WO2020134848A1 - Intelligent detection method and device applied to millimeter wave security check instrument, and storage device - Google Patents

Intelligent detection method and device applied to millimeter wave security check instrument, and storage device Download PDF

Info

Publication number
WO2020134848A1
WO2020134848A1 PCT/CN2019/121764 CN2019121764W WO2020134848A1 WO 2020134848 A1 WO2020134848 A1 WO 2020134848A1 CN 2019121764 W CN2019121764 W CN 2019121764W WO 2020134848 A1 WO2020134848 A1 WO 2020134848A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
area
human body
images
recognition
Prior art date
Application number
PCT/CN2019/121764
Other languages
French (fr)
Chinese (zh)
Inventor
冯智辉
祁春超
Original Assignee
深圳市华讯方舟太赫兹科技有限公司
华讯方舟科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市华讯方舟太赫兹科技有限公司, 华讯方舟科技有限公司 filed Critical 深圳市华讯方舟太赫兹科技有限公司
Publication of WO2020134848A1 publication Critical patent/WO2020134848A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V8/00—Prospecting or detecting by optical means

Definitions

  • This application relates to the field of intelligent security inspection, in particular to an intelligent detection method, device and storage device applied to a millimeter wave security inspection instrument.
  • Millimeter wave imaging technology will use millimeter wave radar.
  • Millimeter-wave radar means that the frequency of the radar emission signal is in the millimeter-wave band, and the frequency of the millimeter wave is 30GHz to 300GHz (wavelength from 1mm to 10mm). In practical engineering applications, the low-end frequency of the millimeter wave is often reduced to 26GHz.
  • the position of the millimeter wave frequency is between microwave and infrared. Compared with microwaves, the typical characteristics of millimeter waves are short wavelength, frequency bandwidth (with a very wide utilization space), and propagation characteristics in the atmosphere. Compared with infrared, millimeter wave has the ability to work around the clock and can be used in harsh environments such as smoke, clouds and fog. In the increasingly crowded microwave frequency band, millimeter wave takes into account the advantages of microwave, and also has some advantages that low-frequency microwave does not.
  • a security inspection device scans an object to be inspected, it is generally imaged on a computer, and then the staff observes and inspects the imaged image to confirm whether it carries dangerous goods, or can only carry out metal and other substances. Alarm, this seems not smart enough, and due to human eyes fatigue problems, its detection efficiency is low, and the detection rate is not high.
  • the present application provides an intelligent detection method, device, and storage device applied to a millimeter wave security detector.
  • the security inspection efficiency is low and the accuracy rate is relatively general.
  • a technical solution adopted by the present application is: to provide an intelligent detection method applied to a millimeter-wave safety inspection instrument, which includes: inputting multiple human body images into a preset dangerous goods detection model for dangerous goods Recognition to obtain multiple recognition images carrying the detection results; input the frontal image of the human body in the multiple human body images to a preset body detection model to perform image segmentation into multiple regions to obtain area identification images; The recognition image is merged with the area identification image to obtain an area recognition image; and the recognition result is output to the area recognition image according to the detection result of the area and the corresponding preset threshold.
  • an intelligent detection device applied to a millimeter wave security inspection instrument characterized in that the intelligent detection device includes a processor coupled to the processor Memory,
  • the memory stores program instructions for implementing the display method described in any one of the above; the processor is used to input multiple human body images into a preset dangerous goods detection model for dangerous goods recognition to obtain multiple pieces of carry detection The resulting recognition image; input the frontal image of the human body into the preset body detection model to perform image segmentation into multiple regions to obtain an area identification image; to identify the multiple images and the area identification The image is fused to obtain an area recognition image; the area recognition image is outputted according to the detection result of the area and the corresponding preset threshold.
  • Another technical solution adopted by the present application is to provide a storage device that stores a program file capable of implementing any of the above methods.
  • this application provides an intelligent detection method, device and storage device applied to a millimeter wave security inspection instrument, which is obtained by inputting multiple human body images into a preset dangerous goods detection model for recognition Recognize the image and input the frontal image of the human body into the preset body detection model to segment the area identification image, and fuse the identification image carrying the detection result with its area identification image to obtain the area identification image to make its area
  • the recognition image carries the detection result, and then the recognition result is output according to the detection result of each region of the region recognition image and the corresponding preset threshold.
  • the image can be directly recognized through deep learning instead of manual methods, which greatly reduces the cost and improves the detection efficiency.
  • by dividing its image into regions and outputting results according to different thresholds of each region it makes its identification more accurate, which reduces the false alarm rate while increasing the detection rate.
  • FIG. 1 is a schematic structural diagram of a first embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application;
  • FIG. 2 is a schematic diagram of a sub-flow of S12 of the first embodiment of the present application.
  • FIG. 3 is a schematic structural view of a second embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application
  • FIG. 4 is a schematic structural block diagram of an embodiment of an intelligent detection device applied to a millimeter wave security inspection instrument of the present application
  • FIG. 5 is a schematic structural diagram of an embodiment of a storage device of the present application.
  • FIG. 1 is a schematic flowchart of a first embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application, which specifically includes the following steps:
  • multiple human body images are input into a preset dangerous goods detection model for detection.
  • the multiple human body images are specifically a terahertz scan image that can be acquired by the security inspection instrument at multiple angles to the inspected object.
  • the security device may specifically be a terahertz cylindrical security device, which can scan the subject in all directions and multiple angles to obtain imaging pictures of the subject at different angles, and use these imaging pictures as its purpose Human images for recognition.
  • the preset dangerous goods detection model is obtained by pre-training, which can be achieved by taking a large number of images of dangerous goods carried by the human body as sample data and using VGGNet as the feature extraction network for optimal parameters. Training.
  • the human body may include people of different genders, such as men and women, people of different clothes, such as winter clothes, summer clothes, spring and autumn clothes, etc., or directly differentiated by thickness, such as short sleeves, sweaters, down jackets, etc. , People with different BMI (Body Mass Index) values, such as different people with BMI less than 18.5, BMI between 18.5 and 24, and BMI greater than 24.
  • BMI Body Mass Index
  • Dangerous goods include some common dangerous goods, which can be such as pistols, metal knives, ceramic knives, rectangular powder explosives, dish-shaped powder explosives, irregular powder explosives, liquid explosives, lighters and others, etc. Some easy to carry and common dangerous goods.
  • the sample database is obtained.
  • the multi-angle can take the human head to footsteps as the central axis, and collect 360 degrees around the axis.
  • a multi-angle image of a large number of different humans carrying at least one kind of dangerous goods is collected as a sample database by a cylindrical security inspection instrument, and is used for input to VGGNet for optimal parameter training, thereby obtaining its preset dangerous goods detection model.
  • VGGNet explored the relationship between the depth of the convolutional neural network and its performance, and successfully constructed a 16-19 layer deep convolutional neural network, proving that increasing the depth of the network can affect the final performance of the network to a certain extent, making errors The rate has dropped significantly, while being very expandable, and the generalization of migration to other image data is also very good.
  • Each input human body image will correspondingly obtain a recognition image carrying the detection result, therefore, after all recognition is completed, multiple recognition images carrying the detection result will be obtained.
  • the identification image carrying the detection result may mean that the identification image carries the detection result integrally, or the identification image and the detection result are separate, but they are related to each other by mapping.
  • S12 Input a frontal image of the human body in the plurality of human body images to a preset body detection model to perform image segmentation into multiple regions to obtain an area identification image.
  • the multiple human body images include the front image of the human body. Because the front image has a better angle, it is conducive to improving the recognition rate and further , The frontal image of the human body is input into a preset body detection model to perform image segmentation, so as to identify different areas thereof, thereby obtaining an area identification image.
  • its preset body detection model is also obtained by using a large number of images of humans carrying dangerous goods as a sample database, using the ZFNet network as a feature extraction network, and then inputting its sample database to its ZFNet network for optimal parameter training. .
  • the front recognition image After inputting its front recognition image into a preset body detection model for recognition, the front recognition image can be segmented according to the region of the human body to obtain different regions and identify the different regions.
  • the first area, the second area, the third area, and the fourth area can be obtained after the division. Specifically, it is divided according to the body area of the subject.
  • the first area is the head area of the subject in the front recognition image, specifically including the head and neck area
  • the second area is the front area in the front recognition image.
  • the third area is the hand area of the subject in the front recognition image, which includes the left arm area and the right arm area
  • the fourth The area is the leg area of the subject in the front recognition image, and also includes the left leg area and the right leg area.
  • FIG. 2 is the sub-step of S12 in the step 1 of FIG. 1 of the intelligent detection method of the present application applied to the millimeter wave security inspection instrument, which specifically includes the following steps:
  • the body detection model After inputting into the body detection model, first recognize its front recognition image to see if it can recognize its second area, that is, the area from the shoulder to the crotch position, so as to judge the recognizability of the front recognition image, specifically If the second area cannot be recognized, it proves that the positive recognition image may be blurry, or the angle is not positive, which may bring greater ambiguity to the specific recognition process, resulting in a higher dangerous goods recognition rate. Low, the recognition effect is not good.
  • the other areas that is, the first area, the third area, and the fourth area are compensated.
  • the other areas are offset, for example, the third area is offset, you can use the second area
  • the area obtains the body midline, and then calculates its offset area, such as the distance of the third area from its body centerline, and then finds its average value, and translates its third area according to its average value.
  • the recognition of the second area fails, it may prove that the front recognition image is not a front image, so its effect may be poor, and it is considered that it needs to be re-recognized, and the staff is prompted to recognize the image recognition abnormality and re-acquire the image.
  • multiple identification images are mapped into the area identification image, so that the multiple identification images are merged with the area identification image, that is, the detection result carried by the identification image is fused into the area identification image, thereby obtaining the area An identification image, wherein the area identification image carries the detection result.
  • its area identification image also completes the identification of each area.
  • S14 Output a recognition result to the area recognition image according to the detection result of the area and the corresponding preset threshold.
  • the area where the human body carries dangerous goods may be mainly concentrated in its chest, span, and other areas. If a threshold is used for the entire body, the false alarm rate may be increased and the detection result may be affected. Therefore, for the entire human body, its Different preset thresholds need to be set for different areas of the human body, and specifically, different preset thresholds are set for different dangerous goods and different areas, such as pistols and other dangerous goods, generally more cross-sections are placed, Correspondingly, the chest can set a lower threshold to improve the recognition accuracy.
  • the threshold can be appropriately increased to prevent false alarms.
  • multiple human body images are recognized through deep learning to improve the accuracy of recognition and obtain more accurate detection results.
  • deep learning processes the frontal image of the human body, based on the human body’s
  • the results are divided into regions, and then the results of the two recognition and segmentation are fused to obtain the detection results of each region on the front image of the human body, and then compared with the preset thresholds of the corresponding regions according to the detection results and further improved Its accuracy rate, thereby reducing the false alarm rate.
  • FIG. 3 is a schematic flowchart of a second embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application, which specifically includes the following steps:
  • ⁇ is the image mean
  • X represents the image matrix
  • ⁇ represents the standard deviation
  • N represents the number of image pixels
  • Per_image_standardizatio is the averaged result.
  • x i represents the pixel value of the image
  • min(x) represent the minimum and maximum value of the image pixel
  • norm represents its normalized result
  • the size of the dangerous goods may be different, such as small, such as a knife, such as a large, such as a pistol, so in the specific training and recognition process, the use of images of the same size may not be able to accurately identify, specific, You can optimize the image based on the components of the preset area and preset ratio to form multiple combined images. Specifically, for a 16*16pt image, you can use 3 preset areas and 3 presets The combination of proportions forms a total of 9 images.
  • the three preset areas may be 8 times, 16 times, or 32 times, and the three preset ratios may be 1:2, 1:1, 2:1, and so on. If the combination of 8 times area and 1:1 ratio is adopted, the 16*16pt image can be formed into 128*128pt image.
  • the focus model is used to optimize the training of the training model.
  • the false alarm rate can be reduced while maintaining a high detection efficiency and detection level. It uses the following loss function for processing.
  • the samples in the sample database that is, the training model needs to be optimized using the focus loss function to obtain the difference between the predicted data and the actual data, so that the loss of the training model must be converged, so that The training result is more accurate, and the redundancy of data is reduced.
  • the body detection model can also perform this process on the human body image during the recognition process.
  • the training process of the body detection model can also be processed similarly, which is not limited here.
  • FIG. 4 is a schematic structural block diagram of an embodiment of an intelligent detection device applied to a millimeter wave security detector provided by this application.
  • the intelligent detection device applied to the millimeter wave security detector provided in this embodiment specifically includes a processor 10 and a memory 11 coupled to the processor.
  • the processor 10 may be a CPU (Central Processing Unit, central processing unit). Or GPU (Graphics Processing Unit), the processor 10 may be an integrated circuit chip with signal processing capabilities.
  • the processor 10 may also be a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components .
  • the general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the processor 10 may input multiple human body images into a preset dangerous goods detection model for dangerous goods recognition to obtain multiple recognition images carrying the detection results; the human body front image among the multiple human body images Input to a preset body detection model to perform image segmentation into multiple areas to obtain an area identification image; perform fusion processing on the multiple identification images and the area identification image to obtain an area identification image; based on the area identification image
  • the detection result of the region and the corresponding preset threshold output the recognition result.
  • the memory 11 stores an instruction file 111 that can implement any of the above embodiments.
  • the other module terminals of the above-mentioned equipment can respectively execute the corresponding steps in the above method embodiments, so the details of each module will not be repeated here.
  • the details of each module please refer to the description of the corresponding steps above.
  • FIG. 5 is a schematic structural diagram of an embodiment of a storage device according to the present application.
  • the instruction file 21 may be stored in the storage device in the form of a software product while still recording each
  • the calculated data includes several instructions to enable a computer device (which may be a personal computer, a server, an intelligent robot, or a network device, etc.) or a processor to execute all or part of the steps of the methods of the embodiments of the present application.
  • the instruction file 21 also has certain independence, and can continue to cooperate with the processor 10 to execute relevant instructions when the operating system and the backup system fail, and will not be replaced, damaged, or emptied during the upgrade, boot program upgrade, and repair.
  • the aforementioned storage devices include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes , Or terminal devices such as computers, servers, mobile phones, and tablets.
  • the present application provides an intelligent detection method, device and storage device provided by the present application applied to a millimeter wave security inspection instrument, and a recognition image is obtained by inputting multiple human body images into a preset dangerous goods detection model for recognition, And input the frontal image of the human body into a preset body detection model to segment the area identification image, and fuse the identification image carrying the detection result with the area identification image to obtain the area identification image, so that the area identification image can be carried After the detection result is detected, the recognition result is then output according to the detection result of each region of the region recognition image and the corresponding preset threshold.
  • the image can be directly recognized through deep learning instead of artificial methods, which greatly reduces the cost and improves the detection efficiency.

Landscapes

  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Geophysics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Geophysics And Detection Of Objects (AREA)

Abstract

An intelligent detection method and device applied to a millimeter wave security check instrument, and a storage device. The method comprises steps of: inputting a plurality of human body images into a preset dangerous goods detection model to perform dangerous goods identification so as to obtain a plurality of identification images carrying detection results (S11); inputting a human body front image in the plurality of human body images into a preset body detection model for image segmentation into a plurality of regions so as to obtain region identification images (S12); performing fusing processing on the plurality of identification images and the region identification images so as to obtain a region identification image (S13); and outputting an identification result to the region identification image according to a detection result of regions and a corresponding preset threshold value (S14). On the one hand, the image can be directly identified by means of depth learning instead of a manual method, and thus the costs are greatly reduced, and the detection efficiency is improved; and on the other hand, region division is performed on the images thereof, and the results are output according to different threshold values of each region, so that the identification is more accurate, and a detection rate is improved while a false alarm rate is reduced.

Description

应用于毫米波安检仪的智能检测方法、装置以及存储装置Intelligent detection method, device and storage device applied to millimeter wave security detector 技术领域Technical field
本申请涉及智能安检领域，特别涉及到一种应用于毫米波安检仪的智能检测方法、装置以及存储装置。This application relates to the field of intelligent security inspection, in particular to an intelligent detection method, device and storage device applied to a millimeter wave security inspection instrument.
背景技术Background technique
近年来，安全问题日益得到世界人民的关注，对安检***的可靠性与智能化也提出了更高的要求。In recent years, security issues have increasingly received the attention of people around the world, and higher requirements have been placed on the reliability and intelligence of security inspection systems.
传统的金属探测器只能对近距离小范围目标进行检测，效率低，已远远不能满足安检的需求。尽管X光等各种射线具有很强的穿透力，但会对被测人体造成辐射伤害，即使当前存在低辐射剂量的X光机，但其依然不容易被公众接受。红外线是靠物体表面温度成像，在有织物遮挡的情况下无法清晰成像。而毫米波成像***不仅可以检测出隐藏在织物下的金属物体，还可以检测出塑料***，***等危险品，获得的信息更加详尽、准确，可以大大地降低误警率。因此，近年来毫米波成像技术在人员安检等方面得到了更加广泛的应用。Traditional metal detectors can only detect short-range and small-range targets, which is inefficient and far from meeting the requirements of security inspection. Although X-rays and other rays have strong penetrating power, they will cause radiation damage to the human body to be measured. Even though there are currently low-radiation dose X-ray machines, they are still not easily accepted by the public. Infrared is imaged by the temperature of the surface of the object, and it cannot be clearly imaged when it is covered by fabric. The millimeter-wave imaging system can not only detect metal objects hidden under the fabric, but also detect plastic pistols, explosives and other dangerous goods. The information obtained is more detailed and accurate, which can greatly reduce the false alarm rate. Therefore, in recent years, millimeter-wave imaging technology has been more widely used in personnel security and other aspects.
毫米波成像技术会使用到毫米波雷达。毫米波雷达是指雷达发射信号频率在毫米波频段，毫米波的频率为30GHz到300GHz(波长从1mm到10mm)，在实际工程应用中，常把毫米波的低端频率降到26GHz。在电磁波谱中，毫米波频率的位置介于微波与红外之间。与微波相比，毫米波的典型特点是波长短、频带宽(具有很广阔的利用空间)以及在大气中的传播特性。与红外相比，毫米波具有全天候工作的能力并且可用于烟尘，云雾等恶劣环境下。在微波频段越来越拥挤的情况下，毫米波兼顾微波的优点，并且还具备低频段微波所不具备的一些优点。Millimeter wave imaging technology will use millimeter wave radar. Millimeter-wave radar means that the frequency of the radar emission signal is in the millimeter-wave band, and the frequency of the millimeter wave is 30GHz to 300GHz (wavelength from 1mm to 10mm). In practical engineering applications, the low-end frequency of the millimeter wave is often reduced to 26GHz. In the electromagnetic spectrum, the position of the millimeter wave frequency is between microwave and infrared. Compared with microwaves, the typical characteristics of millimeter waves are short wavelength, frequency bandwidth (with a very wide utilization space), and propagation characteristics in the atmosphere. Compared with infrared, millimeter wave has the ability to work around the clock and can be used in harsh environments such as smoke, clouds and fog. In the increasingly crowded microwave frequency band, millimeter wave takes into account the advantages of microwave, and also has some advantages that low-frequency microwave does not.
现有技术中，安检装置对于被检对象进行扫描后，一般是在电脑上进行成像，然后工作人员对成像的图片进行观察检测后确认其是否携带危险品，或者仅仅只能对金属等物质进行报警，这样显得不够智能，且由于人眼会存在疲劳等问题，其检测效率较低，检测率也不高。In the prior art, after a security inspection device scans an object to be inspected, it is generally imaged on a computer, and then the staff observes and inspects the imaged image to confirm whether it carries dangerous goods, or can only carry out metal and other substances. Alarm, this seems not smart enough, and due to human eyes fatigue problems, its detection efficiency is low, and the detection rate is not high.
技术解决方案Technical solution
本申请提供一种应用于毫米波安检仪的智能检测方法、装置以及存储装置。以解决现有技术中安检检测效率较低，且准确率较为一般的问题。The present application provides an intelligent detection method, device, and storage device applied to a millimeter wave security detector. In order to solve the problems in the prior art that the security inspection efficiency is low and the accuracy rate is relatively general.
为解决上述技术问题，本申请采用的一个技术方案是：提供一种应用于毫米波安检仪的智能检测方法，该方法包括：将多张人体图像输入到预设危险品检测模型中进行危险品识别以获得多张携带检测结果的识别图像；将所述多张人体图像中的人体正面图像输入到预设的身体检测模型进行图像分割成多个区域以获得区域标识图像；对所述多张识别图像与所述区域标识图像进行融合处理以获取区域识别图像；对所述区域识别图像根据区域的检测结果及对应的预设阈值输出识别结果。In order to solve the above technical problems, a technical solution adopted by the present application is: to provide an intelligent detection method applied to a millimeter-wave safety inspection instrument, which includes: inputting multiple human body images into a preset dangerous goods detection model for dangerous goods Recognition to obtain multiple recognition images carrying the detection results; input the frontal image of the human body in the multiple human body images to a preset body detection model to perform image segmentation into multiple regions to obtain area identification images; The recognition image is merged with the area identification image to obtain an area recognition image; and the recognition result is output to the area recognition image according to the detection result of the area and the corresponding preset threshold.
为解决上述技术问题，本申请采用的另一个技术方案是：提供一种应用于毫米波安检仪的智能检测装置，其特征在于，所述智能检测装置包括处理器与所述处理器耦接的存储器，In order to solve the above technical problems, another technical solution adopted by the present application is to provide an intelligent detection device applied to a millimeter wave security inspection instrument, characterized in that the intelligent detection device includes a processor coupled to the processor Memory,
所述存储器存储有用于实现上述中任一项所述的显示方法的程序指令；所述处理器用于将多张人体图像输入到预设危险品检测模型中进行危险品识别以获得多张携带检测结果的识别图像；将所述多张人体图像中的人体正面图像输入到预设的身体检测模型进行图像分割成多个区域以获得区域标识图像；对所述多张识别图像与所述区域标识图像进行融合处理以获取区域识别图像；对所述区域识别图像根据区域的检测结果及对应的预设阈值输出识别结果。The memory stores program instructions for implementing the display method described in any one of the above; the processor is used to input multiple human body images into a preset dangerous goods detection model for dangerous goods recognition to obtain multiple pieces of carry detection The resulting recognition image; input the frontal image of the human body into the preset body detection model to perform image segmentation into multiple regions to obtain an area identification image; to identify the multiple images and the area identification The image is fused to obtain an area recognition image; the area recognition image is outputted according to the detection result of the area and the corresponding preset threshold.
为解决上述技术问题，本申请采用的另一个技术方案是，提供一种存储装置，该存储装置存储有能够实现以上任一项所述方法的程序文件。To solve the above technical problem, another technical solution adopted by the present application is to provide a storage device that stores a program file capable of implementing any of the above methods.
本申请的有益效果是：区别于现有技术，本申请提供应用于毫米波安检仪的智能检测方法、装置以及存储装置，通过将多张人体图像输入到预设危险品检测模型进行识别以得到识别图像，并将人体正面图像输入到预设的身体检测模型进行分割获取区域标识图像，并通过将携带了检测结果的识别图像与其区域标识图像进行融合，以获取区域识别图像，以使得其区域识别图像携带了检测结果，随后根据其区域识别图像的每个区域的检测结果及对应的预设阈值来输出识别结果。一方面通过深度学习的方式可以直接对图像进行识别，而不是 采用人工的方式，极大的减少了成本，提高了检测效率。一方面通过将其图像进行区域划分，并根据每个区域不同的阈值来输出结果，使得其识别更为精确，使得在提高检出率的同时也降低误报率。The beneficial effects of this application are: different from the prior art, this application provides an intelligent detection method, device and storage device applied to a millimeter wave security inspection instrument, which is obtained by inputting multiple human body images into a preset dangerous goods detection model for recognition Recognize the image and input the frontal image of the human body into the preset body detection model to segment the area identification image, and fuse the identification image carrying the detection result with its area identification image to obtain the area identification image to make its area The recognition image carries the detection result, and then the recognition result is output according to the detection result of each region of the region recognition image and the corresponding preset threshold. On the one hand, the image can be directly recognized through deep learning instead of manual methods, which greatly reduces the cost and improves the detection efficiency. On the one hand, by dividing its image into regions and outputting results according to different thresholds of each region, it makes its identification more accurate, which reduces the false alarm rate while increasing the detection rate.
附图说明BRIEF DESCRIPTION
图1是本申请应用于毫米波安检仪的智能检测方法的第一实施方式的结构示意图；FIG. 1 is a schematic structural diagram of a first embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application;
图2是本申请第一实施例S12的子流程示意图；2 is a schematic diagram of a sub-flow of S12 of the first embodiment of the present application;
图3是本申请应用于毫米波安检仪的智能检测方法的第二实施方式的结构示意图；3 is a schematic structural view of a second embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application;
图4是本申请应用于毫米波安检仪的智能检测装置一实施方式的结构示意框图；4 is a schematic structural block diagram of an embodiment of an intelligent detection device applied to a millimeter wave security inspection instrument of the present application;
图5是本申请存储装置一实施方式的结构示意图。5 is a schematic structural diagram of an embodiment of a storage device of the present application.
本发明的实施方式Embodiments of the invention
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅是本申请的一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without making creative efforts fall within the protection scope of the present application.
另外，若本申请实施例中有涉及“第一”、“第二”等的描述，则该“第一”、“第二”等的描述仅用于描述目的，而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外，各个实施例之间的技术方案可以相互结合，但是必须是以本领域普通技术人员能够实现为基础，当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在，也不在本实用新型要求的保护范围之内。In addition, if there are descriptions related to "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are for descriptive purposes only, and cannot be understood as instructions or hints Its relative importance or implicitly indicates the number of technical features indicated. Thus, the features defined as "first" and "second" may include at least one of the features explicitly or implicitly. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary skilled in the art to achieve, when the combination of technical solutions contradicts each other or cannot be achieved, it should be considered that the combination of such technical solutions does not exist , Nor within the scope of protection required by this utility model.
请参阅图1，图1是本申请应用于毫米波安检仪的智能检测方法的第一实施例流程示意图，其具体包括如下步骤：Please refer to FIG. 1. FIG. 1 is a schematic flowchart of a first embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application, which specifically includes the following steps:
S11，将多张人体图像输入到预设危险品检测模型中进行危险品识别以获得多张携带检测结果的识别图像。S11. Input multiple human body images into a preset dangerous goods detection model to perform dangerous goods recognition to obtain multiple recognition images carrying detection results.
首先将多张人体图像输入到预设的危险品检测模型中进行检测，其多张人体图像具体是可以通过安检仪对被检对象进行多角度采集得到的太赫兹扫描图像。First, multiple human body images are input into a preset dangerous goods detection model for detection. The multiple human body images are specifically a terahertz scan image that can be acquired by the security inspection instrument at multiple angles to the inspected object.
在具体实施例中，其安检仪具体可以是太赫兹圆柱式安检仪，可以对被检对象进行全方位多角度的扫描，从而得到被检对象不同角度的成像图片，将这些成像图片作为其用于识别的人体图像。In a specific embodiment, the security device may specifically be a terahertz cylindrical security device, which can scan the subject in all directions and multiple angles to obtain imaging pictures of the subject at different angles, and use these imaging pictures as its purpose Human images for recognition.
在具体实施例中，其预设的危险品检测模型是通过预先进行训练得到的，其具体可以通过将大量的人体携带危险品的图像为样本数据，并以VGGNet为特征提取网络进行最优参训练得到的。In a specific embodiment, the preset dangerous goods detection model is obtained by pre-training, which can be achieved by taking a large number of images of dangerous goods carried by the human body as sample data and using VGGNet as the feature extraction network for optimal parameters. Training.
在具体实施例中，其人体可以包括不同性别的人，如男女，不同衣着的人，如冬装、夏装、春秋装等等，或者直接以厚度为区别，如短袖、卫衣、羽绒服等等这些，不同人体BMI(体质量指数)值的人，如BMI小于18.5、BMI介于18.5和24之间以及BMI大于24的不同的人。In a specific embodiment, the human body may include people of different genders, such as men and women, people of different clothes, such as winter clothes, summer clothes, spring and autumn clothes, etc., or directly differentiated by thickness, such as short sleeves, sweaters, down jackets, etc. , People with different BMI (Body Mass Index) values, such as different people with BMI less than 18.5, BMI between 18.5 and 24, and BMI greater than 24.
其危险品包括有常见的一些危险品，其具体可以是如***、金属刀、陶瓷刀、矩形粉末***物、碟形粉末***物、不规则粉末***物、液体***物、打火机以及其他等等一些容易携带，且常见的危险品。Dangerous goods include some common dangerous goods, which can be such as pistols, metal knives, ceramic knives, rectangular powder explosives, dish-shaped powder explosives, irregular powder explosives, liquid explosives, lighters and others, etc. Some easy to carry and common dangerous goods.
通过对上述不同的人体与不同的危险品进行随机的组合，并通过圆柱式安检仪对其进行多角度采集，从而获取到样本数据库。Through the random combination of the above different human bodies and different dangerous goods, and the multi-angle collection of the cylindrical security inspection instrument, the sample database is obtained.
在具体实施例中，其多角度可以是以人体头部到脚步为中心轴，围绕该轴进行360度的采集。In a specific embodiment, the multi-angle can take the human head to footsteps as the central axis, and collect 360 degrees around the axis.
通过圆柱式安检仪采集大量不同的人体携带至少一种的危险品的多角度的图像作为其样本数据库，并用于输入到VGGNet做最优参训练，从而得到其预设的危险品检测模型。A multi-angle image of a large number of different humans carrying at least one kind of dangerous goods is collected as a sample database by a cylindrical security inspection instrument, and is used for input to VGGNet for optimal parameter training, thereby obtaining its preset dangerous goods detection model.
其中，在2014年，牛津大学计算机视觉组(Visual Geometry Group)和Google DeepMind公司的研究员一起研发出了新的深度卷积神经网络：VGGNe。Among them, in 2014, the Oxford University Computer Vision Group (Visual Geometry Group) and Google DeepMind researchers jointly developed a new deep convolutional neural network: VGGNe.
VGGNet探索了卷积神经网络的深度与其性能之间的关系，成功地构筑了16～19层深的卷积神经网络，证明了增加网络的深度能够在一定程度上影响网络最终的性能，使错误率大幅下降，同时拓展性又很强，迁移到其它图片数据上的泛化性也非常好。VGGNet explored the relationship between the depth of the convolutional neural network and its performance, and successfully constructed a 16-19 layer deep convolutional neural network, proving that increasing the depth of the network can affect the final performance of the network to a certain extent, making errors The rate has dropped significantly, while being very expandable, and the generalization of migration to other image data is also very good.
输入的每一张人体图像均会相应的获取到一张携带检测结果的识别图像， 因此，在完成全部识别后，会得到多张携带检测结果的识别图像。Each input human body image will correspondingly obtain a recognition image carrying the detection result, therefore, after all recognition is completed, multiple recognition images carrying the detection result will be obtained.
需要知道的是，这里携带检测结果的识别图像可以是指其识别图像一体携带了其检测结果，也可以是识别图像与检测结果是分开的，但是相互映射进行联系。It should be known that the identification image carrying the detection result here may mean that the identification image carries the detection result integrally, or the identification image and the detection result are separate, but they are related to each other by mapping.
S12，将所述多张人体图像中的人体正面图像输入到预设的身体检测模型进行图像分割成多个区域以获得区域标识图像。S12. Input a frontal image of the human body in the plurality of human body images to a preset body detection model to perform image segmentation into multiple regions to obtain an area identification image.
由于是以人体的头部到脚步为中心轴进行多角度的采集，因此其多张人体图像中包括有人体的正面图像，由于正面图像具有更好的角度性，因此利于提高识别率，进一步的，将其人体正面图像输入到预设的身体检测模型中进行图像分割，以对其不同区域进行标识，从而获取到区域标识图像。Because the human head to footsteps are used as the central axis for multi-angle acquisition, the multiple human body images include the front image of the human body. Because the front image has a better angle, it is conducive to improving the recognition rate and further , The frontal image of the human body is input into a preset body detection model to perform image segmentation, so as to identify different areas thereof, thereby obtaining an area identification image.
具体的，其预设的身体检测模型也是通过将大量的人体携带危险品的图像作为样本数据库，以ZFNet网络为特征提取网络，然后将其样本数据库输入到其ZFNet网络做最优参训练得到的。Specifically, its preset body detection model is also obtained by using a large number of images of humans carrying dangerous goods as a sample database, using the ZFNet network as a feature extraction network, and then inputting its sample database to its ZFNet network for optimal parameter training. .
具体的，使用一个多层的反卷积网络来可视化训练过程中特征的演化及发现潜在的问题；同时根据遮挡图像局部对分类结果的影响来探讨对分类任务而言到底那部分输入信息更重要。Specifically, use a multi-layer deconvolution network to visualize the evolution of features during the training process and discover potential problems; at the same time, according to the impact of the partial occlusion image on the classification results, it is more important to classify the input information for the classification task. .
在将其正面识别图像输入到预设的身体检测模型进行识别后，可以根据人体的区域将其正面识别图像进行分割，以得到不同的区域，并对不同区域进行标识。After inputting its front recognition image into a preset body detection model for recognition, the front recognition image can be segmented according to the region of the human body to obtain different regions and identify the different regions.
在具体实施例中，其分割后可以得到第一区域、第二区域、第三区域以及第四区域。其具体是依照被检对象的人体区域进行划分的，其第一区域为正面识别图像中的被检对象的头部区域，具体包括脑袋与颈部区域，其第二区域为正面识别图像中的被检对象的肩部至裆部的区域，即整个胸部到腹部的区域，其第三区域为正面识别图像中被检对象的手部区域，其包括左臂区和右臂区，其第四区域为正面识别图像中被检对象的腿部区域，同时也包括左腿区和右腿区。In a specific embodiment, the first area, the second area, the third area, and the fourth area can be obtained after the division. Specifically, it is divided according to the body area of the subject. The first area is the head area of the subject in the front recognition image, specifically including the head and neck area, and the second area is the front area in the front recognition image. The area from the shoulder to the crotch of the subject, that is, the entire chest to the abdomen, the third area is the hand area of the subject in the front recognition image, which includes the left arm area and the right arm area, and the fourth The area is the leg area of the subject in the front recognition image, and also includes the left leg area and the right leg area.
对这些区域分别进行标识，以获取到区域标识图像。Mark these areas separately to obtain area identification images.
请参阅图2，图2是本申请应用于毫米波安检仪的智能检测方法图1步骤中S12的子步骤，其具体包括如下步骤：Please refer to FIG. 2, which is the sub-step of S12 in the step 1 of FIG. 1 of the intelligent detection method of the present application applied to the millimeter wave security inspection instrument, which specifically includes the following steps:
S121，判断第二区域是否识别成功。S121: Determine whether the second area is successfully identified.
在输入到身体检测模型后，首先对其正面识别图像进行识别，看是否能够识别出其第二区域，即肩部到裆部位置的区域，从而判断该正面识别图像的可识别性，具体的，如果该第二区域无法完成识别，则证明，该正面识别图像可能是较为模糊的，或者角度并非正面，这样可以会给具体的识别过程带来较大的模糊度，导致危险品识别率较低，识别效果不好。After inputting into the body detection model, first recognize its front recognition image to see if it can recognize its second area, that is, the area from the shoulder to the crotch position, so as to judge the recognizability of the front recognition image, specifically If the second area cannot be recognized, it proves that the positive recognition image may be blurry, or the angle is not positive, which may bring greater ambiguity to the specific recognition process, resulting in a higher dangerous goods recognition rate. Low, the recognition effect is not good.
S122，如果识别成功，则对第一区域、第三区域以及第四区域进行补偿。S122: If the recognition is successful, compensate the first area, the third area, and the fourth area.
如果识别正常，则对其他区域，即第一区域、第三区域以及第四区域进行补偿，具体的，如果其他区域出现了偏移，如第三区域发生了偏移，则可以根据其第二区域获取到身体中线，然后计算其偏移的区域，如第三区域离其身体中线的距离，随后求得其平均值，并根据其平均值平移其第三区域。If the recognition is normal, the other areas, that is, the first area, the third area, and the fourth area are compensated. Specifically, if the other areas are offset, for example, the third area is offset, you can use the second area The area obtains the body midline, and then calculates its offset area, such as the distance of the third area from its body centerline, and then finds its average value, and translates its third area according to its average value.
如果是第三区域的一侧，如右臂区发生了缺失，则根据其第二区域获取的身体中线，并计算左臂区离身体中线的距离，并生成对称的区域。从而使得其正面图像为完整且清晰的。If it is on the side of the third area, if the right arm area is missing, the body midline obtained from the second area is calculated, and the distance of the left arm area from the body midline is calculated, and a symmetric area is generated. So that the front image is complete and clear.
S123，如果识别失败；则输出图像识别异常，并重新获取图像。S123, if the recognition fails; output image recognition abnormality, and reacquire the image.
如果第二区域都识别失败，可能证明其正面识别图像并非正面图像，所以其效果可能较差，则认为其需要重新识别，则提示工作人员图像识别异常，并重新进行图像获取。If the recognition of the second area fails, it may prove that the front recognition image is not a front image, so its effect may be poor, and it is considered that it needs to be re-recognized, and the staff is prompted to recognize the image recognition abnormality and re-acquire the image.
S13，对所述多张识别图像与所述区域标识图像进行融合处理以获取区域识别图像。S13. Perform fusion processing on the multiple identification images and the area identification image to obtain an area identification image.
在获取到其区域标识图像与多种识别图像后，需要将其多张识别图像融合到其区域标识图像中，以使得其区域标识图像携带检测结果。After acquiring its area identification image and multiple identification images, it is necessary to fuse its multiple identification images into its area identification image, so that its area identification image carries the detection result.
具体的，将多张识别图像映射到区域标识图像中，以使得多种标识图像与区域标识图像进行融合，即，将其标识图像携带的检测结果融合到其区域标识图像中，从而获取到区域识别图像，其中，所述区域识别图像携带了检测结果。同时，其区域标识图像也完成了对每个区域的标识。Specifically, multiple identification images are mapped into the area identification image, so that the multiple identification images are merged with the area identification image, that is, the detection result carried by the identification image is fused into the area identification image, thereby obtaining the area An identification image, wherein the area identification image carries the detection result. At the same time, its area identification image also completes the identification of each area.
S14，对所述区域识别图像根据区域的检测结果及对应的预设阈值输出识别结果。S14: Output a recognition result to the area recognition image according to the detection result of the area and the corresponding preset threshold.
在具体实施例中，人体携带危险品的区域可能主要集中在其胸部、跨部等区域，如果全身均采用一个阈值，则可能提高误报率，影响检测结果，因此对 于整个人体而言，其需要对人体的不同区域设置不同的预设阈值，且具体的，针对不同的危险品与不同的区域设置不同的预设阈值，比如***等这种危险品，一般是放置跨部的比较多，因此相应的，胸部可以对其设置较为低阈值，以提高识别准确率。In a specific embodiment, the area where the human body carries dangerous goods may be mainly concentrated in its chest, span, and other areas. If a threshold is used for the entire body, the false alarm rate may be increased and the detection result may be affected. Therefore, for the entire human body, its Different preset thresholds need to be set for different areas of the human body, and specifically, different preset thresholds are set for different dangerous goods and different areas, such as pistols and other dangerous goods, generally more cross-sections are placed, Correspondingly, the chest can set a lower threshold to improve the recognition accuracy.
对于第一区域，即头部区域而言，其携带危险品的概率较少，因此可以适当的提高其阈值，以防止出现误报。For the first area, that is, the head area, there is less probability of carrying dangerous goods, so the threshold can be appropriately increased to prevent false alarms.
因此，首先获取其区域识别图像中每个区域的检测结果，根据其各个区域所携带的检测结果与其对应的预设阈值进行比较，如果大于其阈值，则进行警报，并定位到其警报的具体部位。Therefore, firstly obtain the detection result of each area in its area recognition image, compare it with the corresponding preset threshold according to the detection result carried by each area, if it is greater than its threshold, then make an alarm, and locate the specific alarm Location.
上述实施例中，通过深度学习的方式对多张人体图像进行识别，提高其识别的准确率，得到其较为准确的检测结果，一方面深度学习的方式对人体正面图像进行处理，从而根据人体的结果进行区域划分，随后将其两次识别分割的结果进行融合，从而获取到人体正面图像上每个区域的检测结果，随后根据其检测结果与其对应的区域的预设阈值进行比较，进一步的提高其准确率，从而减少误报率。In the above embodiments, multiple human body images are recognized through deep learning to improve the accuracy of recognition and obtain more accurate detection results. On the one hand, deep learning processes the frontal image of the human body, based on the human body’s The results are divided into regions, and then the results of the two recognition and segmentation are fused to obtain the detection results of each region on the front image of the human body, and then compared with the preset thresholds of the corresponding regions according to the detection results and further improved Its accuracy rate, thereby reducing the false alarm rate.
请参阅图3，图3是本申请应用于毫米波安检仪的智能检测方法第二实施例的流程示意图，其具体包括如下步骤：Please refer to FIG. 3. FIG. 3 is a schematic flowchart of a second embodiment of an intelligent detection method applied to a millimeter wave security inspection instrument of the present application, which specifically includes the following steps:
S21，对图像进行均值化处理。S21, the image is averaged.
在具体实施例中，在对其样本数据库进行训练时，需要对其样本数据，即图像进行数据的处理，以使得其训练的方式更有效，从而使得其模型的识别率、精准度越高，误报率越小。In a specific embodiment, when training its sample database, it is necessary to process the sample data, that is, the image, to make its training method more effective, so that its model recognition rate and accuracy are higher, The smaller the false alarm rate.
首先采用如下公式对其图像进行均值化处理，从而根据凸优化理论与数据概率分布相关知识，数据中心化符合数据分布规律，更容易取得训练之后的泛化效果。First, the following formula is used to average the images, so that according to the convex optimization theory and knowledge about the probability distribution of data, data centralization conforms to the data distribution law, and it is easier to obtain the generalization effect after training.
Figure PCTCN2019121764-appb-000001
其中，μ是图像均值，X表示图像矩阵，σ表示标准方差，N表示图像像素数量，Per_image_standardizatio为均值化的后结果。
Figure PCTCN2019121764-appb-000001
Among them, μ is the image mean, X represents the image matrix, σ represents the standard deviation, N represents the number of image pixels, and Per_image_standardizatio is the averaged result.
S22，对图像进行归一化处理。S22, normalize the image.
虽然采用如下公式对图像进行归一化处理，从而保证所有的维度上数据都 在一个变化幅度上。Although the following formula is used to normalize the image, to ensure that the data in all dimensions are in a varying range.
Figure PCTCN2019121764-appb-000002
其中，x i表示图像像素点值,min(x),max(x)分别表示图像像素的最小与最大值，norm表示其归一化后的结果。
Figure PCTCN2019121764-appb-000002
Among them, x i represents the pixel value of the image, min(x), max(x) represent the minimum and maximum value of the image pixel, and norm represents its normalized result.
S23，通过设置多种预设面积与预设比例的组合对图像进行优化。S23. Optimize the image by setting a combination of various preset areas and preset ratios.
由于在具体的过程中，其危险品可能大小不同，如小的如小刀，如大的如***，因此在具体的训练识别过程中，采用同样的大小的图像可能无法精准进行识别，具体的，可以通过根据预设面积与预设比例的组件对其图像进行优化，形成多种组合的图像，具体的，对于某个16*16pt的图像，可以采用3种预设面积的和3种预设比例的组合，一共形成9张图像。In the specific process, the size of the dangerous goods may be different, such as small, such as a knife, such as a large, such as a pistol, so in the specific training and recognition process, the use of images of the same size may not be able to accurately identify, specific, You can optimize the image based on the components of the preset area and preset ratio to form multiple combined images. Specifically, for a 16*16pt image, you can use 3 preset areas and 3 presets The combination of proportions forms a total of 9 images.
在具体实施例中，其3种预设面积可以是8倍、16倍、32倍，其3种预设比例的可以是1:2、1:1、2:1等等。如采用8倍面积，1:1比例的组合，则可以将其16*16pt的图像形成128*128pt的图像。In a specific embodiment, the three preset areas may be 8 times, 16 times, or 32 times, and the three preset ratios may be 1:2, 1:1, 2:1, and so on. If the combination of 8 times area and 1:1 ratio is adopted, the 16*16pt image can be formed into 128*128pt image.
这样极大了增加了整个图像的多样性，无论是对于较大的危险品还是较小的危险品，均能进行良好的扫描，以提供识别率与训练度。This greatly increases the diversity of the entire image, whether it is for larger or smaller dangerous goods, it can be scanned well to provide recognition rate and training degree.
S24，对训练模型采用焦点损失函数进行优化训练。S24, the focus model is used to optimize the training of the training model.
在具体的训练过程中，部分数据较为复杂，处理难度不同，则需要对其训练模型采用焦点损失函数进行优化训练，以减少复杂度，提高识别度。In the specific training process, part of the data is more complex and the processing difficulty is different. Therefore, it is necessary to use the focus loss function to optimize the training of its training model to reduce complexity and improve recognition.
具体是通过降低了大量简单负样本在训练中所占的权重，也可理解为一种困难样本挖掘。从而使得在保持较高的检测效率与检出水平的同时，可以降低其误报率。其具体采用如下的损失函数进行处理。Specifically, by reducing the weight of a large number of simple negative samples in training, it can also be understood as a kind of difficult sample mining. Therefore, the false alarm rate can be reduced while maintaining a high detection efficiency and detection level. It uses the following loss function for processing.
FL(p t)＝-α t(1-p t) γlog(p t)，p t是不同类别的分类概率，γ是个大于0的值，α t是个[0，1]间的小数，γ和α t都是固定值，不参与具体的训练过程。 FL(p t )=-α t (1-p t ) γ log(p t ), p t is the classification probability of different categories, γ is a value greater than 0, α t is a decimal between [0, 1], Both γ and α t are fixed values and do not participate in the specific training process.
通过上述方式，可以将样本数据库中的复杂样本进行分离。In this way, complex samples in the sample database can be separated.
上述实施例中，通过对样本数据库中的样本，也就是需要对其训练模型采用焦点损失函数进行优化训练，从而获取其预测的数据与实际数据的差距程度，使得训练模型的损失要收敛，使得其训练结果更为精确，且减少了数据的冗余度。In the above embodiment, the samples in the sample database, that is, the training model needs to be optimized using the focus loss function to obtain the difference between the predicted data and the actual data, so that the loss of the training model must be converged, so that The training result is more accurate, and the redundancy of data is reduced.
在具体实施例中，其在识别过程中对人体图像也可以进行该处理。对于其 身体检测模型的训练过程也同样可以进行类似的处理，这里不做限定。In a specific embodiment, it can also perform this process on the human body image during the recognition process. The training process of the body detection model can also be processed similarly, which is not limited here.
请参阅图4，图4是本申请提供的应用于毫米波安检仪的智能检测装置的一实施方式结构示意框图。Please refer to FIG. 4. FIG. 4 is a schematic structural block diagram of an embodiment of an intelligent detection device applied to a millimeter wave security detector provided by this application.
本实施例提供的应用于毫米波安检仪的智能检测装置具体包括处理器10以及与该处理器耦接的存储器11。The intelligent detection device applied to the millimeter wave security detector provided in this embodiment specifically includes a processor 10 and a memory 11 coupled to the processor.
其中，处理器10可以是CPU(Central Processing Unit，中央处理单元)。或者GPU(Graphics Processing Unit，图形处理器)，处理器10可能是一种集成电路芯片，具有信号的处理能力。处理器10还可以是通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现成可编程门阵列(FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。The processor 10 may be a CPU (Central Processing Unit, central processing unit). Or GPU (Graphics Processing Unit), the processor 10 may be an integrated circuit chip with signal processing capabilities. The processor 10 may also be a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components . The general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
在本实施中，处理器10可以将多张人体图像输入到预设危险品检测模型中进行危险品识别以获得多张携带检测结果的识别图像；将所述多张人体图像中的人体正面图像输入到预设的身体检测模型进行图像分割成多个区域以获得区域标识图像；对所述多张识别图像与所述区域标识图像进行融合处理以获取区域识别图像；对所述区域识别图像根据区域的检测结果及对应的预设阈值输出识别结果。In this implementation, the processor 10 may input multiple human body images into a preset dangerous goods detection model for dangerous goods recognition to obtain multiple recognition images carrying the detection results; the human body front image among the multiple human body images Input to a preset body detection model to perform image segmentation into multiple areas to obtain an area identification image; perform fusion processing on the multiple identification images and the area identification image to obtain an area identification image; based on the area identification image The detection result of the region and the corresponding preset threshold output the recognition result.
其存储器11存储有能够实现上述任一实施例的指令文件111。The memory 11 stores an instruction file 111 that can implement any of the above embodiments.
上述设备的其他模块终端可分别执行上述方法实施例中对应的步骤，故在此不对各模块进行赘述，详细请参阅以上对应步骤的说明。The other module terminals of the above-mentioned equipment can respectively execute the corresponding steps in the above method embodiments, so the details of each module will not be repeated here. For details, please refer to the description of the corresponding steps above.
参阅图5，图5为本申请存储装置一实施方式的结构示意图，有能够实现上述所有方法的指令文件21，该指令文件21可以以软件产品的形式存储在上述存储装置中，同时还是记录各种计算的数据，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，智能机器人，或者网络设备等)或处理器(processor)执行本申请各个实施方式方法的全部或部分步骤。Referring to FIG. 5, FIG. 5 is a schematic structural diagram of an embodiment of a storage device according to the present application. There is an instruction file 21 capable of implementing all of the above methods. The instruction file 21 may be stored in the storage device in the form of a software product while still recording each The calculated data includes several instructions to enable a computer device (which may be a personal computer, a server, an intelligent robot, or a network device, etc.) or a processor to execute all or part of the steps of the methods of the embodiments of the present application.
所述指令文件21还具有一定独立性，可以在运行***、备份***发生故障时候继续配合处理器10执行相关指令，在升级、引导程序升级以及修复中不会被替换、损坏以及清空。The instruction file 21 also has certain independence, and can continue to cooperate with the processor 10 to execute relevant instructions when the operating system and the backup system fail, and will not be replaced, damaged, or emptied during the upgrade, boot program upgrade, and repair.
而前述的存储装置包括：U盘、移动硬盘、只读存储器(ROM，Read-Only  Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质，或者是计算机、服务器、手机、平板等终端设备。The aforementioned storage devices include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes , Or terminal devices such as computers, servers, mobile phones, and tablets.
综上所述，本申请提供一种本申请提供应用于毫米波安检仪的智能检测方法、装置以及存储装置，通过将多张人体图像输入到预设危险品检测模型进行识别以得到识别图像，并将人体正面图像输入到预设的身体检测模型进行分割获取区域标识图像，并通过将携带了检测结果的识别图像与其区域标识图像进行融合，以获取区域识别图像，以使得其区域识别图像携带了检测结果，随后根据其区域识别图像的每个区域的检测结果及对应的预设阈值来输出识别结果。一方面通过深度学习的方式可以直接对图像进行识别，而不是采用人工的方式，极大的减少了成本，提高了检测效率。一方面通过将其图像进行区域划分，并根据每个区域不同的阈值来输出结果，使得其识别更为精确，减少误报率。另一方面，在对模型进行训练时对数据进行了大量的处理，以提高识别率，并降低误报率，且加快了识别速度。In summary, the present application provides an intelligent detection method, device and storage device provided by the present application applied to a millimeter wave security inspection instrument, and a recognition image is obtained by inputting multiple human body images into a preset dangerous goods detection model for recognition, And input the frontal image of the human body into a preset body detection model to segment the area identification image, and fuse the identification image carrying the detection result with the area identification image to obtain the area identification image, so that the area identification image can be carried After the detection result is detected, the recognition result is then output according to the detection result of each region of the region recognition image and the corresponding preset threshold. On the one hand, the image can be directly recognized through deep learning instead of artificial methods, which greatly reduces the cost and improves the detection efficiency. On the one hand, by dividing its image into regions and outputting results according to different thresholds of each region, it makes its recognition more accurate and reduces the false alarm rate. On the other hand, when the model is trained, a lot of data is processed to improve the recognition rate, reduce the false positive rate, and speed up the recognition speed.
以上仅为本申请的实施方式，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结果或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。The above are only the embodiments of the present application, and therefore do not limit the patent scope of the present application. Any equivalent results or equivalent process transformations made by using the description and drawings of this application, or directly or indirectly used in other related technical fields, The same reason is included in the patent protection scope of this application.

Claims (15)

  1. 一种应用于毫米波安检仪的智能检测方法，其特征在于，所述方法包括：An intelligent detection method applied to a millimeter wave security detector, characterized in that the method includes:
    将多张人体图像输入到预设危险品检测模型中进行危险品识别以获得多张携带检测结果的识别图像；Input multiple human body images into a preset dangerous goods detection model for dangerous goods recognition to obtain multiple recognition images carrying detection results;
    将所述多张人体图像中的人体正面图像输入到预设的身体检测模型进行图像分割成多个区域以获得区域标识图像；Input the frontal image of the human body in the plurality of human body images into a preset body detection model to divide the image into multiple regions to obtain an area identification image;
    对所述多张识别图像与所述区域标识图像进行融合处理以获取区域识别图像；Performing fusion processing on the multiple identification images and the area identification image to obtain an area identification image;
    对所述区域识别图像根据区域的检测结果及对应的预设阈值输出识别结果。The recognition result is output to the region recognition image according to the detection result of the region and the corresponding preset threshold.
  2. 根据权利要求1所述的智能检测方法，其特征在于，所述将多张人体图像输入到预设危险品检测模型中进行危险品识别以获得多张携带检测结果的识别图像之前包括：The intelligent detection method according to claim 1, wherein the step of inputting a plurality of human body images into a preset dangerous goods detection model for dangerous goods recognition to obtain a plurality of recognition images carrying detection results includes:
    通过安检仪采集被检对象多方角度的图像以得到所述多张人体图像。The multi-angle images of the inspected object are collected by the security inspection instrument to obtain the multiple human body images.
  3. 根据权利要求1所述的智能检测方法，其特征在于，所述预设神经网络模型是通过以VGGNet为特征提取网络，并将人体携带危险品的图像作为样本数据库进行最优参训练而得到的。The intelligent detection method according to claim 1, wherein the preset neural network model is obtained by using VGGNet as the feature extraction network, and using the image of the human body carrying dangerous goods as a sample database for optimal parameter training .
  4. 根据权利要求3所述的智能检测方法，其特征在于，所述人体携带危险品的图像是安检仪对携带至少一种危险品的人体进行多方角度的采集所获取的图像。The intelligent detection method according to claim 3, wherein the image of the dangerous goods carried by the human body is an image obtained by a security inspection instrument collecting multiple angles of the human body carrying at least one dangerous goods.
  5. 根据权利要求4所述的智能检测方法，其特征在于，所述人体包括人体BMI值、性别、衣着。The intelligent detection method according to claim 4, wherein the human body includes a human body BMI value, gender, and clothing.
  6. 根据权利要求4所述的智能检测方法，其特征在于，所述危险品至少包括***、金属刀、陶瓷刀、矩形粉末***物、碟形粉末***物、不规则粉末***物、液体***物、打火机。The intelligent detection method according to claim 4, wherein the dangerous goods include at least a pistol, a metal knife, a ceramic knife, a rectangular powder explosive, a dish-shaped powder explosive, an irregular powder explosive, a liquid explosive, lighter.
  7. 根据权利要求3所述的智能检测方法，其特征在于，所述将人体携带危险品的图像作为样本数据库进行最优参训练包括：The intelligent detection method according to claim 3, wherein the optimal parameter training using the images of dangerous goods carried by the human body as a sample database includes:
    对所述图像进行均值化处理；Averaging the image;
    对所述图像进行归一化处理；Normalize the image;
  8. 根据权利要求7所述的智能检测方法，其特征在于，所述方法还包括：The intelligent detection method according to claim 7, wherein the method further comprises:
    通过设置多种预设面积与预设比例的组合对所述图像进行优化；Optimize the image by setting a combination of multiple preset areas and preset ratios;
    对训练模型采用焦点损失函数进行优化训练。Focus training function is used to optimize the training model.
  9. 根据权利要求1所述的智能检测方法，其特征在于，所述对所述预设的身体检测模型是通过以ZFNet为特征提取网络，并将被人体携带危险品的图像作为样本数据库进行最优参训练而得到的。The intelligent detection method according to claim 1, wherein the preset body detection model is optimized by using ZFNet as a feature extraction network, and using images of dangerous goods carried by the human body as a sample database for optimization Obtained from training.
  10. 根据权利要求1所述的智能检测方法，其特征在于，所述将所述多张人体图像中的人体正面图像输入到预设的身体检测模型进行图像分割成多个区域以获得区域标识图像包括：The intelligent detection method according to claim 1, wherein the inputting the frontal image of the human body in the plurality of human body images to a preset body detection model for image segmentation into multiple regions to obtain an area identification image includes :
    对所述人体正面图像进行识别，以对所述人体正面图像的第一区域、第二区域、第三区域、第四区域进行标识；Identify the frontal image of the human body to identify the first area, the second area, the third area, and the fourth area of the frontal image of the human body;
    其中，所述第一区域所述正面识别图像中人体的头部区域；Wherein, in the first area, the head area of the human body in the front recognition image;
    所述第二区域为所述正面识别图像中人体的肩部至裆部区域；The second area is the shoulder to crotch area of the human body in the front recognition image;
    所述第三区域为所述正面识别图像中人体的手部区域；The third area is a hand area of the human body in the front recognition image;
    所述第四区域为所述正面识别图像中人体的腿部区域。The fourth area is the leg area of the human body in the front recognition image.
  11. 根据权利要求10所述的智能检测方法，其特征在于，所述方法还包括：The intelligent detection method according to claim 10, wherein the method further comprises:
    判断所述第二区域是否识别成功，Determine whether the second area is successfully identified,
    如果识别成功，则对所述第一区域、第三区域以及第四区域进行补偿后输出所述区域标识图像；If the recognition is successful, output the area identification image after compensating the first area, the third area, and the fourth area;
    如果识别失败；则输出图像识别异常，并重新获取图像。If the recognition fails; output image recognition abnormality, and re-acquire the image.
  12. 根据权利要求1所述的智能检测方法，其特征在于，所述对所述多张识别图像与所述区域标识图像进行融合处理以获取区域识别图像包括：The intelligent detection method according to claim 1, wherein the fusion processing of the plurality of identification images and the area identification image to obtain an area identification image includes:
    将所述多张识别图像映射到所述区域标识图像中，以使得所述标识图像与所述区域标识图像进行融合，从而获取到所述区域识别图像，其中，所述区域识别图像携带了检测结果。Mapping the plurality of identification images into the area identification image, so that the identification image and the area identification image are fused to obtain the area identification image, wherein the area identification image carries detection result.
  13. 根据权利要求1所述的智能检测方法，其特征在于，所述对所述区域识别图像根据区域的检测结果及对应的预设阈值输出识别结果包括：The intelligent detection method according to claim 1, wherein the output of the recognition result of the region recognition image according to the region detection result and the corresponding preset threshold includes:
    获取所述区域识别图像中每个区域的检测结果；Acquiring the detection result of each area in the area identification image;
    将所述检测结果与对应区域中所述预设阈值进行比较；Comparing the detection result with the preset threshold in the corresponding area;
    根据比较结果输出识别结果。The recognition result is output according to the comparison result.
  14. 一种应用于毫米波安检仪的智能检测装置，其特征在于，所述智能检 测装置包括处理器与所述处理器耦接的存储器，An intelligent detection device applied to a millimeter-wave security detector, characterized in that the intelligent detection device includes a processor and a memory coupled to the processor,
    所述存储器存储有用于实现如权利要求1-13中任一项所述的显示方法的程序指令；The memory stores program instructions for implementing the display method according to any one of claims 1-13;
    所述处理器用于将多张人体图像输入到预设危险品检测模型中进行危险品识别以获得多张携带检测结果的识别图像；将所述多张人体图像中的人体正面图像输入到预设的身体检测模型进行图像分割成多个区域以获得区域标识图像；对所述多张识别图像与所述区域标识图像进行融合处理以获取区域识别图像；对所述区域识别图像根据区域的检测结果及对应的预设阈值输出识别结果。The processor is used for inputting multiple human body images into a preset dangerous goods detection model for dangerous goods recognition to obtain multiple recognition images carrying detection results; inputting human body front images in the multiple human body images into a preset The body detection model performs image segmentation into multiple areas to obtain an area identification image; fuse the multiple identification images with the area identification image to obtain an area identification image; perform the area identification image according to the detection result of the area And the corresponding preset threshold to output the recognition result.
  15. 一种存储装置，其特征在于，存储有能够实现如权利要求1-13中任一项所述方法的程序文件。A storage device, characterized in that a program file capable of implementing the method according to any one of claims 1-13 is stored.
PCT/CN2019/121764 2018-12-28 2019-11-28 Intelligent detection method and device applied to millimeter wave security check instrument, and storage device WO2020134848A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811625395.9 2018-12-28
CN201811625395.9A CN109799544B (en) 2018-12-28 2018-12-28 Intelligent detection method and device applied to millimeter wave security check instrument and storage device

Publications (1)

Publication Number Publication Date
WO2020134848A1 true WO2020134848A1 (en) 2020-07-02

Family

ID=66558106

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/121764 WO2020134848A1 (en) 2018-12-28 2019-11-28 Intelligent detection method and device applied to millimeter wave security check instrument, and storage device

Country Status (2)

Country Link
CN (1) CN109799544B (en)
WO (1) WO2020134848A1 (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112327268A (en) * 2020-11-12 2021-02-05 奥谱毫芯(深圳)科技有限公司 Dangerous target identification method, device, system and medium based on millimeter wave signals
CN113537343A (en) * 2021-07-14 2021-10-22 厦门熵基科技有限公司 Metal classification method, device, equipment and storage medium
CN116071347A (en) * 2023-03-01 2023-05-05 山西戴德测控技术有限公司 Wear degree determination method, device, system and storage medium
CN116189083A (en) * 2023-01-11 2023-05-30 广东汇通信息科技股份有限公司 Dangerous goods identification method for community security inspection assistance

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109799544B (en) * 2018-12-28 2021-03-19 深圳市重投华讯太赫兹科技有限公司 Intelligent detection method and device applied to millimeter wave security check instrument and storage device
CN112016387A (en) * 2019-07-08 2020-12-01 杭州芯影科技有限公司 Contraband identification method and device suitable for millimeter wave security check instrument
CN110443158B (en) * 2019-07-18 2021-10-08 华中科技大学 Millimeter wave image-based human body hidden object detection method
CN110974250B (en) * 2019-12-27 2024-01-16 深圳市华讯方舟光电技术有限公司 Terahertz spectrum-based blood glucose detection method and device and computer storage medium
WO2022083023A1 (en) * 2020-10-23 2022-04-28 公安部第三研究所 Measurement system and measurement method for measuring detection performance of human body safety inspection equipment
CN113743276A (en) * 2021-08-30 2021-12-03 上海亨临光电科技有限公司 Method for judging human body part where target object is located in human body gray level image
CN114049733B (en) * 2021-11-10 2023-07-21 通号通信信息集团有限公司 Intrusion detection method, monitoring platform, intrusion detection system and readable storage medium
CN115311684B (en) * 2022-08-05 2023-03-28 杭州电子科技大学 Method for integrating multi-angle detection results of millimeter wave images
CN115564320B (en) * 2022-12-06 2023-04-07 成都智元汇信息技术股份有限公司 Multi-intelligent-algorithm-oriented scheduling management method, device and medium

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100278426A1 (en) * 2007-12-14 2010-11-04 Robinson Piramuthu Systems and methods for rule-based segmentation for objects with full or partial frontal view in color images
CN102629315A (en) * 2012-02-29 2012-08-08 北京无线电计量测试研究所 Automatic detection and identification apparatus of concealed object
CN102708560A (en) * 2012-02-29 2012-10-03 北京无线电计量测试研究所 Privacy protection method based on millimeter-wave imaging
CN106485268A (en) * 2016-09-27 2017-03-08 东软集团股份有限公司 A kind of image-recognizing method and device
CN108182663A (en) * 2017-12-26 2018-06-19 北京无线电计量测试研究所 A kind of millimeter-wave image effect reinforcing method, equipment and readable storage medium storing program for executing
CN108665509A (en) * 2018-05-10 2018-10-16 广东工业大学 A kind of ultra-resolution ratio reconstructing method, device, equipment and readable storage medium storing program for executing
CN108802840A (en) * 2018-05-31 2018-11-13 北京迈格斯智能科技有限公司 The method and its device of automatic identification object based on artificial intelligence deep learning
CN109086679A (en) * 2018-07-10 2018-12-25 西安恒帆电子科技有限公司 A kind of millimetre-wave radar safety check instrument foreign matter detecting method
CN109799544A (en) * 2018-12-28 2019-05-24 深圳市华讯方舟太赫兹科技有限公司 Intelligent detecting method, device and storage device applied to millimeter wave safety check instrument

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050022252A1 (en) * 2002-06-04 2005-01-27 Tong Shen System for multimedia recognition, analysis, and indexing, using text, audio, and digital video
US8213672B2 (en) * 2007-08-08 2012-07-03 Microsemi Corporation Millimeter wave imaging method and system to detect concealed objects
CN102542570B (en) * 2011-12-30 2014-05-07 北京华航无线电测量研究所 Method for automatically detecting dangerous object hidden by human body in microwave image
CN107238866B (en) * 2017-05-26 2019-05-21 西安电子科技大学 Millimeter wave video imaging system and method based on synthetic aperture technique
CN108520260B (en) * 2018-04-11 2022-02-01 中南大学 Method for identifying visible foreign matters in bottled oral liquid
CN108629319B (en) * 2018-05-09 2020-01-07 北京嘀嘀无限科技发展有限公司 Image detection method and system
CN108764328A (en) * 2018-05-24 2018-11-06 广东工业大学 The recognition methods of Terahertz image dangerous material, device, equipment and readable storage medium storing program for executing

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100278426A1 (en) * 2007-12-14 2010-11-04 Robinson Piramuthu Systems and methods for rule-based segmentation for objects with full or partial frontal view in color images
CN102629315A (en) * 2012-02-29 2012-08-08 北京无线电计量测试研究所 Automatic detection and identification apparatus of concealed object
CN102708560A (en) * 2012-02-29 2012-10-03 北京无线电计量测试研究所 Privacy protection method based on millimeter-wave imaging
CN106485268A (en) * 2016-09-27 2017-03-08 东软集团股份有限公司 A kind of image-recognizing method and device
CN108182663A (en) * 2017-12-26 2018-06-19 北京无线电计量测试研究所 A kind of millimeter-wave image effect reinforcing method, equipment and readable storage medium storing program for executing
CN108665509A (en) * 2018-05-10 2018-10-16 广东工业大学 A kind of ultra-resolution ratio reconstructing method, device, equipment and readable storage medium storing program for executing
CN108802840A (en) * 2018-05-31 2018-11-13 北京迈格斯智能科技有限公司 The method and its device of automatic identification object based on artificial intelligence deep learning
CN109086679A (en) * 2018-07-10 2018-12-25 西安恒帆电子科技有限公司 A kind of millimetre-wave radar safety check instrument foreign matter detecting method
CN109799544A (en) * 2018-12-28 2019-05-24 深圳市华讯方舟太赫兹科技有限公司 Intelligent detecting method, device and storage device applied to millimeter wave safety check instrument

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112327268A (en) * 2020-11-12 2021-02-05 奥谱毫芯(深圳)科技有限公司 Dangerous target identification method, device, system and medium based on millimeter wave signals
CN113537343A (en) * 2021-07-14 2021-10-22 厦门熵基科技有限公司 Metal classification method, device, equipment and storage medium
CN116189083A (en) * 2023-01-11 2023-05-30 广东汇通信息科技股份有限公司 Dangerous goods identification method for community security inspection assistance
CN116189083B (en) * 2023-01-11 2024-01-09 广东汇通信息科技股份有限公司 Dangerous goods identification method for community security inspection assistance
CN116071347A (en) * 2023-03-01 2023-05-05 山西戴德测控技术有限公司 Wear degree determination method, device, system and storage medium
CN116071347B (en) * 2023-03-01 2023-07-28 山西戴德测控技术股份有限公司 Wear degree determination method, device, system and storage medium

Also Published As

Publication number Publication date
CN109799544A (en) 2019-05-24
CN109799544B (en) 2021-03-19

Similar Documents

Publication Publication Date Title
WO2020134848A1 (en) Intelligent detection method and device applied to millimeter wave security check instrument, and storage device
EP3349050B1 (en) Inspection devices and methods for detecting a firearm
CN108956526B (en) Passive terahertz dangerous article detection device, detection method and application thereof
CN109325490B (en) Terahertz image target identification method based on deep learning and RPCA
López-Tapia et al. Using machine learning to detect and localize concealed objects in passive millimeter-wave images
US20160335503A1 (en) Inspection methods and devices
EP3696725A1 (en) Tool detection method and device
CN109492714A (en) Image processing apparatus and its method
Xu et al. YOLO-MSFG: toward real-time detection of concealed objects in passive terahertz images
Xiang et al. Adaptive superpixel generation for SAR images with linear feature clustering and edge constraint
US20220323030A1 (en) Probabilistic image analysis
Reichman et al. Improvements to the Histogram of Oriented Gradient (HOG) prescreener for buried threat detection in ground penetrating radar data
Liu et al. AC‐SDBSCAN: Toward concealed object detection of passive terahertz images
Yang et al. A novel deformable body partition model for MMW suspicious object detection and dynamic tracking
CN114419542A (en) Millimeter wave image detection optimal model selection method and device based on deep learning
Danso et al. An optimal defect recognition security-based terahertz low resolution image system using deep learning network
Su et al. Object recognition for millimeter wave MIMO-SAR images based on high-resolution feature recursive alignment fusion network
CN110751103B (en) Microwave signal classification method and system for detecting inside of object
Ahmed et al. Baggage threat detection under extreme class imbalance
Shafay et al. Programmable broad learning system to detect concealed and imbalanced baggage threats
CN116520315A (en) Target recognition system, target recognition method and target recognition device
CN113822361B (en) SAR image similarity measurement method and system based on Hamming distance
CN108229459A (en) A kind of method for tracking target
Samanta et al. Improved Breast Cancer Detection from Ultrasound Images Using YOLOv8 Model
Wen et al. AI‐Based W‐Band Suspicious Object Detection System for Moving Persons: Two‐Stage Walkthrough Configuration and Recognition Optimization

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19902743

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19902743

Country of ref document: EP

Kind code of ref document: A1