CN115953741B - Edge computing system and method based on embedded algorithm - Google Patents
Edge computing system and method based on embedded algorithm Download PDFInfo
- Publication number
- CN115953741B CN115953741B CN202310240784.4A CN202310240784A CN115953741B CN 115953741 B CN115953741 B CN 115953741B CN 202310240784 A CN202310240784 A CN 202310240784A CN 115953741 B CN115953741 B CN 115953741B
- Authority
- CN
- China
- Prior art keywords
- data
- unit
- article
- personnel
- model
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention provides an edge computing system and a method based on an embedded algorithm, which are characterized in that a data computing module is arranged to remove invalid data to obtain effective data, so that the data acquired by a data acquisition module is filtered and cleaned at the front end, the occupation of resources is reduced, the invalid data is reduced, the investigation time is shortened, and the early warning efficiency and accuracy are improved; determining that the article data is collected by a personnel collecting unit or an article collecting unit according to the distance between the article and personnel, and limiting a data calculation model which can be selected by the article data and an interval time period for collecting the article data according to the distance; therefore, data acquisition and data calculation are carried out by adopting a data calculation model with higher calculation precision and a shorter interval time period aiming at the article information close to the personnel, and the reliability and the credibility of the acquired information are ensured by using a more accurate data calculation model aiming at the article information with larger possibility of dangerous situations.
Description
Technical Field
The invention relates to the technical field of edge calculation, in particular to an edge calculation system and method based on an embedded algorithm.
Background
The model and version of the video monitoring management platform of the existing data center are not uniform, the function is simple, and the problem that video monitoring equipment of a single machine room is distributed in a plurality of management platforms exists. The derived video monitoring data has high invalid data ratio, the retrieval and the inquiry are only completed manually, the working efficiency is low, the warning can not be timely carried out after the dangerous situation occurs, and the requirement of an intelligent machine room can not be met.
The drawbacks and deficiencies of the prior art are mainly manifested in the following aspects:
1) And the useless data generated in the daily production process is excessive, so that the storage resource is occupied. After the dangerous case occurs, the time and effort are consumed for checking the useless data, the tracing process is long, the time point of the dangerous case cannot be positioned at the first time, and the accurate picture cannot be intercepted.
2) After people enter and exit the machine room, the data collected by the front-end camera does not timely carry out deep data analysis and data processing, and the whole process and state cannot be timely controlled.
The prior patent document CN111104841a provides a method for detecting violent behaviors, a system for detecting violent behaviors, a computer device and a readable storage medium, wherein an acquired scene image is firstly passed through a feature pyramid network to detect a frame of a target human body, then according to a detection result, a human body posture of the target human body is estimated by using a cascading pyramid network, according to an estimation result, the estimation result is matched with the violent behaviors stored in a database, so as to judge whether violent behaviors exist or not according to a matching result, and the violent behaviors are classified. However, front-end data elimination is not set in the scheme, the data volume is large, the occupied resources are large, meanwhile, only the estimation on the human body posture is concentrated, further estimation on objects in the surrounding environment of the human body is not performed, and the accuracy and the reliability cannot be guaranteed.
The prior patent document CN113536267A provides a human gait verification method and a cloud server based on artificial intelligence, which are used for screening high-quality human images to generate a picture sequence by evaluating the human image quality of continuous video frame images in a period of time, so that the human region range of each picture can be determined based on the picture sequence to generate a human image sequence and perform human gait feature recognition, and then human gait feature verification is performed through an artificial intelligence model, so that human identity recognition can be performed through human gait under the conditions of face shielding or facial feature blurring of personnel and personnel non-coordination; however, in the scheme provided by the method, feature recognition of human body images and human body gait is still focused, feature recognition is not performed on article information in image information, classification model recognition is not performed on article information with different distances from a human body, and the recognition result is inaccurate and incomplete only by means of recognizing human body information.
Disclosure of Invention
In order to solve the defects and shortcomings in the prior art, the invention provides an edge computing system and method based on an embedded algorithm.
The specific scheme provided by the invention is as follows:
the system comprises a data acquisition module, a data calculation module and a data storage module, wherein the data acquisition module acquires data in an image and transmits the acquired data to the data calculation module through a data transmission module, the data calculation module obtains effective data by calculating and removing invalid data, and the effective data is transmitted to the data storage module through the data transmission module; the method is characterized in that:
the data acquisition module at least comprises a personnel acquisition unit, an article acquisition unit and a smoke and fire acquisition unit; the acquisition priority of each unit satisfies: the smoke and fire collecting unit is more than the personnel collecting unit is more than the article collecting unit; wherein, the liquid crystal display device comprises a liquid crystal display device,
the data collected by the personnel collecting unit at least comprises personnel information data, personnel face data, personnel posture data and personnel accessory data;
the data collected by the article collecting unit at least comprises article information data, article distance data and article reference point data;
the data collected by the smoke and fire collecting unit at least comprises smoke and fire image data, smoke and fire range data and smoke and fire temperature data;
in the data acquisition module, the article data is determined to be acquired by a personnel acquisition unit or an article acquisition unit according to the distance between the article and personnel, and when
When the distance between the object and the person is not higher than a preset distance threshold, the object data is collected by a person collecting unit; defining a first data calculation model which can be selected by the article data, and defining an interval time period for acquiring the article data as a first time period;
when the distance between the article and the person is higher than a preset distance threshold, acquiring the article data by an article acquisition unit; defining a second data calculation model which can be selected by the article data, and defining the interval time period of the article data acquisition as a second time period;
and satisfies the following:
the calculation accuracy of the first data calculation model is higher than that of the second data calculation model;
the first time period is less than the second time period;
the preset distance threshold value can be adjusted in a later period;
the data calculation module comprises a model storage unit, a model training unit, a threshold setting unit, a data judging unit, a data storage unit and a self-learning optimizing unit; the model storage unit stores a plurality of corresponding data calculation models which can be selected by a user, the model training unit trains the data calculation models selected by the user, the threshold setting unit determines a corresponding threshold range according to the data calculation models selected by the user, the data judging unit judges the collected data through the data calculation models to determine the collected data to be effective data or invalid data, the data temporary storage unit stores the effective data and the invalid data in a classified mode, and the self-learning optimizing unit optimizes the calculation models selected by the user once according to the training data, the effective data, the invalid data and the client requirements;
the data storage module comprises a data storage unit, a data analysis unit, a data early warning unit and a data optimization unit; the data pre-warning unit makes corresponding pre-warning prompts according to data analysis results, and the data optimizing unit optimizes the data analyzing unit according to the risk data and the safety data;
and the data optimizing unit transmits the optimizing result of the data analyzing unit to the self-learning optimizing unit so as to secondarily optimize the data calculation model selected by the user.
As a further preferred embodiment of the present invention, the person collecting unit collects face data, posture data, and accessory data of persons at corresponding time points, and collects corresponding change data of face data, posture data, and accessory data of persons at adjacent time points according to a preset time period.
As a further preferred embodiment of the present invention, among the data collected by the item collecting unit, item information data is collected preferentially, surrounding item reference point data is obtained according to the item information data, the item reference point data provides a plurality of reference points for a user to select, and the item collecting unit determines item distance data according to the reference points selected by the user.
In a further preferred embodiment of the present invention, among the data collected by the pyrotechnic collecting unit, pyrotechnic range data and pyrotechnic temperature data are preferentially obtained, pyrotechnic image data are drawn according to the pyrotechnic range data and the pyrotechnic temperature data, and the pyrotechnic image data can individually reflect the pyrotechnic range data and the pyrotechnic temperature data, or simultaneously reflect the pyrotechnic range data and the pyrotechnic temperature data according to a client requirement.
As a further preferable implementation mode of the invention, the model storage unit is stored with a plurality of data calculation models adopting embedded algorithms, and the data calculation models are sorted and stored according to the quality of the training results which pass through the model training unit in the past, so that the data calculation models are sequentially sorted and displayed according to the quality of the training results which pass through the model training unit in the last time when the data calculation models are selected by a user.
As a further preferred embodiment of the present invention, the model training unit classifies the data calculation model into a merit model, a median model and a inferior model according to the merits of the training results of the data calculation model; and the training results of the median model are preferentially and completely transmitted to the self-learning optimizing unit, and the training results of the optimal value model and the inferior value model are delayed and partially transmitted to the self-learning optimizing unit.
As a further preferred embodiment of the present invention, when the self-learning optimization unit performs one-time optimization on the calculation model selected by the user, the priority of the reference is: valid data and invalid data > training data > customer requirements.
As a further preferred embodiment of the present invention, the reliability of the primary optimization by the self-learning optimization unit on the user-selected calculation model is lower than the reliability of the secondary optimization by the self-learning optimization unit on the user-selected calculation model.
As a further preferred embodiment of the present invention, the coverage of the self-learning optimization unit for the primary optimization of the user-selected calculation model is higher than the coverage of the secondary optimization of the user-selected calculation model.
Further, the invention also provides a computing method of the edge computing system based on the embedded algorithm, which is characterized in that: the method comprises the following steps:
1) The data acquisition module acquires corresponding data in the image and transmits the acquired data to the data calculation module: wherein, the liquid crystal display device comprises a liquid crystal display device,
the personnel acquisition unit acquires personnel information data, personnel face data, personnel posture data and personnel accessory data;
the article acquisition unit acquires article information data, article distance data and article reference point data;
the smoke and fire collecting unit collects smoke and fire image data, smoke and fire range data and smoke and fire temperature data;
2) The user selects a data calculation model from the model storage unit;
3) Training a data calculation model selected by a user;
4) Determining a corresponding threshold range according to the data calculation model selected by the user;
5) The collected data is judged through a data calculation model to determine whether the data is valid data or invalid data,
6) Storing valid data and invalid data in a classified manner;
7) The self-learning optimization unit performs primary optimization on the calculation model selected by the user;
8) The valid data is sent to the data storage unit;
9) Performing data analysis on the effective data to determine risk data and safety data in the effective data;
10 Making a corresponding early warning prompt according to the data analysis result;
11 Optimizing the data analysis unit according to the risk data and the safety data;
12 A second optimization is performed on the data calculation model selected by the user.
Compared with the prior art, the invention has the following technical effects:
1) The invention provides an edge computing system and method based on an embedded algorithm, which are characterized in that a data computing module is arranged to exclude invalid data from computation to obtain effective data, so that data acquired by a data acquisition module is filtered and cleaned at the front end, the occupation of resources is reduced, the invalid data is reduced, the investigation time is shortened, and the early warning efficiency and accuracy are improved.
2) The invention provides an edge computing system and a method based on an embedded algorithm, which are used for determining whether article data are collected by a personnel collecting unit or an article collecting unit according to the distance between the article and personnel, and limiting a data computing model which can be selected by the article data and an interval time period for collecting the article data according to the data; therefore, data acquisition and data calculation are carried out by adopting a data calculation model with higher calculation precision and a shorter interval time period aiming at article information close to personnel, so that the reliability and the credibility of acquired information are ensured by adopting a data calculation model with more precise investment aiming at article information with possibly larger dangerous situations, and more comprehensive time-varying data in article data are acquired by adopting a smaller acquisition interval time period, so that effective data are acquired earlier, and early warning and time effectiveness of the dangerous situations are ensured.
3) The invention provides an edge computing system and a method based on an embedded algorithm, which are characterized in that a self-learning optimizing unit is used for carrying out primary optimization and secondary optimization on a computing model selected by a user, the safety data and risk data obtained in an actual computing process are used for carrying out deep secondary optimization on a primary optimizing result while correcting and perfecting the computing model selected by the user, the reliability of optimization is improved, a wider coverage area in the primary optimizing process is reserved and is higher than that of secondary optimization on the computing model selected by the user, and the width and breadth of data acquisition are ensured.
Drawings
Fig. 1 is a schematic diagram of the logic structure of the present invention.
A flowchart of the steps of the present invention is shown in fig. 2.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper", "lower", "inner", "outer", "front", "rear", "both ends", "one end", "the other end", etc. are based on the directions or positional relationships shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific direction, be configured and operated in the specific direction, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it should be noted that, unless explicitly specified and limited otherwise, the terms "mounted," "provided," "connected," and the like are to be construed broadly, and may be fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
First embodiment
Fig. 1 shows a first embodiment of the present invention, which discloses an edge computing system based on an embedded algorithm, the system includes a data acquisition module, a data computing module and a data storage module, the data acquisition module acquires data in an image and transmits the acquired data to the data computing module through a data transmission module, the data computing module obtains effective data by computing and excluding invalid data, and transmits the effective data to the data storage module through the data transmission module; the data calculation module is arranged to remove invalid data to obtain effective data, so that the data acquired by the data acquisition module is filtered and cleaned at the front end, the occupation of resources is reduced, the invalid data is reduced, the investigation time is shortened, and the early warning efficiency and accuracy are improved.
In this embodiment, the data acquisition module at least includes a personnel acquisition unit, an article acquisition unit, and a smoke and fire acquisition unit; wherein, the liquid crystal display device comprises a liquid crystal display device,
the data collected by the personnel collecting unit at least comprises personnel information data, personnel face data, personnel posture data and personnel accessory data; the personnel information data comprise relevant data such as name, sex, job level and the like of a pre-stored system, and personnel information data such as height, body type, color development, wearing and the like which can be obtained from the image; the facial data of the personnel comprise facial organs, expressions, attached glasses, jewelry and other data, and the facial data of the personnel are often changed correspondingly before and after the occurrence of artificial dangerous situations, for example, the facial expressions are changed from peace to a fierce state, the pupils of eyes are changed, the glasses are taken off and the like, and the facial expressions of the personnel are also changed from peace to tension and the like before and after the occurrence of natural dangerous situations, so that the facial data of the personnel are collected particularly necessarily; the gesture data of the person can predict the trend actions of the person in the next interval period, such as walking, lifting arms, lifting elbows and the like, and some gesture actions usually predict the occurrence of human dangerous situations, such as fist making, kicking and the like; and the occurrence of natural dangerous situations, such as fire suppression, watering and other gestures; the personnel accessory data can reflect the possible occurrence of the next interval time to a certain extent, for example, the hand-held dangerous goods can indicate artificial dangerous situations, the hand-held fire extinguisher and the like can indicate natural dangerous situations and the like.
In this embodiment, in the data collected by the personnel collecting unit, personnel face data, personnel posture data and personnel accessory data corresponding to the time points are collected, and corresponding change data of personnel face data, personnel posture data and personnel accessory data of adjacent time points are collected according to a preset time period, so that the dangerous situation which has occurred at present can be obtained through the corresponding change data, and the artificial dangerous situation or the natural dangerous situation which may occur in the next interval time period can be predicted. In this embodiment, the preset time periods for different data may be set to be the same for image data collection and storage, or may be set to be different for focusing on some of the data.
The data collected by the article collecting unit at least comprises article information data, article distance data and article reference point data; the article information data can comprise information data such as the shape, the size, the color and the like of the article, and whether the article has risks and hidden dangers can be confirmed by comparing the article information data with the pre-stored dangerous articles; the object distance data can reflect human dangerous situations possibly happening to some special groups, such as the distances between the socket electrical appliances and infants; the item reference points may provide the user with reference points in several image data for selection by the user.
In this embodiment, among the data collected by the article collecting unit, the article information data is collected preferentially, and the surrounding article reference point data is obtained according to the article information data, where the article reference point data provides a plurality of reference points for the user to select, and the article collecting unit determines the distance data between the rest of the articles and the reference point according to the reference point selected by the user.
The smoke and fire acquisition unit determines whether natural dangerous situations exist at present and whether natural dangerous situations possibly occur in a plurality of subsequent interval time periods; the acquired data at least comprises pyrotechnic image data, pyrotechnic range data and pyrotechnic temperature data. Wherein the pyrotechnic image data is used to determine whether a natural hazard is occurring and to determine a likelihood of the natural hazard occurring; the pyrotechnic range data is used for displaying the range of the current natural dangerous situation and predicting the range possibly swept in the subsequent interval time period; the pyrotechnic temperature data is used to display a temperature range within the current natural risk environment and predict a temperature range likely to be reached during a subsequent interval.
In this embodiment, among the data collected by the smoke and fire collecting unit, the smoke and fire range data and the smoke and fire temperature data are preferentially obtained, and the smoke and fire image data are drawn according to the smoke and fire range data and the smoke and fire temperature data, so that when a natural dangerous situation occurs currently, the smoke and fire image data are intuitively displayed to a user, and the natural dangerous situation possibly occurring in a subsequent interval period is timely predicted; the pyrotechnic image data can independently reflect the pyrotechnic range data and the pyrotechnic temperature data or simultaneously reflect the pyrotechnic range data and the pyrotechnic temperature data according to the requirements of customers so as to meet different use requirements of users.
In this embodiment, the acquisition priority of each unit satisfies: the smoke and fire collecting unit is more than the personnel collecting unit is more than the article collecting unit; the aim of the method is to determine whether natural dangerous situations which are happening or are impending exist in a subsequent interval time period or not through the corresponding data acquired by the smoke and fire acquisition unit, so that users are warned and prompted in real time to make corresponding measures to reduce losses as much as possible; and after the corresponding data collected by the smoke and fire collecting unit exclude the natural dangerous case which is happening currently or is happening in the following interval time period, the collecting priority of the personnel collecting unit is set to be superior to the collecting priority of the article collecting unit because the incidence rate of the artificial dangerous case is far higher than that of the natural dangerous case under normal conditions.
Because the article data in the image is relatively more, the article data cannot be determined to be collected by the personnel collecting unit or the article collecting unit, therefore, in the data collecting module provided by the embodiment, the article data is determined to be collected by the personnel collecting unit or the article collecting unit according to the distance between the article and the personnel, when
When the distance between the object and the person is not higher than a preset distance threshold, the object data is collected by a person collecting unit; defining a first data calculation model which can be selected by the article data, and defining an interval time period for acquiring the article data as a first time period;
when the distance between the article and the person is higher than a preset distance threshold, acquiring the article data by an article acquisition unit; defining a second data calculation model which can be selected by the article data, and defining the interval time period of the article data acquisition as a second time period;
and satisfies the following:
the calculation accuracy of the first data calculation model is higher than that of the second data calculation model;
the first time period is less than the second time period;
this is because when the distance between the object and the person is not higher than the preset distance threshold, the object is relatively close to the person, and the occurrence rate of the artificial dangerous case is far higher than that of the natural dangerous case under normal conditions, so that the possibility that the person generates the artificial dangerous case by using the object in the image is relatively high, and a relatively more accurate calculation model and a shorter time period are required to obtain more data changes in the same time, thereby further improving the calculation accuracy and reliability;
when the distance between the object and the person is higher than the preset distance threshold, the object is relatively far away from the person at the moment, and the occurrence rate of artificial dangerous situations is far higher than that of natural dangerous situations under normal conditions, so that the possibility that the person generates artificial dangerous situations by using the object in the image is relatively low, a calculation model with relatively low accuracy and a longer time period can be selected to obtain less data change in the same time, and therefore calculation accuracy and reliability are guaranteed.
As a further preference, the preset distance threshold can be adjusted later, so as to meet different use requirements of the user.
In this embodiment, as shown in fig. 1, the data calculation module includes a model storage unit, a model training unit, a threshold setting unit, a data determination unit, a data storage unit, and a self-learning optimization unit; the model storage unit stores a plurality of corresponding data calculation models which can be selected by a user, the model training unit trains the data calculation models selected by the user, the threshold setting unit determines a corresponding threshold range according to the data calculation models selected by the user, the data judging unit judges the collected data through the data calculation models to determine the collected data to be effective data or invalid data, the data temporary storage unit stores the effective data and the invalid data in a classified mode, and the self-learning optimizing unit optimizes the calculation models selected by the user once according to the training data, the effective data, the invalid data and the client requirements;
in this embodiment, a plurality of data calculation models adopting an embedded algorithm are stored in the model storage unit, and are stored in a sorted manner according to the quality of the training results of the model training unit in the past to avoid frequent sequential adjustment of the stored models, and when the models are selected by a user, the stored models are displayed in a sorted manner according to the quality of the training results of the model training unit in the last time, so that the data calculation model with the best or inferior training results displayed by the model training unit in the last time is preferentially provided for the user.
In this embodiment, the model training unit classifies the data calculation model into a merit model, a median model and an inferior model according to the merits of the training results of the data calculation model; and the training results of the median model are preferentially and completely transmitted to the self-learning optimizing unit, and the training results of the optimal value model and the inferior value model are delayed and partially transmitted to the self-learning optimizing unit. Similarly, since the newly developed or newly updated data calculation model often shows a better training result, and the actual calculation accuracy, stability and reliability thereof need to be confirmed through multiple training, when the quality is not confirmed, the training result of the median model is preferentially and completely transmitted to the self-learning optimization unit in order to ensure the optimization stability of the self-learning optimization unit, and the training results of the optimal model and the inferior model are delayed and partially transmitted to the self-learning optimization unit, so that the interference of a single training result of the newly developed or newly updated data calculation model to a single optimization process is avoided.
In this embodiment, when the self-learning optimization unit performs primary optimization on the calculation model selected by the user, the priority of reference is valid data and invalid data > training data > client requirement; the priorities of the effective data and the ineffective data determined by actual calculation are set to be higher than the priorities of the training data, so that the interference of a single training result of a newly developed or newly updated data calculation model to a single optimization process is avoided; and simultaneously, the customer requirements are reduced to the minimum, so that the interference caused by the requirements of non-professional users in a primary optimization process of the data calculation model is avoided.
In this embodiment, the data storage module includes a data storage unit, a data analysis unit, a data early warning unit, and a data optimization unit; the data pre-warning unit makes corresponding pre-warning prompts according to data analysis results, and the data optimizing unit optimizes the data analyzing unit according to the risk data and the safety data;
and the data optimizing unit transmits the optimizing result of the data analyzing unit to the self-learning optimizing unit so as to secondarily optimize the data calculation model selected by the user. And the reliability of the primary optimization of the self-learning optimization unit on the calculation model selected by the user is lower than that of the secondary optimization of the self-learning optimization unit on the calculation model selected by the user. The coverage of the self-learning optimization unit for the primary optimization of the user-selected calculation model is higher than that of the self-learning optimization unit for the secondary optimization of the user-selected calculation model. The safety data and the risk data obtained through the actual calculation process are subjected to deep secondary optimization on the primary optimization result, so that the reliability of optimization is improved, the wider coverage area in the primary optimization process is reserved to be higher than the coverage area of secondary optimization on the calculation model selected by the user, and the width and breadth of data acquisition are ensured.
In this embodiment, the selected data calculation model may be a convolutional deep neural network model, which includes an input layer, a convolutional layer, a pooling layer, an output layer, and the like, where each input two-dimensional image in the input layer is convolved with a convolution kernel in the convolutional layer to obtain a plurality of feature maps, and the number of convolution kernels also represents the number of filters. The method comprises the steps that a plurality of feature graphs obtained by carrying out convolution operation on an input layer and a convolution kernel are continuously input into a pooling layer, the pooling layer is used for reducing the size of input data, if the input data is a pooling area of NXN, the 4 input data are compressed into 1 data, and the number of the feature graphs is unchanged in the pooling operation process. Then a convolution layer, a pooling layer and a complete connection layer are passed to obtain the predicted value.
In this embodiment, the graph structure data input is processed by a convolutional deep neural network model, and the state of the whole network can be described by a formula (1) by using a space-based graph neural network calculation unit in the convolutional deep neural network model
Wherein, the liquid crystal display device comprises a liquid crystal display device,
Δt is the interval time; setting in actual calculation
s, all system states are modified and trained by the node.
The activation function formula adopted by the neural network in the convolutional deep neural network model is as follows
Wherein, the liquid crystal display device comprises a liquid crystal display device,
The cost function is defined as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
Second embodiment
Fig. 2 shows a second embodiment of the present invention, which provides a computing method of an edge computing system based on an embedded algorithm, comprising the following steps:
1) The data acquisition module acquires corresponding data in the image and transmits the acquired data to the data calculation module: wherein, the liquid crystal display device comprises a liquid crystal display device,
the personnel acquisition unit acquires personnel information data, personnel face data, personnel posture data and personnel accessory data;
the article acquisition unit acquires article information data, article distance data and article reference point data;
the smoke and fire collecting unit collects smoke and fire image data, smoke and fire range data and smoke and fire temperature data;
2) The user selects a data calculation model from the model storage unit;
3) Training a data calculation model selected by a user;
4) Determining a corresponding threshold range according to the data calculation model selected by the user;
5) The collected data is judged through a data calculation model to determine whether the data is valid data or invalid data,
6) Storing valid data and invalid data in a classified manner;
7) The self-learning optimization unit performs primary optimization on the calculation model selected by the user;
8) The valid data is sent to the data storage unit;
9) Performing data analysis on the effective data to determine risk data and safety data in the effective data;
10 Making a corresponding early warning prompt according to the data analysis result;
11 Optimizing the data analysis unit according to the risk data and the safety data;
12 A second optimization is performed on the data calculation model selected by the user.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (10)
1. The system comprises a data acquisition module, a data calculation module and a data storage module, wherein the data acquisition module acquires data in an image and transmits the acquired data to the data calculation module through a data transmission module, the data calculation module obtains effective data by calculating and removing invalid data, and the effective data is transmitted to the data storage module through the data transmission module; the method is characterized in that:
the data acquisition module at least comprises a personnel acquisition unit, an article acquisition unit and a smoke and fire acquisition unit; the acquisition priority of each unit satisfies: the smoke and fire collecting unit is more than the personnel collecting unit is more than the article collecting unit; wherein, the liquid crystal display device comprises a liquid crystal display device,
the data collected by the personnel collecting unit at least comprises personnel information data, personnel face data, personnel posture data and personnel accessory data;
the data collected by the article collecting unit at least comprises article information data, article distance data and article reference point data;
the data collected by the smoke and fire collecting unit at least comprises smoke and fire image data, smoke and fire range data and smoke and fire temperature data;
in the data acquisition module, the article data is determined to be acquired by a personnel acquisition unit or an article acquisition unit according to the distance between the article and personnel, and when
When the distance between the object and the person is not higher than a preset distance threshold, the object data is collected by a person collecting unit; defining a first data calculation model which can be selected by the article data, and defining an interval time period for acquiring the article data as a first time period;
when the distance between the article and the person is higher than a preset distance threshold, acquiring the article data by an article acquisition unit; defining a second data calculation model which can be selected by the article data, and defining the interval time period of the article data acquisition as a second time period;
and satisfies the following:
the calculation accuracy of the first data calculation model is higher than that of the second data calculation model;
the first time period is less than the second time period;
the preset distance threshold value can be adjusted in a later period;
the data calculation module comprises a model storage unit, a model training unit, a threshold setting unit, a data judging unit, a data temporary storage unit and a self-learning optimizing unit; the model storage unit stores a plurality of corresponding data calculation models which can be selected by a user, the model training unit trains the data calculation models selected by the user, the threshold setting unit determines a corresponding threshold range according to the data calculation models selected by the user, the data judging unit judges the collected data through the data calculation models to determine the collected data to be effective data or invalid data, the data temporary storage unit stores the effective data and the invalid data in a classified mode, and the self-learning optimizing unit optimizes the calculation models selected by the user once according to the training data, the effective data, the invalid data and the client requirements;
the data storage module comprises a data storage unit, a data analysis unit, a data early warning unit and a data optimization unit; the data pre-warning unit makes corresponding pre-warning prompts according to data analysis results, and the data optimizing unit optimizes the data analyzing unit according to the risk data and the safety data;
and the data optimizing unit transmits the optimizing result of the data analyzing unit to the self-learning optimizing unit so as to secondarily optimize the data calculation model selected by the user.
2. An edge computing system based on an embedded algorithm as claimed in claim 1, wherein: and the personnel face data, the personnel posture data and the personnel accessory data of the corresponding time points are acquired in the data acquired by the personnel acquisition unit, and the corresponding change data of the personnel face data, the personnel posture data and the personnel accessory data of the adjacent time points are acquired according to the preset time period.
3. An edge computing system based on an embedded algorithm as claimed in claim 1, wherein: among the data collected by the article collection unit, article information data are collected preferentially, surrounding article reference point data are obtained according to the article information data, the article reference point data provide a plurality of reference points for a user to select, and the article collection unit determines article distance data according to the reference points selected by the user.
4. An edge computing system based on an embedded algorithm as claimed in claim 1, wherein: among the data collected by the firework collecting unit, the firework range data and the firework temperature data are preferentially obtained, the firework image data are drawn according to the firework range data and the firework temperature data, and the firework image data can independently respond to the firework range data and the firework temperature data or simultaneously respond to the firework range data and the firework temperature data according to the requirements of customers.
5. An edge computing system based on an embedded algorithm as claimed in claim 1, wherein: the model storage unit is used for storing a plurality of data calculation models adopting an embedded algorithm, and is used for sorting and storing according to the quality of the training results passing through the model training unit in the past, so that the user can sort and display according to the quality of the training results passing through the model training unit in the last time when selecting.
6. An embedded algorithm based edge computing system according to claim 5, wherein: the model training unit classifies the data calculation model into a merit model, a median model and a inferior model according to the merits and inferiorities of the training result of the data calculation model; and the training results of the median model are preferentially and completely transmitted to the self-learning optimizing unit, and the training results of the optimal value model and the inferior value model are delayed and partially transmitted to the self-learning optimizing unit.
7. An edge computing system based on an embedded algorithm as claimed in claim 6, wherein: when the self-learning optimization unit performs primary optimization on the calculation model selected by the user, the priority of reference is as follows: valid data and invalid data > training data > customer requirements.
8. An edge computing system based on an embedded algorithm as claimed in claim 1, wherein: the reliability of the primary optimization of the self-learning optimization unit on the calculation model selected by the user is lower than that of the secondary optimization of the self-learning optimization unit on the calculation model selected by the user.
9. An edge computing system based on an embedded algorithm as claimed in claim 1, wherein: the coverage area of the self-learning optimization unit for performing primary optimization on the calculation model selected by the user is higher than that of the self-learning optimization unit for performing secondary optimization on the calculation model selected by the user.
10. A method of computing an edge computing system based on an embedded algorithm as claimed in any one of claims 1 to 9, wherein: the method comprises the following steps:
1) The data acquisition module acquires corresponding data in the image and transmits the acquired data to the data calculation module: wherein, the liquid crystal display device comprises a liquid crystal display device,
the personnel acquisition unit acquires personnel information data, personnel face data, personnel posture data and personnel accessory data;
the article acquisition unit acquires article information data, article distance data and article reference point data;
the smoke and fire collecting unit collects smoke and fire image data, smoke and fire range data and smoke and fire temperature data;
2) The user selects a data calculation model from the model storage unit;
3) Training a data calculation model selected by a user;
4) Determining a corresponding threshold range according to the data calculation model selected by the user;
5) The collected data is judged through a data calculation model to determine whether the data is valid data or invalid data,
6) Storing valid data and invalid data in a classified manner;
7) The self-learning optimization unit performs primary optimization on the calculation model selected by the user;
8) The valid data is sent to the data storage unit;
9) Performing data analysis on the effective data to determine risk data and safety data in the effective data;
10 Making a corresponding early warning prompt according to the data analysis result;
11 Optimizing the data analysis unit according to the risk data and the safety data;
12 A second optimization is performed on the data calculation model selected by the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310240784.4A CN115953741B (en) | 2023-03-14 | 2023-03-14 | Edge computing system and method based on embedded algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310240784.4A CN115953741B (en) | 2023-03-14 | 2023-03-14 | Edge computing system and method based on embedded algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115953741A CN115953741A (en) | 2023-04-11 |
| CN115953741B true CN115953741B (en) | 2023-05-12 |
Family
ID=85891410
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310240784.4A Active CN115953741B (en) | 2023-03-14 | 2023-03-14 | Edge computing system and method based on embedded algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115953741B (en) |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190485A (en) * | 2018-07-31 | 2019-01-11 | 李明 | Data processing method, device, computer equipment and storage medium |
| CN111784966A (en) * | 2020-06-15 | 2020-10-16 | 武汉烽火众智数字技术有限责任公司 | Personnel management and control method and system based on machine learning |
| WO2022022368A1 (en) * | 2020-07-28 | 2022-02-03 | 宁波环视信息科技有限公司 | Deep-learning-based apparatus and method for monitoring behavioral norms in jail |
| CN113903058A (en) * | 2021-11-19 | 2022-01-07 | 上海玉贲智能科技有限公司 | Intelligent control system based on regional personnel identification |
| CN114445661B (en) * | 2022-01-24 | 2023-08-18 | 电子科技大学 | Embedded image recognition method based on edge calculation |
-
2023
- 2023-03-14 CN CN202310240784.4A patent/CN115953741B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115953741A (en) | 2023-04-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11710075B2 (en) | Hazard recognition | |
| CN109766755B (en) | Face recognition method and related product | |
| CN108875452A (en) | Face identification method, device, system and computer-readable medium | |
| CN113110207A (en) | Insect pest remote monitoring method and system based on sensor of Internet of things and storage medium | |
| CN108537166B (en) | Method and device for determining shelf browsing amount and analyzing browsing amount | |
| CN107122744A (en) | A kind of In vivo detection system and method based on recognition of face | |
| CN109309808A (en) | A kind of monitoring system and method based on recognition of face | |
| JP5085621B2 (en) | Feature information selection device, image recognition processing device, monitoring system, method, and program applied to image recognition processing | |
| CN109544548A (en) | Defect inspection method, device, server, equipment and the storage medium of cutlery box | |
| CN109598229A (en) | Monitoring system and its method based on action recognition | |
| CN109886555A (en) | The monitoring method and device of food safety | |
| CN107330414A (en) | Act of violence monitoring method | |
| CN114648354A (en) | Advertisement evaluation method and system based on eye movement tracking and emotional state | |
| CN110737201A (en) | monitoring method, device, storage medium and air conditioner | |
| CN109657549A (en) | A kind of safety monitoring method, device, system and computer storage medium | |
| CN115861915A (en) | Fire fighting access monitoring method, fire fighting access monitoring device and storage medium | |
| CN115953741B (en) | Edge computing system and method based on embedded algorithm | |
| CN109670438A (en) | Abnormal behaviour monitoring method, device, system and storage medium for intelligent desk lamp | |
| CN108875488A (en) | Method for tracing object, object tracking device and computer readable storage medium | |
| WO2016038872A1 (en) | Information processing device, display method, and program storage medium | |
| CN114612771A (en) | Fire source monitoring method and system based on neural network | |
| JP2018165966A (en) | Object detection device | |
| JP2018142137A (en) | Information processing device, information processing method and program | |
| KR101413558B1 (en) | Analysis device of user multi-intent, method of analysis user multi-intent | |
| CN111126328A (en) | Intelligent firefighter posture monitoring method and system |
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 |