Model training method and device, suspected object detection method and device, and storage medium
By augmenting and fusion of vehicle backscattered images and training machine learning models, the problem of insufficient imaging clarity of perspective equipment in the prior art is solved, and accurate detection of suspected objects in the vehicle is achieved.
Patent Information
- Application Number
- CN202411957638.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-27
- Publication Date
- 2025-04-15
AI Technical Summary
Existing perspective equipment has the problem of insufficient clarity when imaging highly penetrating substances, especially when there are many types of goods and the placement is messy, the image features are difficult to identify, which increases the difficulty of reviewing pictures.
By collecting and processing the backscattered images of the vehicle, using the generative model to generate more backscattered images, combining the sample suspect images for data augmentation and fusion, and training machine learning models to identify the suspects carried in the vehicle.
Accurate identification of vehicle backscattered images is achieved, the detection accuracy of suspects in the car box is improved, and the difficulty of reviewing pictures is reduced.
Smart Images
Figure CN119832362A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of image processing, and in particular to a model training method and device, a suspect detection method and device, and a storage medium. Background Art
[0002] Security checks play a vital role in key areas such as counter-terrorism and combating drug trafficking and smuggling. The smuggling of dangerous goods such as knives, guns and even liquid explosives undoubtedly provides conditions for violent crimes (such as robbery, kidnapping, terrorist attacks, etc.), which directly threatens the personal and property safety of citizens.
[0003] Currently, each customs port uses X-rays to scan vehicles to obtain perspective images, and then identifies the perspective images to detect suspicious objects hidden in the vehicles. Summary of the invention
[0004] The inventors noticed that in the related art, the perspective image is identified to detect the suspicious objects hidden in the vehicle. However, the perspective device has the problem of insufficient clarity when imaging highly penetrating materials. When there are many types of goods and they are placed in a messy manner, the image features often become difficult to identify due to overlap and interference, which undoubtedly increases the difficulty of image review.
[0005] Accordingly, the present disclosure provides a model training method and a corresponding suspect detection method, which can identify the backscattered image of a vehicle by utilizing a trained machine learning model, thereby accurately identifying the suspects carried in the vehicle's trunk.
[0006] In a first aspect of the present disclosure, a model training method is provided, comprising: acquiring a first backscatter image of each sample vehicle in a plurality of sample vehicles; processing each original vehicle image in a plurality of original vehicle images using a generation model to obtain a second backscatter image corresponding to each original vehicle image; generating a set of images to be processed based on the plurality of first backscatter images and the plurality of second backscatter images; fusing a sample suspect image with each image to be processed in the set of images to be processed to generate a plurality of sample images; identifying each sample image in the plurality of sample images using a machine learning model; and training the machine learning model based on the identification result of each sample image and the sample suspect image.
[0007] In some embodiments, generating the set of images to be processed based on the multiple first backscatter images and the multiple second backscatter images includes: performing contrast stretching processing on each of the first backscatter images to generate multiple third backscatter images; and generating the set of images to be processed based on the multiple third backscatter images and the multiple second backscatter images.
[0008] In some embodiments, generating multiple sample images includes: performing an augmentation transformation on the sample suspect image to generate multiple sample augmented images; performing an augmentation transformation on the background data of each image to be processed to generate multiple background augmented data; fusing each sample augmented image in the multiple sample augmented images with each background augmented data in the multiple background augmented data to generate multiple fused images; and generating the multiple sample images based on the multiple fused images.
[0009] In some embodiments, generating the plurality of sample images according to the plurality of fused images comprises: performing an augmentation transform on each of the plurality of fused images to generate the plurality of sample images.
[0010] In some embodiments, the augmentation transformation includes at least one of the following operations: random grayscale transformation, perspective transformation, random cropping, random flipping, and random size transformation.
[0011] In a second aspect of the present disclosure, a model training device is provided, comprising: an acquisition module configured to acquire a first backscatter image of each sample vehicle in a plurality of sample vehicles; an image processing module configured to process each of a plurality of original vehicle images using a generation model to obtain a second backscatter image corresponding to each of the original vehicle images; a set generation module configured to generate a set of images to be processed based on a plurality of first backscatter images and a plurality of second backscatter images; an image processing module configured to fuse a sample suspect image with each of the images to be processed in the set of images to be processed to generate a plurality of sample images; and a training module configured to identify each of the plurality of sample images using a machine learning model, and train the machine learning model based on the identification result of each sample image and the sample suspect image.
[0012] In a third aspect of the present disclosure, a model training device is provided, comprising: a memory; a processor, coupled to the memory, the processor being configured to execute a model training method as described in any of the above embodiments based on instructions stored in the memory.
[0013] In a fourth aspect of the present disclosure, a method for detecting a suspected object is provided, comprising: obtaining a backscattered image to be detected of a detected vehicle; processing the backscattered image using a machine learning model to detect whether the detected vehicle contains a suspected object, wherein the machine learning model is trained using the model training method described in any of the above embodiments; and when the detected vehicle contains the suspected object, marking the suspected object in the backscattered image.
[0014] In some embodiments, acquiring the backscattered image to be detected of the detected vehicle includes: detecting the vehicle type of the detected vehicle; when the vehicle type of the detected vehicle is a specified vehicle type, acquiring the backscattered image of the detected vehicle; and, among the backscattered images of the detected vehicle, using the backscattered image of the area where the vehicle compartment of the detected vehicle is located as the backscattered image to be detected.
[0015] In a fifth aspect of the present disclosure, a suspicious object detection device is provided, comprising: an image acquisition module, configured to acquire a backscattered image to be detected of a detected vehicle; a detection module, configured to process the backscattered image using a machine learning model to detect whether there is a suspicious object in the detected vehicle, wherein the machine learning model is trained using the model training method described in any of the above embodiments; and a labeling module, configured to label the suspect in the backscattered image when the detected vehicle has the suspect.
[0016] In a sixth aspect of the present disclosure, a model training device is provided, comprising: a memory; a processor, coupled to the memory, the processor being configured to execute a model training method as described in any of the above embodiments based on instructions stored in the memory.
[0017] According to a seventh aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the image segmentation method as described in any of the above embodiments is implemented.
[0018] According to an eighth aspect of the embodiments of the present disclosure, there is provided a computer program product, comprising computer instructions, wherein when the computer instructions are executed by a processor, the image segmentation method as described in any of the above embodiments is implemented.
[0019] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0021] Figure 1 A flowchart of a model training method according to an embodiment of the present disclosure;
[0022] Figure 2A-2CA schematic diagram of image generation effect of some embodiments of the present disclosure;
[0023] Figure 3A-3B A schematic diagram of contrast stretching effect of some embodiments of the present disclosure;
[0024] Figures 4A-4C A schematic diagram of image fusion effects in some embodiments of the present disclosure;
[0025] Figure 5 A schematic diagram of the structure of a model training device according to an embodiment of the present disclosure;
[0026] Figure 6 It is a structural schematic diagram of a model training device according to another embodiment of the present disclosure;
[0027] Figure 7 A schematic diagram of a process of a method for detecting a suspected object according to an embodiment of the present disclosure;
[0028] Figure 8 This is a schematic diagram of the effect of suspect detection according to an embodiment of the present disclosure;
[0029] Fig. 9 This is a schematic diagram of the structure of a suspect detection device according to an embodiment of the present disclosure;
[0030] Fig.10 FIG. 4 is a schematic diagram of the structure of a suspect detection device according to another embodiment of the present disclosure. DETAILED DESCRIPTION
[0031] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present disclosure and its application or use. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.
[0032] Unless specifically stated otherwise, the relative arrangement of components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present disclosure.
[0033] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.
[0034] Technologies, methods, and apparatus known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, such technologies, methods, and apparatus should be considered part of the authorization specification.
[0035] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0036] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.
[0037] Figure 1 The following is a flow chart of a model training method according to an embodiment of the present disclosure. In some embodiments, the following model training method is performed by a model training device, including steps 11-16.
[0038] In step 11 , a first backscatter image of each sample vehicle among a plurality of sample vehicles is acquired.
[0039] It should be noted here that since different substances have different scattering characteristics, the structure and shape of the substance can be effectively understood by using the backscattered image.
[0040] In step 12, each of the multiple original vehicle images is processed using the generation model to obtain a second backscattered image corresponding to each original vehicle image.
[0041] It should be noted here that in order to obtain more image data for training, the original vehicle image can be processed using the existing generation model to obtain the corresponding backscatter image.
[0042] For example, the generative model may include a Pix2Pix model, a GAN (Generative Adversarial Network) model, a Stable Diffusion model, and the like.
[0043] For example, the original vehicle image may include a visible light image, an X-ray image, and the like.
[0044] By utilizing generative models, the diversity of data sets can be effectively enriched, and the problem of model training caused by insufficient data can be effectively solved.
[0045] For example, the perspective image of the car box is Figure 2A As shown, the real backscatter image of the car box is as follows Figure 2B As shown in Figure 2, the backscatter image generated using the perspective image of the vehicle model is as follows: Figure 2C shown.
[0046] By comparison Figure 2B and Figure 2CIt can be seen that by utilizing the generative model, high-quality backscatter images can be generated, thereby effectively enriching the diversity of the dataset.
[0047] In step 13, a set of images to be processed is generated according to the plurality of first backscatter images and the plurality of second backscatter images.
[0048] In some embodiments, each first backscatter image is subjected to contrast stretching processing to generate a plurality of third backscatter images. Next, a set of images to be processed is generated based on the plurality of third backscatter images and the plurality of second backscatter images.
[0049] It should be noted here that by performing contrast stretching on the backscattered image, the pixel values originally concentrated in a certain grayscale range can be more evenly distributed over the entire grayscale, so that both the dark and bright areas in the backscattered image can be presented more delicately, effectively enhancing the contrast and clarity of the backscattered image.
[0050] For example, contrast stretching processing may be performed using linear transformation, log transformation, or a combination of the two according to the grayscale value range of the image to improve the separability of image extraction features.
[0051] For example, the backscatter image collected is Figure 3A As shown. Figure 3A The backscatter image shown in the figure is subjected to contrast stretching, and the result is as follows Figure 3B shown.
[0052] By comparison Figure 3A and Figure 3B It can be clearly seen that both the dark and bright areas in the backscattered image can be presented more delicately, thereby effectively enhancing the contrast and clarity of the backscattered image.
[0053] In step 14, the sample suspect image is fused with each image to be processed in the set of images to be processed to generate a plurality of sample images.
[0054] In some embodiments, the sample suspect image is augmented and transformed to generate multiple sample augmented images. The background data of each image to be processed is augmented and transformed to generate multiple background augmented data. Each sample augmented image in the multiple sample augmented images is fused with each background augmented data in the multiple background augmented data to generate multiple fused images. Next, multiple sample images are generated based on the multiple fused images.
[0055] For example, a sample suspect image such as Figure 4A As shown, the image to be processed is Figure 4B As shown. By comparing the sample suspect image shown in Figure A with Figure 4B The background data of the image to be processed is fused to generate the following Figure 4C The fused image shown.
[0056] It should be noted here that in order to improve the naturalness of image fusion, we perform convolution operations on the foreground image (i.e., the sample suspect image) to achieve the effect of edge grayscale value transition, so that the foreground and background are more naturally connected at the fusion edge. In addition, the brightness of the foreground image is adjusted according to the brightness of the background to ensure the authenticity of the brightness relationship between the foreground and background. At the same time, a flexible dynamic range is set for the grayscale value of the foreground object to simulate the brightness changes that suspects may present in real environments.
[0057] It should also be noted that the obtained fused image can be directly used as a sample image, or the obtained fused image can be further processed and the processed fused image can be used as a sample image.
[0058] In some embodiments, each of the multiple fused images is augmented and transformed to generate multiple sample images.
[0059] For example, the augmentation transformation includes at least one of the following operations: random grayscale transformation, perspective transformation, random cropping, random flipping, and random size transformation.
[0060] It should be noted here that in order to further improve the richness of the data, the inserted sample suspect images can first be subjected to random flipping, random resizing operations, and random grayscale value mean adjustment. These operations are intended to simulate the various postures, sizes, and brightness changes that may occur in the suspects during the actual scanning process, thereby increasing the robustness of the algorithm.
[0061] Subsequently, the background image is also subjected to random grayscale mean adjustment, random flipping and other operations to simulate the changes that may occur in the actual scene. The foreground and background images are then fused.
[0062] Finally, the generated images are randomly transformed, including random grayscale transformation, perspective transformation, random cropping, random flipping, and random size transformation. These transformations not only further enrich the diversity of the image dataset, but also simulate the image features under different scanning angles and directions, providing more comprehensive and realistic image data for the training of the recognition algorithm.
[0063] Through the above series of data generation, fusion, and augmentation operations, we finally built an image database containing multiple scanning angles and multiple directions.
[0064] In step 15, each sample image in the plurality of sample images is identified using a machine learning model.
[0065] For example, machine learning models include the Faster R-CNN (Faster Region-convolutional neuralnetwork) model.
[0066] Faster R-CNN is a typical two-stage target detection framework, which mainly includes two stages, namely RPN (Region Proposal Networks) and RoI Pooling (Region of Interest Pooling). The RPN network introduces multi-scale anchors into the target detection task, replacing the computationally complex sliding window method to generate multiple candidate regions. This step will generate a bunch of anchor boxes, one of which is to crop and filter the anchor box and then use the softmax function to determine whether the anchors belong to the foreground or background, that is, whether they are objects or not. At the same time, the other branch corrects the anchor box through box regression to form a more accurate candidate result proposal. The goal of RoI Pooling is to reduce the dimensionality of the features of the region proposals output by the RPN layer. This is because the region proposal features output by the RPN layer are of different sizes, and the features finally output to the fully connected layer must be of uniform dimension, so RoI Pooling is used to unify the feature scale. Finally, the network uses shared convolution to optimize the categories and positions of multiple targets. This is also divided into two branches. One branch uses the softmax function to determine whether the output of the fully connected layer belongs to the target category (class) or the background (background), that is, whether it is a suspect. At the same time, the other branch corrects the anchor box through box regression to form the final accurate target position prediction.
[0067] In step 16, the machine learning model is trained based on the recognition results of each sample image and the sample suspect image.
[0068] For example, the PyTorch framework is used to train machine learning models.
[0069] In the model training method provided in the above embodiment of the present disclosure, by using sample images from different channels for training, the number and differentiation of samples can be effectively increased, thereby improving the model training effect.
[0070] Figure 5The present invention is a schematic diagram of the structure of a model training device according to an embodiment of the present invention.
[0071] like Figure 5 As shown, the model training device includes an acquisition module 51, an image processing module 52, a set generation module 53, an image processing module 54 and a training module 55.
[0072] The acquisition module 51 is configured to acquire a first backscatter image of each sample vehicle among a plurality of sample vehicles.
[0073] The image processing module 52 is configured to process each of the multiple original vehicle images using the generation model to obtain a second backscatter image corresponding to each original vehicle image.
[0074] It should be noted here that in order to obtain more image data for training, the original vehicle image can be processed using the existing generation model to obtain the corresponding backscatter image.
[0075] For example, the generation model may include a Pix2Pix model, a GAN model, a Stable Diffusion model, and the like.
[0076] For example, the original vehicle image may include a visible light image, an X-ray image, and the like.
[0077] By utilizing generative models, the diversity of data sets can be effectively enriched, and the problem of model training caused by insufficient data can be effectively solved.
[0078] The set generating module 53 is configured to generate a set of images to be processed according to the plurality of first backscatter images and the plurality of second backscatter images.
[0079] In some embodiments, the set generation module 53 performs contrast stretching processing on each first backscatter image to generate a plurality of third backscatter images. Next, a set of images to be processed is generated according to the plurality of third backscatter images and the plurality of second backscatter images.
[0080] It should be noted here that by performing contrast stretching on the backscattered image, the pixel values originally concentrated in a certain grayscale range can be more evenly distributed over the entire grayscale, so that both the dark and bright areas in the backscattered image can be presented more delicately, effectively enhancing the contrast and clarity of the backscattered image.
[0081] The image processing module 64 is configured to fuse the sample suspect image with each image to be processed in the set of images to be processed to generate a plurality of sample images.
[0082] In some embodiments, the image processing module 64 performs augmentation transformation on the sample suspect image to generate multiple sample augmented images, performs augmentation transformation on the background data of each image to be processed to generate multiple background augmented data, and fuses each sample augmented image in the multiple sample augmented images with each background augmented data in the multiple background augmented data to generate multiple fused images. Next, multiple sample images are generated based on the multiple fused images.
[0083] It should also be noted that the obtained fused image can be directly used as a sample image, or the obtained fused image can be further processed and the processed fused image can be used as a sample image.
[0084] In some embodiments, each of the multiple fused images is augmented and transformed to generate multiple sample images.
[0085] For example, the augmentation transformation includes at least one of the following operations: random grayscale transformation, perspective transformation, random cropping, random flipping, and random size transformation.
[0086] The training module 65 is configured to use a machine learning model to identify each sample image among the multiple sample images, and train the machine learning model according to the identification result of each sample image and the sample suspect image.
[0087] Figure 6 It is a structural schematic diagram of a model training device according to another embodiment of the present disclosure.
[0088] like Figure 6 As shown in FIG. 6 , the model training device 60 is in the form of a general-purpose computing device. The model training device 60 includes a memory 61, a processor 62, and a bus 63 connecting different system components.
[0089] The memory 61 may include, for example, a system memory, a non-volatile storage medium, and the like. The system memory may store, for example, an operating system, an application, a boot loader, and other programs. The system memory may include a volatile storage medium, such as a random access memory (RAM) and / or a cache memory. The non-volatile storage medium may store, for example, instructions for a corresponding embodiment of at least one model training method being executed. The non-volatile storage medium includes, but is not limited to, a disk storage, an optical storage, a flash memory, and the like.
[0090] The processor 62 may be implemented by a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistors, and other discrete hardware components. Accordingly, each module such as an acquisition module, a calculation module, and an adjustment module may be implemented by a central processing unit (CPU) running instructions in a memory that execute corresponding steps, or may be implemented by a dedicated circuit that executes corresponding steps.
[0091] For example, the processor 62 is configured to execute instructions stored in the memory to implement the following Figure 1 The method of any one of the embodiments.
[0092] The bus 63 may use any of a variety of bus structures, including, but not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, and a Peripheral Component Interconnect (PCI) bus.
[0093] The interfaces 64, 65, 66 of the model training device 60 and the memory 61 and the processor 62 can be connected via a bus 63. The input and output interface 64 can provide a connection interface for input and output devices such as a display, a mouse, and a keyboard. The network interface 65 provides a connection interface for various networked devices. The storage interface 66 provides a connection interface for external storage devices such as a floppy disk, a USB flash drive, and an SD card.
[0094] Here, various aspects of the present disclosure are described with reference to flowcharts and / or block diagrams of methods, devices, and computer program products according to embodiments of the present disclosure. It should be understood that each frame of the flowchart and / or block diagram and the combination of frames can be implemented by computer-readable program instructions.
[0095] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable device to produce a machine, so that the processor executes the instructions to produce means for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.
[0096] These computer-readable program instructions may also be stored in a computer-readable memory, which cause the computer to work in a specific manner to produce an article of manufacture, including instructions for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.
[0097] The present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.
[0098] The present disclosure also provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the following is implemented: Figure 1The method of any one of the embodiments.
[0099] The present disclosure also provides a computer program product, including computer instructions, wherein when the computer instructions are executed by a processor, the following is achieved: Figure 1 The method of any one of the embodiments.
[0100] Figure 7 FIG. 7 is a flow chart of a suspect detection method according to an embodiment of the present disclosure. In some embodiments, the following suspect detection method is performed by a suspect detection device, including steps 71-73.
[0101] In step 71, a backscatter image to be detected of the detected vehicle is acquired.
[0102] In some embodiments, the step of acquiring the backscattered image to be detected of the detected vehicle includes the following contents.
[0103] 1) Detect the vehicle type of the detected vehicle.
[0104] For example, the solution involved in the present disclosure is only applied to container trucks and vans, so other vehicles will not be processed accordingly, which can effectively save computing resources and speed up the processing process.
[0105] 2) When the vehicle type of the detected vehicle is a specified vehicle type, a backscatter image of the detected vehicle is collected.
[0106] 3) In the backscatter image of the detected vehicle, the backscatter image of the area where the vehicle compartment of the detected vehicle is located is used as the backscatter image to be detected.
[0107] Through the above processing, only the image of the car box area can be processed, which can effectively avoid interference from information outside the car box. In addition, it also effectively improves the resolution of the image input into the machine learning model. Higher resolution means further enhancement of the clarity and details of the suspect, thereby effectively improving the accuracy of model detection.
[0108] In step 72, the backscatter image is processed using a machine learning model to detect whether there is a suspicious object in the detected vehicle.
[0109] It should be noted here that the machine learning model used here utilizes Figure 1 The model can be trained using any of the following model training methods.
[0110] In step 73 , when there is a suspected object in the detected vehicle, the suspected object is marked in the backscatter image.
[0111] For example, Figure 8As shown, the machine learning model processes the backscattered image of the car box and finds that there are three guns in the car box, and then marks these three guns in the image.
[0112] In some embodiments, the solution provided by the above embodiments of the present disclosure is deployed using a torch serving framework.
[0113] Fig. 9 The figure is a schematic diagram of the structure of a suspect detection device according to an embodiment of the present disclosure.
[0114] like Fig. 9 As shown, the suspicious object detection device includes an image acquisition module 91 , a detection module 92 and a labeling module 93 .
[0115] The image acquisition module 91 is configured to acquire a backscattered image to be detected of the detected vehicle.
[0116] In some embodiments, the image acquisition module 91 performs the following operations.
[0117] 1) Detect the vehicle type of the detected vehicle.
[0118] For example, the solution involved in the present disclosure is only applied to container trucks and vans, so other vehicles will not be processed accordingly, which can effectively save computing resources and speed up the processing process.
[0119] 2) When the vehicle type of the detected vehicle is a specified vehicle type, a backscatter image of the detected vehicle is collected.
[0120] 3) In the backscatter image of the detected vehicle, the backscatter image of the area where the vehicle compartment of the detected vehicle is located is used as the backscatter image to be detected.
[0121] The detection module 92 is configured to process the backscatter image using a machine learning model to detect whether there is a suspicious object in the detected vehicle.
[0122] It should be noted here that the machine learning model used here utilizes Figure 1 The model can be trained using any of the following model training methods.
[0123] The marking module 93 is configured to mark the suspected object in the backscatter image when there is a suspected object in the detected vehicle.
[0124] Fig.10 FIG. 4 is a schematic diagram of the structure of a suspect detection device according to another embodiment of the present disclosure.
[0125] like Fig.10As shown, the suspect detection device 100 includes a memory 101 , a processor 102 , a bus 103 , an input / output interface 104 , a network interface 105 , and a storage interface 106 . Fig.10 and Figure 6 The difference is that in Fig.10 In the illustrated embodiment, the processor 102 is configured to execute instructions stored in the memory to implement the following Figure 7 The method of any one of the embodiments.
[0126] The present disclosure also provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the following is implemented: Figure 7 The method of any one of the embodiments.
[0127] The present disclosure also provides a computer program product, including computer instructions, wherein when the computer instructions are executed by a processor, the following is achieved: Figure 7 The method of any one of the embodiments.
[0128] In some embodiments, the functional unit described above may be implemented as a general-purpose processor, a programmable logic controller (PLC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any appropriate combination thereof, for performing the functions described in the present disclosure.
[0129] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.
[0130] The description of the present disclosure is given for the purpose of illustration and description, and is not intended to be exhaustive or to limit the present disclosure to the disclosed form. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present disclosure, and to enable those of ordinary skill in the art to understand the present disclosure and thereby design various embodiments with various modifications suitable for specific uses.
Claims
1. A model training method, comprising: acquiring a first backscattered image of each sample vehicle among a plurality of sample vehicles; Processing each of the plurality of original vehicle images using the generative model to obtain a second backscattered image corresponding to each of the original vehicle images; generating a set of images to be processed according to the plurality of first backscatter images and the plurality of second backscatter images; Fusion of the sample suspect image with each image to be processed in the set of images to be processed to generate a plurality of sample images; Using a machine learning model to identify each sample image in the plurality of sample images; The machine learning model is trained according to the recognition result of each sample image and the sample suspect image.
2. The method according to claim 1, wherein: The step of generating a set of images to be processed according to the plurality of first backscatter images and the plurality of second backscatter images comprises: Performing contrast stretching processing on each of the first backscattered images to generate a plurality of third backscattered images; The to-be-processed image set is generated according to the plurality of third backscatter images and the plurality of second backscatter images.
3. The method according to claim 1, wherein: Generating a plurality of sample images comprises: Performing an augmentation transformation on the sample suspect image to generate a plurality of sample augmented images; Performing an augmentation transformation on the background data of each image to be processed to generate a plurality of background augmented data; Fusing each sample augmented image in the plurality of sample augmented images with each background augmented data in the plurality of background augmented data to generate a plurality of fused images; The plurality of sample images are generated according to the plurality of fused images.
4. The method according to claim 3, wherein: Generating the plurality of sample images according to the plurality of fused images comprises: Performing an augmentation transform on each of the multiple fused images to generate the multiple sample images.
5. The method according to claim 4, wherein: The augmentation transformation includes at least one of the following operations: random grayscale transformation, perspective transformation, random cropping, random flipping and random size transformation.
6. A model training device, comprising: An acquisition module, configured to acquire a first backscattered image of each sample vehicle among a plurality of sample vehicles; An image processing module is configured to process each of the plurality of original vehicle images using a generation model to obtain a second backscattered image corresponding to each of the original vehicle images; A set generating module is configured to generate a to-be-processed image set according to the plurality of first backscatter images and the plurality of second backscatter images; An image processing module, configured to fuse the sample suspect image with each image to be processed in the set of images to be processed to generate a plurality of sample images; The training module is configured to use a machine learning model to identify each sample image in the multiple sample images, and train the machine learning model according to the recognition result of each sample image and the sample suspect image.
7. A model training device, comprising: Memory; A processor is coupled to the memory, and the processor is configured to execute the method according to any one of claims 1 to 5 based on instructions stored in the memory.
8. A method for detecting a suspect, comprising: Acquire a backscattered image to be detected of the detected vehicle; Processing the backscattered image using a machine learning model to detect whether there is a suspicious object in the detected vehicle, wherein the machine learning model is trained using the model training method described in any one of claims 1 to 5; When the detected vehicle contains the suspected object, the suspected object is marked in the backscattered image.
9. The method according to claim 8, wherein: The step of acquiring the backscattered image of the detected vehicle to be detected comprises: Detecting the vehicle type of the detected vehicle; When the vehicle type of the detected vehicle is a specified vehicle type, collecting a backscattered image of the detected vehicle; In the backscatter image of the detected vehicle, the backscatter image of the area where the vehicle compartment of the detected vehicle is located is used as the backscatter image to be detected.
10. A suspicious object detection device, comprising: An image acquisition module is configured to acquire a backscattered image to be detected of a detected vehicle; a detection module configured to process the backscatter image using a machine learning model to detect whether there is a suspect in the detected vehicle, wherein the machine learning model is trained using the model training method according to any one of claims 1 to 5; The marking module is configured to mark the suspect in the backscatter image when the detected vehicle has the suspect.
11. A model training device, comprising: Memory; A processor is coupled to the memory, and the processor is configured to execute the method according to any one of claims 8 to 9 based on instructions stored in the memory.
12. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method according to any one of claims 1-5, 8-9 is implemented.
13. A computer program product, comprising computer instructions, wherein when the computer instructions are executed by a processor, the method according to any one of claims 1-5, 8-9 is implemented.