Processing System

Abstract

To provide a processing system that can implement an inspection using a terahertz wave more advantageously.SOLUTION: A processing system includes: a first imaging system that has a frequency range from 30GHz to 30THz, and captures a first image based on a terahertz from an inspection object; a second imaging system that captures a second image of the inspection object based on an electromagnetic wave of a wavelength different from the terahertz wave; a control unit that controls the first imaging system; and a processor that processes the first image and second image. The control unit is configured to control the first imaging system on the basis of a result of the processing performed on the second image by the processor, and the processor is configured to: detect an inspection area based on the second image; compare the first image with the second image to thereby select an area of the first image corresponding to the inspection area; and perform image processing about the area of the first image corresponding to the inspection area detected from the second image.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a processing system including an imaging system. [Background technology]

[0002] Inspection techniques that utilize terahertz waves are known. Patent Document 1 discloses a method for inspecting for prohibited substances such as narcotics enclosed in sealed letters. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-286716 Summary of the Invention [Problem to be solved by the invention]

[0004] When processing images acquired using terahertz waves for inspection, it may be difficult to obtain sufficient information due to the positional relationship between the object and the detector or the movement of the object. Furthermore, when inspecting a clothed person, the terahertz waves may be scattered by the clothing or the propagation of the terahertz waves may be obstructed by the environment, which may prevent a sufficient inspection. The present invention aims to provide a processing system that can more effectively perform inspections using terahertz waves. [Means for solving the problem]

[0005] The processing system of the present invention includes a first imaging system that captures a first image based on terahertz waves having a frequency range of 30 GHz to 30 THz from an inspection object, a second imaging system that captures a second image of the inspection object based on electromagnetic waves having a wavelength different from the terahertz waves, a control unit that controls the first imaging system, and a processor that processes the first image and the second image, and the control unit controls the first imaging system based on a result of processing the second image by the processor.is a condition when the first image is captured. Control the the condition is one or more selected from the group consisting of an attitude of a camera of the first imaging system, zooming of a camera of the first imaging system, an angle of view of a camera of the first imaging system, a gain for the first image, and a cropping range for the first image; The processor detects an inspection area based on the second image, selects an area of ​​the first image corresponding to the inspection area by comparing the first image with the second image, and performs image processing on the area of ​​the first image corresponding to the inspection area detected from the second image. [Effects of the Invention]

[0006] It is possible to provide a processing system that can more advantageously perform inspections using terahertz waves. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a conceptual diagram of a processing system according to a first embodiment. [Figure 2] 1 is a flowchart of a process according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram of a processing system according to a second embodiment. [Figure 4] 10 is a flowchart of a process according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram of a processing system according to a third embodiment. [Figure 6] 10 is a flowchart of a process according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram of a processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram of a processing system according to a fifth embodiment. [Figure 9] Example of processing system layout. [Figure 10] Example of processing system layout. [Figure 11] Examples of models for machine learning and examples of testing using trained models. [Figure 12] Examples of models for machine learning and examples of control using trained models. DETAILED DESCRIPTION OF THE INVENTION

[0008] The following examples are described in detail with reference to the accompanying drawings. Note that the following examples do not limit the scope of the claimed invention. Although the examples describe multiple features, not all of these features are necessarily essential to the invention, and multiple features may be combined in any desired manner. Furthermore, the same reference numerals are used throughout the accompanying drawings, and redundant explanations will be omitted. Furthermore, in the present invention, terahertz waves include electromagnetic waves in the frequency range of 30 GHz to 30 THz. The concept of electromagnetic waves may also include radio waves such as visible light, infrared light, and millimeter waves.

[0009] (First Example) An outline of a processing system 401 according to the first embodiment will be described with reference to Fig. 1. The processing system 401 includes a first imaging system including a first illumination source 404 and a first camera 402, a second imaging system including a second camera 405, and a processor including a pre-processing unit 406 and a post-processing unit 407.

[0010] A first camera 402 of the first imaging system acquires a first image based on terahertz waves 403 of a first wavelength emitted by a first illumination source 404. The terahertz waves 403 emitted from the first illumination source 404 illuminate an inspection object 410. If the inspection object 410 is a clothed person, the terahertz waves 403 pass through the fibers of the clothing and are reflected by metals or ceramics held by the inspection object 410. In addition, certain substances, such as the explosive RDX (trimethylenetrinitramine), are known to absorb terahertz waves in the range of around 0.8 THz, thereby reducing the reflected waves. A first image is acquired by the first camera 402 based on these reflected waves.

[0011] The second camera 405 of the second imaging system acquires a second image from electromagnetic waves of a different wavelength from the terahertz waves emitted from the first illumination source 404. Visible light, infrared light, or millimeter waves can be used as the electromagnetic waves of a different wavelength. When infrared light is used, an illumination source (not shown) different from the first illumination source 404 may be prepared. The second image acquired by the second camera 405 is subjected to image processing in a pre-processing unit 406. The pre-processing unit 406 performs processing to detect an inspection area from the second image.

[0012] When the second image is acquired using visible light and the subject 410 is a person, the inspection area may be detected by detecting a specific part of the clothing. The inspection area may be identified by creating a model using machine learning and classifying the area of ​​the captured second image using the model. The area may also be identified based on object shape information stored in the database 409. When the second image is acquired using millimeter waves, the inspection area may be detected as a part in the image where the intensity distribution is higher than a predetermined threshold or a part with a large intensity difference. When infrared light is used to acquire the second image, the inspection area may be detected as a part where infrared light radiation is low due to moisture or a specific part of the clothing from an image detected using night vision. The use of infrared light or millimeter waves makes it possible to detect the inspection area even in dark places or when visibility is poor due to weather. When detecting the inspection area from an image of clothing, the inspection area may be detected as an unnatural bulge in the clothing, a person's chest, or a pocket of the clothing.

[0013] The inspection of the inspection object 410 by the processor will be described with reference to FIG. 2. The pre-processing unit 406 detects an inspection area from the second image acquired by the second camera 405 (S421) using the method described above (S422, S423). The post-processing unit 407 processes image data for information on the area of ​​the first image corresponding to the inspection area detected from the second image (S425). The first image is an image acquired by the first camera 402 using terahertz waves (S424) and is an image obtained by looking through clothing, etc. If there is a metal or ceramic object under the clothing, an image from the reflected waves can be obtained. Therefore, the shape of the object can be detected by processing the first image. After the inspection area is detected from the second image, the area in the first image corresponding to the inspection area is selected by comparing the first and second images. Subsequent image processing of the first image is performed on the area corresponding to the inspection area detected from the second image.

[0014] By selecting an area from the first image corresponding to the inspection area and performing image processing, unnecessary information can be reduced and processing can be performed. This reduces the processing load compared to processing the entire image data, enabling faster processing. Therefore, even if the inspection object 410 is moving, features can be detected from the first image multiple times during the short distance and short time it moves. The determination unit 408 estimates the object underneath the clothing based on the multiple detected features (S426). The multiple features may be parts of the object. The determination unit 408 may classify the shape of the object detected from the first image based on data from the database 409. The classification may be performed using a model created by machine learning. When the inspection object 410 moves or due to the positional relationship between the inspection object and the camera, it is possible that the shape information obtained from the image may be part of the object. Even in such cases, the accuracy of the estimation can be improved by classifying the features based on multiple feature information, accumulating the results multiple times, and making a judgment based on the accumulated classification results (S427).

[0015] When this processing system is used in a security monitoring system, the hazard level of the object detected in the inspection area is determined based on the accumulated classification results for the inspection target 410 (S428). The determination may be made based on a machine learning model based on the accumulated classification results. If the inspection target 410 is determined to be carrying a dangerous object, a notification that the inspection target 410 is carrying a dangerous object can be sent to the outside. When the inspection target 410 passes through a gate where the processing system is installed, a warning may be sent from the processing system to the outside. When the inspection target 410 inserts a ticket and passes through a ticket gate, the ticket may be linked to the inspection target 410 to notify the inspection target 410 that it is a subject requiring monitoring. Furthermore, if the second image is generated using visible light, the inspection target 410 can be clearly displayed by displaying an image on a monitor in which the second image and the first image are superimposed. If the determination is pending, the inspection is repeated until a termination condition is met. The termination condition may be the number of inspection repetitions (S429).

[0016] Next, a method for specifying an inspection area from the second image acquired by the second camera 405 using a model (artificial intelligence) created by machine learning will be specifically described.

[0017] 11(a) is a diagram schematically illustrating a model for machine learning, that is, a learning model. In this example, a neural network including an input layer 481, an output layer 483, and at least one intermediate layer 482 is used as the learning model. Image data is input to the input layer 481. Furthermore, the output layer 483 outputs a feature indicating a partial region of the input image.

[0018] The learning model is trained using teacher data with labeled correct answers. That is, the learning model is trained by a method such as backpropagation using a set of data consisting of input image data and labels indicating the inspection target area in the image data. Target areas include, but are not limited to, people, bags, containers, etc. Furthermore, learning may be performed using deep learning by using a convolutional neural network (CNN) as the model.

[0019] FIG. 11(b) is a schematic diagram showing a method for specifying an inspection area using a trained model. A visible light image 484 is input as input. A trained model 485 outputs features indicating the inspection target area. As shown in FIG. 11(b), the output may be in the form of a line surrounding the target area in image 486. Alternatively, the output may be in the form of outputting coordinate information for image processing.

[0020] By using machine learning and artificial intelligence in this way, highly accurate judgments can be made.

[0021] Similarly, when detecting a specific object from the first image, a model (artificial intelligence) created by machine learning may be used. In this case, an image of terahertz waves with the same wavelength as the terahertz waves captured by first camera 402 may be used as training data for learning.

[0022] (Second Example) This embodiment is an example in which a second illumination source 411 that emits terahertz waves is provided in the second imaging system. This embodiment will be described with reference to FIG. 3 . The second illumination source 411 is an illumination source that generates terahertz waves of a second wavelength different from that of the first illumination source 404. As described in the first embodiment, certain substances are known to absorb terahertz waves of specific wavelengths. Therefore, the first illumination source 404 emits terahertz waves of a first wavelength (approximately 0.8 THz in the case of RDX, an explosive) that are easily absorbed by certain substances toward the inspection object 410. If the inspection object 410 possesses a substance that easily absorbs terahertz waves of the first wavelength, reflection from the possessed portion will be reduced. On the other hand, if the terahertz waves of the second wavelength generated by the second illumination source 411 are selected to have a wavelength that is less absorbed by certain substances (approximately 0.5 THz when the first wavelength is 0.8 THz), the certain substance will reflect the terahertz waves of the second wavelength. For the same inspection area, the difference in the reflected waves from a particular material can be used to identify the material.

[0023] The processing of this embodiment will be described with reference to FIG. 4. The pre-processing unit 406 detects an area with high reflection in the second image acquired using terahertz waves of the second wavelength as the inspection area (S431, S432). The post-processing unit 407 acquires a first image (S434) based on terahertz waves of the first wavelength captured by the first camera 402 and starts processing image data for the area of ​​the first image corresponding to the inspection area detected in the second image. The post-processing unit 407 can calculate the difference between information about the inspection area in the second image and information about the area corresponding to the inspection area in the first image (S435). Data for areas with similar reflection and absorption in the second and first images are almost canceled out by calculating the difference between the information. However, data for areas with different reflection and absorption rates at the first and second wavelengths are not canceled out even by calculating the difference between the information. In this way, spectral analysis of the material in the inspection area can be performed by utilizing the difference in absorption rate of terahertz waves between materials. This spectral analysis can be used to estimate the type of material. Furthermore, scattering and reflection from clothing are canceled, so unnecessary signals from clothing and the like can be reduced from the obtained image information, improving the signal-to-noise ratio of the image.

[0024] If the inspector is holding a substance that is highly absorbent of the first wavelength, the substance detected in the inspection area can be classified based on the difference in absorbance between the first wavelength and the second wavelength (S436). The classification can be performed by the determination unit 408 based on the database 409, which stores the relationship between specific substances and wavelengths. The determination unit 408 may also perform classification using a model created by machine learning. In this manner, it can be estimated that the inspection object 410 is holding a substance that absorbs specific wavelengths. It is known that dangerous substances exist among substances that absorb terahertz waves of specific wavelengths. The presence of dangerous substances can be estimated by spectral analysis. The results of multiple spectral analyses can be accumulated to improve detection accuracy (S437). In this way, it is determined that the inspection object 410 may be holding a dangerous substance (S438). This determination may be performed based on a machine learning model based on the accumulated classification results. If it is determined that a dangerous substance is being held, a notification that the inspection object 410 is holding a dangerous substance is sent to the outside. When the inspection target 410 passes through a gate where a processing system is installed, a warning may be issued, or when the person of inspection target 410 inserts a ticket and passes through a ticket gate, the ticket may be linked to inspection target 410, and the person may be notified to the outside as a person requiring monitoring. The wavelength of the terahertz waves emitted by second illumination source 411 may be matched to the absorption spectrum of the substance to be detected, and multiple illumination sources capable of emitting terahertz waves of three or more wavelengths may be combined. If the judgment is pending, the inspection is repeated until a termination condition is met. The termination condition may be the number of inspection repetitions (S439).

[0025] (Third Example) In this embodiment, a specific region is detected in a second image captured by a second imaging system, and then the control unit 412 controls the first illumination source 404 and the first camera 402 of the first imaging system. This embodiment will be described with reference to Figs. 5 and 6.

[0026] The second camera 405 of the second imaging system acquires a second image from electromagnetic waves of a different wavelength from the terahertz waves emitted from the first illumination source 404. The different wavelength electromagnetic waves can be visible light, infrared light, or millimeter waves. The second image acquired by the second camera 405 is subjected to image processing in the pre-processing unit 406. The pre-processing unit 406 detects an inspection area from the second image (S452, S453). The detection of the inspection area is performed as described in the first embodiment.

[0027] The conditions for capturing an image using the first camera are controlled according to the position, range, and state of the inspection area detected from the second image. Control includes controlling the attitude of the first camera, controlling the gain of the captured image, and controlling the imaging range and angle of view, such as zooming and trimming (S454). The output level and wavelength of the terahertz waves emitted from the first illumination source 404 may be changed according to the intensity of the reflected signal from the inspection area and the object in the inspection area. Controlling in this manner improves the accuracy of the inspection. The first imaging system, controlled by the control unit 412, captures a first image based on terahertz waves of the first wavelength (S455). The post-processing unit 407 processes the inspection area based on the captured first image (S456). The determination unit 408 then determines and classifies the object (S457, S458, S459). If the processing system is a security monitoring system, the degree of danger is determined from the accumulation of classification results, and if it is determined that the inspection target 410 is in possession of a dangerous object, a notice that the inspection target 410 is in possession of a dangerous object is sent to the outside. When the inspection target 410 passes through a gate where the processing system is installed, a warning may be sent, and when the inspection target 410 inserts a ticket and passes through a ticket barrier, the ticket may be linked to the inspection target 410, and the inspection target 410 may be designated as a target requiring monitoring. If the determination is pending, the inspection is repeated until a termination condition is met. The termination condition may be the number of inspection repetitions (S460).

[0028] The photographing by the first camera 402 may be controlled by a model (artificial intelligence) created by machine learning from the second image acquired by the second camera 405. This method will be specifically described below.

[0029] 12(a) is a diagram schematically illustrating a model for machine learning, that is, a learning model. In this example, a neural network including an input layer 481, an output layer 483, and at least one intermediate layer 482 is used as the learning model. Image data is input to the input layer 481. Furthermore, the output layer 483 can output a classification result of the subject of the input image.

[0030] The learning model is trained using teacher data with labeled correct answers. That is, the learning model is trained by a method such as backpropagation using a set of data consisting of input image data and labels indicating the inspection target area in the image data. Also, deep learning may be performed by using a convolutional neural network (CNN) as the model.

[0031] The classification of the subject is selected according to the purpose of control. For zoom control, training data labeled with a label indicating whether the subject is small, large, or appropriate size may be used. For gain control, training data labeled with a label indicating whether the subject is underexposed, appropriate, or overexposed may be used. Furthermore, for controlling the switching of the wavelength used by the first camera 402, training data associated with an appropriate wavelength band may be used. Alternatively, for controlling the output of the first illumination source 404, training data classified according to the transmittance of the terahertz waves output by the first illumination source 404 may be used. These training data are merely examples and are not limiting. Furthermore, learning may be performed without training data using deep learning. In this case, learning may be performed by a means for evaluating the results of control performed according to the output for a certain input.

[0032] 12(b) is a schematic diagram showing a method for controlling the first camera 402 using a trained model. A visible light image 487 is input. The trained model 488 can output information indicating the presence or absence of a subject that has low sensitivity to terahertz waves in the wavelength range detected by the first camera 402. Depending on the result, control is performed to increase the output of the first illumination source 404, and an image 489 is obtained.

[0033] By making such a determination using machine learning or artificial intelligence, it is possible to further increase the accuracy of object detection using the first camera 402.

[0034] Similarly, when detecting a specific object from the first image, a model (artificial intelligence) created by machine learning may be used. In this case, an image of terahertz waves with the same wavelength as the terahertz waves captured by first camera 402 may be used as training data for learning.

[0035] (Fourth Example) This embodiment is an example in which an environment monitoring unit 413 is provided to monitor the humidity around the processing unit. This embodiment will be described with reference to Fig. 7. Terahertz waves are easily absorbed by water vapor, but terahertz waves with longer wavelengths are less susceptible to the effects of water vapor. Therefore, the environment monitoring unit 413 is provided to measure humidity and control the imaging system so that it is less susceptible to the effects of the surrounding environment.

[0036] Specifically, when the environment monitoring unit 413 detects that the humidity has increased, it switches the wavelength of the terahertz waves 403 emitted by the first illumination source 404 to a wavelength longer than the currently used wavelength. Alternatively, it may switch to a wavelength that is less susceptible to water vapor depending on the humidity (a wavelength range that exists near 1.2 mm or 0.75 mm and has particularly low atmospheric attenuation). Although the resolution of the image captured by the camera decreases as the wavelength of the terahertz waves increases, the effect of water vapor can be reduced and the inspection can continue.

[0037] (Fifth Example) This embodiment is an example of imaging using terahertz waves of different wavelengths. A second image is acquired using terahertz waves of a second wavelength, which is longer than the wavelength used to capture the first image, and an inspection area is detected from the second image. The inspection area may be determined as an area where an object of a predetermined shape is present or an area where the spectrum of the reflected wave of a predetermined wavelength has changed, using a model created by machine learning.

[0038] This embodiment will be described with reference to FIG. 8. Based on the inspection area detected from the second image, image data of the area of ​​the first image corresponding to the inspection area is processed. The first image captured using terahertz waves of a first wavelength generated from illumination source 404 is Image 1, and the second image captured using a second wavelength generated from illumination source 411 is Image 2. Image 1 is acquired using terahertz waves with a shorter wavelength than Image 2, so it has high resolution and a large amount of information. Therefore, the shape of the object held by the inspection object 410 is clearly visible from the image acquired using terahertz waves. However, because terahertz waves with a short wavelength are used, the depth of field is shallow, making it sensitive to changes in the posture of the inspection object 410. In particular, depending on the posture of the inspection object 410, only a partial shape of the shape of the object held by the inspection object 410 may be acquired.

[0039] On the other hand, the image taken by Image 2 has a lower resolution than Image 1 due to the long wavelength of the terahertz waves, and the shape of the object is unclear. However, because long-wavelength terahertz waves are used, the depth of field is deep and the image is insensitive to changes in the posture of the inspection object 410. In particular, the overall shape of the object held by the inspection object 410 is acquired regardless of the posture of the inspection object 410. By processing Image 2, which has a lower resolution, to identify the position of the object held by the inspection object 410 and then processing the data from Image 1 based on this detected inspection area, the processing load can be reduced and processing can be performed quickly. Therefore, even if the inspection object 410 is moving, the features of the inspection object 410 can be detected multiple times during the short distance and short time it moves, and the object underneath the clothing can be estimated based on the detected features.

[0040] Furthermore, by calculating the difference between Image 1 and Image 2, which were captured using terahertz waves of two different wavelengths, reflections from clothing are canceled out, making it possible to reduce noise in the obtained image information. In particular, the reflections from the entire clothing are mainly composed of scattering, resulting in little difference in intensity and making the image insensitive to changes in the posture of the subject 410 (the acquired image is given random noise overall). Therefore, by calculating the difference between Image 1 and Image 2, the signal from the clothing is canceled out. Furthermore, by calculating the difference, it is possible to obtain an image based on the difference in the absorption rate of terahertz waves for each wavelength of the transmitted material. Therefore, it is also possible to detect the shape of an object containing components other than metal or ceramic from the difference between Image 1 and Image 2.

[0041] The object in the inspection area is estimated by the determination unit 408 by classifying the shape of the object detected from the image 1. When the inspection target 410 moves, the shape of the object obtained from the image is often only partial, so the accuracy of the determination can be improved by accumulating multiple classification results and making a determination based on the accumulated classification results. In the case of a security monitoring system, the level of danger is determined from the accumulation of classification results, and if it is determined that the inspection target 410 is carrying a dangerous object, a notification is issued that the inspection target 410 is carrying a dangerous object. When the inspection target 410 passes through a gate where a processing system is installed, a warning may be issued, or when the inspection target 410 passes through a ticket gate by inserting a ticket, the ticket may be linked to the inspection target 410, and the inspection target 410 may be designated as a target for monitoring.

[0042] (Sixth Example) An application example of the processing system will be described with reference to FIGS. 9 and 10. FIG. 9 shows an example in which a first illumination source 404 of terahertz waves with a first wavelength and a second illumination source 411 of a second wavelength different from the first wavelength are arranged on one side of a doorway 414 of a vehicle or the like. On the other side of the doorway 414, a first camera 402 that captures images based on the first wavelength of terahertz, a second camera 405-1 that captures images based on visible light, infrared light, or millimeter waves, and a second camera 405-2 that captures images based on the second wavelength of terahertz are arranged. Combinations of these cameras and illumination sources can also be used to perform the inspection processes described in Examples 1 to 5. The second camera 405-1 tracks the inspection object 410, and the attitude and angle of view of the first camera 402 can be controlled. Spectral analysis can be performed by adjusting the wavelength of the terahertz waves used by the second camera 405-2 that captures images based on terahertz waves to the absorption rate of the material. Also, by using the second cameras 405-1 and 405-2 to detect the inspection area, it is possible to reduce the processing load on the first image captured by the first camera 402. Furthermore, by utilizing the difference in the absorption rate of materials for the wavelength of terahertz waves, it is also possible to detect the shape of an object containing a material other than metal or ceramic as a component.

[0043] In this embodiment, the second camera 405 uses cameras for visible light, infrared light, millimeter waves, and a second wavelength terahertz wave, but the second camera may use only one of the cameras for visible light, infrared light, millimeter waves, or terahertz. The illumination source and camera may be embedded in a wall, ceiling, or floor so as not to be conspicuous. An illumination source and camera may be located on both the left and right sides of the entrance / exit 414. By locating the illumination source and camera near the entrance / exit 414, the situation where multiple inspection targets 410 overlap is reduced, improving the accuracy of the inspection.

[0044] FIG. 10 illustrates an example in which a processing system is placed near a ticket gate 415 installed at a station. A first illumination source 404 of terahertz waves with a first wavelength and an illumination source 411 of a second wavelength different from the first wavelength are placed on one side of the ticket gate 415. A first camera 402 capturing images based on the first terahertz wavelength, a second camera 405-1 capturing images based on visible light, infrared light, or millimeter waves, and a second camera 405-2 capturing images based on the second terahertz wavelength are placed on the other side of the ticket gate 415. By placing the processing system near the ticket gate 415, overlapping of multiple inspection targets 410 is reduced, improving inspection accuracy. Multiple first and second cameras may be provided. In this case, using multiple cameras can improve detection accuracy, allow for multiple inspection targets, and expand the inspection range.

[0045] The operation of the processing system may be initiated in response to detection of movement of the inspection target 410 by a sensor provided separately from the processing system, opening and closing of a vehicle door, insertion of a ticket into a ticket gate 415, or the like.

[0046] The invention is not limited to the above-described embodiments, and various modifications and variations are possible within the scope of the invention. [Explanation of symbols]

[0047] 401: Processing system, 402: First camera, 403: Terahertz wave, 404: First illumination source, 405: Second camera, 406: Pre-processing unit, 407: Post-processing unit, 408: Determination unit, 409: Database, 410: Inspection object

Claims

1. a first imaging system having a frequency range of 30 GHz to 30 THz and configured to capture a first image based on terahertz waves from an inspection object; a second imaging system that captures a second image of the inspection object based on electromagnetic waves having a wavelength different from the terahertz waves used to capture the first image; a control unit that controls the first imaging system; a processor for processing the first image and the second image; the control unit controls conditions under which the first imaging system captures the first image based on a result of processing performed on the second image by the processor; and the condition is one or more selected from the group consisting of an attitude of a camera of the first imaging system, zooming of a camera of the first imaging system, an angle of view of a camera of the first imaging system, a gain for the first image, and a cropping range for the first image; The processor detects an inspection area from the second image, compares the first image with the second image to select an area of ​​the first image that corresponds to the inspection area, and performs image processing on the area of ​​the first image that corresponds to the inspection area detected from the second image. A processing system comprising:

2. an illumination source that emits the terahertz waves; The processing system according to claim 1 , wherein the control unit controls at least one of a wavelength and an output of the terahertz wave emitted from the illumination source based on a result of the processing.

3. 3. The processing system according to claim 1, wherein the first imaging system and the second imaging system perform imaging based on terahertz waves having different wavelengths.

4. 3. The processing system according to claim 1, wherein the second imaging system performs imaging based on any one of visible light, infrared light, and millimeter waves.

5. 5. The processing system according to claim 1, further comprising a plurality of illumination sources for imaging.

6. 6. The processing system according to claim 1, wherein at least one of the first imaging system and the second imaging system includes a plurality of cameras for imaging.

7. The processing system according to claim 1 , further comprising a determination unit that determines a degree of risk related to the inspection object in accordance with an output of the processor.

8. The processing system according to any one of claims 1 to 7, characterized in that the processor performs processing on the second image based on a learning model learned using at least one of deep learning and machine learning.

9. The processing system according to claim 3 , wherein the processor calculates a difference between information on an area corresponding to the inspection area of ​​the first image and information on an area corresponding to the inspection area of ​​the second image.

10. The processing system according to claim 1 or 9, wherein the detection of the inspection area is based on the shape of the inspection area or information of a spectrum detected from the inspection area.

11. The processing system further includes an environment monitor that measures ambient humidity; 6. The processing system according to claim 4, wherein the control unit controls at least one of a wavelength and an output of the terahertz waves emitted in accordance with the output of the environment monitoring unit.

12. 12. The processing system according to claim 1, wherein the capturing of the first image, the capturing of the second image, and the processing are repeatedly performed.

13. 13. The processing system of claim 1, wherein the processor detects features from the first image.

Images