A real-time automatic detection method and device for explosive devices based on X-ray digital images

Abstract

The present invention provides a method and device for real-time automatic detection of explosives based on X-ray digital images. The detection method is as follows: (1) selecting two types of images, one with ammunition and one without ammunition in the front and side views, from the preprocessed images as prior information to establish an image library, and locating the feature detection area based on the characteristics of the two types of explosives images; (2) extracting the feature values ​​of the image detection area for analysis, establishing the clustering centers of the two types of images, and determining the optimal threshold segmentation index based on maximizing the feature difference between the classes; (3) performing an explosives image test, and further optimizing the setting of the index and threshold in step (2) based on the product detection requirements, to achieve real-time automatic detection of explosives based on X-rays.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and specifically relates to a method and device for real-time automatic detection of explosives based on X-ray digital images. The method is based on X-ray imaging technology, locates the detection area by establishing a priori information image library of explosives products; extracts the characteristic values ​​of the image detection area for analysis, establishes the clustering center of two types of images, and determines the optimal threshold segmentation index based on the maximization of the difference between the characteristics of the classes; realizes the determination of the image judgment threshold that meets the requirements of explosives detection, and achieves the purpose of real-time automatic detection of explosives. Background Art

[0002] In recent years, X-ray imaging has been widely used in industry, medicine, and aerospace. X-ray industrial nondestructive testing based on film imaging technology requires manual judgment, lacks real-time performance, and causes a lot of resource waste. With the development of digital imaging technology, it provides a foundation for X-ray industrial applications to improve real-time dynamic and rapid detection links.

[0003] With the development of the aerospace industry, pyrotechnic products are key units on spacecraft and satellites. The detection of their charge status is related to the success or failure of the product or even the entire model. Using X-ray digital imaging technology to confirm the internal quality of completed assembly products is an effective means to ensure aerospace quality.

[0004] At present, the interpretation of pyrotechnic product inspection images mainly relies on the visual judgment of inspectors, which has the problems of low detection sensitivity, slow detection speed, and the test results are easily affected by the mental state and subjective factors of the inspectors. Some quantitative parameters reflected in the image cannot be determined manually in a short time, and even missed detection may occur. Therefore, it is urgent to establish a fast and reliable fully automatic X-ray detection method for pyrotechnic products.

[0005] The versatility of X-ray sources and the differences in attenuation characteristics of different materials, the perspective of X-ray images and the clutter caused by overlapping and compression of objects have increased the difficulty of target detection in X-ray digital images. The main difficulties in X-ray target detection are: X-ray detection images generally have problems such as blurred edges, and the use of edge detection algorithms can easily lead to missed or wrong target detection; the application of deep learning in industrial automation detection requires the accumulation of a large number of data sets, and the application level needs further evaluation. Summary of the invention

[0006] The technical problem solved by the present invention is to overcome the shortcomings of the existing technology based on X-ray target detection in the detection of pyrotechnics, provide a real-time automatic detection method and device for pyrotechnics based on X-ray digital images, and realize real-time automatic detection of pyrotechnics based on X-ray digital images.

[0007] The technical solution of the present invention is: a real-time automatic detection method of explosive devices based on X-ray digital images, the steps are as follows:

[0008] (1) Obtain the original image I L , to I L Perform preprocessing to obtain the processed X-ray image I u ;

[0009] (2) The preprocessed X-ray image I u According to whether the pyrotechnics have ammunition or not, the image library is divided into two categories, A and B, where A is with ammunition and B is without ammunition;

[0010] (3) Compare the image with ammunition in A and the image without ammunition in B, extract the changed parts of the two images to determine the detection range I h , the matrix size is M*N, and the detection range for the A and B library images is expressed as

[0011] (4) Calculate the mean value μ(I) of the image detection area in the image library ht )、Standard deviation std(I ht ), the gradient matrix mean grad(I ht ), entropy(I ht ) as the eigenvalue, and the eigenvalues ​​of class A and class B images are recorded as

[0012] (5) Determine the eigenvalue clustering centers of the two types of images A and B respectively

[0013] (6) Optimize the image feature value and determine the optimal threshold segmentation index: The corresponding eigenvalues ​​of class A and class B images are recorded as Finally, the modified A and B eigenvalue cluster centers are calculated as follows:

[0014] (7) Input the image to be detected I in , execute step (6) to obtain its eigenvalue judge If Δf<0, then I in It belongs to Category A, that is, it has ammunition; otherwise, it is I in It belongs to category B, which means no ammunition.

[0015] The pretreatment process of step (1) is as follows:

[0016] 11) to I L Perform median filtering, and the filtered output image is I med ;

[0017] 12) Using morphological enhancement method to med Processing is performed to obtain the processed X-ray image I u ,

[0018]

[0019]

[0020] I u =I med +I v -I w ;

[0021] Where H is the structural element, I v For I med The image after top-hat transformation, I w For I med The image after bottom-hat transformation, I u is the output X-ray grayscale image.

[0022] In step (4), the mean value is calculated as follows:

[0023]

[0024] Where μ(I ht ) is image I ht The mean value of each pixel in , i = 1, 2, ... M is the row pixel number of the image, j = 1, 2, ... N is the column pixel number of the image.

[0025] In step (4), the standard deviation is calculated as follows:

[0026]

[0027] Where std(I ht ) is image I ht The standard deviation of all pixels in , i = 1, 2, ... M is the row pixel number of the image, j = 1, 2, ... N is the column pixel number of the image.

[0028] In step (4), the method for calculating the mean of the gradient matrix is:

[0029]

[0030] where grad(I ht ) is image I ht The mean value of the gradient matrix; i = 1, 2, ... M is the row pixel number of the image, j = 1, 2, ... N is the column pixel number of the image;

[0031]

[0032]

[0033] in, For image I ht The gradient matrix along the row and column directions, for digital images, is the difference operation of adjacent pixels in that direction.

[0034] In step (4), the entropy is calculated as follows:

[0035]

[0036] Where entropy(I ht ) is image I ht Entropy, n = 0, 1, ... I ht (max) is image I ht The gray value, P n is the probability of gray value n appearing in the image.

[0037] The step (5) determines the clustering centers of the feature values ​​of the two types of images A and B The specific process is:

[0038]

[0039]

[0040] in, are the cluster centers of A and B, and g1 and g2 are the number of pictures in the two categories.

[0041] The specific process of optimizing the image feature value and determining the optimal threshold segmentation index in step (6) is as follows:

[0042]

[0043] Among them, f is the threshold segmentation index, which is a function of the eigenvalue μ, std, entropy, and grad. By adjusting the feature parameters, when f reaches the maximum value, the optimal threshold segmentation index is obtained:

[0044] The specific process of obtaining the modified A and B eigenvalue cluster centers in step (6) is as follows:

[0045]

[0046]

[0047] in, It is the cluster center of A and B eigenvalues ​​after index correction.

[0048] A real-time automatic detection device for explosives based on X-ray digital images, comprising an image acquisition module, a threshold determination module and an explosives determination module;

[0049] The image acquisition module is equipped with an X-ray detection optical lens and a CCD camera. After the collected analog visible image signal is converted by the image acquisition card A / D, it is transmitted to the threshold determination module for image processing after passing through the computer image generation unit;

[0050] The threshold determination module includes a preprocessing unit, an image classification unit and a threshold calculation unit; the preprocessing unit first processes the input image pair I L Perform median filtering to obtain I med , the processed X-ray image I is obtained by using the morphological image enhancement method u The image classification unit divides the image into two categories, A and B, based on prior information, which represent the two situations of ammunition and no ammunition respectively, and extracts the detection range I of the pyrotechnics real-time automatic detection device to obtain the pyrotechnics charge situation in the image h ; The threshold calculation unit calculates the mean, standard deviation, gradient, and entropy of the image detection area in the picture library as feature values. ht , select the initial eigenvalues ​​and establish the A and B eigenvalue cluster centers respectively According to the difference in Euclidean distance between the characteristic value and the center point of the two types of image clusters, the threshold segmentation index is determined, and the optimal threshold determination index is output;

[0051] The explosive device identification module pre-processes the acquired image, combines the threshold judgment index to determine whether the input image is in a charged state, and outputs the judgment result.

[0052] The advantages of the present invention compared with the prior art are:

[0053] The present invention first locates the detection area and uses the local overall eigenvalue as an indicator, thereby avoiding the problem of difficult detection of target edges based on X-ray images and improving detection efficiency; the initial eigenvalue is extracted, which requires fewer training samples than deep learning and makes the indicator evaluation process clearer.

[0054] The present invention adopts the morphological image enhancement method and designs the morphological structure in combination with the target information and X-ray noise characteristics, which can enhance the contrast of the target information while eliminating interference, and is conducive to further target detection; the index evaluation standard is constructed using prior information, which lays a good foundation for designing an automatic detection method and device for pyrotechnics that meets application requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the structure of the device of the present invention;

[0056] Figure 2Schematic diagram of a flow chart of an embodiment of the present invention.

[0057] Figure 3 These are the median filtering results for five sets of experimental data.

[0058] Figure 4 This is the result after morphological operation.

[0059] Figure 5 Schematic diagram for establishing image libraries A and B. DETAILED DESCRIPTION

[0060] The basic idea of ​​the present invention is as follows: In view of the particularity of explosive charge detection of pyrotechnic products, a real-time automatic detection method and device for pyrotechnic products based on X-ray digital images are designed. Firstly, the morphological image enhancement method is adopted to eliminate interference information and enhance target contrast while keeping the image information of the pyrotechnic product unchanged, so as to facilitate further detection. In the threshold determination link, the characteristics of the prior information are analyzed to construct the optimal threshold that meets the conditions, so as to realize real-time automatic detection of pyrotechnic products.

[0061] The present invention is further described in detail below in conjunction with the accompanying drawings. Figure 1 The structure diagram of a real-time automatic detection device for explosives based on X-ray digital images is shown, which includes an image acquisition module, a threshold determination module and an explosives determination module.

[0062] The image acquisition module is equipped with an X-ray detection optical lens and a CCD camera. The collected analog visible image signal is converted by an image acquisition card A / D and then transmitted to the threshold determination module for image processing after passing through a computer image generation unit.

[0063] The threshold determination module includes a preprocessing unit, an image classification unit and a threshold calculation unit. The preprocessing unit first processes the input image pair I L Perform median filtering to obtain I med , and then use the morphological image enhancement method to obtain the processed X-ray image I u The image classification unit divides the image into two categories, A and B, based on prior information, which represent the two situations of ammunition and no ammunition respectively, and extracts the detection range I of the pyrotechnics real-time automatic detection device to obtain the pyrotechnics charge situation in the image h ; The threshold calculation unit first calculates the mean, standard deviation, gradient, and entropy of the image detection area in the picture library as feature values. ht , select the initial eigenvalue μ(I ht )、std(I ht )、entropy(I ht )、grad(I ht ), calculate the cluster centers of the two eigenvalues ​​A and B respectively According to the difference in Euclidean distance between the feature value and the center point of the two types of image clusters, the optimal threshold segmentation index is output. And the modified A and B eigenvalue cluster centers

[0064] The explosive device identification module pre-processes the acquired image, combines the threshold judgment index to determine whether the input image is in a charged state, and outputs the judgment result.

[0065] The processing flow of the embodiment of the present invention is as follows: Figure 2 As shown in the flowchart, the specific implementation of the present invention is as follows:

[0066] (1) Obtain the original image I L , to I L Perform preprocessing, including the following operations:

[0067] (1.1) to I L Perform median filtering, and the filtered output image is I med ;

[0068] (1.2) Using morphological enhancement method to med Processing is performed to obtain the processed X-ray image I u ,

[0069]

[0070]

[0071] I u =I med +I v -I w

[0072] Where H is the structural element, I v For I med The image after top-hat transformation, I w For I med The image after bottom-hat transformation, I u is the output X-ray grayscale image.

[0073] (2) The pre-processed X-ray images are divided into image libraries A and B according to whether the pyrotechnics have ammunition stored or not, where A is for those with ammunition and B is for those without ammunition.

[0074] (3) Compare the image with ammunition in A and the image without ammunition in B, extract the changed parts of the two images to determine the detection range I h , the matrix size is M*N, and the detection range for the A and B library images is expressed as

[0075] (4) Calculate the mean, standard deviation, gradient matrix mean, and entropy of the image detection area in the image library as eigenvalues. ht have:

[0076]

[0077]

[0078]

[0079]

[0080] Where μ(I ht ) is image I ht The mean value of each pixel in the image, i = 1, 2, ... M is the row pixel number of the image, j = 1, 2, ... N is the column pixel number of the image, std(I ht ) is image I ht The standard deviation of all pixels in the entropy(I ht ) is image I ht Entropy, n = 0, 1, ... I ht (max) is image I ht The gray value, P n is the probability of gray value n appearing in the image, grad(I ht ) is image I ht The mean of the gradient matrix,

[0081]

[0082]

[0083] in, For image I ht The gradient matrix along the row and column directions, for digital images, is the difference operation of adjacent pixels in that direction.

[0084] (5) Determine the clustering centers of the feature values ​​of the two types of images A and B,

[0085]

[0086]

[0087] in, are the cluster centers of A and B, and g1 and g2 are the number of pictures in the two categories.

[0088] (6) The feature value of class A image is The characteristic value of class B image is The threshold segmentation index is a function of the eigenvalue μ, std, entropy, and grad.

[0089] According to the difference in the Euclidean distance between the feature value and the center point of the two types of image clusters, the feature parameters are adjusted. When f reaches the maximum value, the optimal threshold segmentation index is obtained: At this time, the characteristic value of class A image is The characteristic value of class B image is The eigenvalue clustering centers of A and B

[0090]

[0091]

[0092] in, It is the cluster center of A and B eigenvalues ​​after index correction.

[0093] (7) Input the image to be detected I in , calculate the feature value of the image to be detected according to the method in (6) judge If Δf<0, then I in It belongs to Category A, that is, it has ammunition; otherwise, it is I in It belongs to Class B, i.e., no ammunition, and realizes a real-time automatic detection method and device for pyrotechnics based on X-ray digital images.

[0094] The effects of the present invention will be further described below through schematic diagrams.

[0095] to I L Perform median filtering, and the filtered output image is I med ; The median filtering results of five sets of experimental data are as follows Figure 3 As shown;

[0096] The morphological enhancement method was used to med Processing is performed to obtain the processed X-ray image I u ,

[0097]

[0098]

[0099] I u =I med +I v -I w

[0100] Where H is the structural element, I v For I med The image after top-hat transformation, I w For Imed The image after bottom-hat transformation, I u is the output X-ray grayscale image. The result after morphological operation is as follows Figure 4 As shown;

[0101] (3) The pre-processed X-ray images are classified according to the storage conditions of explosive devices, and image libraries A and B are established respectively, where A is for explosive devices with ammunition and B is for explosive devices without ammunition, as shown in the schematic diagram. Figure 5 As shown;

[0102] (4) Examples of calculation results of each group of experimental data are shown in the following table:

[0103]

[0104] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solution of the present invention by using the technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A real-time automatic detection method for explosive devices based on X-ray digital images, characterized in that Here are the steps: (1) Obtain the original image I L , to I L Perform preprocessing to obtain the processed X-ray image I u ; (2) The preprocessed X-ray image I u According to whether the pyrotechnics have ammunition or not, the image library is divided into two categories, A and B, where A is with ammunition and B is without ammunition; (3) Compare the image with ammunition in A and the image without ammunition in B, extract the changed parts of the two images to determine the detection range I h , the matrix size is M*N, and the detection range for the A and B library images is expressed as (4) Calculate the mean value μ(I) of the image detection area in the image library ht )、Standard deviation std(I ht ), the gradient matrix mean grad(I ht ), entropy(I ht ) as the eigenvalue, and the eigenvalues ​​of class A and class B images are recorded as (5) Determine the eigenvalue clustering centers of the two types of images A and B respectively (6) Optimize the image feature value and determine the optimal threshold segmentation index: The corresponding eigenvalues ​​of class A and class B images are recorded as Finally, the modified A and B eigenvalue cluster centers are calculated as follows: The specific process of optimizing the image feature value and determining the optimal threshold segmentation index in step (6) is as follows: Among them, f is the threshold segmentation index, which is a function of the eigenvalue μ, std, entropy, and grad. By adjusting the feature parameters, when f reaches the maximum value, the optimal threshold segmentation index is obtained: (7) Input the image to be detected I in , execute step (6) to obtain its eigenvalue judge If Δf<0, then I in It belongs to Category A, that is, it has ammunition; otherwise, it is I in It belongs to category B, which means no ammunition.

2. The method for real-time automatic detection of explosive devices based on X-ray digital images according to claim 1 is characterized in that: The pretreatment process of step (1) is as follows: 11) to I L Perform median filtering, and the filtered output image is I med ; 12) Using morphological enhancement method to med Processing is performed to obtain the processed X-ray image I u , I u =I med +I v -I w ; Where H is the structural element, I v For I med The image after top-hat transformation, I w For I med The image after bottom-hat transformation, I u is the output X-ray grayscale image.

3. The method for real-time automatic detection of explosive devices based on X-ray digital images according to claim 1 is characterized in that: In step (4), the mean value is calculated as follows: Where μ(I ht ) is image I ht The mean value of each pixel in , i = 1, 2, ... M is the row pixel number of the image, j = 1, 2, ... N is the column pixel number of the image.

4. The method for real-time automatic detection of explosive devices based on X-ray digital images according to claim 1 is characterized in that: In step (4), the standard deviation is calculated as follows: Where std(I ht ) is image I ht The standard deviation of all pixels in , i = 1, 2, ... M is the row pixel number of the image, j = 1, 2, ... N is the column pixel number of the image.

5. The method for real-time automatic detection of explosive devices based on X-ray digital images according to claim 1 is characterized in that: In step (4), the method for calculating the mean of the gradient matrix is: where grad(I ht ) is image I ht The mean value of the gradient matrix; i = 1, 2, ... M is the row pixel number of the image, j = 1, 2, ... N is the column pixel number of the image; in, For image I ht The gradient matrix along the row and column directions, for digital images, is the difference operation of adjacent pixels in that direction.

6. The method for real-time automatic detection of explosive devices based on X-ray digital images according to claim 1 is characterized in that: In step (4), the entropy is calculated as follows: Where entropy(I ht ) is image I ht Entropy, n = 0, 1, ... I ht (max) is image I ht The gray value, P n is the probability of gray value n appearing in the image.

7. The method for real-time automatic detection of explosive devices based on X-ray digital images according to claim 1 is characterized in that: The step (5) determines the clustering centers of the feature values ​​of the two types of images A and B The specific process is: in, are the cluster centers of A and B, and g1 and g2 are the number of pictures in the two categories.

8. The method for real-time automatic detection of explosive devices based on X-ray digital images according to claim 1 is characterized in that: The specific process of obtaining the modified A and B eigenvalue cluster centers in step (6) is as follows: in, It is the cluster center of A and B eigenvalues ​​after index correction.

9. An automated testing device for executing the method of claim 1, characterized in that: It includes an image acquisition module, a threshold determination module and an explosive device determination module; The image acquisition module is equipped with an X-ray detection optical lens and a CCD camera. After the collected analog visible image signal is converted by the image acquisition card A / D, it is transmitted to the threshold determination module for image processing after passing through the computer image generation unit; The threshold determination module includes a preprocessing unit, an image classification unit and a threshold calculation unit; the preprocessing unit first processes the input image pair I L Perform median filtering to obtain I med , the processed X-ray image I is obtained by using the morphological image enhancement method u The image classification unit divides the image into two categories, A and B, based on prior information, which represent the two situations of ammunition and no ammunition respectively, and extracts the detection range I of the pyrotechnics real-time automatic detection device to obtain the pyrotechnics charge situation in the image h ; The threshold calculation unit calculates the mean, standard deviation, gradient, and entropy of the image detection area in the picture library as feature values. ht , select the initial eigenvalues ​​and establish the A and B eigenvalue cluster centers respectively According to the difference in Euclidean distance between the characteristic value and the center point of the two types of image clusters, the threshold segmentation index is determined, and the optimal threshold determination index is output; The explosive device identification module pre-processes the acquired image, combines the threshold judgment index to determine whether the input image is in a charged state, and outputs the judgment result.

Images