Classifier generated using data
By using a generative adversarial network to generate training data and train a classifier, the difficulty of detecting objects of interest in penetrating radiation inspection images is solved, and an automated and efficient detection process is achieved.
Patent Information
- Application Number
- CN202080009750.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-01-17
- Filing Date
- 2020-01-15
- Publication Date
- 2025-02-25
- Estimated Expiration
- 2040-01-15
AI Technical Summary
In inspection images generated using penetrating radiation, detecting objects corresponding to the type of interest (such as threats or contraband) is difficult and even in some cases cannot be detected, resulting in the user having to manually check the container, which is Time-consuming.
Generative adversarial networks (GANs) are used to generate training data, and a classifier is trained through a machine learning algorithm to detect whether there are objects corresponding to the type of interest in the penetrating radiation check image.
It realizes automatic and accurate detection of objects of interest in the inspection image, reducing the need for manual inspection and improving detection efficiency.
Smart Images
Figure CN113508388B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to, but is not limited to, generating a classifier configured to detect an object corresponding to a type of interest in an examination image generated using penetrating radiation. The present invention also relates to, but is not limited to, determining whether an object corresponding to a type of interest is present in an examination image generated using penetrating radiation. The present invention also relates to, but is not limited to, producing an apparatus configured to determine whether an object corresponding to a type of interest is present in an examination image generated using penetrating radiation. The present invention also relates to, but is not limited to, corresponding apparatus and computer programs or computer program products. Background Art
[0002] Penetrating radiation can be used to generate an inspection image of a container containing cargo. In some examples, a user may want to detect objects corresponding to a type of interest on the inspection image, such as threats (such as weapons, explosive materials, or radioactive materials) or contraband (such as pharmaceuticals or cigarettes). Detecting such objects can be difficult. In some cases, the object may not be detected at all. When detection is unclear from the inspection image, the user can manually inspect the container, which can be time-consuming for the user. Summary of the Invention
[0003] Aspects and embodiments of the present invention are set out in the appended claims.These and other aspects and embodiments of the present invention are also described herein.
[0004] Any features in one aspect of the invention may be applied to other aspects of the invention in any appropriate combination. In particular, method aspects may be applied to apparatus and computer program aspects, and vice versa.
[0005] Furthermore, features implemented in hardware can generally be implemented in software, and vice versa.Any references to software and hardware features herein should be interpreted accordingly. BRIEF DESCRIPTION OF THE DRAWINGS
[0006] Embodiments of the present invention will now be described by way of example with reference to the accompanying drawings, in which:
[0007] Figure 1 A flow chart illustrating an example method according to the present disclosure is shown;
[0008] Figure 2 Schematically shows a system configured to implement Figure 1 Example systems and example devices of example methods;
[0009] Figure 3 An example inspection image according to the present disclosure is shown;
[0010] Figure 4The example generated training images are schematically shown;
[0011] Figure 5 Shown in the figure Figure 1 a flowchart of the details of an example method;
[0012] Figure 6 Schematically shows a system configured to implement Figure 1 Example Generative Adversarial Networks for an Example Method;
[0013] Figure 7 An example synthetic object is schematically illustrated;
[0014] Figure 8 shows a flowchart illustrating another example method according to the present disclosure; and
[0015] Figure 9 A flowchart illustrating another example method according to the present disclosure is shown.
[0016] In the drawings, similar elements have the same reference numerals.
[0017] Detailed Description of Exemplary Embodiments
[0018] The present disclosure discloses an example method for generating a classifier configured to detect objects corresponding to a type of interest in an inspection image generated using penetrating radiation (e.g., x-rays, although other penetrating radiation is contemplated). In a typical example, the type of interest may be a threat, such as a weapon (e.g., a gun or rifle), explosive material, radioactive material, and / or the type of interest may be contraband, such as, for example, pharmaceuticals or cigarettes, as non-limiting examples.
[0019] The present disclosure also discloses an example method for determining whether an object corresponding to a type of interest is present in an inspection image generated using penetrating radiation.
[0020] The present disclosure also discloses an example method for producing an apparatus configured to determine whether an object corresponding to a type of interest is present in an inspection image generated using penetrating radiation.
[0021] The present disclosure also discloses a corresponding device and a computer program or computer program product.
[0022] Figure 1 The diagram shows a method for generating Figure 2 and Figure 3 A flowchart of an example method 100 of a classifier 1 is shown in FIG. Figure 2 Also shown are methods that can be configured by method 100 to determine the Figure 31. The device 15 checks whether an object 11 corresponding to the type of interest is present in the image 1000 shown in FIG. The inspection image 1000 is generated by the device 15 using penetrating radiation, for example. Figure 1 The method 100 shown in FIG. 1 also combines Figure 4 Described here, a plurality of generated training images 101 are shown, which include objects 110 corresponding to a type of interest, for training the classifier 1 .
[0023] Figure 1 The method 100 generally includes:
[0024] At S1 , training data including a plurality of training images 101 including objects 110 corresponding to a type of interest is generated; and
[0025] At S2 , the classifier 1 is trained using the generated training data by applying a machine learning algorithm.
[0026] As disclosed in more detail later, the training of the classifier 1 may be performed using generated training data or a combination of observed (ie real) training data and generated training data.
[0027] like Figure 5 As shown in , in some examples, generating training data at S1 may include: at S11, using a generative adversarial network 2, also known as GAN2, which includes a generator 21 and a discriminator 22, both of which are Figure 6 As described in more detail later, the generator 21 includes a neural network, and the discriminator 22 includes a neural network.
[0028] like Figure 5 As shown, in some examples, generating training data at S1 may further include: at S12, training the generator 21 and the discriminator 22 using the initial training data, the initial training data including the following: Figure 6 A plurality of observed images 102 are shown generated using penetrating radiation. In some examples, one or more of the observed images 102 include an observed object 120 corresponding to a type of interest.
[0029] Return Reference Figure 5 In some examples, generating training data at S1 may include: generating a plurality of training images 101 at S13 .
[0030] As will be shown later with reference to method 200 Figure 8 As described in more detail, configuration of device 15 involves, for example, storing classifier 1 at device 15 at S22. In some examples, classifier 1 may be obtained at S21 (e.g., by Figure 1In some examples, obtaining the classifier 1 at S21 may include: receiving the classifier 1 from another data source.
[0031] As described above, the classifier 1 is derived from training data using a machine learning algorithm and is arranged to produce an output indicating whether an object 11 corresponding to a type of interest in the inspection image 1000 is detected.
[0032] The classifier 1 is arranged to be stored in the memory 150 (eg Figure 2 ), it is easier to detect its presence or absence later on, even though the process 100 for deriving the classifier 1 from the training data may be computationally intensive.
[0033] Once configured, the device 15 can provide accurate detection of the object 11 by applying the classifier 1 to the inspection image 1000. The detection process is Figure 9 (described later) is shown (as process 300).
[0034] Computer systems and detection devices
[0035] Figure 2 Schematically illustrates a system configured to at least partially implement Figure 1 10 and device 15. In particular, in a preferred embodiment, computer system 10 executes a machine learning algorithm to generate a classifier 1 for storage on device 15. Although a single device 15 is shown for clarity, computer system 10 can communicate and interact with multiple such devices. The training data itself can be obtained using multiple observed images 102, which are acquired using device 15 and / or other similar devices and / or other sensors and data sources.
[0036] Figure 2 The computer system 10 includes a memory 11 , a processor 12 , and a communication interface 13 .
[0037] System 10 may be configured to communicate with one or more devices 15 via interface 13 and link 30 (eg, a Wi-Fi connection, although other types of connections are also contemplated).
[0038] Memory 11 is configured to store, at least in part, data for use, for example, by processor 12. In some examples, the data stored on memory 11 may include data such as training data (and data used to generate the training data) and / or GAN 2 and / or machine learning algorithms.
[0039] In some examples, processor 12 of system 10 may be configured to at least partially execute Figure 1 Method 100 and / or Figure 8 Method 200 and / or Figure 9 At least some of the steps of method 300.
[0040] Figure 2 The detection device 15 comprises a memory 151 , a processor 152 and a communication interface 153 (for example a Wi-Fi connection, but other types of connections are envisaged) allowing connection to the interface 13 via the link 30 .
[0041] In a non-limiting example, as described in more detail later, the apparatus 15 may also include a device 3 that acts as an inspection system. The device 3 may be integrated into the apparatus 15 or connected to other parts of the apparatus 15 by a wired or wireless connection.
[0042] In some examples, such as Figure 2 As shown in , the present disclosure can be applied to the inspection of a real container 4 containing a real object 111 corresponding to a type of interest. At least some of the methods of the present disclosure may include, for example, obtaining an observation image 102 by irradiating one or more real objects 111 corresponding to the type of interest and / or one or more real containers 4 configured to contain cargo using penetrating radiation, and detecting radiation from the irradiated one or more real objects 111 and / or the irradiated one or more real containers 4. Alternatively or additionally, at least some of the methods of the present disclosure may include obtaining an inspection image 1000 by irradiating one or more real containers 4 configured to contain cargo using penetrating radiation, and detecting radiation from the irradiated one or more real containers 4. The irradiation and / or detection may be performed using one or more devices configured to inspect the real container 4.
[0043] In other words, the apparatus 3 may be used to acquire a plurality of observation images 102 for generating training data and / or to acquire inspection images 1000 .
[0044] In some examples, processor 152 of device 15 may be configured to at least partially execute Figure 1 Method 100 and / or Figure 8 Method 200 and / or Figure 9 At least some of the steps of method 300.
[0045] Obtaining training data
[0046] Return Reference Figure 1 , a classifier 1 is generated based on the training data generated at S1. The training of the classifier 1 may be performed using the generated training data or a combination of observed (ie, real) training data and generated training data.
[0047] The classifier 1 is trained using training data, each training data corresponding to an instance where the classification (e.g., "object present" or "object not present") is known. As described in more detail later, the training data may include:
[0048] Figure 4 and 6 a training image 101 in which an object 110 is present, the training image 101 being used alone or in combination with an observation image 102 in which an object is present (ie, “object not present”), and
[0049] An image 102 is observed in which no object is present (ie, “object not present”).
[0050] Return Reference Figure 5 and 6 , the GAN 2 used in S11 includes a generator 21 and a discriminator 22.
[0051] In some examples, during training of S12, the generator 21 may generate a synthetic image 103 (i.e., a pseudo image) containing a synthetic object 123 (i.e., a pseudo object) from at least random values (e.g., random noise 124) based on initial training data including an observed image 102 (i.e., a real image) containing an observed object 120 (i.e., a real object).
[0052] The observed image 102 may correspond to a Figure 2 3. Alternatively, the observed images 102 may be obtained in a different environment, such as using a similar device (or equivalent sensor set) installed in a different (but preferably similar) environment, or in a controlled test configuration in a laboratory environment.
[0053] In some examples, during training at S12 , discriminator 22 may aim to classify images including synthesized images 103 (ie, fake images) and observed images 102 (ie, real images) according to the nature of the images (ie, real or fake).
[0054] The generator 21 is configured to attempt to deceive the discriminator 22 that the composite image 103 is authentic. The discriminator 22 is configured to attempt to classify the composite image 103 as forged and to classify the observed image 102 as authentic.
[0055] The generator 21 and the discriminator 22 compete with each other based on a loss function. During the training of S12, the success of the generator 21 and the discriminator 22 is defined by the loss function. The discriminator 21 is configured to try to minimize the loss function. The generator 22 is configured to try to maximize the loss function.
[0056] In some examples, the loss function may be an Euler-Lagrangian type function.
[0057] In some examples, the loss function may include at least one of a least mean square function and / or a combination of a weighted Gaussian kernel.
[0058] In some examples, the combination of weighted Gaussian kernels includes combining C(P, Q, D) such that:
[0059]
[0060] Where: n is the number of cores,
[0061] E[X] is the expected value of the function X;
[0062] D(x) is the output of the discriminator 22 for image x,
[0063] D(y) is the output of the discriminator 22 for image y,
[0064] P is the probability distribution of the observed image 102,
[0065] Q is the probability distribution of the synthesized image 103,
[0066] (wk) is the set of positive real numbers such that ∑wk=1, and
[0067] σk is a set of strictly positive numbers that serves as the standard deviation parameter of the Gaussian kernel.
[0068] In some examples, P and Q may be absolutely continuous with respect to the measurement. In a non-limiting example, Q may be absolutely continuous with respect to the Lebessue measurement, and P may or may not be absolutely continuous with respect to the Lebessue measurement.
[0069] In some examples, the combination C(P,Q,D) includes the centers of gravity of any number of Gaussian kernels, and as non-limiting examples, includes one to thirteen Gaussian kernels, such as three Gaussian kernels, five Gaussian kernels, seven Gaussian kernels, or ten Gaussian kernels.
[0070] Return Reference Figure 6 , the generator 21 includes one or more deconvolution layers and transposed convolution layers. The deconvolution layer may have a kernel size of, for example, 4×4 and / or a stride of, for example, 2×2. The kernel size of a layer is the size of the filter applied by the layer to be trained, and the stride of a layer is the distance between two consecutive positions of the kernel when applying the filter to the input of the layer.
[0071] like Figure 6As shown, the generator 21 may include a module 211 configured to project and reshape the noise 124. The reshaping step may fit the input 124 to the correct size for the subsequent deconvolution step. The generator 21 may also include a first module 212, a second module 213, a third module 214, and a fourth module 215. Each module 212, 213, 214, and 215 may be a deconvolution layer. The first module 212 may have a kernel size of, for example, 4×4 and a stride of, for example, 2×2, the second module 213 may have a kernel size of, for example, 4×4 and a stride of, for example, 2×2, the third module 214 may have a kernel size of, for example, 4×4 and a stride of, for example, 2×2, and the fourth module 215 may have a kernel size of, for example, 4×4 and a stride of, for example, 2×2. The generator 21 may also include a fifth module 216. The fifth module 216 may be a transposed convolution layer having a kernel size of, for example, 4×4. In Figure 6 In the example, the generated image 103 may have a size of 80×80.
[0072] Other structures are also conceivable for the generator 21. For example, a deeper structure is conceivable and / or a structure with Figure 6 The structure shown is of the same shape as the structure, which will produce an image 103 having a size different from 80×80 (e.g., 120×120, 150×150, 200×200, or a multi-channel image).
[0073] Return Reference Figure 6 , the discriminator 22 includes one or more convolutional layers, and a fully connected linear activation layer. The convolutional layer may include a kernel size of, for example, 4×4 and a stride of, for example, 2×2 or 1×1.
[0074] The discriminator 22 may include a first module 221, a second module 222, a third module 223, a fourth module 224, and a fifth module 225. The first module 221 may have a kernel size of, for example, 4×4 and a stride of, for example, 2×2, the second module 222 may have a kernel size of, for example, 4×4 and a stride of, for example, 2×2, the third module 223 may have a kernel size of, for example, 4×4 and a stride of, for example, 2×2, and the fourth module 224 may have a kernel size of, for example, 4×4 and a stride of, for example, 1×1. The fifth module 225 may be a fully connected linear activation layer.
[0075] Other structures are also conceivable for the discriminator 22. The discriminator 22 is closely connected to the generator 21 and can be modified similarly to the generator 21. Therefore, the discriminator 22 can be more Figure 6 The illustrated discriminator 22 is deeper and / or adapted for input image sizes other than 80x80.
[0076] Return Reference Figures 4 to 6Generating multiple training images 101 at S13 may include generating synthetic objects 123 corresponding to the type of interest (e.g., synthetic objects 123 are generated by generator 21). In some examples, each synthetic object 123 may be generated using at least a portion of an observed object 120 in one or more observed images 102.
[0077] like Figure 7 As shown, generating synthetic object 123 may include at least one of rotating (as indicated by arrow 71) and / or translating (as indicated by arrow 72) and / or flipping (as indicated by arrow 73) at least a portion of object 120. Alternatively or additionally, in some examples, generating synthetic object 123 may include combining a portion of first object 120 with another portion of second object 120 (as indicated by arrow 74) to obtain a new synthetic object 123 corresponding to the type of interest. In some examples, the combination represented by arrow 74 may utilize an intermediate point in the appearance manifold of object 120.
[0078] Return Reference Figure 2 and Figure 3 As a non-limiting example, the present disclosure may be applied to an inspection image 1000 including a representation 41 of a container 4 containing an object 111 corresponding to a type of interest.
[0079] In such an example, when training generator 21 and discriminator 22 at S12, one or more observed images 102 may include representation 410 of container 4. In such an example, generating composite image 103 (and training image 101 as described below) may include using at least a portion of representation 410 of container 4 in one or more observed images 102. In some examples, the entire representation 410 may be used in composite image 103.
[0080] In some examples, generating the synthetic image 103 (and the training image 101) may also include using one or more synthetic objects 123. Figure 6 As shown in , composite image 103 may include at least a portion of representation 410 and one or more composite objects 123 .
[0081] In some examples, generating the composite image 103 may include combining at least a portion of the representation 410 of the container 4 in the one or more observed images 102 with the one or more composite objects 123 using the Beer-Lambert law.
[0082] A synthetic image 103 classified as real by the discriminator 22 may be used as a training image 101 as part of the training data for the classifier 1 , eg as part of the training data in which the object 110 is present.
[0083] Once the generator 21 has been trained, the generator 21 may generate synthetic images 103 that are sufficiently realistic to be used as training images 101 as part of training data for the classifier 1 , for example as part of training data in which the object 110 is present.
[0084] GAN 2 enables the generation of a large number of training images 101. In some examples, GAN 2 can generate thousands of training images 101, e.g., hundreds of thousands of training images 101, e.g., millions of training images 101, from a relatively small amount of initial training data including observed images 102 (e.g., dozens of observed images 102 or less). GAN 2 enables the generation of training images 101 that include real objects 110. GAN 2 enables the generation of training images 101 that include objects 110 that are not in observed images 102, and GAN 2 enables the addition of variability in the range of objects 110 and / or the acquisition of a larger range of objects 110 of a type of interest.
[0085] The generated training images 101 , given their large number and / or their real objects 110 and / or the variability of their objects 110 , enable the classifier 1 to be improved.
[0086] Generate a classifier
[0087] Return Reference Figure 1 , a classifier is constructed by applying a machine learning algorithm to the training data 1. Any suitable machine learning algorithm can be used to construct the classifier. For example, a method based on a convolutional neural network can be used.
[0088] The learning process is typically computationally intensive and may involve a large amount of training data. In some examples, processor 12 of system 10 may include greater computing power and memory resources than processor 152 of device 15. Classifier generation is therefore performed at least partially at computer system 10, remotely from device 15. In some examples, at least steps S1 and / or S2 of method 100 are performed by processor 12 of computer system 10. However, if sufficient processing power is available locally, classifier learning can be performed (at least partially) by processor 152 of device 15.
[0089] The machine learning step involves inferring behaviors and patterns based on training data and encoding the detected patterns in the form of a classifier1.
[0090] Return Reference Figure 4 and 6 , the training data may include a training image 101 obtained through the process at S1.
[0091] The classification labels (specifying known presence or absence states) of the training data may be known in advance. In some examples, the training data includes generated training data. The training data may include one or more objects 110 (e.g., Figure 4 ). Figure 4 The classification label of the training image 101 is “object 11 exists.” The generated training data can be used alone or in combination with the observation image 102 .
[0092] The training data also includes training images that do not include any objects corresponding to the type of interest. The training images that do not include any objects corresponding to the type of interest may correspond to images acquired by the same device 15 being configured (e.g., using a Figure 2 15 is configured using apparatus 3 of the apparatus 15 depicted in situ (e.g., past data (e.g., past images) acquired in situ. Alternatively, the past data may be obtained in a different environment, for example, using similar equipment (or an equivalent set of sensors) installed in a different (but preferably similar) environment, or in a controlled test configuration in a laboratory environment. The classification labels for training images that do not include any objects corresponding to the type of interest may be known in advance (e.g., when the training images are derived experimentally in a laboratory environment) or may be obtained using auxiliary equipment. In some examples, a domain expert may manually label the training images with a ground truth classification (e.g., “no object” for the image). The classification label for a training image that does not contain any object corresponding to the type of interest is “object not present” (i.e., “no object corresponding to the type of interest”).
[0093] Equipment Manufacturing
[0094] like Figure 8 As shown, a method 200 of producing an apparatus 15 configured to determine whether an object 11 corresponding to a type of interest is present in an inspection image 1000 generated using penetrating radiation may include:
[0095] At S21 , a classifier 1 generated by the method 100 according to any aspect of the present disclosure is obtained; and
[0096] At S22 , the obtained classifier 1 is stored in the memory 151 of the device 15 .
[0097] At S22, the classifier 1 can be stored in the detection device 15. The classifier 1 can be created and stored using any suitable representation, for example, as a data description including data elements specifying the classification criteria and their classification output (e.g., the presence or absence of the object 11). Such a data description can be encoded, for example, using XML or using a customized binary representation. The data description is then interpreted by the processor 152 running on the device 15 when the classifier 1 is applied.
[0098] Alternatively, the machine learning algorithm may generate the classifier 1 directly into executable code (e.g., machine code, virtual machine bytecode, or an interpretable script). This may be in the form of a code routine that the device 15 can call to apply the classifier 1.
[0099] Regardless of how the classifier 1 is represented, the classifier 1 effectively defines a decision algorithm (including a set of rules) for classifying the presence status of the object 11 based on input data (ie, the inspection image 1000 ).
[0100] After the classifier 1 is generated, the classifier 1 is stored in the memory 151 of the device 15. The device 15 can be temporarily connected to the system 10 to transfer the generated classifier (e.g., as a data file or executable code), or the transfer can occur using a storage medium (e.g., a memory card). In a preferred method, the classifier is transmitted from the system 10 to the device 15 via the network connection 30 (this can include transmission from a central location of the system 10 to a local network where the device 15 is located via the Internet). The classifier 1 is then installed on the device 15. The classifier can be installed as part of a firmware update of the device software, or installed independently.
[0101] Installation of the classifier 1 may be performed once (eg at the time of manufacture or installation) or repeatedly (eg as a periodic update). The latter approach may allow the classification performance of the classifier to improve over time as new training data becomes available.
[0102] Apply a classifier to perform object detection
[0103] The presence or absence classification is based on classifier 1.
[0104] After the device 15 has been configured with the classifier 1 , the device 15 can use the classifier based on the locally acquired inspection image 1000 to detect whether the object 11 is present in the inspection image 1000 .
[0105] Generally, the classifier 1 is configured to detect objects 11 corresponding to a type of interest in an inspection image 1000 generated using penetrating radiation, the inspection image 1000 including one or more features at least similar to training data used to generate the classifier 1 by a machine learning algorithm.
[0106] Figure 9 A flow chart illustrating an exemplary method 300 for determining whether an object corresponding to a type of interest is present in an inspection image generated using penetrating radiation. The method 300 is performed by the device 15 (e.g., Figure 2 shown).
[0107] The method 300 includes:
[0108] At S31, an inspection image 1000 is obtained;
[0109] At S32 , applying the classifier 1 generated according to the method of any aspect of the present disclosure to the obtained image 1000 ; and
[0110] At S33 , it is determined based on the application whether an object corresponding to the type of interest exists in the inspection image.
[0111] The classifier includes a plurality of output states. In some examples, the classifier is configured to output one of: a state corresponding to the presence of an object of the type of interest in the inspection image; and / or a state corresponding to the absence of an object of the type of interest in the inspection image.
[0112] Optionally, the method 300 may further include, for example, outputting trigger data to trigger an alarm in response to detecting the object 11 corresponding to the type of interest in the inspection image 1000 at S34 .
[0113] The alarm may comprise an alarm signal (visual or audible), for example for triggering further inspection (eg manual check) of the container 4 (eg for verification).
[0114] Further details and examples
[0115] The present disclosure may be advantageous in, but not limited to, customs and / or security applications.
[0116] The present disclosure generally has application to cargo inspection systems (eg, sea or air cargo).
[0117] Figure 2 The device 3 serves as an inspection system, which is configured to inspect the container 4 , for example by transmitting inspection radiation through the container 4 .
[0118] As a non-limiting example, a container 4 configured to hold cargo may be placed on a vehicle. In some examples, the vehicle may include a trailer configured to carry the container 4.
[0119] Figure 2 The apparatus 3 may comprise a source 5 configured to generate examination radiation.
[0120] The radiation source 5 is configured so that the cargo is inspected through the material (typically steel) of the wall of the container 4, for example for detection and / or identification of the cargo. Alternatively or additionally, a portion of the inspection radiation may be transmitted through the container 4 (the material of the container 4 is therefore transparent to the radiation), while another portion of the radiation may be at least partially reflected by the container 4 (referred to as "backscatter").
[0121] In some examples, apparatus 3 may be mobile and transportable from one location to another (apparatus 3 may include a motor vehicle).
[0122] In the source 5 , electrons are typically accelerated at a voltage between 100 keV and 15 MeV.
[0123] In a mobile inspection system, for steel penetration capability, for example between 40 mm and 400 mm, typically 300 mm (12 inches), the power of the X-ray source 5 can be, for example, between 100 keV and 9.0 MeV, typically 300 keV, 2 MeV, 3.5 MeV, 4 MeV or 6 MeV.
[0124] In a static inspection system, for steel penetration capability, for example, between 300 mm and 450 mm, typically 410 mm (16.1 inches), the power of the X-ray source 5 may be, for example, between 1 MeV and 10 MeV, typically 9 MeV.
[0125] In some examples, the source 5 may emit continuous X-ray pulses. The pulses may be emitted at a given frequency, comprised between 50 Hz and 1000 Hz, for example approximately 200 Hz.
[0126] According to some examples, such as Figure 2 As shown, the detector can be mounted on a frame. The frame can be formed into an inverted "L", for example. In a mobile inspection system, the frame can include an electro-hydraulic boom that can be moved between a retracted position (not shown) in transport mode and an inspection position ( Figure 2 ) is operated. The boom may be operated by a hydraulic actuator (e.g., a hydraulic cylinder). In a static inspection system, the frame may comprise a static structure.
[0127] It should be understood that the inspection radiation source can include other penetrating radiation sources, such as, for example, non-limiting examples, ionizing radiation sources, such as gamma rays or neutrons. The inspection radiation source can also include sources not adapted to be activated by an electrical source, such as radioactive sources, such as those using Co60 or Cs137. In some examples, the inspection system includes a detector, such as an X-ray detector and optionally a gamma and / or neutron detector, adapted to detect the presence of radioactive gamma and / or neutron emitting materials within the payload, for example, concurrently with an X-ray inspection. In some examples, the detector can be positioned to receive radiation reflected by the container 4.
[0128] In the context of the present disclosure, the container 4 may be any type of container, such as a holder or a box, etc. Thus, as non-limiting examples, the container 4 may be a pallet (e.g. a pallet of European standard, American standard or any other standard) and / or a train wagon and / or a tank and / or a trunk of a vehicle and / or a “transport container” (e.g. a tank or an ISO container or a non-ISO container or a unit load device (ULD) container).
[0129] In some examples, one or more memory elements (e.g., the memory of one of the processors) can store data used for the operations described herein. This includes memory elements capable of storing software, logic, code, or processor instructions that are executed to perform the activities described in this disclosure.
[0130] The processor can execute any type of instruction associated with data to realize the operation described in detail herein in this disclosure. In one example, the processor can transform an element or article (e.g., data) from one state or thing to another state or thing. In another example, the activity outlined herein can be implemented with fixed logic or programmable logic (e.g., software / computer instructions executed by a processor), and the elements identified herein can be a certain type of programmable processor, programmable digital logic (e.g., field programmable gate array (FPGA), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)), ASIC including digital logic, software, code, electronic instructions, flash memory, optical disc, CD-ROM, DVD ROM, magnetic card or optical card, other types of machine-readable media suitable for storing electronic instructions, or any suitable combination thereof.
[0131] As one possibility, a computer program, a computer program product, or a computer-readable medium is provided that includes computer program instructions to cause a programmable computer to execute any one or more of the methods described herein. In an example implementation, at least some portions of the activities associated with the processor may be implemented in software. It should be understood that the software components of the present disclosure may be implemented in ROM (read-only memory) form if desired. If desired, the software components may generally be implemented in hardware using conventional techniques.
[0132] In the context of this disclosure, other changes and modifications of the system will be apparent to those skilled in the art, and the various features described above may have advantages with or without the other features described above. The above embodiments are to be understood as illustrative examples, and additional embodiments are contemplated. It should be understood that any feature described with respect to any one embodiment may be used alone or in combination with the other features described, and may also be used in combination with one or more features of any other embodiment, or in any combination of any other embodiment. In addition, equivalents and modifications not described above may also be employed without departing from the scope of the invention as defined in the appended claims.
Claims
1. A method for generating a classifier, the classifier being configured to detect an object corresponding to a type of interest in an inspection image generated using penetrating radiation, the inspection image comprising a representation of a container containing the object corresponding to the type of interest, the method comprising: Obtaining an observed image, comprising: irradiating one or more real objects corresponding to the type of interest and / or one or more real containers configured to contain goods using penetrating radiation, and detecting radiation from the irradiated one or more real objects and / or the irradiated one or more real containers to obtain one or more observed images, wherein the one or more observed images include representations of objects corresponding to the type of interest and containers containing the objects; generating training data including a plurality of training images including objects corresponding to the type of interest using a generative adversarial network including a generator and a discriminator; and By applying a machine learning algorithm, the classifier is trained using the generated training data, Wherein, generating the training data comprises: training the generator and the discriminator using initial training data, the initial training data comprising a plurality of the observed images; and The generator generates synthetic objects corresponding to the type of interest, and uses one or more synthetic objects to generate synthetic images, and the synthetic images classified as real by the discriminator are used as the training data, wherein each synthetic object is generated using at least a portion of an object in the one or more observed images, and the synthetic image is generated using at least a portion of the representation of the container in the one or more observed images.
2. The method according to claim 1, wherein generating the composite object comprises at least one of: rotating and / or translating and / or flipping at least a portion of the object in the one or more observed images and / or combining a portion of a first object in the one or more observed images with another portion of a second object.
3. The method according to claim 1, wherein the generator and the discriminator compete with each other based on a loss function, the loss function comprising at least one of the following: a combination of a least mean square function and / or a weighted Gaussian kernel.
4. The method of claim 3, wherein the combination of the weighted Gaussian kernels comprises a combination C(P, Q, D) such that: in: n is the number of cores, E[X] is the expected value of function X; D(x) is the output of the discriminator (22) for image x, D(y) is the output of the discriminator (22) for image y, P is the probability distribution of the observed image (102), Q is the probability distribution of the synthetic image (103), w k is a set of positive real numbers such that ∑ w k =1, and σ k is a set of strictly positive numbers that serves as the standard deviation parameter of the Gaussian kernel. The method of claim 4 , wherein the combination C(P,Q,D) comprises the centroids of any number of Gaussian kernels. The method according to claim 5 , wherein the number of the Gaussian kernels is three Gaussian kernels, five Gaussian kernels, seven Gaussian kernels or ten Gaussian kernels.
7. The method of claim 1, wherein the generator comprises one or more deconvolution layers and transposed convolution layers.
8. The method according to claim 7, wherein: The one or more deconvolution layers have a kernel size of 4×4 and a stride of 2×2.
9. The method of claim 1, wherein the discriminator comprises one or more convolutional layers and fully connected linear activation layers.
10. The method according to claim 9, wherein: The one or more convolutional layers have a kernel size of 4×4 and a stride of 2×2 or 1×1.
11. The method of claim 1 , wherein generating the composite image comprises using the Beer-Lambert law.
12. The method of claim 1, wherein the illuminating and / or the detecting are performed using one or more devices configured to inspect actual containers.
13. A method according to any one of claims 1 to 12, wherein the classifier is configured to detect objects corresponding to a type of interest in an inspection image generated using penetrating radiation, the inspection image comprising at least one or more features corresponding to the training data used to generate the classifier by the machine learning algorithm.
14. A method according to any one of claims 1-12, wherein the classifier includes multiple output states, and the classifier is configured to output one of the following: a state corresponding to the presence of an object of the type of interest in the inspection image; and / or a state corresponding to the absence of an object of the type of interest in the inspection image.
15. The method according to any one of claims 1 to 12, wherein: The machine learning algorithm includes a convolutional neural network.
16. The method of any one of claims 1-12, wherein the method is performed at a remote computer system separate from an apparatus configured to inspect an actual container.
17. A method for determining whether an object corresponding to a type of interest is present in an inspection image generated using penetrating radiation, the method comprising: obtaining inspection images; applying a classifier generated by the method according to any one of claims 1 to 16 to the obtained image; as well as Based on the application, it is determined whether an object corresponding to the type of interest is present in the inspection image.
18. The method of claim 17, wherein obtaining the inspection image comprises: irradiating one or more actual containers configured to contain cargo using penetrating radiation; as well as Radiation is detected from the irradiated real container or containers.
19. A method of producing a device configured to determine whether an object corresponding to a type of interest is present in an inspection image generated using penetrating radiation, the method comprising: Obtaining a classifier generated by the method according to any one of claims 1 to 16; as well as The obtained classifier is stored in a memory of the device.
20. The method of claim 19, wherein the storing comprises transmitting the generated classifier to the device via a network, the device receiving and storing the classifier.
21. The method of claim 19 or 20, wherein the classifier is generated, stored and / or transmitted in the form of one or more of the following: data representation of the classifier; Executable code for applying the classifier to one or more inspection images.
22. The method of claim 19 or 20, wherein the type of interest comprises at least one of: arms; Explosive materials; Radioactive materials; medicines; and cigarette.
23. The method of claim 19 or 20, wherein using penetrating radiation comprises irradiating by transmission.
24. A device configured to determine whether an object corresponding to a type of interest exists in an inspection image generated using penetrating radiation, the device comprising a memory storing a classifier generated by the method according to any one of claims 1 to 16.
25. The apparatus of claim 24, further comprising a processor, and wherein: The memory of the device further comprises instructions which, when executed by the processor, enable the processor to perform a method according to any one of claims 17 to 23.
26. A computer program product comprising instructions which, when executed by a processor, enable the processor to perform a method according to any one of claims 1 to 23 or to provide an apparatus according to claim 24 or claim 25.