Contraband image intelligent detection method based on double-projection radiation imaging system

Abstract

The invention belongs to the technical field of contraband detection, and particularly relates to an intelligent contraband image detection method based on a double-projection radiation imaging system. According to the scheme, a double-projection image pair containing the contraband is generated through a projection simulation algorithm, then a training data set is constructed, training of a side view image data set and a look-up image data set is achieved through the algorithm, and a contraband detection model of the side view image and the look-up image is obtained; in the side view process, different weights are distributed to the detection results of the side view image and the upward view image according to the distribution characteristics of different types of contrabands in the double-projection image, and the three-dimensional space distribution and existence probability of the contrabands in the vehicle are estimated in combination with the position distribution and size information, so that the detection process is integrally optimized, and the detection efficiency is improved. The detection speed and precision are effectively improved, and the detection rate and the recognition rate are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of contraband detection, and in particular relates to an intelligent detection method for contraband images based on a dual-projection radiation imaging system. Background Art The design system for dual-projection large vehicle detection technology usually uses two sets of radiation sources and detectors to scan and inspect large vehicles using radiation imaging technology. However, due to the complex background structure of large vehicles, the amount of information in the scanned image results is large, while the proportion of contraband in this background is very small, and it is easy to have image stacking. It is often difficult to quickly and accurately judge the entire image through manual image review. Therefore, it is necessary to use image intelligent detection technology for detection.

[0002] The main feature of the dual-projection large vehicle security inspection system is that it can utilize the correlation between the projected images in two directions. For the overlapping problem of objects encountered by a single projection, the stacked part can be effectively expanded through the projection in another direction, thereby improving the detection capability. However, the existing image intelligent detection algorithm mainly relies on a single image. It is difficult to autonomously combine the correlation between projections in two different directions for detection. Therefore, it is necessary to design different detection methods, utilize the characteristics of the dual-projection images themselves and the actual security inspection situation, optimize the detection process, improve the detection performance, increase the detection rate and recognition rate, and reduce the missed detection rate and false detection rate. Summary of the invention

[0003] In view of the above-mentioned problems existing in the prior art, the purpose of the present invention is to provide a method for intelligent detection of contraband images based on a dual-projection radiation imaging system, which can realize correlation detection of two projection images at different angles, improve the performance of the image intelligent detection network, and obtain a higher detection rate.

[0004] In order to solve the above problems, the technical solution adopted by the present invention is as follows: The method for intelligent detection of contraband images based on a dual-projection radiation imaging system of the present invention adopts a dual-projection device, wherein the dual-projection device comprises a gantry, wherein a side radiation source and a two-segment folded line detector are provided on the side of the gantry, and a bottom radiation source and a three-segment folded line detector are provided on the bottom of the gantry. The side radiation source and the bottom radiation source are discretized into spatial coordinates, and the rays are the connecting lines between the side radiation source and the two-segment folded line detector, and between the bottom radiation source and the three-segment folded line detector. The projection of the object section passing through the gantry in the linear array is obtained, and the side view and top view image pairs of the object are obtained by changing the relative positions between the side radiation source, the bottom radiation source, the two-segment folded line detector, the three-segment folded line detector and the object. The method comprises the following steps: S1: Obtaining the 3D volume data of the target contrabandV ; S2: Simulate the scanning process of the dual projection system according to the system parameters of the dual projection device, and perform the scanning on the 3D volume data obtained in step S1. V Perform a simulated scanning simulation to obtain the side view and top view image pairs of the target contraband, i.e., the simulated projection data P ; S3: superimposing the background image index of the measured vehicle to obtain a scanned image pair with the target contraband, and generating a labeling file according to the position of the target contraband in the scanned image pair; S4: Change the type, position and rotation angle of the target contraband and repeat steps S2-S3 to obtain a double-projection image pair with a label to form a training data set; S5: Using a neural network to train the training data set formed in step S4, generating two sets of detection networks for upward and side views, and designing a detection algorithm dedicated to the dual-projection system, the detection algorithm comprising: Define the detection weights for the bottom view and side view according to the type of prohibited goods being targeted ; Define the existence probability of the target contraband in different areas based on the size correlation and position distribution information between the side view and the upward view projection images ; right and The area with the maximum value is taken for analysis to locate the target area; S6: Replace the security inspection image to be detected and repeat steps S4-S5 to achieve intelligent detection of the security inspection image of contraband.

[0005] Furthermore, in step S1, the three-dimensional volume data of the target contraband is analyzed according to the linear attenuation coefficient of the material of the target contraband under the system ray energy. V Assign a value.

[0006] Furthermore, in step S2, the Siddon algorithm is used to obtain the projection of the object section in the line array, thereby obtaining the simulated projection data of the object. P .

[0007] Furthermore, in step S4, according to the category of the target contraband, the coordinate position obtained by superimposing the target contraband in the scanned image pair in step S3 is stored as a txt label file, and the scanned image pair is paired with the label file to form a dual-projection image pair containing multiple target contraband categories to obtain a training data set.

[0008] Furthermore, in step S4, the scanned image pair may be processed by random cropping, filling, scaling, or pasting.

[0009] Further, in step S5, one of the images in the upward view or the side view is selected as the main detection image according to the type of the target contraband, and the detection weights of the upward view and the side view are defined. ,when When , it means that all inspections are performed using the bottom view. , it indicates that all inspections are performed using the side view.

[0010] Furthermore, in step S5, the existence probability of the target contraband in different areas is preset, and the probability coefficient is defined , The larger the value, the higher the possibility that the target contraband exists in the area.

[0011] Further, in step S5, the target contraband is preset to appear in the actual placement position in the vehicle, and the probability coefficient is defined , The larger the value, the higher the possibility that the target contraband exists in the area.

[0012] Furthermore, in step S5, the existence probability of the target contraband in different areas is preset, and the probability coefficient is defined , The larger it is, the higher the possibility that the target contraband exists in this area; Probability of existence , The larger the value, the higher the possibility that the target contraband exists in the area.

[0013] Furthermore, the existence interval of the target contraband is located in the stepping direction of the object, and the existence interval is the intersection between the cone beam formed by the side view and the cone beam formed by the front view.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention can effectively improve the detection efficiency, realize the rapid positioning of the target area by defining and optimizing individual parameters during the detection process, and can also simultaneously improve the detection accuracy, thereby realizing more accurate detection of the target area; (2) The present invention makes full use of the advantages of the dual projection system by jointly detecting the projection images in two directions, thereby maximizing the advantages of the dual projection system, simplifying situations that are difficult to handle with traditional case methods, improving the detection rate, and solving the problem of private possession of contraband to the greatest extent, thereby ensuring the safe and stable operation of society; (3) The present invention can be widely applied to various places and working conditions that require security inspections. It can improve security inspection efficiency and detection accuracy through intelligent monitoring methods, and can be supplemented by manual means to maximize the safety of the target area. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the steps of the present invention; Figure 2 A schematic diagram of prohibited item body data in the present invention; Figure 3 It is a schematic diagram of the structure of the dual projection device in the present invention; Figure 4 A simulation projection data in the present invention P Schematic diagram of Figure 5 A schematic diagram of structural information of a vehicle in the present invention; Figure 6 It is a schematic diagram of the three-dimensional positioning principle of the dual-projection detection system in the present invention; Figure 7 A schematic diagram of a training data set including images and label files in the present invention; Figure 8 These are some of the dataset images in the present invention; In the figure: 1-gantry; 2-side radiation source; 3-two-section broken line detector; 4-bottom radiation source; 5-three-section broken line detector. DETAILED DESCRIPTION

[0016] The present invention is further described below in conjunction with specific embodiments.

[0017] The method for intelligent detection of contraband images based on a dual-projection radiation imaging system of the present invention firstly uses a special dual-projection device, such as Figure 1 As shown, the specific method is as follows.

[0018] First, after the detection process of the dual-projection vehicle inspection system is completed, two projection images of the object, one from the side and the other from above, can be obtained, and these two images can be used for joint intelligent contraband detection.

[0019] S1: According to the main application scenarios of the dual-projection vehicle security inspection system, obtain the 3D data of common contraband such as explosives, police equipment, flammable and explosive contraband, and simulated weapons V The volume data can be obtained by industrial CT or vector design drawing voxelization, and the three-dimensional volume data can be obtained according to the linear attenuation coefficient of the material of the target contraband under the system ray energy. V Assign values, such as Figure 2 As shown, a schematic diagram of contraband body data is given.

[0020] S2: According to the system geometry and system parameters of the dual-projection device, the three-dimensional volume data obtained in step S1 is processed. V Perform simulation scanning to obtain the side and top projection images of contraband, that is, simulated projection dataP , dual projection equipment such as Figure 3 As shown, it includes a gantry 1, a side radiation source 2 and a two-segment fold line detector 3 are arranged on the side of the gantry 1, and a bottom radiation source 4 and a three-segment fold line detector 5 are arranged on the bottom of the gantry 1. The side radiation source 2 and the bottom radiation source 4 are discretized into space coordinates, and the rays are the lines between the side radiation source 2 and the two-segment fold line detector 3, and the bottom radiation source 4 and the three-segment fold line detector 5. The Siddon algorithm is used to obtain the projection of the object section passing through the gantry 1 in the linear array, and the side view and top view image pairs of the object are obtained by changing the relative positions between the side radiation source 2, the bottom radiation source 4, the two-segment fold line detector 3, the three-segment fold line detector 5 and the measured object, that is, the simulated projection data P ,like Figure 4 As shown, a simulation projection data is given P Schematic diagram of .

[0021] S3: Repeat steps S1 and S2 to obtain a large number of upward / side projection image pairs of different contraband, then exponentially superimpose the contraband projection image pairs with the vehicle background to obtain scanned image pairs with the target contraband, and store the superimposed coordinate positions and contraband categories as JSON annotation files in COCO format, pair the scanned image pairs with the label JSON files to produce side and upward contraband image datasets.

[0022] S4: Change the type and position of prohibited items, and perform data enhancement by rotating the angle, randomly cropping, padding, scaling, pasting, etc., to obtain a large number of labeled double-projection scanning image pairs to form a training data set, solve the problems of category imbalance and target sparsity, and select a reasonable detector based on the aspect ratio of the prohibited items in the data set.

[0023] S5: Using a neural network to train the training data set formed in step S4, generating two sets of detection networks for upward and side views, and designing a detection algorithm dedicated to the dual-projection system, the detection algorithm comprising: Define the detection weights for the bottom view and side view according to the type of prohibited goods being targeted , used to indicate the weight of the top view and the side view during the detection process. The weights here are set manually according to different projection image pairs. For example, if the detected object is a pistol, it has a clear feature outline in the top view, while the side view is equivalent to a line. Therefore, the weight of the top view of the contraband can be increased, and the weight of the side view can be reduced. We can also adopt some rules. For example, the more symmetry planes an object has, the higher the side view weight, and the smaller the object, the higher the top view weight. When , it means that all inspections are performed using the bottom view. When , it means that all side views are used for detection. By properly configuring parameter values ​​and adjusting the detection weight coefficients of the two projection images, the detection accuracy and detection capability can be improved.

[0024] Since the position information of the two sets of images of the top view and the side view is highly correlated, for example, although the target contraband can be placed arbitrarily in space, their coordinates in the stepping direction must be strictly consistent. Therefore, when the target contraband is detected in one direction of projection, the contraband in the other projection image must appear in a specific strip. The probability of the target contraband existing in different areas is defined based on the size correlation and position distribution information between the side view and the top view projection images. ; right and The area with the maximum value is taken for analysis, and image information extraction methods such as histogram are used to perform joint detection with the grayscale histogram of the detection target to achieve rapid positioning of the target area; Among them, the probability of the existence of the target contraband in different areas is preset, and the probability coefficient is defined , The larger it is, the higher the possibility that the target contraband exists in this area; In addition to the information of the image itself, the actual placement of the object in the corresponding vehicle is also a factor that limits the focus area of ​​the projected image. For example, for large freight vehicles, the target contraband is more likely to appear in the compartment, and for small cars, the target contraband is more likely to appear in the cab. Based on the structural information of the actual vehicle, the probability of the target contraband existing in different areas is preset, and the probability coefficient is defined , The larger it is, the higher the possibility that the target contraband exists in this area, such as Figure 5 As shown, a schematic diagram of the structural information of a vehicle is given, in which the probability of the existence of dangerous goods in the area of ​​the side view is relatively high, while for the bottom view, the probability of the existence of dangerous goods in the tire part is relatively high.

[0025] Comprehensive probability information of each region of the image, calculate the total probability coefficient of the region, and the existence probability , The larger it is, the higher the possibility that the target contraband exists in this area, thereby improving the detection efficiency.

[0026] S6: Replace the security inspection image to be detected and repeat steps S4-S5 to achieve intelligent detection of the security inspection image of contraband.

[0027] According to the back-projected cone beam of the detection results, the existence interval of the contraband is located in the stepping direction. At the same time, auxiliary means such as manual review can be used to achieve the three-dimensional positioning of the contraband, such as Figure 6As shown, a schematic diagram of the three-dimensional positioning principle of the dual-projection detection system is given. From the figure, it can be seen that the contraband detection area A in the side view and the side source form a cone beam, and the contraband detection area B in the bottom view and the bottom source form a cone beam. The intersection of the two cone beams is the contraband positioning area C.

[0028] In the above step S4, if Figure 7 As shown in FIG. 1 , a schematic diagram of a training data set including images and label files in this embodiment is given. The training data set includes images of several dangerous goods of different categories, and the pixel size is fixed to 320×320. The dangerous goods are randomly distributed in different parts of different images according to their positions in three-dimensional space, such as Figure 8 Part of the dataset images are shown; the training dataset also contains the location, size and category information of several dangerous goods in the image, that is, the label file information, and the label file name is the same as the corresponding image file name.

Claims

1. An intelligent detection method for contraband images based on a dual-projection radiation imaging system, characterized in that: The method adopts a dual-projection device, which includes a gantry, a side portion of which is provided with a side radiation source and a two-segment folded line detector, and a bottom radiation source and a three-segment folded line detector are provided at the bottom of the gantry. The side radiation source and the bottom radiation source are discretized into space coordinates, and rays are the connecting lines between the side radiation source and the two-segment folded line detector, and between the bottom radiation source and the three-segment folded line detector. The projection of the object section passing through the gantry in the linear array is obtained, and the side view and top view image pairs of the object are obtained by changing the relative positions between the side radiation source, the bottom radiation source, the two-segment folded line detector, the three-segment folded line detector and the object. The method comprises the following steps: S1: Obtaining the 3D volume data of the target contraband V ; S2: Simulate the scanning process of the dual projection system according to the system parameters of the dual projection device, and perform the scanning on the 3D volume data obtained in step S1. V Perform a simulated scanning simulation to obtain the side view and top view image pairs of the target contraband, i.e., the simulated projection data P ; S3: superimposing the background image index of the measured vehicle to obtain a scanned image pair with the target contraband, and generating a labeling file according to the position of the target contraband in the scanned image pair; S4: Change the type, position and rotation angle of the target contraband and repeat steps S2-S3 to obtain a double-projection image pair with a label to form a training data set; S5: Using a neural network to train the training data set formed in step S4, generating two sets of detection networks for upward and side views, and designing a detection algorithm dedicated to the dual-projection system, the detection algorithm comprising: Define the detection weights for the bottom view and side view according to the type of prohibited goods being targeted ; Define the existence probability of the target contraband in different areas based on the size correlation and position distribution information between the side view and the upward view projection images ; right and The area with the maximum value is taken for analysis to locate the target area; S6: Replace the security inspection image to be detected and repeat steps S4-S5 to achieve intelligent detection of the security inspection image of contraband.

2. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1 is characterized in that: In step S1, the three-dimensional volume data of the target contraband is converted into the target contraband according to the linear attenuation coefficient of the material of the target contraband under the system ray energy. V Assign a value.

3. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1, characterized in that: In step S2, the Siddon algorithm is used to obtain the projection of the object section in the line array, thereby obtaining the simulated projection data of the object. P .

4. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1, characterized in that: In step S4, according to the category of the target contraband, the coordinate position obtained by superimposing the target contraband in the scanned image pair in step S3 is stored as a txt label file, and the scanned image pair is paired with the label file to form a dual-projection image pair containing multiple target contraband categories to obtain a training data set.

5. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1, characterized in that: In step S4, the scanned image pair may also be processed by random cropping, filling, scaling, or pasting.

6. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1, characterized in that: In step S5, one of the images in the upward view or the side view is selected as the main detection image according to the type of the target contraband, and the detection weights of the upward view and the side view are defined. ,when When , it means that all inspections are performed using the bottom view. , it indicates that all inspections are performed using the side view.

7. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1, characterized in that: In step S5, the probability of the target contraband in different areas is preset, and the probability coefficient is defined , The larger the value, the higher the possibility that the target contraband exists in the area.

8. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1, characterized in that: In step S5, the target contraband appears at the actual placement position in the vehicle, and the probability coefficient is defined , The larger the value, the higher the possibility that the target contraband exists in the area.

9. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 8, characterized in that: In step S5, the probability of the target contraband in different areas is preset, and the probability coefficient is defined , The larger it is, the higher the possibility that the target contraband exists in this area; Probability of existence , The larger the value, the higher the possibility that the target contraband exists in the area.

10. The method for intelligent detection of contraband images based on a dual-projection radiation imaging system according to claim 1, characterized in that: The existence interval of the target contraband is located in the stepping direction of the object, and the existence interval is the intersection between the cone beam formed by the side view and the cone beam formed by the front view.

Images