CN110321851A - A kind of prohibited items detection method, device and equipment - Google Patents
A kind of prohibited items detection method, device and equipment Download PDFInfo
- Publication number
- CN110321851A CN110321851A CN201910603933.2A CN201910603933A CN110321851A CN 110321851 A CN110321851 A CN 110321851A CN 201910603933 A CN201910603933 A CN 201910603933A CN 110321851 A CN110321851 A CN 110321851A
- Authority
- CN
- China
- Prior art keywords
- pixel characteristic
- prohibited items
- picture
- identified
- pixel
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 58
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 238000005070 sampling Methods 0.000 claims description 31
- 238000000034 method Methods 0.000 claims description 25
- 238000010606 normalization Methods 0.000 claims description 21
- 238000012360 testing method Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 230000002708 enhancing effect Effects 0.000 claims description 7
- 230000004927 fusion Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 4
- 238000007689 inspection Methods 0.000 claims description 4
- 230000006835 compression Effects 0.000 claims description 3
- 238000007906 compression Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000012216 screening Methods 0.000 description 5
- 230000001629 suppression Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000011835 investigation Methods 0.000 description 3
- 239000003990 capacitor Substances 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000007499 fusion processing Methods 0.000 description 2
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/05—Recognition of patterns representing particular kinds of hidden objects, e.g. weapons, explosives, drugs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Mathematical Physics (AREA)
- Tourism & Hospitality (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Image Analysis (AREA)
- General Business, Economics & Management (AREA)
- Computer Security & Cryptography (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Educational Administration (AREA)
- Multimedia (AREA)
- Development Economics (AREA)
Abstract
The invention discloses a kind of prohibited items detection methods, due to the pixel characteristic of the picture to be identified extracted by the convolutional neural networks with multilayer convolution kernel, the interference of background characteristics in picture to be identified can be eliminated, only leave and take the pixel characteristic of target object, thus, after the second pixel characteristic figure after normalizing is merged with the first pixel characteristic figure, it can reinforce the pixel characteristic of each target object, to improve the recognition accuracy of various prohibited items, naturally the recognition accuracy of the violated object of small size is also improved, thoroughly prohibited items can be checked, eliminate security risk.The invention also discloses a kind of prohibited items detection device and equipment, have the identical beneficial effect of prohibited items detection method as above.
Description
Technical field
The present invention relates to picture recognition fields, and more particularly to a kind of prohibited items detection method, the invention further relates to one
Kind prohibited items detection device and equipment.
Background technique
The inspection of prohibited items can greatly guarantee the biggish occasion of the density of population (such as subway and train traffic pivot
Knob) the masses the security of the lives and property, (such as x-ray scanning) would generally be scanned to article in the prior art and obtains scanning figure
Piece can classify to the article in picture using classifier later and detect prohibited items, but not had in the prior art
A kind of detection method of maturation, the phenomenon that often will appear the prohibited items that can not identify small volume, for prohibited items
Investigation it is not thorough enough, there are security risks.
Therefore, how to provide a kind of scheme of solution above-mentioned technical problem is that those skilled in the art need to solve at present
Problem.
Summary of the invention
The object of the present invention is to provide a kind of prohibited items detection method, the identification for improving the violated object of small size is accurate
Rate can thoroughly check prohibited items, eliminate security risk；It is a further object of the present invention to provide a kind of violated
Article detection apparatus and equipment improve the recognition accuracy of the violated object of small size, can thoroughly carry out to prohibited items
Investigation, eliminates security risk.
In order to solve the above technical problems, the present invention provides a kind of prohibited items detection methods, comprising:
The pixel characteristic for extracting picture to be identified obtains the first pixel characteristic figure；
The pixel characteristic that the picture to be identified is extracted by the convolutional neural networks with multilayer convolution kernel, obtains second
Pixel characteristic figure；
The first pixel characteristic figure is merged with the second pixel characteristic figure after normalization, it is special to obtain third pixel
Sign figure；
The identification of prohibited items is carried out according to the third pixel characteristic figure.
Preferably, the pixel for extracting the picture to be identified by the convolutional neural networks with multilayer convolution kernel is special
Sign, obtains the second pixel characteristic figure specifically:
Multiple down-sampling operation is carried out to the picture to be identified by the convolutional neural networks with multilayer convolution kernel；
Up-sampling operation is carried out to the picture to be identified after down-sampling；
By the channel information of the to be identified picture of two 1*1 convolution kernel compressions after up-sampling, second is obtained
Pixel characteristic figure.
Preferably, the described pair of picture to be identified after down-sampling carries out up-sampling operation specifically:
Up-sampling operation is carried out to the picture to be identified after down-sampling using bilinear interpolation.
Preferably, described to merge the first pixel characteristic figure with the second pixel characteristic figure after normalization, it obtains
To third pixel characteristic figure specifically:
The first pixel characteristic figure and the second pixel characteristic figure after normalization are subjected to matrix dot product,
The the first pixel characteristic figure enhanced；
The first pixel characteristic figure and the first pixel characteristic figure of enhancing are added, the third picture of fusion is exported
Plain characteristic pattern.
Preferably, the identification that prohibited items are carried out according to the third pixel characteristic figure specifically:
The third pixel characteristic figure input area candidate network RPN network is obtained with preset number candidate regions
The 4th pixel characteristic figure in domain；
The 4th pixel characteristic figure is sequentially inputted to classifier and returns device, is obtained for the picture to be identified
Testing result.
Preferably, the second pixel characteristic figure after the normalization specifically:
The pixel value in the second pixel characteristic figure is converted into the probability value between 0 to 1 by sigmoid function,
The second pixel characteristic figure after obtained normalization.
Preferably, the pixel characteristic for extracting picture to be identified, before obtaining the first pixel characteristic figure, the prohibited items
Detection method further include:
Picture to be identified is carried out to the overturning processing of preset kind corresponding with the quantity of i, the initial value of the i is pre-
If numerical value；
Then after the identification for carrying out prohibited items according to the third pixel characteristic figure, the prohibited items detection method
Further include:
Judge whether i is equal to default value；
If it is not, then by the numerical value+1 of i and return step:
Picture to be identified is carried out to the overturning processing of preset kind corresponding with the quantity of i；
If so, the prohibited items recognition result of all pictures to be identified for different overturning types is carried out
Non-maxima suppression obtains final prohibited items testing result.
Preferably, it is described obtain final prohibited items testing result after, the prohibited items detection method further include:
Control the final prohibited items testing result of prompting device prompt.
In order to solve the above technical problems, the present invention also provides a kind of prohibited items detection devices, comprising:
First extraction module obtains the first pixel characteristic figure for extracting the pixel characteristic of picture to be identified；
Second extraction module, for extracting the picture to be identified by the convolutional neural networks with multilayer convolution kernel
Pixel characteristic obtains the second pixel characteristic figure；
Fusion Module, for the first pixel characteristic figure to be merged with the second pixel characteristic figure after normalization,
Obtain third pixel characteristic figure；
Identification module, for carrying out the identification of prohibited items according to the third pixel characteristic figure.
In order to solve the above technical problems, the present invention also provides a kind of prohibited items detection devices, comprising:
Memory, for storing computer program；
Processor realizes the step of the as above any one prohibited items detection method when for executing the computer program
Suddenly.
The present invention provides a kind of prohibited items detection methods, due to passing through the convolutional neural networks with multilayer convolution kernel
The pixel characteristic of the picture to be identified extracted, can eliminate the interference of background characteristics in picture to be identified, only leave and take object
The pixel characteristic of body, in this way, after the second pixel characteristic figure after normalizing is merged with the first pixel characteristic figure, it can
To reinforce the pixel characteristic of each target object, to improve the recognition accuracy of various prohibited items, also improve naturally small
The recognition accuracy of the violated object of volume, can thoroughly check prohibited items, eliminate security risk.
The present invention also provides a kind of prohibited items detection device and equipment, have prohibited items detection method as above identical
Beneficial effect.
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to institute in the prior art and embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without creative efforts, can also obtain according to these attached drawings
Obtain other attached drawings.
Fig. 1 is a kind of flow diagram of prohibited items detection method provided by the invention；
Fig. 2 is a kind of structural schematic diagram of prohibited items detection device provided by the invention；
Fig. 3 is a kind of structural schematic diagram of prohibited items detection device provided by the invention.
Specific embodiment
Core of the invention is to provide a kind of prohibited items detection method, and the identification for improving the violated object of small size is accurate
Rate can thoroughly check prohibited items, eliminate security risk；Another core of the invention is to provide a kind of violated
Article detection apparatus and equipment improve the recognition accuracy of the violated object of small size, can thoroughly carry out to prohibited items
Investigation, eliminates security risk.
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Referring to FIG. 1, Fig. 1 is a kind of flow diagram of prohibited items detection method provided by the invention, comprising:
Step S1: extracting the pixel characteristic of picture to be identified, obtains the first pixel characteristic figure；
Specifically, being extracted in this step to be identified there are target object and remaining background area in picture to be identified
The purpose of the pixel characteristic of picture is to extract the pixel characteristic of target object have object to utilize in subsequent step
First pixel characteristic figure of volumetric pixel feature carries out the identification of prohibited items.
Wherein, picture to be identified can be a plurality of types of pictures, such as can obtain for subway and railway station screening machine
The X-ray picture etc. obtained, the embodiment of the present invention is it is not limited here.
Specifically, the prohibited items in the embodiment of the present invention can be a variety of, such as can be handed over for subway and train etc.
Logical tool forbids the dangerous goods carried, and including controlled knife etc., the embodiment of the present invention is it is not limited here.
Specifically, the embodiment of the present invention can select the Fasterrcnn using Resnet as convolution feature extractor to detect
Device carries out, which can use Tensorflow frame to construct.
It is, of course, also possible to select other kinds of detector to carry out the feature extraction in the embodiment of the present invention,
The embodiment of the present invention is it is not limited here.
Step S2: the pixel characteristic of picture to be identified is extracted by the convolutional neural networks with multilayer convolution kernel, is obtained
Second pixel characteristic figure；
Specifically, with multilayer convolution kernel convolutional neural networks can to the target object in picture to be identified carry out compared with
The identification of profound pixel characteristic, and background characteristics is weeded out, to extract the pixel of more pure target object
Feature, and as the second pixel characteristic figure.
Wherein, the number of plies of convolution kernel can independently be set according to actual needs, such as can be set to 5 layers etc., this
Inventive embodiments are it is not limited here.
Wherein, an average pondization operation can be connect behind every layer of convolution kernel, the embodiment of the present invention is it is not limited here.
Step S3: the first pixel characteristic figure is merged with the second pixel characteristic figure after normalization, it is special to obtain third pixel
Sign figure；
Specifically, normalized can convert the pixel value of the pixel characteristic of target object in the second pixel characteristic figure
For the probability value within 0-1, that is, representing the pixel value to have great probability is some object, such as some pixel value has 0.8
Probability is fruit knife etc., and the embodiment of the present invention is it is not limited here.
Wherein, the second pixel characteristic figure after normalization is merged with the first pixel characteristic figure, can use the second pixel
The pixel characteristic of more pure target object enhances the first pixel characteristic figure in characteristic pattern, so as to subsequent step utilization
Third pixel characteristic figure more accurately identifies each target object, i.e. prohibited items.
Step S4: the identification of prohibited items is carried out according to third pixel characteristic figure.
Specifically, due to having carried out target object to the first pixel characteristic figure using the second pixel characteristic figure after normalization
The enhancing of pixel characteristic, to obtain third pixel characteristic figure, therefore can be more accurately using third pixel characteristic figure
It identifies various target objects, to also can more accurately identify the prohibited items of small size, reduces omission factor, arrange
In addition to security risk.
The present invention provides a kind of prohibited items detection methods, due to passing through the convolutional neural networks with multilayer convolution kernel
The pixel characteristic of the picture to be identified extracted, can eliminate the interference of background characteristics in picture to be identified, only leave and take object
The pixel characteristic of body, in this way, after the second pixel characteristic figure after normalizing is merged with the first pixel characteristic figure, it can
To reinforce the pixel characteristic of each target object, to improve the recognition accuracy of various prohibited items, also improve naturally small
The recognition accuracy of the violated object of volume, can thoroughly check prohibited items, eliminate security risk.
On the basis of the above embodiments:
Embodiment as one preferred extracts picture to be identified by the convolutional neural networks with multilayer convolution kernel
Pixel characteristic obtains the second pixel characteristic figure specifically:
Multiple down-sampling operation is carried out to picture to be identified by the convolutional neural networks with multilayer convolution kernel；
Up-sampling operation is carried out to the picture to be identified after down-sampling；
By the channel information of to be identified picture of two 1*1 convolution kernel compressions after up-sampling, the second pixel is obtained
Characteristic pattern.
Specifically, carrying out multiple down-sampling operation to picture to be identified by the convolutional neural networks with multilayer convolution kernel
Can the pixel characteristic with relatively deep to target object in picture to be identified carry out excavation extraction.
Specifically, up-sampling operation can amplify the picture after down-sampling operation is reduced, obtain and first
The identical second pixel characteristic figure of pixel characteristic figure size is convenient for subsequent fusion treatment.
Wherein, during carrying out down-sampling using the convolutional neural networks with multilayer convolution kernel, it is possible to cause
The channel information of second pixel characteristic figure is more, inconsistent with the first pixel characteristic figure, influences subsequent fusion process, therefore this
The channel information of the picture to be identified after up-sampling can be compressed in inventive embodiments by two 1*1 convolution kernels, so that
The channel information of second pixel characteristic figure is consistent with the first pixel characteristic figure, is convenient for subsequent fusion process.
Embodiment as one preferred carries out up-sampling operation to the picture to be identified after down-sampling specifically:
Up-sampling operation is carried out to the picture to be identified after down-sampling using bilinear interpolation.
Specifically, bilinear interpolation can rapidly and accurately carry out up-sampling operation.
Certainly, other than bilinear interpolation, it can also adopt and carry out up-sampling operation with other methods, the present invention is implemented
Example is it is not limited here.
Embodiment as one preferred merges the first pixel characteristic figure with the second pixel characteristic figure after normalization,
Obtain third pixel characteristic figure specifically:
First pixel characteristic figure and the second pixel characteristic figure after normalization are subjected to matrix dot product, enhanced
The first pixel characteristic figure；
First pixel characteristic figure and the first pixel characteristic figure of enhancing are added, the third pixel characteristic of fusion is exported
Figure.
Specifically, being equivalent to using the probability value in the second pixel characteristic figure after matrix dot product for the first pixel spy
Pixel value in sign figure has carried out the pixel value A in weight addition, such as the first pixel characteristic figure, in the second pixel characteristic figure
Probability value corresponding with pixel value A is 0.8, then the process of matrix dot product is A*0.8=0.8A, enhanced first pixel is special
The pixel value of pixel value A in sign figure has just become 0.8A from A.
Specifically, the first pixel characteristic figure of enhancing is added with the first pixel characteristic figure, after normalization may be implemented
Enhancing of the second pixel characteristic figure for the first pixel characteristic figure, by taking above-mentioned pixel value A as an example, in third pixel characteristic figure
The pixel value of pixel value A can become 1.8A from the A in the first pixel characteristic figure, be equivalent to can excavate profoundly it is each
Whether pixel value is prohibited items, improves the discrimination of prohibited items.
Embodiment as one preferred carries out the identification of prohibited items according to third pixel characteristic figure specifically:
By third pixel characteristic figure input area candidate network RPN network, obtain that there is preset number candidate region
4th pixel characteristic figure；
4th pixel characteristic figure is sequentially inputted to classifier and returns device, obtains the detection knot for picture to be identified
Fruit.
Specifically, RPN (Region proposal network, region candidate network) network can be played to third picture
Third pixel characteristic figure can be divided into the pixel characteristic of pre-set dimension (such as 7*7) by the effect that plain characteristic pattern is split
Figure, and classifier can carry out the identification of prohibited items, tool of the prohibited items in images to be recognized can be oriented by returning device
Body position coordinate, position coordinates can be multiple types, such as (x1, y1, x2, y2), wherein (x1, y1) can indicate each
Top left co-ordinate of the prohibited items in picture to be identified, (x2, y2) can indicate bottom right angular coordinate.
Specifically, third pixel characteristic figure is sequentially input RPN network, classifier and the process and existing skill that return device
It is similar in art, the recognition result for exporting prohibited items can be played the role of, details are not described herein for the embodiment of the present invention.
Wherein, the preset number can independently be set, such as can be set to 2000 etc., and the embodiment of the present invention is herein not
It limits.
Embodiment as one preferred, the second pixel characteristic figure after normalization specifically:
The pixel value in the second pixel characteristic figure is converted into the probability value between 0 to 1 by sigmoid function, is obtained
Normalization after the second pixel characteristic figure.
Specifically, the pixel value in the second pixel characteristic figure is converted to the probability between 0 to 1 by sigmoid function
Value has many advantages, such as that speed is fast and accuracy is high.
Certainly, it other than this method, can also be normalized using other kinds of method, the embodiment of the present invention
It is not limited here.
Embodiment as one preferred extracts the pixel characteristic of picture to be identified, before obtaining the first pixel characteristic figure,
The prohibited items detection method further include:
Picture to be identified is carried out to the overturning processing of preset kind corresponding with the quantity of i, the initial value of i is present count
Value；
After the identification for then carrying out prohibited items according to third pixel characteristic figure, the prohibited items detection method further include:
Judge whether i is equal to default value；
If it is not, then by the numerical value+1 of i and return step:
Picture to be identified is carried out to the overturning processing of preset kind corresponding with the quantity of i；
If so, the prohibited items recognition result of all pictures to be identified for different overturning types is carried out non-pole
Big value inhibits, and obtains final prohibited items testing result.
Specifically, in view of classifier is likely to different for the processing result of the picture to be identified of different flip angles,
Such as in certain pictures object than comparatively dense and overlapping it is more serious in the case where, under different flip angles, the position of target object
It sets coordinate to be different, affects the identification process of prohibited items, picture to be identified can be carried out in the embodiment of the present invention pre-
If the overturning of number is handled, and will be separated in overturning treated picture to be identified all carries out primary any of the above-described embodiment every time
The process for prohibiting article detection method is equivalent to have obtained and overturns the equal number of recognition result of number, then difference is turned over
The prohibited items recognition result for turning the picture to be identified of type carries out non-maxima suppression, using non-maxima suppression method by institute
Some recognition results are merged, be equivalent to determine for the same pixel highest target object of probability as a result,
Such as the pixel B in the picture to be identified for being located at different overturning types, the probability for being identified as different objects is different, obtains
Pixel B is identified as capacitor batteries by the wherein maximum recognition result of probability value.
Wherein, overturning type can be a variety of, such as can not overturn, flip horizontal, vertical overturning and counterclockwise
It is rotated by 90 °, the embodiment of the present invention is it is not limited here.
Wherein, default value can independently be set, such as be can be set to the initial value of i and added 3 (wherein, initial values
Can independently set, such as can be 0 etc.), it is equivalent to altogether there are four types of type is overturn, the embodiment of the present invention is it is not limited here.
Embodiment as one preferred, after obtaining final prohibited items testing result, the prohibited items detection side
Method further include:
Control prompting device prompts final prohibited items testing result.
It is made according to testing result rapidly not specifically, prohibited items testing result is tipped out and is conducive to staff
Same counter-measure, improves work efficiency, and the type of prompting device can be a variety of, such as can be text prompt or voice
Prompt etc., the embodiment of the present invention is it is not limited here.
Specifically, a kind of specific embodiment provided in an embodiment of the present invention can be with are as follows:
(1), on the tall and handsome computer up to NVIDIA image processor GTX1080Ti, Ubuntu1604 system is installed
System, and CUDA (Compute Unified Device Architecture, the unification for having configured the offer of NVIDIA official are installed
Calculate equipment framework) running environment.
(2), tensorflow deep learning framework platform is built, tensorflow is the symbol based on data flow programming
Number mathematic system, possesses multi-level structure, can be deployed in all kinds of servers, PC (personal computer, personal computer)
Terminal and webpage, and support GPU (Graphics Processing Unit, graphics processor) and TPU (Tensor
Processing Unit, tensor processing unit) high performance numerical computing, it is a outstanding mainstream deep learning frame, and
One of deep learning frame most popular at present.The step of building of tensorflow environment is referred to the installation of official and refers to
South.
(3), the packet image excessively of X-ray screening machine under default number (2000) practical application scene is collected as the detection side
The data set of method makes the object random distribution in every image as far as possible, meets multiple dimensioned requirement, that is, make prohibited items every
Being distributed in image is sparse, and what is had is intensive.
(4), it defines according to the actual situation and needs to detect the prohibited items classification identified in x-ray image, in the present invention
Class declaration share four classes, be controlled knife, lighter, power-supply battery, scissors respectively.It marks out violated in every image
These classifications and coordinate are converted into COCO (Common Objects with the storage of rectangle frame coordinate form by the position coordinates of article
In COntext) data set format, with json formatted file save.
(5), using tensorflow frame, realize that Faster rcnn object detector algorithm, such as the above method are implemented
The testing result of the picture to be identified of different overturning types is carried out non-maxima suppression by the step in example, obtains inspection to the end
Survey result.
(6), realization in the x-ray image collection provided and the data marked, and (5) in (3) and (4) is provided
Faster rcnn algorithm, is trained this model, and in order to accelerate training process, 2 GTX1080Ti are employed herein
Graphics processor parallel training, 2000 original x-ray images become 6000 x-ray images after pretreatment, time consumption for training about may be used
To be 7 hours.
(7), it will be directly used in the Faster rcnn model of prohibited items detection identification obtained in (5), be deployed in a set of
Among software.By video output signals HDMI (High Definition Multimedia Interface, the high definition of screening machine
Multimedia interface) interface connects on image pick-up card, by image pick-up card SDK (Software Development Kit, it is soft
Part development kit) the latest image picture that obtains screening machine, utilize the obtained Faster rcnn model of training in (6), detection
It whether there is prohibited items out from the x-ray image that screening machine intercepts.And the prohibited items type that will test and position
Coordinate is drawn out with rectangle frame, while the prediction probability of model being also plotted on picture.In the practical survey detection deployment of algorithm
In, class probability can select a threshold value, when being greater than 0.7, then it is assumed that there are prohibited items targets, in existing detection technique
In, a pile capacitor batteries target of the left upper of some picture to be identified can not be detected effectively, be treated as sometimes non-violated
Article and neglect.After method in through the embodiment of the present invention can enhance this Partial Feature, so that subsequent classifier
More easily, more accurately classified and positioned with device is returned, to effectively increase the detection of small size object prohibited items
Effect.
Referring to FIG. 2, Fig. 2 is a kind of structural schematic diagram of prohibited items detection device provided by the invention, comprising:
First extraction module 1 obtains the first pixel characteristic figure for extracting the pixel characteristic of picture to be identified；
Second extraction module 2, for extracting the picture of picture to be identified by the convolutional neural networks with multilayer convolution kernel
Plain feature obtains the second pixel characteristic figure；
Fusion Module 3 obtains third for merging the first pixel characteristic figure with the second pixel characteristic figure after normalization
Pixel characteristic figure；
Identification module 4, for carrying out the identification of prohibited items according to third pixel characteristic figure.
Prohibited items detection above-mentioned is please referred to for the introduction of prohibited items detection device provided in an embodiment of the present invention
The embodiment of method, details are not described herein for the embodiment of the present invention.
Referring to FIG. 3, Fig. 3 is a kind of structural schematic diagram of prohibited items detection device provided by the invention, comprising:
Memory 5, for storing computer program；
Processor 6, when for executing computer program realize as above any one of prohibited items detection method the step of.
Prohibited items detection above-mentioned is please referred to for the introduction of prohibited items detection device provided in an embodiment of the present invention
The embodiment of method, details are not described herein for the embodiment of the present invention.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For device disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is said referring to method part
It is bright.
It should also be noted that, in the present specification, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or equipment for including the element.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (10)
1. a kind of prohibited items detection method characterized by comprising
The pixel characteristic for extracting picture to be identified obtains the first pixel characteristic figure；
The pixel characteristic that the picture to be identified is extracted by the convolutional neural networks with multilayer convolution kernel, obtains the second pixel
Characteristic pattern；
The first pixel characteristic figure is merged with the second pixel characteristic figure after normalization, obtains third pixel characteristic
Figure；
The identification of prohibited items is carried out according to the third pixel characteristic figure.
2. prohibited items detection method according to claim 1, which is characterized in that described by with multilayer convolution kernel
Convolutional neural networks extract the pixel characteristic of the picture to be identified, obtain the second pixel characteristic figure specifically:
Multiple down-sampling operation is carried out to the picture to be identified by the convolutional neural networks with multilayer convolution kernel；
Up-sampling operation is carried out to the picture to be identified after down-sampling；
By the channel information of the to be identified picture of two 1*1 convolution kernel compressions after up-sampling, the second pixel is obtained
Characteristic pattern.
3. prohibited items detection method according to claim 2, which is characterized in that described pair after down-sampling described in
Picture to be identified carries out up-sampling operation specifically:
Up-sampling operation is carried out to the picture to be identified after down-sampling using bilinear interpolation.
4. prohibited items detection method according to claim 3, which is characterized in that described by the first pixel characteristic figure
It is merged with the second pixel characteristic figure after normalization, obtains third pixel characteristic figure specifically:
The first pixel characteristic figure and the second pixel characteristic figure after normalization are subjected to matrix dot product, obtained
The first pixel characteristic figure of enhancing；
The first pixel characteristic figure and the first pixel characteristic figure of enhancing are added, the third pixel for exporting fusion is special
Sign figure.
5. prohibited items detection method according to claim 1, which is characterized in that described according to the third pixel characteristic
Figure carries out the identification of prohibited items specifically:
By the third pixel characteristic figure input area candidate network RPN network, obtain that there is preset number candidate region
4th pixel characteristic figure；
The 4th pixel characteristic figure is sequentially inputted to classifier and returns device, obtains the inspection for the picture to be identified
Survey result.
6. prohibited items detection method according to claim 1, which is characterized in that second picture after the normalization
Plain characteristic pattern specifically:
The pixel value in the second pixel characteristic figure is converted into the probability value between 0 to 1 by sigmoid function, is obtained
Normalization after the second pixel characteristic figure.
7. prohibited items detection method according to any one of claims 1 to 6, which is characterized in that the extraction is to be identified
The pixel characteristic of picture, before obtaining the first pixel characteristic figure, the prohibited items detection method further include:
Picture to be identified is carried out to the overturning processing of preset kind corresponding with the quantity of i, the initial value of the i is present count
Value；
Then after the identification for carrying out prohibited items according to the third pixel characteristic figure, which is also wrapped
It includes:
Judge whether i is equal to default value；
If it is not, then by the numerical value+1 of i and return step:
Picture to be identified is carried out to the overturning processing of preset kind corresponding with the quantity of i；
If so, the prohibited items recognition result of all pictures to be identified for different overturning types is carried out non-pole
Big value inhibits, and obtains final prohibited items testing result.
8. prohibited items detection method according to claim 7, which is characterized in that described to obtain final prohibited items inspection
It surveys after result, the prohibited items detection method further include:
Control the final prohibited items testing result of prompting device prompt.
9. a kind of prohibited items detection device, comprising:
First extraction module obtains the first pixel characteristic figure for extracting the pixel characteristic of picture to be identified；
Second extraction module, for extracting the pixel of the picture to be identified by the convolutional neural networks with multilayer convolution kernel
Feature obtains the second pixel characteristic figure；
Fusion Module is obtained for merging the first pixel characteristic figure with the second pixel characteristic figure after normalization
Third pixel characteristic figure；
Identification module, for carrying out the identification of prohibited items according to the third pixel characteristic figure.
10. a kind of prohibited items detection device characterized by comprising
Memory, for storing computer program；
Processor realizes the prohibited items detection side as described in any one of claim 1 to 8 when for executing the computer program
The step of method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910603933.2A CN110321851A (en) | 2019-07-05 | 2019-07-05 | A kind of prohibited items detection method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910603933.2A CN110321851A (en) | 2019-07-05 | 2019-07-05 | A kind of prohibited items detection method, device and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110321851A true CN110321851A (en) | 2019-10-11 |
Family
ID=68122796
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910603933.2A Pending CN110321851A (en) | 2019-07-05 | 2019-07-05 | A kind of prohibited items detection method, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110321851A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860510A (en) * | 2020-07-29 | 2020-10-30 | 浙江大华技术股份有限公司 | X-ray image target detection method and device |
| CN112801070A (en) * | 2021-04-14 | 2021-05-14 | 浙江啄云智能科技有限公司 | Target detection method, device, equipment and storage medium |
| CN116449999A (en) * | 2023-06-16 | 2023-07-18 | 中国铁道科学研究院集团有限公司电子计算技术研究所 | Method, device, equipment and storage medium for registering package security check |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107341517A (en) * | 2017-07-07 | 2017-11-10 | 哈尔滨工业大学 | The multiple dimensioned wisp detection method of Fusion Features between a kind of level based on deep learning |
-
2019
- 2019-07-05 CN CN201910603933.2A patent/CN110321851A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107341517A (en) * | 2017-07-07 | 2017-11-10 | 哈尔滨工业大学 | The multiple dimensioned wisp detection method of Fusion Features between a kind of level based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| F. WANG ET AL.，: "Residual attention network for image classification", 《IN CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
| 李玉: "基于卷积神经网络的射线图像识别系统的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860510A (en) * | 2020-07-29 | 2020-10-30 | 浙江大华技术股份有限公司 | X-ray image target detection method and device |
| CN112801070A (en) * | 2021-04-14 | 2021-05-14 | 浙江啄云智能科技有限公司 | Target detection method, device, equipment and storage medium |
| CN112801070B (en) * | 2021-04-14 | 2021-09-21 | 浙江啄云智能科技有限公司 | Target detection method, device, equipment and storage medium |
| CN116449999A (en) * | 2023-06-16 | 2023-07-18 | 中国铁道科学研究院集团有限公司电子计算技术研究所 | Method, device, equipment and storage medium for registering package security check |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108121997A (en) | Use the object classification in the image data of machine learning model | |
| CN110060237B (en) | Fault detection method, device, equipment and system | |
| CN109101897A (en) | Object detection method, system and the relevant device of underwater robot | |
| CN111179217A (en) | Attention mechanism-based remote sensing image multi-scale target detection method | |
| CN109522966A (en) | A kind of object detection method based on intensive connection convolutional neural networks | |
| KR101930940B1 (en) | Apparatus and method for analyzing image | |
| CN110321851A (en) | A kind of prohibited items detection method, device and equipment | |
| CN108122239A (en) | Use the object detection in the image data of depth segmentation | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN110135424A (en) | Tilt text detection model training method and ticket image Method for text detection | |
| CN110378297A (en) | A kind of Remote Sensing Target detection method based on deep learning | |
| CN100561505C (en) | A kind of image detecting method and device | |
| CN110210482B (en) | Target detection method for improving class imbalance | |
| CN108596944A (en) | A kind of method, apparatus and terminal device of extraction moving target | |
| Visaniy et al. | Icdar 2013 music scores competition: Staff removal | |
| CN106127748A (en) | A kind of characteristics of image sample database and method for building up thereof | |
| CN112884033B (en) | Household garbage classification detection method based on convolutional neural network | |
| Luo et al. | Robot detection and localization based on deep learning | |
| CN105303163B (en) | A kind of method and detection device of target detection | |
| CN109903339A (en) | A kind of video group personage's position finding and detection method based on multidimensional fusion feature | |
| CN113807347A (en) | Kitchen waste impurity identification method based on target detection technology | |
| CN112206541A (en) | Game plug-in identification method and device, storage medium and computer equipment | |
| CN109635755A (en) | Face extraction method, apparatus and storage medium | |
| CN111368865A (en) | Method and device for detecting remote sensing image oil storage tank, readable storage medium and equipment | |
| CN109284752A (en) | A kind of rapid detection method of vehicle |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191011 |