CN109946746A - A kind of safe examination system and method based on deep neural network - Google Patents
A kind of safe examination system and method based on deep neural network Download PDFInfo
- Publication number
- CN109946746A CN109946746A CN201910218654.4A CN201910218654A CN109946746A CN 109946746 A CN109946746 A CN 109946746A CN 201910218654 A CN201910218654 A CN 201910218654A CN 109946746 A CN109946746 A CN 109946746A
- Authority
- CN
- China
- Prior art keywords
- module
- picture
- model
- channel
- image
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 22
- 238000001514 detection method Methods 0.000 claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 40
- 238000013527 convolutional neural network Methods 0.000 claims description 26
- 239000000463 material Substances 0.000 claims description 19
- 238000007689 inspection Methods 0.000 claims description 16
- 230000032258 transport Effects 0.000 claims description 12
- 238000006243 chemical reaction Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 10
- 239000000203 mixture Substances 0.000 claims description 9
- 239000005416 organic matter Substances 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 230000011218 segmentation Effects 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000003384 imaging method Methods 0.000 claims description 5
- 238000012546 transfer Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims description 4
- 102000008297 Nuclear Matrix-Associated Proteins Human genes 0.000 claims description 3
- 108010035916 Nuclear Matrix-Associated Proteins Proteins 0.000 claims description 3
- 210000000299 nuclear matrix Anatomy 0.000 claims description 3
- 238000004846 x-ray emission Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims 1
- 238000007726 management method Methods 0.000 description 21
- 239000003086 colorant Substances 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000012216 screening Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000002360 explosive Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 239000000443 aerosol Substances 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 235000012206 bottled water Nutrition 0.000 description 1
- 230000009172 bursting Effects 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 231100001261 hazardous Toxicity 0.000 description 1
- 239000013056 hazardous product Substances 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 210000000554 iris Anatomy 0.000 description 1
- 239000003350 kerosene Substances 0.000 description 1
- 239000003915 liquefied petroleum gas Substances 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 239000004033 plastic Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V5/00—Prospecting or detecting by the use of ionising radiation, e.g. of natural or induced radioactivity
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- High Energy & Nuclear Physics (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geophysics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
The invention discloses a kind of safe examination system and method based on deep neural network, including X-ray image-forming module, detection model training study module, object identification module and safety management module, the output end of the X-ray image-forming module is connected with the input terminal of object identification module, object identification module and detection model training study module are bi-directionally connected, and the output end of object identification module is connected with the input terminal of safety management module;Color characteristic based on radioscopic image is divided and synthesizes multilevel detection image, establish deep neural network detection model, and feature training and study are carried out to common items with big data, realize identification and classification of the detector to rotation, flexible and deformation object.
Description
Technical field
The invention belongs to safety check technical fields, and in particular to a kind of safe examination system and method based on deep neural network.
Background technique
With rapid economic development, high-speed rail, aircraft etc. has become the essential vehicles of people's daily trip,
However passenger intentionally or unintentionally carries dangerous material and rides public transportation means but into the biggest threat of traffic safety.X is penetrated
Line screening machine has important role for dangerous material safety detection and in terms of ensureing transport facility operational safety.But it passes
The X-ray security inspection machine of system needs staff to check X-ray luggage image carefully conscientiously to discriminate whether containing dangerous goods,
Device intelligence degree is low, manually checks required higher cost, while may also will appear erroneous judgement and fail to judge situation, thus to people
Safety trip cause greatly to threaten, even lead to major accident.
A kind of " contraband safety check automatic identification equipment " (CN 201710233696.6) of patent application publication by image by
RGB color is transformed into three parts of hsv color space and duplication, is divided into three kinds of colors and is identified, after optimizing picture quality,
By contraband template parallel processing under the pattern after three kinds of different colours identifications and the corresponding color prestored, SURF feature is carried out
Match, matching rate 55% or more then thinks luggage, and there are contrabands.
SURF characteristic matching is mainly that the SURF in radioscopic image and the contraband image template prestored is described subnumber amount
It is matched, can only identify the object of same pattern and color, the Detection accuracy of article similar for shape is low (such as to be played
Have as pistol with the shape of true rifle is), generalization ability is poor, and category classification is indefinite.For being rotated in luggage and articles, stretch
Contracting, the object of deformation have certain detectability, but luggage and articles for mess therein or the object to overlap
Body is difficult to carry out accurate detection and differentiation.
" a kind of X-ray screening machine luggage dangerous material fast automatic detecting warning device and method " (CN of patent application publication
201610748757.8), firstly, using gaussian filtering method to image denoising, secondly, carrying out image using non-linear Enhancement Method
Enhancing and dangerous material image segmentation, finally, feature extraction and signature analysis are carried out to doubtful dangerous material image, if discovery dangerous material
It then irises out dangerous material and image data by network interface is transferred to computer and it will be shown on liquid crystal display, selection setting
Make a sound the flashing alarm with LED light.
Existing hazardous material detection technology uses image processing techniques, is mainly split then according to object color to figure
As object features extraction and analysis, the same object of unlike material cannot be handled very well during segmentation, such as: scissors
Tip be blue characteristics, and handle is usually orange color characteristic, divides object in this way and has had to the local feature of object and leads
Cause object accuracy rate is low and object category is indefinite, flexible for rotating in luggage and articles, the object detection accuracy rate of deformation
It is low, and because luggage and articles are stacked in a jumble, the object to overlap is difficult to carry out accurate detection.
With the rapid development of artificial intelligence technology, the intelligent screening machine system for incorporating deep learning algorithm will be significantly
The intelligent program of safety check apparatus is improved, the accuracy rate of dangerous goods identification is improved, while can also effectively mitigate safety check work people
The pressure of member, greatly promote security check passage by efficiency, reduce congestion, ensure that the traffic of people and trip are pacified to the full extent
Entirely.
Summary of the invention
Low for object detection locating accuracy present in existing X-ray luggage detection technique, hazardous classification is unknown
Really, the problems such as detection process intelligence degree is lower proposes that a kind of X based on color segmentation and more plane depth neural networks is penetrated
Line intelligence safety check apparatus and method solve the problems, such as the detection Yu identification of belongings in daily luggage parcel.Based on radioscopic image
Color characteristic divide and synthesize multilevel detection image, establish deep neural network detection model, and use big data pair
Common items carry out feature training and study, realize identification and classification of the detector to rotation, flexible and deformation object.Especially
For mess therein, wind the luggage and articles to overlap mutually and carry out careful detection and examination, study in depth its color,
Shape and textural characteristics improve the accuracy rate of dangerous goods identification to achieve the effect that accurately identify and classify, and promote X-ray
The degree of intelligence of safety check process and security check passage by efficiency, reduce congestion, mitigate the working strength of security staff, maximum journey
Degree ground ensures the traffic trip safety of people.
In order to achieve the above objectives, a kind of safe examination system based on deep neural network of the present invention includes X-ray imaging mould
Block, detection model training study module, object identification module and safety management module, the output end of the X-ray image-forming module and
The input terminal of object identification module connects, and object identification module and detection model training study module are bi-directionally connected, object identification
The output end of module is connected with the input terminal of safety management module;The X-ray image-forming module is for obtaining the X image/video of article
Then sequence obtains digital picture by analog-to-digital conversion, and obtained digital picture is transferred to object identification module;The inspection
Model training study module is surveyed for carrying out picture training, obtains learning model, and learning model is transferred to object identification mould
Block;The object identification module is used to load the learning model in detection model training module, and article is classified and determined
Position, the type and coordinate information that will test the object identified are transmitted to safety management module;It is used in the safety management module
Article different articles is delivered in the kind of object and coordinate information for knowing output according to object identification module to transport in channel.
It further, further include object transport module, object transport module includes that article enters channel, dangerous material output is led to
Road and non-dangerous article output channel.
Further, safety management module includes information management module, alarm module, luggage control module and display mould
Block;Wherein, information management module is used to receive object classification and the location information of object identification module transmission, and according to receiving
Object classification and location information differentiate object whether be dangerous goods;Alarm module is for alarming;Luggage control module is used for
Luggage is delivered in the different channels in object transport module, display module is for showing X-ray picture and testing result.
A kind of safety inspection method based on deep neural network trains image study model first with picture, in article
When detection, the X video stream sequence of article to be detected is acquired, X video stream sequence obtains digital picture by analog-to-digital conversion;So
Load image learning model afterwards utilizes the type and coordinate of article in image study model identification digital picture;Then according to object
Article is according to type divided to different transfer passages by the type and coordinate of product.
Specifically includes the following steps:
Step 1 penetrates the imaging after article using X-ray emission device, obtains X video stream sequence, X image/video sequence
Column obtain digital picture by analog-to-digital conversion;
Step 2, load image learning model are carried out the identification of article by image study model and are determined to digital picture
Position;Described image learning model is obtained by training, specific training method are as follows: the convolutional layer of convolutional neural networks is used first,
Pond layer and full articulamentum build object training model;Then X-ray picture early period obtained is classified by safety check object category,
And the classification and coordinate information of object are marked out, wherein coordinate information includes the coordinate x, y of object central point and the long w of target frame
With wide h;Then the parameter of training pattern, including learning rate, batch processing scale, learning strategy etc. are set;Then it will mark
Picture is sent into convolutional neural networks, is trained with the convolutional neural networks put up to the picture marked, is obtained image
Learning model;Image study model is saved in model learning if achieving the desired results by then authentication image learning model
Library;If falling flat, the parameter of convolutional neural networks is adjusted, continues to train, until image study model reaches pre-
Phase effect.
Article is according to type divided to different transfer passages according to the type and coordinate of article by step 3.
Further, in step 2, for training the picture of learning model using different angle, the picture of position.
Further, the picture marked is sent into convolutional neural networks, with the convolutional neural networks put up to mark
The picture infused is trained, comprising the following steps:
S1, digital picture is partitioned by area to be tested according to X-ray background feature, is detection zone by coloured region
Domain, the digital picture are rgb image;
S2, it the channel r, the channel g and the channel b of digital picture is taken out is stored in the figure that will input convolutional neural networks
First three bit port of piece;Rgb iconic model is converted into hsv color model again, the channel h, the channel s and v for extracting hsv model are logical
Road, and it is stored in three channels behind the rgb of input picture;By the tone H of hsv color model, purity S and brightness V
Value, it is orange that hsv model is divided into organic matter, inorganic matter blue, mixture green and other color segmentations are logical at 4 colors
The training picture in 10 channels is input in convolutional neural networks by road, storage to rear four bit port for inputting picture;
S3, feature extraction is carried out to area to be tested using convolution algorithm, the character representation after convolution are as follows:
Wherein, n_in is the last one-dimensional dimension of tensor.Xk represents k-th of input matrix.Wk represents the kth of convolution kernel
A sub- convolution nuclear matrix.S (i, j) the i.e. value of the corresponding position element of the corresponding output matrix of convolution kernel W, b is bigoted amount;
S4, pond is carried out;
S5, classified using Softmax to each target frame, the bbox after being classified;
S6, loss function use focalloss loss function,
FL (pt)=- αt(1-pt)γlog(pt)
γ is focusing parameter, and γ >=0,1-pt is known as the index of modulation, αtFor adjusting positive sample and negative sample
This ratio, prospect classification use αtWhen, corresponding background classification uses 1- α, and pt is different classes of class probability;
S7, the highest bbox of confidence level is chosen as testing result output.
Further, in S2, when the value of H is at 20 °~60 °, the value of S is when the value of 0.4~1.0, V is 0.4~1.0
The orange channel of organic matter;When 100 °~140 ° of value of H, the value 0.4~1.0 of S is logical for mixture green when value 0.4~1.0 of V
Road;When 220 °~260 ° of value of H, the value 0.4~1.0 of S is inorganic matter blue channel when value 0.4~1.0 of V;When H value not
When within the scope of the orange channel, green channel and blue channel, the value 0.4~1.0 of S, the value 0.4~1.0 of V is other
Color Channel.
Further, in S5, pond is carried out using the method in maximum pond.
Further, in S7, α=0.25, γ=2.
Compared with prior art, the present invention at least has technical effect beneficial below:
(1) multilevel detection image is divided and synthesized to the color characteristic based on radioscopic image, improves article identification
Accuracy rate.
(2) classify in object detection algorithms to different objects picture, by building model, adjusting parameter is utilized
Convolutional neural networks carry out feature learning, the object category in the luggage that can explicitly classify;
(3) dagger, gun, cutter etc. all have blue characteristics, gasoline, lighter, ammunition etc. in radioscopic image detection
With green characteristic, alcohol, gasoline has orange color characteristic, and the color characteristic of dangerous material is mainly blue, green and orange colour,
By object geometrical characteristic, textural characteristics and color characteristic are organically combined, and can greatly reduce cap gun, ornament is to dangerous material
It is interfered when detection, improves object detection and positional accuracy.
(4) in radioscopic image learning process using a large amount of (at least 500) different angles, position image data into
Row study, for fuzzy, rotation, the object of deformation can also be accurately identified, and effectively improve the accuracy rate of object detection.
(5) object classification is clear: for being broadly divided into seven kinds of classifications in terms of object detection, inflammable and explosive type objects are mainly wrapped
It includes: kerosene, liquefied petroleum gas, solid alcohol, compressed gas, firecrackers, fireworks display, fireworks etc.;Firearms and ammunition type objects specifically include that
Gun-simulation, steel ball rifle, stun-gun, gun-type lighter, bullet, blank cartridge, cartridge clip etc.;Explosive type objects specifically include that squama
Sheet TNT, plastic explosive, fuse cord, detonator, timing firecracker apparatus etc.;Controlled knife type objects specifically include that dagger, bullet
Spring knife, knife with three edges, lock knife etc.;Dangerous goods type objects specifically include that scissors, axe, kitchen knife, catapult etc.;Alert tool class article master
It include: electric shock baton, nunchakus, handcuffss, aerosol bomb etc.;
Daily row packet class article specifically includes that bottled water, bottle of liquor, liquid alcohol, glass cement etc..Object classification is clear,
Effective ancillary staff carries out safe investigation and can add object category according to the actual situation, improves safety.
(6) alarm threshold value of dangerous material type is arranged, from model training library in the safe examination system under requiring for different scenes
The middle suitable object detection model of selection is transferred in object detection module, has certain real-time gearing.
Detailed description of the invention
Fig. 1 be the X-ray Intelligent safety inspection system module map based on deep neural network;
Fig. 2 is the flow chart for being object identification detection module and safety management module;
Fig. 3 is the flow chart of model training module;
Fig. 4 is color segmentation algorithm flow chart;
Fig. 5 is X-ray original image;
Fig. 6 a is R plan view;
Fig. 6 b is G plan view;
Fig. 6 c is B plan view;
Fig. 7 a is H plan view;
Fig. 7 b is S plan view;
Fig. 7 c is V plan view;
Fig. 8 a is mixture plan view;
Fig. 8 b is organic matter plan view;
Fig. 8 c is inorganic matter plan view;
Fig. 8 d is other plan views.
Specific embodiment
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
Referring to Fig.1, a kind of safe examination system based on deep neural network includes object transport module, X-ray image-forming module, inspection
Survey model training study module, object identification module and safety management module.Its workflow is object delivery module by luggage
Article be passed to X-ray imager module detection range in, X-ray imager module issue X-ray, through the X-ray after luggage at
Picture obtains X video stream sequence, then obtains digital picture by analog-to-digital conversion, loads safety check article learning model, utilizes volume
Product neural network carries out object classification and positioning, and the classification and position for exporting picture recognition are sent to safety management module, by pacifying
Full management module determines the channel that luggage is flowed to.
Wherein, object transport module is mainly luggage and articles when transmitting safety check;
X-ray image-forming module is mainly that the luggage and articles in the X-ray penetration channel generated using X-ray transmitting tube obtain X figure
As video sequence, digital picture then is obtained by analog-to-digital conversion;
Detection model training study module is used to acquiring and marking object picture, is then fed into convolution mind grade e-learning,
Finally obtain the model of the object detection succeeded in school, and by trained Model Transfer to object identification module;
Object identification module is used to load the X-ray picture learning model of detection model training module, utilizes built-in object
Physical examination method of determining and calculating carries out object identification and positioning, will test the kind of object identified and coordinate information is transmitted to safety management mould
Block;
Alarm module and luggage control module in safety management module are used to know according to object identification module the object of output
Body type and coordinate information determine the need for alarming and article are transmitted to dangerous material channel.Safety management module includes information
Management module, alarm module, luggage control module and display module.
Wherein, information management module be used for receive object identification module transmission object classification and location information, and according to
The object classification received and location information differentiate whether object to be detected is dangerous goods, and alarm module is for alarming, luggage
Control module is for luggage to be delivered in the different channels in object transport module, and display module is for showing that screening machine works
X-ray picture in the process and the result picture that detected.
This system can be used as novel intelligent safe examination system, can also be by object identification module, model training module, bursting tube
It manages module to update into existing safe examination system, auto upgrading transformation is carried out to existing safe examination system.
Intelligent safety inspection method mainly includes detection zone extraction, plane of delineation processing, detector learning training and danger
4 parts of product intelligent measurement.
Detection zone extraction process are as follows: area to be tested is partitioned into according to X-ray background feature, for the time of a large amount of whites
It selects detection zone directly to abandon, avoids the identification operation of subsequent time-consuming, improve the speed of Articles detecting.
Picture plane treatment: mainly by picture by RGB model conversion to hsv model, hsv model includes H, S for pretreatment
Pass through tone H again with tri- planes of color of V and picture is partitioned into orange, green, blue and four planes of other colors again;
The picture for commonly entering convolutional neural networks is indicated by RGB color model, is made of R, G, B3 planes of color.
In the present invention other than this 3 planes of color, H, S, V planes of color of pretreatment stage acquisition are increased, and pass through color
Orange, green, blue and other colors generated after segmentation totally 10 planes of color, it is defeated after panel data fusion treatment
Enter convolutional neural networks and carries out target identification.
Detector learning training: Intelligent safety inspection system is using a large amount of different angles, the X-ray object picture of position, to adopting
The radioscopic image that collection obtains is classified and is marked, and marks the type and coordinate of object, and be classified as in the ratio of 8:2
Learn pictures and test pictures, is marked according to .xml needed for the original image generating algorithm of the radioscopic image of acquisition acquisition
It infuses format (including object category, size and its coordinate position in radioscopic image etc.).
It builds learning model and adjusts suitable parameters, carry out object geometrical characteristic by convolutional neural networks, texture is special
It seeks peace colour feature study, saves learning model, trained X-ray object learning model is then passed through into network communication interface
It is transmitted to object identification module.Luggage and articles are transmitted to X-ray image-forming module by transmission module, and X-ray emission device penetrates luggage
Imaging afterwards obtains X video stream sequence, and X video stream sequence obtains digital picture by analog-to-digital conversion, loads roentgenology
Model is practised, the kind of object and coordinate that will test pass to safety management module by communication interface, according to safety management module
Whether the alarm strategy and confidence threshold value decision-making system of setting alarm and whether are transmitted to dangerous material channel, and confidence threshold value can
It needs oneself to be arranged according to safety check, threshold value is set as percent 70, alarms if object detection confidence level is greater than threshold value.
Object detection module is updated on the basis of original safe examination system and will pass to screening machine by network communication interface,
The type of object and coordinate information are output in existing safety check screen by communication interface using X-ray learning model, if
The alarm threshold value for setting dangerous material suspends object conveyer belt if detecting dangerous material.
Referring to Fig. 3, detector learning training the following steps are included:
Step 1, the convolutional layer using convolutional neural networks, pond layer and full articulamentum build object training model.
Picture in X-ray picture library is classified by safety check object category, and manually marks out object in picture by step 2
Classification and coordinate information.
Step 3, the parameter that training pattern is arranged, parameter includes learning rate, batch processing scale, learning strategy etc..
The picture marked is sent into convolutional neural networks by step 4.
Step 5 is trained the picture marked using the convolutional neural networks put up, and obtains learning model.
Step 6, learning model verifying, if achieving the desired results, are saved in model learning library for learning model;If not reaching
To desired effect, then the parameter of convolutional neural networks is adjusted, continue to train, until learning model achieves the desired results.Object inspection
The MAP (Mean Average Precision) of survey can reach desired effect if value reaches percent 80.
Step 4 the following steps are included:
Step 4.1 detection zone is extracted
It is partitioned into area to be tested according to X-ray background feature, it is direct for the couple candidate detection region of a large amount of blank backgrounds
It abandons, coloured region is detection zone, avoids the identification operation of subsequent time-consuming, improves the speed of Articles detecting.
The processing of step 4.2 plane of delineation
By the HSV of input according to tone, purity, the value range of brightness is divided into the orange channel of organic matter, and inorganic matter is blue
Chrominance channel, the channel of mixture green channel and other colors.
When the value of H is at 20 °~60 °, the value of S is the orange channel of organic matter when the value of 0.4~1.0, V is 0.4~1.0;
When 100 °~140 ° of value of H, the value 0.4~1.0 of S is mixture green channel when value 0.4~1.0 of V;When 220 ° of value of H
~260 °, the value 0.4~1.0 of S is inorganic matter blue channel when value 0.4~1.0 of V;When the value of H is not described orange logical
When within the scope of road, green channel and blue channel, the value 0.4~1.0 of S, the value 0.4~1.0 of V is other Color Channels.
The r of picture, g, b are taken out in channel and are stored in the picture front three that will input convolutional neural networks by step 4.3
Channel, then rgb iconic model is converted into hsv color model, the h of hsv model is extracted, tri- channels s, v are stored in input
Three channels behind the rgb of picture, by hsv tone, purity, picture is divided into organic matter by the different value ranges of brightness
Orange, inorganic matter blue, at 4 Color Channels, storage extremely inputs latter four of picture for mixture green and other color segmentations
The training picture in 10 channels is input in convolutional neural networks by channel.
Color characteristic segmented image of the detection model training study module based on radioscopic image, synthesis collection R, G, B, H, S, V
With material information in more plane monitoring-network images of one, the accuracy rate of detection article identification is improved.
Step 5 the following steps are included:
Step 5.1. carries out feature extraction to area to be tested using convolution algorithm
Picture size 416*416 is inputted, channel 10 carries out convolution algorithm using the convolutional layer of 3*3 and 1*1.After convolution
Character representation
Wherein, n_in is the last one-dimensional dimension of tensor.Xk represents k-th of input matrix.Wk represents the kth of convolution kernel
A sub- convolution nuclear matrix.S (i, j) the i.e. value of the corresponding position element of the corresponding output matrix of convolution kernel W, b indicate bigoted amount.
The pond step 5.2.:
Using the method in maximum pond, i.e., maximum value is chosen as characteristic value to the pond region of 2*2, core step-length is 1.
Step 5.3. classifies to each Bbox using Softmax;
Step 5.4. loss function uses focalloss loss function
FL (pt)=- αt(1-pt)γlog(pt)
γ is referred to as focusing parameter, and γ >=0, (1-pt) is known as the index of modulation, αtFor adjusting positive
With the ratio of negative, prospect classification uses αtWhen, it is that different classes of classification is general that corresponding background classification, which uses 1- α, pt,
Rate;As α=0.25, γ=2, effect is best.α is used to adjust the ratio of positive and negative, when prospect classification uses
α, corresponding background classification use 1- α, and pt is different classes of class probability.
The method that step 5.5. uses local maximum, i.e. the selection highest bbox of the confidence level (rectangle region comprising object
Domain) as testing result output, the i.e. location information of object to be detected.Wherein, Bbox information includes 5 data values, is respectively
X, y, w, h, and confidence.Wherein x, y refer to the centre bit of the bounding box for the object that current grid is predicted
The coordinate set.W, h refer to that the width and height of the bounding box for the object that current grid is predicted, confidence are
Refer to the confidence level of prediction object.
For ease of understanding, picture schematic diagram is given, it is X ray picture that wherein Fig. 5, which is X ray picture, Fig. 6 a after gray processing,
R plan view, Fig. 6 b be G plan view;Fig. 6 c is the B plan view of X ray picture;Fig. 7 a is the H plan view of X ray picture;Fig. 7 b is X
The S plan view of ray diagram;Fig. 7 c is the V plan view of X ray picture;Fig. 8 a is the mixture plan view of X ray picture;Fig. 8 b penetrates for X
The organic matter plan view of line chart;Fig. 8 c is the inorganic matter plan view of X ray picture;Other plan views of X ray picture.
Referring to Fig. 2, safe examination system work the following steps are included:
Step 1: the alarm threshold value of setting different objects.
Step 1: the conveyer belt by object transport module is passed to luggage and articles.
Step 2: X-ray image-forming module issues X-ray, X video stream sequence is obtained through the x-ray imaging of luggage.
Step 3: obtaining the digital picture of luggage and articles by analog-to-digital conversion.
Step 4: the radioscopic image learning model of stress model training module.
Step 5: object is classified and is positioned using learning model.
Step 6: exporting the kind of object and coordinate information in picture and being sent to safety management module.
Step 7: deciding whether that alarm and luggage control module determine that luggage flows to by the alarm module of safety management module
Channel.Technology of the invention mainly uses deep learning to carry out object identification and positioning, in object features learning process
By the geometrical characteristic of object, object color is organically combined in textural characteristics and radioscopic image.Using a large amount of different angles and difference
The image data of position is learnt, and not only be can be very good detection identification and is obscured, rotation, the image of deformation, and can be real
When by trained model modification into a series of screening machines.In management module, administrator can be according to the dangerous journey of object
Degree setting kind of object threshold value, classification and coordinate.Intelligent safety inspection system based on deep learning mainly will be in traditional safe examination system
Artificial cognition dangerous material be improved to by deep learning come assist security staff work process, greatly reduce manually at
This, so that safe examination system is more intelligent.
The functional units in various embodiments of the present invention may be integrated into one processing unit, is also possible to each
Unit physically exists alone, and can also be integrated in one unit with two or more units.
The above content is merely illustrative of the invention's technical idea, and this does not limit the scope of protection of the present invention, all to press
According to technical idea proposed by the present invention, any changes made on the basis of the technical scheme each falls within claims of the present invention
Protection scope within.
Claims (10)
1. a kind of safe examination system based on deep neural network, which is characterized in that including X-ray image-forming module, detection model training
Study module, object identification module and safety management module, the output end of the X-ray image-forming module and object identification module it is defeated
Enter end connection, object identification module and detection model training study module are bi-directionally connected, the output end and peace of object identification module
The input terminal connection of full management module;
Then the X-ray image-forming module obtains digital picture by analog-to-digital conversion for obtaining the X video stream sequence of article,
And obtained digital picture is transferred to object identification module;
The detection model training study module obtains learning model, and learning model is transferred to for carrying out picture training
Object identification module;
The object identification module is used to load the learning model in detection model training module, and article is classified and determined
Position, the type and coordinate information that will test the object identified are transmitted to safety management module;
Kind of object and coordinate information in the safety management module for knowing output according to object identification module is defeated by article
It send to different articles and transports in channel.
2. a kind of safe examination system based on deep neural network according to claim 1, which is characterized in that further include object
Transmission module, object transport module include that article enters channel, dangerous material output channel and non-dangerous article output channel.
3. a kind of safe examination system based on deep neural network according to claim 1, which is characterized in that safety management mould
Block includes information management module, alarm module, luggage control module and display module;Wherein, information management module is for receiving
The object classification and location information that object identification module is sent, and object is differentiated according to the object classification and location information that receive
It whether is dangerous goods;Alarm module is for alarming;Luggage control module is for luggage to be delivered in object transport module
In different channels, display module is for showing X-ray picture and testing result.
4. a kind of safety inspection method based on deep neural network, which is characterized in that train image study mould first with picture
Type acquires the X video stream sequence of article to be detected in Articles detecting, and X video stream sequence is counted by analog-to-digital conversion
Word picture;Then load image learning model utilizes the type and coordinate of article in image study model identification digital picture;So
Article is according to type divided to according to the type of article and coordinate by different transfer passages afterwards.
5. a kind of safety inspection method based on deep neural network according to claim 4, which is characterized in that including following step
It is rapid:
Step 1 penetrates the imaging after article using X-ray emission device, obtains X video stream sequence, X video stream sequence warp
It crosses analog-to-digital conversion and obtains digital picture;
Step 2, load image learning model carry out the identification and positioning of article by image study model to digital picture;
Described image learning model is obtained by training, specific training method are as follows: uses the convolutional layer of convolutional neural networks, Chi Hua first
Layer and full articulamentum build object training model;Then X-ray picture early period obtained is classified by safety check object category, and is marked
The classification and coordinate information of object are outpoured, wherein coordinate information includes the coordinate x, y of object central point and the long w and width of target frame
h;Then the parameter of training pattern, including learning rate, batch processing scale and learning strategy are set;Then the picture marked is sent
Enter in convolutional neural networks, the picture marked is trained with the convolutional neural networks put up, obtains image study mould
Type;Image study model is saved in model learning library if achieving the desired results by then authentication image learning model;If not
It achieves the desired results, then adjusts the parameter of convolutional neural networks, continue to train, until image study model achieves the desired results;
Article is according to type divided to different transfer passages according to the type and coordinate of article by step 3.
6. a kind of safety inspection method based on deep neural network according to claim 5, which is characterized in that in step 2, use
The picture of different angle, position is used in the picture of training learning model.
7. a kind of safety inspection method based on deep neural network according to claim 5, which is characterized in that, will in step 2
The picture marked is sent into convolutional neural networks, is trained with the convolutional neural networks put up to the picture marked,
The following steps are included:
S1, digital picture is partitioned by area to be tested according to X-ray background feature, is detection zone by coloured region,
The digital picture is rgb image;
S2, before the picture for being stored in and will inputting convolutional neural networks is taken out in the channel r, the channel g and the channel b of digital picture
Three bit ports;Rgb iconic model is converted into hsv color model again, extracts the channel h, the channel s and the channel v of hsv model,
And it is stored in three channels behind the rgb of input picture;By the tone H of hsv color model, purity S and brightness V's
It is orange to be divided into organic matter by value for hsv model, inorganic matter blue, mixture green and other color segmentations at 4 Color Channels,
The training picture in 10 channels is input in convolutional neural networks by storage to rear four bit port for inputting picture;
S3, feature extraction is carried out to area to be tested using convolution algorithm, the character representation after convolution are as follows:
Wherein, n_in is the last one-dimensional dimension of tensor, and Xk represents k-th of input matrix, and Wk represents k-th of son of convolution kernel
Convolution nuclear matrix, s (i, j) the i.e. value of the corresponding position element of the corresponding output matrix of convolution kernel W, b is bigoted amount;
S4, pond is carried out;
S5, classified using Softmax to each target frame, the bbox after being classified;
S6, loss function use focalloss loss function,
FL (pt)=- αt(1-pt)γlog(pt)
γ is focusing parameter, and γ >=0,1-pt is known as the index of modulation, αtFor adjusting positive sample and negative sample
Ratio, prospect classification use αtWhen, corresponding background classification uses 1- α, and pt is different classes of class probability;
S7, the highest bbox of confidence level is chosen as testing result output.
8. a kind of safety inspection method based on deep neural network according to claim 7, which is characterized in that in S2, when H's
For value at 20 °~60 °, the value of S is the orange channel of organic matter when the value of 0.4~1.0, V is 0.4~1.0;When H 100 ° of value~
140 °, the value 0.4~1.0 of S is mixture green channel when value 0.4~1.0 of V;When 220 °~260 ° of value of H, the value of S
It is inorganic matter blue channel when value 0.4~1.0 of 0.4~1.0, V;When H value not the orange channel, green channel and
When within the scope of blue channel, the value 0.4~1.0 of S, the value 0.4~1.0 of V is other Color Channels.
9. a kind of safety inspection method based on deep neural network according to claim 7, which is characterized in that in S5, use
The method in maximum pond carries out pond.
10. a kind of safety inspection method based on deep neural network according to claim 7, which is characterized in that in S7, α=
0.25, γ=2.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910218654.4A CN109946746A (en) | 2019-03-21 | 2019-03-21 | A kind of safe examination system and method based on deep neural network |
PCT/CN2020/078425 WO2020187077A1 (en) | 2019-03-21 | 2020-03-09 | Deep neural network-based security check system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910218654.4A CN109946746A (en) | 2019-03-21 | 2019-03-21 | A kind of safe examination system and method based on deep neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109946746A true CN109946746A (en) | 2019-06-28 |
Family
ID=67010557
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910218654.4A Pending CN109946746A (en) | 2019-03-21 | 2019-03-21 | A kind of safe examination system and method based on deep neural network |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109946746A (en) |
WO (1) | WO2020187077A1 (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110286415A (en) * | 2019-07-12 | 2019-09-27 | 广东工业大学 | Safety check contraband detecting method, apparatus, equipment and computer readable storage medium |
CN110488368A (en) * | 2019-07-26 | 2019-11-22 | 中控智慧科技股份有限公司 | A kind of contraband recognition methods and device based on dual intensity X-ray screening machine |
CN110533045A (en) * | 2019-07-31 | 2019-12-03 | 中国民航大学 | A kind of luggage X-ray contraband image, semantic dividing method of combination attention mechanism |
CN110751329A (en) * | 2019-10-17 | 2020-02-04 | 中国民用航空总局第二研究所 | Control method and device for airport security check channel, electronic equipment and storage medium |
CN110794466A (en) * | 2019-07-16 | 2020-02-14 | 中云智慧(北京)科技有限公司 | X-ray machine picture acquisition auxiliary device and processing method |
CN111046908A (en) * | 2019-11-05 | 2020-04-21 | 杭州电子科技大学 | Emulsion explosive package fault real-time monitoring model based on convolutional neural network |
CN111062252A (en) * | 2019-11-15 | 2020-04-24 | 浙江大华技术股份有限公司 | Real-time dangerous article semantic segmentation method and device and storage device |
CN111091150A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon cross rod cover plate fracture detection method |
CN111126238A (en) * | 2019-12-19 | 2020-05-08 | 华南理工大学 | X-ray security inspection system and method based on convolutional neural network |
CN111167733A (en) * | 2020-01-14 | 2020-05-19 | 东莞理工学院 | Automatic visual defect detection system |
CN111429410A (en) * | 2020-03-13 | 2020-07-17 | 杭州电子科技大学 | Object X-ray image material distinguishing system and method based on deep learning |
WO2020187077A1 (en) * | 2019-03-21 | 2020-09-24 | 长安大学 | Deep neural network-based security check system and method |
CN112144150A (en) * | 2020-10-16 | 2020-12-29 | 北京经纬纺机新技术有限公司 | Distributed foreign fiber sorting system applying deep learning image processing |
CN112764120A (en) * | 2020-12-29 | 2021-05-07 | 深圳市创艺龙电子科技有限公司 | Metal detection imaging system and scanning device with same |
CN112861888A (en) * | 2019-11-08 | 2021-05-28 | 中国科学院深圳先进技术研究院 | Automatic identification device and method for limiter |
CN113159110A (en) * | 2021-03-05 | 2021-07-23 | 安徽启新明智科技有限公司 | X-ray-based liquid intelligent detection method |
US20210303924A1 (en) * | 2020-03-31 | 2021-09-30 | Hcl Technologies Limited | Method and system for generating and labelling reference images |
CN114332543A (en) * | 2022-01-10 | 2022-04-12 | 成都智元汇信息技术股份有限公司 | Multi-template security inspection image identification method, equipment and medium |
WO2023087653A1 (en) * | 2021-11-17 | 2023-05-25 | 湖南苏科智能科技有限公司 | Dual-view-angle associated security check method and system based on neural network and multi-source data |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112364903A (en) * | 2020-10-30 | 2021-02-12 | 盛视科技股份有限公司 | X-ray machine-based article analysis and multi-dimensional image association method and system |
CN112444889A (en) * | 2020-11-13 | 2021-03-05 | 北京航星机器制造有限公司 | Rapid security inspection luggage remote centralized interpretation system and method |
CN112991256A (en) * | 2020-12-11 | 2021-06-18 | 中国石油天然气股份有限公司 | Oil pipe material statistical method based on machine vision and deep learning |
CN112560944B (en) * | 2020-12-14 | 2023-08-08 | 广东电网有限责任公司珠海供电局 | Charging pile fire detection method based on image recognition |
CN113762023B (en) * | 2021-02-18 | 2024-05-24 | 北京京东振世信息技术有限公司 | Object identification method and device based on article association relation |
CN113537213B (en) * | 2021-07-14 | 2024-01-30 | 安徽炬视科技有限公司 | Smoke open fire detection algorithm based on variable convolution kernel |
CN113435543B (en) * | 2021-07-22 | 2024-04-09 | 湖南声迅科技有限公司 | Visible light and X-ray image matching method and device based on conveyor belt identification |
CN113759433B (en) * | 2021-08-12 | 2024-02-27 | 浙江啄云智能科技有限公司 | Prohibited article screening method and device and security inspection equipment |
CN115731213B (en) * | 2022-11-29 | 2024-01-30 | 北京声迅电子股份有限公司 | Edge tool detection method based on X-ray image |
CN116610078A (en) * | 2023-05-19 | 2023-08-18 | 广东海力储存设备股份有限公司 | Automatic storage control method and system for stereoscopic warehouse, electronic equipment and storage medium |
CN117197787A (en) * | 2023-08-09 | 2023-12-08 | 海南大学 | Intelligent security inspection method, device, equipment and medium based on improved YOLOv5 |
CN117409199B (en) * | 2023-10-19 | 2024-05-14 | 中南大学 | Growing intelligent security inspection system and method based on cloud big data technology |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105116463A (en) * | 2015-09-22 | 2015-12-02 | 同方威视技术股份有限公司 | Security check channel and security check apparatus |
CN205656321U (en) * | 2016-05-30 | 2016-10-19 | 公安部第一研究所 | A X ray irradiation means for discerning hazardous liquid in parcel |
CN206192923U (en) * | 2016-10-27 | 2017-05-24 | 中云智慧(北京)科技有限公司 | X-ray contraband detection system based on cloud computing |
CN106872498A (en) * | 2017-04-11 | 2017-06-20 | 西安培华学院 | A kind of contraband safety check automatic identification equipment |
CN107871122A (en) * | 2017-11-14 | 2018-04-03 | 深圳码隆科技有限公司 | Safety check detection method, device, system and electronic equipment |
CN108198227A (en) * | 2018-03-16 | 2018-06-22 | 济南飞象信息科技有限公司 | Contraband intelligent identification Method based on X-ray screening machine image |
CN108226196A (en) * | 2018-02-22 | 2018-06-29 | 青岛智慧云谷智能科技有限公司 | A kind of intelligence X-ray machine detection system and detection method |
CN108802840A (en) * | 2018-05-31 | 2018-11-13 | 北京迈格斯智能科技有限公司 | The method and its device of automatic identification object based on artificial intelligence deep learning |
CN109086679A (en) * | 2018-07-10 | 2018-12-25 | 西安恒帆电子科技有限公司 | A kind of millimetre-wave radar safety check instrument foreign matter detecting method |
CN109187598A (en) * | 2018-10-09 | 2019-01-11 | 青海奥越电子科技有限公司 | Prohibited items detection system and method based on Digital Image Processing |
CN109191389A (en) * | 2018-07-31 | 2019-01-11 | 浙江杭钢健康产业投资管理有限公司 | A kind of x-ray image adaptive local Enhancement Method |
CN109447071A (en) * | 2018-11-01 | 2019-03-08 | 博微太赫兹信息科技有限公司 | A kind of mm-wave imaging dangerous goods detection method based on FPGA and deep learning |
CN109472309A (en) * | 2018-11-12 | 2019-03-15 | 南京烽火星空通信发展有限公司 | A kind of X-ray screening machine picture object detecting method |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2003065077A2 (en) * | 2002-01-30 | 2003-08-07 | Rutgers, The State University | Combinatorial contraband detection using energy dispersive x-ray diffraction |
CN106599967A (en) * | 2016-12-08 | 2017-04-26 | 同方威视技术股份有限公司 | Security check item positioning label and security check item positioning method |
CN108303747B (en) * | 2017-01-12 | 2023-03-07 | 清华大学 | Inspection apparatus and method of detecting a gun |
CN107145898B (en) * | 2017-04-14 | 2020-04-10 | 北京航星机器制造有限公司 | Radiographic image classification method based on neural network |
CN107607562B (en) * | 2017-09-11 | 2020-06-02 | 北京匠数科技有限公司 | Forbidden article identification equipment and method and X-ray luggage security inspection system |
CN109446888A (en) * | 2018-09-10 | 2019-03-08 | 唯思科技(北京)有限公司 | A kind of elongated class article detection method based on convolutional neural networks |
CN109946746A (en) * | 2019-03-21 | 2019-06-28 | 长安大学 | A kind of safe examination system and method based on deep neural network |
-
2019
- 2019-03-21 CN CN201910218654.4A patent/CN109946746A/en active Pending
-
2020
- 2020-03-09 WO PCT/CN2020/078425 patent/WO2020187077A1/en active Application Filing
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105116463A (en) * | 2015-09-22 | 2015-12-02 | 同方威视技术股份有限公司 | Security check channel and security check apparatus |
CN205656321U (en) * | 2016-05-30 | 2016-10-19 | 公安部第一研究所 | A X ray irradiation means for discerning hazardous liquid in parcel |
CN206192923U (en) * | 2016-10-27 | 2017-05-24 | 中云智慧(北京)科技有限公司 | X-ray contraband detection system based on cloud computing |
CN106872498A (en) * | 2017-04-11 | 2017-06-20 | 西安培华学院 | A kind of contraband safety check automatic identification equipment |
CN107871122A (en) * | 2017-11-14 | 2018-04-03 | 深圳码隆科技有限公司 | Safety check detection method, device, system and electronic equipment |
CN108226196A (en) * | 2018-02-22 | 2018-06-29 | 青岛智慧云谷智能科技有限公司 | A kind of intelligence X-ray machine detection system and detection method |
CN108198227A (en) * | 2018-03-16 | 2018-06-22 | 济南飞象信息科技有限公司 | Contraband intelligent identification Method based on X-ray screening machine image |
CN108802840A (en) * | 2018-05-31 | 2018-11-13 | 北京迈格斯智能科技有限公司 | The method and its device of automatic identification object based on artificial intelligence deep learning |
CN109086679A (en) * | 2018-07-10 | 2018-12-25 | 西安恒帆电子科技有限公司 | A kind of millimetre-wave radar safety check instrument foreign matter detecting method |
CN109191389A (en) * | 2018-07-31 | 2019-01-11 | 浙江杭钢健康产业投资管理有限公司 | A kind of x-ray image adaptive local Enhancement Method |
CN109187598A (en) * | 2018-10-09 | 2019-01-11 | 青海奥越电子科技有限公司 | Prohibited items detection system and method based on Digital Image Processing |
CN109447071A (en) * | 2018-11-01 | 2019-03-08 | 博微太赫兹信息科技有限公司 | A kind of mm-wave imaging dangerous goods detection method based on FPGA and deep learning |
CN109472309A (en) * | 2018-11-12 | 2019-03-15 | 南京烽火星空通信发展有限公司 | A kind of X-ray screening machine picture object detecting method |
Cited By (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020187077A1 (en) * | 2019-03-21 | 2020-09-24 | 长安大学 | Deep neural network-based security check system and method |
CN110286415A (en) * | 2019-07-12 | 2019-09-27 | 广东工业大学 | Safety check contraband detecting method, apparatus, equipment and computer readable storage medium |
CN110794466A (en) * | 2019-07-16 | 2020-02-14 | 中云智慧(北京)科技有限公司 | X-ray machine picture acquisition auxiliary device and processing method |
CN110488368A (en) * | 2019-07-26 | 2019-11-22 | 中控智慧科技股份有限公司 | A kind of contraband recognition methods and device based on dual intensity X-ray screening machine |
CN110488368B (en) * | 2019-07-26 | 2021-07-30 | 熵基科技股份有限公司 | Contraband identification method and device based on dual-energy X-ray security inspection machine |
CN110533045A (en) * | 2019-07-31 | 2019-12-03 | 中国民航大学 | A kind of luggage X-ray contraband image, semantic dividing method of combination attention mechanism |
CN110533045B (en) * | 2019-07-31 | 2023-01-17 | 中国民航大学 | Luggage X-ray contraband image semantic segmentation method combined with attention mechanism |
CN110751329A (en) * | 2019-10-17 | 2020-02-04 | 中国民用航空总局第二研究所 | Control method and device for airport security check channel, electronic equipment and storage medium |
CN111046908A (en) * | 2019-11-05 | 2020-04-21 | 杭州电子科技大学 | Emulsion explosive package fault real-time monitoring model based on convolutional neural network |
CN112861888A (en) * | 2019-11-08 | 2021-05-28 | 中国科学院深圳先进技术研究院 | Automatic identification device and method for limiter |
CN111062252B (en) * | 2019-11-15 | 2023-11-10 | 浙江大华技术股份有限公司 | Real-time dangerous goods semantic segmentation method, device and storage device |
CN111062252A (en) * | 2019-11-15 | 2020-04-24 | 浙江大华技术股份有限公司 | Real-time dangerous article semantic segmentation method and device and storage device |
CN111091150A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon cross rod cover plate fracture detection method |
CN111126238A (en) * | 2019-12-19 | 2020-05-08 | 华南理工大学 | X-ray security inspection system and method based on convolutional neural network |
CN111126238B (en) * | 2019-12-19 | 2023-06-20 | 华南理工大学 | X-ray security inspection system and method based on convolutional neural network |
CN111167733A (en) * | 2020-01-14 | 2020-05-19 | 东莞理工学院 | Automatic visual defect detection system |
CN111429410A (en) * | 2020-03-13 | 2020-07-17 | 杭州电子科技大学 | Object X-ray image material distinguishing system and method based on deep learning |
CN111429410B (en) * | 2020-03-13 | 2023-09-01 | 杭州电子科技大学 | Object X-ray image material discrimination system and method based on deep learning |
US11989890B2 (en) * | 2020-03-31 | 2024-05-21 | Hcl Technologies Limited | Method and system for generating and labelling reference images |
US20210303924A1 (en) * | 2020-03-31 | 2021-09-30 | Hcl Technologies Limited | Method and system for generating and labelling reference images |
CN112144150A (en) * | 2020-10-16 | 2020-12-29 | 北京经纬纺机新技术有限公司 | Distributed foreign fiber sorting system applying deep learning image processing |
CN112764120A (en) * | 2020-12-29 | 2021-05-07 | 深圳市创艺龙电子科技有限公司 | Metal detection imaging system and scanning device with same |
CN113159110A (en) * | 2021-03-05 | 2021-07-23 | 安徽启新明智科技有限公司 | X-ray-based liquid intelligent detection method |
WO2023087653A1 (en) * | 2021-11-17 | 2023-05-25 | 湖南苏科智能科技有限公司 | Dual-view-angle associated security check method and system based on neural network and multi-source data |
CN114332543B (en) * | 2022-01-10 | 2023-02-14 | 成都智元汇信息技术股份有限公司 | Multi-template security inspection image identification method, equipment and medium |
CN114332543A (en) * | 2022-01-10 | 2022-04-12 | 成都智元汇信息技术股份有限公司 | Multi-template security inspection image identification method, equipment and medium |
Also Published As
Publication number | Publication date |
---|---|
WO2020187077A1 (en) | 2020-09-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109946746A (en) | A kind of safe examination system and method based on deep neural network | |
CN110018524B (en) | X-ray security inspection contraband identification method based on vision-attribute | |
CN108198227A (en) | Contraband intelligent identification Method based on X-ray screening machine image | |
CN109977897A (en) | A kind of ship's particulars based on deep learning recognition methods, application method and system again | |
CN105842173B (en) | A kind of EO-1 hyperion material discrimination method | |
CN108710868A (en) | A kind of human body critical point detection system and method based under complex scene | |
US20070058037A1 (en) | User interface for use in screening luggage, containers, parcels or people and apparatus for implementing same | |
CN109284669A (en) | Pedestrian detection method based on Mask RCNN | |
CN107729818A (en) | A kind of multiple features fusion vehicle recognition methods again based on deep learning | |
CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
CN105809121A (en) | Multi-characteristic synergic traffic sign detection and identification method | |
CA2584683A1 (en) | Apparatus, method and system for screening receptacles and persons | |
CN108288279A (en) | Article discrimination method based on X-ray image foreground target extraction | |
CN110020647A (en) | A kind of contraband object detection method, device and computer equipment | |
CN109190491A (en) | Residual error convolutional neural networks SAR image sea ice classification method | |
CN109284786A (en) | The SAR image terrain classification method of confrontation network is generated based on distribution and structure matching | |
CN109522924A (en) | A kind of broad-leaf forest wood recognition method based on single photo | |
CN110488368A (en) | A kind of contraband recognition methods and device based on dual intensity X-ray screening machine | |
Pechiammal et al. | An efficient approach for automatic license plate recognition system | |
CN112712093A (en) | Security check image identification method and device, electronic equipment and storage medium | |
CN113963222A (en) | High-resolution remote sensing image change detection method based on multi-strategy combination | |
CN110060273A (en) | Remote sensing image landslide plotting method based on deep neural network | |
CN108334937A (en) | A kind of oil film relative thickness extracting method and system waterborne based on convolutional neural networks | |
CN106951863A (en) | A kind of substation equipment infrared image change detecting method based on random forest | |
CN116311078A (en) | Forest fire analysis and monitoring method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190628 |
|
RJ01 | Rejection of invention patent application after publication |