far away from the hydrogen peroxide preparation equipment. Therefore, in the technical solution of the present application, the first feature maps of the respective positions are further encoded by using a second convolutional neural network model of a three-dimensional convolutional kernel to extract a high-dimensional implicit association between hydrogen peroxide solutions of the respective positions in the to-be-monitored place to obtain a first feature vector.
More specifically, in an embodiment of the present application, the channel-associated encoding unit is further configured to: processing the feature maps after the plurality of first feature maps are cascaded along the channel dimension by using a second convolutional neural network of the three-dimensional convolutional kernel according to the following formula to generate the first feature vector;
wherein the formula is:
Figure 853863DEST_PATH_IMAGE035
wherein the content of the first and second substances,
Figure 476782DEST_PATH_IMAGE036
、
Figure 250834DEST_PATH_IMAGE037
and
Figure 459836DEST_PATH_IMAGE038
respectively representing the length, width and height of the three-dimensional convolution kernel,mis shown as
Figure 344747DEST_PATH_IMAGE039
The number of the layer characteristic maps is,
Figure 191480DEST_PATH_IMAGE040
is and
Figure 903477DEST_PATH_IMAGE039
first of a layermA convolution kernel connected with the feature map,
Figure 711027DEST_PATH_IMAGE041
in order to be biased,
Figure 22798DEST_PATH_IMAGE042
representing an activation function.
Specifically, in this embodiment of the present application, the topology information extraction unit 240 is configured to pass topology matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first feature matrix, where a feature value of each position at an off-diagonal position in the topology matrices is a distance between two corresponding gas chromatographs, and a feature value of each position at a diagonal position in the topology matrices is zero.
It should be appreciated that the spatial correlation of the hydrogen peroxide distribution is not fully exploited in view of the fact that only the three-dimensional convolutional neural network is used to extract the high-dimensional implicit associations between hydrogen peroxide at various locations in the site to be monitored. Therefore, in the present invention, the topology information of the plurality of gas chromatographs is further taken into consideration.
Specifically, the third convolutional neural network model is used for encoding the topological matrix of the plurality of gas chromatographs to extract spatial high-dimensional implicit correlation features among the plurality of gas chromatographs to obtain a first feature matrix, wherein feature values of positions on non-diagonal positions in the topological matrix are distances between two corresponding gas chromatographs, and feature values of positions on diagonal positions in the topological matrix are zero.
Accordingly, in one particular example, the layers of the third convolutional neural network are used to convolve input data in a forward pass of layers, a mean pooling along channel dimensions, and an activation process to generate the first feature matrix from a last layer of the third convolutional neural network, wherein an input to the first layer of the third convolutional neural network is a topology matrix of the plurality of gas chromatographs.
Specifically, in this embodiment, the feature fusion unit 250 is configured to fuse the first feature vector and the first feature matrix to obtain a classified feature vector based on a class characterization of a spatial mapping of the first feature vector to the first feature matrix, where the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a binorm of a feature vector obtained by dividing the feature vector obtained by multiplying the first feature matrix by the first feature vector by a feature vector obtained by multiplying the first feature matrix by the first feature vector.
It should be understood that, in the technical solution of the present application, further, the early warning classification may be performed by fusing the first feature vector and the first feature matrix to obtain a classification feature vector including spatial topology information of the hydrogen peroxide solution at each position and composition related information of the hydrogen peroxide solution at each position.
However, in the process of mapping the first feature vector into the feature space where the first feature matrix is located to obtain the classification feature vector, since the first feature vector passes through the channel dimension cascade of the feature map based on the channel attention mechanism, in the process of performing cross-channel feature extraction by the three-dimensional convolution kernel, joint correlation in the channel dimension is generated, and the first feature matrix passes through global pooling in the channel dimension, so that the dimension alignment needs to be ensured as much as possible when mapping is performed. Therefore, in the technical solution of the present application, the first feature vector and the first feature matrix are fused to obtain a classification feature vector further based on class characterization of spatial mapping of the first feature vector to the first feature matrix.
More specifically, in an embodiment of the present application, the feature fusion unit is further configured to: based on the class characterization of the spatial mapping of the first feature vector to the first feature matrix, fusing the first feature vector and the first feature matrix to obtain the classified feature vector in the following formula; wherein the formula is:
Figure 130562DEST_PATH_IMAGE009
wherein the content of the first and second substances,
Figure 747882DEST_PATH_IMAGE010
-representing the first feature vector by means of a first representation,
Figure 432941DEST_PATH_IMAGE011
a first feature matrix representing a first set of features,
Figure 659654DEST_PATH_IMAGE043
representing the two-norm of the feature vector. It should be understood that the classified feature vector realizes the simultaneous projection of the feature vector to the high-dimensional feature space of the feature matrix under the specific label class probability through the class characterization of the spatial mapping of the feature vector to the feature matrix, so as to be adapted to the joint correlation of the feature vector on the channel dimension, improve the alignment of the dimension of the feature distribution of the feature vector in the high-dimensional feature space, and further improve the sensitivity and accuracy of the early warning classification.
Specifically, in this embodiment of the present application, the alarm result generating unit 260 is configured to pass the classification feature vector through a classifier to obtain a classification result, where the classification result is used to characterize whether to generate an alarm prompt.
That is, in the technical solution of the present application, after obtaining the classification feature vector, the classification feature vector is further passed through a classifier to obtain a classification result for characterizing whether an alarm prompt is generated. Accordingly, in one specific example, the classification feature vector is processed using the classifier to obtain the classification result with the following formula:
Figure 979515DEST_PATH_IMAGE044
wherein, in the process,
Figure 266271DEST_PATH_IMAGE045
to
Figure 173047DEST_PATH_IMAGE046
In order to be a weight matrix, the weight matrix,
Figure 704916DEST_PATH_IMAGE047
to
Figure 380748DEST_PATH_IMAGE048
In order to be a vector of the offset,
Figure 336940DEST_PATH_IMAGE049
the classified feature vector is obtained.
In summary, the intelligent toxic and harmful gas alarm system 200 for preparing electronic-grade hydrogen peroxide according to the embodiment of the present application is illustrated, which uses a convolutional neural network model to extract high-dimensional implicit correlation features between hydrogen peroxide concentrations at various positions in a site to be monitored and topological features of a plurality of gas chromatographs to generate a feature vector and a feature matrix, and further generates component class feature vectors by class features of spatial mapping of the feature vectors to the feature matrix in order to fuse feature information of the two to strengthen spatial correlation of hydrogen peroxide distribution based on the topological features so as to improve classification accuracy, so as to realize simultaneous projection of the feature vectors to a high-dimensional feature space of the feature matrix under a specific label class probability, and improve alignment of dimensions of feature distribution of the feature vectors in the high-dimensional feature space, and then improve the sensitivity and accuracy of early warning classification.
As described above, the intelligent toxic and harmful gas alarm system 200 for electronic-grade hydrogen peroxide preparation according to the embodiment of the present application can be implemented in various terminal devices, such as a server of an intelligent toxic and harmful gas alarm algorithm for electronic-grade hydrogen peroxide preparation.
In one example, the intelligent toxic and harmful gas alarm system 200 for electronic-grade hydrogen peroxide preparation according to the embodiment of the present application can be integrated into a terminal device as a software module and/or a hardware module.
For example, the intelligent toxic and harmful gas alarm system 200 for preparing electronic-grade hydrogen peroxide can be a software module in an operating system of the terminal device, or can be an application program developed for the terminal device; of course, the intelligent toxic and harmful gas alarm system 200 for preparing electronic-grade hydrogen peroxide can also be one of numerous hardware modules of the terminal device.
Alternatively, in another example, the intelligent toxic and harmful gas alarm system 200 for electronic-grade hydrogen peroxide preparation and the terminal device can also be separate devices, and the intelligent toxic and harmful gas alarm system 200 for electronic-grade hydrogen peroxide preparation can be connected to the terminal device through a wired and/or wireless network and transmit interactive information according to an agreed data format.
Exemplary method
Fig. 3 illustrates a flow chart of an alarm method of the intelligent toxic and harmful gas alarm system for electronic grade hydrogen peroxide preparation.
As shown in fig. 3, the alarm method of the intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide according to the embodiment of the application includes the steps: s110, acquiring a plurality of gas chromatograms acquired through a plurality of gas chromatographs deployed at a plurality of positions of a place to be monitored; s120, enabling each gas chromatogram in the plurality of gas chromatograms to pass through a first convolution neural network using a channel attention mechanism to obtain a plurality of first feature maps; s130, cascading the plurality of first feature maps along the channel dimension, and then obtaining a first feature vector through a second convolution neural network using a three-dimensional convolution kernel; s140, passing the topological matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first characteristic matrix, wherein the characteristic value of each position on the non-diagonal position in the topological matrices is the distance between the two corresponding gas chromatographs, and the characteristic value of each position on the diagonal position in the topological matrices is zero; s150, fusing the first feature vector and the first feature matrix to obtain a classified feature vector based on class characterization of spatial mapping of the first feature vector to the first feature matrix, wherein the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a second norm of a feature vector obtained by dividing the feature vector obtained by multiplying the first feature matrix by the first feature vector; and S160, enabling the classified feature vectors to pass through a classifier to obtain a classification result, wherein the classification result is used for representing whether an alarm prompt is generated or not.
Fig. 4 is a schematic diagram illustrating an architecture of an alarm method of an intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to an embodiment of the present application.
As shown in fig. 4, in the network architecture of the alarm method of the intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation, first, each of the plurality of gas chromatograms (e.g., P1 as illustrated in fig. 4) is passed through a first convolutional neural network (e.g., CNN1 as illustrated in fig. 4) using a channel attention mechanism to obtain a plurality of first characteristic maps (e.g., F1 as illustrated in fig. 4); then, after cascading the plurality of first feature maps along the channel dimension, obtaining a first feature vector (e.g., VF1 as illustrated in fig. 4) through a second convolutional neural network (e.g., CNN2 as illustrated in fig. 4) using a three-dimensional convolution kernel; then, passing the topological matrices of the plurality of gas chromatographs (e.g., M as illustrated in fig. 4) through a third convolutional neural network (e.g., CNN3 as illustrated in fig. 4) to obtain a first feature matrix (e.g., MF1 as illustrated in fig. 4); then, based on class characterization of the spatial mapping of the first feature vector to the first feature matrix, fusing the first feature vector with the first feature matrix to obtain a classified feature vector (e.g., MF as illustrated in fig. 4); and, finally, passing the classified feature vector through a classifier (e.g., circle S as illustrated in fig. 4) to obtain a classification result, which is used to characterize whether an alarm prompt is generated.
More specifically, in steps S110 and S120, a plurality of gas chromatograms acquired by a plurality of gas chromatographs disposed at a plurality of locations of a site to be monitored are acquired, and each of the plurality of gas chromatograms is passed through a first convolution neural network using a channel attention mechanism to obtain a plurality of first feature maps.
It will be appreciated that since convolutional neural networks have excellent performance in extracting high-dimensional features of images, they can be used to extract high-dimensional implicit features in the gas chromatogram of the air of the site to be monitored, and a classifier is used to perform classification to determine whether an alarm signal needs to be generated.
However, considering that the vapor pressure of hydrogen peroxide is small, the volatility is low, and the distribution of the volatilized hydrogen peroxide at each position in the to-be-monitored place is uneven, in order to improve the sensitivity and accuracy of the alarm, in the technical scheme of the application, gas chromatographs are deployed at a plurality of positions in the to-be-monitored place so as to judge whether to perform intelligent early warning based on the global air analysis of the to-be-monitored place.
That is, specifically, in the technical solution of the present application, a plurality of gas chromatograms are first collected by a plurality of gas chromatographs disposed at a plurality of positions of the site to be monitored.
It should be understood that the gas chromatogram can show the air components and the ratio of each component in the air, but due to the characteristics of small vapor pressure and low volatility of hydrogen peroxide, the hydrogen peroxide existing in the air is weak in image representation of the gas chromatogram.
Therefore, in the technical solution of the present application, a convolutional neural network model of a channel attention mechanism is further adopted to focus on capturing a high-dimensional feature representation of hydrogen peroxide in the gas chromatogram to obtain first feature maps corresponding to the respective positions.
More specifically, in step S130, the plurality of first feature maps are cascaded along the channel dimension and then passed through a second convolutional neural network using a three-dimensional convolutional kernel to obtain a first feature vector.
It should be understood that, because the hydrogen peroxide gas has a characteristic of spatial distribution uniformity in the site to be monitored, that is, the hydrogen peroxide gas concentrations at various positions are correlated, for example, the hydrogen peroxide gas concentration at a place close to the hydrogen peroxide preparation equipment is high, and the hydrogen peroxide gas concentration at a place far away from the hydrogen peroxide preparation equipment is low.
Therefore, in the technical solution of the present application, the first feature maps of the respective positions are further encoded by using a second convolutional neural network model of a three-dimensional convolutional kernel to extract a high-dimensional implicit association between hydrogen peroxide solutions of the respective positions in the to-be-monitored place to obtain a first feature vector.
More specifically, in S140, the topology matrices of the plurality of gas chromatographs are passed through a third convolutional neural network to obtain a first feature matrix, in which the feature value of each position at an off-diagonal position in the topology matrices is the distance between the corresponding two gas chromatographs, and the feature value of each position at a diagonal position in the topology matrices is zero.
It should be appreciated that the spatial correlation of the hydrogen peroxide distribution is not fully exploited in view of the fact that only the three-dimensional convolutional neural network is used to extract the high-dimensional implicit associations between hydrogen peroxide at various locations in the site to be monitored. Therefore, in the present invention, the topology information of the plurality of gas chromatographs is further taken into consideration.
Specifically, the third convolutional neural network model is used for encoding the topological matrix of the plurality of gas chromatographs to extract spatial high-dimensional implicit correlation features among the plurality of gas chromatographs to obtain a first feature matrix, wherein feature values of positions on non-diagonal positions in the topological matrix are distances between two corresponding gas chromatographs, and feature values of positions on diagonal positions in the topological matrix are zero.
Accordingly, in one particular example, the layers of the third convolutional neural network are used to convolve input data in a forward pass of layers, a mean pooling along channel dimensions, and an activation process to generate the first feature matrix from a last layer of the third convolutional neural network, wherein an input to the first layer of the third convolutional neural network is a topology matrix of the plurality of gas chromatographs.
More specifically, in step S150, based on the class characterization of the spatial mapping of the first feature vector to the first feature matrix, the first feature vector and the first feature matrix are fused to obtain a classified feature vector, wherein the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a second norm of a feature vector obtained by dividing the feature vector obtained by multiplying the first feature matrix by the first feature vector.
It should be understood that, in the technical solution of the present application, further, the early warning classification may be performed by fusing the first feature vector and the first feature matrix to obtain a classification feature vector including spatial topology information of the hydrogen peroxide solution at each position and composition related information of the hydrogen peroxide solution at each position.
However, in the process of mapping the first feature vector into the feature space where the first feature matrix is located to obtain the classification feature vector, since the first feature vector passes through the channel dimension cascade of the feature map based on the channel attention mechanism, in the process of performing cross-channel feature extraction by the three-dimensional convolution kernel, joint correlation in the channel dimension is generated, and the first feature matrix passes through global pooling in the channel dimension, so that the dimension alignment needs to be ensured as much as possible when mapping is performed.
Therefore, in the technical solution of the present application, the first feature vector and the first feature matrix are fused to obtain a classification feature vector further based on class characterization of spatial mapping of the first feature vector to the first feature matrix.
More specifically, in step S160, the classified feature vector is passed through a classifier to obtain a classification result, and the classification result is used for characterizing whether an alarm prompt is generated or not.
That is, in the technical solution of the present application, after the classification feature vector is obtained, the classification feature vector is further passed through a classifier to obtain a classification result for characterizing whether an alarm prompt is generated. Accordingly, in one specific example, the classification feature vector is processed using the classifier to obtain the classification result with the following formula:
Figure 871958DEST_PATH_IMAGE050
wherein, in the step (A),
Figure 484816DEST_PATH_IMAGE051
to
Figure 952837DEST_PATH_IMAGE052
In order to be a weight matrix, the weight matrix,
Figure 79931DEST_PATH_IMAGE053
to
Figure 961299DEST_PATH_IMAGE048
In order to be a vector of the offset,
Figure 333506DEST_PATH_IMAGE033
the classified feature vector is obtained.
In summary, the alarm method for the intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide according to the embodiment of the present application is clarified, a convolutional neural network model is used to extract high-dimensional implicit correlation features between hydrogen peroxide concentrations at various positions in a site to be monitored and topological features of a plurality of gas chromatographs to generate an eigenvector and an eigenvector matrix, in order to fuse feature information of the two features and strengthen spatial correlation of hydrogen peroxide distribution based on the topological features so as to improve classification accuracy, a class eigenvector is further generated by class representation of spatial mapping of the eigenvector to the eigenvector matrix so as to realize simultaneous projection of the eigenvector to a high-dimensional eigenspace of the eigenvector matrix under a specific label class probability, and improve alignment of dimensions of the eigenvector in the high-dimensional eigenspace, and then sensitivity and accuracy of early warning classification are improved.
The foregoing describes the general principles of the present application in conjunction with specific embodiments thereof, however, it is noted that the advantages, effects, etc. mentioned in the present application are exemplary only and not limiting, and should not be considered essential to the various embodiments of the present application.
Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and not for the purpose of limitation, and the foregoing details are not to be construed as limiting the present application in any way as it will be readily apparent from the following description.
The block diagrams of devices, apparatuses, devices, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the devices, apparatuses, devices, systems, etc. must be connected, arranged, configured in the manner shown in the block diagrams.
These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by one skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "are used herein to mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".
It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations should be considered as equivalents of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application.
Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims (10)

1. An intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide is characterized by comprising: an air data acquisition unit for acquiring a plurality of gas chromatograms acquired by a plurality of gas chromatographs disposed at a plurality of locations of a site to be monitored; a spatial encoding unit, configured to pass each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first feature maps; the channel correlation coding unit is used for cascading the plurality of first feature maps along the channel dimension and then obtaining a first feature vector through a second convolutional neural network using a three-dimensional convolutional kernel; a topology information extraction unit, configured to pass topology matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first feature matrix, where a feature value of each position at a non-diagonal position in the topology matrices is a distance between two corresponding gas chromatographs, and a feature value of each position at a diagonal position in the topology matrices is zero; a feature fusion unit, configured to fuse the first feature vector and the first feature matrix to obtain a classified feature vector based on a class characterization of a spatial mapping of the first feature vector to the first feature matrix, where the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a second norm of a feature vector obtained by dividing a feature vector obtained by multiplying the first feature matrix by the first feature vector; and the alarm result generating unit is used for enabling the classified characteristic vectors to pass through a classifier to obtain a classification result, and the classification result is used for representing whether an alarm prompt is generated or not.
2. The intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 1, wherein the spatial encoding unit is further configured to: each layer of the first convolutional neural network performs input data in forward transmission of the layer respectively: performing convolution processing on the input data based on a two-dimensional convolution kernel to generate a convolution feature map; pooling the convolved feature map to generate a pooled feature map; performing activation processing on the pooled feature map to generate an activated feature map; calculating a quotient of the feature value mean value of the feature matrix corresponding to each channel in the activation feature map and the sum of the feature value mean values of the feature matrices corresponding to all channels as a weighting coefficient of the feature matrix corresponding to each channel; and weighting the feature matrix of each channel by using the weighting coefficient of each channel in the activation feature map to generate the first feature map.
3. The intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 2, wherein the channel association coding unit is further configured to: processing the feature maps after the plurality of first feature maps are cascaded along the channel dimension by using a second convolutional neural network of the three-dimensional convolutional kernel according to the following formula to generate the first feature vector;
wherein the formula is:
Figure 523497DEST_PATH_IMAGE001
wherein, the first and the second end of the pipe are connected with each other,
Figure 392227DEST_PATH_IMAGE002
、
Figure 349819DEST_PATH_IMAGE003
and
Figure 135766DEST_PATH_IMAGE004
respectively representing the length, width and height of the three-dimensional convolution kernel,mis shown as
Figure 33052DEST_PATH_IMAGE005
The number of the layer characteristic maps is,
Figure 666159DEST_PATH_IMAGE006
is and
Figure 517572DEST_PATH_IMAGE005
first of a layermA convolution kernel connected to each of the feature maps,
Figure 157807DEST_PATH_IMAGE007
in order to be offset,
Figure 145486DEST_PATH_IMAGE008
representing an activation function.
4. The intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 3, wherein the topology information extraction unit is further configured to: performing convolution processing, mean pooling along channel dimensions, and activation processing on input data in forward pass of layers using layers of the third convolutional neural network to generate the first feature matrix from a last layer of the third convolutional neural network, wherein an input of the first layer of the third convolutional neural network is a topology matrix of the plurality of gas chromatographs.
5. The intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 4, wherein the feature fusion unit is further configured to: based on the class characterization of the spatial mapping of the first feature vector to the first feature matrix, fusing the first feature vector and the first feature matrix to obtain the classified feature vector in the following formula;
wherein the formula is:
Figure 713608DEST_PATH_IMAGE009
wherein, in the step (A),
Figure 521159DEST_PATH_IMAGE010
representing the first feature vector in a first set of features,
Figure 334394DEST_PATH_IMAGE011
representing the first feature matrix in a first order,
Figure 943623DEST_PATH_IMAGE012
representing the two-norm of the feature vector.
6. The intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 5, wherein the alarm result generation unit is further configured to: processing the classification feature vector using the classifier to obtain the classification result with the following formula:
Figure 918532DEST_PATH_IMAGE013
wherein, in the step (A),
Figure 977493DEST_PATH_IMAGE014
to
Figure 735364DEST_PATH_IMAGE015
In order to be a weight matrix, the weight matrix,
Figure 556690DEST_PATH_IMAGE016
to
Figure 610490DEST_PATH_IMAGE017
In order to be a vector of the offset,
Figure 392632DEST_PATH_IMAGE018
the classified feature vector is obtained.
7. An alarm method of an intelligent toxic and harmful gas alarm system for preparing electronic grade hydrogen peroxide is characterized by comprising the following steps: acquiring a plurality of gas chromatograms acquired by a plurality of gas chromatographs deployed at a plurality of positions of a site to be monitored; passing each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first feature maps; cascading the plurality of first feature maps along a channel dimension and then obtaining a first feature vector through a second convolutional neural network using a three-dimensional convolutional kernel; enabling the topological matrixes of the plurality of gas chromatographs to pass through a third convolutional neural network to obtain a first characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix are distances between the two corresponding gas chromatographs, and characteristic values of positions on diagonal positions in the topological matrix are zero; based on class characterization of spatial mapping of the first feature vector to the first feature matrix, fusing the first feature vector and the first feature matrix to obtain a classified feature vector, wherein the class characterization of the spatial mapping of the first feature vector to the first feature matrix is generated based on a second norm of a feature vector obtained by dividing the feature vector obtained by multiplying the first feature matrix by the first feature vector; and enabling the classified feature vectors to pass through a classifier to obtain a classification result, wherein the classification result is used for representing whether an alarm prompt is generated or not.
8. The alarm method of the intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 7, wherein the step of passing each of the plurality of gas chromatograms through a first convolutional neural network using a channel attention mechanism to obtain a plurality of first characteristic maps comprises the steps of: each layer of the first convolutional neural network performs input data in forward transmission of the layer respectively: performing convolution processing on the input data based on a two-dimensional convolution kernel to generate a convolution feature map; pooling the convolution feature map to generate a pooled feature map; performing activation processing on the pooled feature map to generate an activated feature map; calculating the quotient of the eigenvalue mean of the eigenvalue matrix corresponding to each channel in the activation characteristic diagram and the sum of the eigenvalue mean of the eigenvalue matrix corresponding to all channels as the weighting coefficient of the eigenvalue matrix corresponding to each channel; and weighting the feature matrix of each channel by using the weighting coefficient of each channel in the activation feature map to generate the first feature map.
9. The alarm method of the intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 8, wherein the obtaining of the first feature vector by the second convolutional neural network using a three-dimensional convolutional kernel after the cascading of the plurality of first feature maps along the channel dimension comprises: processing the feature maps obtained by cascading the plurality of first feature maps along the channel dimension by using a second convolutional neural network of the three-dimensional convolutional kernel according to the following formula to generate a first feature vector; wherein the formula is:
Figure 187151DEST_PATH_IMAGE019
wherein the content of the first and second substances,
Figure 128562DEST_PATH_IMAGE020
、
Figure 87684DEST_PATH_IMAGE021
and
Figure 888281DEST_PATH_IMAGE022
respectively representing the length, width and height of the three-dimensional convolution kernel,mis shown as
Figure 220911DEST_PATH_IMAGE023
The number of the layer characteristic maps is,
Figure 282408DEST_PATH_IMAGE025
is and
Figure 176546DEST_PATH_IMAGE026
first of a layermA convolution kernel connected to each of the feature maps,
Figure DEST_PATH_IMAGE027
in order to be offset,
Figure 727805DEST_PATH_IMAGE028
representing an activation function.
10. The alarm method of the intelligent toxic and harmful gas alarm system for electronic-grade hydrogen peroxide preparation according to claim 9, wherein the passing of the topological matrices of the plurality of gas chromatographs through a third convolutional neural network to obtain a first feature matrix comprises: performing convolution processing, mean pooling along channel dimensions, and activation processing on input data in a forward pass of layers using layers of the third convolutional neural network to generate the first feature matrix from a last layer of the third convolutional neural network, wherein an input of the first layer of the third convolutional neural network is a topological matrix of the plurality of gas chromatographs.
CN202210587131.9A 2022-05-27 2022-05-27 Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide Active CN115100456B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202210587131.9A CN115100456B (en) 2022-05-27 2022-05-27 Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide
PCT/CN2022/116515 WO2023226228A1 (en) 2022-05-27 2022-09-01 Intelligent toxic and harmful gas alarm system for preparation of electronic grade hydrogen peroxide

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210587131.9A CN115100456B (en) 2022-05-27 2022-05-27 Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide

Publications (2)

Publication Number Publication Date
CN115100456A true CN115100456A (en) 2022-09-23
CN115100456B CN115100456B (en) 2023-02-03

Family

ID=83289731

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210587131.9A Active CN115100456B (en) 2022-05-27 2022-05-27 Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide

Country Status (2)

Country Link
CN (1) CN115100456B (en)
WO (1) WO2023226228A1 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115993268A (en) * 2023-03-22 2023-04-21 福建天甫电子材料有限公司 Sampling system and method for anhydrous hydrogen fluoride
CN116106457A (en) * 2023-04-13 2023-05-12 天津海河标测技术检测有限公司 Air sampling and detecting integrated device
CN116859830A (en) * 2023-03-27 2023-10-10 福建天甫电子材料有限公司 Production management control system for electronic grade ammonium fluoride production
CN117129584A (en) * 2023-01-16 2023-11-28 新疆广陆能源科技股份有限公司 Tail gas detection system and method for thermal fluid generator
WO2024098653A1 (en) * 2022-11-08 2024-05-16 福建省龙德新能源有限公司 Automated sampling and analysis system and method for preparation of lithium hexafluorophosphate

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111797911A (en) * 2020-06-22 2020-10-20 浙江大学 Image data multi-label classification method
CN113947739A (en) * 2021-10-15 2022-01-18 深圳市铠硕达科技有限公司 Community safety management monitoring system based on Internet of things and monitoring method thereof

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11544522B2 (en) * 2018-12-06 2023-01-03 University Of Tennessee Research Foundation Methods, systems, and computer readable mediums for determining a system state of a power system using a convolutional neural network
CN113837302B (en) * 2021-09-29 2022-07-26 深圳市铠硕达科技有限公司 City safety management monitoring method and system based on Internet of things and electronic equipment
CN114282003A (en) * 2021-11-30 2022-04-05 中国建设银行股份有限公司 Financial risk early warning method and device based on knowledge graph

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111797911A (en) * 2020-06-22 2020-10-20 浙江大学 Image data multi-label classification method
CN113947739A (en) * 2021-10-15 2022-01-18 深圳市铠硕达科技有限公司 Community safety management monitoring system based on Internet of things and monitoring method thereof

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024098653A1 (en) * 2022-11-08 2024-05-16 福建省龙德新能源有限公司 Automated sampling and analysis system and method for preparation of lithium hexafluorophosphate
CN117129584A (en) * 2023-01-16 2023-11-28 新疆广陆能源科技股份有限公司 Tail gas detection system and method for thermal fluid generator
CN117129584B (en) * 2023-01-16 2024-02-20 新疆广陆能源科技股份有限公司 Tail gas detection system and method for thermal fluid generator
CN115993268A (en) * 2023-03-22 2023-04-21 福建天甫电子材料有限公司 Sampling system and method for anhydrous hydrogen fluoride
CN116859830A (en) * 2023-03-27 2023-10-10 福建天甫电子材料有限公司 Production management control system for electronic grade ammonium fluoride production
CN116859830B (en) * 2023-03-27 2024-01-26 福建天甫电子材料有限公司 Production management control system for electronic grade ammonium fluoride production
CN116106457A (en) * 2023-04-13 2023-05-12 天津海河标测技术检测有限公司 Air sampling and detecting integrated device

Also Published As

Publication number Publication date
WO2023226228A1 (en) 2023-11-30
CN115100456B (en) 2023-02-03

Similar Documents

Publication Publication Date Title
CN115100456B (en) Intelligent toxic and harmful gas alarm system for preparing electronic-grade hydrogen peroxide
CN115356434A (en) Gas monitoring system and monitoring method for hexafluorobutadiene storage place
Yar et al. Vision sensor-based real-time fire detection in resource-constrained IoT environments
US20210248378A1 (en) Spatiotemporal action detection method
CN114782882B (en) Video target behavior anomaly detection method and system based on multi-modal feature fusion
CN115259089B (en) Production management control system for preparing electronic grade hydrofluoric acid and control method thereof
WO2024021253A1 (en) Intelligent toxic and harmful gas alarm system for preparing hexafluorobutadiene
CN115079572A (en) Energy management control system for preparing lithium hexafluorophosphate and control method thereof
CN115791640A (en) Tumor cell detection device and method based on spectroscopic spectrum
CN114870416A (en) Rectification control system and rectification control method for preparing electronic-grade monofluoromethane
CN112037279B (en) Article position identification method and device, storage medium and electronic equipment
CN114912533B (en) State monitoring system and monitoring method applied to transformer
CN115231525B (en) Intelligent separation and purification system for electronic-grade chlorine trifluoride
CN113989927B (en) Method and system for identifying violent behaviors of video group based on bone data
CN115082837B (en) Flow rate control system for filling purified water into PET bottle and control method thereof
CN113343950A (en) Video behavior identification method based on multi-feature fusion
CN114564993A (en) Fatigue detection system based on multi-source information fusion and operation method thereof
CN114745168A (en) Cloud platform inlet real-time flow copying method and system and electronic equipment
CN109002746B (en) 3D (three-dimensional) fire identification method and system
Du et al. Dynamic scene classification using redundant spatial scenelets
Zheng et al. Heteroscedastic sparse representation based classification for face recognition
Liu et al. Genetic Programming-Evolved Spatio-Temporal Descriptor for Human Action Recognition.
CN115902004B (en) Measurement device and measurement method for conductivity of degassed hydrogen
CN115131735A (en) Cooling medium temperature control system and temperature control method after hardware tool heat treatment
CN116269351A (en) Cerebral apoplexy patient rehabilitation effect evaluation system based on movement posture change

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