WO2009070977A1 - Procédé et dispositif d'identification de matériau - Google Patents

Procédé et dispositif d'identification de matériau Download PDF

Info

Publication number
WO2009070977A1
WO2009070977A1 PCT/CN2008/001880 CN2008001880W WO2009070977A1 WO 2009070977 A1 WO2009070977 A1 WO 2009070977A1 CN 2008001880 W CN2008001880 W CN 2008001880W WO 2009070977 A1 WO2009070977 A1 WO 2009070977A1
Authority
WO
WIPO (PCT)
Prior art keywords
energy
pixel
transparency
low
value
Prior art date
Application number
PCT/CN2008/001880
Other languages
English (en)
French (fr)
Inventor
Zhiqiang Chen
Li Zhang
Kejun Kang
Xuewu Wang
Qingping Huang
Yuanjing Li
Yinong Liu
Ziran Zhao
Yongshun Xiao
Original Assignee
Nuctech Company Limited
Tsinghua University
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
Family has litigation
First worldwide family litigation filed litigation Critical https://patents.darts-ip.com/?family=40139670&utm_source=google_patent&utm_medium=platform_link&utm_campaign=public_patent_search&patent=WO2009070977(A1) "Global patent litigation dataset” by Darts-ip is licensed under a Creative Commons Attribution 4.0 International License.
Application filed by Nuctech Company Limited, Tsinghua University filed Critical Nuctech Company Limited
Publication of WO2009070977A1 publication Critical patent/WO2009070977A1/zh

Links

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N23/00Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00
    • G01N23/02Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material
    • G01N23/04Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material and forming images of the material
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N23/00Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00
    • G01N23/02Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material
    • G01N23/06Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material and measuring the absorption
    • G01N23/083Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material and measuring the absorption the radiation being X-rays
    • G01N23/087Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material and measuring the absorption the radiation being X-rays using polyenergetic X-rays
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2223/00Investigating materials by wave or particle radiation
    • G01N2223/40Imaging
    • G01N2223/423Imaging multispectral imaging-multiple energy imaging

Definitions

  • the invention relates to a radiation imaging technology, in particular to a material identification method and device applied to a high-energy X-ray dual-energy imaging inspection system, which can not only obtain a transmission image of an object to be inspected, but also obtain material information in the object to be inspected. . Background technique
  • Containerization has become a major trend in international cargo transportation.
  • WMDs weapons of mass destruction
  • RDDs Radiological Dispersal Devices
  • the current container (radiation security inspection equipment is mainly based on transmission imaging, using X-rays to directly transmit goods, and obtain transmission images of all items covered by X-ray path. Standard transmission imaging technology solves the container
  • the "visualization" problem has been widely used.
  • such devices generally have the following disadvantages: First, the two-dimensional structural information is susceptible to interference by overlapping objects on the ray path; secondly, it does not contain density information; and third, it does not contain material information.
  • the main method is to check the X-ray scan image of the customs declaration and the container, check whether the match is met, the customs declaration is the prior knowledge, and the standard X-ray transmission imaging technology can basically meet the demand.
  • the introduction of CSI has led to the development of container inspection requirements from the inspection of smuggled goods (referred to as “inspection of private”) to the inspection of dangerous goods (referred to as “inspection of dangerous goods”). Due to the wide variety of dangerous species and no fixed shape, it is equal to the inspection There is no prior knowledge of the items in the container, so it is difficult to meet the requirements of container safety inspections solely by standard X-ray transmission imaging technology.
  • the dual-energy transmission technique uses two kinds of X-rays with different energy spectra to penetrate the object, and the difference of the output signals is processed to obtain the atomic number information of the material of the object to be inspected. Therefore, with this technology, the level of safety inspection will be effectively improved to some extent. It is hoped that the high-energy X-ray imaging container inspection system has the material resolving power, which has become a hot spot in international research in recent years.
  • the dual-energy technology is very effective when the X-ray energy is lower than 200KeV, it is widely used in baggage inspection.
  • the X-ray energy penetrating the container reaches several MVs, for different materials of the same quality and thickness. , Al, Fe, the attenuation of the radiation in this energy segment is very small. Therefore, the material resolving power obtained by high-energy X-ray imaging is much worse than that of low-energy dual-energy X-ray technology.
  • Even some experts in the field of container inspection believe that dual-energy imaging technology has little effect when X-ray energy is at 200KeV, so it is not suitable for container inspection systems. Summary of the invention
  • the invention provides a material identification system based on an energy spectrum shaping device and an automatic calibration device, and solves the problem of real-time effective material discrimination in a high-energy X-ray dual-energy imaging container inspection system, and simultaneously utilizes dual-energy gray-scale fusion and colorization. Algorithm to visualize material resolution and grayscale information.
  • the system can obtain the fusion image with high penetration and high contrast sensitivity, and obtain the material information of the cargo, for explosives, drugs and Radioactive materials and other dangerous goods have certain recognition capabilities, thus improving the safety inspection capability of containers.
  • the present invention fully utilizes the advantages of human vision and strives to transmit more information to user.
  • a substance identification method comprising the steps of: transmitting a test object with high energy rays and low energy rays to obtain a high energy transmission image and a low energy transmission image of the object to be inspected, each of the high energy images
  • the value of each pixel represents the high energy transparency of the corresponding portion of the object to be inspected by the high energy ray
  • the value of each pixel in the low energy image represents the low energy transparency of the corresponding portion of the object to be inspected by the low energy ray
  • the calculation a value of a first function of high energy transparency and a value of the second function of the high energy transparency and the low energy transparency; and a value of the first function and the second function by using a previously created classification curve pair
  • the position determined by the value is classified to identify the type of substance in the portion of the object to be examined corresponding to each pixel.
  • the method further comprises the steps of: setting a neighborhood having a predetermined size; and denoising the high energy image and the low energy transmission image in a neighborhood of each pixel.
  • the step of denoising the high energy image and the low energy transmission image in the neighborhood of each pixel comprises: searching for pixels similar to the center pixel in the neighborhood as similar pixels; Similar pixels in the neighborhood are weighted averaged.
  • the difference between the high energy transparency and the low energy transparency of the similar pixels and the high energy transparency and low energy transparency of the center pixel is less than a predetermined value.
  • the object to be inspected is identified as an organic matter, a light metal, an inorganic substance or a heavy metal.
  • the method further comprises the step of colorifying the recognition result.
  • the colorizing display step comprises: performing weighted averaging on the pixel transparency and low energy transparency of each pixel as the fused gray value of the pixel; according to the corresponding part of the object to be inspected a material type to determine a color hue; determining a brightness level of the pixel based on the fused gray value of the pixel; using the color hue and the brightness level as an index to obtain the pixel from a pre-created lookup table
  • the step of determining a color hue according to a material type of a portion corresponding to the pixel in the object to be inspected includes: imparting an orange color to an organic substance, imparting a green color to a light metal, imparting a blue color to an inorganic substance, and imparting a purple color to a heavy metal.
  • the method comprises the step of spectrally shaping the radiation emitted from the source to expand the difference in energy spectrum between the high energy and low energy rays.
  • the classification curve is created for each calibration material by the following steps: illuminating the calibration material of various thicknesses with high energy rays and low energy rays to obtain corresponding high energy transparency and low energy transparency;
  • the first function of transparency acts as the abscissa, and the second function of low energy transparency and high energy transparency is used as the ordinate to form points of the calibration material at different thicknesses; the classification curve is formed based on the points.
  • the step of forming the classification curve based on the points comprises curve fitting the points using a least squares curve fitting method.
  • the step of forming the classification curve based on the point comprises: curve fitting the point using a best fit polynomial in the Chebyshev sense.
  • the method further comprises the step of discretizing the classification curve.
  • a substance recognizing apparatus comprising: an image forming apparatus that transmits an object to be inspected with a xenon ray and a low energy ray to obtain a high energy transmission image and a low energy transmission image of the object to be inspected,
  • the value of each pixel in the high energy image represents the high energy transparency of the high energy ray to the corresponding portion of the object under test, and the value of each pixel in the low energy image represents the low energy transparency of the low energy ray to the corresponding portion of the object to be inspected;
  • the value determined by the value of a function and the value of the second function is classified to identify the type of substance in the portion of the object to be examined
  • the apparatus further comprises: means for setting a neighborhood having a predetermined size; and means for noise reduction of the high energy image and the low energy transmission image in a neighborhood of each pixel.
  • the apparatus for performing noise reduction on the high energy image and the low energy transmission image in a neighborhood of each pixel includes: means for finding a pixel similar to a center pixel as a similar pixel in a neighborhood; A means for weighted averaging of similar pixels in a neighborhood.
  • the substance identification subsystem of the invention is embedded in a high-energy X-ray dual-energy imaging container inspection system, and the ray energy spectrum is shaped by the design of the energy spectrum shaping device, thereby improving the material resolving power.
  • the system status is monitored in real time, and the classification parameters that best match the system status are obtained, which provides a solid foundation for accurate material resolution.
  • the combination of the fast recognition algorithm and the image denoising algorithm in the material resolution module not only ensures the real-time performance of the algorithm, but also greatly reduces the influence of the statistical fluctuation of the X-ray data, and ensures the accuracy of material resolution.
  • the material recognition system obtains a fused image that combines the penetrating power and the contrast sensitivity by designing a gradation fusion algorithm, and can obtain more information than the single-energy system from the gradation.
  • the present invention further designs a color display module, from inputting original dual-energy transmission data to outputting RGB color display data, to ensure the integrity of the entire data processing process.
  • system also fully considers the problem of real-time operation, optimizes various algorithms, and has fast running speed and good real-time performance.
  • the invention can effectively solve the problem of poor material resolving ability of the high energy X-ray dual energy inspection system, has good material resolution effect and color display effect, and has good operability, fast running speed and high practical value. . DRAWINGS
  • Figure 1A shows a basic composition diagram of a high energy X-ray dual energy imaging system
  • Figure 1B shows a schematic diagram of a high energy X-ray dual energy imaging system with a material identification device embedded therein;
  • Figure 1C shows a mass attenuation coefficient curve;
  • FIG. 2 is a schematic diagram of an energy spectrum shaping device according to an embodiment of the present invention, in which a black arrow indicates a high energy ray, a gray arrow indicates a low energy ray, and a black region indicates a shaped material block;
  • FIG. 3 is a schematic diagram of an automatic calibration device according to an embodiment of the present invention, in which black arrows indicate high energy rays and gray arrows indicate low energy rays;
  • FIG. 4A shows a flow chart of an automatic calibration process in accordance with an embodiment of the present invention
  • FIG. 4B shows a flow chart of a substance identification method according to an embodiment of the present invention
  • Figure 5 shows the alpha curve coordinate definition
  • Figure 6A shows a schematic diagram of calibration material training data used in the automatic calibration process
  • Figure 6B shows a schematic diagram of an alpha curve generated from calibration material training data
  • Figure 6C shows the statistical results of the calibration material training data
  • Figure 7A shows a flow chart of the high and low energy transparency noise reduction process
  • Figure 7B shows a flow chart of the noise reduction process performed on the material identification result
  • Figure 8 is a comparison of imaging results produced by the material resolution module for different pre-processing parameters
  • Figure 9 is a schematic view showing a process of generating a color table and a process of generating an RGB image
  • Figure 10 shows a comparison of grayscale and colormap effects.
  • a substance identification device includes a hardware portion and a data processing algorithm.
  • the substance identification device is a sub-system embedded in the high-energy X-ray dual-energy imaging container inspection system. It is based on high-energy dual-energy transmission data for material resolution.
  • the high-energy X-ray dual-energy ray is high-energy X-ray
  • the low-energy ray is low-energy X-ray.
  • a reasonable choice for the energy segment is the high energy X-ray dual-energy imaging container inspection system.
  • the energy segment selection range of high energy X-ray dual energy is generally between 3 MeV and 10 MeV. In theory, the greater the energy difference, the better the material resolving power in the appropriate energy range. However, if the energy difference is too large, the penetration ability of the high-energy and low-energy rays is too large, and the effective range for material resolution is narrow.
  • the high energy X-ray dual energy imaging container inspection system includes a radiation generating device 10, a mechanical transmission device (not shown), an inspected object 20 such as a container, a data acquisition subsystem 30, and a scanning control.
  • Computer and data processing computer (not shown).
  • the ray generating device 10 includes a dual energy accelerator and other auxiliary devices capable of alternately generating X-ray beams of two energies at a high frequency.
  • the mechanical transmission enables the radiation generating device 10 and the data acquisition subsystem 30 to produce a relative movement in the horizontal direction with respect to the container 20 to be inspected.
  • the radiation generating device 10 and the data collecting subsystem 30 may be stationary, and the container 20 to be inspected is moved. Alternatively, it is also possible that the container 20 to be inspected does not move, and the radiation generating device 10 and the data collection subsystem 30 move together.
  • the data acquisition subsystem 30 mainly includes a line array detector for detecting the radiation of the dual-energy X-ray beam generated by the radiation generating device 10 after passing through the detection object, generating dual-energy transmission data, and transmitting the data to the computer (not shown). ).
  • the data acquisition subsystem 30 also includes projection data readout circuitry and logic control units on the detector.
  • the detector can be a solid state detector, a gas detector, or a semiconductor detector.
  • the scan control computer is responsible for checking the main control of the system's operation, including mechanical control, electrical control, and safety interlock control.
  • the data processing computer is responsible for processing and displaying the dual-energy transmission data obtained by the data acquisition subsystem.
  • an energy spectrum shaping device 40 and an automatic calibration device 50 have been introduced, see Fig. 1B.
  • the energy spectrum shaping device 40 includes an energy spectrum shaping material and corresponding auxiliary equipment.
  • the energy spectrum shaping device 40 is placed between the radiation generating device 10 and the object to be inspected 20 for the purpose of shaping the energy spectrum of the radiation output from the radiation generating device 10 before the radiation penetrates the object to be inspected 20, in order to more spectrally distribute the spectrum. Conducive to material resolution.
  • the energy spectrum shaping material is characterized by a large attenuation of low-energy rays and a small attenuation of high-energy rays. The better this feature, the better the spectral shaping effect. As long as this characteristic can be satisfied, it can be used as a spectrum shaping material. Based on the characteristics of the energy spectrum shaping material, after the energy spectrum shaping, the equivalent energy of the radiation is improved. If only the shaping material is applied to the high-energy beam, the equivalent energy of the high-energy beam is improved, and the equivalent energy of the low-energy beam is unchanged, thereby widening the energy difference between the dual energy to improve the material of the system. Resolving power. Based on this characteristic, a graphite material is selected as the shaping material.
  • the thicker the shaped material the better the material resolution.
  • the thicker the shaping material the greater the attenuation of the radiation, and the lower the dose received by the detector, the lower the signal-to-noise ratio of the data. Therefore, the thickness of the shaped material has an optimum value. This optimal value needs to be determined according to the actual situation of the system. According to the distribution of the high energy and low energy ray energy, it is determined that the energy spectrum shaping is performed only for a certain energy file, and FIG. 2A shows a schematic diagram of the energy spectrum shaping device in the form of a turntable. Alternatively, energy spectrum shaping is performed on both energy levels, and Figure 2B shows an energy spectrum shaping device that can perform energy spectrum shaping for both energy levels.
  • the design of the energy spectrum shaping device 40 should be based on the needs of the energy spectrum shaping. It is possible to shape only the high-energy rays and increase the energy resolution of the system by increasing the energy of the high-energy rays to increase the energy difference between the two energy sources. It is also possible to perform energy spectrum shaping on both high and low energy levels. This situation is quite special, generally for low energy ray around 3M. It can be seen from the mass attenuation coefficient curve shown in Fig. 1C that the attenuation coefficient of the low Z material is close to the vicinity of the 3MeV energy section, and the transformation trend is very slow.
  • the energy change has little effect on the resolution of the low Z material, while the attenuation coefficient of the high Z material is an inflection point around 3 MeV. This phenomenon will lead to the inability of the lead material to be distinguished from other materials under this energy choice. Therefore, the energy spectrum of the 3MeV low energy energy is also shaped.
  • the energy spectrum shaping material absorbs the low energy part of the low energy energy, which can improve the distinguishability of the high Z material and has no negative influence on the low Z material.
  • the automatic calibration module consists of two parts: the design of the automatic calibration device 50 on the hardware and the design of the automatic calibration process on the software.
  • the automatic calibration device 50 includes calibration materials in a stepped configuration and corresponding auxiliary devices.
  • the purpose of the automatic calibration device 50 is to collect the calibration data, enter the automatic calibration process, and obtain the classification parameters matching the system status in real time, and save them in the file as the input of the material resolution module.
  • Each material is designed in several steps from thin to thick.
  • the thinnest and thickest thickness is determined by the material resolution range of the system.
  • the number of step series is determined by the accuracy of the calibration and the space in which the automatic calibration device is placed.
  • the auxiliary device mainly provides mechanical transmission to realize positioning scanning to obtain dual-energy transmission data for each step of each material. It is required to continuously scan several columns of dual-energy transmission data at each positioning point. It is recommended to scan more than 256 columns, which can largely eliminate the influence of statistical fluctuation of the signal.
  • the X-ray angular distributions received by the different detectors on the detector boom are different.
  • the energy spectrum distribution is different, resulting in different material resolution parameters. Therefore, considering the influence of the X-ray angular distribution, all the detection heights can be divided into several regions, and each region is independently counted to generate classification parameters. This requires that the calibration material in the automatic calibration device 50 should cover all of the detection intervals of interest.
  • the height of the calibration material is limited by objective factors (processing power, equipment space, etc.), it is impossible to cover all the detector modules on the boom.
  • a simplified way is as follows: In general, the most interesting detection height is placed inside the container. In the position where the goods are placed, the general system will adjust the main beam of the X-ray to the vicinity of the position. Therefore, the direction of the main beam of the ray is the key calibration object.
  • the calibration material can be designed to cover only the region, and the obtained dual-energy transmission data is input as a parameter to the automatic calibration algorithm, and the classification parameter corresponding to the energy spectrum distribution of the X-ray main beam direction is generated as the classification parameter of all the detection regions. This simplified method is within the error tolerance when the X-ray angular distribution is small.
  • the calibration material in the automatic calibration device 50 can be designed in any shape as long as the above requirements are met.
  • the number of steps and thickness of the steps are for illustrative purposes only and do not represent actual meaning.
  • FIG. 4A shows a flow chart of an automatic calibration process in accordance with an embodiment of the present invention.
  • the radiation generating device 10 produces an X-ray beam.
  • the X-ray beam is shaped by the energy spectrum shaping device 40.
  • the automatic calibration process is to be performed, the artificial calibration process is performed and the automatic calibration process is run to obtain the original calibration data in real time.
  • FIG. 4B shows a flow chart of a substance identification method in accordance with an embodiment of the present invention.
  • the ray generating device 10 produces an X-ray beam
  • the spectral shaping device 40 shapes the X-ray beam to obtain an X-ray beam that is advantageous for material resolution.
  • the shaped X-ray beam passes through the object 20 to be inspected to obtain raw dual energy data of the object under test.
  • a data correction module is run to perform data correction on the original dual energy data to eliminate the effects of detector background, detector inconsistency, and radiation dose fluctuations.
  • the corrected data is used for material resolution and dual energy grayscale fusion processing.
  • the classification parameter file generated during the automatic calibration process is input to the material resolution module, and based on the corrected dual energy data, the material of the object to be inspected is identified to generate material information.
  • the corrected dual energy data is input to the dual energy gray level fusion module, and the gray level fusion processing is performed to generate the transmitted image of the detected object after the dual energy data fusion.
  • the material information output by the module is combined with the material information, and at block 280, colorization processing is performed, that is, the transmission image data suitable for the gray scale display is converted into a color display suitable according to the material information contained in the object to be inspected.
  • the RGB data is then displayed 290 on the data processing computer at block 290.
  • the automatic calibration process is triggered by humans, and the automatic calibration device 50 is activated to collect the original calibration data after the energy spectrum shaping, and the data collection subsystem sends the data to the data processing computer.
  • the alpha curve method is used to design the material resolution algorithm. Therefore, the purpose of the automatic calibration algorithm is to calculate the alpha curve classification parameters that match the state of the system.
  • the alpha curve classification parameters of the system state matching are obtained and saved in the file as the parameter input of the material resolution module.
  • the alpha curve coordinates are defined in Figure 5.
  • alphaL and alphaH as follows:
  • alphaL (l-log(T L ))*1000; where T L is low energy transparency;
  • alphaH (l-log(T H ))* 1000; where T H is the transparency of the energy.
  • alphaH Take alphaH as the abscissa alphax of the alpha curve, taking the difference between alphaL and alphaH as the ordinate of the alpha curve.
  • the data correction module is invoked to perform data correction on the original calibration data to eliminate the effects of detector background, detector inconsistency, and radiation dose fluctuations, and to obtain calibration material training data.
  • Figure 6A is a schematic representation of the training data for a certain detection interval on the alpha curve.
  • Fig. 6C is a schematic diagram showing the mean value of the training data in a certain detection area on the alpha curve.
  • the weighted average weight can be calculated simply according to the atomic number, that is, the distinguishability in the range of different atomic numbers is the same.
  • the distinguishability between different atomic numbers since the dual energy is different from the low energy dual energy, the material resolving power is weak, and only materials belonging to different categories can be distinguished. , and can not accurately distinguish materials with different atomic numbers, so this difference is acceptable.
  • material resolution is a feature of dual energy X-ray systems that differ from single-energy X-ray systems. Since the material resolution of high-energy X-ray imaging is much worse than that of low-energy dual-energy X-ray technology, the material resolution module needs to consider not only how to correctly classify, but also how to improve the material resolution.
  • the noise reduction preprocessing is performed on the high and low energy transparency, and then the material is resolved by using the noise reduction result, and finally the material resolution result is further subjected to noise reduction processing. If the system requires very high processing speed, the two-step noise reduction process before and after material identification can only retain one step, which can still ensure better material resolution.
  • the material resolving power is much worse than that of the low energy dual energy X-ray technology, and the statistical fluctuation of the X-ray is inherent. Therefore, the data needs to be pre-processed, and the dual-energy images are separately denoised, otherwise the accuracy of material resolution is greatly limited.
  • the noise reduction the larger the statistical fluctuation is, the lower the classification accuracy is. If the classification accuracy is lower than a certain level, the classification error is determined. In order to improve the classification accuracy and ensure the material resolution effect, it is necessary to design an effective preprocessing algorithm for noise reduction.
  • the purpose of preprocessing is to reduce the noise of the data so that the data is as close as possible to the true value to increase the accuracy of material resolution.
  • the degree of noise reduction is determined by the noise level of the system.
  • the design flexibility of the pre-processing algorithm is relatively large, as long as the purpose of noise reduction can be achieved. In order to discriminate the material as much as possible, the preprocessing algorithm design process needs to focus on two problems of neighborhood selection and similarity point definition:
  • Choice of neighborhood The neighborhood should be made larger, the size is too small, the statistical average points are too small, and the noise reduction effect is not ideal; of course, it can't be too big, it is too big, and the other is to affect the operation. Speed, second, can not further increase the noise reduction effect, and may also cause excessive smoothing.
  • Steps 310 to 340 are: during the execution of the algorithm, the algorithm parameters are set according to the common features of the dual-energy data; steps 350-380 are performed during the execution of the algorithm, and the actual image is reduced according to the algorithm parameters determined in steps 310-340. Noise operation.
  • step 310 high-energy transparency and low-energy transparency are selected as two-dimensional features, because the dual-energy system can directly obtain only two-dimensional features; two-dimensional features are selected instead of only high-energy transparency or low-energy transparency as one-dimensional Features are to ensure that the similarity points are judged more accurately.
  • the key in step 320 is to analyze the statistical fluctuation range of transparency.
  • the statistical fluctuation of transparency is determined by the overall design level of the system. One way to obtain statistical fluctuations is to: Calculate the relative standard deviation of transparency in a flat region. It is generally believed that the noise is 2.355 times the relative standard deviation. The difference in transparency between similar points can be set according to the noise level.
  • step 330 the setting of the neighborhood size is related to the statistical fluctuation range of the data in step 320.
  • the greater the statistical fluctuation the more the noise reduction should be increased accordingly, so the neighborhood area should be set larger.
  • the larger the neighborhood area is set the slower the algorithm runs.
  • step 340 the setting of the weighting coefficients of the similar points different from the center point distance is not a critical step of the algorithm.
  • the system is carried out by means of equal weight addition.
  • a similar point search range is determined for the center pixel according to the neighborhood area determined in step 330. If the difference between the high-energy transparency and the low-energy transparency of a pixel in the neighborhood and the center pixel is smaller than the difference in transparency between the similar points determined in step 320, it is considered to be the similarity of the center pixel.
  • step 360 the weighted average of the high energy transparency and the low energy transparency between the similar points is weighted and averaged according to the weighting average determined in step 340 to obtain a weighted average of the high energy transparency and the low energy transparency.
  • step 370 alphax and alphay are calculated using the noise reduction result outputted in step 360, that is, the weighted average of high energy transparency and low energy transparency. See Figure 5 for the calculation formula.
  • the alpha curve method is used to design the material recognition algorithm.
  • the characteristic axes are alphax and alphay (the alpha curve coordinates are defined in Figure 5).
  • Using a weighted average of high energy transparency and low energy transparency calculate alphax and alphay, and then material resolution.
  • the basis for material resolution is the category boundary line obtained by the automatic calibration module. It can be seen from Fig. 6A that the alpha curve is monotonic in the direction of the atomic number, which is the basis of the dual energy material resolution algorithm. For high energy dual energy, corresponding to the same alphax point, the atomic number of the material increases while the alphay value decreases monotonically.
  • the alpha data of a point to be checked is (alphaxR, alphayR), and the table can obtain three demarcation points corresponding to alphaxR, namely alphayC-Al, alphayAl-Fe, alphayFe-Pb, by comparing alphayR in alphayC-Al, alphayAl -Fe, the size relationship in alphayFe-Pb, can determine the material information of the point to be checked.
  • Material identification is based on a weighted average of similar points. Therefore, the impact of statistical fluctuations on classification accuracy has been reduced to a minimum.
  • the material resolution between the points is not smooth.
  • the noise reduction algorithm is similar to the preprocessing algorithm. The key point is to select the neighborhood size and determine the similarity point. The algorithm steps are illustrated in the flow of Figure 7B.
  • Step 410 440 is to set algorithm parameters according to common features of the dual-energy data during the execution of the algorithm; Steps 450-470 are to perform noise reduction operations on the actual image according to the algorithm parameters determined in steps 410-440 during the execution of the algorithm. .
  • High energy transparency and low energy transparency are still selected in step 410 as a two dimensional feature.
  • the pre-processing algorithm and the material identification algorithm have been added, the feature of preliminary material identification has been added.
  • this increased dimensional feature is based on the first two dimensions and does not increase the amount of information.
  • the noise of the preliminary results of material identification is relatively large, it will affect the judgment of similar points, so this dimension feature is not used.
  • the key in step 420 is to analyze the statistical fluctuations in transparency.
  • the statistical fluctuations in transparency are determined by the overall design level of the system.
  • One way to obtain a statistical fluctuation is to: Count the relative standard deviation of the transparency in a flat area. It is generally believed that the noise is 2.355 times the relative standard deviation. The difference in transparency between similar points can be set according to the noise level.
  • step 430 the setting of the neighborhood size is related to the statistical fluctuation range of the data in step 420.
  • the greater the statistical fluctuation the more the noise reduction should be increased accordingly, so the neighborhood area should be set larger.
  • the larger the neighborhood area is set the slower the algorithm runs.
  • the selected neighborhood area is 11 pixels * 11 pixels.
  • step 440 the setting of the weighting coefficients of the similar points different from the center point distance is not a critical step of the algorithm.
  • the system is carried out by means of equal weight addition.
  • step 450 a similar point search range is determined for the center pixel according to the neighborhood area determined in step 430. If the difference between the transparency of the pixel and the transparency of the central pixel in the neighborhood is less than the difference in the transparency between the similar points determined in step 420, it is considered to be the similarity of the central pixel.
  • step 460 the preliminary results of the material identification between the similar points are weighted and averaged according to the weighting coefficients determined in step 440 to obtain a weighted average of the material identification results.
  • step 470 the weighted average of the material identification results obtained in step 360 is output as the final material identification result.
  • Figure 8 shows a comparison of the effect diagrams of the different pre-processing parameters in the material resolution module. It can be seen that when the pretreatment is not performed, the material identification result is very noisy, and the material resolution accuracy is greatly affected, as shown in (A) of Fig. 8; the smaller neighborhood (3*3) is selected for preprocessing, The noise is reduced, but the degree of noise reduction is very limited, as shown in (B) of Figure 8; the larger neighborhood (11*11) is selected for pre-processing, but the similarity point is not judged, although the image The noise is greatly reduced, but in the edge region of the object, because the data of the two materials are mixed together, the material resolution result is obviously wrong, as shown in Fig. 8(C); select a larger neighborhood (11*11) Pre-processing, and similar point judgment, can play a good role in noise reduction without affecting the edge area, as shown in (D) of Figure 8.
  • the basic idea of the gray-scale fusion module is that the low-energy image is the main proportion for the area that penetrates the low-quality thickness. For the area that penetrates the thick-thickness, the high-energy image is the main proportion.
  • the key is the design of the correspondence between material information and color tones and the design of the color table.
  • color display standards for low-energy X-ray dual-energy systems there are color display standards for low-energy X-ray dual-energy systems in the world. Organic substances are shown in orange, light metals are shown in green, and inorganic substances are shown in blue. The color standard of the high-energy X-ray dual-energy system has not yet been formed. Considering the user's visual habits, the color display standard can be extended by the low-energy X-ray dual-energy system, and the organic substance is still represented by orange, the light metal by green, and the inorganic substance by blue. In addition to organic, light metal, and inorganic substances, the high-energy X-ray dual-energy system can also distinguish heavy metal categories. The display of this category has no ready-made standards. It can be set according to the designer's preference, considering the continuity of color tone usage. The system represents heavy metals in purple tones.
  • the key to the colorization algorithm is the design of the color table and how the atomic numbers and fused gray and color information are mapped.
  • the colorization process is shown in Figure 9. Steps 510-550 show the design principles of the color table; steps 610-650 show the mapping principle of atomic number and fused gray and color information.
  • the atomic number is corresponding to the hue H (H in the HLS color model), and the different atomic numbers are represented by different hue, and the material-resolved categories are equal to the hue categories.
  • the categories can be refined, not only the substances are divided into four categories: organic matter, light metal, inorganic matter and heavy metal, but also further refined into several sub-groups based on the atomic number. category.
  • the design of the hue is based on the principle of color display standards, taking into account the color transition, from orange red to orange yellow, yellow green to blue green, blue blue to dark blue, purple blue to purple.
  • an appropriate saturation value is set for each color tone based on visual theory.
  • the human eye is more sensitive to green tones, generally more sensitive to orange tones, and less sensitive to blue tones, so you can set a lower saturation for the green tones and a medium to high saturation for the orange tones.
  • the hue sets a higher saturation. According to this principle, the corresponding rules of saturation and hue are designed.
  • the key is to design a mapping relationship between the fused gray level and the L value (L in the HLS color model). In order to ensure the harmony of the hue, in the color table of different hues, the brightness perceived by the human eye is equivalent to the color of the same brightness level, and the gradation gradation cannot be simply equivalent to the L value.
  • This step takes advantage of the Y value in the YUV color model.
  • the Y value in the YUV color model represents the brightness perceived by the human eye
  • the L value in the HLS color model does not represent the brightness perceived by the human eye. If the color table is simply designed according to the fusion gray level and the L value, The colors of different tones are not coordinated. Therefore, it is necessary to synthesize the conversion relationship of the YUV model, the RGB model, and the HLS model, and design the mapping relationship between the gradation and the L value according to the principle that the gradation is equivalent to the Y value, thereby obtaining the HLS color table.
  • the Y values of the same brightness level should be equal; therefore the problem translates into a very simple mathematical problem, known in the HLS color model.
  • the hue H and saturation S, as well as the Y value in the YUV model require the determination of the luminance L value in the HLS color model; since the YUV color model does not have a direct conversion relationship with the HLS color model, it can be performed by the RGB model as an intermediary.
  • step 540 the conversion of the HLS color table to the RGB color table is completed.
  • the relationship between the two is calculated in advance, stored in a table, and the RGB values are directly obtained by looking up the table.
  • Construct a two-dimensional mapping table of HLS to RGB the first dimension is the material information index, the second dimension is the fused gray index, and the RGB value is stored in the table.
  • the real-time colorization process is a simple look-up process.
  • the material information output by the material resolution module and the fused gray information output by the gray-scale fusion module are used to find the two-dimensional RGB color table, and the color can be obtained. image.
  • step 610 the material information output by the material resolution module and the fused gray information output by the gray fusion module are obtained as input information of the colorization algorithm.
  • step 620 the first dimensional index value of the RGB two-dimensional mapping table is determined based on the material information (corresponding to the color hue).
  • step 630 the second dimensional index value of the RGB two-dimensional mapping table is determined according to the fused gray level information (color brightness level).
  • step 640 according to the two-dimensional index values determined in steps 620 and 630, the two-dimensional RGB color table is looked up to obtain RGB values.
  • steps 610-640 are repeated to determine the RGB values point by point and output a color image.

Landscapes

  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Toxicology (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)

Description

物质识别方法和设备 技术领域
本发明涉及辐射成像技术,具体涉及一种应用于高能 X射线双能成像检査系统的 物质识别方法和设备, 不仅能够获得被检查物体的透射图像, 而且能够获得被检査物 体中的材料信息。 背景技术
以集装箱为单位进行运输, 是一种现代化、 先进的运输方式。 集装箱化已经成为 国际货物运输的大趋势。 与此同时, 利用集装箱走私, 偷运枪支、 武器、 毒品、 爆炸 物甚至大规模杀伤性武器 (Weapons of Mass Destruction, WMDs) 和放射性散布装置
(Radiological Dispersal Devices, RDDs), 已经成为困扰各国政府、 干扰国际货物运输 正常秩序的国际公害。
美国 911事件之后, 美国政府开始重视货运的潜在风险, 最担心 WMDs和 RDDs 通过集装箱运入美国。 为了防范这种风险, 2001年 1月 17日, 美国海关发布了 "集 装箱安全倡议"(Container Security Initiative, CSI), 要求所有具有直接通航美国港口 业务的外国港口必须装有非侵入式 ( 射线扫描成像设备,对运往美国的集装箱进行 射线扫描检査。 CSI公布 1年后, 就有 18个世界大港口加入该倡议并开始运行。在国 际运输安全要求日益提升的大环境下,世界海关组织全体一致通过决议,要求全体 161 个成员国沿着 CSI模式发展相关集装箱安全检察计划——集装箱安全检查已经成为全 世界共同关注的课题。
目前的集装箱 ( 射线安全检査设备以透射成像为主, 采用 X射线直接透射货 物, 得到 X射线路径覆盖的所有物品的透射图像。标准的透射成像技术解决了集装箱
"可视化" 问题, 得到了广泛应用。 但是这类设备普遍存在以下不足: 首先, 二维结 构信息易受射线路径上重叠物品干扰; 其次, 不包含密度信息; 第三, 不包含材料信 息。
针对 "査私"需求, 主要方式是对照报关单和集装箱的髙能 X射线扫描图像, 检 査是否相符, 报关单就是先验知识, 标准的 X射线透射成像技术基本就能满足需求。 但是, CSI的提出, 使得集装箱检査需求从检査走私物品 (简称 "査私") 向检査危禁 品 (简称 "査危") 发展。 由于危禁品种类繁多, 而且没有固定的形状, 等于对被检 集装箱中的物品没有了先验知识, 因此, 仅仅依靠标准的 X射线透射成像技术, 已经 很难满足集装箱安全检查的需求。
根据 WMDs、 RDDs及其它危禁品的特点, 获取更丰富的被检物特征信息, 才能 实现更准确、有效的安全检查。双能透射技术是利用两种能谱不同的 X射线穿透被检 物, 其输出信号的差异经过处理, 得到被检物的材料原子序数信息。 因此, 利用该技 术, 将在一定程度上有效地提升安全检査水平。希望高能 X射线成像集装箱检查系统 具有材料分辨能力, 己经成了近几年来国际研究的一个热点。
虽然双能技术在 X射线能量低于 200KeV时非常有效,在行李物品检查中得到了 广泛应用, 但是, 穿透集装箱的 X射线能量达到几 MV, 对于相同质量厚度的不同材 料, 如。、 Al、 Fe, 在这个能量段内衰减对射线的衰减差异很小。 因此, 釆用高能 X 射线成像得到的材料分辨能力与低能双能 X射线技术相比要差很多。甚至集装箱检査 领域的部分专家认为, 双能成像技术在 X射线能量髙于 200KeV时, 几乎没有什么效 果, 因此不适合集装箱检查系统。 发明内容
本发明提出了基于能谱整形装置与自动标定装置的物质识别系统, 解决了高能 X 射线双能成像集装箱检查系统中实时有效地进行材料分辨的问题, 同时利用双能灰度 融合和彩色化等算法, 把材料分辨和灰度信息视觉化。
与传统的髙能 X射线成像相比, 本系统利用获取的双能图像, 即能得到兼备高穿 透力与高反差灵敏度的融合图像, 又能得到货物的材料信息, 对于爆炸物、 毒品及放 射性物质等危禁品具有一定的识别能力, 从而提高了集装箱的安全检查能力。
另外, 从传统的高能 X射线成像系统输出的灰色或者伪彩色图像, 到高能 X射 线双能成像系统输出的材料分辨彩色图像, 本发明充分利用人眼视觉的优势, 力求把 更多信息传递给用户。
在本发明的一个方面, 提出了一种物质识别方法, 包括步骤: 用高能射线和低能 射线透射被检物体, 以获得被检物体的高能透射图像和低能透射图像, 所述高能图像 中的每个像素的值表示高能射线对被检物体的相应部分的高能透明度, 而所述低能图 像中每个像素的值表示低能射线对被检物体的相应部分的低能透明度; 针对每个像 素, 计算所述高能透明度的第一函数的值和所述高能透明度和所述低能透明度的第二 函数的值; 以及通过利用预先创建的分类曲线对由所述第一函数的值和所述第二函数 的值所确定的位置进行分类来识别被检物体中与每个像素所对应的那部分的物质的 类型。
根据本发明的实施例, 该方法还包括步骤: 设置具有预定大小的邻域; 以及在每 个像素的邻域中对所述高能图像和所述低能透射图像进行降噪。
根据本发明的实施例,在每个像素的邻域中对所述高能图像和所述低能透射图像 进行降噪的步骤包括: 在邻域中寻找与中心像素相似的像素, 作为相似像素; 对邻域 中的相似像素进行加权平均。
根据本发明的实施例,所述相似像素的高能透明度和低能透明度与所述中心像素 的高能透明度和低能透明度之间的差异小于预定值。
根据本发明的实施例,将所述被检物体识别为有机物、轻金属、无机物或重金属。 根据本发明的实施例, 该方法还包括对识别结果进行彩色化显示的步骤。
根据本发明的实施例,所述彩色化显示步骤包括:对每个像素的髙能透明度和低 能透明度进行加权平均, 作为该像素的融合灰度值; 根据被检物体中与该像素相对应 部分的材料类型来确定颜色色调; 根据该像素的融合灰度值来确定该像素的亮度等 级; 将所述颜色色调和所述亮度等级作为索引来从预先创建的查找表中 ^得该像素的
R值、 G值和 B值。
根据本发明的实施例,根据被检物体中与该像素相对应部分的材料类型来确定颜 色色调的步骤包括: 将橙色赋予有机物、 将绿色赋予轻金属、 将蓝色赋予无机物、 将 紫色赋予重金属。
根据本发明的实施例,该方法包括步骤:对从射线源发出的射线进行能谱整形来 扩大高能射线和低能射线之间的能谱差异。
根据本发明的实施例,所述分类曲线是通过下面的步骤针对每种标定材料而创建 的: 用高能射线和低能射线照射各种厚度下标定材料来获得相应的高能透明度和低能 透明度; 将高能透明度的第一函数作为横坐标, 将低能透明度和高能透明度的第二函 数作为纵坐标来形成标定材料在不同厚度下的点; 基于所述的点形成所述分类曲线。
根据本发明的实施例,基于所述点形成所述分类曲线的步骤包括:采用最小二乘 曲线拟合法对所述点进行曲线拟合。
根据本发明的实施例, 基于所述点形成所述分类曲线的步骤包括: 采用切比雪夫 意义下的最佳拟合多项式对所述点进行曲线拟合。
根据本发明的实施例, 该方法还包括对所述分类曲线进行离散化的步骤。 在本发明的另一方面, 提出了一种物质识别设备, 包括: 图像形成装置, 用髙能 射线和低能射线透射被检物体, 以获得被检物体的高能透射图像和低能透射图像, 所 述高能图像中的每个像素的值表示高能射线对被检物体的相应部分的高能透明度, 而 所述低能图像中每个像素的值表示低能射线对被检物体的相应部分的低能透明度; 计 算装置, 针对每个像素, 计算所述高能透明度的第一函数的值和所述高能透明度和所 述低能透明度的第二函数的值; 以及分类装置, 通过利用预先创建的分类曲线对由所 述第一函数的值和所述第二函数的值所确定的位置进行分类来识别被检物体中与每 个像素所对应的那部分的物质的类型。
根据本发明的实施例, 该设备还包括: 设置具有预定大小的邻域的装置; 以及在 每个像素的邻域中对所述高能图像和所述低能透射图像进行降噪的装置。
根据本发明的实施例,在每个像素的邻域中对所述高能图像和所述低能透射图像 进行降噪的装置包括: 在邻域中寻找与中心像素相似的像素作为相似像素的装置; 对 邻域中的相似像素进行加权平均的装置。
本发明的物质识别分系统内嵌于高能 X射线双能成像集装箱检査系统,通过能谱 整形装置的设计, 对射线能谱进行整形, 提高了材料分辨能力。 另外, 通过标定装置 的设计, 实时监测系统状态, 获取与系统状态最为匹配的分类参数, 为准确进行材料 分辨提供了坚实的基础。 另外, 通过材料分辨模块中快速识别算法与图像降噪算法的 结合, 既保证了算法运行的实时性, 又大幅度地降低了 X射线数据统计涨落的影响, 保证了材料分辨的准确率。
另外, 本物质识别系统通过设计灰度融合算法, 得到兼顾穿透力与反差灵敏度的 融合图像, 单从灰度上就能得到比单能系统更多的信息。
得到了材料分辨的结果以及融合灰度图后, 本发明还进一步设计了彩色显示模 块, 从输入原始双能透射数据, 到输出 RGB彩色显示数据, 保证了整个数据处理过 程的完整性。
另外, 本系统还充分考虑了实时运行的问题, 对各种算法进行了优化设计, 运行 速度快、 实时性好。
本发明可以有效地解决高能 X射线双能检査系统的材料分辨能力差的问题,具有 很好的材料分辨效果和彩色显示效果, 并且可操作性好, 运行速度快, 具有很高的实 用价值。 附图说明
通过下面结合附图说明本发明的优选实施例, 将使本发明的上述及其它目的、 特 征和优点更加清楚, 其中:
图 1A示出了高能 X射线双能成像系统的基本组成示意图;
图 1B示出了内嵌有材料识别设备的高能 X射线双能成像系统的示意图; 图 1C示出了质量衰减系数曲线;
图 2是根据本发明实施例的能谱整形装置的示意图, 图中, 黑色箭头表示高能档 射线, 灰色箭头表示低能档射线, 黑色区域表示整形材料块;
图 3是根据本发明实施例的自动标定装置的示意图, 图中, 黑色箭头表示高能档 射线, 灰色箭头表示低能档射线;
图 4A示出了根据本发明实施例的自动标定过程的流程图;
图 4B示出了根据本发明实施例的物质识别方法的流程图;
图 5示出了 alpha曲线图坐标定义;
图 6A示出了自动标定过程中使用的标定材料训练数据的示意图;
图 6B示出了从标定材料训练数据生成的 alpha曲线的示意图;
图 6C示出了标定材料训练数据的统计结果;
图 7A示出了高低能透明度降噪处理过程的流程图;
图 7B示出了对材料识别结果进行的降噪处理的流程图;
图 8是材料分辨模块针对不同的预处理参数产生的成像结果对比;
图 9是说明颜色表的生成过程和 RGB图像的生成过程的示意图;
图 10示出了灰度图与彩色图效果对比。 具体实施方式
下面参照附图对本发明的优选实施例进行详细说明, 在描述过程中省略了对于本 发明来说是不必要的细节和功能, 以防止对本发明的理解造成混淆。
根据本发明实施例的物质识别设备包括硬件部分和数据处理算法。 物质识别设备 是内嵌于高能 X射线双能成像集装箱检査系统的分系统,它基于高能双能透射数据进 行材料分辨。
为了方便说明, 以下称高能 X射线双能中能量高的射线为高能档 X射线, 能量 低的射线为低能档 X射线。 能量段的合理选择是高能 X射线双能成像集装箱检査系 统的前提。 高能 X射线双能的能量段选择范围一般在 3MeV~10MeV之间。 理论上, 在合适能量范围内, 能量差越大, 材料分辨能力越好。 但是, 能量差太大, 高能档和 低能档射线的穿透能力相差太大, 则能进行材料分辨的有效范围很窄。
图 1A和图 IB示出了高能 X射线双能成像集装箱检査系统的示意图。如图 1A所 示, 高能 X射线双能成像集装箱检査系统包括射线发生装置 10、 机械传动装置 (未 示出)、 诸如集装箱之类的被检査物体 20、 数据采集分系统 30、 扫描控制计算机和数 据处理计算机 (未示出)。
射线发生装置 10包括双能加速器和其他的辅助设备, 能够以很高的频率交替产 生两种能量的 X射线束。 机械传动装置能够使射线发生装置 10及数据采集分系统 30 一起相对于待检集装箱 20产生水平方向的相对运动。
可以是射线发生装置 10和数据采集分系统 30不动, 而待检集装箱 20运动。 作 为另一选择, 也可以是待检集装箱 20不动, 而射线发生装置 10和数据采集分系统 30 一起运动。
数据采集分系统 30主要包括线阵探测器, 用于探测射线发生装置 10产生的双能 量 X射线束穿过检测物体后的射线, 产生双能透射数据, 并将数据传输到计算机(未 示出)。该数据采集分系统 30还包括探测器上的投影数据读出电路和逻辑控制单元等。 探测器可以是固体探测器、 气体探测器、 半导体探测器。
扫描控制计算机负责检査系统运行过程的主控制, 包括机械控制、 电气控制和安 全连锁控制等。 数据处理计算机负责对数据采集分系统获得的双能透射数据进行处 理, 并显示。
为了提高双能系统的材料分辨能力, 以改善物质识别效果, 引进了能谱整形装置 40和自动标定装置 50, 见附图 1B。
能谱整形装置 40包括能谱整形材料以及相应的辅助设备。 能谱整形装置 40放置 在射线发生装置 10与被检物体 20之间, 目的是在射线穿透被检物体 20之前, 对射 线发生装置 10输出的射线的能谱进行整形, 以期能谱分布更有利于材料分辨。
能谱整形材料的特点是对能量低的射线衰减很大、 能量高的射线衰减则很小。 此 特性越好, 则能谱整形的效果越佳。 只要可以满足此特性, 即可作为能谱整形材料。 基于能谱整形材料的特性, 能谱整形后, 射线的等效能量得到提高。 如果只把整形材 料作用在高能档射线上, 则高能档射线等效能量得到提高, 而低能档射线的等效能量 不变, 从而拉大了双能之间的能量差, 以提高系统的材料分辨能力。 基于此特性, 选择石墨材料作为整形材料。 从纯理论角度考虑, 整形材料越厚, 材料分辨能力越好。 但是, 考虑统计涨落, 整形材料越厚, 对射线的衰减程度越大, 探测器接收到的剂量越低, 数据的信噪比越低。 因此, 整形材料的厚度有个最佳值。 这个最佳值需要根据系统的实际情况确定。 根据高能档和低能档射线能量的分布情 况, 确定能谱整形是只针对某能档进行, 图 2A示出了转盘形式的能谱整形装置的示 意图。 或者, 对双能档均进行能谱整形, 图 2B示出了对双能挡都可进行能谱整形的 能谱整形装置。
能谱整形装置 40 的设计应该根据能谱整形的需求而定。 可以只对高能档射线进 行整形, 通过提高高能档射线的等效能量拉大双能之间的能量差, 从而提高系统的材 料分辨力。 也可以对高低能档同时进行能谱整形, 这种情况比较特殊, 一般是对于低 能档射线在 3M附近的情况。从图 1C所示的质量衰减系数曲线上可以看到, 3MeV能 量段附近, 低 Z材料的衰减系数接近, 并且变换趋势都非常缓慢。 因此, 在这个能量 段附近,能量变化对于低 Z材料的分辨能力影响很小,而高 Z材料的衰减系数在 3MeV 附近则是一个拐点。 这个现象将会导致在这种能量选择下铅材料与其它材料不可区 分。 因此, 对 3MeV低能档能量也进行能谱整形, 利用能谱整形材料吸收了低能档能 量中的低能部分, 可以提高了高 Z材料的可区分性, 并且对低 Z材料没有负面影响。
图 3是根据本发明实施例的自动标定装置的示意图。 自动标定模块包括硬件上自 动标定装置 50的设计和软件上自动标定流程的设计两大部分。
自动标定装置 50包括呈阶梯分布的标定材料以及相应的辅助装置。 自动标定装 置 50的目的是采集标定数据, 进入自动标定流程, 实时获取与系统状态匹配的分类 参数, 保存在文件中, 作为材料分辨模块的输入。
这里的标定材料包括各类别的典型材料, 为了保证标定精度, 每种类别至少准备 一种典型材料, 也可每种类别各准备若干种等效原子序数各不同的材料。 如果材料不 好准备, 或者放置自动标定装置 50的空间有限, 中间类别的材料可以省略, 自动标 定算法利用相邻类别的数据插值代替。 标定材料选择与系统的材料分辨要求有关。 高 能 X射线双能要求可以区分有机物、 轻金属、 无机物、 重金属四种类别, 因此, 从四 种类别选择了四种典型材料, 依次为石墨 (Z=6)、 铝 (Z=13)、 铁 (Z=26)、 铅 (Z=82)。 选 择这四种材料基于两种原因, 一是材料比较常见, 二是均属于单质, 性质稳定。
每种材料从薄到厚设计若干级阶梯。 最薄和最厚的厚度由系统的材料分辨范围决 定。 而阶梯级数的数目则由标定的精度及放置自动标定装置的空间共同决定。 辅助装置主要提供机械传动, 实现定位扫描, 以获取每种材料每个阶梯的双能透 射数据。 在每个定位点要求连续扫描若干列双能透射数据, 建议扫描 256列以上, 这 样可以较大程度地消除信号统计涨落的影响。
在高度方向上, 探测器臂架上的不同探测器接收的 X射线角分布是不同的。不同 角分布, 能谱分布是有差异的, 导致材料分辨参数是不同的。 因此, 考虑到 X射线角 分布的影响, 可以把所有探测高度划分成若干区域, 每个区域独立统计, 生成分类参 数。 这就要求自动标定装置 50中标定材料应该覆盖所有感兴趣的探测区间。
如果标定材料的高度受客观因素 (加工能力、 设备空间等) 限制, 无法覆盖臂架 上所有的探测器模块, 一种简化的方式如下: 一般情况下, 最感兴趣的探测高度在集 装箱内摆放货物的位置,一般系统都会把 X射线的主束调整到该位置附近。 因此, 射 线主束方向是重点标定对象。 标定材料可以设计为只覆盖该区域, 得到的双能透射数 据作为参数输入至自动标定算法中, 生成 X射线主束方向的能谱分布对应的分类参 数, 作为所有探测区域的分类参数。 该简化方式在 X射线角分布较小的情况下, 是在 误差允许范围内的。
自动标定装置 50 中标定材料可以设计成任何形状, 只要可以满足上述要求。 在 图 3中, 阶梯的级数和厚度仅仅是用于说明的目的, 并不表示实际的含义。
图 4A示出了根据本发明实施例的自动标定过程的流程图^ 如图 4A所示, 在方 框 110, 射线发生装置 10产生 X射线束。 在方框 120, X射线束被能谱整形装置 40 整形。 在方框 130, 要进行自动标定处理时, 人为触发并且运行自动标定流程, 实时 获取原始标定数据。
然后, 在方框 140, 对原始标定数据进行数据校正处理。 在方框 150, 运行自动 标定算法, 生成分类参数并且保存在文件中。
然后, 在方框 150, 调用自动标定算法计算与当前系统状态匹配的分类参数。 图 4B示出了根据本发明实施例的物质识别方法的流程图。如图 4B所示,在方框 210, 射线发生装置 10产生 X射线束, 然后在方框 220, 能谱整形装置 40对该 X射 线束进行整形, 以获得有利于材料分辨的 X射线束。
在方框 230, 整形后的 X射线束穿透被检物体 20, 以获得被检物体的原始双能数 据。 接下来, 在方框 240, 运行数据校正模块, 对原始双能数据进行数据校正, 以消 除探测器本底、 探测器不一致性及射线剂量波动等的影响。 经过校正后的数据用于材 料分辨和双能灰度融合处理。 接下来, 在方框 250, 在自动标定过程中产生的分类参数文件输入到材料分辨模 块中, 并且基于校正后的双能数据, 对被检物体的材料进行识别, 产生材料信息。
另一方面, 在方框 260, 经过校正的双能数据输入到双能灰度融合模块中, 进行 灰度融合处理, 产生双能数据融合后的被检物体透射图像。 此时, 结合材料分别模块 输出的材料信息, 在方框 280, 进行彩色化处理, 也就是根据被检物体中包含的材料 信息将适合于灰度显示的透射图像数据转换成适用于彩色显示的 RGB数据, 然后在 方框 290, 在数据处理计算机上进行显示 290。
如上所述, 每当系统状态发生改变的时候, 自动标定流程由人为触发, 启动自动 标定装置 50, 采集经过能谱整形的原始标定数据, 由数据采集分系统送至数据处理计 算机。 采用 alpha曲线法来设计材料分辨算法。 因此, 自动标定算法的目的就是计算 与系统状态匹配的 alpha曲线图分类参数。 通过调用自动标定算法, 获得系统状态匹 配的 alpha曲线图分类参数, 并保存在文件中, 作为材料分辨模块的参数输入。 alpha 曲线图坐标定义见图 5。
如图 5所示, 定义 alphaL和 alphaH如下:
alphaL = (l-log(TL))*1000; 其中 TL为低能透明度;
alphaH = (l-log(TH))* 1000; 其中 TH为髙能透明度。
取 alphaH为 alpha曲线的横坐标 alphax,取 alphaL与 alphaH之差作为 alpha曲线 的纵坐标 alphay:
alphax = alphaH = (l-log(TH))*1000;
alphay = alphaL - alphaH = (-log(TL)+log(TH))* 1000。
如上所述, 在方框 130, 调用数据校正模块, 对原始标定数据进行数据校正, 消 除探测器本底、 探测器不一致性及射线剂量波动等的影响, 得到标定材料训练数据。 图 6A为某一探测区间的训练数据在 alpha曲线图上的示意。
下面详细说明从标定材料训练数据生成各类材料之间的类别分界线的过程。
(i)在某一探测区间范围内, 依次对各种材料各个阶梯的若干列校正后的双能数 据进行均值统计, 从而得到标定材料训练数据的一系列均值点。 图 6C为某一探测区 间的训练数据均值点在 alpha曲线图上的示意。
(ii) 在 alpha曲线图 6B上, 连接某种材料的若干个训练数据均值点, 即可得到 该材料的 alpha离散曲线。 但是, 由于标定材料的阶梯数有限, 因此, 直接连接而成 的 alpha离散曲线的精度很低。 为此, 采用最小二乘曲线拟合法 (用最小二乘法求给 定数据点的拟合多项式)进行曲线拟合, 把若干个训练数据均值点作为输入参数, 进 行曲线拟合, 得到该曲线的拟合参数, 即多项式各阶次的系数, 其中, 拟合多项式的 次数根据实际情况选定。 曲线拟合也可以采用其它拟合方法, 如切比雪夫意义下的最 佳拟合多项式。
(iii) 对 alpha曲线 x轴进行离散化, 离散精度根据需要而定。 然后, 利用曲线 拟合参数, 计算每个离散点对应的 y轴数据。 通过这步操作, 得到了该材料的离散化 alpha曲线。
(iv) 重复步骤 (ii) (iii), 直至得到所有材料的离散化 alpha曲线。
(v) 从图 6B可以看到, alpha曲线在原子序数方向是具备单调性的, 这也正是 双能材料分辨算法的依据。 因此, 得到各种材料的离散化 alpha曲线, 就可以依次计 算相邻两条曲线的离散化分界线, 如图 6C示意。
• 四种类别的划分依据为等效原子序数: Z=l~10 划分为有机物类别; Z=10~18划分为轻金属类别; Z=18~57划分为无机物类别; Z > 57划分为 重金属类别。 而四种典型材料分别选用的是石墨 (Z=6)、 铝 (Z=13)、 铁 (Z=26)、 铅 (Z-82)。 根据石墨 (Z=6)材料的离散化 alpha曲线和铝 (Z=13)材 料的离散化 alpha曲线加权平均得到原子序数 Z=10的离散化 alpha曲线, 即得到了有机物和轻金属的类别分界线。其中, 加权平均的权值可以简单 地根据原子序数计算, 即假设不同原子序数范围内的可区分性是相同的。 虽然, 从严格意义上讲, 不同原子序数范围内的可区分性是有差异的, 但 是, 由于髙能双能不同于低能双能, 其材料分辨能力比较弱, 只能区分属 于不同类别的材料, 而不能精确区分原子序数不同的材料, 因此, 这种差 异是可以接受的。
• 同样, 根据铝 (Z=13)材料的离散化 alpha 曲线和铁 (Z=26)材料的离散化 alpha曲线加权平均得到原子序数 的离散化 alpha曲线, 即得到了轻 金属和无机物的类别分界线; 根据铁 (Z-26)材料的离散化 alpha曲线和铅 (Z=82)材料的离散化 alpha 曲线加权平均得到原子序数 Z=57 的离散化 alpha曲线, 即得到了无机物和重金属的类别分界线。
(vi) 重复步骤 ( i:), (ii), (iii), (iv), (v), 直至得到所有探测区间的离散化 类别分界线。
把各个探测区间、 各种典型材料的类别分界线数据按照约定的格式保存在文件 中, 作为材料分辨模块的分类参数。
如上所述, 材料分辨是双能 X射线系统有别于单能 X射线系统的特征。 由于高 能 X射线成像得到的材料分辨能力与低能双能 X射线技术相比要差很多, 因此, 材 料分辨模块不仅需要考虑如何正确分类, 还要考虑如何提升材料分辨效果。
首先, 对高低能透明度进行降噪预处理, 然后利用降噪结果进行材料分辨, 最后 对材料分辨结果进一步进行降噪处理。 如果系统对于处理速度要求非常高, 材料识别 前后两步降噪处理可以只保留一步, 仍可以保证较好的材料分辨效果。
由于高能 X射线双能所处能量段的限制, 其材料分辨能力与低能双能 X射线技 术相比要差很多, 而 X射线的统计涨落是固有的。 因此需要对数据进行预处理, 对双 能图像分别进行降噪, 否则材料分辨的准确率会受到很大的限制。 不进行降噪之前, 在材料分辨能力确定的前提下, 统计涨落幅度越大, 则分类准确率越低, 当分类准确 率低于一定程度的时候, 则判定为分类错误。 为了提高分类准确率, 保证材料分辨效 果, 需要设计有效的预处理算法进行降噪。 预处理的目的是降低数据的噪声, 使得数 据尽量逼近真值, 以增加材料分辨的准确性, 其降噪程度由系统的噪声水平决定。 预 处理算法的设计灵活度比较大, 只要可以实现降噪的目的即可。 本发明为了尽可能地 材料分辨效果, 预处理算法设计过程需要重点考虑邻域选择和相似点定义两个问题:
(a) 邻域的选择: 邻域应该适当地取得大些, 取得过小, 统计平均的点数太少, 降噪效果不理想; 当然, 也不能取得太大, 取得过大, 一是影响运行速度, 二是不能 进一步增加降噪效果, 可能还会引起过度平滑。
(b) 相似点定义: 选择较大的邻域后, 统计平均过程中容易引起边缘模糊问题。 为了避免在降噪的同时, 影响边缘区域, 需要在统计过程中对邻域中的点增加限制条 件, 符合条件的称为相似点, 只有相似点对均值统计有影响, 而与中心点属于不同区 域的点不符合限制条件, 不在统计范围内, 故对均值没有影响。
预处理算法步骤见图 7A流程示意。
其中步骤 310~340是在算法执行过程中,根据双能数据的通用特征设定算法参数; 步骤 350~380是在算法执行过程中, 根据步骤 310~340确定的算法参数, 对实际图像 进行降噪操作。
步骤 310中选定高能透明度和低能透明度为二维特征, 这是因为双能系统可以直 接得到的有且只有这两维特征; 选用两维特征, 而不是只选择高能透明度或者低能透 明度作为一维特征, 是为了保证相似点判断更为准确。 步骤 320中关键是分析透明度统计涨落幅度。 透明度的统计涨落幅度是由系统整 体的设计水平决定的。 获取统计涨落幅度的一种方式为: 统计一个平整区域内透明度 的相对标准差, 一般认为, 噪声为相对标准差的 2.355倍。 根据噪声水平即可设定相 似点间的透明度差异幅度。
步骤 330中, 邻域大小的设置与步骤 320中数据统计涨落幅度相关。 统计涨落幅 度越大, 则降噪力度应相应地增大, 故邻域面积应设置得大些。 另一方面, 邻域面积 设置得越大, 算法的运行速度越慢。 通过对实际数据的处理, 比较不同邻域的处理效 果, 我们系统选择的预处理邻域面积为 5像素 *5像素。
步骤 340中, 与中心点距离不同的相似点的加权系数的设置不是这个算法的关键 步骤。 本系统采用等权相加的方式进行。
步骤 350中, 根据步骤 330中确定邻域面积, 为中心像素确定相似点搜索范围。 如果邻域内某像素与中心像素的高能透明度和低能透明度的差异均小于步骤 320中确 定的相似点间的透明度差异幅度, 则认为是中心像素的相似点。
步骤 360中, 分别对相似点间的高能透明度和低能透明度按照步骤 340确定的加 权系数进行加权平均, 得到高能透明度和低能透明度的加权平均值。
步骤 370中, 利用步骤 360输出的降噪结果, 即高能透明度和低能透明度的加权 平均值, 计算 alphax和 alphay, 计算公式参见图 5。
最后进行材料分辨, 得到材料识别的初步结果。
采用 alpha曲线法来设计材料识别算法, 特征轴为 alphax和 alphay (alpha曲线坐 标定义见图 5 示意)。 利用高能透明度和低能透明度的加权平均值, 计算 alphax和 alphay, 然后进行材料分辨。 材料分辨的依据为自动标定模块得到的类别分界线。 从 图 6A可以看出 alpha曲线在原子序数方向是具备单调性的,这也正是双能材料分辨算 法的依据。对于高能双能,对应于相同的 alphax点,材料的原子序数增加的同时, alphay 值单调下降。已知某待查点 alpha数据为 (alphaxR, alphayR),査表可以得到与 alphaxR 对应的三个分界点, 分别为 alphayC-Al、 alphayAl-Fe、 alphayFe-Pb, 通过比较 alphayR 在 alphayC-Al、 alphayAl-Fe, alphayFe-Pb中的大小关系,即可确定待查点的材料信息。
材料识别是以相似点的加权平均值为依据的。 因此, 数据统计涨落对分类准确性 的影响已经降到了最低。 但是由于是逐点进行材料分辨的, 点与点之间的材料分辨结 果是不平滑的。 为了保证显示效果, 对整图进行材料分辨后, 进一步对相似点的材料 分辨结果进行降噪处理。 降噪算法与预处理算法类似, 关键点是选定邻域大小以及判断相似点。 算法步骤 见图 7B流程示意。
步骤 410 440是在算法执行过程中, 根据双能数据的通用特征设定算法参数; 步 骤 450~470是在算法执行过程中, 根据步骤 410~440确定的算法参数, 对实际图像进 行降噪操作。
步骤 410中仍然选定高能透明度和低能透明度为二维特征。 虽然经过了预处理算 法和材料识别算法, 已经增加了材料初步识别结果这一特征。 但增加的这一维特征是 根据前两维特征得到的, 并没有增加信息量。 反而, 由于材料识别初步结果的噪声较 大, 会影响相似点的判断, 故不选用这一维特征。
步骤 420中关键是分析透明度统计涨落幅度。 透明度的统计涨落幅度是由系统整 体的设计水平决定的。 获取统计涨落幅度的一种方式为: 统计一个平整区域内透明度 的相对标准差。 一般认为, 噪声为相对标准差的 2.355倍。 根据噪声水平即可设定相 似点间的透明度差异幅度。
步骤 430中, 邻域大小的设置与步骤 420中数据统计涨落幅度相关。 统计涨落幅 度越大, 则降噪力度应相应地增大, 故邻域面积应设置得大些。 另一方面, 邻域面积 设置得越大, 算法的运行速度越慢。 为了保证彩色的视觉效果, 我们适当地加大了降 噪力度, 选择的邻域面积为 11像素 *11像素。
步骤 440中, 与中心点距离不同的相似点的加权系数的设置不是这个算法的关键 步骤。 本系统采用等权相加的方式进行。
步骤 450中, 根据步骤 430.中确定邻域面积, 为中心像素确定相似点搜索范围。 如果邻域内某像素与中心像素的髙能透明度和低能透明度的差异均小于步骤 420中确 定的相似点间的透明度差异幅度, 则认为是中心像素的相似点。
步骤 460中, 对相似点间的材料识别初步结果按照步骤 440确定的加权系数进行 加权平均, 得到材料识别结果的加权平均值。
步骤 470中, 输出步骤 360中得到的材料识别结果的加权平均值, 作为最终的材 料识别结果。
图 8给出了材料分辨模块中不同预处理参数的效果图对比。 可以看出, 不进行预 处理时, 材料识别结果噪声非常大, 材料分辨准确率受到很大 响, 见图 8的 (A)示 意; 选择较小的邻域 (3*3 ) 进行预处理, 噪声有所下降, 但是降噪程度很有限, 见 图 8的 (B) 示意; 选择较大的邻域 (11*11)进行预处理, 但不做相似点判断, 虽然图像 的噪声大幅度下降, 但是在物体边缘区域, 由于将两种材料的数据混在一起统计, 材 料分辨结果明显是错误的, 见图 8的 (C) 示意; 选择较大的邻域 (11*11)进行预处理, 且做相似点判断,即可以很好地起到降噪作用,又不影响边缘区域,见图 8的 (D)示意。
灰度融合模块的基本思路是对于穿透低质量厚度的区域, 低能档成像图占主要比 例; 而对于穿透厚质量厚度的区域, 高能档成像图占主要比例。 具体的加权系数可以 根据实际系统的特点灵活调整。 基于此思路, 可采用如下公式进行: resultGray = nil * dFactor + nlh * (l - dFactor), 其中, nil, nlh, resultGray依次为低能 档数据, 高能档数据 融合灰度; dFactor 为加权系数, 可取 nll/max(nll)或者
Figure imgf000016_0001
彩色化模块中, 关键的是材料信息与颜色色调对应关系的设计以及颜色表的设 计。
国际上已有低能 X射线双能系统的彩色显示标准, 有机物用橙色表示, 轻金属用 绿色表示, 无机物用蓝色表示。 高能 X射线双能系统的彩色标准还没有形成, 考虑到 用户的视觉习惯, 可以延用低能 X射线双能系统得彩色显示标准, 仍以橙色表示有机 物, 绿色表示轻金属, 蓝色表示无机物。 除了有机物、 轻金属、 无机物三种类别, 高 能 X射线双能系统还可以区分重金属类别, 该类别的显示没有现成的标准, 可以根据 设计者的喜好设定, 考虑到色调使用的连续性, 本系统以紫色调表示重金属。
彩色化算法的关键点是颜色表的设计以及原子序数和融合灰度与颜色信息如何 映射。彩色化流程见图 9示意。步骤 510~550示出了颜色表的设计原则;步骤 610~650 示出了原子序数和融合灰度与颜色信息的映射原则。
步骤 510、 520中, 把原子序数与色调 H (HLS颜色模型中的 H) 对应, 不同 的原子序数用不同的色调来表示, 材料分辨的类别与色调类别相等。 为了彩色图像的 协调性, 在材料分辨过程中, 可以把类别细化, 不仅把物质分成有机物、 轻金属、 无 机物和重金属四种类别, 而且在各类别内部以原子序数为依据进一步细化成若干个子 类别。 色调的设计在遵循彩色显示标准的大原则下, 兼顾色调过渡, 从橙红到橙黄、 从黄绿到蓝绿、 从青蓝到深蓝、 从紫蓝到紫红。
步骤 530中, 根据视觉理论, 对各种色调设定一个合适的饱和度值。 人眼对绿 色调比较敏感, 对橙色调的敏感度一般, 而对蓝色调比较不敏感, 因此, 可给绿色调 设置较低的饱和度, 给橙色调设置中等偏高的饱和度, 给蓝色调设置较高的饱和度。 根据该原则, 设计饱和度和色调的对应规则。 步骤 540中,关键是设计融合灰度与 L值(HLS颜色模型中的 L)的映射关系。 为了保证色调的协调性, 在不同色调的颜色表中, 对于同一亮度等级的颜色, 人眼感 受到的亮度要相当, 并不能简单地把融合灰度与 L值等效。
• 这一步要借助 YUV颜色模型中的 Y值。 YUV颜色模型中的 Y值代表了 人眼感受到的亮度, 而 HLS颜色模型中的 L值并不能代表人眼感受到的 亮度,如果简单的按照融合灰度与 L值对应来设计颜色表的话,不同色调 的颜色是不协调的。 因此, 需要综合 YUV模型、 RGB模型、 HLS模型的 转换关系, 根据融合灰度与 Y值等效的原则, 设计融合灰度与 L值的映 射关系, 从而得到 HLS颜色表。
• 由于人眼感受到的亮度可以由 YUV颜色模型中的 Y值来表示, 因此同一 亮度级别的 Y值应该相等的; 因此问题就转换成为一个很简单的数学问 题, 已知 HLS颜色模型中的色调 H和饱和度 S, 以及 YUV模型中的 Y 值, 要求确定 HLS颜色模型中的亮度 L值; 由于 YUV颜色模型与 HLS 颜色模型没有直接的转换关系, 可以通过 RGB模型作为中介来进行。 步骤 540中, 完成了 HLS颜色表到 RGB颜色表的转换。 由于 HLS颜色模型到 RGB颜色模型的转换比较耗时, 因此, 事先计算得到两者的关系, 存储在表格中, 直 接査表获取 RGB值。 构建 HLS到 RGB的二维映射表格, 第一维是材料信息索引, 第二维是融合灰度索引, 表格中存储的是 RGB值。 这样, 得到了物质识别后的材料 信息, 以及融合后的灰度后, 通过査表就得到各点的 RGB值, 用于彩色图像显示。
完成了颜色表的设计后, 实时彩色化过程就是简单的査表过程, 利用材料分辨 模块输出的材料信息, 以及灰度融合模块输出的融合灰度信息, 查找二维 RGB颜色 表, 即可得到彩色图像。
步骤 610中, 获取了材料分辨模块输出的材料信息, 以及灰度融合模块输出的 融合灰度信息, 作为彩色化算法的输入信息。
步骤 620中, 根据材料信息 (对应于颜色色调), 确定 RGB二维映射表格的第 一维索引值。
步骤 630中, 根据融合灰度信息(颜色亮度等级), 确定 RGB二维映射表格的 第二维索引值。
步骤 640中,根据步骤 620、 630中确定的两维索引值, 査表二维 RGB颜色表, 得到 RGB值。 步骤 650中, 重复步骤 610~640, 逐点确定 RGB值, 输出彩色图像。
灰度显示与材料分辨后的彩色显示效果见图 10示意。
至此已经结合优选实施例对本发明进行了描述。 应该理解, 本领域技术人员在不 脱离本发明的精神和范围的情况下, 可以进行各种其它的改变、 替换和添加。 因此, 本发明的范围不局限于上述特定实施例, 而应由所附权利要求所限定。

Claims

权 利 要 求 一种物质识别方法, 包括步骤. - 用高能射线和低能射线透射被检物体,以获得被检物体的髙能透射图像和低能透 射图像, 所述高能图像中的每个像素的值表示高能射线对被检物体的相应部分的高能 透明度, 而所述低能图像中每个像素的值表示低能射线对被检物体的相应部分的低能 透明度;
针对每个像素,计算所述高能透明度的第一函数的值和所述高能透明度和所述低 能透明度的第二函数的值; 以及
通过利用预先创建的分类曲线对由所述第一函数的值和所述第二函数的值所确 定的位置进行分类来识别被检物体中与每个像素所对应的那部分的物质的类型。
2、 如权利要求 1所述的方法, 还包括步骤- 设置具有预定大小的邻域; 以及
在每个像素的邻域中对所述高能图像和所述低能透射图像进行降噪。
3、 如权利要求 2所述的方法, 其中在每个像素的邻域中对所述高能图像和所述 低能透射图像进行降噪的步骤包括:
在邻域中寻找与中心像素相似的像素, 作为相似像素;
对邻域中的相似像素进行加权平均。
4、 如权利要求 3所述的方法, 其中所述相似像素的高能透明度和低能透明度与 所述中心像素的高能透明度和低能透明度之间的差异小于预定值。
5、 如权利要求 3所述的方法, 其中将所述被检物体识别为有机物、 轻金属、 无 机物或重金属。 '
6、 如权利要求 5所述的方法, 还包括对识别结果进行彩色化显示的步骤。
7、 如权利要求 6所述的方法, 其中所述彩色化显示步骤包括- 对每个像素的髙能透明度和低能透明度进行加权平均, 作为该像素的融合灰度 值;
根据被检物体中与该像素相对应部分的材料类型来确定颜色色调;
根据该像素的融合灰度值来确定该像素的亮度等级;
将所述颜色色调和所述亮度等级作为索引来从预先创建的査找表中获得该像素 的 R值、 G值和 Β值。
8、 如权利要求 7所述的方法, 其中根据被检物体中与该像素相对应部分的材料 类型来确定颜色色调的步骤包括: 将橙色赋予有机物、 将绿色赋予轻金属、.将蓝色赋 予无机物、 将紫色赋予重金属。
9、 如权利要求 1所述的方法, 还包括步骤: 对从射线源发出的射线进行能谱整 形来扩大高能射线和低能射线之间的能谱差异。
10、如权利要求 1所述的方法,其中所述分类曲线是通过下面的步骤针对每种标 定材料而创建的:
用高能射线和低能射线照射各种厚度下标定材料来获得相应的高能透明度和低 能透明度;
将高能透明度的第一函数作为横坐标,将低能透明度和高能透明度的第二函数作 为纵坐标来形成标定材料在不同厚度下的点;
基于所述的点形成所述分类曲线。
11、如权利要求 10所述的方法,其中基于所述点形成所述分类曲线的步骤包括: 采用最小二乘曲线拟合法对所述点进行曲线拟合。
12、 如权利要求 10所述的方法, 其中基于所述点形成所述分类曲线的步骤包括: 采用切比雪夫意义下的最佳拟合多项式对所述点进行曲线拟合。
13、 如权利要求 10所述的方法, 还包括对所述分类曲线进行离散化的步骤。
14、 一种物质识别设备, 包括- 图像形成装置,用高能射线和低能射线透射被检物体, 以获得被检物体的高能透 射图像和低能透射图像, 所述高能图像中的每个像素的值表示高能射线对被检物体的 相应部分的高能透明度, 而所述低能图像中每个像素的值表示低能射线对被检物体的 相应部分的低能透明度;
计算装置,针对每个像素,计算所述高能透明度的第一函数的值和所述高能透明 度和所述低能透明度的第二函数的值; 以及
分类装置, 通过利用预先创建的分类曲线对由所述第一函数的值和所述第二函数 的值所确定的位置进行分类来识别被检物体中与每个像素所对应的那部分的物质的 类型。
15、 如权利要求 14所述的设备, 还包括:
设置具有预定大小的邻域的装置; 以及
在每个像素的邻域中对所述高能图像和所述低能透射图像进行降噪的装置。
16、 如权利要求 15所述的设备, 其中在每个像素的邻域中对所述高能图像和所 述低能透射图像进行降噪的装置包括:
在邻域中寻找与中心像素相似的像素作为相似像素的装置;
对邻域中的相似像素进行加权平均的装置。
17、 如权利要求 16所述的设备, 其中所述相似像素的高能透明度和低能透明度 与所述中心像素的高能透明度和低能透明度之间的差异小于预定值。
18、 如权利要求 16所述的设备, 其中将所述被检物体识别为有机物、 轻金属、 无机物或重金属。
19、 如权利要求 18所述的设备, 还包括对识别结果进行彩色化显示的装置。
20、 如权利要求 19所述的设备, 其中所述对识别结果进行彩色化显示的装置包 括:
对每个像素的高能透明度和低能透明度进行加权平均作为该像素的融合灰度值 的装置;
根据被检物体中与该像素相对应部分的材料类型来确定颜色色调的装置; 根据该像素的融合灰度值来确定该像素的亮度等级的装置;
将所述颜色色调和所述亮度等级作为索引来从预先创建的查找表中获得该像素 的 R值、 G值和 B值的装置。
21、 如权利要求 20所述的设备, 其中将橙色赋予有机物、 将绿色赋予轻金属、 将蓝色赋予无机物、 将紫色赋予重金属。
22、 如权利要求 14所述的设备, 还包括对从射线源发出的射线进行能谱整形来 扩大高能射线和低能射线之间的能谱差异的装置。
23、 如权利要求 14所述的设备, 其中所述分类曲线是通过下面的步骤针对每种 标定材料而创建的:
用高能射线和低能射线照射各种厚度下标定材料来获得相应的高能透明度和低 能透明度;
将高能透明度的第一函数作为横坐标,将低能透明度和高能透明度的第二函数作 为纵坐标来 成标定材料在不同厚度下的点;
基于所述的点形成所述分类曲线。
24、 如权利要求 23所述的设备, 其中采用最小二乘曲线拟合法对所述点进行曲 线拟合。
25、 如权利要求 23所述的设备, 其中采用切比雪夫意义下的最佳拟合多项式对 所述点进行曲线拟合。
26、 如权利要求 23所述的设备, 其中对所述分类曲线进行离散化。
PCT/CN2008/001880 2007-11-15 2008-11-14 Procédé et dispositif d'identification de matériau WO2009070977A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2007101774052A CN101435783B (zh) 2007-11-15 2007-11-15 物质识别方法和设备
CN200710177405.2 2007-11-15

Publications (1)

Publication Number Publication Date
WO2009070977A1 true WO2009070977A1 (fr) 2009-06-11

Family

ID=40139670

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2008/001880 WO2009070977A1 (fr) 2007-11-15 2008-11-14 Procédé et dispositif d'identification de matériau

Country Status (8)

Country Link
US (1) US8290230B2 (zh)
JP (1) JP4806441B2 (zh)
CN (1) CN101435783B (zh)
AU (1) AU2008243199B2 (zh)
DE (1) DE102008043526B4 (zh)
GB (1) GB2454782B (zh)
RU (1) RU2396550C1 (zh)
WO (1) WO2009070977A1 (zh)

Families Citing this family (55)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101614683B (zh) * 2008-06-27 2011-10-05 清华大学 物质识别系统中的实时标定设备和方法
CN101647706B (zh) * 2008-08-13 2012-05-30 清华大学 高能双能ct系统的图象重建方法
US8422826B2 (en) * 2009-06-05 2013-04-16 Varian Medical Systems, Inc. Method and apparatus to facilitate using fused images to identify materials
US8184769B2 (en) * 2009-06-05 2012-05-22 Varian Medical Systems, Inc. Method and apparatus to facilitate using multiple radiation-detection views to differentiate one material from another
EP2539697A4 (en) * 2010-02-25 2017-05-10 Rapiscan Systems, Inc. A high-energy x-ray spectroscopy-based inspection system and methods to determine the atomic number of materials
AU2011203239B2 (en) * 2010-06-21 2013-09-19 Commonwealth Scientific And Industrial Research Organisation Mineral particle material resolving X-ray imaging
FR2961904B1 (fr) 2010-06-29 2012-08-17 Commissariat Energie Atomique Procede d'identification de materiaux a partir de radiographies x multi energies
JP5562193B2 (ja) * 2010-09-28 2014-07-30 一般社団法人日本アルミニウム協会 アルミ合金判別方法および選別設備
US9207294B1 (en) * 2011-01-30 2015-12-08 Sven Simon Method and apparatus for the contactless determination of electrical quantities
GB201113224D0 (en) 2011-08-01 2011-09-14 Kromek Ltd Method for the radiological investigation of an object
EP2739959B1 (en) 2011-08-01 2017-05-10 Kromek Limited Object monitoring using multispectral radiation
EP2739995B1 (en) * 2011-08-01 2017-05-24 Kromek Limited Detection and/or classification of materials
JP5832889B2 (ja) * 2011-12-28 2015-12-16 株式会社アーステクニカ アルミ合金判別方法と判別装置および選別設備
CN103630947B (zh) * 2012-08-21 2016-09-28 同方威视技术股份有限公司 可监测放射性物质的背散射人体安检系统及其扫描方法
DE102012017872A1 (de) * 2012-09-06 2014-05-15 Technische Universität Dresden Verfahren und Vorrichtung zur bildgebenden Prüfung von Objekten mit Röntgenstrahlung
US8983234B2 (en) * 2012-09-28 2015-03-17 Varian Medical Systems, Inc. Method and apparatus pertaining to using imaging information to identify a spectrum
GB201220418D0 (en) 2012-11-13 2012-12-26 Kromek Ltd Identification of materials
GB201220419D0 (en) 2012-11-13 2012-12-26 Kromek Ltd Identification of materials
US10359375B2 (en) * 2013-10-23 2019-07-23 Nanovision Technology (Beijing) Co., Ltd. Photon count-based radiation imaging system, method and device thereof
CN104749199B (zh) 2013-12-30 2019-02-19 同方威视技术股份有限公司 双能/双视角的高能x射线透视成像系统
CN104749198B (zh) 2013-12-30 2019-08-06 同方威视技术股份有限公司 双通道高能x射线透视成像系统
CN105203569B (zh) * 2014-06-09 2018-06-12 北京君和信达科技有限公司 双能辐射系统和提高双能辐射系统材料识别能力的方法
CN105242322A (zh) * 2014-06-25 2016-01-13 清华大学 探测器装置、双能ct系统和使用该系统的检测方法
CN105808555B (zh) * 2014-12-30 2019-07-26 清华大学 检查货物的方法和系统
CN105806856B (zh) * 2014-12-30 2019-02-19 清华大学 双能射线成像方法和系统
CN104464871B (zh) 2014-12-30 2017-11-21 同方威视技术股份有限公司 射线过滤装置和双能x射线检查系统
CN105806857B (zh) * 2014-12-31 2019-02-19 同方威视技术股份有限公司 双能射线检查系统物质识别及其分类参数处理方法与装置
CN106353828B (zh) * 2015-07-22 2018-09-21 清华大学 在安检系统中估算被检查物体重量的方法和装置
CN105223214B (zh) * 2015-10-22 2019-01-11 同方威视技术股份有限公司 用于物质分辨的标定装置、标定方法和标定系统
CN106706677B (zh) * 2015-11-18 2019-09-03 同方威视技术股份有限公司 检查货物的方法和系统
CN106932414A (zh) 2015-12-29 2017-07-07 同方威视技术股份有限公司 检验检疫用检查系统及其方法
CN105527654B (zh) * 2015-12-29 2019-05-03 中检科威(北京)科技有限公司 一种检验检疫用检查装置
GB2565026B (en) * 2016-05-03 2021-08-18 Rapiscan Systems Inc Radiation signal processing system
WO2017205914A1 (en) * 2016-05-30 2017-12-07 Southern Innovation International Pty Ltd Material characterisation system and method
CN106153091A (zh) * 2016-08-30 2016-11-23 北京华力兴科技发展有限责任公司 多维度车辆信息显示系统
CN108169254A (zh) * 2016-12-07 2018-06-15 清华大学 检查设备和检查方法
CN106841248B (zh) * 2017-04-07 2023-10-31 北京华力兴科技发展有限责任公司 车辆或集装箱的安全检查系统
AU2018272836A1 (en) * 2017-05-22 2019-12-19 Leidos Security Detection & Automation, Inc. Systems and methods for image processing
CN107595311A (zh) * 2017-08-30 2018-01-19 沈阳东软医疗系统有限公司 双能量ct图像处理方法、装置以及设备
CN107478664B (zh) * 2017-09-06 2020-06-26 奕瑞影像科技(太仓)有限公司 线型双能x射线传感器及线型双能x射线检测系统
CN107957428A (zh) * 2017-11-29 2018-04-24 合肥赑歌数据科技有限公司 一种基于dsp自适应图像能量衰退处理方法
CN108254394B (zh) * 2017-12-28 2020-09-01 合肥美亚光电技术股份有限公司 X射线双能检测方法及系统
KR101893557B1 (ko) * 2017-12-29 2018-08-30 (주)제이엘케이인스펙션 영상 처리 장치 및 방법
FR3082945B1 (fr) * 2018-06-22 2020-06-05 Commissariat A L'energie Atomique Et Aux Energies Alternatives Procede de caracterisation d'un objet par imagerie spectrale
JP7476194B2 (ja) 2018-08-10 2024-04-30 レイドス・セキュリティ・ディテクション・アンド・オートメーション・インコーポレイテッド 画像処理のためのシステムおよび方法
US11116462B2 (en) * 2018-08-28 2021-09-14 King Fahd University Of Petroleum And Minerals X-ray system and method for generating x-ray image in color
CN110231005B (zh) * 2019-06-27 2021-04-13 江苏同威信达技术有限公司 一种物品质量厚度检测方法及物品质量厚度检测装置
US10908098B1 (en) * 2019-06-28 2021-02-02 National Technology & Engineering Solutions Of Sandia, Llc Automatic method of material identification for computed tomography
CN111274871B (zh) * 2020-01-07 2020-09-08 西南林业大学 基于轻小型无人机的森林火灾林木受损程度提取方法
KR102230737B1 (ko) * 2020-06-25 2021-03-22 대한민국 4색 X-ray 장비를 이용한 원자의 판별 방법
CN112858167B (zh) * 2021-01-07 2024-01-02 上海奕瑞光电子科技股份有限公司 多排双能线阵探测器扫描方法、系统、介质及装置
JP7480724B2 (ja) 2021-02-22 2024-05-10 株式会社島津製作所 X線撮影装置およびx線撮影方法
KR102293548B1 (ko) * 2021-03-11 2021-08-25 대한민국 인공지능을 이용한 위험물 검출 시스템 및 방법
CN115078414B (zh) * 2022-08-18 2022-11-04 湖南苏科智能科技有限公司 基于多能量x射线的液体成分抗干扰智能检测方法
CN117347396B (zh) * 2023-08-18 2024-05-03 北京声迅电子股份有限公司 基于XGBoost模型的物质种类识别方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5524133A (en) * 1992-01-15 1996-06-04 Cambridge Imaging Limited Material identification using x-rays
JP2004347328A (ja) * 2003-05-20 2004-12-09 Hitachi Ltd X線撮影装置
CN1995993A (zh) * 2005-12-31 2007-07-11 清华大学 一种利用多种能量辐射扫描物质的方法及其装置
CN101019042A (zh) * 2004-03-01 2007-08-15 创新医疗系统技术公司 通过双能量辐射扫描和缓发中子探测来检查物体
CN101074937A (zh) * 2006-05-19 2007-11-21 清华大学 能谱调制装置、识别材料的方法和设备及图像处理方法

Family Cites Families (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4789930A (en) 1985-11-15 1988-12-06 Picker International, Inc. Energy dependent gain correction for radiation detection
DE58903297D1 (de) 1989-04-06 1993-02-25 Heimann Systems Gmbh & Co Materialpruefanlage.
US6160866A (en) * 1991-02-13 2000-12-12 Lunar Corporation Apparatus for bilateral femur measurement
JPH0868768A (ja) 1994-08-30 1996-03-12 Hitachi Medical Corp X線荷物検査装置
FR2740561B1 (fr) * 1995-10-27 1997-12-19 Inst Francais Du Petrole Methode pour evaluer la variation d'intensite d'un rayonnement polychromatique ayant un spectre de frequence connu, apres traversee d'un corps absorbant
US6018562A (en) * 1995-11-13 2000-01-25 The United States Of America As Represented By The Secretary Of The Army Apparatus and method for automatic recognition of concealed objects using multiple energy computed tomography
US6088423A (en) * 1998-06-05 2000-07-11 Vivid Technologies, Inc. Multiview x-ray based system for detecting contraband such as in baggage
DE60109166T2 (de) 2001-09-28 2005-07-28 Motorola, Inc., Schaumburg Kommunikationssystem zur Detektion von ausserhalb des Systems erzeugter Interferenz
WO2004095060A2 (en) * 2003-04-23 2004-11-04 L-3 Communications Security and Detection Systems Corporation X-ray imaging technique
US7092485B2 (en) * 2003-05-27 2006-08-15 Control Screening, Llc X-ray inspection system for detecting explosives and other contraband
US7186023B2 (en) * 2003-06-10 2007-03-06 Shimadzu Corporation Slice image and/or dimensional image creating method
US6950492B2 (en) * 2003-06-25 2005-09-27 Besson Guy M Dynamic multi-spectral X-ray projection imaging
US6987833B2 (en) * 2003-10-16 2006-01-17 General Electric Company Methods and apparatus for identification and imaging of specific materials
US7099433B2 (en) * 2004-03-01 2006-08-29 Spectramet, Llc Method and apparatus for sorting materials according to relative composition
US7257188B2 (en) * 2004-03-01 2007-08-14 Varian Medical Systems Technologies, Inc. Dual energy radiation scanning of contents of an object
WO2006076038A2 (en) * 2004-05-27 2006-07-20 L-3 Communications Security And Detection Systems, Inc. Method and apparatus for detecting contraband using radiated compound signatures
CN100472206C (zh) * 2004-09-15 2009-03-25 中国农业大学 食品中异物的检测方法
CN100582758C (zh) 2005-11-03 2010-01-20 清华大学 用快中子和连续能谱x射线进行材料识别的方法及其装置
DE102007019304A1 (de) 2007-04-24 2008-11-20 Airbus Deutschland Gmbh Überwachungseinrichtung und Überwachungsverfahren für eine Luftfahrzeugeinrichtung

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5524133A (en) * 1992-01-15 1996-06-04 Cambridge Imaging Limited Material identification using x-rays
JP2004347328A (ja) * 2003-05-20 2004-12-09 Hitachi Ltd X線撮影装置
CN101019042A (zh) * 2004-03-01 2007-08-15 创新医疗系统技术公司 通过双能量辐射扫描和缓发中子探测来检查物体
CN1995993A (zh) * 2005-12-31 2007-07-11 清华大学 一种利用多种能量辐射扫描物质的方法及其装置
CN101074937A (zh) * 2006-05-19 2007-11-21 清华大学 能谱调制装置、识别材料的方法和设备及图像处理方法

Also Published As

Publication number Publication date
GB0820542D0 (en) 2008-12-17
JP4806441B2 (ja) 2011-11-02
CN101435783B (zh) 2011-01-26
JP2009122108A (ja) 2009-06-04
GB2454782A (en) 2009-05-20
RU2396550C1 (ru) 2010-08-10
US20090129544A1 (en) 2009-05-21
RU2008144918A (ru) 2010-05-20
US8290230B2 (en) 2012-10-16
DE102008043526A1 (de) 2009-06-10
DE102008043526B4 (de) 2014-07-03
GB2454782B (en) 2009-11-04
AU2008243199B2 (en) 2011-12-01
AU2008243199A1 (en) 2009-06-04
CN101435783A (zh) 2009-05-20

Similar Documents

Publication Publication Date Title
WO2009070977A1 (fr) Procédé et dispositif d'identification de matériau
EP3242126B1 (en) Dual-energy ray imaging method and system
CN102175698B (zh) 物质识别系统中创建分类曲线的方法和设备
Rogers et al. Automated x-ray image analysis for cargo security: Critical review and future promise
US20090052622A1 (en) Nuclear material detection system
US20060098866A1 (en) User interface for inspection system with isoluminant regions
CN109948565B (zh) 一种用于邮政业的违禁品不开箱检测方法
Rogers et al. A deep learning framework for the automated inspection of complex dual-energy x-ray cargo imagery
CN101614683B (zh) 物质识别系统中的实时标定设备和方法
US10386532B2 (en) Radiation signal processing system
US8180139B2 (en) Method and system for inspection of containers
US8290120B2 (en) Dual energy radiation scanning of contents of an object based on contents type
CN110427981A (zh) 基于深度神经网络的sar船舶检测系统及方法
Yang et al. An RGB channel operation for removal of the difference of atmospheric scattering and its application on total sky cloud detection
Wang Enhanced colour encoding of materials discrimination information for multiple view dual-energy x-ray imaging
Wang et al. Precision circular target location in vision coordinate measurement system
Feng et al. Automatic identification of cracks from borehole image under complicated geological conditions
CN109602430A (zh) 骨科射线成像机
US12019203B2 (en) Device and method for scanning items
Worrall et al. Verification Data Pattern Recognition and Change Detection at the Neutron Instrument Level
Çığla et al. Robust material classification on dual-energy x-ray imaging devices
US20090087012A1 (en) Systems and methods for identifying similarities among alarms
Dang et al. A Deep Learning-Based Framework for Estimating Tree Defect Parameters via a Stand-Off Radar
Lalor et al. Direct atomic number reconstruction of dual energy cargo radiographs using a semiempirical transparency model
CN110197145A (zh) 一种联合空间稀疏与相关性的高光谱目标检测算法

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: 08857105

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: 08857105

Country of ref document: EP

Kind code of ref document: A1