Object detection method and system

Abstract

In the case of a liquid material contained in a container without having a shape, it is very difficult to accurately predict the type or physical properties of the liquid, so that there is a problem that the accuracy or detection rate is very low.SOLUTION: The present invention relates to a method and system for detecting an object capable of improving a detection rate of dangerous materials such as liquid explosives as well as existing explosives, and may include (a) preparing an X-ray transmission image of an object, (b) calculating an effective atomic number value (Zeff) using the X-ray transmission image, and (c) classifying a target article image from the X-ray transmission image using the effective atomic number value and reconstruction of a multi-energy image for each energy band.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] [National research and development project that supported this invention] This invention relates to Project Specific Number 1615013162, Project Number RS-2021-KA162822, which was conducted with funding from the Ministry of Land, Infrastructure, Transport and Tourism and supported by the Land, Infrastructure, Transport and Tourism Agency.

[0002] The present invention relates to an object detection method and system, and more particularly to an object detection method and system that can improve the detection rate of hazardous materials such as liquid explosives as well as existing explosives. [Background technology]

[0003] An X-ray generator is a device that generates X-rays by colliding an accelerated electron beam with an anode target. Images of various energy bands can be obtained using such an X-ray generator, i.e., an X-ray detector.

[0004] Such X-ray generators can also be used in computed tomography devices and are used in various clinical fields such as diagnosis, real-time imaging during surgery, and post-operative prognosis assessment. They are widely used not only as medical imaging diagnostic equipment, but also for cargo inspection at airports and non-destructive inspection of manufacturing products such as microstructures.

[0005] For example, computed tomography devices and X-ray security inspection devices can obtain a tomographic image of the subject by irradiating X-rays onto the subject, some of which is absorbed by the subject, and detecting the remaining transmitted radiation with multiple detectors arranged in a line or plane, and then converting the output data of each detector into an electrical signal to reconstruct the image.

[0006] Such a computer tomography apparatus can be used in a wide variety of ways, such as when used for cargo inspection at airports, to detect hazardous substances such as various solid explosives and narcotics and output an alarm signal. Summary of the Invention [Problem to be solved by the invention]

[0007] However, conventional computed tomography devices can only identify objects or infer their physical properties based on the shape of X-ray images, making it extremely difficult to accurately predict the physical properties of objects with similar shapes. It is particularly difficult to accurately detect liquid hazardous substances such as nitroglycerin, EDGN, methyl nitrate, hydrogen peroxide, alcohol, benzyl alcohol, methyl alcohol, ethanol, and sulfur, as well as narcotics such as heroin, cocaine, opium, and morphine, based on their external shapes alone. In particular, for shapeless liquid substances contained in containers, it is extremely difficult to accurately predict the type and physical properties of the liquid, resulting in extremely low accuracy and detection rates.

[0008] The idea of ​​the present invention is to solve the above problems and to provide an object detection method and system that can significantly improve the accuracy and detection rate of identifying and predicting the properties of liquid substances such as liquid explosives, narcotics, as well as existing hazardous substances. However, these problems are merely examples and do not limit the scope of the present invention. [Means for solving the problem]

[0009] To solve the above problem, the object detection method according to the concept of the present invention includes the steps of: (a) preparing an X-ray transmission image of an object; (b) calculating an effective atomic number value (Zeff) using the X-ray transmission image; and (c) classifying a target article image from the X-ray transmission image using the effective atomic number value and reconstruction of a multi-energy image by energy band.

[0010] Furthermore, according to the present invention, the effective atomic number value (Zeff, Effective Z number) may be a value proportional to the cumulative sum of atomic nuclei present on a path through which an X-ray passes.

[0011] In addition, according to the present invention, the step (a) may include: (a-1) loading an X-ray transmission image; (a-2) checking whether the X-ray transmission image is normal or not and outputting an error message if the X-ray transmission image is abnormal; and (a-3) if the X-ray transmission image is normal, correcting and standardizing image distortion based on the value of the background region of the X-ray transmission image or performing histogram correction similar to an actual captured image.

[0012] In addition, according to the present invention, the step (b) may include the steps of: (b-1) first calculating an effective atomic number value (Zeff) of the corrected X-ray transmission image; (b-2) removing unnecessary images of the tray supporting the object by background processing and extracting only the image region of interest; and (b-3) classifying the article image using an artificial neural network model in the X-ray image and a sinogram visualizing the same.

[0013] Furthermore, according to the present invention, the artificial neural network model may include a Cycle-Consistent Generative Adversarial Network (Cycle-GAN).

[0014] In addition, according to the present invention, the step (c) may include a step of (c-1) separating the target item image into images for each energy band in consideration of the proximity or overlap of the object when the basic detection mode is selected, and classifying the target item image into individual item images using the images.

[0015] In addition, according to the present invention, the step (c-1) may include the steps of: (c-1-1) setting an energy band using the effective atomic number value; (c-1-2) removing images other than the object of interest; and (c-1-3) processing the target object image into image images for each energy band of N levels (N is a natural number) from low density to high density according to the set energy band, creating an object list of the objects of interest, and adding and merging the effective atomic number value and density value of the object corresponding to the target object in the object list to a table.

[0016] In addition, according to the present invention, the step (c) may further include a step of (c-2) recalculating the effective atomic number value (Zeff) for each region of interest (ROI) and classifying the ROI into a liquid image of interest if the liquid detection mode is selected.

[0017] Furthermore, according to the present invention, the step (c-2) may include the steps of: (c-2-1) setting regions of interest and extracting reference values ​​of effective atomic number (ZI0, Zeff Initial Value(0)) for each region of interest; (c-2-2) secondarily recalculating the effective atomic number value (Zeff) for each region of interest; (c-2-3) extracting an outline of a container of an item of interest for each region of interest using the effective atomic number value, calculating the effective atomic number value and density value of the container based on the outline, and thereby identifying the shape and type of the container; and (c-2-4) removing noise, extracting a liquid region inside the container, calculating the effective atomic number value and density value of the liquid, and predicting the type and volume of the liquid.

[0018] In addition, according to the present invention, in step (c-2-1), for the purpose of accuracy in detecting an item, the area excluding the object in the tray can be treated as an exception from the area of ​​interest, taking into consideration the size, length, and complexity of the item.

[0019] Furthermore, according to the present invention, in the step (c-2-3), a threshold value of image pixels is specified to separate the extracted region from the background region, and the container region can be separated from the background of the image using the brightness or color difference due to the threshold value.

[0020] In addition, according to the present invention, the step (c-2-4) measures the height of the center of the object to remove noise caused by the difference in physical properties between the container and the sealing cap that seals the container, and can detect the actual volume of liquid based on the degree of filling.

[0021] In addition, according to the present invention, the step (c-2-4) can predict the type of liquid based on dual energy using the high energy and low energy histogram of the inner region among the entire region and the inner region, taking into account materials with similar distribution of dual energy-based attenuation rate (R) in a known material table.

[0022] Furthermore, according to the present invention, in step (c-2-4), the dual-energy-based attenuation ratio (R) has a value similar to the effective atomic number value (Zeff), and may be a ratio of the background PV (Pixel Value) of the high-energy image to the background PV (Pixel Value) of the low-energy image so as to reduce the influence of the penetration depth and density of the object.

[0023] Furthermore, according to the present invention, in the step (c-2-4), in a density / effective atomic number graph in which density (g / cm3) is the first axis (X-axis) and the effective atomic number value (Zeff) is the second axis (Y-axis), physical properties can be predicted based on a physical property table in which substances with similar physical properties are grouped together.

[0024] In addition, according to the present invention, the method may further include (d) comprehensively detecting dangerous items by considering weighting using multi-view images taken at multiple shooting angles.

[0025] Furthermore, according to the present invention, in step (d), the weighting is the number of target items read from the images captured in each view, and in the case of the same object, the number with the higher weighting can be determined as the number of items.

[0026] Meanwhile, an object detection system according to the idea of ​​the present invention for solving the above problem includes an image input unit for inputting an X-ray transmission image of an object, an effective atomic number value calculation unit for calculating an effective atomic number value (Zeff) using the X-ray transmission image, a target classification unit for classifying target object images from the X-ray transmission image using the effective atomic number value and reconstruction of a multi-energy image for each energy band, and a multi-view detection unit for comprehensively detecting dangerous objects by considering weighting using multi-view images taken at multiple shooting angles. The image input unit includes an image loading unit for loading the X-ray transmission image, an error output unit for checking whether the X-ray transmission image is normal and outputting an error message if it is abnormal, and an error output unit for outputting an error message if the X-ray transmission image is normal. The image processing system includes a standardization unit that corrects and standardizes image distortion based on the value of a background region, or performs histogram correction similar to that of an actual captured image. The effective atomic number value calculation unit includes an effective atomic number value primary calculation unit that primarily calculates the effective atomic number value (Zeff) of the corrected X-ray transmission image, a region of interest extraction unit that removes unnecessary images of tray portions supporting the object through background processing and extracts only the image region of interest, and an item image classification unit that classifies item images using an artificial neural network model in the X-ray image and a visualized sinogram thereof. The target classification unit may include, in the basic detection mode, a basic detection unit that separates the target item image into images by energy band taking into account adjacent or overlapping objects and classifies them into individual item images using these, and in the liquid detection mode, a liquid detection unit that recalculates the effective atomic number value (Zeff) for each region of interest (ROI) and classifies them into liquid images of interest.

[0027] In addition, according to the present invention, the basic detection unit may include an energy band setting unit that sets an energy band using the effective atomic number value, an unnecessary image removal unit that removes images other than images of the object of interest, and an energy band-specific image processing unit that processes the target object image into image images for each energy band in N levels (N is a natural number) from low density to high density according to the set energy band, creates an object list of the objects of interest, and adds the effective atomic number value and density value of the object corresponding to the target object in the object list to a table and merges them.

[0028] Furthermore, according to the present invention, the liquid detection unit may include a region of interest setting unit that sets a region of interest and extracts a reference value of the effective atomic number (ZI0, Zeff Initial Value(0)) for each region of interest; an effective atomic number value secondary recalculation unit that secondarily recalculates the effective atomic number value (Zeff) for each region of interest; a container determination unit that extracts an outline of a container of an item of interest for each region of interest using the effective atomic number value, calculates the effective atomic number value and density value of the container based on the outline, and thereby identifies the shape and type of the container; and a liquid determination unit that removes noise, extracts a liquid region inside the container, calculates the effective atomic number value and density value of the liquid, and predicts the type and volume of the liquid. [Effects of the Invention]

[0029] According to some embodiments of the present invention as described above, it is possible to significantly improve the accuracy and detection rate of identifying and predicting the properties of liquid substances such as liquid explosives and narcotics as well as existing hazardous substances by using effective atomic number values ​​obtained from X-ray transmission images and images for each energy band. Of course, the scope of the present invention is not limited to these effects. [Brief explanation of the drawings]

[0030] [Figure 1] 1 is a block diagram illustrating an object detection system according to some embodiments of the present invention. [Figure 2] FIG. 2 is a block diagram illustrating an image input section of the object detection system of FIG. 1. [Figure 3] FIG. 2 is a block diagram illustrating an effective atomic number value calculation unit of the object detection system of FIG. 1. [Figure 4] FIG. 2 is a block diagram illustrating a target classification portion of the object detection system of FIG. 1. [Figure 5] 1 is a flowchart illustrating a method for detecting objects according to some embodiments of the present invention. [Figure 6] 6 is a flowchart showing step (a) of the object detection method of FIG. 5. [Figure 7] 6 is a flowchart showing step (b) of the object detection method of FIG. 5. [Figure 8] 6 is a flowchart showing step (c) of the object detection method of FIG. 5. [Figure 9] 9 is a flowchart showing step (c-1) of the object detection method of FIG. 8. [Figure 10] 9 is a flowchart showing step (c-2) of the object detection method of FIG. 8. [Figure 11] FIG. 1 illustrates an example of an application of the technology for sorting items by energy band in accordance with some embodiments of the present invention. [Figure 12] FIG. 1 illustrates an example of an energy band image processing technique in accordance with some embodiments of the present invention. [Figure 13] FIG. 1 is a diagram illustrating a multi-view weighted object detection technique according to some embodiments of the present invention. [Figure 14] 1 illustrates an example of explosive detection results according to some embodiments of the present invention. [Figure 15] 1 is a chart illustrating improved relative detection rates versus conventional relative detection rates of object detection methods and systems according to some embodiments of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0031] The examples of the present invention are provided to more completely explain the present invention to those skilled in the art, and the following examples can be modified into various other forms, and the scope of the present invention is not limited to the following examples. Rather, these examples are provided to make the present disclosure more complete and complete, and to fully convey the concept of the present invention to those skilled in the art. In addition, the thickness and size of each layer in the drawings are exaggerated for convenience and clarity of explanation.

[0032] Various preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0033] FIG. 1 is a block diagram illustrating an object detection system according to some embodiments of the present invention.

[0034] First, as shown in FIG. 1, an object detection system according to some embodiments of the present invention may include an image input unit 10 that inputs an X-ray transmission image of various objects such as baggage at an airport or cargo on a ship, an effective atomic number value calculation unit 20 that calculates an effective atomic number value (Zeff) using the X-ray transmission image, a target classification unit 30 that classifies target object images from the X-ray transmission image using the effective atomic number value and reconstruction of multi-energy images for each energy band, and a multi-view detection unit 40 that comprehensively detects dangerous objects by considering weighting using multi-view images taken at multiple shooting angles.

[0035] Here, the image input unit 10, effective atomic number value calculation unit 20, target classification unit 30, and multi-view detection unit 40 can be configured in a wide variety of forms, such as a microprocessor, microchip, control circuit, printed circuit board, substrate, central processing unit, arithmetic unit, storage device storing various programs such as a configuration program, program processing device, computer, server computer, cloud, network, smartphone, smart pad, and various smart devices.

[0036] Therefore, according to the object detection system of the present invention, a series of processes can be performed, in which an X-ray transmission image of an object is input, an effective atomic number value (Zeff) is calculated using the X-ray transmission image, an image of a target object is classified from the X-ray transmission image using the effective atomic number value and reconstruction of a multi-energy image for each energy band, and a hazardous object is detected comprehensively by considering weighting using multi-view images taken at multiple shooting angles.

[0037] FIG. 2 is a block diagram illustrating the image input section 10 of the object detection system of FIG.

[0038] More specifically, as shown in FIG. 2, the image input unit 10 may include an image loading unit 11 for loading an X-ray transmission image, an error output unit 12 for checking whether the X-ray transmission image is normal and outputting an error message if the X-ray transmission image is abnormal, and a standardization unit 13 for correcting and standardizing image distortion based on the value of the background region of the X-ray transmission image if the X-ray transmission image is normal, or for performing histogram correction similar to that of an actual captured image.

[0039] Therefore, the image input unit 10 loads an X-ray transmission image, checks whether the X-ray transmission image is normal, and outputs an error message if it is abnormal. If the X-ray transmission image is normal, it can perform a series of processes of correcting and standardizing image distortion based on the value of the background region of the X-ray transmission image, or performing histogram correction similar to that of an actual captured image.

[0040] FIG. 3 is a block diagram showing the effective atomic number value calculation unit 20 of the object detection system of FIG.

[0041] More specifically, as shown in FIG. 3, the effective atomic number value calculation unit 20 may include an effective atomic number value primary calculation unit 21 that primarily calculates the effective atomic number value (Zeff) of the corrected X-ray transmission image, an area of ​​interest extraction unit 22 that removes unnecessary images of the tray supporting the object through background processing and extracts only the image area of ​​interest, and an item image classification unit 23 that classifies item images using an artificial neural network model in the X-ray image and a sinogram that visualizes the X-ray image.

[0042] Therefore, the effective atomic number value calculation unit 20 can perform a series of processes: first, calculate the effective atomic number value (Zeff) of the corrected X-ray transmission image, remove the unnecessary image of the tray portion supporting the object by background processing, extract only the image area of ​​interest, and classify the object image using an artificial neural network model in the X-ray image and its visualized sinogram.

[0043] Here, according to the present invention, a multi-scale structured neural network is used to remove global linear noise that appears due to sparse views, and to overcome the performance limitations of restored images, an artificial neural network model can be used to classify object images in X-ray images and their visualized sinograms.

[0044] FIG. 4 is a block diagram illustrating the target classifier 30 of the object detection system of FIG.

[0045] More specifically, as shown in FIG. 4, the target classification unit 30 may include a basic detection unit 31 that, in the basic detection mode, separates the target object image into images for each energy band, taking into account the proximity or overlap of objects, and classifies them into individual object images using these, or a liquid detection unit 32 that, in the liquid detection mode, recalculates the effective atomic number value (Zeff) for each region of interest (ROI) and classifies them into liquid images of interest.

[0046] Therefore, the target classification unit 30 can selectively perform the basic detection mode and the liquid detection mode, or can perform both of them. In the basic detection mode, the target object image is separated into images by energy band taking into account the proximity or overlap of objects, and these are used to classify into individual object images. In the liquid detection mode, a series of processes can be performed, in which the effective atomic number value (Zeff) is recalculated for each region of interest (ROI) and then classified into an interest liquid image.

[0047] The basic detection unit 31 may include, for example, an energy band setting unit 311 that sets an energy band using an effective atomic number value, an unnecessary image removal unit 312 that removes images other than those of the object of interest, and an energy band-specific image processing unit 313 that processes the target object image into image images for each energy band in N levels (N is a natural number) from low density to high density according to the set energy band, creates an object list of the objects of interest, and adds and merges the effective atomic number value and density value of the object corresponding to the target object in the object list to a table.

[0048] For example, when an X-ray generator, i.e., an X-ray detector with a capacity of 160 kV is used, X-rays in a band of approximately 40 keV (electron volts) to 160 keV can be emitted. For example, by processing a dual-image image using an image in a high-energy band of 140 kV and an image in a low-energy band of 80 kV, the items of interest can be made into an object list, and the effective atomic number values ​​and density values ​​of the objects in the object list corresponding to the target items can be added to a table and merged.

[0049] Therefore, the basic detection unit 31 can perform a series of processes, including setting an energy band using the effective atomic number value, removing images other than those of the object of interest, processing the target object image into image images for each energy band in N levels (N is a natural number) from low density to high density according to the set energy band, creating an object list of the objects of interest, and adding and merging the effective atomic number value and density value of the object corresponding to the target object in the object list to a table.

[0050] The liquid detection unit 32 may include, for example, a region of interest setting unit 321 that sets a region of interest (ROI) and extracts a reference value of the effective atomic number (ZI0, Zeff Initial Value(0)) for each region of interest; an effective atomic number value secondary recalculation unit 322 that secondarily recalculates the effective atomic number value (Zeff) for each region of interest; a container discrimination unit 323 that extracts the outline of the container of the item of interest for each region of interest using the effective atomic number value and calculates the effective atomic number value and density value of the container based on the outline, thereby identifying the shape and type of the container; and a liquid discrimination unit 324 that removes noise, extracts the liquid area inside the container, calculates the effective atomic number value and density value of the liquid, and predicts the type and volume of the liquid.

[0051] In addition, when a general Zeff formula is used, distortions may occur due to the container and volume of the object, but according to the present invention, the penetration depth (corresponding to the volume) can be calculated and corrected using a side view, and the actual volume of a substance inside the container can be detected. In order to remove noise due to differences in the physical properties between the container and the container cap, a valid image is selected from multi-view images, and the center of the object is measured as H to detect the actual volume of the substance. That is, taking into consideration the phenomenon of box processing based on prominent parts due to noise due to differences in the physical properties between the container and the container cap, a range smaller than the actual volume of the substance is detected, but by selecting and enlarging the deepest part, the actual volume size of the substance can be detected with minimal practical error.

[0052] In addition, the range of detectable objects (Liquid explosive) can be limited to within the tray, and background areas other than the tray can be removed to set the object extraction range. For example, to remove the conveyor belt part displayed at the edge of the image, 3% of the edge can be removed.

[0053] Therefore, for training the artificial neural network, an analytically restored image of a sparse view with linear noise and an analytically restored image of a global view without linear noise are used to train an artificial neural network with a multi-scale structure using an artificial neural network training platform, and a dual-energy image can be reconstructed using this.

[0054] Therefore, the liquid detection unit 32 can perform a series of processes: setting regions of interest, extracting reference values ​​of effective atomic number for each region of interest, secondarily recalculating the effective atomic number values ​​for each region of interest, extracting the outline of the container of the item of interest for each region of interest using the effective atomic number values, calculating the effective atomic number value and density value of the container based on the outline, identifying the shape and type of the container through this, removing noise, extracting the liquid region inside the container, calculating the effective atomic number value and density value of the liquid, and predicting the type and volume of the liquid.

[0055] Therefore, according to the present invention, by using the effective atomic number value obtained from the X-ray transmission image and the image for each energy band, it is possible to significantly improve the accuracy and detection rate of identifying and predicting the physical properties of liquid substances such as liquid explosives and narcotics as well as existing hazardous substances.

[0056] FIG. 5 is a flowchart illustrating a method for detecting objects according to some embodiments of the present invention.

[0057] As shown in FIG. 5, an object detection method using an object detection system according to some embodiments of the present invention can be described as follows. The object detection method according to some embodiments of the present invention can include the steps of: (a) preparing an X-ray transmission image of an object; (b) calculating an effective atomic number value (Zeff) using the X-ray transmission image; (c) classifying a target item image from the X-ray transmission image using the effective atomic number value and reconstructing a multi-energy image for each energy band; and (d) comprehensively detecting dangerous items by considering weighting using multi-view images taken at multiple shooting angles.

[0058] Here, the effective atomic number value (Zeff, Effective Z number) can be a value proportional to the cumulative sum of atomic nuclei present on the path through which the X-ray passes.

[0059] FIG. 6 is a flowchart showing step (a) of the object detection method of FIG.

[0060] As shown in FIG. 6, step (a) may include: (a-1) loading an X-ray transmission image; (a-2) checking whether the X-ray transmission image is normal and outputting an error message if it is abnormal; and (a-3) correcting and standardizing image distortion based on the value of the background region of the X-ray transmission image if the X-ray transmission image is normal, or performing histogram correction similar to that of an actual captured image.

[0061] FIG. 7 is a flowchart showing step (b) of the object detection method of FIG.

[0062] Next, as shown in FIG. 7, step (b) can include the steps of (b-1) first calculating the effective atomic number value (Zeff) of the corrected X-ray transmission image, (b-2) removing unnecessary images of the tray portion supporting the object through background processing and extracting only the image area of ​​interest, and (b-3) classifying the article image using an artificial neural network model in the X-ray image and a sinogram visualizing the same.

[0063] Here, the artificial neural network model may include a Cycle-Consistent Generative Adversarial Network (Cycle-GAN).

[0064] FIG. 8 is a flowchart showing step (c) of the object detection method of FIG.

[0065] Next, as shown in FIG. 8, step (c) may include (c-1) in the basic detection mode, separating the target article image into images for each energy band taking into account the proximity or overlap of objects and classifying them into individual article images using these, and / or (c-2) in the liquid detection mode, recalculating the effective atomic number value (Zeff) for each region of interest (ROI) and classifying them into liquid images of interest.

[0066] Therefore, it is possible to set an object extraction range (ROI) and calculate ZIO based on the ROI, and then extract the outline of an object for each ROI using this. It is an area that specifies a specific part within an image, and image processing and analysis can be performed on that part. By focusing only on specific areas suspected to be explosives, it is possible to shorten the explosives analysis processing time and improve the accuracy and efficiency of detection.

[0067] FIG. 9 is a flowchart showing step (c-1) of the object detection method of FIG.

[0068] More specifically, as shown in FIG. 9, step (c-1) may include the steps of: (c-1-1) setting an energy band using an effective atomic number value; (c-1-2) removing images other than the object of interest; and (c-1-3) processing the target object image into image images for N levels of energy bands from low density to high density (N is a natural number) according to the set energy band, creating an object list of the objects of interest, and adding and merging the effective atomic number value and density value of the object corresponding to the target object in the object list to a table.

[0069] FIG. 10 is a flowchart showing step (c-2) of the object detection method of FIG.

[0070] More specifically, as shown in FIG. 10, step (c-2) may include the following steps: (c-2-1) setting a region of interest (ROI) and extracting a reference value of the effective atomic number (ZI0, Zeff Initial Value(0)) for each region of interest; (c-2-2) secondarily recalculating the effective atomic number value (Zeff) for each region of interest; (c-2-3) extracting the outline of the container of the item of interest for each region of interest using the effective atomic number value, calculating the effective atomic number value and density value of the container based on the outline, and thereby identifying the shape and type of the container; and (c-2-4) removing noise, extracting the liquid region inside the container, and calculating the effective atomic number value and density value of the liquid to predict the type and volume of the liquid.

[0071] Here, for example, in step (c-2-1), for the accuracy of item detection, the area excluding the object in the tray can be treated as an exception from the area of ​​interest, taking into account the size, length, and complexity of the item.

[0072] Also, for example, in step (c-2-3), a threshold value of image pixels can be specified to separate the extracted region from the background region, and the container region can be separated from the background of the image using the brightness or color difference due to the threshold value.

[0073] Also, for example, in step (c-2-4), the height of the center of the object can be measured to remove noise caused by differences in physical properties between the container and the sealing cap that seals the container, and the actual liquid volume can be detected based on the degree of filling.

[0074] Also, for example, in step (c-2-4), the type of liquid can be predicted based on dual energy using the high-energy and low-energy histograms of the inner region among the entire region and the inner region, taking into account materials with similar distributions of dual-energy-based attenuation rates (R) in a known material table.

[0075] Also, for example, in step (c-2-4), the dual-energy-based attenuation ratio (R) has a value similar to the effective atomic number value (Zeff), and may be the ratio of the background PV (Pixel Value) of the high-energy image to the background PV (Pixel Value) of the low-energy image so as to reduce the influence of the penetration depth and density of the object.

[0076] Furthermore, for example, in step (c-2-4), in a density / effective atomic number graph in which density (g / cm3) is the first axis (X-axis) and effective atomic number (Zeff) is the second axis (Y-axis), physical properties can be predicted based on a physical property table that groups together substances with similar physical properties.

[0077] On the other hand, in step (d) of Figure 5, the weighting is the number of target items read from the images captured in each view, and if the object is the same, the number with the higher weighting can be determined to be the number of items.

[0078] FIG. 11 illustrates an example application of the technique for sorting items by energy band in accordance with some embodiments of the present invention.

[0079] As shown in FIG. 11, in conventional object segmentation, when objects are adjacent or overlapping as shown in the example diagram, the algorithm recognizes the objects as a single object in the segmentation step, which frequently results in false detection. However, in the case of the present invention, the object segmentation technology by energy band performs segmentation on an image in which objects are etched by energy band, so that only objects in similar energy range bands can be recognized as a single object as shown in the example diagram.

[0080] FIG. 12 illustrates an example of an energy band image processing technique according to some embodiments of the present invention.

[0081] As shown in Figure 12, according to the present invention, in order to detect overlapping and hidden target objects, a target image can be processed into images of seven energy bands ranging from low density to high density. Based on the images of the energy bands, all objects present in the image can be listed, and the Zeff and density values ​​of objects corresponding to the target object in the object list can be added to a table and merged.

[0082] FIG. 13 is a diagram illustrating a multi-view weighted object detection technique according to some embodiments of the present invention.

[0083] As shown in FIG. 13, according to the present invention, in order to improve the detection rate and false detection rate, when performing object detection for one bag set, the detection results of multi-views can be calculated by weighting and displayed as a result.

[0084] For example, weighting can be calculated based on the number of views in which the algorithm determines the same object as a target object. For example, if one object is determined as a target object in three views and another object is determined as a target object in one view, weighting can be applied to the former.

[0085] Therefore, for example, after applying this multi-view object detection technique, it was confirmed that the detection rate of solid hazardous materials improved by more than 80% and the detection rate of liquid hazardous materials improved by more than 50% compared to conventional detection performance.

[0086] FIG. 14 is a diagram illustrating explosive detection results according to some embodiments of the present invention.

[0087] As shown in Figure 14, according to the present invention, it is possible to significantly improve the accuracy and detection rate of various overlapping hazardous materials that could not be detected or had poor accuracy and detection rate in the past, such as Sample A and Sample B, which are solid materials, as well as Sample C and Sample D, which are liquid materials.

[0088] FIG. 15 is a chart illustrating the improved relative detection rate versus the conventional relative detection rate of object detection methods and systems according to some embodiments of the present invention.

[0089] As shown in FIG. 15, assuming that the conventional relative detection rate of the object detection method and system according to some embodiments of the present invention is 100%, the improved relative detection rate of the present invention, expressed in percent (%), shows that the detection rate has increased by approximately 200% or 300% or more for both solid and liquid samples, and in the case of a specific sample (solid sample 2), the detection rate has increased by up to 500% or more.

[0090] The present invention has been described with reference to the embodiments shown in the drawings, but these are merely illustrative, and those skilled in the art will recognize that various modifications and equivalent embodiments are possible. Therefore, the true technical scope of protection of the present invention should be determined by the technical spirit of the appended claims. [Explanation of symbols]

[0091] 10 Image input section 11 Image Loading Section 12 Error output section 13 Standardization Department 20 Effective atomic number calculation section 21 Effective atomic number value primary calculation part 22 Region of interest extraction unit 23. Item Image Classification Unit 30 Target Classification Unit 31 Basic detection section 311 Energy band setting unit 312 Unnecessary image removal unit 313 Energy Band Image Processing Unit 32 Liquid detection section 321 Region of interest setting unit 322 Secondary recalculation of effective atomic number value 323 Container identification section 324 Liquid discrimination section 40 Multi-view detection unit

Claims

1. (a) preparing an X-ray transmission image of an object; (b) calculating an effective atomic number value (Zeff) using the X-ray transmission image; (c) classifying a target article image from the X-ray transmission image using the effective atomic number value and the reconstruction of a multi-energy image for each energy band.

2. The effective atomic number value (Zeff, Effective Z number) is 2. The object detection method according to claim 1, wherein the value is proportional to the cumulative sum of atomic nuclei present on a path through which the X-ray passes.

3. The step (a) includes: (a-1) loading the X-ray transmission image; (a-2) checking whether the X-ray transmission image is normal or not, and outputting an error message if the X-ray transmission image is abnormal; (a-3) if the X-ray transmission image is normal, correcting distortion of the image and standardizing it based on values ​​of a background region of the X-ray transmission image, or performing histogram correction similar to that of an actual captured image.

4. The step (b) includes: (b-1) calculating the effective atomic number value (Zeff) of the corrected X-ray transmission image; (b-2) removing an unnecessary image of the tray portion supporting the object by background processing and extracting only the image region of interest; The object detection method of claim 1, further comprising: (b-3) classifying the article image using an artificial neural network model in the X-ray image and a sinogram visualizing the X-ray image.

5. The object detection method of claim 4 , wherein the artificial neural network model includes a Cycle-Consistent Generative Adversarial Network (Cycle-GAN).

6. The step (c) (c-1) In the basic detection mode, the object detection method of claim 1 includes a step of separating the target item image into images for each energy band, taking into account the proximity or overlap of the object, and classifying the target item image into individual item images using the images.

7. The step (c-1) (c-1-1) setting an energy band using the effective atomic number value; (c-1-2) removing images other than the item of interest; (c-1-3) The object detection method of claim 6, further comprising the steps of: processing the target object image into video images for N energy bands (N is a natural number) from low density to high density according to the set energy bands; creating an object list of the objects of interest; and adding and merging the effective atomic number values ​​and density values ​​of the objects corresponding to the target object in the object list to a table.

8. The step (c) The object detection method of claim 1, further comprising (c-2) recalculating effective atomic number values ​​(Zeff) for each region of interest (ROI) in a liquid detection mode to classify the ROI into liquid images of interest.

9. The step (c-2) (c-2-1) setting a region of interest and extracting a reference value of the effective atomic number (ZI0, Zeff Initial Value(0)) for each region of interest; (c-2-2) secondarily recalculating the effective atomic number value (Zeff) for each region of interest; (c-2-3) extracting an outline of a container of the object of interest for each region of interest using the effective atomic number value, calculating the effective atomic number value and density value of the container based on the outline, and identifying the shape and type of the container through this; (c-2-4) extracting a liquid region inside the container after removing noise, calculating the effective atomic number value and density value of the liquid, and predicting the type and volume of the liquid.

10. The step (c-2-1) The object detection method according to claim 9, wherein an area excluding the object in the tray is treated as an exception from the area of ​​interest, taking into consideration the size, length, and complexity of the object for accuracy of object detection.

11. The step (c-2-3) 10. The object detection method according to claim 9, further comprising: specifying a threshold value for image pixels to separate the container region from the background region; and using the brightness or color difference resulting from the threshold value to separate the container region from the background of the image.

12. The step (c-2-4) The object detection method of claim 9, wherein the height of the center of the object is measured to remove noise caused by differences in physical properties between the container and a sealing cap that seals the container, and the actual volume of the liquid is detected based on the degree of filling.

13. The step (c-2-4) 10. The object detection method of claim 9, wherein the type of liquid is predicted based on dual energy using a high-energy and low-energy histogram of the inner region among the entire region and the inner region, and taking into account materials with similar distributions of dual-energy-based attenuation rates (R) in a known material table.

14. In the step (c-2-4), 14. The object detection method of claim 13, wherein the dual-energy-based attenuation ratio (R) has a value similar to an effective atomic number value (Zeff) and is a ratio of a background PV (Pixel Value) of a high-energy image to a background PV (Pixel Value) of a low-energy image so as to reduce the influence of penetration depth and density of an object.

15. In the step (c-2-4), 15. The object detection method according to claim 14, wherein in a density / effective atomic number graph in which the density (g / cm3) is the first axis (X-axis) and the effective atomic number value (Zeff) is the second axis (Y-axis), the physical properties are predicted based on a physical property table in which substances having similar physical properties are clustered together.

16. 2. The object detection method of claim 1, further comprising the step of (d) detecting a dangerous object comprehensively by taking into consideration weighting using multi-view images taken at multiple shooting angles.

17. In the step (d), The object detection method of claim 16, wherein the weighting is the number of target items read from the images captured in each view, and if the target object is the same, the number with the higher weighting is determined to be the number of items.

18. an image input unit for inputting an X-ray transmission image of the object; an effective atomic number value calculation unit that calculates an effective atomic number value (Zeff) using the X-ray transmission image; a target classification unit for classifying a target object image from the X-ray transmission image using the effective atomic number value and the reconstruction of a multi-energy image for each energy band; and a multi-view detection unit that comprehensively detects dangerous items by considering weighting using multi-view images captured at multiple imaging angles, The image input unit an image loading unit for loading the X-ray transmission image; an error output unit that checks whether the X-ray transmission image is normal or not and outputs an error message if the X-ray transmission image is abnormal; a standardization unit that corrects and standardizes distortion of the image based on a value of a background region of the X-ray image when the X-ray image is normal, or performs histogram correction similar to that of an actual photographed image; The effective atomic number value calculation unit an effective atomic number value primary calculation unit that primarily calculates the effective atomic number value (Zeff) of the corrected X-ray transmission image; an area of ​​interest extraction unit that removes an unnecessary image of a tray portion supporting the object by background processing and extracts only an image area of ​​interest; an article image classifying unit for classifying article images in the X-ray image and a sinogram visualized from the X-ray image using an artificial neural network model; The target classification unit a basic detection unit that separates the target item image into images for each energy band in consideration of the proximity or overlap of the object when in a basic detection mode, and classifies the target item image into individual item images using the images; and a liquid detector that, in a liquid detection mode, recalculates the effective atomic number value (Zeff) for each region of interest (ROI) and classifies the liquid image into an interesting liquid image.

19. The basic detection unit an energy band setting unit that sets an energy band using the effective atomic number value; an unnecessary image removal unit that removes images other than those of the item of interest; 20. The object detection system of claim 18, further comprising an energy band image processing unit that processes the target object image into image images for each of N energy bands (N is a natural number) from low density to high density according to the set energy band, creates an object list of objects of interest, and adds effective atomic number values ​​and density values ​​of objects corresponding to the target object in the object list to a table and merges them.

20. The liquid detection unit is a region of interest setting unit that sets the region of interest and extracts a reference value of the effective atomic number (ZI0, Zeff Initial Value(0)) for each region of interest; an effective atomic number value secondary recalculation unit for secondarily recalculating the effective atomic number value (Zeff) for each region of interest; a container identification unit that extracts an outline of a container of the object of interest for each region of interest using the effective atomic number value, calculates the effective atomic number value and density value of the container based on the outline, and identifies the shape and type of the container through the above calculation; 20. The object detection system of claim 18, further comprising a liquid determination unit that extracts a liquid region inside the container after removing noise, calculates the effective atomic number value and density value of the liquid, and predicts the type and volume of the liquid.

Images