WO2021087962A1 - 一种限制品自动识别装置及方法 - Google Patents

一种限制品自动识别装置及方法 Download PDF

Info

Publication number
WO2021087962A1
WO2021087962A1 PCT/CN2019/116645 CN2019116645W WO2021087962A1 WO 2021087962 A1 WO2021087962 A1 WO 2021087962A1 CN 2019116645 W CN2019116645 W CN 2019116645W WO 2021087962 A1 WO2021087962 A1 WO 2021087962A1
Authority
WO
WIPO (PCT)
Prior art keywords
package
conveyor belt
identification device
identified
restricted
Prior art date
Application number
PCT/CN2019/116645
Other languages
English (en)
French (fr)
Inventor
胡庆茂
张伟烽
张晓东
袁权
Original Assignee
中国科学院深圳先进技术研究院
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 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Priority to PCT/CN2019/116645 priority Critical patent/WO2021087962A1/zh
Publication of WO2021087962A1 publication Critical patent/WO2021087962A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition

Definitions

  • This application belongs to the technical field of security inspection object identification, and in particular relates to a restricted product automatic identification device and method.
  • X-ray baggage safety inspection equipment is widely used to maintain aviation, subway traffic safety and other fields, and provides important technical support for safety inspectors to check the contents of compact, messy and highly variable baggage within a limited time frame.
  • the X-ray baggage security inspection equipment illuminates the contents of the baggage package and displays the organic matter, inorganic matter and mixture as orange, blue, and green pseudo-color images through a series of image processing methods such as image noise removal and image enhancement. The inspector quickly and accurately judges whether there are restricted items in the package.
  • image processing methods such as image noise removal and image enhancement.
  • the inspector quickly and accurately judges whether there are restricted items in the package.
  • the manual interpretation of the security personnel depends on the quality of the false color image, personal experience, and the working status at the time. Missing and misdetection will often occur. Once the peak period is encountered, the phenomenon will be more serious and the efficiency will be low.
  • Chinese patent CN201811174410.2 proposes a "prohibited article detection system and method based on digital image processing”.
  • This patent uses median filtering, color space conversion, threshold segmentation, and two Value processing, contour feature extraction; match the contours of all the extracted items with the contours of prohibited items in the prohibited items library; automatically detect and identify prohibited items according to the results of contour matching, and identify different types of prohibited items Separate marking is performed, and if prohibited items are detected, an audible and visual alarm will be carried out, thereby realizing automatic detection and identification of prohibited items in X-ray security inspection machines.
  • Its disadvantages are: poor mobility and poor accuracy.
  • Chinese patent CN201711126618.2 discloses a "security inspection method, device, system and electronic equipment", which uses a preset deep learning model to extract features from X-ray images after data preprocessing, and uses a deep learning model
  • the trained classifier conducts recognition training on the characteristics of restricted products, obtains the final location and category information of the contraband, generates the recognition result corresponding to the object to be detected, and sends the recognition result of the object to be detected to the security inspection terminal so that the security inspection terminal displays the information
  • the recognition result effectively realizes the automatic detection of restricted items in security inspection images, improves the efficiency and accuracy of manual interpretation, and can effectively prevent potential safety hazards.
  • Chinese patent CN201810551326.1 proposes a "method and device for automatic identification of objects based on artificial intelligence deep learning”.
  • This patent collects X-ray images of the object to be identified by X-ray machines set in different directions in the identification area of the object.
  • the combination of the classification feature with the confidence value greater than the preset minimum value and the corresponding location feature is output as the recognition result.
  • It uses cross-imaging and a preset deep learning model to synthesize multi-dimensionally collected images into a composite image, which greatly improves the recognition rate of contraband, avoids missed inspections, and realizes fully automated security inspections.
  • This application provides an automatic identification device and method for restricted products, which aims to solve one of the above-mentioned technical problems in the prior art at least to a certain extent.
  • An automatic identification device for restricted items including an incoming conveyor belt, an outgoing conveyor belt, a security check and identification device, a control and a display screen; the incoming conveyor belt is located at the entrance of the security check and identification device, and the outgoing conveyor belt is located at the security check and identification device At the exit end of the device, the control and display screen are located above the security check and identification device;
  • the security inspection and identification device includes an X-ray machine and an identification analysis module, the X-ray machine is used to emit an X-ray beam to the package to be identified, and collect high and low energy data and pseudo-color images of the package to be identified;
  • the identification analysis module is connected to the X-ray machine, and is used to use the high and low energy data as the data of the first channel and the second channel respectively, and synthesize the data of the third channel according to the calculation formula to obtain the self-synthesized security image of the package to be identified , Recognizing whether there are restricted products in the self-synthesized security inspection image through the deep learning classifier and regression algorithm, and restricting the output of the product recognition results to the control and display screen;
  • the control and display screen are used to perform template registration of the self-synthesized security check image and the pseudo-color image, and mark the restriction product recognition result on the pseudo-color image according to the registration result and display it;
  • the outgoing conveyor belt is used to convey the parcels that have passed the security check.
  • the technical solution adopted in the embodiment of the present application further includes: the identification analysis module is also used to read the original high and low energy data of the package to be identified, and perform dark current elimination, noise removal and normalization processing on the original high and low energy data.
  • the technical solution adopted in the embodiment of the application further includes a first camera and a second camera, the first camera and the second camera are respectively located above the security inspection and identification device, and the first camera is located near the end of the incoming conveyor belt, The second camera is located at an end close to the outgoing conveyor belt; the first camera and the second camera are used to collect images during the process of passengers placing packages and package delivery, and store the images locally.
  • the technical solution adopted in the embodiment of the application also includes a rotatable separation mechanism, the outgoing conveyor belt includes an inner outgoing conveyor belt and an outer outgoing conveyor belt, and the rotatable separation mechanism is located between the inner outgoing conveyor belt and the outer outgoing conveyor belt; normal conditions Bottom, the rotatable separation structure forwards the conveyor belt toward the outside; when the identification and analysis module recognizes that there are restricted items in the package to be identified, it sends an emergency alarm signal, and if the identification and analysis module recognizes that there are restricted items that cannot be carried in the package , Then send an emergency alarm signal, and turn the rotatable separation structure to the inner outgoing conveyor belt to isolate and inspect the package.
  • the technical solution adopted in the embodiment of this application also includes: when the identification analysis module recognizes that there are restricted items in the package to be identified, read and display the image when the passenger drops the package through the control and display screen, and match the package with the passenger .
  • the technical solution adopted in the embodiment of the present application further includes: the security inspection and identification device further includes a front end lead curtain and a rear end lead curtain, and the front end lead curtain and the rear end lead curtain are used for isolating X-ray rays.
  • the technical solution adopted in the embodiment of this application also includes: the two sides of the X-ray machine are respectively equipped with an infrared transmitter and an infrared receiver that cooperate with each other, and the front-end infrared transmitter is triggered after the package to be identified enters the X-ray machine, The X-ray machine emits an X-ray beam to irradiate the package to be identified, and collects the original high and low energy data and the original pseudo-color image of the package to be identified; after the X-ray machine detects it, the rear infrared receiver is triggered, and the package to be identified leaves the back end
  • the lead curtain is conveyed to the inner outgoing conveyor belt.
  • an automatic identification method for restricted products including the following steps:
  • Step a Collect the high- and low-energy data and pseudo-color images of the package to be identified through the X-ray machine;
  • Step b Use the high-energy and low-energy data as the data of the first channel and the second channel respectively, and synthesize the data of the third channel according to the calculation formula to obtain a self-synthesized security image of the package to be identified;
  • Step c Use a deep learning model to extract the data features of the self-synthesized security check image, use a deep learning classifier and a regression algorithm to identify whether there are restricted products in the self-synthetic security check image, and output the recognition result;
  • Step d Perform template registration on the self-synthesized security check image and the pseudo-color image, and mark the restriction product recognition result on the pseudo-color image according to the registration result and display it.
  • the beneficial effects of the embodiments of the present application are: the automatic identification device and method for restricted products in the embodiments of the present application extract features of the security check images through a deep learning model through a security check image that is self-synthesized based on the original high and low energy data , And output the identification results of restricted products. While realizing the automation of security inspection, this application also has the following beneficial effects:
  • the self-synthesized security inspection image of the original high and low energy data can well reflect the characteristics of restricted products and spatial information, which is more conducive to the feature extraction and training of deep learning, the recognition accuracy is relatively higher, the labor cost of the security inspection is reduced, and the security inspection is improved. s efficiency;
  • the separation structure can quickly isolate the suspected package and wait for further confirmation.
  • the normal security check will continue to avoid blockage of the security check and ensure the efficiency and continuity of the security check.
  • FIG. 1 is a schematic diagram of the structure of an automatic identification device for restricted products according to an embodiment of the present application
  • Fig. 2 is a flowchart of an automatic identification method for restricted products according to an embodiment of the present application.
  • FIG. 1 is a schematic structural diagram of an automatic identification device for restricted products according to an embodiment of the present application.
  • the automatic identification device for restricted products in the embodiment of the application includes an incoming conveyor belt 1, a security check and identification device 2, a first camera 3, a control and display screen 4, a second camera 5, an inner outgoing conveyor belt 6, a rotatable separation mechanism 7, and Outgoing conveyor belt 8.
  • the incoming conveyor belt 1 is located at the entrance end of the security inspection and identification device 2, the inner outgoing conveyor belt 6 and the outer outgoing conveyor belt 8 are respectively located at the exit end of the security inspection and identification device 2, and the rotatable separation mechanism 6 is located at the inner outgoing conveyor belt 6 and the outgoing conveyor belt 8; the first camera 3, the control and display screen 4, and the second camera 5 are respectively located above the security check and identification device 2, and the first camera 3 is located close to the end of the incoming conveyor belt 1, and the second camera 5 is located close to the inside Outgoing one end of the conveyor belt 6.
  • the specific working principle is:
  • the incoming conveyor belt 1 is driven by a motor (not shown) to transfer the parcel to be identified to the security check and identification device 2.
  • the first camera 3 captures the time when the passenger places the parcel in real time The image of the parcel is recorded locally, and the parcel is in one-to-one correspondence with the parcel owner;
  • the security inspection and identification device 2 includes a front lead curtain (not shown in the figure), a rear lead curtain (not shown in the figure), an X-ray machine (not shown in the figure), and an identification analysis module (not shown in the figure).
  • the front lead curtain and the rear lead curtain are used to insulate X-rays and prevent possible radiation hazards to passengers caused by X-ray overflow; both sides of the X-ray machine are equipped with cooperating infrared emitters (not shown) and infrared receivers.
  • the front-end infrared transmitter is triggered, and the X-ray machine emits X-ray beams to irradiate the package to be recognized, and collects the original high and low energy data of the package to be recognized and the original pseudo-color image;
  • the identification analysis module is connected to the X-ray machine to read the original high and low energy data of the package to be identified. After preprocessing the original high and low energy data, such as dark current elimination, noise removal, and normalization, respectively, the high and low energy data are separated As the data of the first channel and the second channel, synthesize the data of the third channel according to the preset calculation formula to obtain the self-synthesis security image of the package to be identified; and use the deep learning model to perform feature selection and extraction on the self-synthesis security image, The deep learning classifier and regression algorithm are used to identify whether there are restricted products in the self-synthesized security inspection image, as well as the category and location information of the existing restricted products, and output the recognition results of the restricted products of the package to the control and display screen 4.
  • the control and display screen 4 matches the self-synthesized security image with the original pseudo-color image, and then prints the restricted product recognition result on the pseudo-color image and displays it on the display screen;
  • the rotatable separation structure 7 faces the outgoing conveyor belt 8; after being detected by the X-ray machine, the rear infrared receiver is triggered, and the package to be identified leaves the rear lead curtain and is transferred to the inner outgoing conveyor belt 6, the inner outgoing conveyor belt 6 is driven by a motor, and the package to be identified is transferred to the outer outgoing conveyor belt 8 through the rotatable separation structure 7 and then sent out; at the same time, the second camera 5 captures the image of the package leaving the rear lead curtain in real time, and Record it locally, leaving important evidence for possible subsequent mishandling of the package and investigation of the responsibility of the package.
  • the identification analysis module If the identification analysis module recognizes that there are uncarryable restricted items in the package, it will send an emergency alarm signal and turn the rotatable separation structure 7 into the outgoing conveyor belt 6 to isolate the package. At the same time, the warning light will light up to remind the security personnel Further unpacking inspection, the control and display screen 4 displays the image when the passenger puts the package, so that the security personnel can match the package with the passenger, and manually turn off the warning light after eliminating the safety hazard. Afterwards, the rotatable separation structure 7 is turned to the outgoing conveyor belt 8 to continue working to avoid jamming of security checks.
  • FIG. 2 is a flowchart of an automatic identification method for restricted products according to an embodiment of the present application.
  • the method for automatically identifying restricted products in the embodiment of the present application includes the following steps:
  • Step 100 Scan the package to be identified by an X-ray machine, and collect the original high and low energy data and the original pseudo-color image of the package to be identified;
  • the high- and low-energy data contains rich original image information, including information such as the category, thickness, and density of the item.
  • Step 200 Perform dark current removal, noise removal, normalization and other preprocessing on the original high-energy and low-energy data respectively;
  • step 200 in order to eliminate the influence of dark current, blank image data is collected as zero point data when the X-ray machine ray source is turned off, and the zero point data is subtracted from the image data collected when the actual ray is turned on. Then a median filter is used to smooth and denoise the high and low energy data, and the normalization process is to reduce the impact of energy differences.
  • Step 300 Use the preprocessed high-energy and low-energy data as the data of the first channel and the second channel respectively, and synthesize the data of the third channel according to a preset calculation formula to obtain a self-synthesized security image of the package to be identified;
  • step 300 because the gray values of the high and low energy imaging of the restricted products of different types, densities and thicknesses are different, the combined information of the high and low energy data can reflect the position and spatial relative characteristics of the restricted products. Therefore, the present invention is based on the original
  • the high and low energy data are obtained from the self-synthesized security inspection image, which can save more information, can well reflect the characteristics of restricted products and spatial information, and is more conducive to the feature extraction and training of deep learning.
  • the recognition accuracy is relatively higher, and the security inspection is reduced. Labor costs and improve the efficiency of security inspections.
  • Step 400 Use the deep learning model to perform feature selection and extraction on the self-synthesized security check image, and use the deep learning classifier and regression algorithm to identify whether there are restricted products in the self-synthesized security check image, as well as the category and location information of the existing restricted products, and output Recognition result
  • step 400 if the recognition result of the self-synthesized security image is that there are no restricted items, the package is directly transmitted through the conveyor belt; if there are restricted items, an emergency alarm signal is sent, and the package is carried out by the rotatable separation mechanism on the conveyor belt. Isolation, at the same time, the warning light is on to remind the security personnel to open the package for further inspection.
  • Step 500 Perform template registration on the self-synthesized security check image and the original pseudo-color image, mark the recognition result on the pseudo-color image according to the registration result, and display it on the display screen;
  • step 500 since the scale of the self-synthesized security image and the original pseudo-color image may be inconsistent, and the pseudo-color image is more convenient for human eyes to observe, this application will perform template matching between the self-synthesized security image and the pseudo-color image. Restricted product identification results (including restricted product category and location) are marked on the pseudo-color image and then displayed, which is more convenient for security personnel to observe and further confirm whether the security check is safe.
  • the images after the package is released can also be collected through the camera in real time to realize one-to-one correspondence between people and packages, leaving important vouchers for possible subsequent package tracking and accountability.
  • the restricted product automatic identification device and method uses the security inspection image that is self-synthesized based on the original high and low energy data, and uses the deep learning model to extract the features of the security inspection image, and output the restriction product recognition result. While realizing the automation of security inspection, this The application also has the following beneficial effects:
  • the self-synthesized security inspection image of the original high and low energy data can well reflect the characteristics of restricted products and spatial information, which is more conducive to the feature extraction and training of deep learning, the recognition accuracy is relatively higher, the labor cost of the security inspection is reduced, and the security inspection is improved. s efficiency;
  • the separation structure can quickly isolate the suspected package and wait for further confirmation.
  • the normal security check will continue to avoid blockage of the security check and ensure the efficiency and continuity of the security check.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)

Abstract

本申请涉及一种限制品自动识别装置及方法。所述方法包括：步骤a：通过X光机采集待识别包裹的高、低能数据以及伪彩色图像；步骤b：将所述高、低能数据分别作为第一通道和第二通道的数据，根据计算公式合成第三通道的数据，得到待识别包裹的自合成安检图像；步骤c：采用深度学习模型提取所述自合成安检图像的数据特征，通过深度学习分类器和回归算法识别所述自合成安检图像中是否存在限制品，并输出识别结果；步骤d：将所述自合成安检图像和伪彩色图像进行模板配准，根据配准结果将限制品识别结果标注在伪彩色图像上并进行显示。本申请识别准确率相对更高，降低了安检的人工成本，提高了安检的效率。

Description

一种限制品自动识别装置及方法 技术领域
本申请属于安检物体识别技术领域，特别涉及一种限制品自动识别装置及方法。
背景技术
如今，火车站、地铁以及机场等场所都会对乘客携带的行李进行检查，不允许乘客携带限制品乘坐交通工具，进行必要的安全风险管控。X光行李安全检查设备广泛用于维护航空、地铁交通安全等领域，为安全检查员在有限的时间范围内检查紧凑、杂乱和高度变化的行李内容提供了重要的技术支持。X光行李安全检查设备通过照射行李包裹里面的物品并经过去除图像噪声、图像增强等一系列图像处理手段将有机物、无机物和混合物分别显示为橙黄色、蓝色和绿色的伪彩色图像，帮助检查员快速准确地判断出包裹中是否存在限制品。但是安检人员的人工判读取决于伪彩色图像的质量、个人经验以及当时的工作状态，会经常出现漏检、误检的情况，一旦遇到高峰期现象更为严重，效率低下。
针对上述问题，中国专利CN201811174410.2提出了一种《基于数字图像处理的违禁物品检测系统及方法》，该专利通过对扫描到的X光图像进行中值滤波、颜色空间转换、阈值分割、二值化处理、轮廓特征提取；将提取到的所有物品的轮廓与违禁物品库中的违禁物品轮廓进行逐个匹配；根据轮廓匹配的结果对违禁物品进行自动检测和识别，并将不同种类的违禁物品进行分别标记，如果检测到违禁物品，会进行声光报警，从而实现了X光安检机中的违禁物品的自动检测和识别。其存在的缺点在于：迁移性比较差，准确率比较差。
中国专利CN201711126618.2公开了一种《安检检测方法、装置、系统及电子设备》，该专利使用了预设的深度学习模型对进行数据预处理后的X光图像进行特征提取，利用深度学习模型训练的分类器对限制品特征进行识别训练，得到违禁品最终的位置和类别信息，生成对应待检测物的识别结果，并将待检测物的识别结果发送至安检终端，以使安检终端显示该识别结果，从而有效地实现了安检图像限制品的自动检测，提高了人工判读的效率和准确度，能够有效预防安全隐患。
中国专利CN201810551326.1提出了一种《基于人工智能深度学习的自动识别物体的方法及其装置》，该专利由设置在物品识别区域不同方位的X光机采集得到待识别物品的X光图像，将多维采集的图像合成一路复合图像，输入预设的深度学习模型，以提取X光图像中对应的待识别物品的多维度数据的分类特征和位置特征；计算分类特征的置信度值，并将置信度值大于预设的最小值的分类特征和相应的位置特征组合作为识别结果输出。其利用交叉成像和预设的深度学习模型，将多维采集的图像合成一路复合图像，大幅提高了违禁品的识别率，避免漏检，且实现了安检的全自动化进行。
上述两件专利虽然采用了深度学习模型实现了限制品的自动检测，但都是基于采集到的伪彩色图像进行特征提取。市面上不同安检设备的图像伪彩色配色的实现方法有所差异，伪彩色的效果也是有所出入的，伪彩色的出发点是按照安检要求将不同类别的物质配色成不同的颜色，具体的颜色深浅和色度并不是严格要求，一一对应的，因此质量参差不齐，干扰极大。伪彩色虽然便于人眼分辨，但也意味着冗余了许多人为添加的信息，丢失了很多原始高低能的大量信息，不能很好体现原始X光的图像限制品特征和空间信息，不利于深度学习的特征提取和训练，识别准确率相对比较劣势。
与此同时，上述专利提出的技术方案与传统的安检方法几乎大同小异，一旦检测到限制品自动报警提醒安检人员，需要安检人员自行定位哪个是携带限制品的包裹并进一步开包检查。遇到高峰出行期间，难免出现机器预警但是无法及时定位发生预警的包裹，开包检查困难大时间长，大大延缓了安检速度甚至错过检查，造成安全隐患。
发明内容
本申请提供了一种限制品自动识别装置及方法，旨在至少在一定程度上解决现有技术中的上述技术问题之一。
为了解决上述问题，本申请提供了如下技术方案：
一种限制品自动识别装置，包括传入传送带、传出传送带、安检及识别装置、控制及显示屏；所述传入传送带位于安检及识别装置的入口端，所述传出传送带位于安检及识别装置的出口端，所述控制及显示屏位于安检及识别装置的上方；
通过所述传入传送带将待识别包裹传送至安检及识别装置内；
所述安检及识别装置包括X光机和识别分析模块，所述X光机用于向待识别包裹发射X光射线束，采集待识别包裹的高、低能数据以及伪彩色图像；
所述识别分析模块连接X光机，用于将所述高、低能数据分别作为第一通道和第二通道的数据，根据计算公式合成第三通道的数据，得到待识别包裹的自合成安检图像，通过深度学习分类器和回归算法识别所述自合成安检图像中是否存在限制品，并限制品识别结果输出到控制及显示屏；
所述控制及显示屏用于将所述自合成安检图像和伪彩色图像进行模板配准，根据配准结果将限制品识别结果标注在伪彩色图像上并进行显示；
所述传出传送带用于传送通过安检的包裹。
本申请实施例采取的技术方案还包括：所述识别分析模块还用于读取待识别包裹的原始高低能数据，并对原始高、低能数据进行暗电流消除、去除噪声和归一化处理。
本申请实施例采取的技术方案还包括第一摄像头和第二摄像头，所述第一摄像头和第二摄像头分别位于安检及识别装置的上方，且所述第一摄像头位于靠近传入传送带的一端，所述第二摄像头位于靠近传出传送带的一端；所述第一摄像头和第二摄像头用于采集旅客投放包裹以及包裹传送过程中的图像，并将图像存储在本地。
本申请实施例采取的技术方案还包括可旋转分离机构，所述传出传送带包括内传出传送带和外传出传送带，所述可旋转分离机构位于内传出传送带和外传出传送带之间；正常情况下，所述可旋转分离结构朝向外传出传送带；当识别分析模块识别到待识别包裹中存在限制品时，则发送紧急报警信号，并通过如果识别分析模块识别出包裹内有不可携带的限制品，则发送紧急报警信号，并将所述可旋转分离结构转向内传出传送带，对该包裹进行隔离检查。
本申请实施例采取的技术方案还包括：当识别分析模块识别到待识别包裹中存在限制品时，通过所述控制及显示屏读取并显示旅客投放包裹时的图像，将包裹和旅客进行匹配。
本申请实施例采取的技术方案还包括：所述安检及识别装置还包括前端铅帘和后端铅帘，所述前端铅帘和后端铅帘用于隔绝X光射线。
本申请实施例采取的技术方案还包括：所述X光机的两侧分别安装有相互配合的红外发射器和红外接收器，所述待识别包裹进入X光机后，触发前端红外发射器，X光机发射X光射线束照射待识别包裹，采集待识别包裹的 原始高低能数据以及原始伪彩色图像；经过X光机检测后，触发后端红外接收器，所述待识别包裹离开后端铅帘被传送至内传出传送带上。
本申请实施例采取的另一技术方案为：一种限制品自动识别方法，包括以下步骤：
步骤a：通过X光机采集待识别包裹的高、低能数据以及伪彩色图像；
步骤b：将所述高、低能数据分别作为第一通道和第二通道的数据，根据计算公式合成第三通道的数据，得到待识别包裹的自合成安检图像；
步骤c：采用深度学习模型提取所述自合成安检图像的数据特征，通过深度学习分类器和回归算法识别所述自合成安检图像中是否存在限制品，并输出识别结果；
步骤d：将所述自合成安检图像和伪彩色图像进行模板配准，根据配准结果将限制品识别结果标注在伪彩色图像上并进行显示。
相对于现有技术，本申请实施例产生的有益效果在于：本申请实施例的限制品自动识别装置及方法通过基于原始高低能数据自合成的安检图像，通过深度学习模型对安检图像进行特征提取，并输出限制品识别结果，在实现安检自动化的同时，本申请还具有以下有益效果：
1、原始高低能数据自合成的安检图像能够很好体现限制品特征和空间信息，更有利于深度学习的特征提取和训练，识别准确率相对更高，降低了安检的人工成本，提高了安检的效率；
2、通过摄像头实时采集投放包裹后的图像，实现了进包和出包时候的人包一一匹配，为后续可能出现的包裹追踪、责任追究留下重要凭证；
3、将自合成安检图像和伪彩色图像进行模板匹配，将识别结果打印在伪彩色图像上，更有利于人眼的观察，进一步确认安检是否安全；
4、一旦发生预警紧急信息，分离结构能够快速将怀疑包裹进行隔离，等待进一步确认，正常安检继续进行，避免安检堵塞，保证安检的效率和持续性。
附图说明
图1是本申请实施例的限制品自动识别装置的结构示意图；
图2是本申请实施例的限制品自动识别方法的流程图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
请参阅图1，是本申请实施例的限制品自动识别装置的结构示意图。本申请实施例的限制品自动识别装置包括传入传送带1、安检及识别装置2、第一摄像头3、控制及显示屏4、第二摄像头5、内传出传送带6、可旋转分离机构7和外传出传送带8。传入传送带1位于安检及识别装置2的入口端，内传出传送带6和外传出传送带8分别位于安检及识别装置2的出口端，可旋转分离机构6位于内传出传送带6和外传出传送带8之间；第一摄像头3、控制及显示屏4和第二摄像头5分别位于安检及识别装置2的上方，且第一摄像头3位于靠近传入传送带1的一端，第二摄像头5位于靠近内传出传送带6的一端。具体工作原理为：
将待识别包裹投放至传入传送带1上，传入传送带1通过电机(图未示)驱动将待识别包裹传送至安检及识别装置2中，同时，通过第一摄像头3实时捕捉旅客投放包裹时的图像，并记录在本地，将包裹和包裹所有者一一对应；
安检及识别装置2包括前端铅帘(图未示)、后端铅帘(图未示)、X光机 (图未示)和识别分析模块(图未示)。前端铅帘和后端铅帘用于隔绝X光射线，防止X光射线溢出可能对旅客造成的辐射危害；X光机两侧分别安装有相互配合的红外发射器(图未示)和红外接收器(图未示)，待识别包裹进机后，触发前端红外发射器，X光机发射X光射线束照射待识别包裹，采集待识别包裹的原始高低能数据以及原始伪彩色图像；
识别分析模块连接X光机，用于读取待识别包裹的原始高低能数据，分别对原始高、低能数据进行暗电流消除、去除噪声、归一化等预处理后，将高、低能数据分别作为第一通道和第二通道的数据，根据预设的计算公式合成第三通道的数据，得到待识别包裹的自合成安检图像；并使用深度学习模型对自合成安检图像进行特征选择和提取，通过深度学习分类器和回归算法识别自合成安检图像中是否存在限制品，以及存在的限制品的类别和位置信息，，并输出该包裹的限制品识别结果到控制及显示屏4。
控制及显示屏4将自合成安检图像和原始伪彩色图像进行模板匹配后，将限制品识别结果打印在伪彩色图像上，并显示在显示屏上；
本申请实施例中，正常情况下，可旋转分离结构7朝向外传出传送带8；经过X光机检测后，触发后端红外接收器，待识别包裹离开后端铅帘被传送至内传出传送带6上，内传出传送带6由电机驱动，将待识别包裹经可旋转分离结构7传送至外传出传送带8后送出；同时，通过第二摄像头5实时捕捉包裹离开后端铅帘的图像，并记录在本地，为后续可能出现的包裹错拿、包裹责任追究留下重要凭证。
如果识别分析模块识别出包裹内有不可携带的限制品，则发送紧急报警信号，将可旋转分离结构7转向内传出传送带6，将该包裹隔离，与此同时，警示灯亮起，提醒安检人员进一步开包检查，控制及显示屏4显示旅客投放包裹 时的图像，便于安检人员将包裹与旅客进行匹配，消除安全隐患后手动关闭警示灯。之后再将可旋转分离结构7转向外传出传送带8继续工作，避免安检堵塞。
请参阅图2，是本申请实施例的限制品自动识别方法的流程图。本申请实施例的限制品自动识别方法包括以下步骤：
步骤100：通过X光机扫描待识别包裹，采集待识别包裹的原始高、低能数据以及原始伪彩色图像；
步骤100中，高、低能数据中蕴含着丰富的原始图像信息，其中包括物品的类别、厚度、密度等信息。
步骤200：分别对原始高、低能数据进行暗电流消除、去除噪声、归一化等预处理；
步骤200中，为了消除暗电流的影响，在X光机射线源关闭时采集空白图像数据作为零点数据，将实际射线开启时采集的图像数据减去零点数据。然后使用中值滤波器对高、低能数据进行平滑去噪，归一化处理是为了减小能量差异的影响。
步骤300：将预处理后的高、低能数据分别作为第一通道和第二通道的数据，根据预设的计算公式合成第三通道的数据，得到待识别包裹的自合成安检图像；
步骤300中，由于不同类别、密度和厚度的限制品的高低能成像中灰度值是不一样的，高、低能数据的组合信息可以反映限制品的位置和空间相对特征，因此本发明基于原始高、低能数据得到自合成安检图像，可以保存更多的信息，能够很好的体现限制品特征和空间信息，更有利于深度学习的特征提取和训练，识别准确率相对更高，降低安检的人工成本，并提高安检效率。
步骤400：使用深度学习模型对自合成安检图像进行特征选择和提取，通过深度学习分类器和回归算法识别自合成安检图像中是否存在限制品，以及存在的限制品的类别和位置信息，并输出识别结果；
步骤400中，如果自合成安检图像的识别结果为不存在限制品，则通过传送带直接传出包裹；如果存在限制品，则发送紧急报警信号，并通过传送带上的可旋转分离机构将该包裹进行隔离，与此同时，警示灯亮起，提醒安检人员进一步开包检查。
步骤500：将自合成安检图像和原始伪彩色图像进行模板配准，根据配准结果将识别结果标注在伪彩色图像上，并通过显示屏进行显示；
步骤500中，由于自合成安检图像的尺度和原始伪彩色图像可能存在不一致的情况，而伪彩色图像更有利人眼观察，因此本申请将自合成安检图像和伪彩色图像进行模板匹配后，将限制品识别结果(包括限制品类别和位置)标注在伪彩色图像上后进行显示，更便于安检人员观察，进一步确认安检是否安全。
本申请实施例中，还可以通过摄像头实时采集包裹投放后的图像，实现人和包裹一一对应，为后续可能出现的包裹追踪、责任追究留下重要凭证。
本申请实施例的限制品自动识别装置及方法通过基于原始高低能数据自合成的安检图像，通过深度学习模型对安检图像进行特征提取，并输出限制品识别结果，在实现安检自动化的同时，本申请还具有以下有益效果：
1、原始高低能数据自合成的安检图像能够很好体现限制品特征和空间信息，更有利于深度学习的特征提取和训练，识别准确率相对更高，降低了安检的人工成本，提高了安检的效率；
2、通过摄像头实时采集投放包裹后的图像，实现了进包和出包时候的人包一一匹配，为后续可能出现的包裹追踪、责任追究留下重要凭证；
3、将自合成安检图像和伪彩色图像进行模板匹配，将识别结果打印在伪彩色图像上，更有利于人眼的观察，进一步确认安检是否安全；
4、一旦发生预警紧急信息，分离结构能够快速将怀疑包裹进行隔离，等待进一步确认，正常安检继续进行，避免安检堵塞，保证安检的效率和持续性。
以上所述仅是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本发明的保护范围。

Claims (8)

  1. 一种限制品自动识别装置，包括传入传送带和传出传送带，其特征在于，还包括安检及识别装置、控制及显示屏；所述传入传送带位于安检及识别装置的入口端，所述传出传送带位于安检及识别装置的出口端，所述控制及显示屏位于安检及识别装置的上方；
    通过所述传入传送带将待识别包裹传送至安检及识别装置内；
    所述安检及识别装置包括X光机和识别分析模块，所述X光机用于向待识别包裹发射X光射线束，采集待识别包裹的高、低能数据以及伪彩色图像；
    所述识别分析模块连接X光机，用于将所述高、低能数据分别作为第一通道和第二通道的数据，根据计算公式合成第三通道的数据，得到待识别包裹的自合成安检图像，通过深度学习分类器和回归算法识别所述自合成安检图像中是否存在限制品，并限制品识别结果输出到控制及显示屏；
    所述控制及显示屏用于将所述自合成安检图像和伪彩色图像进行模板配准，根据配准结果将限制品识别结果标注在伪彩色图像上并进行显示；
    所述传出传送带用于传送通过安检的包裹。
  2. 根据权利要求1所述的限制品自动识别装置，其特征在于，所述识别分析模块还用于读取待识别包裹的原始高低能数据，并对原始高、低能数据进行暗电流消除、去除噪声和归一化处理。
  3. 根据权利要求1所述的限制品自动识别装置，其特征在于，还包括第一摄像头和第二摄像头，所述第一摄像头和第二摄像头分别位于安检及识别装置的上方，且所述第一摄像头位于靠近传入传送带的一端，所述第二摄像头位于靠近传出传送带的一端；所述第一摄像头和第二摄像头用于采集旅客投放包裹 以及包裹传送过程中的图像，并将图像存储在本地。
  4. 根据权利要求1至3任一项所述的限制品自动识别装置，其特征在于，还包括可旋转分离机构，所述传出传送带包括内传出传送带和外传出传送带，所述可旋转分离机构位于内传出传送带和外传出传送带之间；正常情况下，所述可旋转分离结构朝向外传出传送带；当识别分析模块识别到待识别包裹中存在限制品时，则发送紧急报警信号，并通过如果识别分析模块识别出包裹内有不可携带的限制品，则发送紧急报警信号，并将所述可旋转分离结构转向内传出传送带，对该包裹进行隔离检查。
  5. 根据权利要求4所述的限制品自动识别装置，其特征在于，当识别分析模块识别到待识别包裹中存在限制品时，通过所述控制及显示屏读取并显示旅客投放包裹时的图像，将包裹和旅客进行匹配。
  6. 根据权利要求1所述的限制品自动识别装置，其特征在于，所述安检及识别装置还包括前端铅帘和后端铅帘，所述前端铅帘和后端铅帘用于隔绝X光射线。
  7. 根据权利要求6所述的限制品自动识别装置，其特征在于，所述X光机的两侧分别安装有相互配合的红外发射器和红外接收器，所述待识别包裹进入X光机后，触发前端红外发射器，X光机发射X光射线束照射待识别包裹，采集待识别包裹的原始高低能数据以及原始伪彩色图像；经过X光机检测后，触发后端红外接收器，所述待识别包裹离开后端铅帘被传送至内传出传送带上。
  8. 一种限制品自动识别方法，其特征在于，包括以下步骤：
    步骤a：通过X光机采集待识别包裹的高、低能数据以及伪彩色图像；
    步骤b：将所述高、低能数据分别作为第一通道和第二通道的数据，根据计算公式合成第三通道的数据，得到待识别包裹的自合成安检图像；
    步骤c：采用深度学习模型提取所述自合成安检图像的数据特征，通过深度学习分类器和回归算法识别所述自合成安检图像中是否存在限制品，并输出识别结果；
    步骤d：将所述自合成安检图像和伪彩色图像进行模板配准，根据配准结果将限制品识别结果标注在伪彩色图像上并进行显示。
PCT/CN2019/116645 2019-11-08 2019-11-08 一种限制品自动识别装置及方法 WO2021087962A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/116645 WO2021087962A1 (zh) 2019-11-08 2019-11-08 一种限制品自动识别装置及方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/116645 WO2021087962A1 (zh) 2019-11-08 2019-11-08 一种限制品自动识别装置及方法

Publications (1)

Publication Number Publication Date
WO2021087962A1 true WO2021087962A1 (zh) 2021-05-14

Family

ID=75848171

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/116645 WO2021087962A1 (zh) 2019-11-08 2019-11-08 一种限制品自动识别装置及方法

Country Status (1)

Country Link
WO (1) WO2021087962A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113284201A (zh) * 2021-05-27 2021-08-20 杭州睿影科技有限公司 安检图像的生成方法、安检系统及存储介质
CN114295649A (zh) * 2021-12-31 2022-04-08 杭州睿影科技有限公司 一种信息关联方法、装置、电子设备及存储介质
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
CN116311085A (zh) * 2023-05-19 2023-06-23 杭州睿影科技有限公司 一种图像处理方法、系统、装置及电子设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101696947A (zh) * 2009-10-13 2010-04-21 公安部第一研究所 一种x射线双能量透射与康普顿背散射图像智能融合方法
US20110170661A1 (en) * 2008-08-26 2011-07-14 General Electric Company Inspection system and method
CN104991282A (zh) * 2015-07-22 2015-10-21 北京睿思厚德辐射信息科技开发有限公司 具有行李包外形同步显示的x射线行李包检查系统
CN205404838U (zh) * 2016-03-08 2016-07-27 龙靖 一种用于配合快递物流x光安检机安检的全自动系统
CN110133739A (zh) * 2019-04-04 2019-08-16 南京全设智能科技有限公司 一种x射线安检设备及其自动识图方法
CN110412657A (zh) * 2019-06-06 2019-11-05 上琛安防科技（上海）有限公司 智能化x光安全检测设备人像集成平台系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110170661A1 (en) * 2008-08-26 2011-07-14 General Electric Company Inspection system and method
CN101696947A (zh) * 2009-10-13 2010-04-21 公安部第一研究所 一种x射线双能量透射与康普顿背散射图像智能融合方法
CN104991282A (zh) * 2015-07-22 2015-10-21 北京睿思厚德辐射信息科技开发有限公司 具有行李包外形同步显示的x射线行李包检查系统
CN205404838U (zh) * 2016-03-08 2016-07-27 龙靖 一种用于配合快递物流x光安检机安检的全自动系统
CN110133739A (zh) * 2019-04-04 2019-08-16 南京全设智能科技有限公司 一种x射线安检设备及其自动识图方法
CN110412657A (zh) * 2019-06-06 2019-11-05 上琛安防科技（上海）有限公司 智能化x光安全检测设备人像集成平台系统

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113284201A (zh) * 2021-05-27 2021-08-20 杭州睿影科技有限公司 安检图像的生成方法、安检系统及存储介质
CN114295649A (zh) * 2021-12-31 2022-04-08 杭州睿影科技有限公司 一种信息关联方法、装置、电子设备及存储介质
CN114295649B (zh) * 2021-12-31 2023-11-03 杭州睿影科技有限公司 一种信息关联方法、装置、电子设备及存储介质
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
CN116311085A (zh) * 2023-05-19 2023-06-23 杭州睿影科技有限公司 一种图像处理方法、系统、装置及电子设备
CN116311085B (zh) * 2023-05-19 2023-09-01 杭州睿影科技有限公司 一种图像处理方法、系统、装置及电子设备

Similar Documents

Publication Publication Date Title
WO2021087962A1 (zh) 一种限制品自动识别装置及方法
CN110133739B (zh) 一种x射线安检设备及其自动识图方法
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
CN204287075U (zh) 一体化安全检查系统
CN106251347B (zh) 地铁异物检测方法、装置、设备及地铁屏蔽门系统
EP3796051A1 (en) Personnel security inspection method and personnel security inspection system
CN109977877A (zh) 一种安检智能辅助判图方法、系统以及系统控制方法
CN206193265U (zh) 跨境电子商务智能安检系统
CN108279244B (zh) 一种城市轨道交通安检系统
CN111612020B (zh) 一种异常被检物的定位方法以及安检分析设备、系统
CN205157812U (zh) 新型智能监测安检机
CN101696947B (zh) 一种x射线双能量透射与康普顿背散射图像智能融合方法
JP2016126014A (ja) 一体型セキュリティ検査システム
CN108957578A (zh) 旋转式太赫兹安检装置及安检方法
CN105068135A (zh) 行邮智能检测系统及检测方法
CN110133740A (zh) 一种双视角智能化x射线安检设备
AU2017261639A1 (en) Security check system and method
CN110850493A (zh) 一种扫描成像机器视觉判图的线阵列高速安检机
JP2017091498A (ja) ゲート装置
CN111310635B (zh) 基于TensorFlow的安检违禁品识别系统及方法
CN114160447B (zh) 一种先期机检系统及方法
CN113706497B (zh) 一种违禁品智能识别装置和系统
CN208537736U (zh) 基于人车物的综合安检系统
CN206672198U (zh) 一种安检通行系统
CN109425623A (zh) 一种新型自动化安检通道

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: 19951595

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: 19951595

Country of ref document: EP

Kind code of ref document: A1