WO2020173021A1 - 一种基于人工智能的违禁物识别方法、装置、设备和存储介质 - Google Patents

一种基于人工智能的违禁物识别方法、装置、设备和存储介质 Download PDF

Info

Publication number
WO2020173021A1
WO2020173021A1 PCT/CN2019/092678 CN2019092678W WO2020173021A1 WO 2020173021 A1 WO2020173021 A1 WO 2020173021A1 CN 2019092678 W CN2019092678 W CN 2019092678W WO 2020173021 A1 WO2020173021 A1 WO 2020173021A1
Authority
WO
WIPO (PCT)
Prior art keywords
package
ray image
prohibited
pixel
pixels
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.)
Ceased
Application number
PCT/CN2019/092678
Other languages
English (en)
French (fr)
Inventor
吴壮伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020173021A1 publication Critical patent/WO2020173021A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V5/00—Prospecting or detecting by the use of ionising radiation, e.g. of natural or induced radioactivity
    • G01V5/20—Detecting prohibited goods, e.g. weapons, explosives, hazardous substances, contraband or smuggled objects
    • G01V5/22—Active interrogation, i.e. by irradiating objects or goods using external radiation sources, e.g. using gamma rays or cosmic rays
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/12—Edge-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Definitions

  • This application relates to the field of image recognition, in particular to methods, devices, equipment, and storage media for identifying prohibited objects based on artificial intelligence.
  • the security inspection machine (or called the security inspection instrument) is an electronic device that sends the checked luggage into the X-ray inspection channel by means of a conveyor belt to complete the inspection.
  • the package detection sensor After the baggage enters the X-ray inspection channel, the package detection sensor will be blocked; the package detection sensor sends a detection signal, and the detection signal is sent to the system control part; the system control part generates an X-ray trigger signal to trigger the X-ray source to emit an X-ray beam.
  • a fan-shaped X-ray beam passing through the collimator passes through the security-checked items on the conveyor belt. The X-rays are absorbed by the security-checked items, and finally bombard the dual-energy semiconductor detector installed in the channel.
  • the dual-energy semiconductor detector converts X-rays into signals. These weak signals are amplified and sent to the signal processing box for further processing to form X-ray video.
  • X-rays are electromagnetic waves that can penetrate opaque objects such as wood, cardboard, and leather.
  • the security inspection system can present images of different colors on the screen according to the degree of absorption of X-rays by the object. At this time, the security inspector quickly checked the X-ray video of the X-ray scan on the screen, and judged whether there were any prohibited objects based on experience.
  • the main purpose of this application is to solve the technical problems of low efficiency and high labor cost in the existing method of identifying prohibited items.
  • the probability value of each suspected prohibited item is obtained.
  • An artificial intelligence-based method for identifying contrabands includes: collecting X-ray images from X-ray video taken by a security inspection machine at a preset sampling time interval; identifying the area where each package in the X-ray image is located, according to each The area where the package is located is divided into X-ray images corresponding to each package from the X-ray image; the target data is extracted from the X-ray image through the convolutional layer of the convolutional neural network model, and the volume
  • the pooling layer of the product neural network model performs de-redundancy processing on the target data to obtain item feature information; inputs the item feature information to the prohibited object recognition model; and uses the prohibited object recognition model to analyze the X-ray image Identify each security-checked item in the package in the X-ray image, and if it is identified that there is a suspicious prohibited item in the package in the X-ray image, then match each suspicious prohibited item with a corresponding preset prohibited probability value;
  • the accumulator superimposes the prohibited probability values
  • the method before the collecting X-ray images from the X-ray video shot by the security inspection machine at a preset sampling time interval, the method further includes:
  • the sampling time interval is set according to the time required for the security-checked article to pass through the shooting area of the security screening machine.
  • the setting the sampling time interval according to the time required for the security-checked article to pass through the shooting area of the security inspection machine includes:
  • Obtain the length of the shooting area of the security inspection machine obtain the conveying speed of the conveyor belt of the security inspection machine; divide the length by the conveying speed to obtain a sampling time reference value; compare the sampling time reference value to a preset
  • the constant a of is multiplied to obtain the sampling time interval; the constant a is greater than 0 and less than or equal to 1.
  • the identifying the area of each package in the X-ray image, and dividing the X-ray image from the X-ray image according to the area where each package is located includes:
  • a corresponding X-ray image includes:
  • the first pixel is any pixel of the package; traverse each pixel around the first pixel; If there are pixels in the background area among the four adjacent pixels on the upper, lower, left, and right of the second pixel, it is determined that the second pixel is a pixel on the outline of the package; The second pixel is any pixel around the first pixel; all pixels on the contour of the package to which the first pixel belongs are traversed; the package to which the first pixel belongs is extracted The image of the area enclosed by all pixels on the contour of is used as the X-ray sub-image corresponding to the package to which the first pixel belongs.
  • the method further includes:
  • the issuing of a prohibited object prompt includes at least one of the following implementation methods: emitting a prompt sound; turning on a prompt light; popping up a prohibited object prompt box on the screen of the security inspection machine; and marking the prohibited object in the X-ray image.
  • the expression of the prohibited object recognition model is:
  • I is the dimension of the input vector
  • V is the dimension of the vectorized item in the X-photon image
  • H is the number of neurons in the hidden layer
  • K is the number of neurons in the output layer
  • x is the volume
  • the item feature information includes item shape feature information and item color feature information.
  • this application also provides an artificial intelligence-based contraband identification device, which includes a transceiver module and a processing module.
  • the processing module is used to control the transceiving operation of the transceiving module.
  • the transceiver module is used to collect X-ray images from the X-ray video shot by the security inspection machine at a preset sampling time interval.
  • the processing module is used to identify the area where each package in the X-ray image is located, and divide the X-ray image from the X-ray image according to the area where each package is located;
  • the convolutional layer of the convolutional neural network model extracts target data from the X-photon image, and the target data is de-redundantly processed through the pooling layer of the convolutional neural network model to obtain item feature information;
  • the item feature information is input to the prohibited object recognition model; through the prohibited object recognition model, each security-checked object in the package in the X-ray image is recognized, if it is recognized that there is in the package in the X-ray image
  • the suspicious prohibited objects match the preset prohibited probability value corresponding to each suspicious prohibited object; the total probability value corresponding to each suspicious prohibited object is superimposed by the accumulator to obtain the total probability value; if the total probability value is summed If it is greater than the prohibited object threshold, it is determined that there is a prohibited object in the package in the X-ray image.
  • the processing module is specifically configured to use the gray value of the background area in the X-ray image as a reference to identify the pixel points of the package in the X-ray image; respectively identify the X-ray image According to all the pixels on the outline of each package, the X-ray image corresponding to each package is divided from the X-ray image according to all the pixels on the outline of each package.
  • the processing module is specifically configured to randomly extract any first pixel of the package that has not been traversed in the X-ray image; the first pixel is any pixel of the package; Pixels around the first pixel; if there is a pixel in the background area among the four adjacent pixels on the upper, lower, left, and right of the second pixel, then the second pixel is determined A point is a pixel on the outline of the package; the second pixel is any pixel around the first pixel; all pixels on the outline of the package to which the first pixel belongs are traversed; Extracting an image of an area surrounded by all pixels on the contour of the package to which the first pixel belongs is taken as the X-ray image corresponding to the package to which the first pixel belongs.
  • the processing module is further configured to, if it is determined that there is a prohibited object in the package in the X-ray image, send a prohibited object prompt through the transceiver module.
  • the issuing of a prohibited object prompt includes at least one of the following implementation methods: emitting a prompt sound; turning on a prompt light; popping up a prohibited object prompt box on the screen of the security inspection machine; and marking the prohibited object in the X-ray image.
  • the expression of the prohibited object recognition model is:
  • I is the dimension of the input vector
  • V is the dimension of the vectorized item in the X-photon image
  • H is the number of neurons in the hidden layer
  • K is the number of neurons in the output layer
  • x is the volume
  • the item feature information includes item shape feature information and item color feature information.
  • the present application also provides a computer device, including a transceiver, a memory, and a processor.
  • the memory stores computer-readable instructions.
  • the processor is caused to execute the steps of the artificial intelligence-based method for identifying prohibited objects.
  • the present application also provides a storage medium storing non-volatile computer-readable instructions.
  • the computer-readable instructions When executed by one or more processors, the one or more processors Perform the steps of the aforementioned artificial intelligence-based method for identifying prohibited objects.
  • FIG. 1 is a schematic flowchart of a method for identifying prohibited objects based on artificial intelligence in an embodiment of the application.
  • Fig. 2 is a schematic diagram of the flow of step S2 in Fig. 1.
  • Fig. 3 is a schematic structural diagram of an artificial intelligence-based contraband identification device in an embodiment of the application.
  • Figure 4 is a schematic structural diagram of a computer device in an embodiment of the application.
  • Figure 1 is a flow chart of an artificial intelligence-based method for identifying prohibited objects in some embodiments of the application.
  • the method for identifying prohibited objects is executed by a prohibited object identification device.
  • the prohibited object identification device can be integrated in the security inspection machine or in conjunction with the security inspection.
  • the electrical connection of the machine, as shown in Figure 1, may include the following steps S1-S6:
  • S1 Collect X-ray images from X-ray videos taken by the security inspection machine at a preset sampling time interval.
  • the X-ray video taken by the security inspection machine is sampled at a preset sampling time interval to obtain X-ray images.
  • the X-ray image carries the item characteristic information; according to the item characteristic information, the security-checked item in the X-ray image is identified to identify whether the security-checked item is a suspicious prohibited item.
  • step S1 before step S1, the method further includes step S11: setting the sampling time interval according to the time required for the security-checked article to pass through the imaging area of the security screening machine.
  • each security-checked item is driven by the conveyor belt of the security inspection machine to pass through the security inspection machine's shooting area at a constant speed, the time required for the security-checked item to pass through the security inspection machine's shooting area is unchanged, which means that each security-checked item is in X
  • the appearance time in the light video is constant and the same. In order to collect the X-ray image of each security-checked item in the X-ray video, the sampling time interval needs to be set reasonably.
  • the time to pass through the shooting area of the security inspection machine is 5 seconds; if the sampling time interval is greater than 5 seconds, for example, set to 7 seconds or 8 seconds, in this way, the X-ray image collection time is longer than the appearance of the security-checked item in the X-ray video Time, the X-ray images of some security-checked items cannot be collected from the X-ray video, resulting in missed inspection; if the sampling time interval is too small, for example, set to 1 second, in this way, because the X-ray image acquisition time is less than
  • the appearance time of the security-checked item in the X-ray video results in that the X-ray image of the same security-checked item is collected 5 times within the 5-second appearance time of the X-ray video, which means the same security-checked item
  • the security check has been done 5 times, which undoubtedly increases the amount of data processing and calculations during the security check process (including multiple unnecessary repetitive calculations), resulting in the inability to feed back the X-ray image
  • sampling time interval is set to 4 seconds, this can basically ensure that each security-checked item in the X-ray video is only collected once X-ray image within the 5 second display time of the X-ray video. A missed inspection occurred. Therefore, the sampling time interval cannot be greater than the time that the security-checked article appears in the X-ray video, and the sampling time interval should not be too short.
  • the sampling time interval of X-ray images can be adjusted and set according to actual needs.
  • the time required for the security-checked article to pass through the area captured by the X-ray video can be calculated from the conveying speed of the conveyor belt of the security inspection machine and the length of the part that the conveyor belt can show in the X-ray image.
  • step S11 includes the following steps S111-S114:
  • the shooting area of the security inspection machine is also the area shown by the X-ray video.
  • the length of the shooting area of the security check machine is determined, and the actual distance of the security check item through the shooting area of the security check machine is also determined.
  • the conveying speed of the conveyor belt can be set in advance. Once the conveying speed of the conveyor belt is obtained, the speed at which the security-checked items pass through the shooting area of the security inspection machine is also obtained.
  • sampling time reference value represents the time required for the security-checked article to pass through the shooting area of the security screening machine.
  • the constant a is greater than 0 and less than or equal to 1.
  • the sampling time interval is less than or equal to the sampling time reference value.
  • the sampling time interval of the X-ray image can be automatically calculated and set, so that the sampling time interval of the X-ray image in this method is the same as that of the security-checked article.
  • the moving speed is compatible, saving the trouble of manually setting the sampling interval.
  • the packaging is used to contain the security-checked items.
  • a frame of X-ray image may have multiple passengers' security-checked items.
  • Each passenger's security-checked items are generally placed in their own packaging, and the security inspection machine performs security checks on the items in each package.
  • multiple packages are usually passed through the security inspection machine one after another, so there may be multiple packages in one frame of X-ray image.
  • security inspectors identify suspicious contraband in the X-ray video package, and then confirm and inspect the package containing the suspicious contraband.
  • the security personnel actually identify the packages in the X-ray video one by one.
  • to determine which passenger’s package contains suspicious prohibited items in this application it is necessary to distinguish each package first, and then identify whether each package contains suspicious prohibited items. When a package is found When there are suspicious contrabands in the possession, only further inspection of the packaging containing the suspicious contrabands is enough.
  • Packages include devices such as barrels, boxes, baskets, bottles, jars, cans, bags, etc., used to contain items. It should be understood that some items that have passed through the security inspection machine alone, such as equipment, etc., although they are not in the package itself, may contain prohibited items inside, so the items that have passed the security inspection machine alone are also regarded as packaging. .
  • step S2 includes the following steps S21-S22:
  • the background picture displayed by the X-ray video is fixed, so the gray value of each pixel in the background area in the X-ray image is also fixed.
  • the background area in the X-ray image and the gray value of the package are different, so the background area in the X-ray image is easy to distinguish from the package.
  • Each pixel in the background area in the X-ray image is marked as logic 0, and each pixel in the X-ray image belonging to the package is marked as logical 1, to distinguish each pixel in the background area from each pixel in the package.
  • the background area in the X-ray image separates each package. Find the edge position of each package and the background area, and then divide the area of each package in the X-ray image, that is, get the X-ray image corresponding to each package, and then compare the package in the X-ray image Identify prohibited items.
  • step S22 includes the following steps S221-S223:
  • the first pixel is any pixel of the package.
  • a random walk method is used to traverse the X-ray image. First, randomly obtain the first pixel point of any package that has not been traversed in the X-ray image matrix as the starting pixel point of the package in the traversed X-ray image. Starting from the starting pixel position, each pixel in the traversed package is traversed.
  • step S221 find other packages that have not been traversed, and continue to extract the package that has not been traversed until the X-ray image of each package in the X-ray image is one One divided out.
  • the item feature information includes item shape feature information, item color feature information, and the like.
  • the convolutional neural network model starts from the position of the starting pixel of the X-photon image and gradually traverses the data of the X-photon image with 1 pixel as a step. Run the convolution operation to extract the item feature information in the X-ray image.
  • the convolutional neural network model splices item feature information into continuous data, and inputs the spliced item feature information into the contraband identification model.
  • the expression of the prohibited object recognition model is:
  • I is the dimension of the input vector
  • V is the dimension of the vectorized item in the X-photon image
  • H is the number of neurons in the hidden layer
  • K is the number of neurons in the output layer
  • x is the volume
  • the prohibited object recognition model in advance. Specifically, a certain amount of X-ray images for training are collected from X-ray videos, and the X-ray images for training contain feature information of prohibited objects. Manually mark the prohibited objects in each training X-ray image, and assign the corresponding prohibited probability value to each marked prohibited object. Input the manually labeled training X-ray image data into the contraband identification model.
  • the prohibited object recognition model memorizes the feature information of the prohibited objects marked by the X-ray image for training, remembers the weight value of each feature information of each prohibited object, so as to realize the accurate identification of each prohibited object and output each prohibited object. The preset probability of violation corresponding to the object.
  • Convolutional neural networks extract item feature information in X-ray images and send the extracted item feature information to the recurrent neural network;
  • Neural network recognition recognizes prohibited objects in X-ray images based on item feature information.
  • the training principle of the recognition model is generally the same, so I won't repeat it here.
  • the water bottle in the package is not enough to determine that the water bottle contains flammable and explosive liquid, but there are also flammable and explosive liquids in the water bottle. Possibility of liquid. Therefore, set a corresponding banned probability value for items suspected of water bottles.
  • a list of the identification probability of a variety of prohibited objects is established, and the prohibited objects are identified by assigning probability values to each suspected prohibited object.
  • the prohibited item is not necessarily a complete item, but can also be a part of a complete item.
  • a complete item can be divided into multiple specific parts, and a corresponding prohibited probability value can be assigned to each part.
  • the strips in the package may be prohibited knives, or other normal items.
  • the strips similar to the prohibited knives are assigned a corresponding prohibited probability value; the prohibited sticks will be suitable for holding
  • the handle object assigns a corresponding prohibited probability value to the handle object similar to the prohibited tool; superimpose the prohibited probability value corresponding to the bar similar to the prohibited tool with the prohibited probability value corresponding to the handle object similar to the prohibited tool,
  • the probability value obtained can more accurately reflect whether the identified overall item is a prohibited item. The more specific the position of the item is identified, the higher the accuracy of the identification.
  • An accumulator is set in advance, and the probabilistic value corresponding to each possible prohibited item in the package is summed through the accumulator. The higher the sum of the probability values obtained, the greater the possibility that the package contains prohibited items.
  • a very low probability of prohibition is not enough to accurately determine the presence of prohibited objects in the package, and it is likely to be the prohibited probability value obtained by identifying non-prohibited objects similar to the prohibited objects.
  • Set a prohibited object threshold and compare the sum of the probability values with the preset prohibited object threshold. When the sum of the prohibited probability values exceeds the preset prohibited object threshold, it is determined that the package contains prohibited objects, which can be more accurate Identify whether the package contains prohibited substances.
  • the issuing of a prohibited object prompt includes at least one of the following implementations: emitting a prompt sound; turning on a prompt light; popping up a prohibited object prompt box on the screen of the security inspection machine; and marking in the X-ray image Prohibited items.
  • the background area and the package in the X-ray image are identified by gray value, and the X-ray image corresponding to each package is divided from the X-ray image. Identify the security-checked items in the package, identify the prohibited items that may exist in each package, and assign a corresponding prohibited probability value to each suspected prohibited object, and sum the prohibited probability values to obtain the sum of the probability values , Compare the total probability value with the preset threshold of prohibited objects, and accurately determine whether there are prohibited objects in the prohibited objects. The labor cost in the security inspection process is reduced, and the efficiency of the security inspection is improved.
  • the present application also provides a device for identifying contraband based on artificial intelligence.
  • the device includes a transceiver module 1 and a processing module 2.
  • the processing module 2 is used to control the transceiving operation of the transceiving module 1.
  • the transceiver module 1 is used to collect X-ray images from X-ray videos taken by the security inspection machine at a preset sampling time interval.
  • the processing module 2 is used to identify the area where each package is located in the X-ray image, and divide the X-ray image from the X-ray image according to the area where each package is located;
  • the convolutional layer of the convolutional neural network model extracts target data from the X-photon image, and the target data is de-redundantly processed through the pooling layer of the convolutional neural network model to obtain item feature information;
  • the item feature information is input to the prohibited object recognition model; each security-checked item in the package in the X-ray image is identified through the prohibited object recognition model, if the package memory in the X-ray image is recognized In the case of suspicious prohibited objects, each suspicious prohibited object is matched with the corresponding preset prohibited probability value; through the accumulator, the prohibited probability values corresponding to each suspicious prohibited object are superimposed to obtain the total probability value; if the probability value is If the sum is greater than the prohibited object threshold, it is determined that there is a prohibited object in the package in the X-ray image.
  • the processing module 2 is specifically configured to identify each pixel in the background area of the X-ray image and each pixel belonging to the package in the X-ray image according to the gray value of the pixel; respectively identify the X-ray image According to all the pixels on the outline of each package, the X-ray image corresponding to each package is divided from the X-ray image according to all the pixels on the outline of each package.
  • the processing module 2 is specifically configured to randomly extract a pixel point of any package in the X-ray image that has not been traversed, and use the randomly extracted pixel point as the starting pixel point; The pixels around the starting pixel; determine whether there is a pixel in the background area among the four adjacent pixels on the top, bottom, left, and right of the traversed pixel; if the traversed pixel is above, If there are pixels in the background area in the four adjacent pixels on the bottom, left, and right, it is determined that the traversed pixel is the pixel on the outline of the package; the package to which the starting pixel belongs is traversed Extract all pixels on the outline of the object; extract the image of the area enclosed by all pixels on the outline of the package to which the start pixel belongs, and use it as the X-ray image of the package to which the start pixel belongs.
  • the processing module 2 is further configured to send out a prohibited object reminder through the transceiver module 1 if it is determined that there is a prohibited object in the package in the X-ray image.
  • the expression of the prohibited object recognition model is:
  • I is the dimension of the input vector
  • V is the dimension of the vectorized item in the X-photon image
  • H is the number of neurons in the hidden layer
  • K is the number of neurons in the output layer
  • x is the volume
  • the item feature information includes item shape feature information and item color feature information.
  • the background area and the package in the X-ray image are identified by gray value, and the X-ray image corresponding to each package is divided from the X-ray image. Identify the security-checked items in the package, identify the prohibited items that may exist in each package, and assign a corresponding prohibited probability value to each suspected prohibited object, and sum the prohibited probability values to obtain the sum of the probability values , Compare the total probability value with the preset threshold of prohibited objects, and accurately determine whether there are prohibited objects in the prohibited objects. The labor cost in the security inspection process is reduced, and the efficiency of the security inspection is improved.
  • the present application also provides a computer device.
  • the computer device includes a transceiver 901, a processor 902, and a memory 903.
  • the memory 903 stores computer-readable instructions.
  • the processor 902 executes the steps of the artificial intelligence-based method for identifying contraband in the foregoing embodiments.
  • the physical device corresponding to the transceiver module 1 shown in FIG. 3 is the transceiver 901 shown in FIG. 4, and the transceiver 901 can implement part or all of the transceiver module 1 and the same or similar functions.
  • the physical device corresponding to the processing module 2 shown in FIG. 3 is the processor 902 shown in FIG. 4, which can implement part or all of the processing module 2 and the same or similar functions.
  • the present application also provides a storage medium storing non-volatile computer-readable instructions.
  • the computer-readable instructions When executed by one or more processors, the one or more processors Perform the steps of the artificial intelligence-based method for identifying prohibited objects in each of the foregoing embodiments.
  • the method of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be achieved by hardware, but in many cases the former is better. ⁇
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product.
  • the computer software product is stored in a storage medium (such as ROM/RAM), including Several instructions are used to make a terminal (which may be a mobile phone, a computer, a server, or a network device, etc.) execute the method described in each embodiment of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Theoretical Computer Science (AREA)
  • Quality & Reliability (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geophysics (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)

Abstract

本申请涉及基于人工智能的违禁物识别方法、装置、设备和存储介质，方法包括：采集X光图像；从X光图像中划分出与各包装物一一对应的X光子图像；若识别出X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值；对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和；若概率值总和大于违禁物阈值，则判定X光子图像中的包装物内存在违禁物。通过对X光图像中包装物内的各被安检物品进行识别，得到各可疑违禁物品的违禁概率值，并对各违禁概率值进行求和，根据概率值总和判断包装物中是否有违禁物，实现了安检的自动检测，提高了安检效率，降低了人力成本。

Description

一种基于人工智能的违禁物识别方法、装置、设备和存储介质
本申请要求于2019年2月25日提交中国专利局、申请号为201910136323.6、发明名称为“基于人工智能的违禁物识别方法、装置、设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像识别领域，尤其涉及基于人工智能的违禁物识别方法、装置、设备和存储介质。
背景技术
安检机(或称安全检查仪)是一种借助于传送带将被检查行李送入X射线检查通道而完成检查的电子设备。行李进入X射线检查通道后，将阻挡包裹检测传感器；包裹检测传感器发出检测信号，检测信号被送往系统控制部分；系统控制部分产生X射线触发信号，触发X射线源发射X射线束。一束经过准直器的扇形X射线束穿过传送带上的被安检物品，X射线被安检物品吸收，最后轰击安装在通道内的双能量半导体探测器。双能量半导体探测器把X射线转变为信号，这些很弱的信号被放大，并送到信号处理机箱做进一步处理，形成X光视频。X射线是一种可以穿透木材、纸板、皮革等不透明物体的电磁波。安检机能根据物体对X射线的吸收程度，在荧屏上呈现不同颜色的影像。这时，安检员通过荧屏快速查看X射线扫描的X光视频，凭借经验判断是否有违禁物。
但是，发明人意识到通过人工识别X光视频中违禁物的方法并不可靠，因为地铁、机场等场所每天过往的人员量很大，安检人员很难时刻集中精力去观察X光图像，稍不留神就可能漏掉装有可能违禁物的行李。因此，现有的违禁物识别方式存在违禁物识别效率低，且耗费人力成本的问题。
发明内容
本申请的主要目的在于解决现有的违禁物识别方式效率低和人力成本高的技术问题，通过对X光图像中包装物内的各被安检物品进行识别，得到各可疑违禁物品的违禁概率值，并对各违禁概率值进行求和，根据概率值总和判断包装物中是否有违禁物，实现了安检的自动检测，提高了安检效率，降低了人力成本。
一种基于人工智能的违禁物识别方法，包括：以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像；识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像；通过卷积神经网络模型 的卷积层从所述X光子图像中提取目标数据，通过所述卷积神经网络模型的池化层对所述目标数据进行去冗余处理，得到物品特征信息；将所述物品特征信息输入至违禁物识别模型；通过所述违禁物识别模型对所述X光子图像中的包装物内的各被安检物品进行识别，若识别出所述X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值；通过累加器对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和；若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物。
可选地，在所述以预设的采样时间间隔从所述安检机拍摄的X光视频中采集X光图像之前，所述方法还包括：
根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔。
可选地，所述根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔，包括：
获取所述安检机的拍摄区域的长度；获取所述安检机的传送带的传送速度；将所述长度与所述传送速度相除，得到采样时间参考值；将所述采样时间参考值与预设的常数a相乘，得到所述采样时间间隔；所述常数a大于0，且小于或等于1。
可选地，所述识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像，包括：
以所述X光图像中背景区域的灰度值为参考，识别出所述X光图像中的包装物的像素点；分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像。
可选地，所述分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像，包括：
随机提取所述X光图像中任一未被遍历的包装物的第一像素点；所述第一像素点为包装物的任一像素点；遍历所述第一像素点周围的各像素点；若遍历到第二像素点的上、下、左、右四个相邻的像素点中存在所述背景区域的像素点，则确定所述第二像素点为包装物轮廓上的像素点；所述第二像素点为所述第一像素点周围的任一像素点；遍历出所述第一像素点所属的包装物的轮廓上的所有像素点；提取所述第一像素点所属的包装物的轮廓上的所有像素点所围成的区域图像，作为所述第一像素点所属的包装物对应的所述X光子图像。
可选地，在所述若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物之后，所述方法还包括：
发出违禁物提示。
所述发出违禁物提示至少包括以下实现方式之一：发出提示声音；开启提示灯；在所述安检机的荧屏上弹出违禁物提示框；在所述X光图像中标识出违禁物。
可选地，所述违禁物识别模型的表达式为：
Figure PCTCN2019092678-appb-000001
Figure PCTCN2019092678-appb-000002
Figure PCTCN2019092678-appb-000003
Figure PCTCN2019092678-appb-000004
其中，I为输入向量的维度，V为所述X光子图像中经过向量化的物品的维度，H为隐层的神经元个数，K为输出层的神经元个数，x为所述卷积神经网络模型提取出来的所述物品特征信息，v为所述违禁物识别模型对所述物品特征信息识别结果转化成的向量数据，
Figure PCTCN2019092678-appb-000005
为所述违禁物识别模型中隐含层神经元当前时刻的输入，
Figure PCTCN2019092678-appb-000006
为所述违禁物识别模型中隐含层神经元当前时刻的输出；
Figure PCTCN2019092678-appb-000007
为所述违禁物识别模型中输出层神经元当前时刻的输入；
Figure PCTCN2019092678-appb-000008
为所述违禁物识别模型中输出层神经元当前时刻的输出，
Figure PCTCN2019092678-appb-000009
为违禁概率值。
可选地，所述物品特征信息包括物品形状特征信息和物品颜色特征信息。
基于相同的技术构思，本申请还提供了一种基于人工智能的违禁物识别装置，包括收发模块和处理模块。所述处理模块用于控制所述收发模块的收发操作。
所述收发模块，用于以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像。
所述处理模块，用于识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像；通过卷积神经网络模型的卷积层从所述X光子图像中提取目标数据，通过所述卷积神经网络模型的池化层对所述目 标数据进行去冗余处理，得到物品特征信息；将所述物品特征信息输入至违禁物识别模型；通过所述违禁物识别模型对所述X光子图像中的包装物内的各被安检物品进行识别，若识别出所述X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值；通过累加器对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和；若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物。
可选地，所述处理模块具体用于以所述X光图像中背景区域的灰度值为参考，识别出所述X光图像中的包装物的像素点；分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像。
可选地，所述处理模块具体用于随机提取所述X光图像中任一未被遍历的包装物的第一像素点；所述第一像素点为包装物的任一像素点；遍历所述第一像素点周围的各像素点；若遍历到第二像素点的上、下、左、右四个相邻的像素点中存在所述背景区域的像素点，则确定所述第二像素点为包装物轮廓上的像素点；所述第二像素点为所述第一像素点周围的任一像素点；遍历出所述第一像素点所属的包装物的轮廓上的所有像素点；提取所述第一像素点所属的包装物的轮廓上的所有像素点所围成的区域图像，作为所述第一像素点所属的包装物对应的所述X光子图像。
可选地，所述处理模块还用于若判定所述X光子图像中的包装物内存在违禁物，则通过所述收发模块发出违禁物提示。
所述发出违禁物提示至少包括以下实现方式之一：发出提示声音；开启提示灯；在所述安检机的荧屏上弹出违禁物提示框；在所述X光图像中标识出违禁物。
可选地，所述违禁物识别模型的表达式为：
Figure PCTCN2019092678-appb-000010
Figure PCTCN2019092678-appb-000011
Figure PCTCN2019092678-appb-000012
Figure PCTCN2019092678-appb-000013
其中，I为输入向量的维度，V为所述X光子图像中经过向量化的物品的维度，H为隐层的神经元个数，K为输出层的神经元个数，x为所述卷积神经网络模型提取出来的所述物品特征信息，v为所述违禁物识别模型对所述物品特征信息识别结果转化成的向量数据，
Figure PCTCN2019092678-appb-000014
为所述违禁物识别模型中隐含层神经元当前时刻的输入，
Figure PCTCN2019092678-appb-000015
为所述违禁物识别模型中隐含层神经元当前时刻的输出；
Figure PCTCN2019092678-appb-000016
为所述违禁物识别模型中输出层神经元当前时刻的输入；
Figure PCTCN2019092678-appb-000017
为所述违禁物识别模型中输出层神经元当前时刻的输出，
Figure PCTCN2019092678-appb-000018
为违禁概率值。
可选地，所述物品特征信息包括物品形状特征信息和物品颜色特征信息。
基于相同的技术构思，本申请还提供了一种计算机设备，包括收发器、存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行如上述的基于人工智能的违禁物识别方法的步骤。
基于相同的技术构思，本申请还提供了一种存储有非易失性计算机可读指令的存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行如上述基于人工智能的违禁物识别方法的步骤。
本申请的有益效果：通过对X光图像中包装物内的各被安检物品进行识别，得到各可疑违禁物品的违禁概率值，并对各违禁概率值进行求和，根据概率值总和判断包装物中是否有违禁物，实现了安检的自动检测，提高了安检效率，降低了人力成本。
附图说明
图1为本申请实施例中基于人工智能的违禁物识别方法的流程示意图。
图2为图1中步骤S2的流程示意图。
图3为本申请实施例中基于人工智能的违禁物识别装置的结构示意图。
图4为本申请实施例中计算机设备的结构示意图。
具体实施方式
下面详细描述本申请的实施例，所述实施例的示例在附图中示出，其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的，仅用于解释本申请，而不能解释为对本申请的限制。
图1为本申请一些实施方式中一种基于人工智能的违禁物识别方法的流程图，该违禁物识别方法由违禁物识别装置执行，违禁物识别装置可以集成在安检机内，也可以与安检机电气连接，如图1所示，可以包括如下步骤S1-S6：
S1、以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像。
以预设的采样时间间隔对安检机拍摄的X光视频进行采样，获取X光图像。X光图像携带了物品特征信息；根据物品特征信息对X光图像中被安检物品进行识别，识别出被安检物品是否是可疑的违禁物。
一些实施方式中，在步骤S1之前，所述方法还包括步骤S11：根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔。
由于各被安检物品是在安检机的传送带带动下匀速通过安检机拍摄区域的，因此，被安检物品通过安检机的拍摄区域所需要的时间是不变的，也就是说各被安检物品在X光视频中的显现时间是不变且相同的。为了能采集到X光视频中每个被安检物品的X光图像，需要合理设定采样时间间隔。假设通过安检机的拍摄区域的时间为5秒；若采样时间间隔大于5秒，例如设置为7秒或8秒，这样，由于采集X光图像的时间大于被安检物品在X光视频中的显现时间，则导致无法从X光视频中采集到某些被安检物品的X光图像，发生漏检情况；若采样时间间隔过小，例如设置为1秒，这样，由于采集X光图像的时间小于被安检物品在X光视频中的显现时间，则导致含有同一个被安检物品在X光视频5秒的显现时间内，其X光图像被采集了5次，也就意味着同一个被安检物品被做了5次安检识别，无疑增加了安检过程中的数据处理运算量(其中包括多次不必要的重复运算)，导致无法及时反馈被安检物品的X光图像，甚至出现被安检物品已经通过传送带且被用户带走，而未来记得及拦下可疑物品。如果将采样时间间隔设定为4秒，这样可以在X光视频5秒的显现时间内，基本上可以保证X光视频中每个被安检物品都只被采集了一次X光图像，且不会发生漏检的情况。因此，采样时间间隔不能大于被安检物品在X光视频中所显现的时间，且采样时间间隔不宜太短。当然，实际应用中，X光图像的采样时间间隔可根据实际需要进行调试设定。
由于安检机内的X射线检查设备的位置是固定的，所以X光视频的所能拍摄的区域范围也是固定的。被安检物品经过X光视频所拍摄的区域范围所需要的时间可以由安检机的传送带的传送速度及传送带在X光图像中所能显现出来的部位的长度计算求得。
一些实施方式中，步骤S11包括以下步骤S111-S114：
S111、获取所述安检机的拍摄区域的长度。
安检机的拍摄区域也就是X光视频所显现的区域范围。确定了安检机的拍摄区域的长度，也就确定了被安检物品通过安检机的拍摄区域的实际路程。
S112、获取所述安检机的传送带的传送速度。
传送带的传送速度是可以预先设置的，获取了传送带的传送速度，也就得到了被安检物品通过安检机的拍摄区域的速度。
S113、将所述长度与所述传送速度相除，得到采样时间参考值。
将所述长度除以所述传送速度，便可得到所述采样时间参考值。所述采样时间参考值表示了被安检物品通过安检机的拍摄区域所需要的时间。
S114、将所述采样时间参考值与预设的常数a相乘，得到所述采样时间间隔。
所述常数a大于0，且小于或等于1。所述采样时间间隔小于或等于所述采样时间参考值。
根据采样时间参考值设定合理的采样时间间隔，使得同一个被安检物品在过安检机过程中只被采集一次X光图像，且不会发生漏检的情况。
本实施例中，在获取传送带的传送速度及安检机的拍摄区域的长度后，可以自动计算并设置X光图像的采样时间间隔，使得本方法的X光图像的采样时间间隔与被安检物品的移动速度相适应，省去手动设置采样时间间隔的麻烦。
S2、识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像。
包装物用于容纳被安检物品。
一帧X光图像中可能会有多个乘客的被安检物品，每个乘客的被安检物品一般是放在各自的包装物中，安检机则是对一个个包装物内的物品进行安检的。安检时，一般都是多个包装物陆续过安检机的，因此一帧X光图像中可能有多个包装物。在传统的人工监视X光视频的方式中，安检人员辨识出X光视频的包装物中有可疑的违禁物，然后对含有可疑的违禁物包装物进行开包确认及检查。安检人员实际上是对X光视频中的包装物逐一进行辨识。同理，本申请要确定哪一个乘客的包装物中含有可疑的违禁物，就需要先将各包装物区分开来，然后分别识别各包装物中是否含有可疑的违禁物，当发现某个包装物藏有可疑的违禁物时，只需对藏有可疑的违禁物的包装物做进一步检查就可以了。
包装物包括如桶、箱、筐、瓶、坛、罐、袋等用于容纳物品的装置。需要理解的是，某些单独过安检机的物品，如仪器设备等，虽然它本身不是放包装物中的，但其内部可能容纳有违禁物，因此单独过安检机的物品也视为包装物。
一些实施方式中，如图2所示，步骤S2包括以下步骤S21-S22：
S21、以所述X光图像中背景区域的灰度值为参考，识别出所述X光图像中的包装物的像素点。
X光视频所显示的背景画面是固定的，因此，X光图像中的背景区域的各像素点的灰度值也是固定的。X光图像中的背景区域与包装物灰度值是不同的，因此X光图像中的背景区域与包装物是容易区分开的。将X光图像中背景区域的各像素点标注为逻辑0，将X光图像中属于包装物各像素点标注为逻辑1，用以区分背景区域的各像素点和属于包装物各像素点。
S22、分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像。
在过安检机的过程中，前后的包装物之间一般会留有一定的间隔，因此，X光图像中的背景区域将各包装物分割开来。找到各包装物与背景区域的边沿位置，也就划分出了X光图像中的各包装物的区域，即得到与各包装物一一对应的X光子图像，然后对X光子图像中的包装物进行违禁物识别。
一些实施方式中，步骤S22包括以下步骤S221-S223：
S221、随机提取所述X光图像中任一未被遍历的包装物的第一像素点。
所述第一像素点为包装物的任一像素点。
本申请中采用随机游走的方法，对X光图像进行遍历。首先在X光图像的矩阵中随机获取任一未被遍历的包装物的第一像素点，作为遍历X光图像中包装物起始像素点。从起始像素点位置开始对其所在的被遍历的包装物内的各像素点进行遍历。
S222、遍历所述起始像素点周围的各像素点；若遍历到第二像素点的上、下、左、右四个相邻的像素点中存在所述背景区域的像素点，则确定所述第二像素点为包装物轮廓上的像素点；所述第二像素点为所述第一像素点周围的任一像素点；遍历出所述第一像素点所属的包装物的轮廓上的所有像素点。
识别包装物内某一像素点旁边的像素点是否存在背景区域的像素点，若存在，则说明该像素点在包装物边沿处。采用该方式对包装物内的像素点进行遍历，直到遍历到包装物所有方向都是背景区域。
S223、提取所述第一像素点所属的包装物的轮廓上的所有像素点所围成的区域图像，作为所述第一像素点所属的包装物对应的所述X光子图像。
当遍历到包装物所有方向都是背景区域时，也就确定了属于该包装物所有像素点的区 域，也就确定了该包装物X光子图像。X光图像中一个包装物被遍历后，返回步骤S221，找到其它未被遍历的包装物，继续对该未被遍历的包装物进行提取，直到将X光图像中的各包装物X光子图像一一划分出来。
S3、通过卷积神经网络模型的卷积层从所述X光子图像中提取目标数据，通过所述卷积神经网络模型的池化层对所述目标数据进行去冗余处理，得到物品特征信息；将所述物品特征信息输入至违禁物识别模型；通过所述违禁物识别模型对所述X光子图像中的包装物内的各被安检物品进行识别，若识别出所述X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值。
所述物品特征信息包括物品形状特征信息和物品颜色特征信息等。
预先设置3*3宽度的卷积神经网络模型，卷积神经网络模型从X光子图像的起始像素点的位置开始，以1个像素点为步幅，逐步对X光子图像的数据进行遍历，运行卷积运算，提取X光子图像中的物品特征信息。卷积神经网络模型将物品特征信息拼接成连续的数据，并将拼接后的物品特征信息输入至违禁物识别模型。
一些实施方式中，违禁物识别模型的表达式为：
Figure PCTCN2019092678-appb-000019
Figure PCTCN2019092678-appb-000020
Figure PCTCN2019092678-appb-000021
Figure PCTCN2019092678-appb-000022
其中，I为输入向量的维度，V为所述X光子图像中经过向量化的物品的维度，H为隐层的神经元个数，K为输出层的神经元个数，x为所述卷积神经网络模型提取出来的所述物品特征信息，v为所述违禁物识别模型对所述物品特征信息识别结果转化成的向量数据，
Figure PCTCN2019092678-appb-000023
为所述违禁物识别模型中隐含层神经元当前时刻的输入，
Figure PCTCN2019092678-appb-000024
为所述违禁物识别模型中隐含层神经元当前时刻的输出；
Figure PCTCN2019092678-appb-000025
为所述违禁物识别模型中输出层神经元当前时刻的输 入；
Figure PCTCN2019092678-appb-000026
为所述违禁物识别模型中输出层神经元当前时刻的输出，
Figure PCTCN2019092678-appb-000027
为违禁概率值。
预先对违禁物识别模型进行训练。具体地，从X光视频中采集一定量的训练用X光图像，训练用X光图像中包含违禁物的特征信息。人工标注出各训练用X光图像中的违禁物，并为标注的各违禁物分配对应的违禁概率值。将做好人工标注的训练用X光图像的数据输入到违禁物识别模型。违禁物识别模型通过对训练用X光图像所标注的违禁物的特征信息进行记忆，记住识别各违禁物的各特征信息的权重值，从而实现对各违禁物的准确识别，并输出各违禁物对应的预设违禁概率值。用于识别物品的模型有很多，可以采用卷积神经网络和循环神经网络实现，卷积神经网络对X光图像中的物品特征信息进行提取，将提取的物品特征信息发送给循环神经网络；循环神经网络识别根据物品特征信息对X光图像内的违禁物进行识别。识别模型的训练大体原理一致，在此不再累述。
不同的物品为违禁物的可能性不一样，例如，包装物中的水瓶，仅凭包装物中有水瓶不足以判定水瓶内有易燃、易爆的液体，但也有装有易燃、易爆液体的可能性。因此，为疑似水瓶的物品设置一个对应的违禁概率值。为了更准确的判断出包装物中是否存在为违禁物，建立多种违禁物的识别概率列表，采用对各可疑违禁物分配概率值的方式进行违禁物识别。
违禁物不一定是某种完整的物品，也可以是某个完整的物品的某个部位。为了更准确的判断物品是否为违禁物，在训练违禁物识别模型时，可以将一个完整的物品划分成多个具体部位，并为各部位分配一对应的违禁概率值。例如，包装物内的条状物有可能是违禁的刀具，也可能是其它正常物品，为类似于违禁的刀具的条状物分配一对应的违禁概率值；违禁的棍棒都会有适于把持的手柄物，为类似于违禁刀具的手柄物分配一对应的违禁概率值；将类似于违禁的刀具的条状物对应的违禁概率值与类似于违禁刀具的手柄物对应的违禁概率值进行叠加，所得到的概率值，更能准确的反映所识别的整体物品是否为违禁物。对物品的部位识别的越具体，识别的准确率就越高。
S4、通过累加器对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和。
预先设置累加器，通过累加器对包装物中各可能的违禁物对应的违禁概率值进行求和，得到的概率值总和越高，说明包装物内含有违禁物的可能性越大。
S5、若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物。
很低的违禁概率值不足以准确判断包装物内存在违禁物，很可能是对与违禁物相似的非违禁物进行识别所得到的违禁概率值。设定一个违禁物阈值，比较所述概率值总和与预 设的违禁物阈值，当各违禁概率值之和超过预设的违禁物阈值时，才判定包装物中含有违禁物，这样可以更准确的识别出包装物是否含有违禁物。
S6、若判定所述X光子图像中的包装物内存在违禁物，则发出违禁物提示。
一些实施方式中，所述发出违禁物提示至少包括以下实现方式之一：发出提示声音；开启提示灯；在所述安检机的荧屏上弹出违禁物提示框；在所述X光图像中标识出违禁物。
上述实施例中，通过灰度值识别出X光图像中的背景区域与包装物，并从X光图像中划分出各包装物对应的X光子图像，通过违禁物识别模型对X光子图像中各包装物内的被安检物品进行识别，识别出各包装物内可能存在的违禁物，并为各可疑的违禁物分配对应的违禁概率值，并对各违禁概率值进行求和，得到概率值总和，将概率值总和与预设的违禁物阈值进行比较，准确判断违禁物内是否存在违禁物。降低了安检过程中的人力成本，提高了安检的效率。
基于相同的技术构思，本申请还提供了一种基于人工智能的违禁物识别装置，如图3所示，该装置包括收发模块1和处理模块2。所述处理模块2用于控制所述收发模块1的收发操作。
所述收发模块1，用于以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像。
所述处理模块2，用于识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像；通过卷积神经网络模型的卷积层从所述X光子图像中提取目标数据，通过所述卷积神经网络模型的池化层对所述目标数据进行去冗余处理，得到物品特征信息；将所述物品特征信息输入至违禁物识别模型；通过所述违禁物识别模型对所述X光子图像中的包装物内的各被安检物品进行识别，若识别出所述X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值；通过累加器对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和；若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物。
一些实施方式中，所述处理模块2具体用于根据像素点的灰度值识别出所述X光图像中背景区域的各像素点和属于包装物各像素点；分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像。
一些实施方式中，所述处理模块2具体用于随机提取所述X光图像中任一未被遍历的 包装物一个像素点，将随机提取的该像素点作为起始像素点；遍历所述起始像素点周围的各像素点；判断所遍历的像素点的上、下、左、右四个相邻的像素点中是否存在所述背景区域的像素点；若所遍历的像素点的上、下、左、右四个相邻的像素点中存在所述背景区域的像素点，则判定该所遍历的像素点为包装物轮廓上的像素点；遍历出所述起始像素点所属的包装物轮廓上的所有像素点；提取所述起始像素点所属的包装物轮廓上的所有像素点所围成的区域图像，作为所述起始像素点所属的包装物所述X光子图像。
一些实施方式中，所述处理模块2还用于若判定所述X光子图像中的包装物内存在违禁物，则通过所述收发模块1发出违禁物提示。
一些实施方式中，违禁物识别模型的表达式为：
Figure PCTCN2019092678-appb-000028
Figure PCTCN2019092678-appb-000029
Figure PCTCN2019092678-appb-000030
Figure PCTCN2019092678-appb-000031
其中，I为输入向量的维度，V为所述X光子图像中经过向量化的物品的维度，H为隐层的神经元个数，K为输出层的神经元个数，x为所述卷积神经网络模型提取出来的所述物品特征信息，v为所述违禁物识别模型对所述物品特征信息识别结果转化成的向量数据，
Figure PCTCN2019092678-appb-000032
为所述违禁物识别模型中隐含层神经元当前时刻的输入，
Figure PCTCN2019092678-appb-000033
为所述违禁物识别模型中隐含层神经元当前时刻的输出；
Figure PCTCN2019092678-appb-000034
为所述违禁物识别模型中输出层神经元当前时刻的输入；
Figure PCTCN2019092678-appb-000035
为所述违禁物识别模型中输出层神经元当前时刻的输出，
Figure PCTCN2019092678-appb-000036
为违禁概率值。
一些实施方式中，所述物品特征信息包括物品形状特征信息和物品颜色特征信息。
上述实施例中，通过灰度值识别出X光图像中的背景区域与包装物，并从X光图像中划分出各包装物对应的X光子图像，通过违禁物识别模型对X光子图像中各包装物内的被安检物品进行识别，识别出各包装物内可能存在的违禁物，并为各可疑的违禁物分配对应 的违禁概率值，并对各违禁概率值进行求和，得到概率值总和，将概率值总和与预设的违禁物阈值进行比较，准确判断违禁物内是否存在违禁物。降低了安检过程中的人力成本，提高了安检的效率。
基于相同的技术构思，本申请还提供了一种计算机设备，如图4所示，该计算机设备包括收发器901、处理器902和存储器903，所述存储器903中存储有计算机可读指令，所述计算机可读指令被所述处理器902执行时，使得所述处理器902执行上述各实施方式中的所述的基于人工智能的违禁物识别方法的步骤。
图3中所示的收发模块1对应的实体设备为图4所示的收发器901，该收发器901能够实现收发模块1部分或全部，且相同或相似的功能。
图3中所示的处理模块2对应的实体设备为图4所示的处理器902，该处理器902能够实现处理模块2部分或全部，且相同或相似的功能。
基于相同的技术构思，本申请还提供了一种存储有非易失性计算机可读指令的存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行上述各实施方式中的所述的基于人工智能的违禁物识别方法的步骤。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM)中，包括若干指令用以使得一台终端(可以是手机，计算机，服务器或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述，但是本申请并不局限于上述的具体实施方式，上述的具体实施方式仅仅是示意性的，而不是限制性的，本领域的普通技术人员在本申请的启示下，在不脱离本申请宗旨和权利要求所保护的范围情况下，还可做出很多形式，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，这些均属于本申请的保护之内。

Claims (20)

  1. 一种基于人工智能的违禁物识别方法，包括：
    以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像；
    识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像；
    通过卷积神经网络模型的卷积层从所述X光子图像中提取目标数据，通过所述卷积神经网络模型的池化层对所述目标数据进行去冗余处理，得到物品特征信息；将所述物品特征信息输入至违禁物识别模型；通过所述违禁物识别模型对所述X光子图像中的包装物内的各被安检物品进行识别，若识别出所述X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值；
    通过累加器对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和；
    若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物。
  2. 根据权利要求1所述的基于人工智能的违禁物识别方法，
    在所述以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像之前，所述方法还包括：
    根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔。
  3. 根据权利要求2所述的基于人工智能的违禁物识别方法，
    所述根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔，包括：
    获取所述安检机的拍摄区域的长度；
    获取所述安检机的传送带的传送速度；
    将所述长度与所述传送速度相除，得到采样时间参考值；
    将所述采样时间参考值与预设的常数a相乘，得到所述采样时间间隔；所述常数a大于0，且小于或等于1。
  4. 根据权利要求1所述的基于人工智能的违禁物识别方法，
    所述识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像，包括：
    以所述X光图像中背景区域的灰度值为参考，识别出所述X光图像中的包装物的像素点；
    分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像。
  5. 根据权利要求4所述的基于人工智能的违禁物识别方法，
    所述分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像，包括：
    随机提取所述X光图像中任一未被遍历的包装物的第一像素点；所述第一像素点为包装物的任一像素点；
    遍历所述第一像素点周围的各像素点；若遍历到第二像素点的上、下、左、右四个相邻的像素点中存在所述背景区域的像素点，则确定所述第二像素点为包装物轮廓上的像素点；所述第二像素点为所述第一像素点周围的任一像素点；遍历出所述第一像素点所属的包装物的轮廓上的所有像素点；
    提取所述第一像素点所属的包装物的轮廓上的所有像素点所围成的区域图像，作为所述第一像素点所属的包装物对应的所述X光子图像。
  6. 根据权利要求1至5任一所述的基于人工智能的违禁物识别方法，
    在所述若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物之后，所述方法还包括：
    发出违禁物提示；
    所述发出违禁物提示至少包括以下实现方式之一：
    发出提示声音；
    开启提示灯；
    在所述安检机的荧屏上弹出违禁物提示框；
    在所述X光图像中标识出违禁物。
  7. 根据权利要求1至5任一所述的基于人工智能的违禁物识别方法，
    所述违禁物识别模型的表达式为：
    Figure PCTCN2019092678-appb-100001
    Figure PCTCN2019092678-appb-100002
    Figure PCTCN2019092678-appb-100003
    Figure PCTCN2019092678-appb-100004
    其中，I为输入向量的维度，V为所述X光子图像中经过向量化的物品的维度，H为隐层的神经元个数，K为输出层的神经元个数，x为所述卷积神经网络模型提取出来的所述物品特征信息，v为所述违禁物识别模型对所述物品特征信息识别结果转化成的向量数据，
    Figure PCTCN2019092678-appb-100005
    为所述违禁物识别模型中隐含层神经元当前时刻的输入，
    Figure PCTCN2019092678-appb-100006
    为所述违禁物识别模型中隐含层神经元当前时刻的输出；
    Figure PCTCN2019092678-appb-100007
    为所述违禁物识别模型中输出层神经元当前时刻的输入；
    Figure PCTCN2019092678-appb-100008
    为所述违禁物识别模型中输出层神经元当前时刻的输出，
    Figure PCTCN2019092678-appb-100009
    为违禁概率值。
  8. 一种基于人工智能的违禁物识别装置，包括：
    收发模块，用于以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像；
    处理模块，用于识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像；通过卷积神经网络模型的卷积层从所述X光子图像中提取目标数据，通过所述卷积神经网络模型的池化层对所述目标数据进行去冗余处理，得到物品特征信息；将所述物品特征信息输入至违禁物识别模型；通过所述违禁物识别模型对所述X光子图像中的包装物内的各被安检物品进行识别，若识别出所述X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值；通过累加器对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和；若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物。
  9. 根据权利要求8所述的基于人工智能的违禁物识别装置，
    在所述以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像之前，所述处理模块还用于：
    根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔。
  10. 根据权利要求9所述的基于人工智能的违禁物识别装置，
    所述根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔，包括：
    获取所述安检机的拍摄区域的长度；
    获取所述安检机的传送带的传送速度；
    将所述长度与所述传送速度相除，得到采样时间参考值；
    将所述采样时间参考值与预设的常数a相乘，得到所述采样时间间隔；所述常数a大于0，且小于或等于1。
  11. 根据权利要求8所述的基于人工智能的违禁物识别装置，
    所述识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像，包括：
    以所述X光图像中背景区域的灰度值为参考，识别出所述X光图像中的包装物的像素点；
    分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像。
  12. 根据权利要求1 1所述的基于人工智能的违禁物识别装置，
    所述分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像，包括：
    随机提取所述X光图像中任一未被遍历的包装物的第一像素点；所述第一像素点为包装物的任一像素点；
    遍历所述第一像素点周围的各像素点；若遍历到第二像素点的上、下、左、右四个相邻的像素点中存在所述背景区域的像素点，则确定所述第二像素点为包装物轮廓上的像素点；所述第二像素点为所述第一像素点周围的任一像素点；遍历出所述第一像素点所属的包装物的轮廓上的所有像素点；
    提取所述第一像素点所属的包装物的轮廓上的所有像素点所围成的区域图像，作为所述第一像素点所属的包装物对应的所述X光子图像。
  13. 一种基于人工智能的违禁物识别的装置，所述装置包括：
    至少一个处理器、存储器和输入输出单元；
    其中，所述存储器用于存储程序代码，所述处理器用于调用所述存储器中存储的程序代码来执行以下操作：
    以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像；
    识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像；
    通过卷积神经网络模型的卷积层从所述X光子图像中提取目标数据，通过所述卷积神经网络模型的池化层对所述目标数据进行去冗余处理，得到物品特征信息；将所述物品特征信息输入至违禁物识别模型；通过所述违禁物识别模型对所述X光子图像中的包装物内的各被安检物品进行识别，若识别出所述X光子图像中的包装物内存在可疑的违禁物，则为各可疑的违禁物匹配对应的预设的违禁概率值；
    通过累加器对各可疑的违禁物对应的违禁概率值进行叠加，得到概率值总和；
    若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物。
  14. 根据权利要求13所述的装置，所述处理器具体执行以下操作：
    在所述以预设的采样时间间隔从安检机拍摄的X光视频中采集X光图像之前，所述方法还包括：
    根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔。
  15. 根据权利要求14所述的装置，所述处理器具体执行以下操作：
    所述根据被安检物品通过所述安检机的拍摄区域所需要的时间设定所述采样时间间隔，包括：
    获取所述安检机的拍摄区域的长度；
    获取所述安检机的传送带的传送速度；
    将所述长度与所述传送速度相除，得到采样时间参考值；
    将所述采样时间参考值与预设的常数a相乘，得到所述采样时间间隔；所述常数a大于0，且小于或等于1。
  16. 根据权利要求13所述的装置，所述处理器具体执行以下操作：
    所述识别所述X光图像中的各包装物所在区域，根据各包装物所在区域从所述X光图像中划分出与各包装物一一对应的X光子图像，包括：
    以所述X光图像中背景区域的灰度值为参考，识别出所述X光图像中的包装物的像素点；
    分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像。
  17. 根据权利要求16所述的处理器，所述处理器具体执行以下操作：
    所述分别识别出所述X光图像中的各包装物轮廓上的所有像素点，根据各包装物轮廓上的所有像素点从所述X光图像中划分出与各包装物一一对应的所述X光子图像，包括：
    随机提取所述X光图像中任一未被遍历的包装物的第一像素点；所述第一像素点为包装物的任一像素点；
    遍历所述第一像素点周围的各像素点；若遍历到第二像素点的上、下、左、右四个相邻的像素点中存在所述背景区域的像素点，则确定所述第二像素点为包装物轮廓上的像素点；所述第二像素点为所述第一像素点周围的任一像素点；遍历出所述第一像素点所属的包装物的轮廓上的所有像素点；
    提取所述第一像素点所属的包装物的轮廓上的所有像素点所围成的区域图像，作为所述第一像素点所属的包装物对应的所述X光子图像。
  18. 根据权利要求13-17中任意一项所述的装置，所述处理器具体执行以下操作：
    在所述若所述概率值总和大于所述违禁物阈值，则判定所述X光子图像中的包装物内存在违禁物之后，所述方法还包括：
    发出违禁物提示；
    所述发出违禁物提示至少包括以下实现方式之一：
    发出提示声音；
    开启提示灯；
    在所述安检机的荧屏上弹出违禁物提示框；
    在所述X光图像中标识出违禁物。
  19. 根据权利要求13-17中任意一项所述的装置，
    所述违禁物识别模型的表达式为：
    Figure PCTCN2019092678-appb-100010
    Figure PCTCN2019092678-appb-100011
    Figure PCTCN2019092678-appb-100012
    Figure PCTCN2019092678-appb-100013
    其中，I为输入向量的维度，V为所述X光子图像中经过向量化的物品的维度，H为隐层的神经元个数，K为输出层的神经元个数，x为所述卷积神经网络模型提取出来的所述物品特征信息，v为所述违禁物识别模型对所述物品特征信息识别结果转化成的向量数据，
    Figure PCTCN2019092678-appb-100014
    为所述违禁物识别模型中隐含层神经元当前时刻的输入，
    Figure PCTCN2019092678-appb-100015
    为所述违禁物识别模型中隐含层神经元当前时刻的输出；
    Figure PCTCN2019092678-appb-100016
    为所述违禁物识别模型中输出层神经元当前时刻的输入；
    Figure PCTCN2019092678-appb-100017
    为所述违禁物识别模型中输出层神经元当前时刻的输出，
    Figure PCTCN2019092678-appb-100018
    为违禁概率值。
  20. 一种存储有非易失性计算机可读指令的存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行如权利要求1至7中的任一所述的基于人工智能的违禁物识别方法的步骤。
PCT/CN2019/092678 2019-02-25 2019-06-25 一种基于人工智能的违禁物识别方法、装置、设备和存储介质 Ceased WO2020173021A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910136323.6A CN109978827A (zh) 2019-02-25 2019-02-25 基于人工智能的违禁物识别方法、装置、设备和存储介质
CN201910136323.6 2019-02-25

Publications (1)

Publication Number Publication Date
WO2020173021A1 true WO2020173021A1 (zh) 2020-09-03

Family

ID=67077268

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/092678 Ceased WO2020173021A1 (zh) 2019-02-25 2019-06-25 一种基于人工智能的违禁物识别方法、装置、设备和存储介质

Country Status (2)

Country Link
CN (1) CN109978827A (zh)
WO (1) WO2020173021A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112184676A (zh) * 2020-09-30 2021-01-05 中云智慧（北京）科技有限公司 一种从x光机视频流中获取包裹图像的方法
CN113554598A (zh) * 2021-06-24 2021-10-26 浙江大华技术股份有限公司 物品检测方法、装置、电子设备、计算机可读存储介质
CN114255436A (zh) * 2020-09-11 2022-03-29 同方威视技术股份有限公司 基于人工智能的安检图像识别系统及其方法
CN114758239A (zh) * 2022-04-22 2022-07-15 安徽工业大学科技园有限公司 基于机器视觉的物品飞离预定行进线路的监测方法和系统
CN114994100A (zh) * 2022-06-06 2022-09-02 河南双晟智能科技有限公司 一种自助式铁路旅客安检系统及其检测方法
CN118330760A (zh) * 2024-04-07 2024-07-12 北京安核技术有限公司 智能化安检方法和系统
CN119339130A (zh) * 2024-09-24 2025-01-21 杭州睿影科技有限公司 安检机扫描条带数据的处理方法、装置、设备及存储介质
WO2025035728A1 (zh) * 2023-08-15 2025-02-20 同方威视技术股份有限公司 安检机操作方法、系统、装置、设备、介质和程序产品

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112016387A (zh) * 2019-07-08 2020-12-01 杭州芯影科技有限公司 适用于毫米波安检仪的违禁品识别方法和装置
CN110543857A (zh) * 2019-09-05 2019-12-06 安徽启新明智科技有限公司 基于图像分析的违禁物识别方法、装置、系统和存储介质
CN110751079A (zh) * 2019-10-16 2020-02-04 北京海益同展信息科技有限公司 物品检测方法、装置、系统和计算机可读存储介质
CN110852248A (zh) * 2019-11-07 2020-02-28 江苏弘冉智能科技有限公司 一种基于机器视觉的易燃易爆区域违禁设备及动作监测方法
CN111062252B (zh) * 2019-11-15 2023-11-10 浙江大华技术股份有限公司 一种实时危险物品语义分割方法、装置及存储装置
CN111325114B (zh) * 2020-02-03 2022-07-19 重庆特斯联智慧科技股份有限公司 一种人工智能识别分类的安检图像处理方法和装置
CN111290040A (zh) * 2020-03-12 2020-06-16 安徽启新明智科技有限公司 一种基于图像识别的主动式双视角关联的方法
CN112364903A (zh) * 2020-10-30 2021-02-12 盛视科技股份有限公司 基于x光机的物品分析及多维图像关联方法与系统
CN114689358A (zh) * 2020-12-31 2022-07-01 清华大学 违禁品检测机器人及检测方法
CN113792665B (zh) * 2021-09-16 2023-08-08 山东大学 一种针对不同角色权限的违禁区域侵入检测方法
CN114758259B (zh) * 2022-06-15 2022-09-06 科大天工智能装备技术（天津）有限公司 一种基于x光物体图像识别的包裹检测方法及系统
CN120510557B (zh) * 2025-07-22 2025-11-04 北京智芯微电子科技有限公司 视频处理方法、系统、装置、设备及可读存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9547821B1 (en) * 2016-02-04 2017-01-17 International Business Machines Corporation Deep learning for algorithm portfolios
CN106485268A (zh) * 2016-09-27 2017-03-08 东软集团股份有限公司 一种图像识别方法及装置
CN107871122A (zh) * 2017-11-14 2018-04-03 深圳码隆科技有限公司 安检检测方法、装置、系统及电子设备
CN108198227A (zh) * 2018-03-16 2018-06-22 济南飞象信息科技有限公司 基于x光安检机图像的违禁品智能识别方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105654130A (zh) * 2015-12-30 2016-06-08 成都数联铭品科技有限公司 一种基于递归神经网络的复杂图像文字序列识别系统
PL3764281T3 (pl) * 2016-02-22 2025-02-10 Rapiscan Systems, Inc. Sposoby identyfikacji broni palnej na obrazach radiograficznych
CN107607562B (zh) * 2017-09-11 2020-06-02 北京匠数科技有限公司 一种违禁物品识别设备及方法、x光行李安检系统
CN108665509A (zh) * 2018-05-10 2018-10-16 广东工业大学 一种超分辨率重构方法、装置、设备及可读存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9547821B1 (en) * 2016-02-04 2017-01-17 International Business Machines Corporation Deep learning for algorithm portfolios
CN106485268A (zh) * 2016-09-27 2017-03-08 东软集团股份有限公司 一种图像识别方法及装置
CN107871122A (zh) * 2017-11-14 2018-04-03 深圳码隆科技有限公司 安检检测方法、装置、系统及电子设备
CN108198227A (zh) * 2018-03-16 2018-06-22 济南飞象信息科技有限公司 基于x光安检机图像的违禁品智能识别方法

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114255436A (zh) * 2020-09-11 2022-03-29 同方威视技术股份有限公司 基于人工智能的安检图像识别系统及其方法
CN114255436B (zh) * 2020-09-11 2024-03-19 同方威视技术股份有限公司 基于人工智能的安检图像识别系统及其方法
CN112184676A (zh) * 2020-09-30 2021-01-05 中云智慧（北京）科技有限公司 一种从x光机视频流中获取包裹图像的方法
CN113554598A (zh) * 2021-06-24 2021-10-26 浙江大华技术股份有限公司 物品检测方法、装置、电子设备、计算机可读存储介质
CN114758239A (zh) * 2022-04-22 2022-07-15 安徽工业大学科技园有限公司 基于机器视觉的物品飞离预定行进线路的监测方法和系统
CN114758239B (zh) * 2022-04-22 2024-06-04 安徽工业大学科技园有限公司 基于机器视觉的物品飞离预定行进线路的监测方法和系统
CN114994100A (zh) * 2022-06-06 2022-09-02 河南双晟智能科技有限公司 一种自助式铁路旅客安检系统及其检测方法
WO2025035728A1 (zh) * 2023-08-15 2025-02-20 同方威视技术股份有限公司 安检机操作方法、系统、装置、设备、介质和程序产品
CN118330760A (zh) * 2024-04-07 2024-07-12 北京安核技术有限公司 智能化安检方法和系统
CN119339130A (zh) * 2024-09-24 2025-01-21 杭州睿影科技有限公司 安检机扫描条带数据的处理方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN109978827A (zh) 2019-07-05

Similar Documents

Publication Publication Date Title
WO2020173021A1 (zh) 一种基于人工智能的违禁物识别方法、装置、设备和存储介质
CN111145177B (zh) 图像样本生成方法、特定场景目标检测方法及其系统
US8781066B2 (en) Method and apparatus for assessing characteristics of liquids
CN104636707B (zh) 自动检测香烟的方法
CN108303747B (zh) 检查设备和检测枪支的方法
CN110751079A (zh) 物品检测方法、装置、系统和计算机可读存储介质
CN111612020B (zh) 一种异常被检物的定位方法以及安检分析设备、系统
CN106874845B (zh) 图像识别的方法和装置
CN110119734A (zh) 刀具检测方法及装置
EP3349049B1 (en) Inspection devices and methods for inspecting a container
CN103744120A (zh) 一种违禁品辅助鉴别方法及装置
WO2021087962A1 (zh) 一种限制品自动识别装置及方法
CN113706497B (zh) 一种违禁品智能识别装置和系统
CN110850493A (zh) 一种扫描成像机器视觉判图的线阵列高速安检机
CN111382725A (zh) 违禁快递包裹的处理方法、装置、设备及存储介质
CN106872498A (zh) 一种违禁品安检自动识别装置
CN107999413A (zh) 一种高速x光快递分拣方法
CN114462487A (zh) 目标检测网络训练及检测方法、装置、终端及存储介质
CN110427898B (zh) 包裹安检识别方法、系统、装置及计算机可读存储介质
Rogers et al. Detection of cargo container loads from X-ray images
CN114548230A (zh) 基于rgb色彩分离双路特征融合的x射线违禁物品检测方法
KR20220077528A (ko) Ｘ-ｒａｙ 영상 다중 판독 시스템 및 그 방법
CN113160341A (zh) 一种含有目标物品的x光图像生成方法、系统及设备
Vekariya et al. A survey on oil spill detection using SAR images in machine learning
CN116563628A (zh) 安检判图识别方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19917276

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19917276

Country of ref document: EP

Kind code of ref document: A1