CN104165896A - Liquid goods safety inspection method and device - Google Patents

Abstract

The invention discloses a method and device for safety inspection of liquid articles. By performing dual-view X-ray imaging on liquid articles placed in a special inspection tray, only using the X-ray projection information of two viewing angles, by analyzing the projection image of the container content, reconstruct the shape of the cross-section of the container, and use the classifier method to efficiently realize the automatic detection of the danger of the inspected liquid article; the liquid article safety inspection device of the present invention includes two perspectives and consists of 2 groups of X-ray sources and detectors , each group of X-ray sources and detectors constitutes a detection unit; the invention has the advantages of high inspection accuracy and imaging quality and can perform safety inspection on multiple liquid objects at one time, with fast inspection speed and high cost performance.

Description

Method and device for safety inspection of liquid articles

technical field

The invention relates to a method and device for safety inspection of liquid articles, belonging to the technical field of safety inspection of liquid articles.

Background technique

At present, in the existing liquid article safety inspection technology, the non-destructive inspection method based on X-ray transmission imaging technology has the characteristics of high accuracy of inspection results, low sensitivity to container materials, and convenient operation, and is receiving more and more attention. Security inspection manufacturers attach great importance to it. Among them, two invention patents with publication numbers CN101140247A and CN101629916A are more representative. What they have in common is that they are all based on CT tomography technology. The method can be briefly described as follows: the radiation source emits X The ray transmission is inspected liquid object, the detector is used to receive the ray beam transmitted through the liquid object, and forms hundreds of multi-angle projection data, and calculates by inverting the hundreds of multi-angle projection data Obtain the ray absorption coefficient of the inspected liquid article. Among them, the invention patent of CN101629916A can obtain the density and material information of the inspected liquid article at the same time due to the use of dual-energy X-rays. Finally, the absorption coefficient of the liquid article or the density of the liquid article, The material information is compared with the preset database to complete the inspection of the inspected liquid items. The biggest advantage of this type of liquid article safety inspection method based on CT tomography technology is its high inspection accuracy, because it has obtained hundreds of multi-angle projection data during the inspection process, and then uses relatively mature various projection reconstruction technologies , the ideal fault cross-section data can be obtained.

Although the above-mentioned CT tomography technology has high detection performance, it also has obvious deficiencies: 1) The inspection efficiency is low, and usually only one liquid object of a conventional size can be inspected at a time, and the inspection process takes a long time; 2) The inspection object Coverage is limited, except that it is not suitable for detecting super-large and super-high liquids, and it can only specifically inspect liquid objects and is not compatible with inspection packages; these two shortcomings will limit the application of CT-type liquid object safety inspection methods and equipment in places with large flow of people , Such as airports, railway stations, large gatherings and other places.

Therefore, for the safety inspection of common liquid articles, it is necessary to research and develop a method and device for safety inspection of liquid articles, which can not only ensure high inspection accuracy, but also realize rapid inspection of multiple liquid articles, and is also compatible with passengers. Take your package with you for inspection.

Contents of the invention

The purpose of the present invention is to provide a method and device for safety inspection of liquid articles that can overcome the above-mentioned technical problems. The present invention can perform dual-view X-ray imaging on liquid articles placed in a special inspection tray, without damaging the original packaging. Under the circumstances, only use the X-ray projection information of two viewing angles to quickly inspect multiple liquid items in the tray at one time, and efficiently realize the automatic detection of the danger of the inspected liquid items. In the liquid article safety inspection method of the present invention, the liquid is placed in the tray in the order that the front and back do not block each other, and during the detection process, the package and the tray appear alternately, the liquid article and the non-liquid article appear alternately in the tray, and multiple liquid articles Several cases of head-to-tail connection will not affect the detection ability; the method of the present invention is based on the dual-view image of the liquid object, analyzes the attributes of the container, reconstructs the container section, and then obtains various characteristic information of the liquid in the container, and judges accordingly Are liquid objects dangerous.

The liquid article safety inspection method of the present invention can distinguish between packages and trays, liquid articles in the tray and non-liquid articles, and multiple liquid articles connected head to tail, and then select a suitable detection slice position of the liquid container; in each slice , first analyze the properties of the container, and then reconstruct the cross-section of the dual-view container based on the properties of the container, use the reconstructed cross-section to obtain various characteristics such as the material and density of the liquid, and use a suitable classifier such as a support vector machine to judge whether the liquid is dangerous ; The method of the present invention can be applied to a variety of channel-type X-ray security inspection equipment with no less than 2 viewing angles and side-viewing viewing angles.

The method for safety inspection of liquid articles of the present invention only uses dual-view X-ray images to reconstruct container sections, firstly analyzes container attributes, including information such as container material and basic types of container sections; then combines nonlinear least squares method and conjugate gradient method , to realize the reconstruction of the cross-sectional shape of the container under the condition of two viewing angles; the method for reconstructing the cross-sectional shape of the container can be applied to various channel-type X-ray security inspection equipment with not less than two viewing angles and a side-viewing viewing angle.

The method for security inspection of liquid articles of the present invention distinguishes whether it is a package or a tray according to the content of the X-ray image, and the method for distinguishing the package and the tray can be applied to various traditional channel X-ray security inspections including single-view and multi-view devices equipment.

The method for security inspection of liquid articles of the present invention distinguishes whether each article in the tray is a liquid article or a non-liquid article according to the shape, grayscale, and material information of the article area in the tray. The method for distinguishing liquid articles and non-liquid articles can be applied to include Various traditional channel-type X-ray security inspection equipment including single-view and multi-view equipment.

The liquid article security inspection method of the present invention divides a plurality of liquid articles connected head to tail according to the shape, grayscale, and material information of the liquid article area, and finally selects the position of the detection slice on each liquid container for subsequent detection; The method for segmenting the head-to-tail connected liquid items and the method for selecting the position of the detection slice can be applied to various traditional channel X-ray safety inspection equipment including single-view and multi-view equipment.

After acquiring the dual-view image of the liquid item, the liquid detection method of the present invention specifically includes the following steps:

1) Image classification: Divide the collected security images into two categories, one is package images, and the other is tray images, in which liquid items, belts, wallets, mobile phones, coats, ladies’ handbags, etc. may be placed in the trays. Items subject to separate security checks;

2) Image segmentation: The present invention allows non-liquid items and liquids to be mixed on the same tray for security inspection, but requires that liquid items and other items do not overlap in any viewing angle, and images from 2 viewing angles of one tray are divided into Multi-segment, each segment of the image is either full of liquid items, called the container image segment, or full of non-liquid items, called the non-liquid image segment;

3) Detection position selection: multiple liquid items in the same container image segment in the tray are likely to be connected end to end, and multiple containers that may be connected end to end are separated according to material property changes, volume, local shape, etc. Then find several slice positions with stable local material properties in each container for subsequent detection;

4) Container attribute estimation: First, determine whether the liquid container is a high-density container. If it is a high-density container, it is necessary to determine the basic shape classification of the container section. If it is a low-density container, it is also necessary to analyze the local morphological characteristics of the container projection. Obtain various necessary container information;

5) Section reconstruction of low-density containers: For low-density containers such as plastic bottles, model the section of the container, propose an algorithm combining nonlinear least squares method and conjugate gradient method, iteratively estimate whether each grid in the section belongs to the container area, Finally, the cross-sectional shape is obtained;

6) High-density container cross-section reconstruction: For high-density containers such as glass bottles, the container cross-section is modeled, and an algorithm based on the nonlinear least squares method combined with the conjugate gradient method is proposed to iteratively estimate whether each grid in the cross-section belongs to the container, liquid, etc. or air, and finally obtain the cross-sectional shape, utilizing the container attribute information obtained in step 4) in this process;

7) Decision-making: Based on the cross-sectional shape of the container obtained in step 5) or 6), calculate various characteristics such as the material and density of the liquid, and use the support vector machine as a classifier to give a judgment on whether the liquid is dangerous.

Finally, as long as one slice of a container is judged as dangerous by step 7), the system will give an alarm to the dangerous container.

The method of the present invention allows liquid and non-liquid items to be alternately placed in the tray without deliberate distance between the liquids. In addition, before reconstructing the cross-section of the container, more abundant container property information can be obtained through the X-ray projection image.

The liquid article safety inspection device of the present invention adopts a brand-new layout design of dual viewing angles. The two viewing angles are composed of two sets of X-ray sources and detectors, and each set of X-ray sources and detectors is called a detection unit. The detection unit is a central bottom-illuminating angle of view unit V1 and a side-illuminating angle of view unit V2 installed in the conveying channel; the central bottom-illuminating angle of view unit V1 includes a central bottom-illuminating X-ray source and a first detector; the side-illuminating angle of view Unit V2 includes a side-illuminated X-ray source and a second detector; the images of the detected items collected from the two viewing angles are processed and judged according to the above-mentioned 7 liquid detection steps of the present invention; The two perspectives described are truly orthogonal.

The advantage of the present invention is that by reconstructing the section of the liquid object, and then calculating various characteristic information such as the material and density of the liquid, in addition to checking the liquid in the low-density (such as plastic) container, it can also check the high-density (such as glass) The liquid in the container is inspected, which has high inspection accuracy and imaging quality. In addition, compared with the prior art, the present invention can perform safety inspection on multiple liquid objects at one time, with fast inspection speed and high cost performance, and is suitable for people Application of safety inspection of liquid items in places with large flow such as airports, railway stations, large gatherings and other places.

In addition, the present invention also has the function of parcel security inspection, enabling security inspectors to observe parcels through two orthogonal angles, and the probability of occlusion between objects is low, and the imaging deformation is small. In short, the most functions are realized with the relatively least number of perspectives.

Description of drawings

Fig. 1 is the flow chart of the method for safety inspection of liquid articles in the present invention;

Fig. 2 is a schematic structural view of the liquid article safety inspection device of the present invention;

Fig. 3 is a schematic top view of the liquid-carrying plastic tray according to the present invention, wherein the liquid storage tank is in the middle of the tray;

Fig. 4 is a schematic front view of the conveying channel and the position of the X-ray source in the liquid article safety inspection device according to the present invention, wherein ABCD represents the circumscribed polygon of the section of the inspected object;

Fig. 5 is a schematic diagram of the lower half of the projection curves of the slices of the circular container (Fig. 5(a)) and the square container (Fig. 5(b)) in the side view angle V2;

Fig. 6 is a schematic diagram of the initial shape of the section of the container, i.e. the gray part, wherein ABCD is the circumscribed polygon of the section of the object in Fig. 4;

Fig. 7 is the schematic diagram of the image that a plurality of containers are connected end to end;

Fig. 8 is a schematic cross-sectional view of a high-density container;

FIG. 9 is a schematic diagram of a projection curve of a side-viewing angle V2 of a low-density container slice.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

Fig. 1 has shown the flow chart of a kind of liquid article safety inspection method described in the present invention; First, liquid is placed flat in the liner groove in the plastic pallet of certain size, and each article is placed in order before and after, carries out X-ray scanning; Then , the device analyzes the liquid items therein according to the acquired images of the tray from two perspectives, so as to realize the safety inspection of the liquid items. The detection process mainly includes the following modules: image classification module 1, image segmentation module 2, detection location selection module 3, container attribute estimation module 4, low-density container section reconstruction module 5, high-density container section reconstruction module 6, and decision-making module 7. Finally, return the liquid detection result to the control program of the safety inspection system. If flammable and explosive dangerous goods are found, the program will alarm.

Fig. 2 is a schematic structural view of the liquid article safety inspection device according to the present invention, wherein 11 and 12 are lead door curtains at the exit of the channel, 13 is a detector group with a side view angle V2, 14 is a collimator with a side view angle V2, and 15 16 is the detector group at the bottom viewing angle V1 in the middle, 17 is the channel of the safety inspection device, 18 and 19 are lead door curtains at the channel entrance, 20 is the plastic tray, and 21 is the light sensor at the channel exit 22 is the ray source of the middle bottom viewing angle V1, 23 is the collimator of the middle bottom viewing angle V1, 24 is the light sensor at the entrance of the channel, 25 is the conveyor, 26 is the liquid object, 27 is the system control and Signal processing circuit unit, 28 is a comprehensive processing computer. Liquid articles 26 will be placed flat in the groove 32 shown in Figure 3 in the plastic tray 20 according to regulations.

When the plastic tray 20 carrying liquid items 26 enters from the entrance on the right side of the passage, and passes through the scanning of the bottom irradiation source 22 in the middle of V1 and the side irradiation source 15 in V2, the system control and signal processing circuit unit 27 obtains images from two perspectives respectively. , and sent to the computer 28 for analysis and processing. Among them, please refer to Figure 3 for the form of the plastic tray. There is a lining groove in the center of the tray for placing liquid items.

The following describes the functions and implementation methods of each module in Figure 1 one by one for the process shown in Figure 1;

In the image classification module 1, the collected security inspection images need to be divided into two categories, one is package images, and the other is pallet images. Trays may contain liquids, belts, wallets, mobile phones, outerwear, ladies' handbags and other items that require separate security checks. It is also required that liquid items and other non-liquid items will not be overlapped and imaged in two viewing angles, that is, the safety inspection device of the present invention does not allow non-liquid items and liquid items to be placed side by side in the tray along the conveying direction. .

When distinguishing whether the currently inspected object is a package or a pallet, the size of the inspected object must first be determined according to the image content. Find the four sides of the inspected object through edge detection and Hough transform. Since the size of the pallet is known, the size of the inspected object can be deduced according to the above four-side boundary information, so that it can be identified that a large part of the packages are not pallets. Then, for the remaining images, it can be judged whether it is a pallet based on the following features: 1) If there are more metal pixels on the four sides of the V1 perspective or on the front and rear boundaries of the V2 perspective, it is a package rather than a Plastic tray; 2) The overall average grayscale, usually the average grayscale of each pixel in the package is low, and if the average grayscale is low to a certain extent, it must be a package; 3) There is a grayscale in both the V1 and V2 perspectives. The threshold value T _tray corresponds to the background of an empty tray. If the grayscale of a certain number of pixels within the four sides of the current viewing angle image is higher than T _tray , it is considered a tray. In fact, if there is a liquid in the tray, due to the shape of the liquid and the rules for placing items mentioned in the previous paragraph, it cannot completely block the background of the tray near its position, and some pixels of the background of the tray will always be exposed. It is easy to judge as a tray. In case there is no liquid in the tray and it is covered with various sundries, if it is misjudged as a package, the tray will be treated as a package for subsequent explosive detection without damaging the detection capability of the equipment; on the contrary, very few are almost empty, And a package whose size is similar to a pallet may be misjudged as a pallet. This problem will be explained in subsequent steps (referring to the description of the image segmentation module 2).

In the image segmentation module 2, each column of the pallet image is analyzed as an analysis unit, and the image is judged from which column to which column, and the object in between is liquid. This segment of image is called a container image segment; and the image segment From which column to which column, during which the object is non-liquid (personal item), this segment of image is called a non-liquid image segment; wherein, each row of pixels of the image corresponds to the X-ray fan of the viewing angle V1 or V2 in one scan The value of each detector. On each column of the image, the tray background is used as the threshold for segmentation: advance backward from the first pixel of this column, and when a pixel whose gray level is lower than the tray background is encountered, it is considered to have encountered the boundary of the object in the tray; from this column The last pixel of the image is moved forward. When a pixel whose grayscale is lower than the background of the tray is encountered, it is considered to have encountered the boundary of the other end of the object; if the object is after the boundary of the image column, background grayscale removal processing can be performed. Then, use the object pixels between the two ends of the boundary to identify whether the current column is a column of pixels of liquid items according to the following characteristics:

1) The distance between the borders at both ends (especially useful for the V1 perspective), and whether these pixels are close to the position of the liquid placement tank.

2) Average gray level: the average gray level between the borders at both ends and the central 1/3 part, these two average gray levels are too high to be regarded as liquid.

3) The average value of P(liquid|G _p ,M _p ) of each pixel p between the two ends of the boundary, where P(liquid|G _p ,M _p ) means that when the gray level of p is G _p and the material value is M When _p , the probability that point p belongs to the liquid item; the probability table P(liquid|G _p , M _p ) comes from the statistics of the image of the liquid item in advance, and the material value comes from the dual-energy grayscale information.

4) The average value/variance of the local gradient of the gray level; including calculating the gradient in the column direction and the row direction respectively.

5) Consecutive high grayscale pixels close to the grayscale of the tray background are encountered between the borders at both ends.

6) Between the borders at both ends, continuously encounter pixels whose material values correspond to metals (the color of such pixels in the security inspection field is dark blue).

If the above 6 columns conform to the liquid characteristics, and the number of consecutive columns is sufficient, these columns are considered to represent the columns occupied by liquid items in the image, and the continuous liquid columns and mixed blank columns are regarded as a container image segment , the so-called blank column is a column in which no object pixel can be found; then, the continuous non-liquid column and the mixed blank column are regarded as a non-liquid image segment. For the non-liquid image segment, ordinary package explosive detection is carried out; for the container image segment, the subsequent liquid detection process will be carried out.

In the image classification module 1, there is a special case, that is, a very few packages that are almost empty and similar in size to a pallet may be misjudged as a pallet. Liquid image segment, so as to detect ordinary package explosives; or if there is liquid in the package and the liquid image segment can be separated, then the subsequent liquid detection process can be carried out.

In the detection position selection module 3, each liquid container in a container image segment can be separated according to the liquid object image from the V1 or V2 perspective to form individual containers for subsequent analysis. In the aforementioned image segmentation module 2, the upper and lower boundaries of the object are found for each column of the image. If multiple liquid items in a tray are connected end to end, a large liquid area consisting of many consecutive columns will be formed. It is necessary to separate each container in this large area. The main basis for the segmentation is the following criteria: 1) the gap; 2) the bottom of the glass bottle; 3) the bottleneck; 4) the sudden change in material properties; Mutations in item shape.

Gaps are the simplest basis for item segmentation, as shown in 71 in FIG. 7 . In particular, the bottom of most containers is not flat, such as plastic Coke bottles. When the ends are connected, a certain gap will be formed. The grayscale of the pixels in the gap is very different from that of the liquid pixels, which is easy to identify. In addition, the bottom of the glass bottle has prominent grayscale characteristics, its grayscale is obviously lower, and it is far away from the range of liquid organic matter in material value (even the main part of the glass bottle is dominated by the material properties of liquid organic matter ), and form a short thick line that is nearly perpendicular to the advancing direction of the pallet, as shown in 73 in FIG.

The bottleneck/cap is usually much thinner than the bottle body (the bottleneck 72 shown in FIG. 7 ), and the analysis is carried out with the direction of the tray advancing as the axis and each column of the image as the unit. First calculate the width of the liquid area between the upper and lower boundaries of each column; if the width of each column in several consecutive columns is narrow enough and the width shows a smooth and thinning trend, then these columns can be regarded as bottlenecks/caps; In particular, when the widths of adjacent columns of liquid regions change significantly, this generally means that a bottle bottom has been encountered. Also, for side-illuminated images, the projected position of the bottleneck should be higher than the slot in the tray, which is also a distinctive feature. The gaps in the container image segment, the bottom of the bottle, and the areas between the neck/cap are first regarded as a container.

Of course, the above criteria cannot completely separate all liquid objects connected, for example, when some liquid objects with no obvious bottleneck and similar width are connected end to end and can be seamlessly docked by chance. To this end, a material value expression model in the liquid region on each column i of the image is defined Divide the pixels of the container area on this column into 6 equal parts, Indicates the average material value of pixels in the range of the first copy here, and so on. Then calculate MD(i)=|M _i -M _i+Δi |, which represents the difference in the distribution of material values between two adjacent columns i and i+Δi. The position where MD(i) undergoes a sufficiently large mutation is considered to encounter a new container. Finally, for each segmented container interior, a plurality of detection slices are collected, which respectively correspond to a column of V1/V2 perspective images. It is required that the intervals between these slices are as equal as possible, and the MD values of each column before and after each slice are very small, so as to ensure the stability of the liquid properties near the detected position.

The so-called slice is the section of the container that the ray fan passes through when the ray fan of the ray source scans the container perpendicular to the conveying direction; The gray value of a column of pixels is generated in the perspective image, and each pixel corresponds to the projected gray value of a ray; the pixel values of the same slice in the corresponding columns of the two perspective images respectively form the projection curves of the two perspectives of the slice . In the present invention, the danger of the liquid in the container will be judged based on the section of the container, that is, the cross section, and the corresponding projection curves from two angles of view.

Next, for each selected detection slice, the container attribute estimation module 4 is executed; in the container attribute estimation module 4, it is first necessary to judge whether it is a high-density container, and the method is to calculate the material value of the container and the R component of the RGB pseudo-color image value, a container material feature vector is obtained by a statistical method, and the vector contains 4 feature variables. The first two characteristic variables are based on the detection slice, and the mean value of the material value and the difference between the mean value and the mean value of the material value of the neighborhood of the same size in the middle area of the container are calculated in the neighborhood belonging to the edge of the container. The latter two characteristic variables are based on The edge mean of the R component calculated by the same method and the difference between the mean and the middle mean of the container. Normalize this feature matrix so that different feature quantization values are comparable. Through a lot of training, the priori can distinguish plastic containers and high-density containers in this characteristic matrix. The threshold value is expressed as an interval, which is greater than the upper limit of the interval as a high-density container, and less than the lower limit of the interval is a plastic container. In the interval, the container material should be further identified by the gray-scale geometric features of the detection slice. The specific method is that the high-density container has gray valley points near the edges of both sides of the projection curve of the detection slice. This is because according to the projection relationship, the path of the relevant rays on the outer wall of the high-density container is long, the absorption is large, and the grayscale The value is low, and plastic containers do not have this feature.

After it is judged as a high-density container, it is necessary to judge the type of container cross-section shape; the type of container cross-section is mainly divided into 4 types: round, oval, square, and rectangular. Note that these are only general categories, such as hexagonal or An octagon can be approximated as a circle or an ellipse. As shown in Figure 4, the rectangle in the center of the figure represents a section of the container; four rays emitted from two ray sources tangent to the section form the circumscribed polygon ABCD of the section. First of all, use the circumscribed polygon of the section of the container to determine whether the aspect ratio of the container is approximately 1:1 by whether the inscribed circle of the polygon can be found. If so, it is a circle or a square, otherwise it is an ellipse or a rectangle. Furthermore, pattern recognition is performed using the lower half of the projection curve of the slice's V2 viewing angle. Fig. 5 is a schematic diagram of the lower half of the projection curve of the container detection slice in the side view angle V2. Fig. 5(a) corresponds to a circular container, and Fig. 5(b) corresponds to a square container. There is a significant difference between the two; in the projection curve, The left direction of the grayscale axis represents high grayscale (less X-ray attenuation), whereas the right direction represents low grayscale (large X-ray attenuation). Appropriate pattern recognition methods, such as specimen matching, can be used to distinguish whether the projection curve of the current section is closer to a circular container or a square container; for containers whose aspect ratio is obviously not 1:1, similar Pattern recognition method to judge whether it is an oval container or a rectangular container.

The reason for judging the classification of container sections is to obtain sufficient prior knowledge in the following work of high-density container section reconstruction module 6 . The space plane where the section of the container is located is equally divided into a large number of grids of appropriate scale. Generally speaking, the reconstruction of the container section starts from the circumscribed polygon of the section (as shown in Figure 4), and gradually marks the grids that should belong to the air in the circumscribed polygon as air, and finally the remaining grids that belong to the container form the shape of the container section . The method of the present invention introduces new information here—classification of vessel sections. For example, when the section of a container is classified as an ellipse, using the four sides of the circumscribed polygon, a formula for forming an ellipse is formed by data fitting. Enlarge the ellipse moderately, superimpose it with the circumscribed polygon, and the common grid of the two forms the initial shape of the container section, which greatly reduces the number of grids that need to be marked as air, reduces the variables of the subsequent method, and increases The reliability of cross-sectional reconstruction is improved. As shown in Figure 6, assuming that the section of the container has been classified as an ellipse, the grid points in the gray part of the figure are the initial shape of the section. Compared with starting the analysis directly with the circumscribed polygon ABCD, many variables are reduced. Similar processing can also be performed if the section type is judged to be rectangular or circular.

Next, the high-density container section reconstruction module 6 is firstly introduced, the purpose of which is to obtain the actual shape of the container section through the initial shape of the above section. Herein, the present invention introduces a new iterative optimization method—an optimization method based on nonlinear least squares and FR conjugate gradient method. The overall objective function of the optimization method is shown in formula (1):

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ~</span>\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ~</span>\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; wall</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; wall</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; liq</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; liq</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Where M is the number of all rays of two viewing angles in the current slice, A _i is the attenuation of ray i (this is obtained according to the grayscale of ray projection), is the length of the ray i passing through the outer wall of the container in the section, is the length of the ray i passing through the liquid in the container in the section, μ _wall is the X-ray attenuation coefficient of the outer wall, and μ _liq is the X-ray attenuation coefficient of the liquid. In addition, the height S of the liquid level of the liquid, the thickness of the left side of the container d ₁ , the thickness of the upper side of the container d ₂ , the thickness of the right side of the container d ₃ , and the thickness of the lower side of the container d ₄ are all parameters in the optimization model. appear directly in formula (1), but obviously, their changes in each iteration will affect the shape of the container section. For details, please refer to Figure 8. The dark gray part in the figure is the outer wall of the high-density container, the light gray part represents the liquid, 81 represents d ₁ , 82 represents d ₂ , 83 represents d ₃ , 84 represents d ₄ , and 85 represents the liquid level. The surface height is S. The reason for designing four wall thickness parameters is mainly because the wall thickness of the glass container is not uniform, and setting the wall thickness of each bottle to be unequal can reduce the dispersion of the final result. In the following, t is used to represent the round of optimization iterations.

When t=1, modify the above seven parameters μ _wall , μ _liq , S, d ₁ , d ₂ , d ₃ and d ₄ , and calculate the matrix shown in formula (2) according to the nonlinear least square method J,

$\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; =</span>\left[\begin{array}{ccccccc}\frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; wall</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; liq</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; S</span>}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}\\ <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>\\ <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>\\ <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>\\ \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; wall</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; liq</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; S</span>}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Note that although S, d ₁ , d ₂ , d ₃ and d ₄ do not appear in the calculation formulas of each R(i), the step size Δ of each of these five variables can be designed. In the current cross-sectional shape, when the When S, d ₁ , d ₂ , d ₃ and d ₄ independently increase and decrease Δ, the value of each R(i) changes, so that the partial derivatives at the back of the matrix in formula (2) can be calculated. In addition, it is necessary to calculate the formula (3) shown in .

$<span\; class="notranslate">\; \&dtri;</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>{\left[\begin{array}{ccccccc}\frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; wall</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; liq</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; S</span>}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}& \frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}\end{array}\right]}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Obviously, the calculation of formula (3) can be carried out according to the content of formula (2), for example available by The combination is calculated. Then calculate the correction amount of each parameter Among them, s is a 7-dimensional vector, and its 7 vector elements represent the correction amount of 7 parameters in round t, and adjust the identity of each grid inside the section according to the new values of these parameters.

In the t+1 round, the shape of each container section will be corrected according to the FR conjugate gradient method, that is, the identity of each grid in the container section, including air, outer wall, and liquid. Correct the identity of the container grid on the outermost layer of the cross-section and the "air" grid on the outer layer of these container grids. Let there be a total of N of these two types of grids p ₁ to p _N. calculate Similarly, although p ₁ ~ p _N do not appear in the calculation formula of F, if the identity of each grid p _j is reversed, for example, container => air, or air => container, the value of F will change, so that The partial derivatives of each item in the formula can be calculated. for If p _j is large enough, its identity is reversed. Once the shape of the container section is adjusted, because the parameters of S, d ₁ , d ₂ , d ₃ and d ₄ should be kept unchanged at this time, the identities of the grids inside the section also need to undergo a new round of correction, for example After some of the grid-identities belonging to the outer wall of the container are turned into air, some of the corresponding identities in the position are the grids of liquid, and their identities will become the outer wall of the container.

The two processes of round t and round t+1 are alternately performed, and the model parameters are corrected first, and then the shape of the section is corrected until the improvement of the objective function F is less than a certain threshold.

For the low-density container section reconstruction module 5, a similar principle is used, but the process is simpler because there is no outer wall factor. Simply use the circumscribed polygon as the starting point of the section shape, and then optimize the objective function shown in formula (4).

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ~</span>\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ~</span>\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; liq</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; liq</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

When t=1, the above-mentioned one parameter μ _liq is corrected, and the J matrix becomes a simple form of one column and M rows accordingly. In the t+1 round, still according to the FR conjugate gradient method, the shape of the container section is corrected—the identity of the outermost container grid of the section and the air grid of the outer layer of these container points are corrected, the method Similar, no more details. What needs to be emphasized is that although it is not necessary to estimate the section type, it is also necessary to analyze the local morphological characteristics of the container projection. For example, if there is a liquid surface, the projection curve of the angle of view V2 corresponds to the upper part of the container, and the projection effect shown in Figure 9 will appear; In the projection curve, the left direction of the gray scale axis represents high gray scale (less X-ray attenuation), the right direction of gray scale axis represents low gray scale (more X-ray attenuation), and the projection curve marked by the small circle in the upper right corner The segment corresponds to those rays that pass through the liquid surface, which is easy to identify; since the rays corresponding to the uppermost part of the projection curve pass through the liquid surface, the outer container grid p _j located on these rays will be in each round On the basis of , a greater weight is added, so that the identity of the grid such as p _j can be quickly changed into air, so that the shape of the liquid surface can be formed as soon as possible.

After obtaining the cross-sectional shape of the container slice, in the decision-making module 7, various characteristics such as the material, density, volume, and container wall thickness of the liquid can be calculated using existing methods in the art, so each container slice can be obtained 1 multidimensional feature vector. In the training phase, for the container slices loaded with explosive liquids, the obtained feature vectors are added to the set T ^p ; for other safe liquid container slices, the obtained feature vectors are added to the set T ⁿ . According to the pattern recognition theory, a suitable classifier can be used to distinguish the vector belonging to T ^p from the vector belonging to T ⁿ . The present invention selects a support vector machine to distinguish containers loaded with explosive liquids and safe liquids; other mature classifiers in the field of pattern recognition, such as decision trees or neural networks, can also perform classification.

Finally, if according to the above-mentioned multiple characteristics, for the liquid container with slices judged as dangerous by the classifier, the integrated processing computer 28 as shown in Figure 2 will display the results on the screen of the security inspector according to the aforementioned detection results and container segmentation results. Draw an alarm frame at the corresponding position.

To sum up, the image classification module 1 enables the method and device for safety inspection of liquid articles of the present invention to be compatible with the actual demand for alternate detection of packages, personal items and liquid articles; the image segmentation module 2 enables security inspectors to Flexible placement of liquid and non-liquid items in the inspection tray, especially when the number of liquid items carried by passengers is small or orders are placed, other personal items can also be placed in the same tray, thereby improving the efficiency of the security inspection process; detection location selection Module 3 can divide the liquid items connected head to tail, so that security inspectors can place liquid items in the inspection tray less restricted, without deliberately keeping the distance between liquid items; container attribute estimation module 4, obtained a wealth of information about The prior knowledge of the container; the low-density container section reconstruction module 5 and the high-density container section reconstruction module 6 can accurately obtain the cross-sectional shape of the container when there are only 2 viewing angle data, and the prior knowledge of the container plays a crucial role. It plays an important role; the extraction of various features of the container and the use of classifiers in the decision-making module 7 make the system no longer limited to visually comparing the values of liquid materials and densities when making decisions, but from a large Abstract classification rules are mined from large-scale training data to achieve better identification of dangerous liquids.

A liquid article safety inspection device according to the present invention is shown in FIG. Bottom-illuminated angle of view unit V1; wherein, the side-illuminated angle of view unit V2 includes a side-illuminated X-ray source 15 and a second detector 13; the middle bottom-illuminated angle of view unit V1 includes a central bottom-illuminated X-ray source 22 and a first detector 16.

As shown in Figure 4, the above-mentioned side-illuminated X-ray source 15 and the middle bottom-illuminated X-ray source 22 are respectively located in different orientations of the transport channel 17, that is, in the front view direction of the transport channel 17, the side-illuminated X-ray source 15 is arranged on the side of the conveying channel 17, and the central bottom-illuminated X-ray source 22 is arranged below the middle of the conveying channel 17, so that two angles at the side and below the middle of the conveying channel 17 form an orthogonal dual-view layout model.

Similarly, corresponding to the above-mentioned central bottom-illuminated X-ray source 22 and side-illuminated X-ray source 15 , the corresponding first detector 16 and second detector 13 are also attached to different positions of the conveying channel 17 . The first and second detectors are both dual-energy (ie, high and low energy) detectors, so as to provide material property information of the inspected liquid for the liquid detection algorithm.

The first detector 16 is a door-type detector, and the second detector 13 is an L-type detector.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the disclosure of the present invention are all It should be covered within the protection scope of the claims of the present invention.

Claims (8)

1. A method for safety inspection of liquid articles, characterized in that the method first distinguishes the liquid articles in the package and the tray, the tray and the non-liquid articles, and divides a plurality of liquid articles connected head to tail, and then selects a suitable liquid container In each slice, first analyze the properties of the container, and then reconstruct the cross-section of the dual-view container based on the properties of the container, use the reconstructed cross-section to obtain various characteristics such as the material and density of the liquid, and use an appropriate classifier For example, a support vector machine can be used to judge whether a liquid is dangerous; the method for safety inspection of liquid objects can be applied to various channel-type X-ray safety inspection equipment with not less than 2 viewing angles and side-viewing viewing angles. 2. A method for safety inspection of liquid articles according to claim 1, characterized in that, only using dual-view X-ray images to reconstruct the section of the container, first analyzing the properties of the container, including information such as the material of the container, the basic type of the section of the container, etc. ; and then combine the nonlinear least squares method with the conjugate gradient method to realize the reconstruction of the cross-sectional shape of the container under the condition of two viewing angles; Channel type X-ray security inspection equipment. 3. A method for security inspection of liquid goods according to claim 1, characterized in that, according to the content of the X-ray image, it is distinguished whether it is a package or a tray, and the method for distinguishing between a package and a tray can be applied to include single-view, multi-view Various traditional channel X-ray safety inspection equipment including equipment. 4. A method for safety inspection of liquid articles according to claim 1, characterized in that, according to the shape, grayscale, and material information of the article area in the tray, it is distinguished whether each article in the tray is a liquid article or a non-liquid article, and said The method for distinguishing liquid items from non-liquid items can be applied to various traditional channel-type X-ray security inspection equipment including single-view and multi-view equipment. 5. A method for safety inspection of liquid articles according to claim 1, characterized in that, according to the shape, gray scale, and material information of the liquid article area, a plurality of liquid articles connected head to tail are divided, and finally each liquid article The location of the detection slice is selected on the container for subsequent detection; the segmentation method of the head-to-tail connected liquid item and the selection method of the detection slice position can be applied to various traditional channel X-rays including single-view and multi-view equipment Safety inspection equipment. 6. The method for security inspection of liquid items according to claim 1, wherein after acquiring the dual-view images of liquid items, liquid detection specifically comprises the following steps: 1) Image classification: Divide the collected security images into two categories, one is package images, and the other is tray images, in which liquid items, belts, wallets, mobile phones, coats, ladies’ handbags, etc. may be placed in the trays. Items subject to separate security checks; 2) Image segmentation: Non-liquid items and liquids are allowed to be mixed on the same tray for security inspection, but liquid items and other items are required to be imaged without overlapping in any viewing angle, and the images of 1 tray and 2 viewing angles are divided into multiple segments, Each segment of the image is either full of liquid items, called a container image segment, or full of non-liquid items, called a non-liquid image segment; 3) Detection position selection: multiple liquid items in the same container image segment in the tray are likely to be connected end to end, and multiple containers that may be connected end to end are separated according to material property changes, volume, local shape, etc. Then find several slice positions with stable local material properties in each container for subsequent detection; 4) Container attribute estimation: First, determine whether the liquid container is a high-density container. If it is a high-density container, it is necessary to determine the basic shape classification of the container section. If it is a low-density container, it is also necessary to analyze the local morphological characteristics of the container projection. Obtain various necessary container information; 5) Section reconstruction of low-density containers: For low-density containers such as plastic bottles, model the section of the container, propose an algorithm combining nonlinear least squares method and conjugate gradient method, iteratively estimate whether each grid in the section belongs to the container area, Finally, the cross-sectional shape is obtained; 6) Sectional reconstruction of high-density containers: For high-density containers such as glass bottles, the section of the container is modeled, and an algorithm of nonlinear least squares method combined with the conjugate gradient method is proposed to iteratively estimate whether each grid in the section belongs to the container, liquid, etc. or air, and finally obtain the cross-sectional shape, utilizing the container attribute information obtained in step 4) in this process; 7) Decision-making: Based on the cross-sectional shape of the container obtained in step 5) or 6), calculate various characteristics such as the material and density of the liquid, and use the support vector machine as a classifier to give a judgment on whether the liquid is dangerous; Finally, as long as one slice of a container is judged as dangerous by step 7), the system will give an alarm to the dangerous container. 7. A liquid article safety inspection device, characterized in that it adopts a brand-new dual-view layout design, and the two view angles are composed of 2 sets of X-ray sources and detectors, and each set of X-ray sources and detectors is called A detection unit, wherein the detection unit is a middle bottom viewing angle unit V1 and a side viewing angle unit V2 installed in the conveying channel; the middle bottom viewing angle unit V1 includes a middle bottom X-ray source and a first detection device; the side-view unit V2 includes a side-illuminated X-ray source and a second detector. 8. The liquid article safety inspection device according to claim 7, wherein the two viewing angles are truly orthogonal.