CN108764328A - The recognition methods of Terahertz image dangerous material, device, equipment and readable storage medium storing program for executing - Google Patents

Abstract

The invention discloses a kind of Terahertz image dangerous material recognition methods, and this approach includes the following steps：Terahertz image to be identified is pre-processed, and extracts the marginal information in Terahertz image；Using marginal information and default dangerous material comparison database, judge to whether there is dangerous material in Terahertz image；If it is, being split to the dangerous material edge in Terahertz image, obtains dangerous material and divide image；Sorter network is preset in dangerous material segmentation image input to handle, obtains classification information.The quantity that the Terahertz image for carrying out dangerous material identification can be reduced reduces the occupancy of computer resource, promotes recognition speed.The invention also discloses a kind of Terahertz image dangerous material identification device, equipment and readable storage medium storing program for executing, have corresponding technique effect.

Description

Terahertz image dangerous goods identification method, device, equipment and readable storage medium

technical field

The present invention relates to the technical field of safety assurance, in particular to a method, device, equipment and a readable storage medium for identifying dangerous goods in a terahertz image.

Background technique

At present, the more popular new security inspection method is to use terahertz imaging algorithm and terahertz image dangerous goods extraction and identification method to identify dangerous goods.

In the real-time security inspection process, the terahertz imaging device continuously generates terahertz images to be recognized, and then the processor uses object extraction and recognition algorithms to detect each terahertz image, and finally outputs the recognition result. Due to the complexity of the current object extraction and recognition algorithms, it takes a long time to detect terahertz images, which in turn leads to slow security checks. In practical applications, especially in security check places with a large flow of people such as stations or airports, when the boarding time of the passengers to be checked is imminent, the slow security check speed may cause passengers to fail to board the bus or board the plane smoothly.

To sum up, how to effectively improve the speed of dangerous goods recognition in terahertz images is a technical problem urgently needed to be solved by those skilled in the art.

Contents of the invention

The object of the present invention is to provide a terahertz image dangerous goods identification method, device, equipment and readable storage medium, so as to improve the speed of terahertz image dangerous goods identification.

In order to solve the above technical problems, the present invention provides the following technical solutions:

A terahertz image dangerous goods identification method, comprising:

Preprocessing the terahertz image to be identified, and extracting edge information in the terahertz image;

Using the edge information and a preset dangerous goods comparison library to determine whether there are dangerous goods in the terahertz image;

If so, segmenting the edge of the dangerous goods in the terahertz image to obtain a segmented image of the dangerous goods;

Inputting the segmented image of dangerous goods into a preset classification network for processing to obtain classification information.

Preferably, the dangerous goods segmentation image is input into a preset classification network for processing to obtain classification information, including:

The dangerous goods segmentation image is used as the input variable of the ZFNet convolutional neural network, and the classification information output by the ZFNet convolutional neural network is obtained.

Preferably, after inputting the segmented image of dangerous goods into the preset classification network for processing and obtaining classification information, the method further includes:

Projecting the position information of the dangerous goods edge and the classification information in the edge information into the terahertz image.

Preferably, the preprocessing of the terahertz image to be identified includes:

Perform grayscale transformation on the terahertz image to be recognized;

The terahertz image is smoothed by median filtering.

Preferably, extracting edge information in the terahertz image includes:

Using the Canny edge detection operator to extract the edge information of the terahertz image.

Preferably, using the edge information and the preset dangerous goods comparison library to determine whether there are dangerous goods in the terahertz image includes:

Compare the similarity between the geometric features in the extracted edge information and the statistical learning results of the edge geometric features of dangerous goods in the preset dangerous goods comparison library;

When the similarity is greater than a preset threshold, it is determined that dangerous goods exist in the terahertz image.

Preferably, it also includes:

Obtain color images of dangerous goods;

extracting edge contour information of the color image of the dangerous goods;

Create a new dangerous goods comparison library, and store the edge profile information into the dangerous goods comparison library.

A terahertz image dangerous goods identification device, comprising:

An edge information extraction module, configured to preprocess the terahertz image to be identified, and extract edge information in the terahertz image;

A judging module, configured to judge whether dangerous goods exist in the terahertz image by using the edge information and a preset dangerous goods comparison library;

A segmentation module, configured to segment the edge of the dangerous goods in the terahertz image to obtain a dangerous goods segmentation image when there are dangerous goods in the terahertz image;

The classification information determination module is configured to input the segmented images of dangerous goods into a preset classification network for processing to obtain classification information.

A terahertz image dangerous goods identification device, comprising:

memory for storing computer programs;

A processor, configured to implement the steps of the above-mentioned method for identifying dangerous goods with terahertz images when executing the computer program.

A readable storage medium, where a computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned method for identifying dangerous goods in terahertz images are realized.

Apply the method provided by the embodiment of the present invention to preprocess the terahertz image to be recognized, and extract the edge information in the terahertz image; use the edge information and the preset dangerous goods comparison library to judge whether there are dangerous goods in the terahertz image ; If so, segment the edge of the dangerous goods in the terahertz image to obtain the dangerous goods segmentation image; input the dangerous goods segmentation image into the preset classification network for processing to obtain the classification information. Before identifying dangerous goods on the terahertz image, the edge information is used for pre-judgment. Only when dangerous goods are detected in the edge information, the terahertz image is segmented, and then the dangerous goods segmented image is input into the preset classification network for classification. processing to obtain the final classification information, that is, the recognition result. That is to say, in the actual security inspection process, for a large number of terahertz images to be identified, only the terahertz images with dangerous goods are identified for the category of dangerous goods, and the category of dangerous goods for the terahertz images without dangerous goods does not need to be identified . In this way, the number of terahertz images for dangerous goods identification can be reduced, the occupation of computer resources can be reduced, and the identification speed can be improved.

Correspondingly, the embodiments of the present invention also provide a terahertz image dangerous goods identification device, equipment, and readable storage medium corresponding to the above-mentioned terahertz image dangerous goods identification method, which have the above-mentioned technical effects and will not be repeated here.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is an implementation flow chart of a method for identifying dangerous goods in a terahertz image in an embodiment of the present invention;

FIG. 2 is a schematic structural diagram of another method for identifying dangerous goods in a terahertz image in an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of another method for identifying dangerous goods in a terahertz image in an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of another method for identifying dangerous goods in a terahertz image in an embodiment of the present invention;

Fig. 5 is a comparison diagram of edge detection effects of different edge detection operators;

Fig. 6 is a schematic diagram of the terahertz dangerous goods identification result obtained after applying the technical solution provided by the embodiment of the present invention;

7 is a schematic diagram of a comparison of terahertz image preprocessing and filtering methods in an embodiment of the present invention;

Fig. 8 is a schematic diagram of the classification process of a terahertz dangerous goods identification method in an embodiment of the present invention;

FIG. 9 is a schematic flowchart of a method for identifying dangerous goods in a terahertz image in an embodiment of the present invention;

Fig. 10 is a schematic structural diagram of a terahertz image dangerous goods identification device in an embodiment of the present invention;

Fig. 11 is a schematic structural diagram of a terahertz image dangerous goods identification device in an embodiment of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. Apparently, the described embodiments are only some of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Embodiment one:

Please refer to FIG. 1. FIG. 1 is a flow chart of a method for identifying dangerous goods in a terahertz image in an embodiment of the present invention. The method includes the following steps:

S101. Perform preprocessing on the terahertz image to be recognized, and extract edge information in the terahertz image.

Wherein, the terahertz image to be recognized may be a terahertz image obtained during real-time security inspection, or may be a terahertz image in a preset image library. The preprocessing of the terahertz image to be detected may be converting it from a non-grayscale image to a grayscale image, performing denoising processing, or processing to increase resolution, etc.

Then the edge information is extracted from the preprocessed terahertz image. Among them, the edge refers to the collection of those pixels whose gray levels of the surrounding pixels change, which is mainly manifested as the discontinuity of the local adjustment of the image, that is, the place where the signal changes ambiguously. The edge information can include the geometric shape of the image edge contour, the position of the edge in the terahertz image, the number of edges, the size of the closed edge and other information. In this embodiment, common examples such as Roberts (a first-order differential operator that uses a local differential operator to find edges), Sobel (a first-order differential operator that uses the gradient value of a pixel adjacent area to calculate the pixel gradient ), Prewitt (a first-order differential operator that suppresses noise through pixel averaging), Laplacian-Gauss (a second-order differential operator that performs Gaussian filtering first and then Laplacian detection), Canny (a not easy Affected by noise, edge detection operators such as operators sensitive to weak edges) perform edge detection, and how to extract edge information will not be repeated here.

S102. Using edge information and a preset dangerous goods comparison library to determine whether dangerous goods exist in the terahertz image.

In this embodiment, a dangerous goods comparison database can be established in advance, and the characteristic information of dangerous goods such as outline information, size information, and geometric features (such as shape) of dangerous goods can be stored in the dangerous goods comparison database.

After the edge information is obtained, the edge information can be compared with the characteristic information of dangerous goods in the pre-established dangerous goods comparison library to determine whether there are dangerous goods in the terahertz image. Specifically, a corresponding threshold can be set, for example, if the similarity between the edge profile in the current edge information and the profile information of the dangerous goods in the dangerous goods comparison library is greater than the preset similarity threshold, it is considered that there are dangerous goods in the terahertz image.

When the judgment result is yes, the operation of step S103 is performed, and if no, no operation is performed on the current terahertz image. For example, when the judgment result is no, the next terahertz image can be processed.

S103. Segment the edge of the dangerous goods in the terahertz image to obtain a segmented image of the dangerous goods.

After it is determined in step S102 that there is a dangerous article in the terahertz image, that is, when there is a dangerous article in the terahertz image, the position of the dangerous article in the terahertz image can be determined based on the edge information. Then, the terahertz image is segmented, and the part of the image occupied by the edge of the dangerous goods is segmented from the terahertz image to obtain a segmented image of the dangerous goods. Wherein, the segmented image of dangerous goods is an image including edges of dangerous goods. When performing segmentation, in order to maximize the segmentation of the edge of dangerous goods from the terahertz image, a certain extension can be made on the basis of the minimum area occupied by the edge of dangerous goods before segmentation.

S104. Inputting the segmented images of dangerous goods into a preset classification network for processing to obtain classification information.

In this embodiment, a classification network may be preset, and a class library corresponding to the classification network may be established. Specifically, the classification network can be a common classification network (classification algorithm) such as Faster R-CNN, R-FCN, YOLO, SSD, and ZFNet. In this embodiment, there is no limitation as to which classification network is specifically used to perform classification processing.

That is to say, after obtaining the edge image of dangerous goods, the segmented image of dangerous goods can be input into a preset classification network for processing to obtain classification information, that is, the final recognition result.

Apply the method provided by the embodiment of the present invention to preprocess the terahertz image to be recognized, and extract the edge information in the terahertz image; use the edge information and the preset dangerous goods comparison library to judge whether there are dangerous goods in the terahertz image ; If so, segment the edge of the dangerous goods in the terahertz image to obtain the dangerous goods segmentation image; input the dangerous goods segmentation image into the preset classification network for processing to obtain the classification information. Before identifying dangerous goods on the terahertz image, the edge information is used for pre-judgment. Only when dangerous goods are detected in the edge information, the terahertz image is segmented, and then the dangerous goods segmented image is input into the preset classification network for classification. processing to obtain the final classification information, that is, the recognition result. That is to say, in the actual security inspection process, for a large number of terahertz images to be identified, only the terahertz images with dangerous goods are identified for the category of dangerous goods, and the category of dangerous goods for the terahertz images without dangerous goods does not need to be identified . In this way, the number of terahertz images for dangerous goods identification can be reduced, the occupation of computer resources can be reduced, and the identification speed can be improved.

It should be noted that, based on the first embodiment above, the embodiment of the present invention also provides a corresponding improvement solution. In subsequent embodiments, the same steps as in the first embodiment above or corresponding steps may be referred to each other, and the corresponding beneficial effects may also be referred to each other, and will not be repeated in the improved embodiments below.

Embodiment two:

Please refer to FIG. 2. FIG. 2 is a flow chart of another method for identifying dangerous goods in a terahertz image in an embodiment of the present invention. The method includes the following steps:

S201. Perform grayscale transformation on the terahertz image to be recognized.

When the terahertz image to be recognized is not a grayscale image, the terahertz image is converted to grayscale, that is, the terahertz image is displayed with black tones.

S202. Perform smoothing processing on the terahertz image by using a median filter.

Please refer to Figure 7. It can be seen from the figure that after using the median filter, the changes between the pixels in the image are relatively smooth. Compared with the mean filter, the median filter can retain the pixel change characteristics of the original image such as image edge. Therefore, in this embodiment, the terahertz image is smoothed by median filtering. Specifically, the pixel I'(i, j) of the noise image is centered by using adaptive median filtering.

Because the median filter will replace the pixel in the center of the selected pixel window with the average value of the selected pixel window, if the selected pixel window is too large, the edge pixels may be blurred; if the selected pixel window is too small, it will not be possible to Noise removal. Preferably, in this embodiment, considering that the terahertz image has the characteristics of mixed noise and the experimental data show that using a window with selected pixels of 3×3 for median filtering, the smoothed terahertz image obtained is more conducive to Dangerous goods identification. Of course, in other embodiments of the present invention, other sizes of selected pixel windows may also be selected according to actual conditions.

Based on the characteristics of mixed noise, first judge the noise model in the filtering window, and calculate the variance of the pixels in the window:

in

Among them, σ ² is the variance, and M is the mean value in the filter window. When σ ² is greater than the threshold, the median filter is used, and the traditional median filter is used:

where (s, t)∈S _xy , That is, the terahertz image after median filtering, f(x, y) is the gray value of the center pixel, and S _xy represents the coordinate group of a rectangular sub-image window whose center is at (x, y) and whose size is m×n.

S203. Extract edge information in the terahertz image.

S204. Using edge information and a preset dangerous goods comparison library to determine whether there are dangerous goods in the terahertz image.

If there are dangerous goods, the operation of step S205 is performed, and if there are no dangerous goods, no operation is performed.

S205. Segment the edges of the dangerous goods in the terahertz image to obtain a segmented image of the dangerous goods.

S206. Using the segmented image of the dangerous goods as an input variable of the ZFNet convolutional neural network to obtain classification information output by the ZFNet convolutional neural network.

Among them, the number of layers of the ZFNet convolutional neural network is much smaller than that of other popular classifier networks such as VGG-16. The smaller convolution kernel and fewer layers make the calculation amount of training and testing doubled, greatly improving the Classification speed. Since the terahertz image in the embodiment of the present invention is preprocessed, the background has less influence on the classification process, so fewer layers can accurately classify the contours of dangerous goods. Using the dangerous goods segmentation image as the input variable of the relatively simple ZFNet convolutional neural network, the classification information can be quickly obtained.

Embodiment three:

Please refer to FIG. 3. FIG. 3 is a flow chart of another method for identifying dangerous goods in a terahertz image in an embodiment of the present invention. The method includes the following steps:

S301. Create a dangerous goods comparison library.

Based on the possible insufficient number of terahertz dangerous goods images, a color dangerous goods image library can be established to extract dangerous goods edge contour information from color images. Specifically, the implementation process of creating a dangerous goods comparison library includes the following steps:

Step 1. Obtain a color image of dangerous goods;

Step 2, extracting the edge contour information of the color image of the dangerous goods;

Step 3: Create a new dangerous goods comparison library, and store the edge contour information into the dangerous goods comparison library.

For ease of description, the above three steps will be described in combination below.

The color image of dangerous goods is acquired, and the color image of dangerous goods may specifically be a color photo of common dangerous goods, for example, a color image of common controlled appliances. Then extract the edge contour information of dangerous goods in the color image of dangerous goods, and store the edge contour information in the created dangerous goods comparison library, and finally obtain the dangerous goods comparison library that can be used for dangerous goods pre-judgment. Among them, the common edge information extraction algorithm can be used to extract the edge contour information in the color image of the dangerous goods, which will not be repeated here.

S302. Perform preprocessing on the terahertz image to be recognized, and extract edge information in the terahertz image.

S303. Compare the similarity between the geometric features in the extracted edge information and the statistical learning results of the edge geometric features of dangerous goods in the preset dangerous goods comparison library.

In this embodiment, when pre-judging dangerous goods, the geometric features in the edge information are mainly used to compare the similarity with the statistical learning results of the edge geometric features of dangerous goods. Wherein, the statistical learning result may be a geometric feature obtained after performing statistical learning on the same dangerous goods from multiple dangerous goods sample contours. Obviously, when the similarity is higher, the possibility of dangerous goods with higher similarity appearing in the terahertz image corresponding to the edge geometric features of the dangerous goods is greater.

S304. When the similarity is greater than a preset threshold, it is determined that dangerous goods exist in the terahertz image.

In this embodiment, a threshold can be set for the edge pre-judgment of dangerous goods. The specific size of the threshold can be preset, or can be determined and adjusted according to the actual situation, which is not limited here.

When the calculated relative degree is greater than the preset threshold, it can be considered that dangerous goods exist in the terahertz image.

Of course, in other embodiments of the present invention, when the threshold value is used to judge whether there may be dangerous goods in the terahertz image, the value of the connected region in the edge detection image can be obtained, and the area and shape of the connected region can be calculated based on The above area and shape information will remove other edges to leave the edge of dangerous goods, and record the position information of the edge of dangerous goods in the image, and finally mark it with a green rectangular frame. Part of the code is as follows:

"[L,N]=bwlabel(Image_canny,4);% returns the connected regions and their numbers in the image

s=regionprops(L,'Area');% Calculate the total number of pixels in each region of the image

Image_only_edge = ismember(L, find([s.Area]>=100&[s.Area]<=500));

% Determine whether the connected area is a non-human body contour by setting the range

[B,L,N]=bwboundaries(Image_only_edge);

% Filter non-human body contours

rectangle('position',[left-5,top-5,right-left+10,bottom-top+10],'edgecol','g');

%Draw a rectangular frame outside the outline of the human body, which is convenient for similarity comparison later.”

Among them, % is a comment match.

After pre-judging whether there are dangerous goods in the terahertz image, some terahertz images without grayscale mutation can be filtered, which can reduce the number of dangerous goods category recognition for terahertz images, reduce the amount of calculation and energy consumption, and further Improve the speed of dangerous goods recognition in terahertz images.

S305. When there are dangerous goods in the terahertz image, segment the edge of the dangerous goods in the terahertz image to obtain a segmented image of the dangerous goods.

S306. Inputting the segmented images of dangerous goods into a preset classification network for processing to obtain classification information.

Embodiment four:

Please refer to Fig. 4, Fig. 4 is a flow chart of another method for identifying dangerous goods in a terahertz image in an embodiment of the present invention, the method includes the following steps:

S401. Perform preprocessing on the terahertz image to be recognized, and extract edge information of the terahertz image by using a Canny edge detection operator.

Please refer to Figure 5. After the preprocessing of the terahertz image to be recognized, the Canny edge detection operator with relatively good edge detection effect can be used to extract the edge of the terahertz image. Specifically, 0.489 may be used as the ratio of strong and weak edges. Of course, in other embodiments of the present invention, the ratio of strong and weak edges may also be other values.

Among them, the first-order differential convolution template of the Canny edge detection operator is:

ρ[i,j]=(f[i,j+1]-f[i,j]+f[i+1,j+1]-f[i+1,j])/2;

Q[i,j]=(f[i+1,j]-f[i,j]+f[i+1,j+1]-f[i,j+1])/2;

Among them, ρ _x and ρ _y are partial derivative matrices, M[i, j] is the magnitude of the gradient, and the gradient direction of θ is the angle, where f is the gray value of the image, P represents the gradient magnitude in the X direction, and Q represents the Y direction Gradient magnitude. As for how the Canny edge detection operator performs edge detection specifically, refer to the specific process of the existing Canny edge detection operator, which will not be repeated here.

S402. Using edge information and a preset dangerous goods comparison library to determine whether dangerous goods exist in the terahertz image.

When the judgment result is yes, the operation of step S403 is performed; if no, no operation is performed.

S403. Segment the edge of the dangerous goods in the terahertz image to obtain a segmented image of the dangerous goods.

S404. Inputting the segmented images of dangerous goods into a preset classification network for processing to obtain classification information.

S405. Project the location information and classification information of the edge of the dangerous goods in the edge information into the terahertz image.

After the classification information is obtained, in order to view the identification results of dangerous goods, the location information and classification information of the dangerous goods edge in the edge information can be projected into the terahertz image. Specifically, please refer to FIG. 6 , where "Gun" is the category information, and the boxes represent the location information of dangerous goods. Of course, in practical applications, the position information and category information projected into the terahertz image can also be marked with other colors (other colors other than black tone, such as red, green, etc.).

In order to facilitate those skilled in the art to understand the technical solutions provided by the embodiments of the present invention, the technical solutions provided by the above embodiments will be described in detail below taking a specific application scenario of real-time security inspection as an example.

Please refer to FIG. 9 , during real-time security inspection, after the terahertz image is obtained, the terahertz image is firstly preprocessed. Then use the Canny edge detection operator to detect the edge of the preprocessed terahertz image to judge whether there is a gray-scale mutation. If there is no gray-scale mutation, stop the detection of the terahertz ellipse image and continue to process the next terahertz image. Hertzian images, that is, a large number of terahertz images are screened to reduce time-consuming dangerous goods identification for non-dangerous goods terahertz images. If there is a gray-scale mutation, mark the location information of the gray-scale mutation, and set a candidate frame for segmentation or cropping. The candidate boxes are then cropped from the terahertz image. Input the image obtained after cropping into the ZFNet classification network for processing as shown in Figure 8 to obtain classification information.

After the classification information and coordinate information are obtained, the classification information and coordinate information are projected into the terahertz image, and the processing result diagram shown in FIG. 6 can be obtained. After seeing the processing results, the security inspectors can take corresponding protective measures to protect the lives and property of the public.

Before using the ZFNet classification network to identify dangerous goods on the image, a RGB sample library can also be pre-established to train the ZFNet classification network. Among them, RGB refers to the RGB color mode. Specifically, all the colors on the computer screen are formed by mixing red, green and blue colors in different proportions. A group of red, green and blue is the smallest display unit. Any color on the screen can be recorded and represented by a set of RGB values. Therefore, red, green, and blue are also called the three primary colors, which are R (red), G (green), and B (blue) in English. That is to say, the RGB dangerous goods sample library is the dangerous goods sample library of color images. Then the color image is preprocessed, and the ZFNet classification network is trained using the Canny edge image. Specifically, the Canny edge image is input into the ZFNet classification network, and algorithms such as backpropagation are used to train under the GPU (Graphics Processing Unit, image processor). Among them, the backpropagation algorithm (BP, Backpropagation algorithm) mainly has two links (incentive propagation, weight update) repeated loop iterations until the response of the network to the input reaches the predetermined target range.

It should be noted that the preprocessing of the images in the color sample library can be grayscale transformation, denoising processing, and then edge detection to obtain Canny edge images. For example, the common formula for converting a color image to a grayscale image is: Gray=R*0.299+G*0.587+B*0.114.

Corresponding to the above method embodiment, the embodiment of the present invention also provides a terahertz image dangerous goods identification device, the terahertz image dangerous goods identification device described below and the terahertz image dangerous goods identification method described above can correspond to each other refer to.

Referring to shown in Figure 10, the device includes the following modules:

An edge information extraction module 101, configured to preprocess the terahertz image to be identified, and extract edge information in the terahertz image;

The judging module 102 is used to judge whether there are dangerous goods in the terahertz image by using the edge information and the preset dangerous goods comparison library;

The segmentation module 103 is used to segment the edge of the dangerous goods in the terahertz image to obtain the dangerous goods segmentation image when there are dangerous goods in the terahertz image;

The classification information determination module 104 is configured to input the segmented image of dangerous goods into a preset classification network for processing to obtain classification information.

Apply the device provided by the embodiment of the present invention to preprocess the terahertz image to be recognized, and extract the edge information in the terahertz image; use the edge information and the preset dangerous goods comparison library to judge whether there are dangerous goods in the terahertz image ; If so, segment the edge of the dangerous goods in the terahertz image to obtain the dangerous goods segmentation image; input the dangerous goods segmentation image into the preset classification network for processing to obtain the classification information. Before identifying dangerous goods on the terahertz image, the edge information is used for pre-judgment. Only when dangerous goods are detected in the edge information, the terahertz image is segmented, and then the dangerous goods segmented image is input into the preset classification network for classification. processing to obtain the final classification information, that is, the recognition result. That is to say, in the actual security inspection process, for a large number of terahertz images to be identified, only the terahertz images with dangerous goods are identified for the category of dangerous goods, and the category of dangerous goods for the terahertz images without dangerous goods does not need to be identified . In this way, the number of terahertz images for dangerous goods identification can be reduced, the occupation of computer resources can be reduced, and the identification speed can be improved.

In a specific embodiment of the present invention, the classification information determination module 104 is specifically configured to use the segmented image of dangerous goods as an input variable of the ZFNet convolutional neural network, and obtain the classification information output by the ZFNet convolutional neural network.

In a specific embodiment of the present invention, it also includes:

The recognition result projection module is used to project the position information and classification information of the edge of the dangerous goods in the edge information to the terahertz image after inputting the dangerous goods segmentation image into the preset classification network for processing and obtaining the classification information.

In a specific implementation manner of the present invention, the edge information extraction module 101 specifically includes:

A grayscale transformation unit, configured to perform grayscale transformation on the terahertz image to be recognized;

The smoothing processing unit is used for smoothing the terahertz image by using a median filter.

In a specific implementation manner of the present invention, the edge information extraction module 101 specifically includes:

The edge information extraction unit is used to extract the edge information of the terahertz image by using the Canny edge detection operator.

In a specific implementation manner of the present invention, the judging module 102 specifically includes:

The similarity calculation unit is used to compare the similarity between the geometric features in the extracted edge information and the statistical learning results of the edge geometric features of dangerous goods in the preset dangerous goods comparison library;

A judging unit, configured to determine that dangerous goods exist in the terahertz image when the similarity is greater than a preset threshold.

In a specific embodiment of the present invention, it also includes:

The dangerous goods comparison library creation module is used to obtain the color image of dangerous goods; extract the edge contour information of the color image of dangerous goods; create a new dangerous goods comparison library, and store the edge contour information in the dangerous goods comparison library.

Corresponding to the above method embodiment, the embodiment of the present invention also provides a terahertz image dangerous goods recognition device, a terahertz image dangerous goods recognition device described below and a terahertz image dangerous goods recognition described above The methods can be referred to each other.

As shown in Figure 11, the terahertz image dangerous goods identification equipment includes:

memory D1 for storing computer programs;

The processor D2 is configured to implement the steps of the method for identifying dangerous goods in a terahertz image in the above method embodiment when executing a computer program.

Corresponding to the above method embodiment, the embodiment of the present invention also provides a readable storage medium, and a readable storage medium described below and a method for identifying dangerous goods in a terahertz image described above can be referred to in correspondence.

A readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for identifying dangerous goods by terahertz image in the above method embodiment are realized.

Specifically, the readable storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and the like that can store program codes. readable storage media.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same or similar parts of each embodiment can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related information, please refer to the description of the method part.

Professionals can further realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software or a combination of the two. In order to clearly illustrate the possible For interchangeability, in the above description, the composition and steps of each example have been generally described according to their functions. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present invention.

The steps of the methods or algorithms described in connection with the embodiments disclosed herein may be directly implemented by hardware, software modules executed by a processor, or a combination of both. Software modules can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other Any other known storage medium.

In this paper, specific examples are used to illustrate the principles and implementation methods of the present invention, and the descriptions of the above embodiments are only used to help understand the technical solutions and core ideas of the present invention. It should be pointed out that for those skilled in the art, without departing from the principle of the present invention, some improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims (10)

1. A terahertz image dangerous goods identification method, characterized in that, comprising: Preprocessing the terahertz image to be identified, and extracting edge information in the terahertz image; Using the edge information and a preset dangerous goods comparison library to determine whether there are dangerous goods in the terahertz image; If so, segmenting the edge of the dangerous goods in the terahertz image to obtain a segmented image of the dangerous goods; Inputting the segmented image of dangerous goods into a preset classification network for processing to obtain classification information. 2. The terahertz image dangerous goods identification method according to claim 1, wherein the dangerous goods segmentation image is input into a preset classification network for processing to obtain classification information, including: The dangerous goods segmentation image is used as the input variable of the ZFNet convolutional neural network, and the classification information output by the ZFNet convolutional neural network is obtained. 3. The terahertz image dangerous goods identification method according to claim 1, characterized in that, after inputting the dangerous goods segmented image into a preset classification network for processing and obtaining classification information, the method further includes: Projecting the position information of the dangerous goods edge and the classification information in the edge information into the terahertz image. 4. The terahertz image dangerous article identification method according to claim 1, wherein the preprocessing of the terahertz image to be identified includes: Perform grayscale transformation on the terahertz image to be recognized; The terahertz image is smoothed by median filtering. 5. The method for identifying dangerous goods in terahertz images according to claim 1, wherein extracting edge information in the terahertz images includes: Using the Canny edge detection operator to extract the edge information of the terahertz image. 6. The method for identifying dangerous articles in a terahertz image according to claim 1, wherein the edge information is compared with a preset dangerous article library to determine whether there are dangerous articles in the terahertz image, including: Compare the similarity between the geometric features in the extracted edge information and the statistical learning results of the edge geometric features of dangerous goods in the preset dangerous goods comparison library; When the similarity is greater than a preset threshold, it is determined that dangerous goods exist in the terahertz image. 7. The terahertz image dangerous goods identification method according to any one of claims 1 to 6, further comprising: Obtain color images of dangerous goods; extracting edge contour information of the color image of the dangerous goods; Create a new dangerous goods comparison library, and store the edge profile information into the dangerous goods comparison library. 8. A terahertz image dangerous goods identification device, characterized in that it comprises: An edge information extraction module, configured to preprocess the terahertz image to be identified, and extract edge information in the terahertz image; A judging module, configured to judge whether dangerous goods exist in the terahertz image by using the edge information and a preset dangerous goods comparison library; A segmentation module, configured to segment the edge of the dangerous goods in the terahertz image to obtain a dangerous goods segmentation image when there are dangerous goods in the terahertz image; The classification information determination module is configured to input the segmented images of dangerous goods into a preset classification network for processing to obtain classification information. 9. A terahertz image dangerous goods identification device, characterized in that it comprises: memory for storing computer programs; A processor, configured to implement the steps of the method for identifying dangerous goods in terahertz images according to any one of claims 1 to 7 when executing the computer program. 10. A readable storage medium, characterized in that a computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the terahertz image risk described in any one of claims 1 to 7 is realized. The steps of the product identification method.