X-ray image processing system, method and recording medium

Through multi-sensors, X-ray transmission data is synthesized and learning data is generated. Deep learning technology is used to solve the problem of automatic item recognition in X-ray carry-on luggage inspection, achieving high-precision item recognition and reducing the load of inspectors.

CN112346138BActive Publication Date: 2024-11-12HITACHI SOFTWARE ENG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010126421.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2019-08-09
Filing Date
2020-02-28
Publication Date
2024-11-12
Estimated Expiration
2040-02-28

AI Technical Summary

Technical Problem

In X-ray carry-on luggage inspection, the prior art is difficult to automatically identify overlapping items, and requires a large number of trained inspectors to make image judgments, making it difficult to ensure efficient and accurate in large-scale activities.

Method used

By using multiple sensors to obtain X-ray transmission data, synthesize material information and generate color images, and use deep learning technology to construct learning data to achieve high-precision recognition of overlapping items.

Benefits of technology

It improves the accuracy of X-ray image recognition, can identify items with high accuracy when items overlap, reduces the load of inspectors, and is suitable for dynamically changing item combinations and inspection activities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112346138B_ABST
    Figure CN112346138B_ABST
Patent Text Reader

Abstract

The object of the present invention is to provide an X-ray image processing system and method and a recording medium for improving the accuracy of image recognition using X-ray images obtained in X-ray inspection services. As a solution, information of an area determined to contain an object is obtained, and the X-ray transmission amounts obtained by multiple sensors in the same area (for example, the same background, the same luggage) are synthesized, and material information is estimated based on the re-synthesized X-ray transmission amounts to generate a color image. As the object of synthesis, for example, the combination information of the objects stored in operation is used, and the combination of objects contained in the same luggage is set to a combination of more objects. The color image generated by the synthesis is used as learning data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to an X-ray image processing system and method thereof, and a recording medium, and in particular to a technique for determining an object contained in an X-ray image acquired by an X-ray inspection device used for carrying baggage inspection. Background Art

[0002] X-ray carry-on baggage inspection devices are used in carry-on baggage inspections at airports or large-scale event venues. X-ray carry-on baggage inspection devices usually generate grayscale images that indicate the amount of X-ray transmission, and color images that are colored for each material after determining the material. Inspectors visually inspect the images to confirm the presence of dangerous items. In general, if inspectors find dangerous items, they will open the baggage for inspection.

[0003] In order to confirm whether or not dangerous goods are included in X-ray images, highly trained inspectors are required. Therefore, it is difficult to temporarily secure a large number of inspectors, for example, in the case of a large-scale event, from the perspective of prior training and cost. Therefore, in order to reduce the burden on inspectors even a little bit, attempts have been made to automate the detection of dangerous goods.

[0004] As a solution to automate image recognition, image recognition technology using deep learning based on AI (artificial intelligence) can be cited. Deep learning is widely used in image analysis and other applications, and can achieve higher recognition accuracy, so its popularity is progressing. However, in order to learn, a large amount of learning data is required, and in order to obtain higher recognition accuracy, it is necessary to prepare learning data containing multiple styles. As a technology using machine learning for image recognition, for example, Patent Document 1 discloses a method of synthesizing images of areas obtained from two images after correcting color information.

[0005] Patent Document 1: Japanese Patent Application Publication No. 2017-45441 Summary of the invention

[0006] The technology described in Patent Document 1 is generally based on images taken by crime prevention cameras or vehicle-mounted cameras. Therefore, when the synthesized objects overlap, it is possible to consider which image is in the front. In addition, since the type and position of the image of the synthesized object are limited based on the information of the installation position of the camera, various conditions for image recognition can be set in advance.

[0007] On the other hand, in image synthesis with X-ray images as the object, since X-rays pass through objects, the image in the area where the objects overlap becomes an image containing the features of multiple objects. Therefore, the conventional image synthesis technology cannot be used as it is. In addition, when applied to X-ray carry-on baggage inspection, the combination of objects that is preferably used varies depending on the type and period of the activity to be inspected, so it is difficult to set these in advance.

[0008] An object of the present invention is to improve the accuracy of image recognition using X-ray images acquired in X-ray inspection work.

[0009] According to a preferred example of the X-ray image processing system of the present invention, information of an area determined to contain an article is obtained, and the X-ray transmission amounts obtained by multiple sensors in the same area (for example, the same background, the same luggage) are synthesized, and material information is estimated again based on the synthesized X-ray transmission amounts to generate a color image. As the object of synthesis, in one example, the combination information of the articles stored in operation is used, and as the combination of articles contained in the same luggage, for example, the combination of more articles is preferred. The color image generated by the synthesis is used as learning data.

[0010] Effects of the Invention

[0011] According to the present invention, the accuracy of image recognition using X-ray images obtained in X-ray inspection work can be improved. In addition, by constructing learning data using images synthesized into X-ray images, objects can be recognized with high accuracy even when objects overlap. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a structural diagram of an X-ray inspection device.

[0013] Figure 2 It is a functional structure diagram of the processing device 102.

[0014] Figure 3 It is a diagram showing the structure and data examples of the co-occurrence data table 220.

[0015] Figure 4 It is a diagram showing the structure and data examples of the article data table 221.

[0016] Figure 5 This is a flowchart showing the processing operation of object recognition.

[0017] Figure 6 This is a flowchart showing the generation of the learning data 218 and the learning operation.

[0018] Figure 7 This is a diagram showing the relationship between the color image and the transmission amount in Example 2.

[0019] Figure 8 This is a diagram showing an example of transmission amount data of two types of articles in Example 2.

[0020] Fig. 9 This is a flowchart showing the processing actions for synthesizing items in Example 2.

[0021] Fig.10 This is a diagram showing an example of transmission amount data of two types of articles in Example 2.

[0022] Fig.11 This is a diagram showing an example of transmission amount data of two types of articles in Example 2.

[0023] Fig.12 It is a diagram showing the structure and data example of the energy conversion table.

[0024] Description of symbols

[0025] 100: X-ray inspection device

[0026] 101: Device body

[0027] 102: Processing device

[0028] 103: Display unit

[0029] 104: Input section

[0030] 201: CPU

[0031] 202: Main memory

[0032] 205: Ministry of Communications

[0033] 210: Storage

[0034] 211: OS

[0035] 212: Item Identification Procedure

[0036] 213: Image synthesis program

[0037] 214: Screen display program

[0038] 215: Co-occurrence Data Management Program

[0039] 216: X-ray Data Management Program

[0040] 217: Item Learning Program

[0041] 218: Learning Data

[0042] 219: Baggage photo data

[0043] 220: Co-occurrence data table

[0044] 221: Item Data Table

[0045] 222: Learning Model DETAILED DESCRIPTION

[0046] Hereinafter, preferred embodiments of the X-ray image processing system of the present invention will be described.

[0047] [Example 1]

[0048] Figure 1 This is a structural diagram of an X-ray inspection apparatus according to an embodiment of an X-ray image processing system.

[0049] X-ray inspection apparatus 100 is widely used as a carry-on baggage inspection apparatus in airport security inspection, for example, and includes an X-ray apparatus body (hereinafter referred to as an apparatus body) 101, a processing apparatus 102, a display unit 103, and an input unit 104. Processing apparatus 102 is, for example, a personal computer (PC).

[0050] The device body 101 has an irradiation mechanism for irradiating X-rays and an X-ray photographing mechanism for photographing an object such as carry-on baggage to measure the amount of X-ray transmission, and outputs data on the amount of X-ray transmission (hereinafter, sometimes simply referred to as transmission data or transmission). The processing device 102 determines whether the carry-on baggage is safe based on the object recognition result of the X-ray image photographed by the device body 101, and uses the stored object information to learn a model for image recognition. The display unit 103 is a display terminal that displays the X-ray image, and the inspector can visually confirm the X-ray image displayed on the screen.

[0051] The device body 101 has a conveying mechanism with a belt conveyor for conveying luggage, and the conveying mechanism is controlled by the control unit to drive and stop the belt conveyor. The device body 101 has a display light, and when the control unit determines that the object is dangerous carry-on luggage (alarm object) as a result of object identification, the display light is turned on to notify the inspector that it is an alarm object.

[0052] Two types of X-ray sensors (hereinafter simply referred to as sensors) for measuring the amount of X-ray transmission are arranged in the conveying mechanism to obtain two types of data. That is, one sensor obtains data of lower energy, and the other sensor obtains data of higher energy. The processing device 102 determines the material of the object based on the difference between the high-energy data and the low-energy data obtained by the two sensors. As an X-ray sensor, any X-ray sensor can be used as long as it can obtain X-ray data that can determine the material, and the detection method of the X-ray sensor is not limited. For example, a backscatter material determination sensor can be used. In addition, the acquisition of X-ray data by the X-ray sensor can be called an X-ray data acquisition unit.

[0053] Figure 2 It is a functional structure diagram of the processing device 102.

[0054] The processing device 102 is an information processing device including a processing unit (CPU: Central Processing Unit) 201 , a main memory 202 , a display interface (I / F) unit 203 connected to the display unit 103 , an input interface (I / F) unit 204 connected to the input unit 104 , a communication unit 205 , and a storage unit 210 .

[0055] The processing unit 201 executes the program to realize a predetermined function or action. The storage unit 210 stores an item recognition program 212, an image synthesis program 213, a screen display program 214, a co-occurrence data management program 215, an X-ray data management program 216, an item learning program 217, learning data 218, baggage photographing data 219, a co-occurrence data table 220, an item data table 221, and a learning model 222.

[0056] Here, the object recognition program 212 uses the deep learning segmentation technology to recognize the objects contained in the image in pixel units, taking as the object a color image obtained by visualizing the material information of the baggage and the density information of the objects generated based on the X-ray transmission data captured by the device main body 101. The details of the operation of the object recognition program 212 will be described later.

[0057] The image synthesis program 213 generates a combination of objects using the co-occurrence data table 220, and synthesizes the image at a designated or randomly designated position using the high and low energies of the object regions. The operation of the image synthesis program 214 will be described in detail later.

[0058] The screen display program 214 displays a color image or grayscale image generated from the X-ray transmission data captured by the apparatus main body 101 and the object information of the inspection object on the display unit 103. In addition, if there is an object of the inspection object that cannot be identified, the area and type of the object are designated from the input unit 203 such as a mouse, and additional object information is obtained. The operation of the screen display program 214 will be described in detail later.

[0059] The co-occurrence data management program 215 counts the combination of articles brought together based on the information of articles recognized in the carry-on baggage inspection, and generates the combination information of the articles used for the synthesis of the image. The operation of the co-occurrence data management program 215 will be described in detail later.

[0060] The X-ray data management program 216 generates a color image representing material and density information based on the X-ray transmission data measured by the two sensors in the device main body 101, and registers the transmission data and the color image in the baggage imaging data 219. That is, the material of the article is determined based on the difference in the X-ray transmission data of the two sensors, and a color image is generated in which the material information is the type of color and the transmission is the color density (the higher density part is darker, and the lower density part is lighter). In addition, when there is a request to obtain the baggage imaging data 219, the data of the specified conditions are provided to the requesting party.

[0061] The object learning program 217 uses the input color image and label data, performs learning using segmentation processing of deep learning, and generates a learning model 222 .

[0062] In addition, if the above-mentioned object recognition program 212, image synthesis program 213, screen display program 214, co-occurrence data management program 215, X-ray data management program 216, and object learning program 217 stored in the storage unit 210 are executed by the processing unit 201, they will respectively function as the object recognition unit 212′, image synthesis unit 213′, screen display unit 214′, co-occurrence data management unit 215′, X-ray data management unit 216′, and object learning unit 217′.

[0063] The learning data 218 registers the color image and label data used in the learning. The learning data 218 refers to a data area that stores a collection of data or learning data. The color image is image data generated by the X-ray data management program 216. The label data is data representing the outline of the object, that is, data representing the type of the object by the outline, and is associated with the object ID of the object data table 221 (described later).

[0064] The baggage imaging data 219 stores X-ray baggage images, namely, high and low transmission data, color images, and grayscale images, imaged by the device main body 101. The baggage imaging data 291 is assigned a baggage ID of the co-occurrence data table 220 (described later).

[0065] The co-occurrence data table 220 is as follows Figure 3 As shown, information of the article recognition result of each baggage obtained by the co-occurrence data management program 215 is stored. That is, the shooting date and time, the presence or absence of detected articles such as smartphones, wallets, plastic bottles, cameras, watches, etc. are stored in correspondence with the baggage ID assigned to each baggage.

[0066] Item data table 221 Figure 4As shown, information such as the type, overlap, frequency of use, and method of adding of the items identified by the X-ray baggage image (baggage imaging data 219) is registered. That is, in the item data table 221, for each item ID that identifies the item, a type ID indicating the type of the item, an overlap flag indicating whether the item overlaps with other items, a frequency of use indicating the number of times the item is used in learning, and an addition flag indicating whether the addition of item information is a recognition result or an additional registration are registered. Here, any method can be used to determine whether there is a possibility of overlap when the pixels of two items are in contact. Furthermore, for each item ID, a color image, a grayscale image, high and low transmission data of the item area, and mask data of pixels indicating the existence of the item are separately registered in the baggage imaging data 219 of the storage unit 210, and are associated with the item ID of the item data table 221. The mask data stores a value corresponding to the number of pixels of the rectangle surrounding the item area, and 1 is registered in the pixels of the item and 0 is registered in the other pixels. In addition, as long as the position of the object can be determined in pixel units, it can be saved in other formats.

[0067] In the learning model 222 , parameter information of the model learned by the item learning program 217 is registered.

[0068] Next, refer to Figure 5 Next, the processing operation for identifying and displaying an object will be described. This processing operation is performed in daily inspection work using the X-ray inspection apparatus 100.

[0069] First, the X-ray data management program 216 obtains the X-ray transmission data captured by multiple sensors in order to determine the material, and determines the material of the pixel unit of the image based on the differential information of the transmission data of the multiple sensors. The material can be determined by, for example, a method known in the art of classifying the material into four types: metal, inorganic, organic, and others based on the differential information of two types of transmission, high and low energy. Next, using the material information and the X-ray transmission of the specified sensor, a color image is generated in which the material information is the type of color and the transmission is the color density (the denser part is darker, and the less dense part is lighter), and the X-ray transmission data of the two sensors and the color image are registered in the baggage imaging data 219 (S501).

[0070] Next, the object recognition program 212 obtains the color image obtained in step S501, and uses deep learning segmentation processing to recognize the object in pixel units (S502). Here, for the recognition of the object, a library such as "Fully Convolutional Instance-aware Semantic Segmentation" which is widely known as an OSS (open source software) library can be used. In addition, as long as the object can be determined in pixel units, there is no limitation on the recognition method.

[0071] Next, when the inspection object includes the article set as the alarm object in the recognition result of step S502, the screen display program 214 displays an image with the corresponding part highlighted on the display unit 103 (S503). Here, the highlighting is realized by, for example, surrounding the alarm object with a red rectangle. In addition, there is no limitation on the method of highlighting as long as the area of ​​the alarm object can be known. In addition, when the inspector confirms that there is no prohibited object highlighted, the input of the outline and type of the article can be received by the input unit 104 such as a keyboard or a mouse, and registered in the article data table 221 as additional article information. Here, the designation of the outline of the article can be input by surrounding the outline with a plurality of points or by manually drawing the outline using the display unit 103 such as a touch panel. In addition, for the alarm object and other articles, it is possible to define in advance which of the grayscale image and the color image is easier to observe in the unit of the article, and change the color and grayscale coloring in the unit of the article based on the definition to display. In the usual inspection business, sometimes the inspection is performed while switching the display between color and grayscale as a whole image. In this case, by coloring each item in accordance with the ease of observation, it is expected that the inspection will be more efficient. The coloring method can be achieved by switching the color of the pixel determined to have an item between color and grayscale, so the detailed description is omitted.

[0072] Next, the X-ray data management program 216 registers the number of articles in the co-occurrence data table 220 for each baggage identified in step 502 (S504). In addition, the number of articles additionally designated in step S503 is also registered in the co-occurrence data table 220. As a result, the type and number of articles (the number of articles is sometimes referred to as frequency information) are registered in the co-occurrence data table 220 for each baggage ID.

[0073] Next, the X-ray data management program 216 registers the image information and transmission amount data of the object recognized in step S502 and the image information and transmission amount data of the object additionally designated in step S503 in the object data table 221 (S505). As a result, in the object data table 221, the object type ID, the object overlap flag, the use frequency indicating the number of times the object has been used in learning, the object information addition method flag, etc. are registered in correspondence with the object ID.

[0074] Furthermore, the screen display program 214 terminates the processing when receiving an instruction to terminate the shooting, and executes the processing of steps S501 to S505 until receiving an end instruction (S506) when no instruction is given. In addition, the contents of the co-occurrence data table 220 and the article data table 221 are rewritten each time the data of luggage or articles is registered, so it can also be called an update of the table contents.

[0075] Next, refer to Figure 6 The generation of the learning data 218 and the learning operation are described.

[0076] First, the co-occurrence data management program 215 obtains co-occurrence data from the co-occurrence data table 220, collects the combination information of items that frequently appear within a specified period (e.g., one week), and selects the combination of items with higher frequency as the item list (S601). For example, data with the same number of items can be arranged in order of frequency, and the order of frequency from high to low can be set as candidates, and the first item combination is selected. Figure 3 In the example of the co-occurrence data table 220, the combination of a smartphone and a wallet is high frequency. In addition, the first list may be used and then the second list may be used, and the priority of the list used in the past may be lowered, so as not to be limited to a combination of specific items.

[0077] Next, the image synthesis program 213 selects the article data of the article selected in step S601 from the article data table 221 (S602). Here, the article data is obtained by giving priority to images of articles that are less frequently used in learning and manually added (unrecognized) articles, and randomly selecting from candidate articles that meet the conditions determined in advance. In addition, the article data obtained from the image where the article overlaps with other articles may be partially missing or part of other articles may be included in the article data, so it is excluded from the selection object.

[0078] Next, the image synthesis program 213 selects the position in the image where the item obtained in step S602 is to be placed (S603). Regarding the image position, a random number is used to obtain the rotation angle of the image (0 to 360 degrees), and the starting coordinates X and Y (0 < X < the width of the luggage image, 0 < Y < the height of the luggage image). For X and Y, selection is made in such a way that the image of the item does not protrude from the area of the luggage image. In addition, the size of the luggage image is defined in advance, and the item position relative to the X-ray luggage image of a bag or tray etc. without items determined in advance is selected.

[0079] Next, the image synthesis program 213 synthesizes the item data selected in step S602 at the position and rotation angle selected in step S603 (S604). Since a normal luggage image has items added to a bag or tray etc., a large amount of shooting data of only the tray or bag is prepared in advance here, and synthesis is implemented by adding to the image selected from among them. In the case of multiple items, for example, the first item is the synthesis of the image of only the tray or bag and the item image, and the second item is the synthesis for the image after adding the first item. Synthesis is implemented by adding the items one by one. In addition, regarding items that are the objects of alarms such as dangerous items, even if the frequency is low, they need to be actively included in the learning data, so it is also possible to intentionally add an item that is the object of an alarm once in a pre-determined number of times.

[0080] Here, for the synthesis of the image, the position of the item is obtained based on the mask data of the item data, and two types of transmission amount data including the coordinates of the item are obtained. For the transmission amount A of the item and the transmission amount B of the luggage image (including the case where items have been added), the value calculated by the following calculation formula is calculated as the new transmission amount when the transmission amount A and the transmission amount B overlap. In addition, the following calculation formula is an example, and other calculation formulas can also be used as long as the change in the transmission amount when the items overlap can be approximated.

[0081] [Formula] (Transmission amount A / Transmission amount MAX) × (Transmission amount B / Transmission amount MAX) × Transmission amount MAX

[0082] Here, Transmission amount MAX is the maximum value of the sensor of the device obtained when there is no luggage, and is set to a value such as "65535" for example. Regarding the calculation of the transmission amount in the case where two items overlap, by implementing the transmission amount data of the two sensors, the change in the transmission amount when the item and the luggage image overlap can be approximated.

[0083] Next, for the two transmissions after all the objects are synthesized, the X-ray data management program 216 determines the material of each pixel based on the difference information of the two transmissions, generates a color image, and registers it with the pixels of the objects and the label data indicating the type in the learning data 218 (S605). By using the synthesis method of this embodiment, it is possible to perform synthesis taking into account overlap that cannot be synthesized in color images, so that more natural learning data can be generated.

[0084] Next, the processing from step S602 to step S605 is repeated a preset number of times to generate N learning data, which are registered in the learning data 218 (S606). Even for the same item list, since the item data used in the synthesis and the synthesis position are different, a different baggage image is generated each time. Therefore, when N baggage images are generated for M item lists, N can be set so that M×N is included in the time available for learning data generation.

[0085] Next, the processing from step S601 to step S606 is executed until there is an instruction to end the item list selection, and the processing is terminated when there is an end instruction (S607).

[0086] Next, the object learning program 217 reads the learning data 218, and performs model learning (S608) by a method that can identify objects in pixel units, such as a segmentation method of deep learning. In learning, image data and label data indicating the coordinates at which objects are located are required, but since the position of the object can be identified during synthesis, learning can be performed by using this information as label data.

[0087] In addition, the Figure 5 and Figure 6 The processing action is Figure 2 Although the several programs shown are performed in cooperation, it is also possible to configure such that these operations are implemented as one program.

[0088] [Example 2]

[0089] Reference Figures 7 to 12 , the synthesis of X-ray images related to Example 2 is explained.

[0090] Example 2 is based on the relevant example 1 Figure 2 The processing device 102 is a prerequisite, but the image synthesis method is different from that of embodiment 1. Figure 7 As shown in the figure, the positions of multiple objects are determined in the color image ( Figure 7 (1)) to synthesize objects using X-ray transmission ( Figure 7 (2)), and then generate a color image based on the X-ray transmission ( Figure 7(3)), thereby generating the color image required for deep learning. In addition, since the transformation from X-ray transmission to color image is an irreversible transformation, the synthesis of Example 2 is achieved by maintaining both the X-ray transmission and the color image. Figure 8 As shown, using (a) high-energy transmission data and (b) low-energy transmission data, the following formulas 1 and 2 are used to synthesize the transmission data (2 types) of the image (for example, with a resolution of 640×480) serving as the background of the synthetic image to be synthesized and the transmission data (2 types) of the object in pixel units.

[0091] [Formula 1]

[0092]

[0093] [Formula 2]

[0094]

[0095] Among them, P1 and P2 are the transmission amounts of the pixels of interest of each image. When two or more images are synthesized simultaneously, P3 is accumulated in the same manner. Here, x and y represent the position of the pixel, with x=0 and y=0 at the top left of the image, and x increases toward the right and y increases toward the bottom.

[0096] Next, refer to Fig. 9 The flowchart of Example 2 is used to illustrate the image synthesis processing action.

[0097] The image synthesis program 213 obtains the following from the baggage photographic data 219: Fig.10 As in (a) of FIG. 1 , the transmission amount data of the background such as a tray or a bag is captured (S901). Alternatively, the luggage image data of an object other than a tray or a bag may be acquired.

[0098] Next, the image synthesis program 213 refers to the object data table 221 to obtain the transmission amount data of the synthesis target object (S902). In addition, since the shape of the object is not limited to a rectangle, the transmission amount data of the object includes the following: Fig.11 The information that identifies pixels with items, such as the mask data (detection result of color image) shown in (b) of FIG. 1 , is used to identify pixels with items. For example, pixels with items are registered as 1, and pixels without items are registered as 0. When multiple items are represented by one mask data, pixels without items are registered as 0, and other pixels are registered with IDs (>=1) that identify the items.

[0099] Next, the image synthesis program 213 synthesizes the transmission amount by a synthesis method based on the equations shown in [Equation 1] and [Equation 2] (S903). In addition, when the transmission amount is not 65536 levels, the 65535 part is adjusted to the level number.

[0100] Next, when there is another object to be synthesized, the processing from step S902 to step S903 is repeated (S904).

[0101] Next, the X-ray data management program 216 generates a color image based on the synthesized transmission amount (two types of high and low energy) data, and outputs the color image as learning data together with information indicating the type of article in each pixel (S905). Fig.10 The learning data is stored in the learning data 218 together with the type information of the article of each pixel.

[0102] In addition, when the input data from the device main body 101 is not two kinds of high and low energy data but only a color image, the color image can be grayed out, and the image can be converted such that the portion with higher density is a darker image and the portion with lower density is a lighter image, and the value of each pixel of the gray image is set to high energy or low energy. However, since two kinds of transmission amount data are required for the synthesis of the object, for example, using Fig.12 The energy conversion table for converting the transmission amount (assuming high energy) to low energy is used to estimate the transmission amount. In addition, when there is no corresponding value in the energy conversion table, the conversion value of the transmission amount value of the closest value can be used. In addition, the energy conversion table is prepared in advance in the storage unit 210, and the value of the table can be generated by obtaining statistics of the combination of high and low energy in advance.

[0103] According to this embodiment, since learning data can be automatically generated using the object data and additional data stored while performing the X-ray carry-on baggage inspection, there is an effect of improving the accuracy while performing the X-ray carry-on baggage inspection. In addition, the combination of objects can be made close to the combination of objects in operation, so that generation of learning data that matches the operation can be expected. Furthermore, by synthesizing the image using the transmission amount of the area of ​​the object determined by the color image and returning it to the color image, there is an effect of being able to construct learning data that can reproduce the overlap of the objects.

Claims

1. An X-ray image processing system, using a processing device having a processing unit for processing data and a storage unit to process X-ray images, characterized in that: The processing unit has: An acquisition unit that acquires a plurality of X-ray transmission amount data from a plurality of sensors; An image generating unit determines the material of the image in units of pixels based on the difference between the plurality of X-ray transmission data obtained, and generates a color image according to the material of the object; A recognition unit that recognizes an object in pixel units based on the generated color image; as well as A synthesis unit synthesizes the background image and the image of the object using the X-ray transmission data for each identical background; storing the generated color image and the X-ray transmission data in the storage unit; storing information of the above-mentioned articles contained in the same luggage as the background unit in the above-mentioned storage unit; The synthesis unit generates a color image by synthesizing the image of the luggage and the image of the article using the X-ray transmission data based on the information of the article stored in the storage unit.

2. The X-ray image processing system according to claim 1, characterized in that: The storage unit includes: a first table for registering the number of articles contained in each piece of luggage acquired by the acquisition unit; and A second table registers image information of the object recognized by the recognition unit and transmission amount data in association with each other; The synthesizing unit refers to the first table, selects image information of the articles from the second table for a combination of articles having a high frequency of combination of articles, and synthesizes the images.

3. The X-ray image processing system according to claim 1, characterized in that: The color image generated by the above-mentioned synthesis unit is used as learning data.

4. An X-ray image processing method, using a processing device having a processing unit for processing data and a storage unit to process an X-ray image, characterized in that: have: The step of obtaining a plurality of X-ray transmission data by a plurality of sensors; A step of determining the material of the image in pixel units based on the difference between the plurality of X-ray transmission amount data obtained; Based on the X-ray transmission data, a color image is generated according to the material of the object; an identification step of identifying an object in pixel units based on the generated color image; as well as A synthesis step of synthesizing a background image and an image of the object using the X-ray transmission data for each identical background; storing the generated color image and the X-ray transmission data in the storage unit; storing information of the above-mentioned articles contained in the same luggage as the background unit in the above-mentioned storage unit; In the synthesis step, the image of the luggage and the image of the article are synthesized using the X-ray transmission data based on the information of the article stored in the storage unit to generate a color image.

5. The X-ray image processing method according to claim 4, characterized in that: storing frequency information of the above-mentioned items contained in the above-mentioned same luggage in a storage unit; In the synthesis step, based on the frequency information of the articles stored in the storage unit, for the articles with a high frequency of the combination of the articles, the image of the luggage and the image of the articles are synthesized using the X-ray transmission data to generate a color image.

6. The X-ray image processing method according to claim 4, characterized in that: The method further includes a step of adding an article that has not been recognized in the recognition step by specifying it through an input unit.

7. The X-ray image processing method according to claim 4, characterized in that: The color image generated by the above-mentioned synthesis step is used as learning data.

8. The X-ray image processing method according to claim 4, characterized in that: Pre-storing a plurality of background images in the storage unit; In the synthesis step, one background image selected from a plurality of background images stored in the storage unit is synthesized with the image of the object.

9. A computer-readable recording medium recording an X-ray image processing program for causing a processing device that processes data to execute the X-ray image processing method according to claim 4.