CN109447162A - A kind of real-time Activity recognition system and its working method based on Lora and Capsule - Google Patents
A kind of real-time Activity recognition system and its working method based on Lora and Capsule Download PDFInfo
- Publication number
- CN109447162A CN109447162A CN201811294526.XA CN201811294526A CN109447162A CN 109447162 A CN109447162 A CN 109447162A CN 201811294526 A CN201811294526 A CN 201811294526A CN 109447162 A CN109447162 A CN 109447162A
- Authority
- CN
- China
- Prior art keywords
- behavioural information
- information
- unit
- module
- capsule
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The present invention relates to a kind of real-time Activity recognition system and its working method based on Lora and Capsule, which includes behavioural information physical layer, behavioural information access layer, behavioural information podium level, behavioural information application layer.The transmission of behavioural information access layer is used Lora node, the base station Lora by the present invention, realizes remote, low-power consumption behavioural information transmission;Processing in terms of having carried out inconsistent and incompleteness to behavioural information uncertainty improves the confidence level of behavioural information;It obtains the spatial relationship between useful feature and feature automatically using Capsule to carry out Activity recognition, is greatly improved in precision aspect.
Description
Technical field
The present invention relates to a kind of real-time Activity recognition system and its working method based on Lora and Capsule, belong to people
The technical field of work intelligence and pattern-recognition.
Background technique
Activity recognition system is the behavioural information by obtaining people, by the system of reasonable model realization behavior.With
The development and maturation of the advanced technologies such as Internet of Things, artificial intelligence, big data and cloud computing, more and more scholars begin to focus on row
For the research for identifying direction.Activity recognition has become artificial intelligence and very powerful and exceedingly arrogant grinds with one in pattern identification research field
Study carefully direction, the development of wearable device provides excellent opportunity for Human bodys' response in addition.Nowadays, Human bodys' response skill
Art has obtained tentatively in the fields such as game, human motion analysis, smart home, human-computer interaction and medical diagnosis and monitoring
Using.
Behavioural information required for Activity recognition system is mainly from following two aspect:
1, the behavioural information of view-based access control model --- visual behaviour information is acquired by picture pick-up device.
2, sensor-based behavioural information --- sign behavioural information is acquired by Intelligent hardware.
Currently, the Activity recognition technology of mainstream mainly identifies the behavioural information of view-based access control model on the market.Although
The acquisition modes of two kinds of behavioural informations can realize the real-time identification of human body behavior, but view-based access control model row by corresponding algorithm
For acquisition of information mode there are certain drawbacks, many behaviors in the more dim scene in the blind area of picture pick-up device or environment
Information can not obtain.Such as: when being fought in toilet, manager can not understand in toilet there is a situation where,
So carrying out Activity recognition by single visual behaviour information, there are many flaws.
In addition, the transmission technology of information is mainly in Activity recognition system with bluetooth, ZigBee, WiFi, 3G, 4G at present
It is main.The transmission technology of mainstream can not get both in transmission range and power consumption, therefore, to realize real-time Activity recognition, need
A kind of low-power consumption, long transmission distance transmission technology.Loa is the radio modem technology issued by Semtech company.It should
Technology has Lora-WAN agreement, Lora proprietary protocol, four CLASS agreement, data penetration transmission classifications, compared to other low-power consumption
Wide area networking of things network technology, the technology power consumption, in terms of exist very big advantage.
The model that system carries out Activity recognition mainly uses machine learning, deep learning scheduling algorithm.Currently, the machine of mainstream
Study mainly has k nearest neighbor algorithm (KNN), support vector machines (SVM), random forest, neural network etc.;The deep learning of mainstream is calculated
Method mainly has deep neural network (DNN), convolutional neural networks (CNN), circulation (recurrence) neural network (RNN), depth conviction
Network (DBN) etc..Mainstream algorithm be all according to behavioural information feature carry out Activity recognition, so input behavior information simply by the presence of
Certain feature will be divided into a certain class behavior.Such as: behavioural information 1100 and 0011, which respectively indicates, stands up and sits down, and 1010
Behavior of fighting is indicated with 0101, as long as the algorithm of mainstream may will be to fight there are two the behavioural information judgement of 1 and two 0 features
Behavior, so if it will all can be considered as the behavior of fighting that the feature of input behavior information, which is 0101,1100,0011 or 1010,
But actually 1100 or 0011 expression is behavior of standing up or sit down.This is mainly due to existing mainstream algorithms only to consider behavior
Whether information includes certain features, the spatial character without considering these features, so will cause erroneous judgement to a certain extent, is dropped
The accuracy rate of low Activity recognition.Geoffrey Hinton proposed the concept of capsule network in 2017, and capsule network not only can be with
Whether mark behavioural information has certain features, while can also identify the spatial relationship between each feature.The present invention uses
Capsule network algorithm model have higher promotion in terms of the accuracy rate of Activity recognition.
Summary of the invention
For the erroneous judgement of unicity, the particularity of transmission mode and prediction model that current behavioural information obtains, this hair
It is bright to propose a kind of real-time Activity recognition system based on Lora and Capsule.
The invention also discloses the working methods of above system.
Summary of the invention:
Real-time Activity recognition system based on Lora and Capsule, firstly, by equipment quality (QoD) parameter to intelligence
Hardware is screened and is designed, and is acquired by smart machine to behavioural information, then, behavioural information is passed through Lora node
It is transmitted, the base station Lora will receive behavioural information at this time.Since behavioural information at this time is primitive behavior information, system is needed
Uncertain detection is carried out to primitive behavior information, will be had incomplete or inconsistent information in behavioural information and passed through up and down
The methods of text prediction filling, benefit 0, deletion are handled, to improve the confidence level of behavioural information.Next, not true to process
The behavioural information of qualitative processing is standardized and the interception based on time series, and standardization is the accuracy rate in order to improve model
And generalization ability, it is normalization and the raising model for mode input by the interception that sliding window mechanism carries out behavioural information
Accuracy rate.It subsequently, is to be trained the behavioural information set with label under the network architecture model built, not
Best model is found while disconnected optimization loss function.The behavioural information acquired in real time is passed to model, realizes the real-time of behavior
Identification.Finally, system can according to user experience quality (QoE) and service quality (QoS) to preset threshold value and model parameter into
Row adjustment, improves the stability, accuracy rate and applicability of system.
The present invention provides a kind of feasible scheme for sensor-based real-time Activity recognition, compensates for based on video line
Caused defect when to identify, multi-source behavioural information are that the accuracy rate of raising Activity recognition is laid a good foundation.Based on Capsule
Application of the deep learning model of network in terms of Activity recognition also substantially increases the accuracy rate of Activity recognition.
The technical solution of the present invention is as follows:
A kind of real-time Activity recognition system based on Lora and Capsule, including sequentially connected behavioural information physical layer,
Behavioural information access layer, behavioural information podium level, behavioural information application layer;
The behavioural information physical layer is used for: it perceives, acquire from environment, storage, the behavioural information for transmitting user, behavior
Information includes: acceleration, angular speed, heart rate;
The behavioural information access layer is used for: the behavioural information of acquisition is carried out networking biography by low-power consumption wide area network
It is defeated;
The behavioural information podium level is used for: when successively carrying out uncertain detection, standardization to behavioural information and being based on
Between sequence interception, will be trained with the behavioural information set of label under the network architecture model built, constantly excellent
Best model is found while changing penalty values;Uncertainty detection refers to: by information incomplete in behavioural information or inconsistent
It is handled by context-prediction filling, benefit 0, delet method, improves the confidence level of behavioural information;Standardization is logarithm type
Data are normalized, to improve the accuracy rate and generalization ability of model;Interception based on time series, which refers to, to be passed through
The interception that sliding window mechanism carries out behavioural information improves the accuracy rate of model to guarantee the normalization of mode input;
The behavioural information application layer is used for: adjusting entire Activity recognition system stability and adaptivity in real time.
The present invention proposes one relatively at four aspects such as transmission technology, information processing, Activity recognition and behavior application
The system of optimization, compensate at present Activity recognition real-time on the market it is poor, can not be in the deficiency that specific region uses, in behavior knowledge
It is promoted in terms of other accuracy rate there has also been further, system is made to have stability.
Preferred according to the present invention, the behavioural information physical layer, that is, behavior information acquisition module, the behavioural information is adopted
Collecting module includes sensor module and several Intelligent hardware modules;The sensor module includes several different types of sensings
Device, the Intelligent hardware module are separately connected several different types of sensors, and the Intelligent hardware module is for controlling sensing
Device perceives the different types of behavioural information of user, and the behavioural information perceived is stored.
Son is carried out by QoD parameter, application scenarios and the user demand of each submodule in behavioural information acquisition module
The selection of module and the design of smart machine, QoD parameter mainly include the sample frequency, service life, precision of sensor module
Deng.
Intelligent hardware module uses the transmission that Lora carries out behavioural information.Transmission technology, which refers to, makes full use of different channels
Transmittability constitute a complete Transmission system, enable the technology of information reliable transmission.With social progress and nothing
The development of line technology, under the premise of packet loss is of less demanding, the convenience of wireless transmission is further amplified.Mainstream at present
Wireless technology mainly have a WiFi, bluetooth, ZigBee, 3G, 4G etc., each wireless technology transmission range and power consumption all in
The situation that can not be got both, but in order to realize real-time Activity recognition, the transmission of behavioural information needs a kind of low in energy consumption, transmission range
Remote wireless technology.Low-power consumption local area network (LPWAN) is the major technique for solving existing situation, therefore present invention employs Lora
Carry out the transmission of behavioural information.
Preferred according to the present invention, the behavioural information access layer, that is, behavior information transmission modular, the behavioural information passes
Defeated module includes behavioural information sending module and behavioural information receiving module;The behavioural information sending module connects the intelligence
Hardware module, for behavioural information to be sent to the behavioural information receiving module.
It is further preferred that the information sending module is Lora node, the behavioural information receiving module is Lora base
It stands.
It is preferred according to the present invention, the behavioural information podium level, that is, behavior information pre-processing module, the behavioural information
Preprocessing module includes sequentially connected behavioural information detection module, behavioural information uncertainty cancellation module, at behavioural information
Manage module, network architecture module;
The behavioural information detection module includes inconsistency detection/quantifying unit and incompleteness detection/quantifying unit;
The behavioural information uncertainty cancellation module includes that inconsistency eliminates unit and imperfection elimination unit;
The behavioural information processing module includes sequentially connected behavioural information Standardisation Cell, behavioural information sliding window list
Member;
The network architecture module includes sequentially connected convolution layer unit, Capsule layer Unit one, Capsule layer two
Unit connects layer unit entirely;
The behavioural information receiving module, that is, gateway connects the behavioural information detection module;
The behavioural information that the behavioural information receiving module receives i.e. primitive behavior information input is to the behavioural information
Detection module believes primitive behavior by the inconsistency detection/quantifying unit and incompleteness detection/quantifying unit
Breath carries out probabilistic detection, and the inconsistency detection/quantifying unit detects the different types of behavioural information of synchronization
With the presence or absence of objection, the incompleteness detection/quantifying unit detection synchronization perception behavioural information is with the presence or absence of loss;
If it find that behavioural information has uncertainty, then unit and the inconsistency are eliminated by the imperfection
It eliminates unit and carries out probabilistic elimination, the imperfection is eliminated unit and lost to existing for synchronization perception behavioural information
It loses situation to be handled by elimination method, 0 method of benefit, context-prediction completion method, the inconsistency eliminates unit to inconsistent
Information by way of vote by ballot, the QoD principle of optimality of hardware, D-S (Dempster-Shafer) Evidence, fuzzy set into
Row processing, into the behavioural information Standardisation Cell;If it find that behavioural information is then directly entered institute there is no uncertainty
State behavioural information Standardisation Cell;At the behavioural information Standardisation Cell and the behavioural information sliding window unit
Reason, the behavioural information Standardisation Cell are handled by standardization, method for normalizing, are improved recognition accuracy and are applicable in
Property;The behavioural information sliding window unit by adjust sliding window size and sliding window sliding type to behavioural information into
Interception of the row based on time series;
The behavioural information being disposed is input in trained network architecture model, network architecture model realization is passed through
Activity recognition;The convolution layer unit extracts feature to behavioural information, carries out the conversion of characteristic scalar to vector, described
Capsule layer Unit one is used to the behavioural information of input being converted into the behavioural information with spatial character;It is Capsule layers described
Unit two are handled behavioural information by dynamic routing protocol;Behavioural information feature is changed by the full connection layer unit
All features are carried out operation finally by Softmax classifier, identify current behavior by orderly one-dimensional characteristic.
Network architecture module groundwork is that identification is made according to behavioural information, in artificial intelligence, area of pattern recognition
In, what the proposition of machine learning can be practical realize the powerful of artificial intelligence, and the proposition of deep learning has in terms of discrimination
One significant progress.But whether machine learning model and the emphasis of deep learning model concern are wrapped in input information
Some characteristic values are contained.The network architecture based on Capsule used in the present invention is not only to feature possessed by behavioural information
It is paid close attention to, is also added into the spatial relationship of behavioural information feature, improve the accuracy rate of Activity recognition.
Behavior preprocessing module is mainly that the confidence level of behavioural information is improved by the pretreatment to behavioural information.It compares
Directly for primitive behavior information carries out Activity recognition, the present invention's some systems can exist after pre-processing by behavioural information
Stability and accuracy rate etc. have greatly improved.The present invention mainly carries out analysis of uncertainty to raw information, by right
The probabilistic classification of behavioural information and degree perform corresponding processing.In terms of information standardization, the present invention provides rule
Generalized method, method for normalizing.By adjusting the size of sliding window and the sliding type of sliding window to behavioural information progress
Interception based on time series.
Preferred according to the present invention, the behavioural information application layer includes behavioural information threshold setting module, behavior application
Layer adjustment module, the behavior application layer adjustment module includes sequentially connected Activity recognition unit, user feedback unit, mistake
Amending unit;
Behavioural information threshold setting module is used to adjust the threshold value in behavioural information uncertainty cancellation module, thus to prison
Measured data adjusts Uncertainty Management module selection data processing method with the presence or absence of uncertainty;Activity recognition unit is used
In doing real-time identification to current behavior;The user feedback unit according to different scenes and user demand to preset threshold value and
The parameter of network architecture module is adjusted, and improves the applicability of system to a certain extent;Error correction unit is continuous
Network architecture module is adjusted, network architecture module is allowed to be constantly in optimum state.
Parameter adjustment, the user feedback list are carried out when higher to identification error rate by the error correction unit
First user is by regulating system parameter to be suitable for different scenes.
Above-mentioned real-time Activity recognition system and its working method based on Lora and Capsule, comprises the following steps that
Step S01: sensor perceives behavioural information
The QoD choice of parameters different manufacturers of primitive behavior information, sensor according to required for Activity recognition, different type
Sensor, the QoD parameter of sensor includes: sample frequency, service life, precision, such as: the use for needing important mornitoring
Family can carry out the perception of behavioural information using sample frequency height, sensor with high accuracy, and general user can be adopted
Sample frequency is common, precision ordinary sensors carry out the perception of behavioural information;Sensor perceives the different types of behavioural information of user;
Step S02: design Intelligent hardware module
It is required to select suitable Intelligent hardware module according to scheme, each sensor is controlled by Intelligent hardware module, and
Behavioural information needed for acquiring Activity recognition system;
Step S03: the transmission of behavioural information
According to the requirement of real-time Activity recognition, some wireless transmission methods of present mainstream can be excluded, such as 3G, 4G,
ZigBee, bluetooth etc..Low-power consumption wide area network is a kind of more suitable transmission mode, since NB-IoT will put into civilian, is examined
Consider privacy, this system carries out the transmission of behavioural information using Lora node;
Step S04: the reception of behavioural information
Using the base station Lora reception behavior features information;The sending device of behavioural information uses Lora node, corresponding
The reception of behavioural information uses the base station Lora.
Step S05: the uncertain detection of behavioural information
Behavioural information threshold range is set, such as the accuracy of setting behavioural information is not less than 85%, when behavioural information
When accuracy is lower than 85%, then it is assumed that information is uncertain behavioural information, by inconsistency detection/quantifying unit, endless
Standby property detection/quantifying unit successively carries out inconsistency detection/quantization, incompleteness detection/quantization to primitive behavior information, obtains
To testing result, when primitive behavior information there are it is inconsistent, incomplete when, execute step S05, otherwise, execute step S06;It is former
Beginning behavioural information refers to the different types of behavioural information of user of step S01 sensor perception;
Step S06: the uncertain of behavioural information is eliminated
Incompleteness eliminates the threshold value that unit is detected by behavioural information uncertainty, and different methods is selected to believe behavior
Breath handled, when the accuracy of behavioural information be 85%-90% when, for behavioural information using context-prediction completion method into
Row processing is handled for behavioural information using 0 method of mending when the accuracy of behavioural information is 90%-95%, when behavior is believed
When the accuracy of breath is 95%-100%, handled for behavioural information using elimination method;
Inconsistency eliminate unit inconsistent information is handled, processing method include vote by ballot, hardware QoD most
Excellent principle, D-S (Dempster-Shafer) Evidence, fuzzy set;Improve the confidence level of primitive behavior information;
Step S07: the processing of behavioural information
The higher behavioural information of confidence level is standardized by behavioural information Standardisation Cell;The standard of behavioural information
Change and use different standardized ways for different types of data, comprising: for the data of classification type feature, is compiled using only heat
Code (one-hot coding) standardization;For the data of numeric type feature, standardized using normalized;For order type spy
The data of sign are standardized using order type numeric coding;Standardization can allow system to have good scalability.
Referring to the parameter of user preset, the parameter of user preset includes: the size of sliding window and the sliding type of window,
Sliding window processing is carried out to the behavioural information after standardization by behavioural information sliding window unit, behavioural information is made to become defeated
Enter the block of information of network structure module;
Step S08: the behavioural information network architecture
Pass through convolution layer unit, Capsule layer Unit one, Capsule layer Unit two, full articulamentum building unit one four
The network architecture model of layer, referring to the parameter of user setting, the parameter of user setting specifically includes that the status of input data, big
The series of parameters such as small, convolutional layer core size, number are trained the behavioural information with label by iteration several times,
The association of the dynamic routing in model parameter and Capsule layer unit is continued to optimize in training process by reducing loss function
View, finally obtains the high network architecture model of discrimination;
Step S09: the identification of behavioural information
The behavioural information acquired in real time is input to the reality that current behavior is carried out in trained network architecture model
When identify;
Step S10: error detection
Judge that current behavior identifies whether that mistake occurs, if there are mistakes for discovery, thens follow the steps S11, otherwise execute step
Rapid S12;
Step S11: error correction
Error correction unit is adjusted the relevant parameter of behavioural information threshold range, behavioural information processing module;Row
It include the threshold range of uncertain detection for information threshold range, the relevant parameter of behavioural information processing module includes behavior letter
Cease the size of the sliding window in sliding window unit and the sliding type of window;When identification mistake is more, raising behavior appropriate
The threshold range of information, while the sliding type of the size of sliding window and window being become smaller.
Step S12: user feedback detection
Judgement system whether there is field feedback, such as there is feedback information, thens follow the steps S13.
Step S13: user feedback
User feedback unit carries out feedback tune to the relevant parameter to behavioural information threshold range, behavioural information processing module
It is whole.
It is preferred according to the present invention, the step S08,
The network architecture module includes sequentially connected convolution layer unit, Capsule layer Unit one, Capsule layer two
Unit connects layer unit entirely;
It is N that convolution kernel number in convolution layer unit, which is arranged,1, each convolution kernel size is 1 × Nuclear_Size1, step-length is
L1;
It is N that convolution kernel number in Capsule layer Unit one, which is arranged,2, each convolution kernel size is 1 × Nuclear_Size2,
Step-length is L2;
The output length that Capsule layer Unit two is arranged is that Num_Output ties up behavioural information, and each dimension uses Vec_
Lenv behavioural information feature;
Output length is Output_Length in the full connection layer unit of setting;
It comprises the following steps that
(1) behavioural information of a size of Batch_Size × 1 × Window_Size × 3 is inputted, Batch_Size refers to
The number of the behavioural information once run in network architecture module, Window_Size refer to input network architecture module every time
Length;
(2) after the behavioural information of the size of Batch_Size × 1 × Window_Size × 3 passes through convolution layer unit, lead to
It crosses formula (I) and the behavioural information of input is converted into vector by scalar:
In formula (I), XiRefer to behavioural information by uncertain, standardization, the sliding window processing based on time series
Each information later;WijRefer to the weight parameter of convolution layer unit, initial value is defaulted as generating the random of cutting gearbox
Number;
bjRefer to that the offset parameter of convolution layer unit, initial value are defaulted as 0.0;
The number of n expression convolution kernel;
YjIt is to indicate that convolutional layer exports;
Output information size are as follows:Wherein need
The result for guaranteeing preceding formula mid-score is positive integer.Output at this time is the result is that a vector behavioural information, meets
The input requirements of Capsule network;
(3) it enablesThe above M group convolutional layer is encapsulated in
In Capsule network, by the behavioural information Y of vectorjIt is input to Capsule layer Unit one, by formula (II) by the behavior of input
Information is converted into the behavioural information with spatial character;
In formula (II), WjlRefer to the weight parameter of Capsule layer Unit one, initial value is defaulted as generating cutting gearbox
Random number;
blRefer to that the offset parameter of Capsule layer Unit one, initial value are defaulted as 0.0;
Squsah () function is a kind of new nonlinear function, non-thread similar to tanh () common before, relu () etc.
Property function, squsah () function is the Nonlinear Processing towards Vector Message;And other nonlinear functions are primarily directed to scalar
The processing of information;
Refer to the vector behavioural information feature of Capsule network output;
The information size exported after Capsule layer Unit one are as follows:
(4) behavioural information with spatial character is input to Capsule layer Unit two, passes through dynamic routing protocol, that is, formula
(III), (IV) handles behavioural information;
In formula (III), (IV),
bikRefer in Capsule layer Unit one in i-th of neuron and Capsule layer Unit two the dynamic of k-th of neuron
State routing weights;
bijRefer in Capsule layer Unit one in i-th of neuron and Capsule layer Unit two the dynamic of j-th of neuron
State routing weights;
Refer to each Capsule layers of output;
SjRefer to behavioural information feature of the Capsule layer Unit two by output after dynamic routing protocol.
It is the vector output for pointing out the network architecture;
The information size exported after two cell processing of Capsule layer are as follows: Batch_Size × Num_Output × Vec_
Lenv×1;
(5) by full connection layer unit by behavioural information by vector median filters at scalar;
The information size exported after connecting layer unit entirely are as follows:
Batch_Size×Output_Length×1;
(6) Softmax classifier is added, the Classification and Identification of behavioural information is carried out by Softmax classifier;By that will believe
It ceases the behavioural information feature that size is Batch_Size × Output_Length × 1 and carries out each behavior probability by classifier
Solution, find out the corresponding maximum behavior of all kinds of probability numbers, the final recognition result, that is, probability numbers of network architecture module
Maximum behavior.
The invention has the benefit that
1, practicability:
Real-time body's Activity recognition has higher requirement for transmission medium and accuracy rate, and the present invention realizes well
The low-power consumption of behavioural information, remote transmission;Also has certain advantage in terms of the accuracy rate of Activity recognition simultaneously.
2, adaptivity:
For different application scenarios, by user feedback unit (QoE) and error correction unit (QoS) in system
Parameter is adjusted, and improves the adaptivity of system, provides personalized, intelligentized service for user.Wherein adjustable ginseng
Number includes: inconsistency detection/quantifying unit threshold value and incompleteness detection/quantifying unit threshold value, behavioural information standard
Change the mode of the standardized way in unit and sliding window size and sliding in behavioural information sliding window unit, network architecture mould
The parameters such as the number of iterations, learning rate and training the number of iterations in block.
3, high reliability:
View-based access control model behavioural information single piece of information source is after being added sensor-based behavioural information, Activity recognition system
It can be more perfect;Activity recognition is carried out in contrast with the model algorithm of mainstream simultaneously, and the present invention has in terms of accuracy rate
It is further to be promoted.Also there is good stability in terms of real-time.
Detailed description of the invention
Fig. 1 is that the present invention is based on the real-time Activity recognition system main module structural framings of Lora and Capsule and connection to close
It is schematic diagram.
Fig. 2 is that the module composition realized the present invention is based on the real-time Activity recognition system of Lora and Capsule and connection are closed
It is schematic diagram.
Fig. 3 is the real-time Activity recognition working-flow schematic diagram the present invention is based on Lora and Capsule.
Fig. 4 is the Activity recognition schematic diagram of the real-time Activity recognition system the present invention is based on Lora and Capsule.
Fig. 5 is that the present invention is based on Capsule layers one in the Activity recognition of the real-time Activity recognition system of Lora and Capsule
Unit, two cell operation schematic illustration of Capsule layer.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments and specification
Attached drawing 1-5 carries out clear, complete description to technical solution of the present invention, it is clear that specific embodiment described herein is only used
To explain the present invention, it is not intended to limit the present invention.
Embodiment 1
A kind of real-time Activity recognition system based on Lora and Capsule, as shown in Figure 1, including sequentially connected behavior
Information physical layer, behavioural information access layer, behavioural information podium level, behavioural information application layer;
Behavioural information physical layer is used for: it perceives, acquire from environment, storage, the behavioural information for transmitting user, behavioural information
It include: acceleration, angular speed, heart rate;
Behavioural information access layer is used for: by the behavioural information of acquisition by the Lora technology in low-power consumption wide area network into
Row networking transmission;
Behavioural information podium level is used for: uncertain detection, standardization is successively carried out to behavioural information and based on time sequence
The interception of column will be trained under the network architecture model built with the behavioural information set of label, continue to optimize damage
Best model is found while mistake value;Uncertainty detection refers to: information incomplete in behavioural information or inconsistent is passed through
Context-prediction filling, benefit 0, delet method are handled, and the confidence level of behavioural information is improved;Standardization is logarithm type data
It is normalized, to improve the accuracy rate and generalization ability of model;Interception based on time series, which refers to, passes through sliding
The interception that windowing mechanism carries out behavioural information improves the accuracy rate of model to guarantee the normalization of mode input;
Behavioural information application layer is used for: adjusting entire Activity recognition system stability and adaptivity in real time.
The present invention proposes one relatively at four aspects such as transmission technology, information processing, Activity recognition and behavior application
The system of optimization, compensate at present Activity recognition real-time on the market it is poor, can not be in the deficiency that specific region uses, in behavior knowledge
It is promoted in terms of other accuracy rate there has also been further, system is made to have stability.
Embodiment 2
According to a kind of real-time Activity recognition system based on Lora and Capsule described in embodiment 1, as shown in Fig. 2, its
Difference is,
Behavioural information physical layer, that is, behavior information acquisition module, behavioural information acquisition module include sensor module and several
A Intelligent hardware module;Sensor module includes several different types of sensors, Intelligent hardware module be separately connected it is several not
The sensor of same type, Intelligent hardware module is for controlling the sensor perception different types of behavioural information of user, and perception
To behavioural information stored.
Son is carried out by QoD parameter, application scenarios and the user demand of each submodule in behavioural information acquisition module
The selection of module and the design of smart machine, QoD parameter mainly include the sample frequency, service life, precision of sensor module
Deng.
Intelligent hardware module uses the transmission that Lora carries out behavioural information.Transmission technology, which refers to, makes full use of different channels
Transmittability constitute a complete Transmission system, enable the technology of information reliable transmission.With social progress and nothing
The development of line technology, under the premise of packet loss is of less demanding, the convenience of wireless transmission is further amplified.Mainstream at present
Wireless technology mainly have a WiFi, bluetooth, ZigBee, 3G, 4G etc., each wireless technology transmission range and power consumption all in
The situation that can not be got both, but in order to realize real-time Activity recognition, the transmission of behavioural information needs a kind of low in energy consumption, transmission range
Remote wireless technology.Low-power consumption local area network (LPWAN) is the major technique for solving existing situation, therefore present invention employs Lora
Carry out the transmission of behavioural information.
Behavioural information access layer, that is, behavior information transmission modular, behavioural information transmission module include behavioural information sending module
With behavioural information receiving module;Behavioural information sending module connects the Intelligent hardware module, for behavioural information to be sent to
Behavioural information receiving module.
Information sending module is Lora node, and the behavioural information receiving module is the base station Lora.In view of real-time and field
The demand of scape, the present invention choose Lora node and the base station Lora as transmission medium.
Behavioural information podium level, that is, behavior information pre-processing module, behavioural information preprocessing module include sequentially connected row
For information detecting module, behavioural information uncertainty cancellation module, behavioural information processing module, network architecture module;
Behavioural information detection module includes inconsistency detection/quantifying unit and incompleteness detection/quantifying unit;Behavior
Information uncertainty cancellation module includes that inconsistency eliminates unit and imperfection elimination unit;Behavioural information processing module packet
Include sequentially connected behavioural information Standardisation Cell, behavioural information sliding window unit;Network architecture module includes sequentially connected volume
Lamination unit, Capsule layer Unit two, connects layer unit at Capsule layer Unit one entirely;
Behavioural information receiving module, that is, gateway connects behavioural information detection module;
The behavioural information that behavioural information receiving module receives i.e. primitive behavior information input to behavioural information detection module,
Primitive behavior information is carried out by inconsistency detection/quantifying unit and incompleteness detection/quantifying unit probabilistic
Detection, inconsistency detection/quantifying unit detection different types of behavioural information of synchronization whether there is objection, incompleteness
Detection/quantifying unit detection synchronization perception behavioural information is with the presence or absence of loss;
If it find that behavioural information has uncertainty, then unit is eliminated by imperfection and the inconsistency is eliminated
Unit carries out probabilistic elimination, and it is logical to loss situation existing for synchronization perception behavioural information that imperfection eliminates unit
It crosses elimination method, 0 method of benefit, context-prediction completion method to be handled, inconsistency eliminates unit and passes through ballot to inconsistent information
Election, the QoD principle of optimality of hardware, D-S (Dempster-Shafer) Evidence, fuzzy set mode handled, into row
For information standardization unit;QoC index refers to the index for describing behavior information quality, including integrality, confidence level and more
New degree etc.;If it find that behavioural information is then directly entered behavioural information Standardisation Cell there is no uncertainty;Believed by behavior
Breath Standardisation Cell and behavioural information sliding window unit are handled, and behavioural information Standardisation Cell passes through standardization, normalization side
Method is handled, and recognition accuracy and applicability are improved;The size and cunning that behavioural information sliding window unit passes through adjusting sliding window
The sliding type of dynamic window carries out the interception based on time series to behavioural information;
The behavioural information being disposed is input in trained network architecture model, network architecture model realization is passed through
Activity recognition;The convolution layer unit to behavioural information extract feature, carry out characteristic scalar to vector conversion, Capsule layers
Unit one is used to the behavioural information of input being converted into the behavioural information with spatial character;Capsule layer Unit two passes through dynamic
State Routing Protocol handles behavioural information;Behavioural information feature is changed into orderly one-dimensional characteristic by full connection layer unit,
All features are subjected to operation finally by Softmax classifier, identify current behavior.
Network architecture module groundwork is that identification is made according to behavioural information, in artificial intelligence, area of pattern recognition
In, what the proposition of machine learning can be practical realize the powerful of artificial intelligence, and the proposition of deep learning has in terms of discrimination
One significant progress.But whether machine learning model and the emphasis of deep learning model concern are wrapped in input information
Some characteristic values are contained.The network architecture based on Capsule used in the present invention is not only to feature possessed by behavioural information
It is paid close attention to, is also added into the spatial relationship of behavioural information feature, improve the accuracy rate of Activity recognition.
Behavior preprocessing module is mainly that the confidence level of behavioural information is improved by the pretreatment to behavioural information.It compares
Directly for primitive behavior information carries out Activity recognition, the present invention's some systems can exist after pre-processing by behavioural information
Stability and accuracy rate etc. have greatly improved.The present invention mainly carries out analysis of uncertainty to raw information, by right
The probabilistic classification of behavioural information and degree perform corresponding processing.In terms of information standardization, the present invention provides rule
Generalized method, method for normalizing.By adjusting the size of sliding window and the sliding type of sliding window to behavioural information progress
Interception based on time series.
Behavioural information application layer includes behavioural information threshold setting module, behavior application layer adjustment module, behavior application layer
Adjusting module includes sequentially connected Activity recognition unit, user feedback unit, error correction unit;
Behavioural information threshold setting module is used to adjust the threshold value in behavioural information uncertainty cancellation module, thus to prison
Measured data adjusts Uncertainty Management module selection data processing method with the presence or absence of uncertainty;Activity recognition unit is used
In doing real-time identification to current behavior;The user feedback unit according to different scenes and user demand to preset threshold value and
The parameter of network architecture module is adjusted, and improves the applicability of system to a certain extent;Error correction unit is continuous
Network architecture module is adjusted, network architecture module is allowed to be constantly in optimum state.Activity recognition unit is mainly according to behavioural information
Activity recognition is carried out with reasonable model;QoS index value refers to service quality, generates corresponding adjustment letter according to service quality
Breath, then feeds back to behavioural information preprocessing module;User is obtained to the QoE index value of entire application service, and generates feedback
Information is transferred to behavioural information preprocessing module;QoE index value is for expressing user to the user of application service satisfactory degree
Score index, main function are to adjust preset QoC index value.
Embodiment 3
Real-time Activity recognition system and its working method described in embodiment 2 based on Lora and Capsule, as shown in figure 3,
By taking Activity recognition of fighting as an example, criminal in prison may all slightly different with ordinary person in terms of psychology and physiology, right
It might have extreme behavior to problem angle.Aggressive behavior causes serious influence and harm in order to prevent, and system is by adding
Velocity sensor S1, angular-rate sensor S2, heart rate sensor S3 obtain behavioural information of the criminal in one day, by information
The confidence level that information is improved after pretreatment, then carries out real-time Activity recognition by trained model.Prison administration person can
Real-time Activity recognition is carried out so that different parameters to be arranged according to different scenes and different criminals.Specific steps include:
Step S01: QoD parameter is obtained
Main QoD parameter includes: the precision of sensor, the material in sampling interval and bracelet.The precision of sensor is distinguished
It is 0.94,0.80,0.88, the sampling interval is respectively 0.02s-1s, and material is mainly made of materials such as rubber, alloys.
Step S02: design acquisition equipment
It according to sensor QoD parameter, the demand of criminal and puts in prison grade and carries out the design of Intelligent hardware, for emphasis
Supervised entities, the design that behavioural information acquires equipment can be using sample frequency is high, accuracy of identification is high, survivable material
Matter, for showing good, slight supervised entities, design it is contemplated that, identification lower using sample frequency it is common, cost
Slightly lower material is designed.
Step S03: the transmission of behavioural information
According to confidentiality and the size of scope of activities, the transmission of behavioural information is carried out using Lora node;
Step S04: the reception of behavioural information
Using the base station Lora reception behavior features information;The sending device of behavioural information uses Lora node, corresponding
The reception of behavioural information uses the base station Lora.
Step S05: the uncertain detection of behavioural information
Incompleteness detection/quantifying unit sets a threshold to 0.85, i.e., has 85% in the behavioural information per second received
Loss of learning then think primitive behavior information there are incomplete;
Inconsistency detection/quantifying unit sets a threshold to 0.8, i.e., the behavioural information similarity per second received is lower than
Then determine that there are inconsistent for this group of primitive behavior information when 0.8.
Inconsistency detection/quantization, incomplete is carried out to primitive behavior information according to the behavioural information threshold range of setting
Property detection/quantization primitive behavior information is analyzed, there are inconsistent, incomplete etc. uncertain for discovery primitive behavior information
Property when, execute step S06, it is no to then follow the steps S07.
Step S06: the uncertain of behavioural information is eliminated
If it was found that primitive behavior information incomplete behavioural information can be deleted there are when incompleteness, in system or
Incomplete information is carried out benefit 0 or contextual information prediction to be filled incomplete behavioural information, system default choosing
Contextual information prediction is selected to be filled incomplete behavioural information;
If it was found that primitive behavior information, there are inconsistency, system can be by inconsistency information according to vote by ballot principle
It modifies or inconsistent information is modified according to the QoD principle of optimality of acquisition of information hardware or by inconsistent information
It finds out confidence level using the method for D-S Evidence to modify, system default selection is by inconsistency information according to vote by ballot
Principle is modified.
After the uncertain elimination of behavioural information, after greatly improving the confidence level of primitive behavior information and being
The identification of processing and the behavior of continuous behavioural information provides reliability.
Step S07: the processing of behavioural information
The Standardisation Cell of behavioural information is mainly standardized same class behavioural information, workable standard in system
Changing is mainly normalization method or method for normalizing, and system default is standardized as method for normalizing;
Behavioural information sliding window unit mainly carries out the interception based on time series to behavioural information, and system mainly provides cunning
The size of dynamic window and the two class parameter of mode of sliding, the size of sliding window have 40,60,80,100, and sliding type mainly has
Sliding based on half the time sequence and the sliding based on All Time sequence, the size of the sliding window of system default are 80,
Sliding type is the sliding based on half the time sequence.
Step S08: the behavioural information network architecture
Pass through convolution layer unit, Capsule layer Unit one, Capsule layer Unit two, full articulamentum building unit one four
The network architecture module of layer carries out the behavioural information with label by n times iteration referring to some parameters of user preset
It trains, constantly optimizes the dynamic routing that loss function comes in optimization module parameter and Capsule layer unit in training process
Agreement finally obtains the higher module of discrimination.Wherein training set can choose proprietary behavioural information, can also choose certain
The behavioural information of one people is trained module, then to a certain personal progress Activity recognition.Due to the program need it is huge
Behavioural information and the biggish resource of needs are supported, it is proposed that are carried out just for moiety by weight grade criminal using system default is chosen whole
The behavioural information database of a prison criminal is trained module.As shown in figure 4, the specific implementation of module employed in this example
Process is as follows:
It is 256 that convolution kernel number in convolution layer unit, which is arranged, and each convolution kernel size is 1 × 41, step-length 1;
It is 32 that convolution kernel number in Capsule layer Unit one, which is arranged, and each convolution kernel size is 1 × 21, step-length 2;
The output length that Capsule layer Unit two is arranged is 8 dimension behavioural informations, and each dimension uses 16 behavioural information spies
Sign;
It is 6 that the length exported in full connection layer unit, which is arranged,;
It comprises the following steps that
(1) behavioural information of 5 × 1 × 80 × 3 sizes is inputted;
(2) after the behavioural information of 5 × 1 × 80 × 3 sizes passes through convolution layer unit, by formula (I) by the row of input
Vector is converted by scalar for information:
In formula (I), XiRefer to behavioural information by uncertain, standardization, the sliding window processing based on time series
Each information later;WijRefer to the weight parameter of convolution layer unit, initial value is defaulted as generating the random of cutting gearbox
Number;
bjRefer to that the offset parameter of convolution layer unit, initial value are defaulted as 0.0;
The number of n expression convolution kernel;
YjIt is to indicate that convolutional layer exports;
Output information size are as follows: 5 × 1 × 40 × 256;The result for guaranteeing preceding formula mid-score is wherein needed to be positive whole
Number.Output at this time is the result is that a vector behavioural information, meets the input requirements of Capsule network;
(3) as shown in figure 5, above 8 groups of convolutional layers are encapsulated in Capsule, the result that convolution layer unit is exported is defeated
Enter to Capsule layer Unit one, the behavioural information of input is converted into the behavioural information with spatial character by formula (II);
In formula (II), WjlRefer to the weight parameter of Capsule layer Unit one, initial value is defaulted as generating cutting gearbox
Random number;
blRefer to that the offset parameter of Capsule layer Unit one, initial value are defaulted as 0.0;
Squsah () function is a kind of new nonlinear function, non-thread similar to tanh () common before, relu () etc.
Property function, squsah () function is the Nonlinear Processing towards Vector Message;And other nonlinear functions are primarily directed to scalar
The processing of information;
Refer to the vector behavioural information feature of Capsule network output;
The information size exported after Capsule layer Unit one are as follows: 5 × 320 × 8 × 1;
(4) using Capsule layer Unit one output result as the input information of Capsule layer Unit two, pass through dynamic road
Behavioural information is handled by agreement, that is, formula (III), (IV);
In formula (III), (IV),
bikRefer in Capsule layer Unit one in i-th of neuron and Capsule layer Unit two the dynamic of k-th of neuron
State routing weights;
bijRefer in Capsule layer Unit one in i-th of neuron and Capsule layer Unit two the dynamic of j-th of neuron
State routing weights;
Refer to each Capsule layers of output;
SjRefer to behavioural information feature of the Capsule layer Unit two by output after dynamic routing protocol.
It is the vector output for pointing out the network architecture;
The information size exported after two cell processing of Capsule layer are as follows: 5 × 12 × 16 × 1;
(5) by full connection layer unit by behavioural information by vector median filters at scalar;
The information size exported after connecting layer unit entirely are as follows: 5 × 192 × 1;
(6) Softmax classifier is added, the Classification and Identification of behavioural information is carried out by Softmax classifier;By that will believe
The behavioural information feature that breath size is 5 × 192 × 1 carries out the solution of each behavior probability by classifier, finds out corresponding each
The maximum behavior of class probability numbers, the final maximum behavior of recognition result, that is, probability numbers of network architecture module.
System customized parameter mainly has the parameters such as dynamic routing the number of iterations, learning rate and training the number of iterations, moves
State route iteration number is set as 1-10;Learning rate is set as 0.1,0.01,0.001;With setting for training the number of iterations
It is set to 1-50.System default parameter is followed successively by 5,0.01,40.
Step S09: the identification of behavioural information
The behavioural information acquired in real time is input to the reality that current behavior is carried out in trained network architecture model
When identify;
Step S10: error detection
Judge that current behavior identifies whether that mistake occurs, if there are mistakes for discovery, thens follow the steps S11, otherwise execute step
Rapid S12;
Step S11: error correction
Error correction unit is adjusted the relevant parameter of behavioural information threshold range, behavioural information processing module;Row
It include the threshold range of uncertain detection for information threshold range, the relevant parameter of behavioural information processing module includes behavior letter
Cease the size of the sliding window in sliding window unit and the sliding type of window;When identification mistake is more, raising behavior appropriate
The threshold range of information, while the sliding interval of the size of sliding window and window being become smaller;
Step S12: user feedback detection
Judgement system whether there is field feedback, such as there is feedback information, thens follow the steps S13.
Step S13: user feedback
A feedback information is carried out according to varying environment locating for different user, some parameters that adjustment user preset is set are come
Adjust behavioural information processing module and network architecture module.Adjustable parameter includes: in primitive behavior information detecting module
Inconsistency detection/quantifying unit threshold value and incompleteness detection/quantifying unit threshold value, in behavioural information processing module
Behavioural information Standardisation Cell in standardized way and behavioural information sliding window unit in sliding window size and sliding
Mode, the number of iterations, learning rate and training the number of iterations in network architecture module etc..
Claims (8)
1. a kind of real-time Activity recognition system based on Lora and Capsule, which is characterized in that believe including sequentially connected behavior
Cease physical layer, behavioural information access layer, behavioural information podium level, behavioural information application layer;
The behavioural information physical layer is used for: it perceives, acquire from environment, storage, the behavioural information for transmitting user, behavioural information
It include: acceleration, angular speed, heart rate;
The behavioural information access layer is used for: the behavioural information of acquisition is carried out networking transmission by low-power consumption wide area network;
The behavioural information podium level is used for: uncertain detection, standardization is successively carried out to behavioural information and based on time sequence
The interception of column will be trained under the network architecture model built with the behavioural information set of label, continue to optimize damage
Best model is found while mistake value;Uncertainty detection refers to: information incomplete in behavioural information or inconsistent is passed through
Context-prediction filling, benefit 0, delet method are handled, and the confidence level of behavioural information is improved;Standardization is logarithm type data
It is normalized;Interception based on time series refers to the interception that behavioural information is carried out by sliding window mechanism;
The behavioural information application layer is used for: adjusting entire Activity recognition system stability and adaptivity in real time.
2. a kind of real-time Activity recognition system based on Lora and Capsule according to claim 1, which is characterized in that
Behavioural information physical layer, that is, behavior the information acquisition module, the behavioural information acquisition module include sensor module and several
A Intelligent hardware module;The sensor module includes several different types of sensors, and the Intelligent hardware module connects respectively
Several different types of sensors are connect, the Intelligent hardware module is for controlling the different types of behavior letter of sensor perception user
Breath, and the behavioural information perceived is stored.
3. a kind of real-time Activity recognition system based on Lora and Capsule according to claim 2, which is characterized in that
Behavioural information access layer, that is, behavior the information transmission modular, the behavioural information transmission module include behavioural information sending module
With behavioural information receiving module;The behavioural information sending module connects the Intelligent hardware module, for sending out behavioural information
It send to the behavioural information receiving module.
4. a kind of real-time Activity recognition system based on Lora and Capsule according to claim 3, which is characterized in that
The information sending module is Lora node, and the behavioural information receiving module is the base station Lora.
5. a kind of real-time Activity recognition system based on Lora and Capsule according to claim 3, which is characterized in that
The behavioural information podium level, that is, behavior information pre-processing module, the behavioural information preprocessing module include sequentially connected row
For information detecting module, behavioural information uncertainty cancellation module, behavioural information processing module, network architecture module;
The behavioural information detection module includes inconsistency detection/quantifying unit and incompleteness detection/quantifying unit;
The behavioural information uncertainty cancellation module includes that inconsistency eliminates unit and imperfection elimination unit;
The behavioural information processing module includes sequentially connected behavioural information Standardisation Cell, behavioural information sliding window unit;
The network architecture module include sequentially connected convolution layer unit, Capsule layer Unit one, Capsule layer Unit two,
Full connection layer unit;
The behavioural information receiving module, that is, gateway connects the behavioural information detection module;
The behavioural information that the behavioural information receiving module receives i.e. primitive behavior information input is detected to the behavioural information
Module, by the inconsistency detection/quantifying unit and incompleteness detection/quantifying unit to primitive behavior information into
The probabilistic detection of row, whether the inconsistency detection/quantifying unit detection different types of behavioural information of synchronization
There are objection, the incompleteness detection/quantifying unit detection synchronization perception behavioural information is with the presence or absence of loss;
If it find that behavioural information has uncertainty, then unit is eliminated by the imperfection and the inconsistency is eliminated
Unit carries out probabilistic elimination, and the imperfection eliminates unit and loses feelings to existing for synchronization perception behavioural information
Condition is handled by elimination method, 0 method of benefit, context-prediction completion method, and the inconsistency eliminates unit to inconsistent information
It is handled by way of vote by ballot, the QoD principle of optimality of hardware, D-S Evidence, fuzzy set, is believed into the behavior
Cease Standardisation Cell;If it find that behavioural information is then directly entered the behavioural information Standardisation Cell there is no uncertainty;
It is handled by the behavioural information Standardisation Cell and the behavioural information sliding window unit, the behavioural information standardization is single
Member is handled by standardization, method for normalizing, improves recognition accuracy and applicability;The behavioural information sliding window unit is logical
The sliding type of the size and sliding window of overregulating sliding window carries out the interception based on time series to behavioural information;
The behavioural information being disposed is input in trained network architecture model, network architecture model realization behavior is passed through
Identification;The convolution layer unit extracts feature to behavioural information, carries out the conversion of characteristic scalar to vector, Capsule layers described
Unit one is used to the behavioural information of input being converted into the behavioural information with spatial character;Capsule layer Unit two is logical
Dynamic routing protocol is crossed to handle behavioural information;Behavioural information feature is changed into orderly one by the full connection layer unit
All features are carried out operation finally by Softmax classifier, identify current behavior by dimensional feature.
6. a kind of real-time Activity recognition system based on Lora and Capsule according to claim 5, which is characterized in that
The behavioural information application layer includes behavioural information threshold setting module, behavior application layer adjustment module, and the behavior is answered
It include sequentially connected Activity recognition unit, user feedback unit, error correction unit with layer adjustment module;
Behavioural information threshold setting module is used to adjust the threshold value in behavioural information uncertainty cancellation module, thus to monitoring number
It is uncertain according to whether there is, and adjust Uncertainty Management module selection data processing method;Activity recognition unit for pair
Current behavior does real-time identification;The user feedback unit is according to different scenes and user demand to preset threshold value and network
The parameter of structure module is adjusted, and error correction unit constantly adjusts network architecture module, allow network architecture module always
In optimum state.
7. the working method of the real-time Activity recognition system as claimed in claim 6 based on Lora and Capsule, feature exist
In comprising the following steps that
Step S01: sensor perceives behavioural information
The QoD choice of parameters different manufacturers of primitive behavior information, sensor according to required for Activity recognition, different types of biography
Sensor, the QoD parameter of sensor include: sample frequency, service life, precision, and sensor perceives the different types of behavior of user
Information;
Step S02: design Intelligent hardware module
It is required to select suitable Intelligent hardware module according to scheme, each sensor is controlled by Intelligent hardware module, and acquire
Behavioural information needed for Activity recognition system;
Step S03: the transmission of behavioural information
The transmission of behavioural information is carried out using Lora node;
Step S04: the reception of behavioural information
Using the base station Lora reception behavior features information;
Step S05: the uncertain detection of behavioural information
Set behavioural information threshold range, when primitive behavior information there are it is inconsistent, incomplete when, execute step S05, otherwise,
Execute step S06;Primitive behavior information refers to the different types of behavioural information of user of step S01 sensor perception;
Step S06: the uncertain of behavioural information is eliminated
Incompleteness eliminates the threshold value that detects by behavioural information uncertainty of unit, select different methods to behavioural information into
Row processing, when the accuracy of behavioural information is 85%-90%, for behavioural information using at context-prediction completion method
Reason is handled for behavioural information using 0 method of mending, when the accuracy of behavioural information is 90%-95% when behavioural information
When accuracy is 95%-100%, handled for behavioural information using elimination method;
Inconsistency is eliminated unit and is handled inconsistent information, and processing method includes the optimal original of QoD of vote by ballot, hardware
Then, D-S Evidence, fuzzy set;Improve the confidence level of primitive behavior information;
Step S07: the processing of behavioural information
The higher behavioural information of confidence level is standardized by behavioural information Standardisation Cell;The standardization needle of behavioural information
Different standardized ways is used to different types of data, comprising: for the data of classification type feature, using one-hot coding mark
Standardization;For the data of numeric type feature, standardized using normalized;For the data of order type feature, using orderly
Type numeric coding standardization;
Referring to the parameter of user preset, the parameter of user preset includes: the size of sliding window and the sliding type of window, is passed through
Behavioural information sliding window unit carries out sliding window processing to the behavioural information after standardization, and behavioural information is made to become inputting net
The block of information of network structure module;
Step S08: the behavioural information network architecture
By convolution layer unit, Capsule layer Unit one, Capsule layer Unit two, one four layers of full articulamentum building unit
Network architecture model is trained the behavioural information with label by iteration several times referring to the parameter of user setting, instructs
The dynamic routing protocol in model parameter and Capsule layer unit is continued to optimize by reducing loss function during white silk,
Finally obtain the high network architecture model of discrimination;
Step S09: the identification of behavioural information
The behavioural information acquired in real time is input to the real-time knowledge that current behavior is carried out in trained network architecture model
Not;
Step S10: error detection
Judge that current behavior identifies whether that mistake occurs, if there are mistakes for discovery, thens follow the steps S11, it is no to then follow the steps
S12;
Step S11: error correction
Error correction unit is adjusted the relevant parameter of behavioural information threshold range, behavioural information processing module;Behavior letter
Breath threshold range includes the threshold range of uncertain detection, and the relevant parameter of behavioural information processing module includes that behavioural information is sliding
The size of sliding window in window unit and the sliding type of window;
Step S12: user feedback detection
Judgement system whether there is field feedback, such as there is feedback information, thens follow the steps S13;
Step S13: user feedback
User feedback unit carries out feedback adjustment to the relevant parameter to behavioural information threshold range, behavioural information processing module.
8. the working method of the real-time Activity recognition system according to claim 7 based on Lora and Capsule, feature
Be, the step S08, the network architecture module include sequentially connected convolution layer unit, Capsule layer Unit one,
Capsule layer Unit two connects layer unit entirely;
It is N that convolution kernel number in convolution layer unit, which is arranged,1, each convolution kernel size is 1 × Nuclear_Size1, step-length L1;
It is N that convolution kernel number in Capsule layer Unit one, which is arranged,2, each convolution kernel size is 1 × Nuclear_Size2, step-length
For L2;
The output length that Capsule layer Unit two is arranged is that Num_Output ties up behavioural information, and each dimension uses Vec_Lenv
A behavioural information feature;
Output length is Output_Length in the full connection layer unit of setting;
It comprises the following steps that
(1) behavioural information of a size of Batch_Size × 1 × Window_Size × 3 is inputted, Batch_Size refers to once
The number of the behavioural information run in network architecture module, Window_Size refer to the length of input network architecture module every time
Degree;
(2) after the behavioural information of the size of Batch_Size × 1 × Window_Size × 3 passes through convolution layer unit, pass through formula
(I) behavioural information of input is converted into vector by scalar:
In formula (I), XiRefer to behavioural information after uncertain, standardization, the sliding window processing based on time series
Each information;WijRefer to the weight parameter of convolution layer unit, initial value is defaulted as generating the random number of cutting gearbox;
bjRefer to that the offset parameter of convolution layer unit, initial value are defaulted as 0.0;
The number of n expression convolution kernel;
YjIt is to indicate that convolutional layer exports;
Output information size are as follows:
(3) it enablesThe above M group convolutional layer is encapsulated in
In Capsule network, by the behavioural information Y of vectorjIt is input to Capsule layer Unit one, by formula (II) by the behavior of input
Information is converted into the behavioural information with spatial character;
In formula (II), WjlRefer to the weight parameter of Capsule layer Unit one, initial value be defaulted as generating cutting gearbox with
Machine number;
blRefer to that the offset parameter of Capsule layer Unit one, initial value are defaulted as 0.0;
Squsah () function is a kind of new nonlinear function;
Refer to the vector behavioural information feature of Capsule network output;
The information size exported after Capsule layer Unit one are as follows:
(4) behavioural information with spatial character is input to Capsule layer Unit two, passes through dynamic routing protocol, that is, formula
(III), (IV) handles behavioural information;
In formula (III), (IV),
bikRefer to the dynamic road of k-th of neuron in i-th of neuron and Capsule floor Unit two in Capsule floor Unit one
By weight;
bijRefer to the dynamic road of j-th of neuron in i-th of neuron and Capsule floor Unit two in Capsule floor Unit one
By weight;
Refer to each Capsule layers of output;
SjRefer to behavioural information feature of the Capsule layer Unit two by output after dynamic routing protocol;
It is the vector output for pointing out the network architecture;
The information size exported after two cell processing of Capsule layer are as follows: Batch_Size × Num_Output × Vec_Lenv ×
1;
(5) by full connection layer unit by behavioural information by vector median filters at scalar;
The information size exported after connecting layer unit entirely are as follows:
Batch_Size×Output_Length×1;
(6) Softmax classifier is added, the Classification and Identification of behavioural information is carried out by Softmax classifier;By the way that information is big
The small behavioural information feature for Batch_Size × Output_Length × 1 carries out asking for each behavior probability by classifier
Solution, finds out the corresponding maximum behavior of all kinds of probability numbers, and the final recognition result of network architecture module, that is, probability numbers are maximum
Behavior.
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811294526.XA CN109447162B (en) | 2018-11-01 | 2018-11-01 | Real-time behavior recognition system based on Lora and Capsule and working method thereof |
AU2019371325A AU2019371325B2 (en) | 2018-11-01 | 2019-03-23 | Real-time activity recognition system and working method based on lora and capsule |
PCT/CN2019/079387 WO2020087848A1 (en) | 2018-11-01 | 2019-03-23 | Real-time behavior identification system based on lora and capsule, and operating method therefor |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811294526.XA CN109447162B (en) | 2018-11-01 | 2018-11-01 | Real-time behavior recognition system based on Lora and Capsule and working method thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109447162A true CN109447162A (en) | 2019-03-08 |
CN109447162B CN109447162B (en) | 2021-09-24 |
Family
ID=65549883
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811294526.XA Active CN109447162B (en) | 2018-11-01 | 2018-11-01 | Real-time behavior recognition system based on Lora and Capsule and working method thereof |
Country Status (3)
Country | Link |
---|---|
CN (1) | CN109447162B (en) |
AU (1) | AU2019371325B2 (en) |
WO (1) | WO2020087848A1 (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110852382A (en) * | 2019-11-12 | 2020-02-28 | 山东大学 | Behavior recognition system based on space-time multi-feature extraction and working method thereof |
WO2020087848A1 (en) * | 2018-11-01 | 2020-05-07 | 山东大学 | Real-time behavior identification system based on lora and capsule, and operating method therefor |
CN111881793A (en) * | 2020-07-20 | 2020-11-03 | 东北大学 | Non-invasive load monitoring method and system based on capsule network |
WO2021227294A1 (en) * | 2020-05-09 | 2021-11-18 | 西北工业大学 | Identity recognition method based on expert feedback mechanism |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111695342B (en) * | 2020-06-12 | 2023-04-25 | 复旦大学 | Text content correction method based on context information |
CN111931882B (en) * | 2020-07-20 | 2023-07-21 | 五邑大学 | Automatic goods checkout method, system and storage medium |
CN112327189B (en) * | 2020-10-14 | 2023-06-09 | 北方工业大学 | Comprehensive judging method for health state of energy storage battery based on KNN algorithm |
CN116204844B (en) * | 2023-04-28 | 2023-07-04 | 西南石油大学 | Electrical equipment abnormal data cleaning method based on uncertainty |
CN116661330B (en) * | 2023-07-31 | 2023-12-01 | 深圳市云图数字科技有限公司 | Collaborative operation method of intelligent home system and intelligent home system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104615983A (en) * | 2015-01-28 | 2015-05-13 | 中国科学院自动化研究所 | Behavior identification method based on recurrent neural network and human skeleton movement sequences |
CN107092894A (en) * | 2017-04-28 | 2017-08-25 | 孙恩泽 | A kind of motor behavior recognition methods based on LSTM models |
CN108345846A (en) * | 2018-01-29 | 2018-07-31 | 华东师范大学 | A kind of Human bodys' response method and identifying system based on convolutional neural networks |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040098367A1 (en) * | 2002-08-06 | 2004-05-20 | Whitehead Institute For Biomedical Research | Across platform and multiple dataset molecular classification |
KR20100058833A (en) * | 2008-11-25 | 2010-06-04 | 삼성전자주식회사 | Interest mining based on user's behavior sensible by mobile device |
CN102348101A (en) * | 2010-07-30 | 2012-02-08 | 深圳市先进智能技术研究所 | Examination room intelligence monitoring system and method thereof |
US11074495B2 (en) * | 2013-02-28 | 2021-07-27 | Z Advanced Computing, Inc. (Zac) | System and method for extremely efficient image and pattern recognition and artificial intelligence platform |
CN104268577B (en) * | 2014-06-27 | 2017-05-03 | 大连理工大学 | Human body behavior identification method based on inertial sensor |
CN104200113A (en) * | 2014-09-10 | 2014-12-10 | 山东农业大学 | Internet of Things data uncertainty measurement, prediction and outlier-removing method based on Gaussian process |
CN104598880A (en) * | 2015-03-06 | 2015-05-06 | 中山大学 | Behavior identification method based on fuzzy support vector machine |
CN105139029B (en) * | 2015-08-14 | 2018-11-02 | 哈尔滨华夏矿安科技有限公司 | A kind of Activity recognition method and device of prison prisoner |
CN109447162B (en) * | 2018-11-01 | 2021-09-24 | 山东大学 | Real-time behavior recognition system based on Lora and Capsule and working method thereof |
-
2018
- 2018-11-01 CN CN201811294526.XA patent/CN109447162B/en active Active
-
2019
- 2019-03-23 AU AU2019371325A patent/AU2019371325B2/en not_active Ceased
- 2019-03-23 WO PCT/CN2019/079387 patent/WO2020087848A1/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104615983A (en) * | 2015-01-28 | 2015-05-13 | 中国科学院自动化研究所 | Behavior identification method based on recurrent neural network and human skeleton movement sequences |
CN107092894A (en) * | 2017-04-28 | 2017-08-25 | 孙恩泽 | A kind of motor behavior recognition methods based on LSTM models |
CN108345846A (en) * | 2018-01-29 | 2018-07-31 | 华东师范大学 | A kind of Human bodys' response method and identifying system based on convolutional neural networks |
Non-Patent Citations (1)
Title |
---|
SARA SABOUR ET.AL.: ""Dynamic Routing Between Capsules"", 《CS.CV》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020087848A1 (en) * | 2018-11-01 | 2020-05-07 | 山东大学 | Real-time behavior identification system based on lora and capsule, and operating method therefor |
CN110852382A (en) * | 2019-11-12 | 2020-02-28 | 山东大学 | Behavior recognition system based on space-time multi-feature extraction and working method thereof |
CN110852382B (en) * | 2019-11-12 | 2023-04-18 | 山东大学 | Behavior recognition system based on space-time multi-feature extraction and working method thereof |
WO2021227294A1 (en) * | 2020-05-09 | 2021-11-18 | 西北工业大学 | Identity recognition method based on expert feedback mechanism |
CN111881793A (en) * | 2020-07-20 | 2020-11-03 | 东北大学 | Non-invasive load monitoring method and system based on capsule network |
CN111881793B (en) * | 2020-07-20 | 2024-03-01 | 东北大学 | Non-invasive load monitoring method and system based on capsule network |
Also Published As
Publication number | Publication date |
---|---|
AU2019371325A1 (en) | 2020-10-15 |
CN109447162B (en) | 2021-09-24 |
WO2020087848A1 (en) | 2020-05-07 |
AU2019371325B2 (en) | 2022-01-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109447162A (en) | A kind of real-time Activity recognition system and its working method based on Lora and Capsule | |
Kato et al. | Ten challenges in advancing machine learning technologies toward 6G | |
Din et al. | Machine learning in the Internet of Things: Designed techniques for smart cities | |
CN108764059B (en) | Human behavior recognition method and system based on neural network | |
CN109698836A (en) | A kind of method for wireless lan intrusion detection and system based on deep learning | |
KR20190104487A (en) | Method and apparatus for recommending food and drink based on artificial intelligence-based user status | |
CN109362066A (en) | A kind of real-time Activity recognition system and its working method based on low-power consumption wide area network and capsule network | |
CN110198237A (en) | A kind of wireless configuration method for being intended to driving network | |
He et al. | Intrusion detection based on stacked autoencoder for connected healthcare systems | |
CN110139067A (en) | A kind of wild animal monitoring data management information system | |
CN109951363A (en) | Data processing method, apparatus and system | |
US20220156587A1 (en) | Multi-head deep metric machine-learning architecture | |
CN108551491A (en) | The mobile Internet of Things cloud system of heterogeneous network spatial multi time based on artificial intelligence | |
CN113191479A (en) | Method, system, node and storage medium for joint learning | |
CN105100735A (en) | People intelligent monitoring management system and management method | |
KR20190106946A (en) | Artificial device and method for controlling the same | |
CN110244854A (en) | A kind of artificial intelligence approach of multi-class eeg data identification | |
CN107679484A (en) | A kind of Remote Sensing Target automatic detection and recognition methods based on cloud computing storage | |
CN114612813A (en) | Identity recognition method, model training method, device, equipment and storage medium | |
CN104092503B (en) | A kind of artificial neural network frequency spectrum sensing method optimized based on wolf pack | |
Succetti et al. | A fast deep learning technique for Wi-Fi-based human activity recognition | |
CN113379045A (en) | Data enhancement method and device | |
CN117201733B (en) | Real-time unmanned aerial vehicle monitoring and sharing system | |
Zhao et al. | A moving object detection method using deep learning-based wireless sensor networks | |
CN108875030B (en) | Context uncertainty eliminating system based on hierarchical comprehensive quality index QoX and working method thereof |
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 |