Comprehensive liquid qualitative method, X-ray security inspection equipment and device

Abstract

The invention relates to the technical field of security check, and discloses a comprehensive liquid qualitative method and X-ray security check equipment and device.The method comprises the steps that a transmission image and a back scattering image of a to-be-checked object are obtained, and the coordinate association relation of the same associated container in the transmission image and the back scattering image is established; respectively performing liquid segmentation on the transmission image and the back scattering image to correspondingly obtain an unshielded liquid contour and a liquid contour, and extracting original data of a liquid area and original data of a container shell area; based on the liquid area original data and the container shell area original data, transmission liquid features, back scattering liquid features, transmission container features and back scattering container features are obtained through calculation, a liquid feature matrix is constructed through fusion, and a liquid attribute classification result is obtained through a liquid attribute judgment model. According to the method, the liquid characteristics in the transmission and back scattering data and the container characteristics of liquid superposition are comprehensively extracted, the attribute discrimination of the target organic liquid is completed, and the substance qualitative accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of security inspection technology, specifically to a comprehensive liquid qualitative method, and X-ray security inspection equipment and apparatus. Background Technology

[0002] Currently, baggage inspection uses X-ray transmission technology. This involves scanning and acquiring high-energy and low-energy images of the object, calculating its effective atomic number, and using pseudo-color to classify substances into three categories: orange for organic matter, green for mixtures and light metals, and blue for inorganic matter. However, because items in security checks are often in a superimposed state, the boundaries between mixtures / light metals and organic or inorganic matter are relatively blurred. This results in a mixture of green and orange or blue colors when organic and inorganic matter are superimposed, making it difficult to accurately analyze the composition of the mixture. When the organic matter is too thick, the colors also appear intertwined, limiting the detection accuracy for non-overlapping organic matter. For conventional security scanners, X-ray transmission is insufficient for detailed analysis of organic matter; while security CT equipment can calculate the density distribution of matter, it still cannot accurately determine the material properties, and its slow speed makes it unsuitable for rapid security checks. Security identification determines the category of the object based on its shape and color, but it cannot determine the material properties.

[0003] Therefore, existing transmission security inspection technology can only identify the shape of the object being inspected, but cannot effectively identify organic substances such as liquids, explosives, and drugs. This results in problems such as the inability to qualitatively identify items, insufficient identification accuracy, and difficulty in judging the properties of substances. Summary of the Invention

[0004] This invention provides a comprehensive liquid qualitative method, X-ray security inspection equipment and device to solve problems such as the inability to qualitatively identify items, insufficient identification accuracy, and difficulty in judging the properties of substances.

[0005] In a first aspect, the present invention provides a comprehensive liquid qualitative method, the method comprising: Acquire the transmission and backscatter images of the item to be inspected, and establish the coordinate relationship between the same associated container in the transmission and backscatter images based on the transmission and backscatter images; Liquid segmentation is performed on the transmission image and backscattered image respectively to obtain the unobstructed liquid contour and the liquid contour. Based on the unobstructed liquid contour and the liquid contour, the original data of the liquid region and the original data of the container shell region are extracted. Based on the original data of the liquid region and the original data of the container shell region, the transmitted liquid features, backscattered liquid features, transmitted container features and backscattered container features are calculated and fused to construct a liquid feature matrix; The liquid feature matrix is ​​input into a preset liquid property discrimination model to obtain the liquid property classification result.

[0006] This invention provides a comprehensive liquid characterization method that integrates X-ray transmission and backscatter dual-modal imaging data to construct a complete processing flow from image registration, container association, liquid segmentation to multi-feature extraction and modeling. This achieves high-precision identification of liquid properties within a container by comprehensively extracting liquid features from transmission and backscatter data, as well as container features superimposed on the liquid, to identify the properties of the target organic liquid. This improves the accuracy of substance characterization and effectively overcomes the limitations of traditional single transmission technology, which has low resolution for organic matter and cannot distinguish superimposed substances. It also enhances the automatic identification capability and reliability of hazardous liquids in security inspection scenarios.

[0007] In one optional implementation, establishing a coordinate association relationship between the same associated container in the transmitted image and the backscattered image based on the transmitted image and the backscattered image includes: Spatial registration is performed between the transmitted image and the backscattered image to obtain the registered coordinates. Container detection and segmentation are performed on the transmitted image and the backscattered image respectively to identify the container region and obtain the container coordinates of the transmitted image and the container bounding box of the backscattered image; the container region includes the container outline and the bounding box. The coordinates of the container in the transmission image are mapped to the backscatter coordinate system using the registered coordinates to generate the transmission container mapping box. Based on the transmission container mapping box and the backscatter container result box, the coordinate relationship between the same associated container in the transmission image and the backscatter image is established.

[0008] This invention provides a comprehensive liquid qualitative method that achieves precise positional association of the same container in transmission and backscatter images through spatial registration, dual-modal container detection, and coordinate mapping. This ensures that subsequent liquid segmentation and feature extraction are based on the same target object, thereby improving the accuracy and reliability of multimodal data fusion from the source and laying a solid spatial correspondence foundation for the accurate identification of liquid properties.

[0009] In one optional implementation, spatial registration is performed between the transmitted image and the backscattered image to obtain registered coordinates, including: Using preset translation and scaling factors, the coordinates in the transmitted image are mapped to the coordinates in the backscattered image. The registered coordinate formula is expressed as: ; ; in, , It is the translation coefficient that maps the transmitted image to the backscattered image. , It is the scaling factor for mapping the transmitted image to the backscattered image. This represents the coordinates of the transmission point in the transmission image.

[0010] In one optional implementation, a fast coordinate mapping is achieved by using preset linear transformation parameters. Spatial alignment between the two modal images is completed through simple and efficient calculations. While ensuring registration accuracy, the computational complexity is significantly reduced, providing a stable and reliable geometric correspondence basis for subsequent container association and feature extraction, thereby supporting the real-time performance and accuracy of the overall discrimination process.

[0011] In one optional implementation, establishing a coordinate association between the same associated container in the transmission image and the backscattered image based on the transmission container mapping frame and the backscattered container result frame includes: Calculate the cross-union ratio (CUB) and the full cross-union ratio (MCR) between the transmission container mapping frame and the backscatter container result frame. If both the CUB and MCR are greater than a preset threshold, they are determined to be the same container, and a pair of correlations is established between the same container in the transmission image coordinate system and the backscatter container coordinate system.

[0012] The present invention provides a comprehensive liquid qualitative method that uses a dual-index judgment mechanism that simultaneously calculates the cross-union ratio and the complete cross-union ratio, combined with a preset threshold, to achieve strict screening of container associations. While ensuring matching accuracy, it effectively avoids erroneous associations that may be caused by a single overlap metric, thus providing a stable and reliable data alignment foundation for subsequent liquid feature fusion and discrimination based on the same target.

[0013] In one optional implementation, liquid segmentation is performed on the transmitted image and the backscattered image respectively to obtain the unobstructed liquid contour and the liquid outline, and the original data of the liquid region and the original data of the container shell region are extracted based on the unobstructed liquid contour and the liquid outline, including: Liquid segmentation is performed on containers in the transmission image, the outline of the unobstructed liquid is extracted, and it is matched with the associated containers; Liquid segmentation is performed on containers in the backscattered image, liquid contours are extracted, and they are matched with associated containers; The corresponding raw data of the liquid region is extracted from the pre-acquired raw transmission data and raw backscatter data, and the raw data of the container shell region is extracted from the raw backscatter data.

[0014] The present invention provides a comprehensive liquid qualitative method that extracts the unobstructed liquid contour from the transmission image and the high-sensitivity liquid contour from the backscatter image, and performs precise matching based on the established container association. This achieves effective segmentation and data alignment of the dual-modal liquid region, thereby providing an accurate and consistent numerical basis for subsequent fusion of the original transmission and backscatter data and extraction of highly discriminative statistical features.

[0015] In one optional embodiment, the characteristics of the transmitted liquid include: a first mean, a first standard deviation, a first quartile Q1, a first quartile Q3, a first skewness, and a first coefficient of variation; the characteristic of the transmitted container is the ratio of the transmitted liquid to the container. Backscattering liquid characteristics include the second mean, second standard deviation, second quartile Q1', second quartile Q3', second skewness, and second coefficient of variation; backscattering container characteristics include the container mean and the percentage of backscattering liquid in the container.

[0016] This invention provides a comprehensive qualitative method for liquids. The feature extraction scheme systematically combines multi-dimensional statistics such as mean, standard deviation, quartiles, skewness, and coefficient of variation to comprehensively characterize the distribution, numerical, and morphological properties of liquids and containers in dual-modal imaging. At the same time, it introduces related features such as the ratio of liquid to container and the container mean to effectively characterize the relationship between the liquid and the carrying environment. This provides a rich, complementary, and physically meaningful input feature set for subsequent discrimination models, significantly improving the accuracy and robustness of liquid attribute classification.

[0017] In one alternative implementation, the method further includes: Based on the characteristics of transmitted liquid, backscattered liquid, transmitted container, and backscattered container, an occlusion feature matrix is ​​constructed. The occlusion feature matrix is ​​then input into a preset occlusion discrimination model to obtain the liquid occlusion state determination result.

[0018] This invention provides a comprehensive liquid qualitative method that reuses extracted bimodal liquid and container statistical features to construct a targeted occlusion feature matrix and input it into a dedicated discrimination model. This enables independent and specialized judgment of liquid occlusion status, significantly enhancing the overall system's adaptability and recognition reliability in complex and overlapping security inspection scenarios without the need for additional data collection and processing.

[0019] In one optional implementation, the transmission image includes a high-energy transmission image and a low-energy transmission image; the method further includes: A pseudo-color image is calculated based on the high-energy transmission image and the low-energy transmission image; the pseudo-color image includes three channels: R, G, and B. Transmission high-energy images, transmission low-energy images, and transmission false color images Figure 3 Each channel is fused with the registered and mapped backscatter image to generate multi-channel fused image data; Multi-channel fused data is input into a preset classification model to simultaneously obtain classification results for liquid properties and occlusion states. These classification results are used to supplement or verify the liquid property classification results and liquid occlusion state determination results.

[0020] This invention provides a comprehensive liquid qualitative method that constructs multi-channel fused data containing material attenuation characteristics, atomic number classification information, and surface scattering features by fusing raw transmission high / low energy data, pseudo-color image decomposition channels derived from them, and registered backscattered images. Furthermore, it achieves synchronous and integrated discrimination of liquid properties and occlusion states by using a multi-task classification model, thereby enhancing the richness of feature representation while improving the overall synergy, efficiency, and reliability of the system's judgment.

[0021] In a second aspect, the present invention provides an X-ray security inspection device, which includes: an X-ray transmission imaging module, a Compton backscatter imaging module, an image feature extraction module, and a first liquid property discrimination module; X-ray transmission imaging module, used to acquire transmission images of the item to be inspected; Compton backscatter imaging module, used to obtain backscattered images of the object under inspection; The image feature extraction module is used to perform the steps described in the first aspect above: acquiring the transmission image and backscatter image of the item to be inspected, and establishing the coordinate relationship between the same associated container in the transmission image and backscatter image based on the transmission image and backscatter image; performing liquid segmentation on the transmission image and backscatter image respectively to obtain the unobstructed liquid contour and liquid outline, and extracting the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and liquid outline; and calculating the transmission liquid features, backscatter liquid features, transmission container features, and backscatter container features based on the original data of the liquid region and the original data of the container shell region, and fusing them to construct a liquid feature matrix. The first liquid property discrimination module is used to perform the step in the first aspect above of inputting the liquid feature matrix into the preset liquid property discrimination model to obtain the liquid property classification result.

[0022] Thirdly, the present invention provides a comprehensive liquid qualitative analysis apparatus, the apparatus comprising: The module for establishing the correlation between transmission images and backscattered images is used to acquire transmission images and backscattered images of the item to be inspected, and to establish the coordinate correlation between the same associated container in the transmission images and backscattered images based on the transmission images and backscattered images. The liquid contour and container shell extraction module is used to perform liquid segmentation on the transmission image and backscatter image respectively, to obtain the unobstructed liquid contour and liquid outline, and to extract the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and liquid outline. The liquid feature matrix construction module is used to calculate the transmitted liquid features, backscattered liquid features, transmitted container features, and backscattered container features based on the original data of the liquid region and the original data of the container shell region, and then fuse them to construct the liquid feature matrix. The second liquid property discrimination module is used to input the liquid feature matrix into the preset liquid property discrimination model to obtain the liquid property classification result.

[0023] Fourthly, the present invention provides an electronic device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the comprehensive liquid qualitative method of the first aspect or any corresponding embodiment described above.

[0024] Fifthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the comprehensive liquid qualitative method of the first aspect or any corresponding embodiment described above.

[0025] In a sixth aspect, the present invention provides a computer program product, including computer instructions for causing a computer to execute the comprehensive liquid qualitative method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the first process of the comprehensive liquid qualitative method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a second process for a comprehensive liquid qualitative method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the third process of the comprehensive liquid qualitative method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the fourth process of the comprehensive liquid qualitative method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the statistical liquid property classification results based on the comprehensive liquid qualitative method according to an embodiment of the present invention; Figure 6 This is a structural block diagram of an X-ray security inspection device according to an embodiment of the present invention; Figure 7 This is a structural block diagram of the integrated liquid qualitative apparatus according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0030] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0031] Currently, the security inspection industry can only identify the shape of items, but cannot identify their qualitative characteristics. This results in insufficient identification accuracy and difficulty in judging the properties of materials.

[0032] Therefore, this invention provides a comprehensive liquid characterization method that comprehensively extracts liquid features from transmission and backscatter data, as well as container features superimposed on the liquid, to complete the property identification of the target organic liquid and improve the accuracy of substance characterization.

[0033] According to an embodiment of the present invention, a comprehensive liquid qualitative method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0034] This embodiment provides a comprehensive liquid qualitative method that can be used in the aforementioned electronic device. Figure 1 This is a flowchart of a comprehensive liquid qualitative method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the transmission image and backscatter image of the item to be inspected, and establish the coordinate relationship between the same associated container in the transmission image and backscatter image based on the transmission image and backscatter image.

[0035] Specifically, items to be inspected generally refer to luggage, bags, parcels, and other containers that need to be inspected through security checkpoints.

[0036] Transmission images: The X-ray source and detector are located on opposite sides of the object. After the X-rays penetrate the object, the detector generates an image based on the intensity (attenuation) of the received X-rays. Transmission images include: High-energy transmission images: Images formed after X-rays of higher energy penetrate the object, offering stronger penetration for different materials. Low-energy transmission images: Images formed after X-rays of lower energy penetrate the object, more sensitive to differences in materials. Transmission pseudo-color images: A type of derivative image. By processing the attenuation data of high- and low-energy images using a computer, the "effective atomic number" of the scanned object is calculated, and pseudo-color encoding is performed according to preset rules (e.g., organic matter - orange, mixtures - green, inorganic matter - blue). This generated image is used for rapid visual differentiation of major material categories.

[0037] Backscattered image: The X-ray source and detector are located on the same side of the object. The detector is specifically designed to receive X-ray photons (mainly Compton scattered photons) scattered (reflected) back by the material inside the object, and to generate an image based on these photons.

[0038] Simultaneously acquire X-ray transmission images and Compton backscatter images of the item to be inspected; through spatial registration technology, map and associate the regions in the two images that correspond to the same physical container, establish a unified spatial correspondence of the container under the "transmission-backscatter" dual-modal view, and provide an accurate geometric benchmark for subsequent fusion analysis.

[0039] Step S102: Perform liquid segmentation on the transmission image and backscattered image respectively to obtain the unobstructed liquid contour and the liquid contour, and extract the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and the liquid contour.

[0040] Specifically, based on the associated container region, liquid target segmentation is performed on the transmission image and backscatter image respectively: in the transmission image, the liquid contour that is not obscured by other objects is extracted, while in the backscatter image, the complete liquid contour is directly extracted by taking advantage of its high sensitivity to organic matter; then, based on the segmented contour, the original numerical information of the liquid region and the container shell region is extracted from the original transmission and backscatter data respectively, so as to complete the location and acquisition of key data.

[0041] Step S103: Based on the original data of the liquid region and the original data of the container shell region, calculate the transmitted liquid features, backscattered liquid features, transmitted container features and backscattered container features, and fuse them to construct a liquid feature matrix.

[0042] Statistical analysis is performed on the extracted raw data to calculate the characteristics of the transmitted liquid (such as mean and standard deviation), the characteristics of the backscattered liquid, the characteristics of the transmitted container (such as the liquid area ratio of the container), and the characteristics of the backscattered container (such as the container mean). These features obtained from different physical principles and imaging perspectives are fused to construct a comprehensive liquid feature matrix, thereby comprehensively and deeply characterizing the properties of the target liquid.

[0043] Step S104: Input the liquid feature matrix into the preset liquid attribute discrimination model to obtain the liquid attribute classification result.

[0044] The constructed liquid feature matrix is ​​input into a pre-trained liquid attribute discrimination model (such as logistic regression, random forest, LightGBM, etc.); the model learns and compares feature patterns, and outputs the attribute classification result of the liquid (such as the specific liquid type and its probability), ultimately achieving automated and highly accurate qualitative identification of liquids.

[0045] The comprehensive liquid qualitative method provided in this embodiment integrates X-ray transmission and Compton backscattering techniques to complete material scanning imaging; machine vision technology to complete target detection and semantic segmentation recognition based on the global receptive field; and image registration technology to complete spatial registration of transmission images (high-energy and low-energy) and backscattered images. By combining the above raw data, the liquid is segmented and its spatial information is obtained. The container, transmission properties, and backscattering properties of the liquid are extracted, and the model is trained to complete the qualitative prediction of the liquid. This improves the accuracy of material qualitative analysis and effectively overcomes the limitations of traditional single transmission technology, such as low resolution for organic matter and inability to distinguish superimposed substances. It enhances the automatic identification capability and reliability of liquid hazardous materials in security inspection scenarios.

[0046] This embodiment provides a comprehensive liquid qualitative method, which can be used in the aforementioned electronic devices. Figure 2 This is a flowchart of a comprehensive liquid qualitative method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the transmission image and backscatter image of the item to be inspected, and establish the coordinate relationship between the same associated container in the transmission image and backscatter image based on the transmission image and backscatter image.

[0047] Specifically, the container outline and bounding box are identified by extracting the transmission image and backscatter image. The backscatter coordinates corresponding to each transmission point are calculated using spatial registration information (area scaling ratio and offset). The above step S201 includes: Step S2011: Spatial registration is performed between the transmitted image and the backscattered image to obtain the registered coordinates.

[0048] Using preset translation coefficients , and preset scaling factor , The coordinates in the transmitted image are mapped to the coordinates in the backscattered image. The registered coordinate formula is expressed as: (1); (2); in, , It is the translation coefficient that maps the transmitted image to the backscattered image. , It is the scaling factor for mapping the transmitted image to the backscattered image. This represents the coordinates of the transmission point in the transmission image. Represents the coordinates of the backscattered image.

[0049] Step S2012: Container detection and segmentation are performed on the transmitted image and the backscattered image respectively, the container region is identified, and the container coordinates of the transmitted image and the container bounding box of the backscattered image are obtained; the container region includes the container outline and the bounding box.

[0050] Specifically, different computer vision models are used to process bimodal images: for transmission images with strong color contrast, target detection and instance segmentation algorithms are used to quickly locate the container and obtain its precise outline, while for backscattered images with clear edges, semantic segmentation models are used to directly extract the continuous boundary of the container shell. This allows for efficient and accurate acquisition of the container's position coordinates and outline information from the two imaging modes, laying a reliable foundation for subsequent container association, liquid segmentation, and feature fusion.

[0051] Step S2013: Using the registered coordinates, the container coordinates in the transmission image are mapped to the backscatter coordinate system to generate a transmission container mapping box. Based on the transmission container mapping box and the backscatter container result box, the coordinate association relationship between the same associated container in the transmission image and the backscatter image is established.

[0052] In some optional implementations, step S2013 above includes: Step a: Calculate the cross-union ratio and the full cross-union ratio between the transmission container mapping frame and the backscattering container result frame. If both the cross-union ratio and the full cross-union ratio are greater than the preset threshold, they are determined to be the same container, and a pair of correlation relationships are established between the same container in the transmission image coordinate system and the backscattering container coordinate system.

[0053] Specifically, for both the transmission image and the backscattered image (the original scanned image is segmented to find the container, and the transmitted container is mapped onto the backscattered image to obtain a bounding box), the recognition result is used as the best reference box (the result box obtained after image segmentation in the backscattered coordinate system). The complete intersection-union ratio (CIoU) and intersection-union ratio (IoU) are calculated for the bounding box and the result box. If there is a result box at the position corresponding to the bounding box, and both are greater than a certain threshold (a decimal between 0 and 1), the container is considered to have matched successfully. The unique coordinates of each container on the transmission image and the backscattered image are obtained (coordinate values ​​in the transmission coordinate system and the backscattered image coordinate system, respectively).

[0054] The formulas for calculating the Intersection over Union (IoU) and the Complete Intersection over Union (CIoU) are as follows: (3); (4); (5); (6); in, It's a mapping box. It's the results box. and These represent the center coordinates of the mapping box and the result box, respectively. It represents the Euclidean distance between two center points (the distance between two points in Euclidean space). This represents the diagonal length of the smallest bounding rectangle of the two images. It is a weighting function. It's the similarity in aspect ratio. It is the width of the results box. It is the height of the result box. It is the width of the mapping box. It is the height of the mapping box.

[0055] Step S202: Perform liquid segmentation on the transmission image and backscattered image respectively to obtain the unobstructed liquid contour and the liquid outline, and extract the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and the liquid outline.

[0056] Specifically, step S202 includes: Step S2021: Perform liquid segmentation on the container in the transmission image, extract the unobstructed liquid contour, and match it with the associated container.

[0057] This step involves finding the associated liquid within the transmission container. Specifically, for the container in the transmission image, instance segmentation is used to extract the unobstructed liquid contour (identifying both the container outline and the liquid). Unobstructed means that only the container itself is superimposed on the liquid, without any other objects. In cases where transmission recognition is performed using instance segmentation, the liquid contour has already been output; only the association between the liquid and its container needs to be established.

[0058] Step S2022: Perform liquid segmentation on the container in the backscattered image, extract the liquid contour, and match it with the associated container.

[0059] For backscattered images, since the liquid depth information has already been reconstructed from the scattering angle and the energy information from the photon count during the image generation process, the liquid data in the backscattered image has a small error compared to the real liquid. Therefore, the liquid data can be directly extracted without considering occlusion. A segmentation algorithm is used to extract the liquid contour range within each container.

[0060] Number of photons received by the detector Counting formula: (7); in: It is photon flux; , These are incident attenuation and outgoing attenuation, respectively. It is electron density; It is the effective volume; It is a solid angle; It is the KN cross section of the Compton scattering.

[0061] The formula for the scattering KN cross section of Compton scattering within a unit solid angle is: (8); in: It is the scattering cross section within a unit solid angle; It is the classical electron radius; It is the energy of the incident photon. It is the energy of the scattered photon; It is the scattering angle.

[0062] The formula for calculating the scattering KN cross section is: (9); Preferably, global segmentation based on the YOLO framework or container local segmentation based on U-Net is used. For the U-Net framework, dilated convolution is introduced to expand the receptive field of the image, which saves computational resources and improves speed while ensuring that the convolution output contains a larger range of information.

[0063] Step S2023: Extract the corresponding liquid region raw data from the pre-acquired transmission raw data and backscatter raw data respectively, and extract the container shell region raw data from the backscatter raw data.

[0064] Specifically, this step extracts the original values ​​of the corresponding regions from the pre-acquired raw transmission data (high / low energy images) and raw backscatter data based on the two sets of contour coordinates mentioned above, forming liquid region data, and simultaneously extracts the original values ​​of the container shell region from the backscatter data, thereby providing accurate and aligned underlying data for subsequent dual-modal feature calculation.

[0065] Preferably, instance segmentation based on the YOLO (You Only Look Once) framework is used. Considering speed optimization, an attention mechanism is introduced to maintain high efficiency while further reducing the number of parameters and computational load, so as to meet the real-time processing requirements of security checks.

[0066] Step S203: Based on the original data of the liquid region and the original data of the container shell region, the transmitted liquid features, backscattered liquid features, transmitted container features, and backscattered container features are calculated and fused to construct a liquid feature matrix. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0067] Step S204: Input the liquid feature matrix into the preset liquid property discrimination model to obtain the liquid property classification result. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0068] The comprehensive liquid characterization method provided in this embodiment extracts the unobstructed liquid contour from the transmission image and the high-sensitivity liquid contour from the backscatter image, and performs precise matching based on the established container association. This achieves effective segmentation and data alignment of the dual-modal liquid region, thus providing an accurate and consistent numerical basis for subsequent fusion of the original transmission and backscatter data and extraction of highly discriminative statistical features.

[0069] This embodiment provides a comprehensive liquid qualitative method, which can be used in the aforementioned electronic devices. Figure 3 This is a flowchart of a comprehensive liquid qualitative method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Acquire the transmission image and backscatter image of the item to be inspected, and establish the coordinate relationship between the same associated container in the transmission image and backscatter image based on the transmission image and backscatter image. For details, please refer to [link to details]. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0070] Step S302 involves performing liquid segmentation on the transmitted image and the backscattered image respectively, obtaining the unobstructed liquid contour and the liquid outline, and extracting the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and the liquid outline. For details, please refer to [link to relevant documentation]. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0071] Step S303: Based on the original data of the liquid region and the original data of the container shell region, calculate the transmitted liquid features, backscattered liquid features, transmitted container features and backscattered container features, and fuse them to construct a liquid feature matrix.

[0072] Specifically, step S303 includes: For the raw transmission data (raw high-energy and low-energy data images), extract the raw data of the transmitted liquid region based on the liquid contour. For the raw backscatter data, extract the raw data of the backscatter liquid region based on the unobstructed liquid contour. Extract the non-liquid parts inside the backscatter container to obtain the raw information of the backscatter container (i.e., find the raw data of the container shell).

[0073] Considering that X-ray transmission categorizes substances into three types—organic, mixtures, and inorganic—the container containing the liquid should be an organic container, a glass mixture container, or a metal container. The extracted liquid data is a superposition of the liquid and a container of varying thicknesses. The formulas for the original data of the transmitted liquid region and the original data of the backscattered liquid region are as follows: (9); (10); in: It is the raw data of the region of liquid transmission in the transmission image; It is the transmission air scan value (preset value). It is the thickness factor multiplied by the attenuation value of the transmission container material (a known fixed factor determined based on the container material). It is the attenuation value of thickness coefficient × liquid i (unknown value in transmission image); This is the raw data of the backscattered liquid region in the backscattered image. It is the thickness × backscatter container reflection value (based on a known fixed value of the container material); It is the backscattering thickness coefficient × the reflection value of liquid i (the liquid properties and liquid thickness are known in the backscattering image).

[0074] Given that the container properties are known in the transmission image, and the liquid properties and thickness are known in the backscattered image, we can extract the raw transmission and backscattered data, calculate the feature values ​​of the liquid and container, eliminate the container's influence on the liquid, and obtain more accurate liquid data. This involves extracting transmission liquid features, backscattered liquid features, transmission container features, and backscattered container features, including: For each data point in the transmission image, five features were extracted. The selected liquid features were the first mean, the first standard deviation, the first quartiles Q1 and Q3, the first skewness, and the first coefficient of variation (CV). The ratio of liquid to container area was used as the transmission container feature value (i.e., the sixth feature, liquid area / container area).

[0075] For backscattered images, feature information of the liquid and container was calculated, and five features were extracted for each data point. The selected backscattered liquid features were the second mean, second standard deviation, second quartiles Q1' and Q3', second skewness, and second coefficient of variation (CV). The container mean (mean of a single container, the sixth feature) and the liquid-to-container ratio were used as backscattered container features (the seventh feature). These features describe the statistical characteristics and distribution of the security inspection liquid data, improving multicollinearity while maintaining the model's predictive ability.

[0076] The 13 extracted features are concatenated in a preset order (e.g., 5D of transmitted liquid → 1D of transmitted container → 5D of backscattered liquid → 2D of backscattered container) to form a 13-dimensional feature vector representing the liquid sample. When processing multiple samples, these vectors are combined into the final liquid feature matrix, which can be directly input into a machine learning model for training or prediction. This process systematically integrates statistical and geometric information from dual-modal imaging.

[0077] Step S304: Input the liquid feature matrix into the preset liquid attribute discrimination model to obtain the liquid attribute classification result.

[0078] Specifically, the liquid feature matrix is ​​input into a regression-based discrimination model to obtain liquid discrimination results, which include liquid category and probability.

[0079] When using a regression-based model, the liquid with the highest probability is identified.

[0080] When using a decision tree-based discriminative model, the result is the liquid class with the highest probability.

[0081] When using a clustering-based model, the result is the class to which the product belongs, and the probability is approximated by the distance to the center of the class.

[0082] Preferably, the preset liquid property discrimination model uses logistic regression, random forest, LightGBM (LightGradient Boosting Machine), K-means clustering, and GMM (Gaussian Mixture Model Clustering).

[0083] Step S305: Based on the characteristics of the transmitted liquid, the characteristics of the backscattered liquid, the characteristics of the transmitted container, and the characteristics of the backscattered container, construct an occlusion feature matrix and input the occlusion feature matrix into a preset occlusion discrimination model to obtain the liquid occlusion state determination result.

[0084] Specifically, from the 13 features of transmitted and backscattered liquids and containers calculated above, a key subset sensitive to occlusion (such as the mean and standard deviation of transmitted liquid, the mean of backscattered liquid, the ratio of liquid area to container area, and the mean of backscattered container) is selected and recombined into a lower-dimensional occlusion feature matrix. Subsequently, the occlusion feature matrix is ​​input into a pre-trained binary classification model (such as logistic regression, support vector machine, or lightweight gradient boosting tree). Based on the learned mapping relationship between features and occlusion state, the model outputs a judgment result indicating whether the liquid is occluded (such as "occluded" or "not occluded"), thereby achieving a specialized evaluation of liquid visibility in complex superimposed scenarios.

[0085] A pseudo-color image is calculated based on the high-energy transmission image and the low-energy transmission image; the pseudo-color image includes three channels: R, G, and B.

[0086] Step S306: Transmit high-energy image, transmission low-energy image, and transmission pseudo-color image are processed together. Figure 3 Each channel is fused with the registered and mapped backscattered image to generate multi-channel fused image data. The multi-channel fused data is then input into a preset classification model to simultaneously obtain classification results for liquid properties and occlusion states. These classification results are used to supplement or verify the liquid property classification results and liquid occlusion state determination results.

[0087] Specifically, for the discrimination model based on multimodal fusion image data, a transmission and backscatter multi-channel fusion matrix is ​​generated. For dual-energy transmission images, spatial registration information is used to map and align the backscatter image with the transmission image, redundant boundaries are removed, and missing background data is filled in.

[0088] The transmission RGB pseudocolor image is split into three images: R, G, and B. These four images (i.e., the three transmission RGB images, and the backscattered transmission mapping image (the image obtained by mapping the original backscattered image to the transmission coordinate system, with redundant boundaries removed, and the same size and object position as the transmission R, G, and B pseudocolor images)) are then normalized and resized along with the original high-energy and low-energy transmission images. Finally, they are fused into multimodal fused image data.

[0089] Image attribute discrimination is an image-based discrimination method. Two models (occlusion and liquid attributes) are completed using a fully connected neural network. The generated multi-channel fused data is input, and the liquid category and liquid occlusion category are obtained simultaneously.

[0090] Multimodal fused image data is input into a multi-task discrimination model for synchronous analysis and cross-validation, thus forming a complete technical closed loop from data acquisition and fusion processing to intelligent discrimination. Ultimately, in complex and superimposed security inspection scenarios, it achieves rapid, accurate and robust automated qualitative identification of liquid properties in sealed containers. Liquid category - liquid occlusion category is used to supplement or verify the above single liquid property classification results and single liquid occlusion state judgment results.

[0091] The comprehensive liquid qualitative method provided in this embodiment constructs multi-channel fused data that includes material attenuation characteristics, atomic number classification information, and surface scattering features by fusing raw transmission high / low energy data, pseudo-color image decomposition channels derived from them, and registered backscattered images. It also uses a multi-task classification model to achieve synchronous and integrated discrimination of liquid properties and occlusion states, thereby improving the richness of feature representation while enhancing the overall synergy, efficiency, and reliability of the system's judgment.

[0092] As one or more specific application embodiments of the present invention, combined with Figure 4 and Figure 5 A further detailed description of the comprehensive liquid qualitative method set provided by the present invention is as follows: Figure 4 As shown, the specific process is as follows: I. Using a discriminative model based on statistical information: Step 1: Extract the bounding box bbox_T of the liquid container from the transmission recognition result and the bounding box bbox_B of the liquid container from the backscatter recognition result. Use the spatial registration information offset_x, offset_y, scale_x, and scale_y obtained by registering the transmission and backscatter images to map the coordinates of the transmission container to the backscatter coordinate system to obtain bbox_T_B, and map the coordinates of the transmission container to the backscatter coordinate system to obtain bbox_B_T.

[0093] Step 2: Using the recognition result as the best bounding box, merge bbox_T with bbox_B_T and bbox_B with bbox_T_B using IoU and CIoU parameters to obtain the associated coordinates bbox_coord of the same container in the transmission and backscatter images. Each container contains a set of mutually associated transmission coordinates bbox_coord_T and backscatter coordinates bbox_coord_B.

[0094] Step 3: For the transmission image, since the input is the recognition result based on the YOLO framework, which already contains liquid information, extract the unobstructed liquid range information container_liquids_T in each container and associate it with the corresponding container in bbox_coord.

[0095] Step 4: For the containers in the backscattered image, use semantic segmentation based on the U-Net framework to extract the liquid range information (container_liquids_B) and the non-liquid container range (container_sur_B) within each container. Then, associate these information with the corresponding container in the bbox_coord.

[0096] Step 5: Using container_liquids_T, extract the original values ​​of the unobstructed liquid from the original data of the high-energy image, low-energy image, and pseudo-color image, normalize them, and then statistically obtain 4 sets of transmitted liquid characteristic values.

[0097] Step 6: Use container_liquids_B to extract the raw liquid data from the backscattered original image, normalize it, and then statistically obtain the backscattered liquid feature values.

[0098] Step 7: Use container_sur_B to extract the original container data from the backscattered image, normalize it, and then statistically obtain the backscattered container feature values.

[0099] Step 8: Associate the extracted feature values ​​with the containers in bbox_coord.

[0100] Step 9: For each container in bbox_coord, apply PCA (Principal Component Preservation) to optimize the variance of the liquid classification model. Input this optimization into three classification models: logistic regression, random forest, and LightGBM, respectively, to obtain three classification results and their confidence scores. The class with the highest confidence score is taken as the final liquid classification result.

[0101] Step 10: For each container in `bbox_coord`, extract the mean of transmitted liquid, standard deviation of transmitted liquid, mean of backscattered liquid, standard deviation of backscattered liquid, mean of backscattered container, percentage of transmitted liquid in container, and percentage of backscattered liquid in container. Integrate these into container features and use an occlusion discrimination model to output the occlusion results. A schematic diagram of the statistical liquid attribute classification results is shown below. Figure 5 As shown.

[0102] II. Using a discriminative model based on multimodal fusion image data: Step 1: Extract the bounding box bbox_T of the liquid container from the transmission recognition result and the bounding box bbox_B of the liquid container from the backscatter recognition result. Use the spatial registration information offset_x, offset_y, scale_x, and scale_y obtained by registering the transmission and backscatter images to map the coordinates of the transmission container to the backscatter coordinate system to obtain bbox_T_B, and map the coordinates of the transmission container to the backscatter coordinate system to obtain bbox_B_T.

[0103] Step 2: Using the recognition result as the best bounding box, merge bbox_T with bbox_B_T and bbox_B with bbox_T_B using IoU and CIoU parameters to obtain the associated coordinates bbox_coord of the same container in the transmission and backscatter images. Each container contains a set of mutually associated transmission coordinates bbox_coord_T and backscatter coordinates bbox_coord_B.

[0104] Step 3: For the containers in the transmission image, use instance segmentation based on the YOLO framework to extract the unobstructed liquid range information container_liquids_T in each container and associate it with the corresponding container in bbox_coord.

[0105] Step 4: For the containers in the backscattered image, use instance segmentation based on the YOLO framework to extract the liquid range information (container_liquids_B) and the non-liquid container range (container_sur_B) within each container. Then, associate these information with the corresponding container in the bbox_coord.

[0106] Step 5: For each container in the backscattered original image, extract the backscattered container image container_B.

[0107] Step 6: Using the spatial registration information offset_x, offset_y, scale_x, scale_y, map the backscattered container map to the size of the transmission container, and obtain the backscattered container map container_B_T.

[0108] Step 7: For the dual-energy transmission original image, for each container, extract the high-energy original container image (container_H), the low-energy original container image (container_L), and the pseudo-color container image (container_RGB). Then, split the transmission pseudo-color container image (container_RGB) into three images: container_R, container_G, and container_B. After normalizing the values ​​and adjusting the sizes of these images, fuse them with the high-energy container image (container_H), the transmission low-energy container image (container_L), and the backscattered container mapping image (container_B_T) to form the multimodal fused image data (container_data). Finally, associate container_data with the containers in bbox_coord.

[0109] Step 8: For the multimodal fusion image data of each container in bbox_coord, use a fully connected network image classifier to output the liquid classification results and confidence scores.

[0110] Step 9: For the multimodal fused image data of each container in bbox_coord, use a fully connected network occlusion classifier to output the occlusion results and confidence scores.

[0111] The comprehensive liquid qualitative method provided in this embodiment has the following beneficial effects: 1. By combining X-ray transmission and backscattering, more accurate organic data is introduced from backscattering, which improves the low accuracy of traditional transmission data in the organic range and enhances the accuracy of organic property identification.

[0112] 2. By introducing feature statistics and discrimination of raw data, the method improves the situation where inaccurate liquid values ​​make it difficult to determine liquid composition when liquids are superimposed on containers. Building upon the recognition based on graphic contour color, it enhances the accuracy of liquid attribute recognition. Introducing container feature values ​​and combining them with liquid feature values ​​improves the recognition effect when containers are occluded. The method has low complexity, fast computation speed, meets the real-time requirements of security checks, and improves the accuracy of security check recognition.

[0113] 3. Combining X-ray transmission and backscattering, and introducing normalized multimodal fusion images, multiple results are output in a single inference, improving the accuracy and speed of attribute discrimination when combining transmission and backscattering.

[0114] This embodiment also provides an X-ray security inspection device, such as... Figure 6 As shown, the device includes: an X-ray transmission imaging module 601, a Compton backscatter imaging module 602, an image feature extraction module 603, and a first liquid property discrimination module 604.

[0115] The X-ray transmission imaging module is used to acquire transmission images of the object under inspection; the Compton backscatter imaging module is used to acquire backscatter images of the object under inspection; and the image feature extraction module is used to perform the above operations. Figure 1 The process involves acquiring the transmission and backscatter images of the item to be inspected, establishing coordinate relationships between the same container in the transmission and backscatter images, performing liquid segmentation on the transmission and backscatter images to obtain unobstructed liquid contours and liquid outlines, extracting raw data of the liquid region and raw data of the container shell region based on the raw data of the unobstructed liquid contours and liquid outlines, calculating transmission liquid features, backscatter liquid features, transmission container features, and backscatter container features based on the raw data of the liquid region and the raw data of the container shell region, and fusing them to construct a liquid feature matrix; the first liquid attribute discrimination module is used to perform... Figure 1 The steps shown are to input the liquid feature matrix into a preset liquid property discrimination model to obtain the liquid property classification result.

[0116] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0117] This embodiment also provides a comprehensive liquid qualitative apparatus for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0118] This embodiment provides a comprehensive liquid qualitative analysis device, such as... Figure 7 As shown, it includes: The module 701 for establishing the correlation between transmission image and backscattered image is used to acquire the transmission image and backscattered image of the item to be inspected, and to establish the coordinate correlation between the same associated container in the transmission image and backscattered image based on the transmission image and backscattered image.

[0119] The liquid contour and container shell extraction module 702 is used to perform liquid segmentation on the transmission image and backscatter image respectively, to obtain the unobstructed liquid contour and liquid outline, and to extract the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and liquid outline.

[0120] The liquid feature matrix construction module 703 is used to calculate the transmitted liquid features, backscattered liquid features, transmitted container features, and backscattered container features based on the original data of the liquid region and the original data of the container shell region, and then fuse them to construct the liquid feature matrix.

[0121] The second liquid attribute discrimination module 704 is used to input the liquid feature matrix into a preset liquid attribute discrimination model to obtain the liquid attribute classification result. The second liquid attribute discrimination module 704 has the same function as the first liquid attribute discrimination module 604.

[0122] In some optional implementations, the correlation establishment module 701 between the transmitted image and the backscattered image includes: The spatial registration unit is used to spatially register the transmitted image and the backscattered image to obtain the registered coordinates.

[0123] The container detection and segmentation unit is used to detect and segment containers in the transmitted image and the backscattered image respectively, identify container regions, and obtain the container coordinates in the transmitted image and the container bounding box in the backscattered image; the container region includes the container outline and the bounding box.

[0124] The mapping and association establishment unit is used to map the container coordinates in the transmission image to the backscatter coordinate system using the registered coordinates, generate the transmission container mapping box, and establish the coordinate association relationship between the same associated container in the transmission image and the backscatter image based on the transmission container mapping box and the backscatter container result box.

[0125] In some alternative implementations, the spatial registration unit includes: The registration subunit is used to map coordinates in the transmitted image to coordinates in the backscattered image using preset translation and scaling factors. The coordinate formula after registration is expressed as: ; ; in, , It is the translation coefficient that maps the transmitted image to the backscattered image. , It is the scaling factor for mapping the transmitted image to the backscattered image. This represents the coordinates of the transmission point in the transmission image.

[0126] In some optional implementations, the mapping and association establishment unit includes: The Cross-Union Ratio (CUI) and Correlation Establishment Sub-unit is used to calculate the CUI and Full Cross-Union Ratio between the transmission container mapping frame and the backscatter container result frame. If both the CUI and Full Cross-Union Ratio are greater than a preset threshold, they are determined to be the same container, and a pair of correlation relationships between the same container in the transmission image coordinate system and the backscatter container coordinate system are established.

[0127] In some alternative implementations, the liquid profile and container shell extraction module 702 includes: The transmission image liquid segmentation unit is used to segment the liquid in the container in the transmission image, extract the unobstructed liquid contour, and match it with the associated container.

[0128] The backscattered image liquid segmentation unit is used to segment the liquid in containers in the backscattered image, extract the liquid contour, and match it with the associated container.

[0129] The raw data extraction unit is used to extract the corresponding raw data of the liquid region from the pre-acquired raw transmission data and raw backscatter data, and at the same time extract the raw data of the container shell region from the raw backscatter data.

[0130] In some optional embodiments, the characteristics of the transmitted liquid include: a first mean, a first standard deviation, a first quartile Q1, a first quartile Q3, a first skewness, and a first coefficient of variation; the characteristic of the transmitted container is the ratio of the transmitted liquid to the container. Backscattering liquid characteristics include the second mean, second standard deviation, second quartile Q1', second quartile Q3', second skewness, and second coefficient of variation; backscattering container characteristics include the container mean and the percentage of backscattering liquid in the container.

[0131] In some alternative embodiments, the device further includes: The liquid occlusion state determination module is used to construct an occlusion feature matrix based on the characteristics of the transmitted liquid, the characteristics of the backscattered liquid, the characteristics of the transmitted container, and the characteristics of the backscattered container, and input the occlusion feature matrix into the preset occlusion discrimination model to obtain the liquid occlusion state determination result.

[0132] In some alternative embodiments, the transmission image includes a transmission high-energy image and a transmission low-energy image; the apparatus further includes: The multi-task decision module is used to calculate a pseudo-color image based on the high-energy transmission image and the low-energy transmission image; the pseudo-color image includes three channels: R, G, and B; and it combines the high-energy transmission image, the low-energy transmission image, and the pseudo-color image. Figure 3Each channel is fused with the registered and mapped backscattered image to generate multi-channel fused image data. The multi-channel fused data is then input into a preset classification model to simultaneously obtain classification results for liquid properties and occlusion states. These classification results are used to supplement or verify the liquid property classification results and liquid occlusion state determination results.

[0133] The integrated liquid qualitative analysis apparatus provided in this embodiment of the invention can execute the integrated liquid qualitative analysis method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0134] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0135] The following is a detailed reference. Figure 8 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 801, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 802 or a program loaded from memory 808 into random access memory (RAM) 803. The RAM 803 also stores various programs and data required for the operation of the electronic device. The processor 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0136] Typically, the following devices can be connected to I / O interface 805: input devices 806 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 807 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 808 including, for example, magnetic tapes, hard disks, etc.; and communication devices 809. Communication device 809 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0137] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 809, or installed from a memory 808, or installed from a ROM 802. When the computer program is executed by the processor 801, it performs the functions defined in the comprehensive liquid qualitative method of the embodiments of the present invention.

[0138] Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0139] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the comprehensive liquid qualitative method shown in the above embodiments is implemented.

[0140] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0141] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A comprehensive qualitative method for liquids, characterized in that, The method includes: Acquire the transmission and backscatter images of the item to be inspected, and establish the coordinate relationship between the same associated container in the transmission and backscatter images based on the transmission and backscatter images; Liquid segmentation is performed on the transmission image and backscattered image respectively to obtain the unobstructed liquid contour and the liquid contour. Based on the unobstructed liquid contour and the liquid contour, the original data of the liquid region and the original data of the container shell region are extracted. Based on the original data of the liquid region and the original data of the container shell region, the transmitted liquid characteristics, backscattered liquid characteristics, transmitted container characteristics and backscattered container characteristics are calculated and fused to construct a liquid characteristic matrix; The liquid feature matrix is ​​input into a preset liquid property discrimination model to obtain the liquid property classification result.

2. The method according to claim 1, characterized in that, Establish the coordinate relationship between the same associated container in the transmitted image and the backscattered image, including: The transmitted image and the backscattered image are spatially registered to obtain the registered coordinates. Container detection and segmentation are performed on the transmitted image and the backscattered image respectively to identify the container region and obtain the container coordinates of the transmitted image and the container bounding box of the backscattered image; the container region includes the container outline and the bounding box. The coordinates of the container in the transmission image are mapped to the backscatter coordinate system using the registered coordinates to generate a transmission container mapping box. Based on the transmission container mapping box and the backscatter container result box, the coordinate association relationship between the same associated container in the transmission image and the backscatter image is established.

3. The method according to claim 2, characterized in that, Spatial registration of the transmitted image and the backscattered image is performed to obtain the registered coordinates, including: Using preset translation and scaling factors, the coordinates in the transmitted image are mapped to the coordinates in the backscattered image. The registered coordinate formula is expressed as: ； ； in, , It is the translation coefficient that maps the transmitted image to the backscattered image. , It is the scaling factor for mapping the transmitted image to the backscattered image. This represents the coordinates of the transmission point in the transmission image.

4. The method according to claim 2, characterized in that, Based on the transmission container mapping frame and the backscatter container result frame, establish the coordinate association relationship between the same associated container in the transmission image and the backscatter image, including: Calculate the cross-union ratio and the full cross-union ratio between the transmission container mapping frame and the backscatter container result frame. If both the cross-union ratio and the full cross-union ratio are greater than a preset threshold, they are determined to be the same container, and a pair of correlation relationships between the same container in the transmission image coordinate system and the backscatter container coordinate system are established.

5. The method according to claim 1, characterized in that, Liquid segmentation is performed on the transmitted image and the backscattered image respectively, resulting in the unobstructed liquid contour and the liquid outline. Based on these unobstructed liquid contours and the liquid outline, raw data of the liquid region and the container shell region are extracted, including: Liquid segmentation is performed on the container in the transmission image, the outline of the unobstructed liquid is extracted, and it is matched with the associated container; Liquid segmentation is performed on the container in the backscattered image, the liquid contour is extracted, and it is matched with the associated container; The corresponding raw data of the liquid region is extracted from the pre-acquired raw transmission data and raw backscatter data, and the raw data of the container shell region is extracted from the raw backscatter data.

6. The method according to claim 1, characterized in that, The characteristics of the transmitted liquid include: a first mean, a first standard deviation, a first quartile Q1, a first quartile Q3, a first skewness, and a first coefficient of variation; the characteristic of the transmitted container is the ratio of the transmitted liquid to the container. The backscattering liquid characteristics include the second mean, the second standard deviation, the second quartile Q1', the second quartile Q3', the second skewness, and the second coefficient of variation; the backscattering container characteristics include the container mean and the ratio of backscattering liquid to container volume.

7. The method according to claim 1, characterized in that, The method further includes: Based on the transmitted liquid characteristics, backscattered liquid characteristics, transmitted container characteristics, and backscattered container characteristics, an occlusion feature matrix is ​​constructed, and the occlusion feature matrix is ​​input into a preset occlusion discrimination model to obtain the liquid occlusion state determination result.

8. The method according to claim 7, characterized in that, The transmission image includes a high-energy transmission image and a low-energy transmission image; the method further includes: A pseudo-color image is calculated based on transmission high-energy and transmission low-energy images; the pseudo-color image includes three channels: R, G, and B. The three channels of transmission high-energy image, transmission low-energy image and transmission pseudo-color image are fused with the backscattered image after registration and mapping to generate multi-channel fused image data; Multi-channel fused data is input into a preset classification model to simultaneously obtain classification results of liquid properties and occlusion states. These classification results are used to supplement or verify the liquid property classification results and liquid occlusion state determination results.

9. An X-ray security inspection device, characterized in that, The device includes: an X-ray transmission imaging module, a Compton backscatter imaging module, an image feature extraction module, and a first liquid property discrimination module; The X-ray transmission imaging module is used to acquire transmission images of the item to be inspected; The Compton backscatter imaging module is used to obtain backscatter images of the item under inspection. The image feature extraction module is used to perform the steps of acquiring the transmission image and backscatter image of the item to be inspected in claim 1, and establishing the coordinate relationship between the same associated container in the transmission image and backscatter image based on the transmission image and backscatter image; performing liquid segmentation on the transmission image and backscatter image respectively to obtain the unobstructed liquid contour and liquid contour, and extracting the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and liquid contour; calculating the transmission liquid feature, backscatter liquid feature, transmission container feature and backscatter container feature based on the original data of the liquid region and the original data of the container shell region, and fusing them to construct a liquid feature matrix. The first liquid property discrimination module is used to perform the step in claim 1 of inputting the liquid feature matrix into a preset liquid property discrimination model to obtain the liquid property classification result.

10. A comprehensive liquid qualitative analysis device, characterized in that, The device includes: The module for establishing the correlation between transmission images and backscattered images is used to acquire transmission images and backscattered images of the item to be inspected, and to establish the coordinate correlation between the same associated container in the transmission images and backscattered images based on the transmission images and backscattered images. The liquid contour and container shell extraction module is used to perform liquid segmentation on the transmission image and backscatter image respectively, to obtain the unobstructed liquid contour and liquid outline, and to extract the original data of the liquid region and the original data of the container shell region based on the unobstructed liquid contour and liquid outline. The liquid feature matrix construction module is used to calculate the transmitted liquid features, backscattered liquid features, transmitted container features, and backscattered container features based on the original data of the liquid region and the original data of the container shell region, and to fuse them to construct the liquid feature matrix. The second liquid property discrimination module is used to input the liquid feature matrix into a preset liquid property discrimination model to obtain the liquid property classification result.

Images