KR20230074354A - Electronic device detecting dangerous object on x-ray image based on artificial intelligence model and operating method thereof - Google Patents

Abstract

The present invention relates to a method for detecting a dangerous object and an electronic device performing the same. To this end, the method for an electronic device to detect a dangerous object according to one embodiment of the present invention may comprise: a step of acquiring an X-ray image including at least one object to be inspected; a step of dividing the acquired X-ray image into a plurality of frame images with a predetermined frame interval; a step of acquiring gamma-conversion images by performing gamma correction with respect to the plurality of divided frame images; a step of acquiring an object detection result from the artificial intelligence model by inputting the gamma-conversion images in the artificial intelligence model outputting the dangerous object detection result within the X-ray image when the gamma conversion images are inputted; a step of visualizing the object detection result to generate visual contents; and a step of outputting the generated visual contents.

Description

Electronic device for detecting a dangerous object on an X-ray image based on an artificial intelligence model and its operating method

The present disclosure relates to X-ray inspection equipment and an operating method thereof. More specifically, it relates to a method for detecting a dangerous object on an X-ray image and an electronic device for performing the same.

Types of X-ray inspection equipment can be divided into small baggage inspection equipment and large cargo inspection equipment. In the case of inspection using X-ray scanning equipment, there is an advantage in that unauthorized transport goods and concealed goods can be easily identified in real time by passing through the X-ray without opening the cargo.

An X-ray image is an image generated by projecting an object using X-rays having strong penetrating power and then absorbing the beam using a detector. X-ray images are a type of electromagnetic wave and are used for security checks at airports.

X-rays are a type of electromagnetic wave and are characterized by a short wavelength, high frequency, and strong penetrating power. However, when objects overlap or overlap each other, the absorption power of X-rays may vary due to differences in density of the objects. Objects with low density may not be able to transmit light, and conversely, if the density is too high, images may not be easily revealed, which may limit security screening using X-rays.

Therefore, there is a demand for developing a technology for accurately identifying an object using an X-ray image even when the objects overlap or overlap.

Korea Patent No. 2063859

According to an embodiment, a method of detecting a dangerous object using an X-ray image by an electronic device and an electronic device performing the same may be provided.

According to an embodiment, a method for an electronic device to detect a dangerous object on an X-ray image using an artificial intelligence model and an electronic device for performing the detection may be provided.

According to an embodiment of the present disclosure for achieving the above technical problem, a method for detecting a dangerous object by an electronic device includes obtaining an X-ray image including at least one object to be inspected; Dividing the obtained X-ray image into a plurality of frame images having a preset frame interval; acquiring gamma-converted images by performing gamma correction on the divided plurality of frame images; obtaining an object detection result from the artificial intelligence model by inputting the gamma-converted images to an artificial intelligence model that outputs a dangerous object detection result in the X-ray image when the gamma-converted images are input; generating visual content by visualizing the object detection result; and outputting the generated visual content. Including, a method may be provided.

According to another embodiment for achieving the above-described technical problem, an electronic device for detecting a dangerous object includes a display; network interface; a memory that stores one or more instructions; and at least one processor executing the one or more instructions; wherein the at least one processor obtains an X-ray image including at least one object to be inspected by executing the one or more instructions, and converts the acquired X-ray image into a plurality of frames having a preset frame interval. An artificial intelligence model that acquires gamma-converted images by dividing into images and performing gamma correction on the divided plurality of frame images, and outputs a dangerous object detection result in the X-ray image when the gamma-converted images are input. , An electronic device may be provided that obtains an object detection result from the artificial intelligence model by inputting the gamma-converted images, generates visual content by visualizing the object detection result, and outputs the generated visual content. .

According to another embodiment for achieving the above-described technical problem, a method for detecting a dangerous object by an electronic device includes obtaining an X-ray image including at least one object to be inspected; Dividing the obtained X-ray image into a plurality of frame images having a preset frame interval; acquiring gamma-converted images by performing gamma correction on the divided plurality of frame images; obtaining an object detection result from the artificial intelligence model by inputting the gamma-converted images to an artificial intelligence model that outputs a dangerous object detection result in the X-ray image when the gamma-converted images are input; generating visual content by visualizing the object detection result; and outputting the generated visual content. A computer-readable recording medium recording a program for executing the method on a computer, including, may be provided.

According to an embodiment, a dangerous object may be accurately detected on an X-ray image.

According to an embodiment, even when objects overlap or overlap, a dangerous object can be effectively detected based on an artificial intelligence model.

1 is a diagram schematically illustrating a process of detecting a dangerous object by an electronic device according to an exemplary embodiment.
2 is a flowchart of a method of detecting a dangerous object by an electronic device according to an exemplary embodiment.
3 is a diagram for explaining a process of learning an artificial intelligence model by an electronic device according to an embodiment.
4 is a diagram for explaining structures of a Faster R-CNN model and a Cascade R-CNN model according to an embodiment.
5 is a diagram for explaining the structure of Res2Net according to an embodiment.
6 is a diagram for explaining a process of performing gamma correction by an electronic device according to an exemplary embodiment.
7 is a diagram illustrating examples of X-ray images of dangerous objects used by an electronic device according to an exemplary embodiment.
8 is a diagram for explaining a comparison between a loss value of an artificial intelligence model used by an electronic device and performance of each model according to an exemplary embodiment.
9 is a block diagram of an electronic device according to an embodiment.
10 is a flowchart of a method of detecting a dangerous object by an electronic device according to another embodiment.

Terms used in this specification will be briefly described, and the present disclosure will be described in detail.

The terms used in the present disclosure have been selected from general terms that are currently widely used as much as possible while considering the functions in the present disclosure, but they may vary according to the intention or precedent of a person skilled in the art, the emergence of new technologies, and the like. In addition, in a specific case, there is also a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the invention. Therefore, terms used in the present disclosure should be defined based on the meaning of the term and the general content of the present disclosure, not simply the name of the term.

When it is said that a certain part "includes" a certain component throughout the specification, it means that it may further include other components without excluding other components unless otherwise stated. In addition, terms such as "...unit" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software. .

Hereinafter, with reference to the accompanying drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily carry out the present disclosure. However, the present disclosure may be implemented in many different forms and is not limited to the embodiments described herein. And in order to clearly describe the present disclosure in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

1 is a diagram schematically illustrating a process of detecting a dangerous object by an electronic device according to an exemplary embodiment.

In general, since x-rays have high transmittance, all objects in an x-ray image can be imaged through perspective. However, since x-rays have different densities when different objects overlap each other, light absorbing power may be different, and objects with low density may not transmit light. Conversely, if the density of the object is too high, the image may not be easily revealed, making it difficult to search for baggage.

In addition, the size of various types of dangerous objects that appear in security screening varies greatly depending on the type of dangerous object. In the case of conventional security search technology, there is a limitation in that it is difficult to accurately recognize it.

In addition, in the case of the conventional security search technology, the background of the X-ray image is complicated and there is a limit to finding an area with a dangerous object, and due to overlapping or overlapping of dangerous objects in the image, objects made of the same material have the same color, making it difficult to recognize , there was a limitation that it was difficult to observe the object when objects of the same density overlapped.

However, the electronic device 1000 according to the present disclosure may accurately detect an object using a Cascade R-CNN model to solve object recognition limitations due to object density or object overlap, and improve the accuracy of the artificial intelligence model. In order to improve, the data labeling problem can be solved by mixing CP algorithm, supervised learning and semi-supervised learning.

The electronic device 1000 according to an embodiment obtains an X-ray image obtained using X-rays or an X-ray image 102 including X-ray images at a plurality of frame intervals, and obtains a predetermined object from the obtained X-ray image or X-ray image. can be detected. For example, the electronic device 1000 includes a memory 120 that stores one or more instructions and at least one processor 140 that executes the one or more instructions, and an artificial intelligence model is stored in the memory 120. It can be. The electronic device 1000 according to the present disclosure may detect a dangerous object appearing in the X-ray image or the X-ray image by analyzing the X-ray image or the X-ray image using the artificial intelligence model 122 .

According to an embodiment, the electronic device 1000 may detect a dangerous object such as a dangerous substance, a dangerous object, a storage medium, a gun, or a knife from an X-ray image or an X-ray image. According to an embodiment, the electronic device 1000 may be used to maintain the safety of an airport or target building by being located at an airport or building search counter that inspects baggage using X-rays and detects a dangerous object.

An artificial intelligence model used by the electronic device 1000 according to the present disclosure may be a Cascade R-CNN model. The electronic device 1000 can detect a dangerous object from an X-ray image using a cascade R-CNN model, and cp (copy -paste) algorithm can be used to increase training data samples, and the data labeling problem can be solved by using a mixture of supervised learning and semi-supervised learning.

According to an embodiment, the electronic device 1000 may include an X-ray scanning device, but an X-ray image generated by generating X-rays for baggage from the X-ray scanning device connected to the electronic device 1000 and absorbing the beam with a detector. (102) Alternatively, an X-ray image may be acquired. The electronic device 1000 may classify whether an object in the X-ray image is organic or inorganic by analyzing the X-ray image or the X-ray image using the artificial intelligence model 122 .

According to an embodiment, the electronic device 1000 may generate visual content 142 by analyzing the X-ray image 102 using an artificial intelligence model. According to an embodiment, the electronic device 1000 may provide the generated visual content 1420 to a user through a display or other output unit, and may transmit information about the visual content to an external device connected to the electronic device. Also, according to an embodiment, the electronic device 1000 may provide a detection result 144, which is a result of analyzing an X-ray image using an artificial intelligence model 122, along with providing visual content 142. . According to an embodiment, the detection result 144 may include the type of dangerous object detected from the X-ray image, the location of the dangerous object, and location information of a bounding box around the dangerous object.

According to an embodiment, the electronic device 1000 may display organic/inorganic materials having a low density in a faint and light manner, and display organic/inorganic materials having a high density in a dark color. Also, according to an embodiment, the electronic device 1000 may generate a color image by analyzing an X-ray image. The electronic device 1000 can display organic materials in orange, a mixture of organic and inorganic materials in green, and inorganic materials in blue in a color image, and can display objects that cannot transmit X-rays due to their thickness and density in black. can However, it is not limited to the above example, and the electronic device 1000 may display objects in different ways based on at least one of density of the object or whether the object is an organic/inorganic material using an artificial intelligence model. .

2 is a flowchart of a method of detecting a dangerous object by an electronic device according to an exemplary embodiment.

In S210, the electronic device 1000 may obtain an X-ray image including at least one object to be inspected. For example, the electronic device 1000 may obtain an X-ray image from an X-ray scanning device connected to the electronic device, or may obtain an X-ray image using an X-ray scanning device included in the electronic device 1000. An X-ray image obtained by the electronic device 1000 may include a plurality of X-ray images.

In S220, the electronic device 1000 may divide the obtained X-ray image into a plurality of frame images having a preset frame interval. In S230, the electronic device 1000 may obtain gamma-converted images by performing gamma correction on the divided plurality of frame images. The electronic device 1000 according to the present disclosure provides gamma-correction (Gamma-enhanced Data Enhancement) method can be used to improve the color and sharpness of the part where objects overlap in the X-ray image.

According to an embodiment, the electronic device 1000 may improve color sharpness by correcting an X-ray image with a gray level that is too high or too low, and color contrast between bright and dark parts in the X-ray image by using a nonlinear processing method. can increase

In S240, when the gamma-converted images are input, the electronic device 1000 may obtain an object detection result from the artificial intelligence model by inputting the gamma-converted images to the artificial intelligence model that outputs the dangerous object detection result in the X-ray image.

In S250, the electronic device 1000 may generate visual content by visualizing the object detection result output from the artificial intelligence model. In S260, the electronic device 1000 may output the generated visual content.

3 is a diagram for explaining a process of learning an artificial intelligence model by an electronic device according to an embodiment.

In S310, the electronic device 1000 may obtain training X-ray images that may be identified as a predetermined dangerous object prior to acquiring the X-ray images. For example, the electronic device 1000 may obtain a learning X-ray image including a predetermined dangerous object. Also, the electronic device 1000 may obtain a learning X-ray image that does not include a predetermined dangerous object.

In S320, the electronic device 1000 may train an artificial intelligence model based on the learning X-ray images. For example, the electronic device 1000 may train an artificial intelligence model based on training data including all training X-ray images including a dangerous object or training X-ray images not including a predetermined dangerous object. According to an embodiment, the training data used by the electronic device 1000 to train the artificial intelligence model may be an image or video-type data including a plurality of images.

When the artificial intelligence model is learned, the electronic device 1000 may obtain a dangerous object detection result in the X-ray image from the artificial intelligence model by inputting gamma-converted images according to gamma conversion into the learned artificial intelligence model.

4 is a diagram for explaining structures of a Faster R-CNN model and a Cascade R-CNN model according to an embodiment.

Figures 410 and 420 of FIG. 4 show the structures of the Faster R-CNN model and the Cascade R-CNN model. The Cascade R-CNN model used by the electronic device 1000 according to the present disclosure is very similar to the Faster R-CNN model and can largely perform two-step operations. For example, when X-ray images or gamma-correction converted images are input to the Cascade R-CNN model, a feature map for the X-ray image may be obtained through feature extraction using a feature extraction network inside the Cascade R-CNN model.

A Region Proposal Network (RPN) can generate corresponding samples by finding candidate targets in a feature map and concatenating different Intersection over Union (IOU) thresholds. The IOU threshold according to the present disclosure is an index for evaluating the accuracy of object detection, and generally, the electronic device 1000 can determine whether object detection is successful or not based on the IOU threshold having a range of 0 to 1. .

According to an embodiment, the Cascade R-CNN model used by the electronic device 1000 may use multiple searchers, but Faster R-CNN uses one searcher. Each searcher in the Cascade R-CNN model can be learned based on positive and negative samples of different IoU thresholds, and the output value of the previous searcher can be used as the value of the next searcher to be input.

Each of the plurality of searchers is learned based on positive and negative samples having different Intersection over Union (IOU) thresholds, and the plurality of searchers detects the artificial intelligence model by gradually increasing the different IOU thresholds. It is possible to improve the accuracy of the result and suppress the overfitting problem of the artificial intelligence model.

In Figure 410, pool is a pooling layer, FC1 is a fully connected layer, BO is a bounding box of a candidate region, B1 is a predicted bounding box, and C1 is the final classification predicted value. In Figure 420, pool is a pooling layer for feature maps, FC1, FC2, and FC3 are fully connected layers, and B0, B1, and B2 may represent bounding boxes of candidate regions. B3 represents a predicted bounding box in the structure, C1 and C2 represent predicted classification results, and C3 may represent a final classification prediction value.

Cascade R-CNN uses the cascade method, so by using the output value of the previous classifier as the input value of the next classifier, it can provide better data to the next classifier, resulting in higher efficiency.

According to an embodiment, Cascade R-CNN uses three cascade detectors in the classification and regression stages and can continuously optimize candidate frames and make detection results more accurate by gradually increasing the IoU threshold. Through this, the electronic device 1000 according to the present disclosure can effectively prevent overfitting and false positives caused by a fixed IoU threshold that is too high or too low.

According to an embodiment, an artificial intelligence model used by an electronic device may include a Region Proposal Network (RPN), each RPN model uses a plurality of searchers, and a previous searcher among the plurality of searchers is the above search engine within the plurality of searchers. It can be prepared as an input value of the next searcher connected to the previous searcher.

According to an embodiment, since the ratio between the high resolution of the X-ray image and the feature part is very small in many cases, the detector of the first stage is a high-density detector, and the detector of the second stage is a large number of background regions. It is used because it is difficult to train the model as it is generated. In the second step, detection can be performed by filtering out many irrelevant background regions using a Region Proposal Network (RPN). Using a fixed IoU threshold for detection in RPN may not be helpful in detecting dangerous objects considering the many false positives that can occur in a complex background.

Therefore, the electronic device 1000 according to the present disclosure has the advantage of using the Cascade R-CNN model, filtering irrelevant background areas using the Cascade prediction method, and obtaining better detection position coordinates. . In addition, the Cascade R-CNN model according to the present disclosure not only gradually improves the accuracy of the final output of the artificial intelligence model by using the more accurate candidate region output of the previous detector as the input of the next detector, but also solves the overfitting problem. It also has the advantage of being able to effectively suppress it.

5 is a diagram for explaining the structure of Res2Net according to an embodiment.

Referring to Figure 510, the Res2Net structure is shown. The artificial intelligence model used by the electronic device 1000 according to the present disclosure is a backbone network based on the structure shown in FIG. 5 and may include a feature extraction network. features can be extracted.

The Res2Net network is a new type of target detection network jointly proposed in 2019 by researchers from Nankai University and several well-known external universities. The Res2Net network module can be well integrated with other network models with good performance. Res2Net uses a finer convolution kernel to extract finer multi-scale features and can extend the receive field range of each network layer, and uses multiple smaller convolution kernels to replace the traditional 3*3 convolution kernel. can be used

A small convolutional kernel can reduce feature loss because it increases output features connected to each other like a hierarchical residual network, and the basic structure of Res2Net is shown in Figure 510.

According to an embodiment, Res2Net uses a connection method by dividing a 3*3 convolution from the residual block of ResNet into several feature subsets, and combining features extracted from multiple feature subsets can enrich image features, The ability to extract scale features can be improved. Therefore, there is an advantage in that detection performance can be improved without increasing the computational load, so the electronic device 1000 according to the present disclosure can use the Res2Net-101 model to extract the features of a dangerous object from an X-ray image. It is not limited.

6 is a diagram for explaining a process of performing gamma correction by an electronic device according to an exemplary embodiment.

The electronic device 1000 according to the present disclosure provides gamma-correction (Gamma-enhanced Data Enhancement) method can be used to improve the color and sharpness of the part where objects overlap in the X-ray image.

For example, in S610, the electronic device 1000 may convert pixel values within a plurality of frame images into gray scale values above or below a predetermined threshold by using a predetermined gamma correction function. A gamma correction function used by the electronic device 1000 is expressed in Equation 1 below.

The gamma correction function includes, as variables, a pixel gradation value within the frame images, a maximum gray scale value within the frame images, and a parameter determining a degree of change in pixel gradation within the frame images as variables, and It may be a nonlinear processing function that includes the gray scale value of as an output value.

In Equation 1, a may be a pixel gradation value within the input frame images, G may be a gray scale value of a pixel within the frame images, and amax may be a maximum gray scale value of a pixel within the frame images. In addition, according to an embodiment, the gamma correction function may further include a user parameter for determining the degree of gradation change in an image by gamma conversion, and according to an embodiment, it may be set to 0.3, but is not limited thereto. .

In S620, the electronic device 1000 may obtain a plurality of frame images including pixels converted to a predetermined gray scale value as gamma converted images by using the gamma correction function according to Equation 1 above.

7 is a diagram illustrating examples of X-ray images of dangerous objects used by an electronic device according to an exemplary embodiment.

Referring to a drawing 720 of FIG. 7 , examples of dangerous objects that the electronic device 1000 detects through an X-ray image are shown. According to an embodiment, the electronic device 1000 analyzes an X-ray image using an artificial intelligence model, such as a knife, scissors, sharp tools, small pieces of glass, electrical equipment, batteries, umbrellas, plastic beverage bottles, and plastics with nozzles. A bottle, a rod that can be extended in length (eg, a self-defense rod), etc. can be detected as a dangerous object.

8 is a diagram for explaining a comparison between a loss value of an artificial intelligence model used by an electronic device and performance of each model according to an exemplary embodiment.

The electronic device 1000 according to the present disclosure may construct an artificial intelligence model for detecting dangerous substances using the Cascade R-CNN model, and a loss value during training of the artificial intelligence model is as shown in Figure 810. It can be seen that the electronic device 1000 performs training of the artificial intelligence model about 160,000 times, and as the loss value reaches 0.4511 in the first stage, the loss value does not change significantly and the model shows a tendency to stabilize.

Also, referring to figure 820, a table comparing the Faster R-CNN model and the Cascade R-CNN model is shown. Cascade R-CNN's AP (IoU=0.5) is 4.2% higher than Faster R-CNN's, AP (IoU=0.75) is 2.7%, and Recall is 6.8% higher. Accordingly, it can be seen that Cascade R-CNN used by the electronic device 1000 according to the present disclosure exhibits better performance than Faster R-CNN.

9 is a block diagram of an electronic device according to an embodiment.

An electronic device 1000 according to an embodiment may include a display 1200, a processor 1300, a network interface 1500, and a memory 1700. However, it is not limited to the above example, and it goes without saying that the electronic device 1000 may include more components required to perform the method of detecting a dangerous object. According to another embodiment, the electronic device 1000 may include fewer components than those shown in FIG. 9 .

The processor 1300 typically controls overall operations of the electronic device 1000 . For example, the processor 1300 may generally control the network interface 1500 and the display 1200 by executing programs stored in the memory 1700 .

According to an embodiment, the processor 1300 obtains an X-ray image including at least one object to be inspected by performing one or more instructions, and converts the obtained X-ray image into a plurality of frames having a preset frame interval. An artificial intelligence model that acquires gamma-converted images by dividing into images and performing gamma correction on the divided plurality of frame images, and outputs a dangerous object detection result in the X-ray image when the gamma-converted images are input. , Obtain an object detection result from the artificial intelligence model by inputting the gamma-converted images, generate visual content by visualizing the object detection result, and output the generated visual content.

In addition, according to an embodiment, the processor 1300 acquires learning X-ray images that can be identified as a predetermined dangerous object prior to the acquiring of the X-ray images, and the artificial intelligence based on the acquired learning X-ray images. A dangerous object detection result in the X-ray image may be acquired by learning an intelligence model and inputting the gamma-converted images to the learned artificial intelligence model.

Also, according to an embodiment, the processor 1300 converts pixel values in the plurality of frame images into gray scale values greater than or equal to a predetermined threshold value or less than or equal to a predetermined threshold value by using a predetermined gamma correction function, and A plurality of frame images including pixels converted to scale values may be obtained as the gamma-converted images.

Also, according to an embodiment, the processor 1300 may perform at least a part of the method of detecting a dangerous object by the electronic device 1000 described above with reference to FIGS. 1 to 8 .

The network interface 1500 may acquire at least one X-ray image and an X-ray image including the X-ray images from the X-ray scanning device under the control of a processor. Also, the network interface 1500 may further obtain learning X-ray images and data for learning an artificial intelligence model from a server connected to the electronic device. Also, the network interface 1500 may transmit visualization content generated by the electronic device and information on detection results to an external device connected to the electronic device 1000 .

The memory 1700 may store programs for processing and control of the processor 1300 and may store data input to or output from the electronic device 1000 . Also, the memory 1700 may store information about an artificial intelligence model used by the electronic device 1000 . According to an embodiment, the artificial intelligence model stored in the memory 1700 is a neural network model, and may further store weight information about at least one layer and connection strength of the layers.

According to an embodiment, the memory 1700 is a flash memory type, a hard disk type, a multimedia card micro type, or a card type memory (eg SD or XD memory, etc.), RAM (RAM, Random Access Memory) SRAM (Static Random Access Memory), ROM (ROM, Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory) , a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium.

The display 1200 may output visualized content generated by the electronic device 1000 or information on detection results on a screen.

10 is a flowchart of a method of detecting a dangerous object by an electronic device according to another embodiment.

According to an embodiment, the electronic device 1000 may include a detection image enhancement module 1002 , a detection module 1004 , and a visualization module 1006 . According to an embodiment, the detected image enhancement module 1002, the detection module 1004, and the visualization module 1006 mean a set of instructions stored in a memory in the electronic device 1000 and executable under the control of a processor. can

For example, in S1002, the electronic device 1000 acquires a real-time X-ray image. In S1004, according to an embodiment, the detection image enhancement module 1002 divides the images into frames. In S1006, the detection image enhancement module 1002 improves the quality of the image by performing gamma correction on the image. In S1008, the detection image enhancement module generates a detection image (eg, a gamma conversion image).

In S1010, the detection module 1004 inputs the detection image obtained from the detection image enhancement module 1002 to the Cascade R-CNN model. In operation S1012, the detection module 1012 may identify whether a dangerous object is detected based on the output value of the Cascade R-CNN model. In operation S1014, when the dangerous object is not detected, the detection module 1012 generates dangerous object type information and transfers the generated dangerous object type information to the visualization module 1006. In S1016, the visualization module 1006 may generate a warning message based on the dangerous object type information. In S1018, if it is identified that the dangerous object is detected based on the output value of the Cascade R-CNN model, the visualization module 1006 may output the visualization content generated by visualizing the dangerous object together with the warning message generated in S10106. there is.

The method according to an embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the present disclosure, or may be known and usable to those skilled in computer software.

In addition, a computer program device including a recording medium in which a program for performing a different method according to the above embodiment is stored may be provided. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler.

Although the embodiments of the present disclosure have been described in detail above, the scope of the present disclosure is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present disclosure defined in the following are also the scope of the present disclosure. belongs to

Claims (15)

A method for detecting a dangerous object by an electronic device,
obtaining an X-ray image including at least one object to be inspected;
Dividing the obtained X-ray image into a plurality of frame images having a preset frame interval;
acquiring gamma-converted images by performing gamma correction on the divided plurality of frame images;
obtaining an object detection result from the artificial intelligence model by inputting the gamma-converted images to an artificial intelligence model that outputs a dangerous object detection result in the X-ray image when the gamma-converted images are input;
generating visual content by visualizing the object detection result; and
outputting the generated visual content; Including, method. The method of claim 1, wherein the method
Acquiring learning X-ray images that can be identified as a predetermined dangerous object before acquiring the X-ray image; and
learning the artificial intelligence model based on the obtained learning x-ray images; Including more,
Obtaining the object detection result
obtaining a dangerous object detection result in the X-ray image by inputting the gamma-converted images to the learned artificial intelligence model; Including, method. 3. The method of claim 2, wherein obtaining the gamma converted images comprises:
converting pixel values in the plurality of frame images into gray scale values greater than or equal to a predetermined threshold by using a predetermined gamma correction function; and
acquiring a plurality of frame images including pixels converted to gray scale values as the gamma converted images; Including, method. 4. The method of claim 3, wherein the gamma correction function is
A pixel gradation value in the frame images, a maximum gray scale value in the frame images, and a parameter determining a degree of change in pixel gradation in the frame images as variables, and a gray scale value of a pixel in the frame images Characterized in that a non-linear processing function, including as an output value, method. The method of claim 3, wherein the artificial intelligence model
A method characterized in that it is a cascade R-CNN model and includes a Res2Net-101 model as a backbone network for extracting features from the gamma-converted images. The method of claim 5, wherein the artificial intelligence model
It includes a Region Proposal Network (RPN) model, the RPN model uses a plurality of searchers, and a previous searcher among the plurality of searchers is connected to the previous searcher in the plurality of searchers and is provided as an input value of the next searcher Characterized in that, the method. The method of claim 6, wherein each of the plurality of searchers is learned based on positive and negative samples having different intersection over union (IOU) thresholds, and the plurality of searchers gradually increase the different IOU thresholds. Characterized in that, the accuracy of the detection result of the artificial intelligence model is improved, and the overfitting problem of the artificial intelligence model is suppressed. 8. The method of claim 7, wherein the method
transmitting the generated visualization content to an external device connected to the electronic device; Including, method. In an electronic device for detecting a dangerous object,
display;
network interface;
a memory that stores one or more instructions; and
at least one processor to execute the one or more instructions; including,
By executing the one or more instructions, the at least one processor:
Obtaining an X-ray image including at least one object to be inspected;
Dividing the obtained X-ray image into a plurality of frame images having a preset frame interval;
Obtaining gamma-converted images by performing gamma correction on the divided plurality of frame images;
When the gamma-converted images are input, an object detection result is obtained from the artificial intelligence model by inputting the gamma-converted images to an artificial intelligence model that outputs a dangerous object detection result in the X-ray image;
visual content is generated by visualizing the object detection result;
An electronic device that outputs the generated visual content. 10. The method of claim 9, wherein the at least one processor
Acquiring learning X-ray images that can be identified as a predetermined dangerous object prior to the acquiring of the X-ray image,
Learning the artificial intelligence model based on the acquired learning X-ray images;
An electronic device that obtains a dangerous object detection result in the X-ray image by inputting the gamma-converted images to the learned artificial intelligence model. 11. The method of claim 10, wherein the at least one processor
converting pixel values in the plurality of frame images into gray scale values above or below a predetermined threshold by using a predetermined gamma correction function;
Acquiring a plurality of frame images including pixels converted to gray scale values as the gamma converted images. 12. The method of claim 11, wherein the gamma correction function is
A pixel gradation value in the frame images, a maximum gray scale value in the frame images, and a parameter determining a degree of change in pixel gradation in the frame images as variables, and a gray scale value of a pixel in the frame images An electronic device, characterized in that it is a non-linear processing function included as an output value. The method of claim 11, wherein the artificial intelligence model
An electronic device characterized in that it is a cascade R-CNN model and includes a Res2Net-101 model as a backbone network for extracting features from the gamma-converted images. The method of claim 13, wherein the artificial intelligence model
It includes a Region Proposal Network (RPN) model, the RPN model uses a plurality of searchers, and a previous searcher among the plurality of searchers is connected to the previous searcher in the plurality of searchers and is provided as an input value of the next searcher Characterized in that, the electronic device. A method for detecting a dangerous object by an electronic device,
obtaining an X-ray image including at least one object to be inspected;
Dividing the obtained X-ray image into a plurality of frame images having a preset frame interval;
acquiring gamma-converted images by performing gamma correction on the divided plurality of frame images;
obtaining an object detection result from the artificial intelligence model by inputting the gamma-converted images to an artificial intelligence model that outputs a dangerous object detection result in the X-ray image when the gamma-converted images are input;
generating visual content by visualizing the object detection result; and
outputting the generated visual content; A computer-readable recording medium recording a program for executing a method on a computer, including a.

Images