Millimeter wave image target detection method and system based on hidden object size distribution

Abstract

The invention provides a hidden object size distribution-based millimeter wave image target detection method and system. The hidden object size distribution-based millimeter wave image target detection method comprises the following steps of carrying out statistics on size distribution of hidden objects of the same type; fitting a real size distribution curve by using the distribution function, and smoothing the real size distribution curve; detecting an input image by using the target detection model to obtain hidden object position size information and a category probability based on deep learning; determining a hidden object category determination probability based on the size; and determining a joint category determination probability based on the category determination probability of deep learning and the category determination probability based on size distribution. The method has the advantages that the classification precision of target detection in millimeter wave human body security inspection and the classification accuracy of the whole model can be effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of millimeter wave image target detection, and in particular to a millimeter wave image target detection method and system based on hidden object size distribution. Background Art

[0002] Millimeter-wave holographic imaging technology can effectively detect objects hidden in various parts of the human body under clothing without direct contact with the human body, especially non-metallic objects. In addition, millimeter-wave human body imaging equipment is harmless to the human body and has strong penetration. Its transmission power is less than one thousandth of the electromagnetic radiation of a mobile phone. Therefore, it can effectively improve the objectivity, accuracy and pertinence of the inspection, thereby effectively reducing the labor intensity of security inspectors and improving security inspection efficiency.

[0003] Compared with visible light, the imaging resolution of millimeter-wave imaging technology is relatively low, and changes in the surrounding environment will also affect the imaging quality. At the same time, hidden objects are usually small in size and have limited texture information. The human body security inspection task requires detecting whether the inspected person is carrying dangerous goods, so the target detection model must not only detect hidden objects, but also accurately classify the hidden objects. The current target detection model can accurately detect hidden objects in most cases, but how to correctly classify hidden objects is still a challenge. This problem restricts the widespread application of millimeter-wave human body hidden object security inspection. Summary of the invention

[0004] On the one hand, an embodiment of the present application provides a method for detecting targets using millimeter wave images based on hidden object size distribution, which may include the following steps: Statistical size distribution of hidden objects of the same category; Fitting a true size distribution curve using a distribution function to smooth the true size distribution curve; Use the target detection model to detect the input image and obtain the location and size information of the hidden object and the category probability based on deep learning; Determine the probability of concealed object classification based on size; A joint category decision probability is determined based on the category decision probability obtained through deep learning and the category decision probability obtained through size distribution.

[0005] The solution of one aspect of the present application can be further configured as follows in a preferred embodiment: In the step of counting the size distribution of hidden objects of the same category, the statistical method of the size distribution of hidden objects of the same category includes the following expression: , (1) In the formula, Indicates the size of the hidden object, Represents the normalized value of the width of the hidden object target box, Represents the normalized value of the height of the hidden object target box, Represents the horizontal pixel value of the millimeter wave grayscale image, Represents the vertical pixel value of the millimeter wave grayscale image, Represents the pixel value of the width of the hidden object target box. Represents the pixel value of the hidden object target box height.

[0006] The solution of one aspect of the present application can be further configured as follows in a preferred embodiment: In the step of fitting the real size distribution curve using the distribution function to smooth the real size distribution curve, the fitting method of the real size distribution curve includes the following expression: , (2) In the formula, represents the sum of squared errors, represents the original distribution, represents the fitting function, i Represents the pixel value corresponding to the size of the hidden object, and its value range is [1, n ], n Represents the maximum value in the range of pixel values ​​corresponding to the size of the hidden object.

[0007] The solution of one aspect of the present application can be further configured as follows in a preferred embodiment: In the step of determining the hidden object category determination probability based on size, the method for determining the hidden object category determination probability based on size includes the following expression: , (3) In the formula, Indicates that when the hidden object size value is Belong to a category l The probability of represents the hidden object size value, Δ represents the small size offset, f l (x) Indicates category l The corresponding fitting function is, f k (x) Indicates category k The corresponding fitting function is, l Indicates category l , k Indicates category k , k The value range is [1 ,N ].

[0008] The solution of one aspect of the present application can be further configured as follows in a preferred embodiment: In the step of determining the joint category determination probability based on the category determination probability of deep learning and the category determination probability based on the size distribution, the method for determining the joint category determination probability includes the following expression: , (4) In the formula, represents the probability of joint category determination, P deep represents the probability of category determination based on deep learning, P size Represents the probability of class decision based on size distribution.

[0009] The solution of one aspect of the present application can be further configured as follows in a preferred embodiment: The final category prediction probability of the hidden object is the joint category determination probability And the weighted mean of the category judgment probability Pdeep based on deep learning: , (5) In the formula, represents the final category prediction probability of the hidden object, Expressed as weighting coefficient, represents the probability of joint category determination, P deep Represents the category judgment probability based on deep learning.

[0010] The solution of one aspect of the present application can be further configured as follows in a preferred embodiment: The weighting coefficients are set to: , In the formula, Expressed as weighting coefficient, represents the model learning parameters, represents a natural constant, Can be a fixed constant.

[0011] On the other hand, the embodiment of the present application provides a millimeter wave image target detection system based on hidden object size distribution, which may include: Statistics module, used to count the size distribution of hidden objects of the same category; A fitting module, used for fitting a real size distribution curve using a distribution function, and smoothing the real size distribution curve; The detection module is used to detect the input image using the target detection model to obtain the location and size information of the hidden object and the category probability based on deep learning; A judgment probability determination module, used to determine the probability of hidden object category judgment based on size; The joint category determination probability determination module is used to determine the joint category determination probability based on the deep learning category determination probability and the size distribution category determination probability.

[0012] The embodiments of the present application have at least the following beneficial effects: 1. The embodiment of the present application is based on the millimeter wave image target detection method based on the size distribution of hidden objects. In view of the fact that the size distribution of hidden objects in the imaging image is relatively fixed in millimeter wave body inspection, the category probability of the object is judged according to its size information, and the category probability based on deep learning is enhanced with the category probability. This can effectively improve the classification accuracy of target detection in millimeter wave body inspection, and has great practical significance for the wide promotion and application of this technology.

[0013] 2. The millimeter wave image target detection method based on hidden object size distribution in the embodiment of the present application points out that the human body security inspection task has significant characteristics compared with other target detection tasks, that is, the inspected person usually stands at the designated position of the security scanner, the distance between the inspected person and the near-field radar is fixed, and the imaging size of the items carried by the inspected person remains unchanged. At the same time, although the same type of dangerous goods to be detected, such as cigarettes, lighters, mobile phones, etc., have different brands and models, their size variation range is extremely small. Therefore, based on historical information, the probability of hidden object targets belonging to different categories can be calculated from the perspective of size, and the original classification probability based on the deep learning target detection model can be enhanced, thereby effectively improving the overall classification accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of the millimeter wave image target detection method based on hidden object size distribution in this application.

[0015] Figure 2 This is a statistical diagram of the overall size distribution in the data set of this application.

[0016] Figure 3 This is a diagram showing the function fitting results of the size distribution curve for each category in this application.

[0017] Figure 4 Schematic diagram of calculating category probability based on size distribution for this application. DETAILED DESCRIPTION

[0018] Example Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0019] The millimeter wave image target detection method based on hidden object size distribution in the embodiment of the present application is as follows: Figure 1 As shown, the following steps may be specifically included: Step S1, counting the size distribution of hidden objects of the same category; the statistical method for the size distribution of hidden objects of the same category includes the following expression: , (1) In the formula, Indicates the size of the hidden object, Represents the normalized value of the width of the hidden object target box, Represents the normalized value of the height of the hidden object target box, Represents the horizontal pixel value of the millimeter wave grayscale image, Represents the vertical pixel value of the millimeter wave grayscale image, Represents the pixel value of the width of the hidden object target box. Represents the pixel value of the hidden object target box height.

[0020] The accumulation of different types of hidden objects at different sizes in the statistical data set, such as Figure 2 As shown, Figure 2 The horizontal axis in is the pixel value occupied by the long side of the object annotation box, and the vertical axis is the number of objects of this type at a specific size.

[0021] Step S2, fitting the real size distribution curve using the distribution function to smooth the real size distribution curve; the fitting method of the real size distribution curve includes the following expression: , (2) In the formula, represents the sum of squared errors, represents the original distribution, represents the fitting function, i Represents the pixel value corresponding to the size of the hidden object, and its value range is [1, n ], n Represents the maximum value in the range of pixel values ​​corresponding to the size of the hidden object. The function fitting of the size distribution curve of each category is shown in Table 1:

[0022] In Table 1 above, laplace (Laplace distribution) is the name of a specific type of distribution function, loc (control offset) is a parameter in the Laplace distribution function, scale (control scaling) is another parameter in the Laplace distribution function, foldnorm represents the folded normal distribution, c represents a shape parameter, gumbel_l represents the Gumbel distribution, exponnorm represents the modified Gaussian distribution, and k represents another shape parameter. The distribution curve function is fitted as follows: Figure 3 shown.

[0023] Step S3: Use the YOLOv5 target detection model to detect the input image and obtain the hidden object location and size information [x, y, w, h] and the category probability based on deep learning. P deep YOLOv5 (full name: You Only Look Once V5) is a classic target detection model. YOLOv5 can detect the position of the target from the image and determine the category of the target. The YOLOv5 model consists of three parts: backbone, neck and head. The backbone is responsible for extracting features from the image, the neck is responsible for feature combination, and the detection head is responsible for determining the target position and category.

[0024] Step S4: Determine the probability of hidden object classification based on size, and calculate the classification probability based on size distribution. Figure 4 As shown; the method for determining the probability of hidden object category based on size includes the following expressions: , (3) In the formula, Indicates that when the hidden object size value is Belong to a category l The probability of represents the hidden object size value, Δ represents the small size offset, f l (x) Indicates category l The corresponding fitting function is, f k (x) Indicates category k The corresponding fitting function is, l Indicates category l , k Indicates category k , k The value range is [1 ,N ].

[0025] Step S5: Classification probability based on deep learning P deep and the probability of class determination based on size distribution P size , determine the joint category determination probability; the method for determining the joint category determination probability includes the following expression: , (4) In the formula, represents the probability of joint category determination, P deep represents the probability of category determination based on deep learning, Psize Represents the probability of class decision based on size distribution.

[0026] The final category prediction probability of the hidden object is the joint category determination probability Classification probability based on deep learning P deep The weighted mean of : , (5) In the formula, represents the final category prediction probability of the hidden object, Expressed as weighting coefficient, represents the probability of joint category determination, P deep Represents the category judgment probability based on deep learning.

[0027] To ensure the probability of class determination based on deep learning P deep The joint category determination probability For reference only, the weighting coefficients are set to: , In the formula, Expressed as weighting coefficient, represents the model learning parameters, Represents a natural constant.

[0028] The embodiment of the present application also provides a millimeter wave image target detection system based on hidden object size distribution, which may specifically include: Statistics module, used to count the size distribution of hidden objects of the same category; A fitting module, used for fitting a real size distribution curve using a distribution function, and smoothing the real size distribution curve; The detection module is used to detect the input image using the YOLOv5 target detection model to obtain the location and size information of the hidden object [x, y, w, h] and the category probability based on deep learning P deep ; A judgment probability determination module, used to determine the probability of hidden object category judgment based on size; Joint category decision probability determination module for category decision probability based on deep learning P deep and the probability of class determination based on size distribution P size , determine the joint category decision probability.

[0029] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but only as an example of an embodiment. Other modifications within the scope and spirit of the present application will occur to those skilled in the art.

[0030] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0031] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.

[0032] It should also be understood that although the present application has been described with reference to some specific examples, those skilled in the art will be able to readily implement many other equivalent forms of the present application.

[0033] The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.

[0034] Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments applied for are merely examples of the present application, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that obscure the present application. Therefore, the specific structural and functional details applied for herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present application in a variety of ways with substantially any suitable detailed structure.

[0035] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present application.

[0036] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A millimeter wave image target detection method based on hidden object size distribution, characterized in that: The following steps are involved: Count the size distribution of hidden objects of the same category; Fitting a true size distribution curve using a distribution function to smooth the true size distribution curve; Use the target detection model to detect the input image and obtain the location and size information of the hidden object and the category probability based on deep learning; Determine the probability of concealed object classification based on size; A joint category decision probability is determined based on the category decision probability obtained through deep learning and the category decision probability obtained through size distribution.

2. The millimeter wave image target detection method based on hidden object size distribution according to claim 1 is characterized in that: In the step of counting the size distribution of hidden objects of the same category, the statistical method of the size distribution of hidden objects of the same category includes the following expression: ，(1) In the formula, Indicates the size of the hidden object, Represents the normalized value of the width of the hidden object target box, Represents the normalized value of the height of the hidden object target box, Represents the horizontal pixel value of the millimeter wave grayscale image, Represents the vertical pixel value of the millimeter wave grayscale image, Represents the pixel value of the width of the hidden object target box. Represents the pixel value of the hidden object target box height.

3. The millimeter wave image target detection method based on hidden object size distribution according to claim 1 is characterized in that: In the step of fitting the real size distribution curve using the distribution function to smooth the real size distribution curve, the fitting method of the real size distribution curve includes the following expression: ，(2) In the formula, represents the sum of squared errors, represents the original distribution, represents the fitting function, i Represents the pixel value corresponding to the size of the hidden object, and its value range is [1, n ], n Represents the maximum value in the range of pixel values ​​corresponding to the size of the hidden object.

4. The millimeter wave image target detection method based on hidden object size distribution according to claim 1 is characterized in that: In the step of determining the hidden object category determination probability based on size, the method of determining the hidden object category determination probability based on size includes the following expression: ，(3) In the formula, Indicates that when the hidden object size value is Belong to a category l The probability of represents the hidden object size value, Δ represents the small size offset, f l (x) Indicates category l The corresponding fitting function is, f k (x) Indicates category k The corresponding fitting function is, l Indicates category l , k Indicates category k , k The value range is [1 ,N ].

5. The millimeter wave image target detection method based on hidden object size distribution according to claim 1 is characterized in that: In the step of determining the joint category determination probability based on the category determination probability of deep learning and the category determination probability based on the size distribution, the method for determining the joint category determination probability includes the following expression: ，(4) In the formula, represents the probability of joint category determination, P deep represents the category judgment probability based on deep learning, P size Represents the probability of class decision based on size distribution.

6. The method for detecting targets in millimeter wave images based on hidden object size distribution according to claim 5, characterized in that: The final category prediction probability of the hidden object is the joint category determination probability Classification probability based on deep learning P deep The weighted mean of : ，(5) In the formula, represents the final category prediction probability of the hidden object, Expressed as weighting coefficient, represents the probability of joint category determination, P deep Represents the category judgment probability based on deep learning.

7. The method for detecting targets in millimeter wave images based on hidden object size distribution according to claim 6, characterized in that: The weighting coefficients are set to: ， In the formula, Expressed as weighting coefficient, represents the model learning parameters, Represents a natural constant.

8. A millimeter wave image target detection system based on hidden object size distribution, characterized in that: include: Statistics module, used to count the size distribution of hidden objects of the same category; A fitting module, used for fitting a real size distribution curve using a distribution function, and smoothing the real size distribution curve; The detection module is used to detect the input image using the target detection model to obtain the location and size information of the hidden object and the category probability based on deep learning; A judgment probability determination module, used to determine the probability of hidden object category judgment based on size; The joint category determination probability determination module is used to determine the joint category determination probability based on the deep learning category determination probability and the size distribution category determination probability.

Images