CN109063887A - A kind of crime hotspot prediction method and system - Google Patents
A kind of crime hotspot prediction method and system Download PDFInfo
- Publication number
- CN109063887A CN109063887A CN201810619599.5A CN201810619599A CN109063887A CN 109063887 A CN109063887 A CN 109063887A CN 201810619599 A CN201810619599 A CN 201810619599A CN 109063887 A CN109063887 A CN 109063887A
- Authority
- CN
- China
- Prior art keywords
- crime
- convolutional
- hotspot
- hotspot graph
- data
- 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
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 38
- 239000000284 extract Substances 0.000 claims description 13
- 238000003475 lamination Methods 0.000 claims description 8
- 230000007613 environmental effect Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 abstract description 5
- 230000009286 beneficial effect Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 206010039203 Road traffic accident Diseases 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000001537 neural effect Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000004821 distillation Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
- G06Q50/265—Personal security, identity or safety
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Computer Security & Cryptography (AREA)
- Educational Administration (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of crime hotspot prediction method and system, are related to public safety field.This programme solves the technical issues of how finding different regions collaboration crime, is suitable for predicting regional crime probability.The present invention is the following steps are included: S1: obtaining source data, generates multiple kde hotspot graphs according to the source data；S2: creation convolutional neural networks model is trained the convolutional neural networks model according to the kde hotspot graph；S3: using the common trait of the kde hotspot graph of trained convolutional neural networks model extraction different regions, the cnn hotspot graph comprising collaboration crime information is obtained.The present invention is suitable for predicting the crime probability in each area.
Description
Technical field
The present invention relates to public safety field, in particular to a kind of crime hotspot prediction method and system.
Background technique
In order to safeguard public safety, the crime probability to area is needed to predict, so that the area big to probability increases
Police strength avoids the benefit damage of the masses to reduce or even avoid the generation of crime dramas to greatest extent.By to somewhere
Image carries out crime analysis of central issue, obtains crime hotspot graph, can intuitively check the crime distribution situation in somewhere, thus alert
Power can be allocated according to crime distribution situation.
Based on the crime hotspot graph prediction technique of Density Estimator (KDE, Kernel Density Estimation), it is
A kind of classical method of crime prediction based on space time information of comparison, it passes through history previous conviction given area and geography
Location information combines, using a probability function calculate this area's future occur crime a possibility that size.But it is existing
It is intended to the different location multi-person synergy crime in a certain area for criminal activity, is made the collaboration of different regions using above scheme
Cases extract difficulty, can be only done single regional Event Distillation.
Summary of the invention
The technical problem to be solved by the present invention is to how find different regions collaboration crime.
The technical scheme to solve the above technical problems is that a kind of crime hotspot prediction method, including following step
It is rapid:
S1: obtaining source data, and the kde hotspot graph of different regions is generated according to the source data；
S2: creation convolutional neural networks model instructs the convolutional neural networks model according to the kde hotspot graph
Practice；
S3: utilizing trained convolutional neural networks model, extracts the common trait of different regions kde hotspot graph, obtains
Cnn hotspot graph comprising collaboration crime information.
The source data for being used to generate kde hotspot graph is obtained by the input of user, then creates convolutional neural networks model,
Kde hotspot graph is handled as multi-channel digital image input convolutional neural networks model, from the angle pair of image procossing
Kde hotspot graph is trained, and obtains trained convolutional neural networks model, and extracting in the kde hotspot graph of different regions needs
The common trait wanted, collaboration crime are to be committed a crime jointly in different regions, therefore have in different area and commit a crime jointly
Trend occurs, and extracts feature by the cnn hotspot graph to different regions, it can be found that collaboration crime information, by collaboration crime letter
Breath and cnn hotspot graph, which combine, obtains the cnn hotspot graph comprising collaboration crime information.
The beneficial effects of the present invention are: the crime hotspot prediction method of this programme utilizes convolutional neural networks model extraction
Feature in kde hotspot graph, Multi-channel hot point diagram is analogized to image by this method, and is modeled using convolutional neural networks, thus
The feature in hotspot graph is automatically extracted, feature is extracted compared to artificial, can be avoided the interference caused by subjective factors of people, so that extract
Feature can more comprehensively；Due to kde hotspot graph include Crime Information, the feature found in kde hotspot graph with criminal
The related common feature of guilty information can be recorded as collaboration crime information, how find that different regions are assisted to solve
The technical issues of with crime.
Based on the above technical solution, the present invention can also be improved as follows.
Further, it in step S1, specifically includes:
S11: obtaining source data, and the source data includes crime hotspot graph, and the crime hotspot graph is miscellaneous category of offenses type
In the crime hotspot graph of the hotspot's distribution set of graphs of multiple predetermined times；
S12: determining criminal type quantity to be predicted, and the port number of crime hotspot graph is arranged according to the quantity of criminal type
Amount generates multichannel crime hotspot graph；
S13: the multichannel crime hotspot graph is normalized；
S14: kde processing is carried out to the number of channels of the multichannel crime hotspot graph at each moment, generates multiple kde
Hotspot graph.
Beneficial effect using above-mentioned further scheme is,；The crime hotspot graph of acquisition is miscellaneous category of offenses type multiple
When the hotspot's distribution set of graphs of predetermined time, the crime hotspot graph of single criminal type can once be carried out miscellaneous category of offenses type
Statistics；By crime hotspot graph as the multichannel crime hotspot graph of multi-channel digital image, using convolutional neural networks model into
Row processing, after setting the quantity of criminal type to the number of channels of multichannel crime hotspot graph, by convolutional neural networks mould
The multichannel crime hotspot graph of type processing will not lose the information of criminal type；Place is normalized in multichannel crime hotspot graph
Brought into after reason convolutional neural networks model compare without normalized multichannel crime hotspot graph in model training receipts
It is more preferable to hold back effect；It, can be by the previous conviction in multichannel crime hotspot graph after crime hotspot graph is carried out kde processing by multichannel
It is connected with geographical location.
Further, in step S12, specifically: criminal type quantity to be predicted is determined, according to the hair of each criminal type
Raw number arrangement, takes n criminal type, remaining the criminal type conduct being not included in the n criminal type from big to small
An independent criminal type, obtains n+1 criminal type, and the number of channels of crime hotspot graph is arranged according to the quantity of criminal type
Obtain multichannel crime hotspot graph.
Beneficial effect using above-mentioned further scheme is that step S12 carries out the crime hotspot graph for needing to participate in training
Statistics allows users to be quickly found out the criminal type of n before ranking, and can be convenient user will be other than preceding n criminal type
Other criminal types compare as an individual criminal type.
Further, it in step S2, specifically includes:
S21: each kde hotspot graph is sorted according to preset time period, spatio-temporal data is obtained, by the sky
M- time data are divided into three groups of a importations according to preset time；
S22: a importation obtains three groups of different a importations according to different preset times；
S23: four convolutional networks of creation, respectively convolutional network a, convolutional network b, convolutional network c and convolutional network d,
The convolutional network a, the convolutional network b and the convolutional network c include the first convolutional layer and the second convolutional layer, defeated by a
Enter part to arrange sequentially in time, while first group of a importation is input to the convolutional network a, by second group
A importation be input to the convolutional network b, a importation of third group is input to the convolutional network c, and
Among respectively obtain the first intermediate data of convolutional network a, the first intermediate data of convolutional network b and convolutional network c first
Data；
The first intermediate data of the convolutional network a: being input to the second convolutional layer of the convolutional network a by S24, by institute
The first intermediate data for stating convolutional network b is input to the second convolutional layer of the convolutional network b, by the of the convolutional network c
One intermediate data is input to the second convolutional layer of the convolutional network c, and respectively obtain convolutional network a the second intermediate data,
The second intermediate data of convolutional network b and the second intermediate data of convolutional network c；
S25: by the second intermediate data of the convolutional network a, the second intermediate data and the volume of the convolutional network b
The second intermediate data of product network c is concatenated, and concatenated data is obtained；
S26: the convolutional network d includes third convolutional layer and Volume Four lamination；
S27: the concatenated data is input to third convolutional layer and obtains third intermediate data；
S28: the third intermediate data is input to Volume Four lamination and obtains the 4th intermediate data；
S29: the source data further includes history metadata, and the history metadata includes environmental state information, and creation is complete
The history metadata is input to the full articulamentum and the dimension of the history metadata is adjusted to and is exported by articulamentum
Data dimension is consistent, then merges to obtain output data with the 4th intermediate data；
S210: the output data is brought into preset tanh activation primitive and calculates and obtain preliminary cnn hotspot graph；
S211: the mean square error for calculating the preliminary cnn hotspot graph obtains cnn when the number of iterations reaches preset times
Hotspot graph.
Four convolutional networks of creation collectively constitute convolutional neural networks model, using having for above-mentioned further scheme
Beneficial effect is, kde hotspot graph is arranged rear sequentially in time continues after an action of the bowels and calculate, by spatio-temporal data according to it is default when
Between be divided into three groups of a importations after, below by spatio-temporal data input convolutional neural networks model be trained when,
A different parts can be selected to be trained, so that the training degree of each convolutional network is similar, but the sky that training is used
M- time data are different, obtain different convolutional networks, so that calculating error is reduced, the string that comprehensive three convolutional networks obtain
It connects data and is input to convolutional neural networks model d and obtain the 4th intermediate data, be eventually adding the history including environmental state information
Metadata is trained, and obtains the output data merged with the 4th intermediate data, compared to it is no be added history metadata scheme,
The information that the output data of this programme includes is more, can calculate more influence factors for influencing crime and occurring.
Further, the first convolutional layer convolution kernel size 3 × 3, convolution kernel number are 32, the second convolutional layer convolution
Core size 3 × 3, convolution kernel number are 16, the third convolutional layer convolution kernel size 3 × 3, and convolution kernel number is 16, described the
Four convolutional layer convolution kernel sizes 3 × 3, convolution kernel number are 4.
Beneficial effect using above-mentioned further scheme is that the scheme less than the convolution kernel size in this programme cannot calculate
The figure of calculative this complexity of kde hotspot graph, the scheme greater than the convolution kernel size in this programme are compared in this programme
The calculating of this programme is more complicated, reduces computational efficiency.
Further, a kind of crime hotspot prediction system, comprising:
Input unit generates multiple kde hotspot graphs for obtaining source data, and according to the source data；
Training unit, for creating convolutional neural networks model, and according to the kde hotspot graph to the convolutional Neural net
Network model training；
Reasoning element, for by weight share in the way of extract different regions cnn hotspot graph common trait, obtain
Cnn hotspot graph comprising collaboration crime information.
Beneficial effect using above-mentioned further scheme is, after obtaining source data by input unit, obtains multiple kde heat
Then point diagram brings kde hotspot graph into convolutional neural networks model and is trained to obtain cnn hotspot graph, recycles weight shared
Mode extract different regions cnn hotspot graph common trait, what common trait herein indicated to occur in different regions
Crime, therefore can be used as collaboration crime information and stored, after common trait and cnn hotspot graph are combined, obtain comprising association
With the cnn hotspot graph of crime information.
Further, a kind of storage medium is stored with instruction in the storage medium, when computer reads described instruction,
The computer is set to execute any method in above scheme.
The advantages of additional aspect, will be set forth in part in the description, and partially will become from the following description bright
It is aobvious, or practice is recognized through the invention.
Detailed description of the invention
Fig. 1 is the flow diagram of the embodiment of crime hotspot prediction method of the present invention；
Fig. 2 is the flow diagram of the other embodiments of crime hotspot prediction method of the present invention；
Fig. 3 is the flow diagram of the other embodiments of crime hotspot prediction method of the present invention；
Fig. 4 is the system structure diagram of the embodiment of crime hotspot prediction system of the present invention；
Fig. 5 is the structural representation of the convolutional neural networks model of the other embodiments of crime hotspot prediction method of the present invention
Figure.
Specific embodiment
The principle and features of the present invention will be described below with reference to the accompanying drawings, and the given examples are served only to explain the present invention, and
It is non-to be used to limit the scope of the invention.
Embodiment is substantially as shown in Fig. 1:
Crime hotspot prediction method in the present embodiment, comprising the following steps:
S1: obtaining source data, generates multiple kde hotspot graphs according to source data, 5 years crime numbers have been used in the present embodiment
According to, one day 2 crime data, 5 years about 3650 crime datas in total, number of days increases if having the leap year, the kde in the present embodiment
Hotspot graph is the crime hotspot prediction figure obtained based on Density Estimator, and source data is the history crime note that record has specific region
The data of record and geographical location estimate that history previous conviction and geographical location information by specific region combine by kde,
Kde estimation can be carried out by the way of Gaussian kernel, and bandwidth can use the half of Scott criterion, it may be assumed that
Wherein n is the number of data point in a hotspot graph channel, i.e. hotspot graph pixel wide × height, and d is data
Dimension, d=2 in the present embodiment, dimension are the 2 two-dimensional map grids for indicating longitude and latitude composition in geography information；
S2: creation convolutional neural networks model is trained to obtain according to kde hotspot graph to convolutional neural networks model
Cnn hotspot graph, the cnn hotspot graph in the present embodiment are to pass through convolutional Neural net for kde hotspot graph as multi-channel digital image
The hotspot graph obtained after network model training can extract the spy of the multi-channel digital image of input by convolutional neural networks model
Sign, extracting feature compared to manpower can be avoided the subjective impact extraction of people as a result, to obtain the sheet that can be preferably directed to problem
Body；
S3: by weight share in the way of extract different regions cnn hotspot graph common trait, a convolutional Neural net
One feature of network model extraction image needs to use a filter, be arranged in a convolutional neural networks model it is multiple not
Same filter can extract multiple and different features, utilize a filter to the cnn hot spot of different regions in the present embodiment
Figure carries out feature extraction, so that feature common in different cnn hotspot graphs is obtained, due to being identical feature in different regions,
Therefore the identical crime processing that can be used as different regions can be used as collaboration crime information processing, to obtain comprising collaboration
The cnn hotspot graph for information of committing a crime.
The source data for being used to generate kde hotspot graph is obtained by the input of user, then creates convolutional neural networks model,
Kde hotspot graph is handled as multi-channel digital image input convolutional neural networks model, from the angle pair of image procossing
Kde hotspot graph is trained, and generates the cnn hotspot graph of different regions, by weight share in the way of, cnn in different regions
Hotspot graph extracts the common trait needed, and collaboration crime is committed a crime jointly in different regions, therefore in different area meetings
There is common trait appearance, common trait is extracted by the cnn hotspot graph to different regions, it can be found that collaboration crime information, it will
Collaboration crime information and cnn hotspot graph, which combine, obtains the cnn hotspot graph comprising collaboration crime information.
Optionally, in some other embodiments, as shown in Fig. 2, being specifically included in step S1:
S11: obtaining source data, and source data includes crime hotspot graph, which is miscellaneous category of offenses type multiple
The crime hotspot graph of the hotspot's distribution set of graphs of predetermined time, the predetermined time in the present embodiment are using 12h as interval average mark
3650 predetermined times of cloth, criminal type can be theft, robbery or swindle；
S12: determining criminal type quantity to be predicted, and the port number of crime hotspot graph is arranged according to the quantity of criminal type
Amount generates multichannel crime hotspot graph；
S13: multichannel crime hotspot graph is normalized；
S14: kde processing is carried out to the number of channels of the multichannel crime hotspot graph at each moment, the kde in the present embodiment
Processing is the crime hotspot graph obtained based on Density Estimator, generates multiple kde hotspot graphs, utilizes 3650 criminals in the present embodiment
Guilty data generate 3650 kde hotspot graphs.
The crime hotspot graph of acquisition is miscellaneous category of offenses type in the hotspot's distribution set of graphs of multiple predetermined times, single criminal
The crime hotspot graph of guilty type can once count miscellaneous category of offenses type；Crime hotspot graph is converted into multi-channel digital
The multichannel crime hotspot graph of image, could be handled using convolutional neural networks model, and the quantity of criminal type is arranged
It, will not by the multichannel crime hotspot graph of convolutional neural networks model treatment after number of channels for multichannel crime hotspot graph
Lose the information of criminal type；Bring the progress of convolutional neural networks model after multichannel crime hotspot graph is normalized into
It calculates quicker compared to the multichannel crime hotspot graph without normalized；Multichannel crime hotspot graph is subjected to kde
After processing, can by multichannel crime hotspot graph previous conviction and geographical location connect.
Optionally, in some other embodiments, in step S12, specifically: determine criminal type quantity to be predicted,
It is arranged according to the frequency of each criminal type, takes n criminal type, n=3 in the present embodiment, 3 crime classes from big to small
Type is respectively to steal, plunder and swindle, and is not included in remaining criminal type in n criminal type as an independent crime class
Type is pressed for example, criminal offence included in multichannel crime hotspot graph has theft, robbery, swindle, traffic accident and has a fist fight
Frequency arrangement, has a fist fight 10 times altogether, steals 5 times, swindles 4 times, traffic accident 2 times, plunders 1 time, then can will struggle against
It beats up, steal and swindles as 3 kinds of independent criminal types, by remaining traffic accident and plunder as a kind of independent crime class
Type, i.e. other types.N+1 criminal type, i.e. 4 criminal types are obtained, crime hot spot is arranged according to the quantity of criminal type
The number of channels of figure obtains multichannel crime hotspot graph.
Step S12 allows users to be quickly found out ranking to needing the crime hotspot graph for participating in training to be counted
The criminal type of preceding n, then using other criminal types other than preceding n criminal type as an individual criminal type for
Compare at family.
Optionally, in some other embodiments, as shown in Figure 3 and Figure 5, in step S2, specifically include:
S21: each kde hotspot graph is sorted according to preset time period, spatio-temporal data is obtained, by space-time number
It is divided into a importation according to according to preset time, it is respectively [X that a in the present embodiment, which can be 7,7 importations,t-7s,
Xt-6s,…,Xt-s], wherein the value of t and s is continuous integer, such as 1,2,3 etc., the reality between each group of hotspot graph
Interval is 12 hours.Current time is t, previous moment t-1, but the t-1 moment is actually the hotspot graph before 12 hours,
In this way, t-2 is actually the hotspot graph before 24 hours, and so on, s=1 at this time；If taking s=2, according to description, t-s
Moment is the hotspot graph before 24 hours, and t-2s is the hotspot graph before 48 hours；If s=8, t-s are before one week
Hotspot graph, it is s=8, s=2 and s=1 that the s in the present embodiment, which can distinguish value, obtains [Xt-56,Xt-48,…,Xt-8]、
[Xt-14,Xt-12,…,Xt-2] and [Xt-7,Xt-6,…,Xt-1], different according to the value of s, the convolutional network a1 in the present embodiment is
Tendency information 2, convolutional network b8 are cycle information 3, and convolutional network c9 is recent information 4；
S22: a importation obtains three groups of different a importations according to three groups of different preset times；
S23: four convolutional networks of creation, respectively convolutional network a1, convolutional network b8, convolutional network c9 and convolutional network
D, convolutional network a1, convolutional network b8 and convolutional network c9 include the first convolutional layer and the second convolutional layer, and a importation is pressed
It is arranged according to time sequencing, while first group of a1 importation is input to convolutional network a1, by second group of a1 input
Part is input to convolutional network b8, a1 importation of third group is input to convolutional network c9, and respectively obtain convolution net
First intermediate data of the first intermediate data of network a1, the first intermediate data of convolutional network b8 and convolutional network c9；
S24: the first intermediate data of convolutional network a1 is input to the second convolutional layer of convolutional network a1, by convolutional network
The first intermediate data of b8 is input to the second convolutional layer of convolutional network b8, and the first intermediate data of convolutional network c9 is input to
The second convolutional layer of convolutional network c9, and respectively obtain the second intermediate data of convolutional network a1, convolutional network b8 second in
Between data and convolutional network c9 the second intermediate data；
S25: by the second intermediate data of convolutional network a1, the second intermediate data of convolutional network b8 and convolutional network c9
Second intermediate data is concatenated, and concatenated data 5 is obtained；
S26: convolutional neural networks model d7 includes third convolutional layer and Volume Four lamination；
S27: concatenated data 5 is input to third convolutional layer and obtains third intermediate data；
S28: third intermediate data is input to Volume Four lamination and obtains the 4th intermediate data；
S29: source data further includes history metadata, and history metadata includes environmental state information, the ring in the present embodiment
Border status information include weather condition, whether festivals or holidays, whether night crime and guilty place description, encoded using one-hot
Source data is pre-processed, full articulamentum 6 is created, history metadata is input to full articulamentum 6 and utilizes Reshape function
The dimension of history metadata is adjusted to consistent with output data dimension, then merges to obtain output data with the 4th intermediate data；
S210: output data being brought into preset tanh activation primitive and calculates and obtain preliminary cnn hotspot graph, in the present embodiment
Tanh activation primitive are as follows:
Wherein, x represents output data, and the value of y constitutes preliminary cnn hotspot graph；
S211: the mean square error of preliminary cnn hotspot graph is calculated, the formula of mean square error is calculated in the present embodiment are as follows:
Wherein, MSE indicates mean square error,Indicate output data, X indicates true crime hotspot graph.
When the number of iterations reaches preset times, training terminates, and the number of iterations is set as 60 wheels in the present embodiment.Every wheel iteration
Batch processing amount be 32.Obtain cnn hotspot graph.
Kde hotspot graph is arranged into rear sequentially in time and continues calculating after an action of the bowels, spatio-temporal data is divided into a input
Behind part, when spatio-temporal data input convolutional neural networks model is trained below, can select different a
Part is trained, so that the training degree of each convolutional network is similar, but the spatio-temporal data difference that training is used,
Different convolutional networks is obtained, to reduce calculating error, the concatenated data 5 that comprehensive three convolutional networks obtain is input to convolution
Neural network model d7 obtains the 4th intermediate data, is eventually adding the history metadata including environmental state information and is trained,
The output data merged with the 4th intermediate data is obtained, compared to no scheme that history metadata is added, the output number of this programme
According to comprising information it is more, more influence factors for influencing crime and occurring can be calculated,
Optionally, in some other embodiments, the first convolutional layer convolution kernel size 3 × 3, convolution kernel number are 32, the
Two convolutional layer convolution kernel sizes 3 × 3, convolution kernel number are 16, third convolutional layer convolution kernel size 3 × 3, and convolution kernel number is
16, Volume Four lamination convolution kernel size 3 × 3, convolution kernel number is 4.
It is this that calculative kde hotspot graph in this programme cannot be calculated less than the scheme of the convolution kernel size in this programme
Complicated figure, it is more complicated compared to the calculating of this programme greater than the scheme of the convolution kernel size in this programme, reduce calculating
Efficiency.
It optionally, in some other embodiments, in conjunction with Fig. 4, can also include a kind of crime hotspot prediction system, it should
System includes:
Input unit 7 generates multiple kde hotspot graphs for obtaining source data, and according to source data；
Training unit 8, for creating convolutional neural networks model, and according to kde hotspot graph to convolutional neural networks model
Training obtains cnn hotspot graph；
Reasoning element 9, for by weight share in the way of extract different regions cnn hotspot graph common trait, obtain
To the cnn hotspot graph comprising collaboration crime information.
After obtaining source data by input unit 7, multiple kde hotspot graphs are obtained, then bring kde hotspot graph into convolution mind
Be trained to obtain cnn hotspot graph through network model, by weight share in the way of extract the cnn hotspot graphs of different regions and be total to
Same feature, common trait herein indicate the crime occurred in different regions, therefore can be used as collaboration crime information and carry out
Storage after combining common trait and cnn hotspot graph, obtains the cnn hotspot graph comprising collaboration crime information.
Optionally, in some other embodiments, it can also include a kind of storage medium, be stored with finger in storage medium
It enables, when computer, which is read, to be instructed, computer is made to execute method any in above scheme.
The advantages of additional aspect, will be set forth in part in the description, and partially will become from the following description bright
It is aobvious, or practice is recognized through the invention.
The foregoing is merely a prefered embodiment of the invention, is not intended to limit the invention, all in the spirit and principles in the present invention
Within, any modification, equivalent replacement, improvement and so on should all be included in the protection scope of the present invention.
Claims (7)
1. a kind of crime hotspot prediction method, it is characterised in that the following steps are included:
S1: obtaining source data, and the kde hotspot graph of different regions is generated according to the source data；
S2: creation convolutional neural networks model is trained the convolutional neural networks model according to the kde hotspot graph；
S3: trained convolutional neural networks model is utilized, the common trait of the kde hotspot graph of different regions is extracted, is wrapped
The cnn hotspot graph of the information of crime containing collaboration.
2. crime hotspot prediction method according to claim 1, it is characterised in that: in step S1, specifically include:
S11: obtaining source data, and the source data includes crime hotspot graph, and the crime hotspot graph is miscellaneous category of offenses type more
The hotspot's distribution set of graphs of a predetermined time；
S12: determining criminal type quantity to be predicted, and the number of channels of crime hotspot graph is arranged according to the quantity of criminal type,
Generate multichannel crime hotspot graph；
S13: the multichannel crime hotspot graph is normalized；
S14: kde processing is carried out to each channel of the multichannel crime hotspot graph at each moment, generates multiple kde hot spots
Figure.
3. crime hotspot prediction method according to claim 2, it is characterised in that: in step S12, specifically: determine to
The criminal type quantity of prediction arranges according to the frequency of each criminal type, takes n criminal type from big to small, do not wrap
Remaining criminal type in the n criminal type is contained in as an independent criminal type, obtains n+1 criminal type, root
Multichannel crime hotspot graph is obtained according to the number of channels of the quantity setting crime hotspot graph of criminal type.
4. crime hotspot prediction method according to claim 2, it is characterised in that: in step S2, specifically include:
S21: each kde hotspot graph being sorted according to preset time period, obtains spatio-temporal data, by it is described space-when
Between data a importation be divided into according to preset time；
S22: a importation obtains three groups of different a importations according to three groups of different preset times；
S23: four convolutional networks of creation, respectively convolutional network a, convolutional network b, convolutional network c and convolutional network d are described
Convolutional network a, the convolutional network b and the convolutional network c respectively include the first convolutional layer and the second convolutional layer, by a
Importation arranges sequentially in time, while first group of a importation is input to the convolutional network a, by second
A importation of group is input to the convolutional network b, and a importation of third group is input to the convolutional network c,
And respectively obtain the first intermediate data of convolutional network a, the first intermediate data of convolutional network b and convolutional network c first in
Between data；
The first intermediate data of the convolutional network a: being input to the second convolutional layer of the convolutional network a by S24, by the volume
The first intermediate data of product network b is input to the second convolutional layer of the convolutional network b, will be in the first of the convolutional network c
Between data be input to the second convolutional layer of the convolutional network c, and respectively obtain the second intermediate data of convolutional network a, convolution
The second intermediate data of network b and the second intermediate data of convolutional network c；
S25: by the second intermediate data of the convolutional network a, the second intermediate data and the convolution net of the convolutional network b
The second intermediate data of network c is concatenated, and concatenated data is obtained；
S26: the convolutional network d includes third convolutional layer and Volume Four lamination；
S27: the concatenated data is input to third convolutional layer and obtains third intermediate data；
S28: the third intermediate data is input to Volume Four lamination and obtains the 4th intermediate data；
S29: the source data further includes history metadata, and the history metadata includes environmental state information, creates full connection
The history metadata is input to the full articulamentum and is adjusted to the dimension of the history metadata and output data by layer
Dimension is consistent, then merges to obtain output data with the 4th intermediate data；
S210: the output data is brought into preset tanh activation primitive and calculates and obtain preliminary cnn hotspot graph；
S211: the mean square error for calculating the preliminary cnn hotspot graph obtains cnn hot spot when the number of iterations reaches preset times
Figure.
5. crime hotspot prediction method according to claim 4, it is characterised in that: the first convolutional layer convolution kernel size
3 × 3, convolution kernel number is 32, the second convolutional layer convolution kernel size 3 × 3, and convolution kernel number is 16, the third convolution
Layer convolution kernel size 3 × 3, convolution kernel number are 16, and the Volume Four lamination convolution kernel size 3 × 3, convolution kernel number is 4.
6. a kind of crime hotspot prediction system characterized by comprising
Input unit generates multiple kde hotspot graphs for obtaining source data, and according to the source data；
Training unit, for creating convolutional neural networks model, and according to the kde hotspot graph to the convolutional neural networks mould
Type training；
Reasoning element, for by weight share in the way of extract different regions cnn hotspot graph common trait, included
The cnn hotspot graph of collaboration crime information.
7. a kind of storage medium, it is characterised in that: instruction is stored in the storage medium, when computer reads described instruction
When, so that the computer is executed the method as described in any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810619599.5A CN109063887B (en) | 2018-06-13 | 2018-06-13 | Crime hotspot prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810619599.5A CN109063887B (en) | 2018-06-13 | 2018-06-13 | Crime hotspot prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063887A true CN109063887A (en) | 2018-12-21 |
| CN109063887B CN109063887B (en) | 2021-09-17 |
Family
ID=64820990
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810619599.5A Active CN109063887B (en) | 2018-06-13 | 2018-06-13 | Crime hotspot prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063887B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110309935A (en) * | 2019-03-26 | 2019-10-08 | 浙江工业大学 | A kind of method of crime prediction based on improvement STARMA model |
| CN114418244A (en) * | 2022-03-29 | 2022-04-29 | 北京零点远景网络科技有限公司 | Case prediction analysis method and device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150379413A1 (en) * | 2014-06-30 | 2015-12-31 | Palantir Technologies, Inc. | Crime risk forecasting |
| CN105678428A (en) * | 2016-01-28 | 2016-06-15 | 温州职业技术学院 | Criminal suspicion probability prediction method and system |
| CN106952208A (en) * | 2017-03-17 | 2017-07-14 | 讯飞智元信息科技有限公司 | Crime automatic prediction method and system |
| CN106991510A (en) * | 2017-05-31 | 2017-07-28 | 福建江夏学院 | A kind of method based on the traffic accident of spatial-temporal distribution characteristic predicted city |
| CN107563122A (en) * | 2017-09-20 | 2018-01-09 | 长沙学院 | The method of crime prediction of Recognition with Recurrent Neural Network is locally connected based on interleaving time sequence |
| US20180053110A1 (en) * | 2016-08-22 | 2018-02-22 | The Catholic University Of Korea Industry-Academic Cooperation Foundation | Method of predicting crime occurrence in prediction target region using big data |
-
2018
- 2018-06-13 CN CN201810619599.5A patent/CN109063887B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150379413A1 (en) * | 2014-06-30 | 2015-12-31 | Palantir Technologies, Inc. | Crime risk forecasting |
| CN105678428A (en) * | 2016-01-28 | 2016-06-15 | 温州职业技术学院 | Criminal suspicion probability prediction method and system |
| US20180053110A1 (en) * | 2016-08-22 | 2018-02-22 | The Catholic University Of Korea Industry-Academic Cooperation Foundation | Method of predicting crime occurrence in prediction target region using big data |
| CN106952208A (en) * | 2017-03-17 | 2017-07-14 | 讯飞智元信息科技有限公司 | Crime automatic prediction method and system |
| CN106991510A (en) * | 2017-05-31 | 2017-07-28 | 福建江夏学院 | A kind of method based on the traffic accident of spatial-temporal distribution characteristic predicted city |
| CN107563122A (en) * | 2017-09-20 | 2018-01-09 | 长沙学院 | The method of crime prediction of Recognition with Recurrent Neural Network is locally connected based on interleaving time sequence |
Non-Patent Citations (2)
| Title |
|---|
| ERIK JOHANSSON，CHRISTOFFER GAHLIN，ET AL.: "Crime Hotspots: An Evaluation of the KDE Spatial", 《2015 EUROPEAN INTELLIGENCE AND SECURITY INFORMATICS CONFERENCE》 * |
| 吕雪梅: "美国犯罪情报预测分析技术的特点", 《情报杂志》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110309935A (en) * | 2019-03-26 | 2019-10-08 | 浙江工业大学 | A kind of method of crime prediction based on improvement STARMA model |
| CN114418244A (en) * | 2022-03-29 | 2022-04-29 | 北京零点远景网络科技有限公司 | Case prediction analysis method and device, electronic equipment and storage medium |
| CN114418244B (en) * | 2022-03-29 | 2022-07-08 | 北京零点远景网络科技有限公司 | Case prediction analysis method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063887B (en) | 2021-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | SeqST-GAN: Seq2Seq generative adversarial nets for multi-step urban crowd flow prediction | |
| CN107609708B (en) | User loss prediction method and system based on mobile game shop | |
| Johnstone et al. | Homicide reporting in Chicago dailies | |
| Schonteich et al. | Crime in South Africa: A country and cities profile | |
| CN109189991A (en) | Repeat video frequency identifying method, device, terminal and computer readable storage medium | |
| CN110084103A (en) | A kind of same pedestrian's analysis method and system based on face recognition technology | |
| CN110084603A (en) | Method, detection method and the corresponding intrument of training fraudulent trading detection model | |
| CN109919624B (en) | Network loan fraud group recognition and early warning method based on space-time aggregation | |
| CN109063887A (en) | A kind of crime hotspot prediction method and system | |
| Montolio et al. | How time shapes crime: The temporal impacts of football matches on crime | |
| CN109345375A (en) | A kind of suspicious money laundering Activity recognition method and device | |
| CN107038858A (en) | Method is recommended in the dynamic share-car of the private car that commutes | |
| CN109615573B (en) | Foreign personnel flow channeling analysis method and system based on big data | |
| CN108536572A (en) | Smart mobile phone App based on AppUsage2Vec models uses prediction technique | |
| CN108521588A (en) | A kind of main broadcaster's arrangement method and system based on time slicing, server and storage medium | |
| CN110019963B (en) | Method for searching suspect relation personnel | |
| CN111603158A (en) | Fatigue driving warning method and system based on electrophysiological signal artificial intelligence analysis | |
| Small | Female crime in the United States, 1963–1998: An update | |
| Miles | Estimating the effect of America's Most Wanted: A duration analysis of wanted fugitives | |
| CN106777215A (en) | A kind of image searching result display methods and device | |
| EZEGWU et al. | Audience perception of broadcast media coverage of banditry and kidnapping in Nigeria: A study of Benin residents | |
| CN110399399B (en) | User analysis method, device, electronic equipment and storage medium | |
| CN113298017A (en) | Behavior proposal generation method for video behavior detection | |
| CN111768031A (en) | Method for predicting crowd gathering tendency based on ARMA algorithm | |
| Fabel et al. | How Much Violence Does Football Hooliganism Cause? |
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 |