CN110020647A - A kind of contraband object detection method, device and computer equipment - Google Patents
A kind of contraband object detection method, device and computer equipment Download PDFInfo
- Publication number
- CN110020647A CN110020647A CN201810020069.9A CN201810020069A CN110020647A CN 110020647 A CN110020647 A CN 110020647A CN 201810020069 A CN201810020069 A CN 201810020069A CN 110020647 A CN110020647 A CN 110020647A
- Authority
- CN
- China
- Prior art keywords
- image
- contraband
- determined
- target
- detected
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 152
- 230000002708 enhancing effect Effects 0.000 claims abstract description 121
- 238000000034 method Methods 0.000 claims abstract description 86
- 239000000463 material Substances 0.000 claims abstract description 82
- 238000012545 processing Methods 0.000 claims abstract description 39
- 238000000605 extraction Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 46
- 238000007689 inspection Methods 0.000 claims description 32
- 230000003321 amplification Effects 0.000 claims description 23
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 23
- 238000013135 deep learning Methods 0.000 claims description 21
- 239000000284 extract Substances 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 230000001965 increasing effect Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 8
- 230000009466 transformation Effects 0.000 description 6
- 238000010606 normalization Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 229920003023 plastic Polymers 0.000 description 2
- 239000004033 plastic Substances 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 208000003464 asthenopia Diseases 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 239000013077 target material Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013519 translation Methods 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10116—X-ray image
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
The embodiment of the invention provides a kind of contraband object detection method, device and computer equipment, method includes: to carry out enhancing processing to the specific characteristic in radioscopic image to be detected by a variety of pre-set image Enhancement Methods, obtains multiple enhancing images；Using goal-selling detection method, each enhancing image is detected, is judged with the presence or absence of contraband target to be determined in each enhancing image, and determine the location information in region locating for contraband target to be determined；If statistics there are the quantity of the enhancing image of contraband target to be determined to reach preset threshold, corresponding with the location information region of interest area image of extraction from radioscopic image to be detected；Obtain and according to the attributive character of region of interest area image, determine the material classification of contraband target to be determined；If material classification meets default material classification, it is determined that the contraband target to be determined detected is determining contraband target.The omission factor and false detection rate of contraband target detection can be reduced by this programme.
Description
Technical field
The present invention relates to safety check technical fields, more particularly to a kind of contraband object detection method, device and computer
Equipment.
Background technique
With the fast development of logistic industry, by express delivery, logistics carry secretly in the way of transport contraband the phenomenon that layer go out not
Thoroughly, the mode concealment of this entrainment transport is very strong, and the difficulty of investigation is larger, endangers social security.Tradition uses X-ray safety check
Contraband in machine testing luggage, package carries out article by can achieve not unpack to X-ray transmission luggage, package imaging
The purpose of inspection is widely applied in places such as airport, customs, railway station, subway stations.However, using X-ray security inspection machine
The method for detecting contraband needs staff to participate in image discriminating, and the degree of automation is low, and staff observes image for a long time
It is be easy to cause visual fatigue, the case where missing inspection, erroneous detection, especially logistic industry easily occurs, freight traffic volume is big, is pressed for time, and gives
Staff brings great pressure, also brings to supervision department difficult to regulate.
In recent years, with the fast development of artificial intelligence, target detection is carried out using the method for machine learning and is increasingly becoming
Mainstream.Theory based on machine learning can determine figure by deep learning operation for image formed by X-ray security inspection machine
The target for meeting preset condition with the similarity of contraband objective contour feature trained in advance as in is determining violated items
Mark, to reach the detection to contraband target.
But the defect as existing for X-ray scanning, the image scanned, which often exists, to be obscured, shows imperfect etc. ask
Topic, also, due to the complicated variety of contraband target, only contraband target is detected by contour feature, detection
As a result there is higher omission factor and false detection rate.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of contraband object detection method, device and computer equipment, with
Reduce the omission factor and false detection rate of contraband target detection.Specific technical solution is as follows:
In a first aspect, the embodiment of the invention provides a kind of contraband object detection methods, which comprises
By a variety of pre-set image Enhancement Methods, the specific characteristic in the radioscopic image to be detected of acquisition is enhanced
Processing, obtains multiple enhancing images；
Using goal-selling detection method, each enhancing image is detected, judge in each enhancing image with the presence or absence of to
It determines contraband target, and determines the essential attribute information of the contraband target to be determined, wherein the essential attribute information
Location information including region locating for the contraband target to be determined；
The quantity for counting the enhancing image there are the contraband target to be determined, if the quantity reaches preset threshold,
Region of interest area image corresponding with the location information is then extracted from the radioscopic image to be detected；
Obtain and according to the attributive character of the region of interest area image, determine the substance of the contraband target to be determined
Classification；
If the material classification meets default material classification, it is determined that the contraband target to be determined detected is true
Fixed contraband target.
Optionally, the acquisition modes of the radioscopic image to be detected, comprising:
Obtain the source radioscopic image acquired by X-ray security inspection machine；
Operation is normalized to the source radioscopic image, obtains radioscopic image to be detected.
Optionally, goal-selling detection method is utilized described, each enhancing image is detected, judges each enhancing image
In whether there is contraband target to be determined, it is described and before determining the essential attribute information of the contraband target to be determined
Method further include:
Obtain multiple image patterns comprising any contraband target；
According to the coordinate information of contraband target area in each image pattern of calibration, extract in each contraband target area
Image, obtain multiple sample exposure masks；
Default amplification processing is carried out to each sample exposure mask, the mask image after obtaining multiple amplifications；
Each mask image is projected into multiple pre-set images, multiple training samples are obtained；
The multiple training sample is trained using predetermined deep learning method, obtains detection model；
It is described to utilize goal-selling detection method, each enhancing image is detected, judges whether deposit in each enhancing image
In contraband target to be determined, and determine the essential attribute information of the contraband target to be determined, comprising:
Using the detection model, each enhancing image is detected, is judged in each enhancing image with the presence or absence of contraband
Target, and determine the essential attribute information of the contraband target to be determined.
Optionally, if meeting default material classification in the material classification, it is determined that is detected is described to be determined
After contraband target is determining contraband target, the method also includes:
By the sample in given scenario sample set and include the determination contraband target image setting for candidate
Sample includes the sample for obeying target distribution under given scenario in the given scenario sample set；
According to the correlation of each candidate samples, weight is distributed for each candidate samples；
Maximum first candidate samples of weight in each candidate samples are selected, and selection and institute from the multiple training sample
The similarity for stating the first candidate samples is greater than the second candidate samples of default similarity；
Updating the sample in the given scenario sample set is first candidate samples and second candidate samples；
The sample in updated given scenario sample set is trained using the predetermined deep learning method, is obtained
The detection model of update, it is right when obtaining radioscopic image to be detected next time, to execute the detection model using the update
Each enhancing image is detected, and is judged with the presence or absence of contraband target to be determined in each enhancing image, and determination is described to be determined
The essential attribute information of contraband target.
Optionally, it is described acquisition and according to the attributive character of the region of interest area image, determine described to be determined violated
Items target material classification, comprising:
Obtain the gray value of the region of interest area image；
Determine default tonal range locating for the gray value；
Based on the default tonal range, the material classification of the contraband target to be determined is determined.
Optionally, the essential attribute information further includes the type information of the contraband target to be determined；
If meeting default material classification in the material classification, it is determined that the violated items to be determined detected
It is designated as after determining contraband target, the method also includes:
According to the type information, the type of the contraband target of the determination is obtained.
Second aspect, the embodiment of the invention provides a kind of contraband object detecting device, described device includes:
Enhance processing module, is used for by a variety of pre-set image Enhancement Methods, in the radioscopic image to be detected of acquisition
Specific characteristic carries out enhancing processing, obtains multiple enhancing images；
Detection module detects each enhancing image, judges each enhancing image for utilizing goal-selling detection method
In whether there is contraband target to be determined, and determine the essential attribute information of the contraband target to be determined, wherein described
Essential attribute information includes the location information in region locating for the contraband target to be determined；
Statistical module, for counting the quantity of the enhancing image there are the contraband target to be determined, if the quantity
Reach preset threshold, then extracts region of interest area image corresponding with the location information from the radioscopic image to be detected；
Module is obtained, for obtaining and according to the attributive character of the region of interest area image, is determined described to be determined separated
The material classification of contraband goods target；
Determining module, if meeting default material classification for the material classification, it is determined that is detected is described to be determined
Contraband target is determining contraband target.
Optionally, the acquisition module, is also used to:
Obtain the source radioscopic image acquired by X-ray security inspection machine；
Operation is normalized to the source radioscopic image, obtains radioscopic image to be detected.
Optionally, described device further include:
Image pattern obtains module, for obtaining multiple image patterns comprising any contraband target；
Extraction module extracts each disobey for the coordinate information of contraband target area in each image pattern according to calibration
Image in contraband goods target area obtains multiple sample exposure masks；
Processing module is expanded, for carrying out default amplification processing to each sample exposure mask, the exposure mask figure after obtaining multiple amplifications
Picture；
Projection module obtains multiple training samples for projecting each mask image into multiple pre-set images；
First training module is obtained for being trained using predetermined deep learning method to the multiple training sample
Detection model；
The detection module, is specifically used for:
Using the detection model, each enhancing image is detected, is judged in each enhancing image with the presence or absence of contraband
Target, and determine the essential attribute information of the contraband target to be determined.
Optionally, described device further include:
Setting module, for by the sample in given scenario sample set and include the determination contraband target figure
It include the sample for obeying target distribution under given scenario in the given scenario sample set as being set as candidate samples；
Distribution module distributes weight for the correlation according to each candidate samples for each candidate samples；
Selecting module, for selecting maximum first candidate samples of weight in each candidate samples, and from the multiple training
The second candidate samples for being greater than default similarity with the similarity of first candidate samples are selected in sample；
Update module is first candidate samples and described for updating the sample in the given scenario sample set
Two candidate samples；
Second training module, for utilizing the predetermined deep learning method in updated given scenario sample set
Sample is trained, the detection model updated, to execute described in utilizing when obtaining radioscopic image to be detected next time
The detection model of update detects each enhancing image, judges to whether there is contraband target to be determined in each enhancing image,
And determine the essential attribute information of the contraband target to be determined.
Optionally, the acquisition module, is specifically used for:
Obtain the gray value of the region of interest area image；
Determine default tonal range locating for the gray value；
Based on the default tonal range, the material classification of the contraband target to be determined is determined.
Optionally, the essential attribute information further includes the type information of the contraband target to be determined；
Described device further include:
Determination type module, for obtaining the type of the contraband target of the determination according to the type information.
The third aspect, the embodiment of the invention provides a kind of computer equipments, including processor and memory, wherein
The memory, for storing computer program；
The processor when for executing the program stored on the memory, realizes first party of the embodiment of the present invention
Method and step described in face.
A kind of contraband object detection method, device and computer equipment provided in an embodiment of the present invention, by a variety of pre-
If image enchancing method, the specific characteristic in the radioscopic image to be detected of acquisition is enhanced, obtains multiple enhancing images,
Using goal-selling detection method, each enhancing image is detected, is judged in each enhancing image with the presence or absence of to be determined violated
Items mark, and determine the location information in region locating for contraband target to be determined, statistics there are contraband targets to be determined
The quantity of enhancing image when reaching preset threshold, extracted from radioscopic image to be detected with locating for contraband target to be determined
The corresponding region of interest area image of the location information in region, since there may be non-contraband mesh in contraband target to be determined
Mark, then according to the attributive character of region of interest area image, can determine the material classification of contraband target to be determined, according to substance
Classification can determine whether contraband target to be determined is determining contraband target.It may be deposited for radioscopic image to be detected
Obscuring, showing the problems such as imperfect, by pre-set image Enhancement Method, to the specific characteristic in radioscopic image to be detected into
Row enhancing, in this way, the content in radioscopic image to be detected can clear as much as possible, be completely shown in obtained enhancing image,
It avoids due to obscuring, showing the problems such as imperfect caused missing inspection, reduces the omission factor of contraband target detection；By more
Kind of pre-set image Enhancement Method enhances radioscopic image to be detected, then counts that there are the enhancings of contraband target to be determined
The quantity of image is made whether as the judgement of contraband target when the quantity of statistics reaches threshold value, in this way, even if occurring individual
Enhance image erroneous detection, due to being provided with threshold value, can not still be made whether as the judgement of contraband target, to reduce erroneous detection
Rate；Also, for there may be non-contraband targets identical with the contour feature of contraband target, due to the material class of the two
Not there is difference, then determine the material classification of contraband target to be determined by attributive character, if it is separated that material classification, which meets,
The pre-set categories of contraband goods material classification, it is determined that be contraband target, reduce the false detection rate of contraband target detection.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is the flow diagram of the contraband object detection method of the embodiment of the present invention；
Fig. 2 a is the source radioscopic image that the X-ray security inspection machine of the embodiment of the present invention scans；
Fig. 2 b is the embodiment of the present invention to the enhancing image obtained after source radioscopic image progress logarithmic transformation；
Fig. 2 c is the increasing to obtaining after the progress adaptive histogram equalization processing of source radioscopic image of the embodiment of the present invention
Strong image；
Fig. 2 d is the embodiment of the present invention to the enhancing image obtained after source radioscopic image progress gamma transformation；
Violated items in the source radioscopic image that Fig. 3 a scans for the different X-ray security inspection machines of the embodiment of the present invention
Logo image；
Fig. 3 b is the embodiment of the present invention by contraband target figure in obtained radioscopic image to be detected after normalization
Picture；
Fig. 4 a is the image pattern schematic diagram of the embodiment of the present invention；
Fig. 4 b is the sample exposure mask schematic diagram of the embodiment of the present invention；
Fig. 4 c is the mask image after the amplification of the embodiment of the present invention；
Fig. 4 d is the pre-set image of the embodiment of the present invention；
Fig. 4 e is the training sample schematic diagram of the embodiment of the present invention；
Fig. 5 is the structural schematic diagram of the contraband object detecting device of one embodiment of the invention；
Fig. 6 is the structural schematic diagram of the contraband object detecting device of another embodiment of the present invention；
Fig. 7 is the structural schematic diagram of the contraband object detecting device of further embodiment of this invention；
Fig. 8 is the structural schematic diagram of the computer equipment of the embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
In order to reduce the omission factor and false detection rate of contraband target detection, the embodiment of the invention provides a kind of violated items
Mark detection method, device and computer equipment.
It is introduced in the following, being provided for the embodiments of the invention a kind of contraband object detection method first.
A kind of executing subject of contraband object detection method provided by the embodiment of the present invention can be calculated to execute intelligence
The computer equipment of method, the computer equipment can integrate in X-ray security inspection machine, can also be independently of X-ray security inspection machine, also
It can be remote server, include at least the processor equipped with kernel processor chip in executing subject.Realize that the present invention is implemented
A kind of mode of contraband object detection method provided by example can be the software being set in executing subject, hardware circuit and
At least one of logic circuit mode.
As shown in Figure 1, a kind of contraband object detection method provided by the embodiment of the present invention, may include walking as follows
It is rapid:
S101 carries out the specific characteristic in the radioscopic image to be detected of acquisition by a variety of pre-set image Enhancement Methods
Enhancing processing, obtains multiple enhancing images.
Pre-set image Enhancement Method is the method that enhancing processing is carried out for the specific characteristic in radioscopic image to be detected,
Wherein, enhancing processing is exactly the specific characteristic purposefully emphasized in radioscopic image to be detected, expands radioscopic image to be detected
Difference between middle different target feature.The pre-set image Enhancement Method includes but are not limited to: median filtering, mean filter,
High-pass filtering, gray scale normalization, gray scale logarithmic transformation, square root transformation, adaptive histogram equalization etc..X-ray to be detected
Specific characteristic in image can be understood as interested feature, and specific characteristic can be the whole special of radioscopic image to be detected
Sign, or the local feature of radioscopic image to be detected.By a variety of pre-set image Enhancement Methods, to X ray picture to be detected
After specific characteristic as in carries out a series of parallel enhancing processing, the enhancing image of available multiple and different effects.For example,
Logarithmic transformation is carried out to source radioscopic image as shown in Figure 2 a, obtains enhancing image as shown in Figure 2 b, it is bright in the enhancing image
It spends directly proportional to the thickness of object；In another example being carried out at adaptive histogram equalization to source radioscopic image as shown in Figure 2 a
Reason, obtains enhancing image as shown in Figure 2 c, brightness is directly proportional to the quality of object in the enhancing image；For another example to such as scheming
Source radioscopic image shown in 2a carries out gamma transformation, obtains enhancing image as shown in Figure 2 d, the letter of the enhancing objects in images
It makes an uproar than the square root for being proportional to image pixel intensities.It can be seen that by these examples through pre-set image Enhancement Method, can be enhanced
Specified target in radioscopic image to be detected.
Radioscopic image to be detected can be the source radioscopic image directly scanned by X-ray security inspection machine, still, by
It is had differences between the X-ray security inspection machine of various models, obtained image grayscale is different, in order to adapt to different model
X-ray security inspection machine, the image grayscale approach for scanning each X-ray security inspection machine is consistent, increases algorithm for natural trend
Robustness needs the image scanned to all kinds of X-ray security inspection machines that operation is normalized, then carries out image analysis, normalizing
Changing operation may include doses change, perspective, material composition and object towards variation etc..
Therefore, optionally, the acquisition modes of radioscopic image to be detected may include:
Obtain the source radioscopic image acquired by X-ray security inspection machine；
Operation is normalized to source radioscopic image, obtains radioscopic image to be detected.
Contraband target image such as Fig. 3 a in the source radioscopic image that different X-ray security inspection machines scans, it can be seen that
The gray scale for the source radioscopic image that each X-ray security inspection machine scans is different, passes through the X ray picture to be detected obtained after normalization
Contraband target image is as shown in Figure 3b as in, and the source radioscopic image that each X-ray security inspection machine scans obtains after normalization
The radioscopic image to be detected arrived has consistent gray scale.
S102 detects each enhancing image using goal-selling detection method, judges whether deposit in each enhancing image
In contraband target to be determined, and determine the essential attribute information of contraband target to be determined.
Wherein, essential attribute information includes the location information in region locating for contraband target to be determined.Obtaining multiple increasings
After strong image, target detection can be carried out to each enhancing image, judged in each enhancing image with the presence or absence of violated items to be determined
Mark, goal-selling detection method can be the object detection method based on deep learning, such as convolutional neural networks algorithm can also
Think the classifier classification method that aspect ratio pair is carried out with contraband target template.For instantly popular deep learning method,
It before carrying out target detection, needs first to train to obtain a detection model, carries out target inspection using the detection model that training obtains
It surveys.Contraband target to be determined can be understood as doubtful contraband target, such as with having with contrabands such as controlled knife, pistols
There is the target of the features such as similar color, shape, wherein may include cap gun, plastic knife etc..
Therefore, optionally, goal-selling detection method is being utilized, each enhancing image is being detected, judging each enhancing figure
It whether there is contraband target to be determined as in, and before the step of determining the essential attribute information of contraband target to be determined,
It can also include the following steps:
The first step obtains multiple image patterns comprising any contraband target；
Second step extracts each violated items according to the coordinate information of contraband target area in each image pattern of calibration
The image in region is marked, multiple sample exposure masks are obtained；
Third step carries out default amplification processing to each sample exposure mask, the mask image after obtaining multiple amplifications；
4th step projects each mask image into multiple pre-set images, obtains multiple training samples；
5th step is trained multiple training samples using predetermined deep learning method, obtains detection model.
Image pattern can be the multiple images comprising any contraband target acquired under experimental situation, such as Fig. 4 a institute
Show, since deep learning training needs a large amount of sample, and include any contraband target amount of images it is seldom, procurement cost
Height, therefore, the multiple images sample got are only the sample that part includes contraband target, then need to expand sample
Increase, image (the i.e. violated items in contraband target area should be extracted according to the coordinate information of the contraband target area of calibration
Logo image), as shown in Figure 4 b, using the image in the contraband target area of extraction as sample exposure mask, in order to guarantee sample
Diversity, needs to carry out sample exposure mask default amplification processing, and the mask image after amplification as illustrated in fig. 4 c, presets amplification processing
May include but be not limited only to: random cropping, rotation, translation, random overturning, distortion, contrast variation, brightness adjustment, add with
Machine noise, the superposition of multiple exposure masks etc..If the case where size and diversity are restricted, are easy to appear over-fitting, then
Over-fitting, boosting algorithm performance can be reduced by default amplification processing.
Mask image after amplification is projected into pre-set image as shown in figure 4d, training sample as shown in fig 4e is obtained
This, being trained using predetermined deep learning method to training sample can be obtained detection model.Wherein, predetermined depth study side
Method can be each neural network computing method of such as convolutional neural networks, and trained process can use transfer learning technology.
Since contraband target is usually very small in image pattern, even if having passed through amplification processing, it is desirable to which training obtains one accurately
Detection model is still extremely difficult, because of the training sample that amplification treated mask image is obtained by projection, although quantity
It greatly increases, but the diversity of training sample is insufficient, direct the problem of training neural network to be easy to produce locally optimal solution.Cause
This, can use transfer learning technology, first with a large amount of visible images data (such as 100,000 visible images data,
Million visible images data etc.) carry out neural network training, obtain a training pattern；Then, visible images are used
The training pattern initialization detection model that data training obtains, then the training sample projected after amplification processing is inputted, it carries out
The fine tuning of detection model obtains final detection model by constantly finely tuning.
Then by the way that each enhancing image is inputted detection model respectively, export in result may include in each enhancing image whether
There are contraband target and the essential attribute information of contraband target.That is, using goal-selling detection method, to each enhancing image
It is detected, is judged with the presence or absence of contraband target to be determined in each enhancing image, and determine the base of contraband target to be determined
The step of this attribute information may include:
Using detection model, each enhancing image is detected, judges to whether there is contraband target in each enhancing image,
And determine the essential attribute information of contraband target to be determined.
It, not only can be with when carrying out the detection of contraband target to be determined using the object detection method based on deep learning
The location information in region locating for contraband target to be determined is obtained, the type information of contraband target to be determined can also be obtained,
I.e. essential attribute information further includes the type information for having contraband target to be determined.The type message reflection contraband to be determined
The doubtful type of target, for example, which contraband target to be determined may can for controlled knife, which contraband target to be determined
It can be pistol etc..
S103 counts the quantity of the enhancing image there are contraband target to be determined, if the quantity reaches preset threshold,
Region of interest area image corresponding with location information is extracted from radioscopic image to be detected.
In the method for carrying out target classification, the method for the area-of-interest plus depth study being often used, this method needs
It first proposes area-of-interest, the realization of the dividing methods such as color, edge can be used, then the region being partitioned into is classified.This
If the content shown in image is fuzzy, imperfect in kind method, easily there is the case where erroneous detection, it is therefore, in the present embodiment, right
There are the quantity of the enhancing image of contraband target to be determined to be counted, can if the quantity of statistics reaches preset threshold
To extract region of interest area image corresponding with location information from radioscopic image to be detected, that is, the operation of region recommendation is carried out,
The region that statistical magnitude reaches preset threshold is recommended as area-of-interest, false detection rate can be reduced.Wherein, preset threshold can be with
It is set according to actual conditions, for example, preset threshold is set as to enhance the half of the sum of image, that is, a small number of obediences are more
Number, if it exceeds detect contraband target to be determined in the enhancing image of half, then extracted from radioscopic image to be detected with
The corresponding region of interest area image of location information；In another example preset threshold is set as 1, it is equivalent to veto by one vote system, as long as having
Contraband target to be determined is detected in one enhancing image, then is extracted from radioscopic image to be detected corresponding with location information
Region of interest area image.
S104, obtain and according to the attributive character of region of interest area image, determine the material class of contraband target to be determined
Not.
Due in contraband target to be determined there may be non-contraband target, for example, cap gun, plastics knife etc.,
According to the physical property of X-ray, the substance of different atomic numbers be to the damping capacity of X-ray it is different, one true pistol
Showing completely different, different material on the x-ray image with a wooden cap gun has different attributive character, because
This needs to obtain the attributive character of region of interest area image in order to filter out non-contraband target, according to the attributive character
Determine the material classification of contraband target to be determined, wherein the object of unlike material on the x-ray image tonal range have it is bright
Aobvious difference.
Therefore, optionally, obtain and according to the attributive character of region of interest area image, determine contraband target to be determined
The step of material classification may include:
Obtain the gray value of region of interest area image；
Determine default tonal range locating for gray value；
Tonal range is preset based on this, determines the material classification of contraband target to be determined.
Wherein, the corresponding tonal range of object that tonal range is unlike material is preset, if region of interest area image
Gray value is exactly in the corresponding default tonal range of pistol, it is determined that the material classification of contraband target to be determined is hand
Rifle.
S105, if material classification meets default material classification, it is determined that the contraband target to be determined detected is to determine
Contraband target.
Wherein, default material classification can be the corresponding classification of pre-set contraband, such as metal, stone implement, ammunition
Etc. classifications.If material classification does not meet default material classification, it is determined that contraband target to be determined is non-contraband target.
For using deep learning method to carry out target detection, since detection model trained in advance is in concrete scene application
When, the detection data of actual scene and the sample of training have bigger difference, then the performance of detection model is poor.If it is intended to training
One high performance detection model, needs to establish a sufficiently large sample set, this sample set needs to cover various scales, view
Angle, image resolution ratio, X-ray intensity sample, still, due to various X-ray security inspection machines use environment variation very greatly, can not
It realizes and obtains so many sample；Alternatively, be one dedicated detection model of given scenario of training there are also a kind of settling mode,
The detection model can provide performance more higher than general detection model, and still, being trained for each given scenario is
Arduous and time-consuming work.In order to cope with the above problem, it may be considered that from general detection model specialization one for specified field
The detection model of scape.
Therefore, optionally, the step of the contraband target to be determined confirmly detected is determining contraband target it
Afterwards, contraband object detection method can also include the following steps:
The first step, by the sample in given scenario sample set and include determining contraband target image setting for wait
Sampling sheet, wherein include the sample for obeying target distribution under given scenario in given scenario sample set；
Second step distributes weight according to the correlation of each candidate samples for each candidate samples；
Third step, selects maximum first candidate samples of weight in each candidate samples, and from the detection model pair trained
Answer the second candidate samples for selecting to be greater than default similarity with the similarity of the first candidate samples in multiple training samples；
4th step, updating the sample in given scenario sample set is the first candidate samples and the second candidate samples；
5th step is trained the sample in updated given scenario sample set using predetermined deep learning method,
The detection model updated, it is right when obtaining radioscopic image to be detected next time, to execute using the detection model updated
Each enhancing image is detected, and is judged with the presence or absence of contraband target to be determined in each enhancing image, and determination is to be determined violated
Items target essential attribute information.
After determining contraband target, according to the Sample Refreshment in actual contraband target and given scenario sample set
Training detection model passes through constantly iteration, the property of detection model for carrying out the detection and identification of contraband target next time
Can be continuously improved, even if also, actual scene change, still can adjust detection model in time, guarantee detection performance.
Optionally, essential attribute information further includes the type information of contraband target to be determined；
If then meeting default material classification in material classification, it is determined that the contraband target to be determined detected is determining
After the step of contraband target, can also include:
According to type information, the type of determining contraband target is obtained.
It, not only can be with when carrying out the detection of contraband target to be determined using the object detection method based on deep learning
The location information in region locating for contraband target to be determined is obtained, the type information of contraband target to be determined can also be obtained,
I.e. essential attribute information further includes the type information for having contraband target to be determined.Then confirmly detecting determining violated items
After mark, the type of the contraband target can also be obtained according to type information, can directly export the class of the contraband target
Type informs that supervisor detects that type is the contraband target of controlled knife, still detects that type is the violated items of pistol
Mark, supervisor's monitoring of being more convenient in this way.
Using the present embodiment, for radioscopic image to be detected there may be obscuring, showing the problems such as imperfect, by pre-
If image enchancing method, the specific characteristic in radioscopic image to be detected is enhanced, in this way, energy in obtained enhancing image
It is enough it is clear as much as possible, completely show content in radioscopic image to be detected, avoid due to obscuring, showing imperfect etc. ask
Missing inspection caused by inscribing, reduces the omission factor of contraband target detection；X to be detected is penetrated by a variety of pre-set image Enhancement Methods
Line image is enhanced, then counts the quantity of the enhancing image there are contraband target to be determined, reaches threshold in the quantity of statistics
It is made whether when value as the judgement of contraband target, in this way, even if there are individual enhancing image erroneous detections, due to being provided with threshold value,
It can not still be made whether as the judgement of contraband target, to reduce false detection rate；Also, for there may be with contraband
The identical non-contraband target of the contour feature of target, it is since there are difference for the material classification of the two, then true by attributive character
The material classification of fixed contraband target to be determined, if material classification meets for the pre-set categories of contraband material classification, really
It is set to contraband target, reduces the false detection rate of contraband target detection.
Corresponding to above method embodiment, the embodiment of the invention provides a kind of contraband object detecting devices, such as Fig. 5 institute
Show, which may include:
Enhance processing module 510, is used for by a variety of pre-set image Enhancement Methods, to the radioscopic image to be detected of acquisition
In specific characteristic carry out enhancing processing, obtain multiple enhancing images；
Detection module 520 detects each enhancing image, judges each enhancing for utilizing goal-selling detection method
It whether there is contraband target to be determined in image, and determine the essential attribute information of the contraband target to be determined, wherein
The essential attribute information includes the location information in region locating for the contraband target to be determined；
Statistical module 530, for counting the quantity of the enhancing image there are the contraband target to be determined, if the number
Amount reaches preset threshold, then area-of-interest figure corresponding with the location information is extracted from the radioscopic image to be detected
Picture；
Module 540 is obtained to determine described to be determined for obtaining and according to the attributive character of the region of interest area image
The material classification of contraband target；
Determining module 550, if meeting default material classification for the material classification, it is determined that is detected is described to true
Determining contraband target is determining contraband target.
Optionally, the acquisition module 540, can be also used for:
Obtain the source radioscopic image acquired by X-ray security inspection machine；
Operation is normalized to the source radioscopic image, obtains radioscopic image to be detected.
Optionally, the acquisition module 540, specifically can be used for:
Obtain the gray value of the region of interest area image；
Determine default tonal range locating for the gray value；
Based on the default tonal range, the material classification of the contraband target to be determined is determined.
Optionally, the essential attribute information further includes the type information of the contraband target to be determined；
Described device can also include:
Determination type module, for obtaining the type of the contraband target of the determination according to the type information.
Using the present embodiment, for radioscopic image to be detected there may be obscuring, showing the problems such as imperfect, by pre-
If image enchancing method, the specific characteristic in radioscopic image to be detected is enhanced, in this way, energy in obtained enhancing image
It is enough it is clear as much as possible, completely show content in radioscopic image to be detected, avoid due to obscuring, showing imperfect etc. ask
Missing inspection caused by inscribing, reduces the omission factor of contraband target detection；X to be detected is penetrated by a variety of pre-set image Enhancement Methods
Line image is enhanced, then counts the quantity of the enhancing image there are contraband target to be determined, reaches threshold in the quantity of statistics
It is made whether when value as the judgement of contraband target, in this way, even if there are individual enhancing image erroneous detections, due to being provided with threshold value,
It can not still be made whether as the judgement of contraband target, to reduce false detection rate；Also, for there may be with contraband
The identical non-contraband target of the contour feature of target, it is since there are difference for the material classification of the two, then true by attributive character
The material classification of fixed contraband target to be determined, if material classification meets for the pre-set categories of contraband material classification, really
It is set to contraband target, reduces the false detection rate of contraband target detection.
Based on embodiment illustrated in fig. 5, the embodiment of the invention also provides a kind of contraband object detecting devices, such as Fig. 6 institute
Show, which may include:
Enhance processing module 610, is used for by a variety of pre-set image Enhancement Methods, to the radioscopic image to be detected of acquisition
In specific characteristic carry out enhancing processing, obtain multiple enhancing images；
Image pattern obtains module 620, for obtaining multiple image patterns comprising any contraband target；
Extraction module 630 extracts each for the coordinate information of contraband target area in each image pattern according to calibration
Image in contraband target area obtains multiple sample exposure masks；
Processing module 640 is expanded, for carrying out default amplification processing to each sample exposure mask, the exposure mask after obtaining multiple amplifications
Image；
Projection module 650 obtains multiple training samples for projecting each mask image into multiple pre-set images；
First training module 660 is obtained for being trained using predetermined deep learning method to the multiple training sample
To detection model；
Detection module 670 detects each enhancing image, judges each enhancing image for utilizing the detection model
In whether there is contraband target, and determine the essential attribute information of the contraband target to be determined, wherein the basic category
Property information includes the location information in region locating for the contraband target to be determined；
Statistical module 680, for counting the quantity of the enhancing image there are the contraband target to be determined, if the number
Amount reaches preset threshold, then area-of-interest figure corresponding with the location information is extracted from the radioscopic image to be detected
Picture；
Module 690 is obtained to determine described to be determined for obtaining and according to the attributive character of the region of interest area image
The material classification of contraband target；
Determining module 6100, if meeting default material classification for the material classification, it is determined that detect it is described to
Determine that contraband target is determining contraband target.
Optionally, the essential attribute information further includes the type information of the contraband target to be determined；
Described device can also include:
Determination type module, for obtaining the type of the contraband target of the determination according to the type information.
Using the present embodiment, for radioscopic image to be detected there may be obscuring, showing the problems such as imperfect, by pre-
If image enchancing method, the specific characteristic in radioscopic image to be detected is enhanced, in this way, energy in obtained enhancing image
It is enough it is clear as much as possible, completely show content in radioscopic image to be detected, avoid due to obscuring, showing imperfect etc. ask
Missing inspection caused by inscribing, reduces the omission factor of contraband target detection；X to be detected is penetrated by a variety of pre-set image Enhancement Methods
Line image is enhanced, then counts the quantity of the enhancing image there are contraband target to be determined, reaches threshold in the quantity of statistics
It is made whether when value as the judgement of contraband target, in this way, even if there are individual enhancing image erroneous detections, due to being provided with threshold value,
It can not still be made whether as the judgement of contraband target, to reduce false detection rate；Also, for there may be with contraband
The identical non-contraband target of the contour feature of target, it is since there are difference for the material classification of the two, then true by attributive character
The material classification of fixed contraband target to be determined, if material classification meets for the pre-set categories of contraband material classification, really
It is set to contraband target, reduces the false detection rate of contraband target detection.Also, by being expanded, being projected to image pattern
Processing, limited image pattern is expanded and is trained for the biggish sample of sample size, it is possible to reduce over-fitting promotes detection property
Energy.
Based on embodiment illustrated in fig. 6, the embodiment of the invention also provides a kind of contraband object detecting devices, such as Fig. 7 institute
Show, which may include:
Enhance processing module 710, is used for by a variety of pre-set image Enhancement Methods, to the radioscopic image to be detected of acquisition
In specific characteristic carry out enhancing processing, obtain multiple enhancing images；
Image pattern obtains module 720, for obtaining multiple image patterns comprising any contraband target；
Extraction module 730 extracts each for the coordinate information of contraband target area in each image pattern according to calibration
Image in contraband target area obtains multiple sample exposure masks；
Processing module 740 is expanded, for carrying out default amplification processing to each sample exposure mask, the exposure mask after obtaining multiple amplifications
Image；
Projection module 750 obtains multiple training samples for projecting each mask image into multiple pre-set images；
First training module 760 is obtained for being trained using predetermined deep learning method to the multiple training sample
To detection model；
Detection module 770 detects each enhancing image, judges each enhancing image for utilizing the detection model
In whether there is contraband target, and determine the essential attribute information of the contraband target to be determined, wherein the basic category
Property information includes the location information in region locating for the contraband target to be determined；
Statistical module 780, for counting the quantity of the enhancing image there are the contraband target to be determined, if the number
Amount reaches preset threshold, then area-of-interest figure corresponding with the location information is extracted from the radioscopic image to be detected
Picture；
Module 790 is obtained to determine described to be determined for obtaining and according to the attributive character of the region of interest area image
The material classification of contraband target；
Determining module 7100, if meeting default material classification for the material classification, it is determined that detect it is described to
Determine that contraband target is determining contraband target；
Setting module 7110, for by the sample in given scenario sample set and include the determination contraband target
Image setting be candidate samples, include the sample for obeying target distribution under given scenario in the given scenario sample set；
Distribution module 7120 distributes weight for the correlation according to each candidate samples for each candidate samples；
Selecting module 7130, for selecting maximum first candidate samples of weight in each candidate samples, and from the multiple
The second candidate samples for being greater than default similarity with the similarity of first candidate samples are selected in training sample；
Update module 7140, for updating the sample in the given scenario sample set for first candidate samples and institute
State the second candidate samples；
Second training module 7150, for utilizing the predetermined deep learning method to updated given scenario sample set
In sample be trained, the detection model updated, to execute utilization when obtaining radioscopic image to be detected next time
The detection model of the update detects each enhancing image, judges in each enhancing image with the presence or absence of contraband to be determined
Target, and determine the essential attribute information of the contraband target to be determined.
Optionally, the essential attribute information further includes the type information of the contraband target to be determined；
Described device can also include:
Determination type module, for obtaining the type of the contraband target of the determination according to the type information.
Using the present embodiment, for radioscopic image to be detected there may be obscuring, showing the problems such as imperfect, by pre-
If image enchancing method, the specific characteristic in radioscopic image to be detected is enhanced, in this way, energy in obtained enhancing image
It is enough it is clear as much as possible, completely show content in radioscopic image to be detected, avoid due to obscuring, showing imperfect etc. ask
Missing inspection caused by inscribing, reduces the omission factor of contraband target detection；X to be detected is penetrated by a variety of pre-set image Enhancement Methods
Line image is enhanced, then counts the quantity of the enhancing image there are contraband target to be determined, reaches threshold in the quantity of statistics
It is made whether when value as the judgement of contraband target, in this way, even if there are individual enhancing image erroneous detections, due to being provided with threshold value,
It can not still be made whether as the judgement of contraband target, to reduce false detection rate；Also, for there may be with contraband
The identical non-contraband target of the contour feature of target, it is since there are difference for the material classification of the two, then true by attributive character
The material classification of fixed contraband target to be determined, if material classification meets for the pre-set categories of contraband material classification, really
It is set to contraband target, reduces the false detection rate of contraband target detection.Also, by being expanded, being projected to image pattern
Processing, limited image pattern is expanded and is trained for the biggish sample of sample size, it is possible to reduce over-fitting promotes detection property
Energy；After determining contraband target, according to the Sample Refreshment training in actual contraband target and given scenario sample set
Detection model, for carrying out the detection and identification of contraband target next time, by constantly iteration, the performance of detection model is not
It is disconnected to improve, even if also, actual scene change, still can adjust detection model in time, guarantee detection performance.
The embodiment of the invention also provides a kind of computer equipments, as shown in figure 8, including processor 801 and memory
802, wherein
Memory 802, for storing computer program；
Processor 801 when for executing the program stored on memory 802, realizes such as above-mentioned contraband target detection
All steps of method.
Above-mentioned memory may include RAM (Random Access Memory, random access memory), also may include
NVM (Non-Volatile Memory, nonvolatile memory), for example, at least a magnetic disk storage.Optionally, memory
It can also be that at least one is located remotely from the storage device of aforementioned processor.
Above-mentioned processor can be general processor, including CPU (Central Processing Unit, central processing
Device), NP (Network Processor, network processing unit) etc.；Can also be DSP (Digital Signal Processor,
Digital signal processor), ASIC (Application Specific Integrated Circuit, specific integrated circuit),
FPGA (Field-Programmable Gate Array, field programmable gate array) or other programmable logic device are divided
Vertical door or transistor logic, discrete hardware components.
In the present embodiment, the processor of the computer equipment is led to by reading the computer program stored in memory
It crosses and runs the computer program, can be realized: for radioscopic image to be detected there may be obscuring, show the problems such as imperfect,
By pre-set image Enhancement Method, the specific characteristic in radioscopic image to be detected is enhanced, in this way, obtained enhancing figure
The content in radioscopic image to be detected can clear as much as possible, be completely shown as in, avoided endless due to obscuring, showing
The problems such as whole caused missing inspection, reduces the omission factor of contraband target detection；It is treated by a variety of pre-set image Enhancement Methods
Detection radioscopic image is enhanced, then counts the quantity of the enhancing image there are contraband target to be determined, in the quantity of statistics
It is made whether when reaching threshold value as the judgement of contraband target, in this way, even if there are individual enhancing image erroneous detections, due to being provided with
Threshold value can not still be made whether as the judgement of contraband target, to reduce false detection rate；Also, for there may be with
The identical non-contraband target of the contour feature of contraband target then passes through attribute since there are difference for the material classification of the two
Feature determines the material classification of contraband target to be determined, if the default class that material classification meets for contraband material classification
Not, it is determined that be contraband target, reduce the false detection rate of contraband target detection.
In addition, the embodiment of the invention provides one corresponding to contraband object detection method provided by above-described embodiment
Kind storage medium when the computer program is executed by processor, realizes above-mentioned contraband target for storing computer program
All steps of detection method.
In the present embodiment, storage medium is stored with executes the inspection of contraband target provided by the embodiment of the present invention at runtime
The application program of survey method, therefore can be realized: for radioscopic image to be detected, there may be obscure, show imperfect etc. ask
Topic, by pre-set image Enhancement Method, enhances the specific characteristic in radioscopic image to be detected, in this way, obtained enhancing
The content in radioscopic image to be detected can clear as much as possible, be completely shown in image, avoided due to fuzzy, display not
The problems such as complete caused missing inspection, reduces the omission factor of contraband target detection；Pass through a variety of pre-set image Enhancement Methods pair
Radioscopic image to be detected is enhanced, then counts the quantity of the enhancing image there are contraband target to be determined, in the number of statistics
Amount is made whether when reaching threshold value as the judgement of contraband target, in this way, even if there are individual enhancing image erroneous detections, due to setting
Threshold value can not still be made whether as the judgement of contraband target, to reduce false detection rate；Also, for there may be
Non-contraband target identical with the contour feature of contraband target then passes through category since there are difference for the material classification of the two
Property feature determines the material classification of contraband target to be determined, if the default class that material classification meets for contraband material classification
Not, it is determined that be contraband target, reduce the false detection rate of contraband target detection.
For computer equipment and storage medium embodiment, method content as involved in it is substantially similar to
Embodiment of the method above-mentioned, so being described relatively simple, the relevent part can refer to the partial explaination of embodiments of method.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device,
For computer equipment and storage medium embodiment, since it is substantially similar to the method embodiment, so the comparison of description is simple
Single, the relevent part can refer to the partial explaination of embodiments of method.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all
Any modification, equivalent replacement, improvement and so within the spirit and principles in the present invention, are all contained in protection scope of the present invention
It is interior.
Claims (13)
1. a kind of contraband object detection method, which is characterized in that the described method includes:
By a variety of pre-set image Enhancement Methods, enhancing processing is carried out to the specific characteristic in the radioscopic image to be detected of acquisition,
Obtain multiple enhancing images；
Using goal-selling detection method, each enhancing image is detected, is judged in each enhancing image with the presence or absence of to be determined
Contraband target, and determine the essential attribute information of the contraband target to be determined, wherein the essential attribute information includes
The location information in region locating for the contraband target to be determined；
The quantity for counting the enhancing image there are the contraband target to be determined, if the quantity reaches preset threshold, from
Region of interest area image corresponding with the location information is extracted in the radioscopic image to be detected；
Obtain and according to the attributive character of the region of interest area image, determine the material class of the contraband target to be determined
Not；
If the material classification meets default material classification, it is determined that the contraband target to be determined detected is determining
Contraband target.
2. the method according to claim 1, wherein the acquisition modes of the radioscopic image to be detected, comprising:
Obtain the source radioscopic image acquired by X-ray security inspection machine；
Operation is normalized to the source radioscopic image, obtains radioscopic image to be detected.
3. the method according to claim 1, wherein goal-selling detection method is utilized described, to each enhancing
Image is detected, and is judged with the presence or absence of contraband target to be determined in each enhancing image, and determine the contraband to be determined
Before the essential attribute information of target, the method also includes:
Obtain multiple image patterns comprising any contraband target；
According to the coordinate information of contraband target area in each image pattern of calibration, the figure in each contraband target area is extracted
Picture obtains multiple sample exposure masks；
Default amplification processing is carried out to each sample exposure mask, the mask image after obtaining multiple amplifications；
Each mask image is projected into multiple pre-set images, multiple training samples are obtained；
The multiple training sample is trained using predetermined deep learning method, obtains detection model；
It is described utilize goal-selling detection method, each enhancing image is detected, judge in each enhancing image whether there is to
It determines contraband target, and determines the essential attribute information of the contraband target to be determined, comprising:
Using the detection model, each enhancing image is detected, judges to whether there is contraband target in each enhancing image,
And determine the essential attribute information of the contraband target to be determined.
4. if according to the method described in claim 3, it is characterized in that, meet default material class in the material classification
Not, it is determined that after the contraband target to be determined detected is determining contraband target, the method also includes:
By the sample in given scenario sample set and include the determination contraband target image setting be candidate samples,
It include the sample for obeying target distribution under given scenario in the given scenario sample set；
According to the correlation of each candidate samples, weight is distributed for each candidate samples；
Maximum first candidate samples of weight in each candidate samples are selected, and the selection and described the from the multiple training sample
The similarity of one candidate samples is greater than the second candidate samples of default similarity；
Updating the sample in the given scenario sample set is first candidate samples and second candidate samples；
The sample in updated given scenario sample set is trained using the predetermined deep learning method, is updated
Detection model the detection model using the update is executed, to each increasing with when obtaining radioscopic image to be detected next time
Strong image is detected, and is judged with the presence or absence of contraband target to be determined in each enhancing image, and determination is described to be determined violated
Items target essential attribute information.
5. the method according to claim 1, wherein described obtain and according to the category of the region of interest area image
Property feature, determines the material classification of the contraband target to be determined, comprising:
Obtain the gray value of the region of interest area image；
Determine default tonal range locating for the gray value；
Based on the default tonal range, the material classification of the contraband target to be determined is determined.
6. method according to any one of claims 1 to 5, which is characterized in that the essential attribute information further include it is described to
Determine the type information of contraband target；
If meeting default material classification in the material classification, it is determined that the contraband target to be determined detected is
After determining contraband target, the method also includes:
According to the type information, the type of the contraband target of the determination is obtained.
7. a kind of contraband object detecting device, which is characterized in that described device includes:
Enhance processing module, is used for through a variety of pre-set image Enhancement Methods, to specified in the radioscopic image to be detected of acquisition
Feature carries out enhancing processing, obtains multiple enhancing images；
Detection module, for utilize goal-selling detection method, each enhancing image is detected, judge be in each enhancing image
No there are contraband targets to be determined, and determine the essential attribute information of the contraband target to be determined, wherein described basic
Attribute information includes the location information in region locating for the contraband target to be determined；
Statistical module, for counting the quantity of the enhancing image there are the contraband target to be determined, if the quantity reaches
Preset threshold then extracts region of interest area image corresponding with the location information from the radioscopic image to be detected；
It obtains module and determines the contraband to be determined for obtaining and according to the attributive character of the region of interest area image
The material classification of target；
Determining module, if meeting default material classification for the material classification, it is determined that is detected is described to be determined violated
Items are designated as determining contraband target.
8. device according to claim 7, which is characterized in that the acquisition module is also used to:
Obtain the source radioscopic image acquired by X-ray security inspection machine；
Operation is normalized to the source radioscopic image, obtains radioscopic image to be detected.
9. device according to claim 7, which is characterized in that described device further include:
Image pattern obtains module, for obtaining multiple image patterns comprising any contraband target；
Extraction module extracts each contraband for the coordinate information of contraband target area in each image pattern according to calibration
Image in target area obtains multiple sample exposure masks；
Processing module is expanded, for carrying out default amplification processing to each sample exposure mask, the mask image after obtaining multiple amplifications；
Projection module obtains multiple training samples for projecting each mask image into multiple pre-set images；
First training module is detected for being trained using predetermined deep learning method to the multiple training sample
Model；
The detection module, is specifically used for:
Using the detection model, each enhancing image is detected, judges to whether there is contraband target in each enhancing image,
And determine the essential attribute information of the contraband target to be determined.
10. device according to claim 9, which is characterized in that described device further include:
Setting module, for by the sample in given scenario sample set and include that the image of contraband target of the determination is set
It is set to candidate samples, includes the sample for obeying target distribution under given scenario in the given scenario sample set；
Distribution module distributes weight for the correlation according to each candidate samples for each candidate samples；
Selecting module, for selecting maximum first candidate samples of weight in each candidate samples, and from the multiple training sample
The similarity of middle selection and first candidate samples is greater than the second candidate samples of default similarity；
Update module is that first candidate samples and described second wait for updating the sample in the given scenario sample set
Sampling sheet；
Second training module, for utilizing the predetermined deep learning method to the sample in updated given scenario sample set
It is trained, the detection model updated, to execute and utilize the update when obtaining radioscopic image to be detected next time
Detection model, each enhancing image is detected, is judged in each enhancing image with the presence or absence of contraband target to be determined, and really
The essential attribute information of the fixed contraband target to be determined.
11. device according to claim 7, which is characterized in that the acquisition module is specifically used for:
Obtain the gray value of the region of interest area image；
Determine default tonal range locating for the gray value；
Based on the default tonal range, the material classification of the contraband target to be determined is determined.
12. according to any device of claim 7 to 11, which is characterized in that the essential attribute information further includes described
The type information of contraband target to be determined；
Described device further include:
Determination type module, for obtaining the type of the contraband target of the determination according to the type information.
13. a kind of computer equipment, which is characterized in that including processor and memory, wherein
The memory, for storing computer program；
The processor when for executing the program stored on the memory, realizes any side claim 1-6
Method step.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810020069.9A CN110020647B (en) | 2018-01-09 | 2018-01-09 | Contraband target detection method and device and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810020069.9A CN110020647B (en) | 2018-01-09 | 2018-01-09 | Contraband target detection method and device and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110020647A true CN110020647A (en) | 2019-07-16 |
| CN110020647B CN110020647B (en) | 2021-03-02 |
Family
ID=67187822
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810020069.9A Active CN110020647B (en) | 2018-01-09 | 2018-01-09 | Contraband target detection method and device and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110020647B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389000A (en) * | 2018-10-18 | 2019-02-26 | 龙马智芯（珠海横琴）科技有限公司 | A kind of barcode recognition method and the computer using this method |
| CN110514671A (en) * | 2019-09-24 | 2019-11-29 | 广西贺州市红星纸业有限公司 | A kind of novel pulpboard cleanliness detection method |
| CN110531435A (en) * | 2019-08-29 | 2019-12-03 | 公安部第一研究所 | It is a kind of for testing the test box and test method of contraband probe algorithm |
| CN110619600A (en) * | 2019-09-17 | 2019-12-27 | 南京旷云科技有限公司 | Neural network model training method and device, storage medium and electronic equipment |
| CN110751079A (en) * | 2019-10-16 | 2020-02-04 | 北京海益同展信息科技有限公司 | Article detection method, apparatus, system and computer readable storage medium |
| CN110910467A (en) * | 2019-12-03 | 2020-03-24 | 浙江啄云智能科技有限公司 | X-ray image sample generation method, system and application |
| CN111783812A (en) * | 2019-11-18 | 2020-10-16 | 北京沃东天骏信息技术有限公司 | Method and device for identifying forbidden images and computer readable storage medium |
| CN112037155A (en) * | 2020-11-03 | 2020-12-04 | 长沙海信智能系统研究院有限公司 | Method, device, equipment and medium for determining target attribute structured data |
| CN112329497A (en) * | 2019-07-18 | 2021-02-05 | 杭州海康威视数字技术股份有限公司 | Target identification method, device and equipment |
| CN113297888A (en) * | 2020-09-18 | 2021-08-24 | 阿里巴巴集团控股有限公司 | Method and device for checking image content detection result |
| CN113449556A (en) * | 2020-03-26 | 2021-09-28 | 顺丰科技有限公司 | Contraband detection method and device, edge computing equipment and storage medium |
| CN114742979A (en) * | 2022-04-15 | 2022-07-12 | 推想医疗科技股份有限公司 | Position identification method and device based on positioning patch, storage medium and equipment |
| CN115564320A (en) * | 2022-12-06 | 2023-01-03 | 成都智元汇信息技术股份有限公司 | Multi-intelligent-algorithm-oriented scheduling management method, device and medium |
| CN116958646A (en) * | 2023-06-12 | 2023-10-27 | 北京声迅电子股份有限公司 | Forbidden article detection method based on X-ray image |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080152082A1 (en) * | 2006-08-16 | 2008-06-26 | Michel Bouchard | Method and apparatus for use in security screening providing incremental display of threat detection information and security system incorporating same |
| CN103530629A (en) * | 2013-11-01 | 2014-01-22 | 四川九洲电器集团有限责任公司 | Method and system for detecting extended range of small infrared signal target |
| JP2015184908A (en) * | 2014-03-24 | 2015-10-22 | 東芝アルパイン・オートモティブテクノロジー株式会社 | Image processor and image processing program |
| CN105096350A (en) * | 2014-05-21 | 2015-11-25 | 腾讯科技（深圳）有限公司 | Image detection method and device |
| CN105844657A (en) * | 2016-04-26 | 2016-08-10 | 江南大学 | Dual-channel fabric fault detection method based on two-dimensional empirical mode decomposition |
| CN105869148A (en) * | 2016-03-24 | 2016-08-17 | 北京小米移动软件有限公司 | Target detection method and device |
| CN106250936A (en) * | 2016-08-16 | 2016-12-21 | 广州麦仑信息科技有限公司 | Multiple features multithreading safety check contraband automatic identifying method based on machine learning |
| US20160375904A1 (en) * | 2015-06-29 | 2016-12-29 | The Regents Of The University Of California | Multiple-parts based vehicle detection integrated with lane detection for improved computational efficiency and robustness |
-
2018
- 2018-01-09 CN CN201810020069.9A patent/CN110020647B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080152082A1 (en) * | 2006-08-16 | 2008-06-26 | Michel Bouchard | Method and apparatus for use in security screening providing incremental display of threat detection information and security system incorporating same |
| CN103530629A (en) * | 2013-11-01 | 2014-01-22 | 四川九洲电器集团有限责任公司 | Method and system for detecting extended range of small infrared signal target |
| JP2015184908A (en) * | 2014-03-24 | 2015-10-22 | 東芝アルパイン・オートモティブテクノロジー株式会社 | Image processor and image processing program |
| CN105096350A (en) * | 2014-05-21 | 2015-11-25 | 腾讯科技（深圳）有限公司 | Image detection method and device |
| US20160375904A1 (en) * | 2015-06-29 | 2016-12-29 | The Regents Of The University Of California | Multiple-parts based vehicle detection integrated with lane detection for improved computational efficiency and robustness |
| CN105869148A (en) * | 2016-03-24 | 2016-08-17 | 北京小米移动软件有限公司 | Target detection method and device |
| CN105844657A (en) * | 2016-04-26 | 2016-08-10 | 江南大学 | Dual-channel fabric fault detection method based on two-dimensional empirical mode decomposition |
| CN106250936A (en) * | 2016-08-16 | 2016-12-21 | 广州麦仑信息科技有限公司 | Multiple features multithreading safety check contraband automatic identifying method based on machine learning |
Non-Patent Citations (3)
| Title |
|---|
| S NERCESSIAN 等: "Automatic detection of potential threat objects in X-ray luggage scan images", 《2008 IEEE CONFERENCE ON TECHNOLOGIES FOR HOMELAND SECURITY》 * |
| 宋修竹: "X光安检图像处理方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 栾孟杰 等: "小波融合的乳腺图像增强方法", 《计算机工程与应用》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389000A (en) * | 2018-10-18 | 2019-02-26 | 龙马智芯（珠海横琴）科技有限公司 | A kind of barcode recognition method and the computer using this method |
| CN112329497A (en) * | 2019-07-18 | 2021-02-05 | 杭州海康威视数字技术股份有限公司 | Target identification method, device and equipment |
| CN110531435A (en) * | 2019-08-29 | 2019-12-03 | 公安部第一研究所 | It is a kind of for testing the test box and test method of contraband probe algorithm |
| CN110531435B (en) * | 2019-08-29 | 2024-05-31 | 公安部第一研究所 | Test box and test method for testing contraband detection algorithm |
| CN110619600A (en) * | 2019-09-17 | 2019-12-27 | 南京旷云科技有限公司 | Neural network model training method and device, storage medium and electronic equipment |
| CN110619600B (en) * | 2019-09-17 | 2023-12-08 | 南京旷云科技有限公司 | Neural network model training method and device, storage medium and electronic equipment |
| CN110514671A (en) * | 2019-09-24 | 2019-11-29 | 广西贺州市红星纸业有限公司 | A kind of novel pulpboard cleanliness detection method |
| CN110751079A (en) * | 2019-10-16 | 2020-02-04 | 北京海益同展信息科技有限公司 | Article detection method, apparatus, system and computer readable storage medium |
| WO2021073370A1 (en) * | 2019-10-16 | 2021-04-22 | 北京海益同展信息科技有限公司 | Article inspection method, apparatus and system and computer readable storage medium |
| CN111783812A (en) * | 2019-11-18 | 2020-10-16 | 北京沃东天骏信息技术有限公司 | Method and device for identifying forbidden images and computer readable storage medium |
| CN110910467B (en) * | 2019-12-03 | 2023-04-18 | 浙江啄云智能科技有限公司 | X-ray image sample generation method, system and application |
| CN110910467A (en) * | 2019-12-03 | 2020-03-24 | 浙江啄云智能科技有限公司 | X-ray image sample generation method, system and application |
| CN113449556A (en) * | 2020-03-26 | 2021-09-28 | 顺丰科技有限公司 | Contraband detection method and device, edge computing equipment and storage medium |
| CN113449556B (en) * | 2020-03-26 | 2024-06-04 | 顺丰科技有限公司 | Contraband detection method and device, edge computing equipment and storage medium |
| CN113297888A (en) * | 2020-09-18 | 2021-08-24 | 阿里巴巴集团控股有限公司 | Method and device for checking image content detection result |
| CN113297888B (en) * | 2020-09-18 | 2024-06-07 | 阿里巴巴集团控股有限公司 | Image content detection result checking method and device |
| CN112037155A (en) * | 2020-11-03 | 2020-12-04 | 长沙海信智能系统研究院有限公司 | Method, device, equipment and medium for determining target attribute structured data |
| CN114742979A (en) * | 2022-04-15 | 2022-07-12 | 推想医疗科技股份有限公司 | Position identification method and device based on positioning patch, storage medium and equipment |
| CN115564320A (en) * | 2022-12-06 | 2023-01-03 | 成都智元汇信息技术股份有限公司 | Multi-intelligent-algorithm-oriented scheduling management method, device and medium |
| CN115564320B (en) * | 2022-12-06 | 2023-04-07 | 成都智元汇信息技术股份有限公司 | Multi-intelligent-algorithm-oriented scheduling management method, device and medium |
| CN116958646A (en) * | 2023-06-12 | 2023-10-27 | 北京声迅电子股份有限公司 | Forbidden article detection method based on X-ray image |
| CN116958646B (en) * | 2023-06-12 | 2024-01-30 | 北京声迅电子股份有限公司 | Forbidden article detection method based on X-ray image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110020647B (en) | 2021-03-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110020647A (en) | A kind of contraband object detection method, device and computer equipment | |
| US10885531B2 (en) | Artificial intelligence counterfeit detection | |
| JP6878575B2 (en) | Banknote management methods, systems, programs and recording media | |
| CN106295601B (en) | A kind of improved Safe belt detection method | |
| CN105518709B (en) | The method, system and computer program product of face for identification | |
| Huang et al. | A new building extraction postprocessing framework for high-spatial-resolution remote-sensing imagery | |
| CN101142584B (en) | Method for facial features detection | |
| CN105303193B (en) | A kind of passenger number statistical system based on single-frame images processing | |
| US10509979B2 (en) | Inspection methods and systems | |
| Jaccard et al. | Tackling the X-ray cargo inspection challenge using machine learning | |
| Yang et al. | CNN with spatio-temporal information for fast suspicious object detection and recognition in THz security images | |
| CN111145177A (en) | Image sample generation method, specific scene target detection method and system thereof | |
| CN101901346B (en) | Method for identifying unsuitable content in colour digital image | |
| CN110060237A (en) | A kind of fault detection method, device, equipment and system | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN108291876A (en) | System and method for the authenticity for detecting product | |
| CN108108751A (en) | A kind of scene recognition method based on convolution multiple features and depth random forest | |
| CN105160355B (en) | A kind of method for detecting change of remote sensing image based on region correlation and vision word | |
| CN104200478B (en) | Low-resolution touch screen image defect detection method based on sparse representation | |
| Björklund et al. | Automatic license plate recognition with convolutional neural networks trained on synthetic data | |
| CN108288279A (en) | Article discrimination method based on X-ray image foreground target extraction | |
| CN102542293A (en) | Class-I extraction and classification method aiming at high-resolution SAR (Synthetic Aperture Radar) image scene interpretation | |
| CN103020649A (en) | Forest type identification method based on texture information | |
| CN106952393A (en) | Paper Currency Identification and device, electronic equipment and storage medium | |
| Jaccard et al. | Using deep learning on X-ray images to detect threats |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |