CN116842141B - Alarm smoke linkage based digital information studying and judging method - Google Patents
Alarm smoke linkage based digital information studying and judging method Download PDFInfo
- Publication number
- CN116842141B CN116842141B CN202311083689.4A CN202311083689A CN116842141B CN 116842141 B CN116842141 B CN 116842141B CN 202311083689 A CN202311083689 A CN 202311083689A CN 116842141 B CN116842141 B CN 116842141B
- Authority
- CN
- China
- Prior art keywords
- feature
- training
- text
- vector
- predicted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 239000000779 smoke Substances 0.000 title claims abstract description 36
- 238000012549 training Methods 0.000 claims abstract description 66
- 230000008569 process Effects 0.000 claims abstract description 20
- 239000013598 vector Substances 0.000 claims description 62
- 238000013528 artificial neural network Methods 0.000 claims description 26
- 230000007246 mechanism Effects 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 238000001514 detection method Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000011160 research Methods 0.000 description 17
- 241000208125 Nicotiana Species 0.000 description 11
- 235000002637 Nicotiana tabacum Nutrition 0.000 description 11
- 230000000694 effects Effects 0.000 description 10
- 230000006399 behavior Effects 0.000 description 7
- 238000004891 communication Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 229940036051 sojourn Drugs 0.000 description 2
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- 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
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06V10/764—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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a digital information studying and judging method based on smoke alarm linkage, which comprises the following steps: the pre-training is to describe images and characters in a database, encode and decode the images and characters to obtain corresponding pre-training characteristics, encode and decode the images and/or characters to be studied and judged in the recognition process to obtain corresponding characteristics, recognize the characteristics in the pre-training database, judge the information related to smoke, acquire the corresponding pre-training characteristics through pre-training, and judge the information related to smoke through the recognition process, so that a more accurate result is obtained.
Description
Technical Field
The invention mainly relates to a digital information research and judgment method, in particular to a digital information research and judgment method based on smoke alarm linkage.
Background
In the Internet era, online shopping gradually replaces offline transactions to become a mainstream consumption mode, and smoke-related staff focus on the wide Internet. In recent years, the transaction mode of the Internet and logistics delivery increases the difficulty of studying and judging the task conditions of smoke-related personnel, the communication relationship of the smoke-related personnel, the sojourn relationship of the smoke-related personnel, the loss of regulation of related personnel, the activity area of related personnel, the smoke-related vehicles, the transportation of the smoke-related vehicles, key bayonets and the like, and the traditional obstacle of tobacco monopoly management work is heavy.
Disclosure of Invention
The invention aims to solve the problem that the prior art cannot carry out research and judgment on the task conditions of smoke-related personnel, the communication relation of the smoke-related personnel, the sojourn relation of the smoke-related personnel, the smoke-related vehicles, the transportation of the smoke-related vehicles, key bayonets and the like, and provides a digital information research and judgment method based on smoke-related linkage.
In order to achieve the above purpose, the present invention adopts the following technical scheme: a method for studying and judging digital intelligence based on smoke linkage includes: the pre-training and recognition process is performed,
the pre-training is to describe the images and characters in the database, encode and decode the images and characters to obtain the corresponding pre-training characteristics, the recognition process obtains the corresponding characteristics by encoding and decoding the images and/or characters to be studied and judged, recognizes the characteristics in the pre-training database, judges the information related to smoke,
wherein the pre-training comprises the steps of:
step S100, given a task, performing word segmentation processing and attention on a text by adopting a multi-head attention mechanism and a gating circulation unit mechanism;
step S200, given a task, identifying an image by adopting an LSTM neural network, and comparing the acquired characteristics with the characteristics of a characteristic library;
step S300, associating the text and the image after the pre-training to form a text-image pair training set containing the characteristics.
The identification process comprises the following steps:
step S400, identifying the input text and/or image to be predicted, and obtaining corresponding characteristics;
step S500, comparing the feature to be predicted and the text-image pair containing the feature obtained by pre-training, and if the feature is within the threshold range, identifying and obtaining the task type.
Further, in step S100, the attention mechanism based on the deep neural network assists the network model in screening the information with the highest degree of association with the current stage task from the complex input information. The gating circulation unit can enable the neural network to not only memorize past information, but also selectively delete unimportant information. Meanwhile, in order to increase the accuracy of pre-training, a multi-head attention mechanism is adopted to conduct comparison learning on the same text information, and the situation of text mismatch is reduced.
The specific process of focusing attention on the text of the word segmentation by adopting the attention mechanism in the step S100 includes:
step S101, adopting an ebedding model pairTreating to obtain->Wherein->Representing a vector corresponding to a word or phrase of the text, < >>In order to distribute the attention of the person,Unumber of single and phrases in the text;
step S102, obtaining the query vector in the multi-graph attention mechanismKey vector->Sum vector->
Wherein,、/>、/>is a matrix of U×T->，，/>T is the number of network channels where the multi-head attention mechanism is located;
step S103, each query vectorMultiplying all key vectors to obtain corresponding score +.>；
Step S104, obtaining each query vector by using Softmax functionAttention probability distribution +.>I.e. weight coefficient
Step S105, for each query vectorCorresponding value vector +.>Weighted summation is carried out to obtain the attention output after comparison learning>
Step S105, the attention output obtained by all the query vectors is spliced to obtain contrast learningAttention to laterb
。
Further, in the step S100, the text of interest is also different for different tasks. For a smoke-related person studying and judging task, the concerned text vector comprises { personnel basic information, trip data, call data and express data }; for a smoke-related person communication relation research and judgment task, the concerned text vector comprises { the name, sex, age of a member, the interaction frequency between the member and the social network, and the relation density }; the smoke-related personnel living relation study and judgment task should pay attention to travel records, and the text vector of attention comprises { shift, carriage, adjacent seat, departure place, purpose, check in and check out }; for related personnel out-of-specification research and judgment tasks, the concerned text vector comprises { type of out-of-specification behavior, frequency, time, place, call, related personnel, environment }; for a specific activity research and judgment task of a smoke-related person, the concerned text vector comprises { person name, identity card number, location, travel information, contact person condition, travel history, communication record, social network behavior }; for a relevant personnel activity area research task, the concerned text vector comprises { the historical activity range of personnel, time node }; for the smoke-related vehicle research and judgment task, the text vector concerned comprises { vehicle basic information, bayonet data, high-speed data and time sequence data }; for the tobacco-related vehicle transportation research and judgment task, the text vector concerned comprises { high-speed toll station data, high-speed bayonet data, truck passing data, a running behavior mode of the tobacco-related vehicle, full load weight, weight change condition before unloading and after unloading }.
Further, in the step S200, the characteristics of the image obtained by encoding the image using an LSTM neural network (long-short-term memory network), the image encoding is performed using the LSTM neural network, the dimension is first reduced, and then the modeling is performed,
based on the LSTM neural network, the specific process of training the image in step S200 includes:
step S201, describing set of image IXPerforming multi-layer convolution to generate successive encoded vectorsZ
Wherein,is onedThe vector of the vector is the one that,Zis onem×m×dIs a vector of (2);
step S202, adopting an ebedding model for eachPerforming adjacent search to obtain corresponding encoding table vectorZ q The method comprises the steps of carrying out a first treatment on the surface of the Wherein the ebedding model contains the coding table +.>By proximity search, willZMap to thisKOne of the vectors, i.e
Step S203, adopting decoder model pairZ q Reconstructing to obtain an imageIIs encoded image of (a)
Step S204, setting an objective functionFor coded pictures->Training is performed
Wherein,、/>is a superparameter and->；
Further, in step S200, the trained image acquires features, and compares the features with the features of the database to determine the feature type.
Further, in the step S400, for the text portion to be predicted, the feature of interest is obtained by performing encoding and recognition in the same manner as in the step S100; for the image part to be predicted, the same method as in step S200 is used for encoding and identifying to obtain the features, but the comparison of the feature library is not performed.
Further, in step S500, the obtained characteristics of the text portion to be predicted and the text characteristics in the training set are respectively input into two twin branches in the LSTM neural network for training. The specific process is as follows:
step S501, respectively inputting the features to be predicted and the features of the training set into two twin branches in the LSTM neural network for training;
step S502, obtaining the trained loss, and determining a first hidden variable feature of feature codes to be predicted and a second hidden variable feature of feature codes of a training set based on the loss;
and step S503, performing iterative training on the neural network according to the feature codes to be predicted and the corresponding first hidden variable features, and the feature codes of the training set and the corresponding second hidden variable features until the loss is minimum, and obtaining a similarity detection model.
Further, the loss function in step S502 is
Wherein m is a sampleThe number of the product is the number,for training set feature vector, ++>Is the feature vector to be predicted.
Further, the first hidden variable in step S502 is characterized by a loss function pairDeriving, the second hidden variable is characterized by a loss function pair +.>And (5) deriving.
Further, in step S503, the corresponding first hidden variable feature is superimposed according to the feature code to be predicted, so as to obtain an updated first feature code; superposing corresponding second hidden variable features according to the feature codes of the training set to obtain updated second feature codes; and performing iterative training on the neural network by adopting the updated first feature codes and the updated second feature codes.
Compared with the prior art, the invention has the following advantages: (1) In the invention, the images and the characters in the database are described and encoded and decoded to obtain corresponding pre-training characteristics, thereby achieving better information acquisition effect and obtaining corresponding characteristics; (2) In the identification process, the image and/or the text to be researched and judged are encoded and decoded to obtain corresponding features, the corresponding features are identified with the features in the pre-training database, and the information about the smoke is judged, so that more accurate information research and judgment is obtained.
The invention is further described below with reference to the drawings.
Drawings
FIG. 1 is a schematic diagram of the method of the present invention.
Detailed Description
It should be noted that, the information (including but not limited to user equipment information, user personal information, etc.), data (including but not limited to data for analysis, stored data, presented data, etc.), and signals related to the present disclosure are all authorized by the user or are fully authorized by the parties, and the collection, use, and processing of relevant data is required to comply with relevant laws and regulations and standards of relevant countries and regions. In the embodiment, the behavior of revealing personal information and the phenomenon of infringing personal privacy do not exist, and the national legal regulations are met.
Referring to fig. 1, a method for studying and judging digital information based on smoke alarm linkage includes pre-training and recognition processes. And describing images and characters in the database through pre-training, and encoding and decoding to obtain corresponding pre-training characteristics. The identification process is to encode and decode the image and/or text to be studied and judged to obtain the corresponding characteristics, and to identify the characteristics in the pre-training database to judge the information related to the smoke.
In the embodiment, the method and the device can be used for studying and judging the task situations of the tobacco involving staff, the communication relation of the tobacco involving staff, the living relation of the tobacco involving staff, the loss of regulation of related staff, the specific activities of the tobacco involving staff, the activity areas of related staff, the tobacco involving vehicles, the transportation of the tobacco involving vehicles, key bayonets and the like.
Wherein the pre-training comprises the steps of:
step S100, given a task, performing word segmentation processing and attention on a text by adopting a multi-head attention mechanism and a gating circulation unit mechanism;
step S200, given a task, identifying an image by adopting an LSTM neural network, and comparing the acquired characteristics with the characteristics of a characteristic library;
step S300, associating the text and the image after the pre-training to form a text-image pair training set containing the characteristics.
The identification process comprises the following steps:
step S400, identifying the input text and/or image to be predicted, and obtaining corresponding characteristics;
step S500, comparing the feature to be predicted and the text-image pair containing the feature obtained by pre-training, and if the feature is within the threshold range, identifying and obtaining the task type.
In step S100, when a person views an image or text containing a lot of information, the person does not pay attention to all the information at the same time, but first sees the key area, so in order to simulate the cognitive attention of the person, the attention mechanism based on the deep neural network assists the network model in screening out the information with the highest degree of association with the task at the present stage from the complex input information. The gating circulation unit can enable the neural network to not only memorize past information, but also selectively delete unimportant information. Meanwhile, in order to increase the accuracy of pre-training, the embodiment adopts a multi-head attention mechanism to carry out comparison learning on the same text information, and reduces the situation of text mismatch.
Based on the above principle, the specific process of focusing attention on the text of the word segmentation in step S100 by adopting the attention mechanism includes:
step S101, adopting an ebedding model pairTreating to obtain->Wherein->Representing a vector corresponding to a word or phrase of the text, < >>In order to distribute the attention of the person,Unumber of single and phrases in the text;
step S102, obtaining the query vector in the multi-graph attention mechanismKey vector->Sum vector->
Wherein,、/>、/>is a matrix of U×T->，，/>T is the number of network channels where the multi-head attention mechanism is located;
step S103, each query vectorMultiplying all key vectors to obtain corresponding score +.>；
Step S104, obtaining each query vector by using Softmax functionAttention probability distribution +.>I.e. weight coefficient
Step S105, for each query vectorCorresponding value vector +.>Weighted summation is carried out to obtain the attention output after comparison learning>
Step S105, the attention output obtained by all the query vectors is spliced to obtain the attention after the contrast learningb
In step S100, the text of interest is also different for different tasks. For a smoke-related person studying and judging task, the concerned text vector comprises { personnel basic information, trip data, call data and express data }; for a smoke-related person communication relation research and judgment task, the concerned text vector comprises { the name, sex, age of a member, the interaction frequency between the member and the social network, and the relation density }; the smoke-related personnel living relation study and judgment task should pay attention to travel records, and the text vector of attention comprises { shift, carriage, adjacent seat, departure place, purpose, check in and check out }; for related personnel out-of-specification research and judgment tasks, the concerned text vector comprises { type of out-of-specification behavior, frequency, time, place, call, related personnel, environment }; for a specific activity research and judgment task of a smoke-related person, the concerned text vector comprises { person name, identity card number, location, travel information, contact person condition, travel history, communication record, social network behavior }; for a relevant personnel activity area research task, the concerned text vector comprises { the historical activity range of personnel, time node }; for the smoke-related vehicle research and judgment task, the text vector concerned comprises { vehicle basic information, bayonet data, high-speed data and time sequence data }; for the tobacco-related vehicle transportation research and judgment task, the text vector concerned comprises { high-speed toll station data, high-speed bayonet data, truck passing data, a running behavior mode of the tobacco-related vehicle, full load weight, weight change condition before unloading and after unloading }.
In step S200, the image is encoded using an LSTM neural network (long short term memory network) to obtain features. The existing autoregressive method is used for coding and identifying the images, and because the images are generated pixel by pixel, each pixel needs to be randomly sampled, and the operation is slow. And the LSTM neural network is adopted for image coding, firstly dimension reduction is carried out, and then modeling is carried out. Based on the LSTM neural network, the specific process of training the image in step S200 includes:
step S201, describing set of image IXPerforming multi-layer convolution to generate successive encoded vectorsZ
Wherein,is onedThe vector of the vector is the one that,Zis onem×m×dIs a vector of (2);
step S202, adopting an ebedding model for eachPerforming adjacent search to obtain corresponding encoding table vectorZ q The method comprises the steps of carrying out a first treatment on the surface of the Wherein the ebedding model contains the coding table +.>By proximity search, willZMap to thisKOne of the vectors, i.e
Step S203, adopting decoder model pairZ q Reconstructing to obtain an imageIIs encoded image of (a)
Step S204, setting an objective functionFor coded pictures->Training is performed
Wherein,、/>is a superparameter and->；
Further, in step S200, the trained image acquires features, and compares the features with the features of the database to determine the feature type.
In step S400, for the text portion to be predicted, coding recognition is performed by the same method as in step S100 to obtain a feature of interest; for the image part to be predicted, the same method as in step S200 is used for encoding and identifying to obtain the features, but the comparison of the feature library is not performed.
In step S500, the obtained characteristics of the text portion to be predicted and the text characteristics in the training set are respectively input into two twin branches in the LSTM neural network for training. The specific process is as follows:
step S501, respectively inputting the features to be predicted and the features of the training set into two twin branches in the LSTM neural network for training;
step S502, obtaining the trained loss, and determining a first hidden variable feature of feature codes to be predicted and a second hidden variable feature of feature codes of a training set based on the loss;
and step S503, performing iterative training on the neural network according to the feature codes to be predicted and the corresponding first hidden variable features, and the feature codes of the training set and the corresponding second hidden variable features until the loss is minimum, and obtaining a similarity detection model.
The loss function in step S502 is
Wherein m is the number of samples,for training set feature vector, ++>Is the feature vector to be predicted.
The first hidden variable in step S502 is characterized by a loss function pairDeriving, the second hidden variable is characterized by a loss function pair +.>And (5) deriving.
In step S503, according to the feature code to be predicted, the corresponding first hidden variable feature is superimposed, so as to obtain an updated first feature code; superposing corresponding second hidden variable features according to the feature codes of the training set to obtain updated second feature codes; and performing iterative training on the neural network by adopting the updated first feature codes and the updated second feature codes.
Claims (3)
1. A method for studying and judging digital information based on smoke alarm linkage is characterized by comprising a pre-training and recognition process, wherein
The pre-training comprises the following steps:
step S100, given a task, performing word segmentation processing and attention on a text by adopting a multi-head attention mechanism and a gating circulation unit mechanism;
step S200, given a task, identifying an image by adopting an LSTM neural network, and comparing the acquired characteristics with the characteristics of a characteristic library;
step S300, associating the text and the image after pre-training to form a text-image pair training set containing characteristics;
the identification process comprises the following steps:
step S400, identifying the input text and/or image to be predicted, and obtaining corresponding characteristics;
step S500, comparing the feature to be predicted and the text-image pair containing the feature obtained by pre-training, and if the feature is within the threshold range, identifying and obtaining the task type;
the specific process of focusing attention on the text of the word segmentation by adopting the attention mechanism in the step S100 includes:
step S101, adopting an ebedding model pairProcessing to obtainWherein->Representing a vector corresponding to a word or phrase of the text, < >>In order to distribute the attention of the person,Uis the number of words and phrases in the text;
step S102, obtaining the query vector in the multi-head attention mechanismKey vector->Sum vector-> Wherein (1)>、/>、/>Is a matrix of U×T->，，/>，TThe number of channels of the neural network where the multi-head attention mechanism is located;
step S103, each query vectorMultiplying all key vectors to obtain corresponding score +.>；
Step S104, obtaining each query vector by using Softmax functionAttention probability distribution +.>I.e. weight coefficient
Step S105 +.>Corresponding value vector +.>Weighted summation is carried out to obtain the attention output after comparison learning> Step S106, the attention output obtained by all the query vectors is spliced to obtain the attention after the contrast learningb
；
The specific process of training the image in step S200 includes:
step S201, describing set of image IXPerforming multi-layer convolution to generate successive encoded vectorsZ
Wherein (1)>Is onedThe vector of the dimensions is used to determine,Zis onem×m×dIs a vector of (2);
step S202, adopting an ebedding model for eachPerforming adjacent search to obtain corresponding encoding table vectorZ q The method comprises the steps of carrying out a first treatment on the surface of the Wherein the ebedding model contains the coding table +.>By proximity search, willZMap to thisKOne of the vectors, i.e
Step S203, adopting decoder model pairZ q Reconstructing to obtain an imageICoded picture +.> Step S204, set the objective function +.>For coded pictures->Training is performed
Wherein (1)>、/>Is a superparameter and->；
In the step S500, the obtained characteristics of the text portion to be predicted and the text characteristics in the training set are respectively input into two twin branches in the LSTM neural network for training; the specific process is as follows:
step S501, respectively inputting the features to be predicted and the features of the training set into two twin branches in the LSTM neural network for training;
step S502, obtaining the trained loss, and determining a first hidden variable feature of feature codes to be predicted and a second hidden variable feature of feature codes of a training set based on the loss;
step S503, carrying out iterative training on the neural network according to the feature codes to be predicted and the corresponding first hidden variable features, and the feature codes of the training set and the corresponding second hidden variable features until the loss is minimum, and obtaining a similarity detection model;
the loss function in the step S502 is
Wherein m is the number of samples, +.>For training set feature vector, ++>The feature vector is to be predicted;
the first hidden variable in the step S502 is characterized by a loss function pairDeriving, the second hidden variable is characterized by a loss function pairSeeking a derivative;
in step S503, according to the feature code to be predicted, the corresponding first hidden variable feature is superimposed, so as to obtain an updated first feature code; superposing corresponding second hidden variable features according to the feature codes of the training set to obtain updated second feature codes; and performing iterative training on the neural network by adopting the updated first feature codes and the updated second feature codes.
2. The method for studying and judging digitized information based on smoke alarm linkage according to claim 1, wherein in the step S200, the trained image acquires the features and compares the features with the features of the database to judge the feature type.
3. The method for studying and judging digitized information based on smoke linkage according to claim 2, wherein in the step S400, coding and recognizing the text part to be predicted by the same method as the step S100 to obtain the feature of interest; for the image part to be predicted, the same method as in step S200 is used for encoding and identifying to obtain the features, but the comparison of the feature library is not performed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311083689.4A CN116842141B (en) | 2023-08-28 | 2023-08-28 | Alarm smoke linkage based digital information studying and judging method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311083689.4A CN116842141B (en) | 2023-08-28 | 2023-08-28 | Alarm smoke linkage based digital information studying and judging method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116842141A CN116842141A (en) | 2023-10-03 |
| CN116842141B true CN116842141B (en) | 2023-11-07 |
Family
ID=88165482
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311083689.4A Active CN116842141B (en) | 2023-08-28 | 2023-08-28 | Alarm smoke linkage based digital information studying and judging method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116842141B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112257445A (en) * | 2020-10-19 | 2021-01-22 | 浙大城市学院 | Multi-modal tweet named entity recognition method based on text-picture relation pre-training |
| CN115134559A (en) * | 2022-01-12 | 2022-09-30 | 北京环球森林科技有限公司 | Bayonet cigarette end identification and suspicious behavior intelligent studying and judging method for forest fire prevention |
| CN115239937A (en) * | 2022-09-23 | 2022-10-25 | 西南交通大学 | Cross-modal emotion prediction method |
| CN115982350A (en) * | 2022-12-07 | 2023-04-18 | 南京大学 | False news detection method based on multi-mode Transformer |
| WO2023093574A1 (en) * | 2021-11-25 | 2023-06-01 | 北京邮电大学 | News event search method and system based on multi-level image-text semantic alignment model |
| CN116611021A (en) * | 2023-04-19 | 2023-08-18 | 齐鲁工业大学(山东省科学院) | Multi-mode event detection method and system based on double-transducer fusion model |
-
2023
- 2023-08-28 CN CN202311083689.4A patent/CN116842141B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112257445A (en) * | 2020-10-19 | 2021-01-22 | 浙大城市学院 | Multi-modal tweet named entity recognition method based on text-picture relation pre-training |
| WO2023093574A1 (en) * | 2021-11-25 | 2023-06-01 | 北京邮电大学 | News event search method and system based on multi-level image-text semantic alignment model |
| CN115134559A (en) * | 2022-01-12 | 2022-09-30 | 北京环球森林科技有限公司 | Bayonet cigarette end identification and suspicious behavior intelligent studying and judging method for forest fire prevention |
| CN115239937A (en) * | 2022-09-23 | 2022-10-25 | 西南交通大学 | Cross-modal emotion prediction method |
| CN115982350A (en) * | 2022-12-07 | 2023-04-18 | 南京大学 | False news detection method based on multi-mode Transformer |
| CN116611021A (en) * | 2023-04-19 | 2023-08-18 | 齐鲁工业大学(山东省科学院) | Multi-mode event detection method and system based on double-transducer fusion model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116842141A (en) | 2023-10-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111666588B (en) | Emotion differential privacy protection method based on generation countermeasure network | |
| CN112015901A (en) | Text classification method and device and warning situation analysis system | |
| CN112016313A (en) | Spoken language element identification method and device and alarm situation analysis system | |
| CN113343640B (en) | Method and device for classifying customs commodity HS codes | |
| CN114091462B (en) | Case fact mixed coding based criminal case risk mutual learning assessment method | |
| CN113836896A (en) | Patent text abstract generation method and device based on deep learning | |
| CN117993499B (en) | Multi-mode knowledge graph construction method for four pre-platforms for flood control in drainage basin | |
| CN116304984A (en) | Multi-modal intention recognition method and system based on contrast learning | |
| CN116610818A (en) | Construction method and system of power transmission and transformation project knowledge base | |
| CN115269836A (en) | Intention identification method and device | |
| CN111815485A (en) | Sentencing prediction method and device based on deep learning BERT model | |
| CN115063612A (en) | Fraud early warning method, device, equipment and storage medium based on face-check video | |
| CN117251685B (en) | Knowledge graph-based standardized government affair data construction method and device | |
| CN114461760A (en) | Method and device for matching case fact with law bar | |
| CN116842141B (en) | Alarm smoke linkage based digital information studying and judging method | |
| CN118035440A (en) | Enterprise associated archive management target knowledge feature recommendation method | |
| CN116578734B (en) | Probability embedding combination retrieval method based on CLIP | |
| CN117314623A (en) | Loan fraud prediction method, device and storage medium integrating external knowledge | |
| CN117037990A (en) | Intelligent classification storage method and device based on medical records quality | |
| CN116205350A (en) | Reinforcement personal risk analysis and prediction system and method based on legal documents | |
| CN117037017A (en) | Video emotion detection method based on key frame erasure | |
| KR102556450B1 (en) | Customized wine recommendation method based on artificial intelligence and operating server for the method | |
| CN115982388A (en) | Case quality control map establishing method, case document quality testing method, case quality control map establishing equipment and storage medium | |
| CN117935784A (en) | Voice processing model training method, voice recognition method and device | |
| CN117077680A (en) | Question and answer intention recognition method and device |
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 | ||
| CP02 | Change in the address of a patent holder | ||
| CP02 | Change in the address of a patent holder |
Address after: Room 119, 1st Floor, Building 3, No. 20 Yong'an Road, Shilong Economic Development Zone, Mentougou District, Beijing, 102300 Patentee after: Beijing Zhongan Technology Development Co.,Ltd. Address before: A502, No. 28 Dongjiaomin Lane, Dongcheng District, Beijing, 100010 Patentee before: Beijing Zhongan Technology Development Co.,Ltd. |