Millimeter-wave image target detection method and system based on the size distribution of hidden objects

Abstract

The present application provides a millimeter-wave image target detection method and system based on the size distribution of concealed objects. The millimeter-wave image target detection method based on the size distribution of concealed objects includes the following steps: statistically analyzing the size distribution of concealed objects of the same category; fitting the true size distribution curve using a distribution function and smoothing the true size distribution curve; using a target detection model to detect an input image to obtain the position and size information of the concealed objects and the category probability based on deep learning; determining the category determination probability of the concealed objects based on size; and determining the joint category determination probability based on the category determination probability based on deep learning and the category determination probability based on size distribution. It has the advantages of being able to effectively improve the classification accuracy of target detection in millimeter-wave human body security inspection and the overall classification accuracy of the model, etc.

Description

Technical Field

[0001] This application relates to the technical field of millimeter-wave image target detection, and particularly relates to a millimeter-wave image target detection method and system based on the size distribution of concealed objects. Background Art

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

[0003] Compared with visible light, millimeter-wave imaging technology has relatively low imaging resolution, and changes in the surrounding environment will also affect the imaging quality. At the same time, concealed objects are usually small in size and have limited texture information of their own. The human body security inspection task requires detecting whether the inspected person is carrying dangerous goods. Therefore, the target detection model not only needs to detect concealed items but also accurately classify the concealed objects. Most current target detection models can accurately detect concealed objects in most cases, but how to correctly classify concealed objects still faces challenges, and this problem restricts the widespread application of millimeter-wave human concealed object security inspection. Summary of the Invention

[0004] One aspect of the embodiments of this application provides a millimeter-wave image target detection method based on the size distribution of concealed objects, which may include the following steps:

[0005] Statistically analyze the size distribution of concealed objects of the same category;

[0006] Use a distribution function to fit the true size distribution curve and smooth the true size distribution curve;

[0007] Use a target detection model to detect the input image to obtain the position and size information of the concealed object and the category probability based on deep learning;

[0008] Determine the category determination probability of the concealed object based on its size;

[0009] Based on the category determination probability based on deep learning and the category determination probability based on size distribution, determine the joint category determination probability.

[0010] The solution of one aspect of this application can be further configured in a preferred embodiment as follows:

[0011] In the step of statistically analyzing the size distribution of concealed objects of the same category, the statistical method for the size distribution of concealed objects of the same category includes the following expression:

[0012] , (1)

[0013] Wherein, represents the size of the hidden object, represents the normalized width value of the target box of the hidden object, represents the normalized height value of the target box of the hidden object, represents the pixel value in the horizontal direction of the millimeter-wave grayscale image, represents the pixel value in the vertical direction of the millimeter-wave grayscale image, represents the pixel value occupied by the width of the target box of the hidden object, represents the pixel value occupied by the height of the target box of the hidden object.

[0014] In a preferred embodiment, the solution of one aspect of the present application can be further configured as:

[0015] In the step of fitting the true size distribution curve by using the distribution function and smoothing the true size distribution curve, the fitting method of the true size distribution curve includes the following expression:

[0016] , (2)

[0017] Wherein, 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 the value range is [1, n , n represents the maximum value in the value range of the pixel value corresponding to the size of the hidden object.

[0018] In a preferred embodiment, the solution of one aspect of the present application can be further configured as:

[0019] In the step of determining the size-based hidden object category determination probability, the method for determining the size-based hidden object category determination probability includes the following expression:

[0020] , (3)

[0021] Wherein, represents the probability that when the size value of the hidden object is belongs to a certain category l , represents the size value of the hidden object, Δ represents the small size offset, f l (x) represents the category l corresponding fitting function, f k(x) Indicates the category k The corresponding fitting function l Indicates the category l , k Indicates the category k , k The value range of is [1 ,N .

[0022] In one preferred embodiment, the solution of one aspect of the present application can be further configured as follows:

[0023] In the step of determining the joint category determination probability based on the category determination probability based on deep learning and the category determination probability based on size distribution, the method for determining the joint category determination probability includes the following expression:

[0024] , (4)

[0025] In the formula Indicates the joint category determination probability P deep Indicates the category determination probability based on deep learning P size Indicates the category determination probability based on size distribution

[0026] In one preferred embodiment, the solution of one aspect of the present application can be further configured as follows:

[0027] The final category prediction probability of the hidden object is the weighted mean of the joint category determination probability And the category determination probability Pdeep based on deep learning:

[0028] , (5)

[0029] In the formula Indicates the final category prediction probability of the hidden object Indicates the weighting coefficient Indicates the joint category determination probability P deep Indicates the category determination probability based on deep learning

[0030] In one preferred embodiment, the solution of one aspect of the present application can be further configured as follows:

[0031] The weighting coefficient is set to:

[0032] ,

[0033] In the formula Indicates the weighting coefficient Indicates the model learning parameter denotes the natural constant It can be a fixed constant.

[0034] On the other hand, an embodiment of the present application provides a millimeter-wave image target detection system based on the size distribution of concealed objects, which may include:

[0035] A statistical module for statistically analyzing the size distribution of concealed objects of the same category;

[0036] A fitting module for fitting the true size distribution curve using a distribution function and smoothing the true size distribution curve;

[0037] A detection module for detecting the input image using a target detection model to obtain the position and size information of the concealed object and the category probability based on deep learning;

[0038] A determination probability determination module for determining the category determination probability of the concealed object based on the size;

[0039] A joint category determination probability determination module for determining the joint category determination probability based on the category determination probability based on deep learning and the category determination probability based on the size distribution.

[0040] The embodiment of the present application has at least the following beneficial effects:

[0041] 1. The millimeter-wave image target detection method based on the size distribution of concealed objects in the embodiment of the present application, aiming at the characteristic that the size distribution of concealed objects in the imaging image is relatively fixed in millimeter-wave body security inspection, judges the category probability according to the size information of the object, and uses this category probability to enhance the category probability based on deep learning, which can effectively improve the target detection and classification accuracy in millimeter-wave body security inspection and has great practical significance for the wide promotion and application of this technology.

[0042] 2. The millimeter-wave image target detection method based on the size distribution of concealed objects in the embodiment of the present application points out that the 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 inspection instrument, the distance between the inspected person and the near-field radar is fixed, the imaging size of the items carried by the inspected person remains unchanged, and at the same time, for the same type of dangerous goods to be detected, such as cigarettes, lighters, mobile phones, etc., although there are differences in brands and models, their size variation ranges are extremely small. Therefore, based on historical information, the probability that the concealed object target belongs to different categories can be calculated from the perspective of size, and the classification probability of the original deep learning-based target detection model can be enhanced, thereby effectively improving the overall classification accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flowchart of the millimeter-wave image target detection method based on the size distribution of concealed objects in the present application.

[0044] Figure 2 It is a statistical schematic diagram of the overall size distribution in the dataset of this application.

[0045] Figure 3 It is a graph of the fitting results of the size distribution curves for various categories in this application.

[0046] Figure 4 It is a schematic diagram of calculating the category probability based on the size distribution in this application. Detailed implementation manners

[0047] Embodiment

[0048] Reference is made here to the various solutions and features of this application with reference to the accompanying drawings.

[0049] The millimeter-wave image target detection method based on the size distribution of hidden objects in the embodiment of this application, as Figure 1 shown, may specifically include the following steps:

[0050] Step S1: Statistically analyze 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:

[0051] , (1)

[0052] In the formula, represents the size of the hidden object, represents the width normalization value of the hidden object target box, represents the height normalization value of the hidden object target box, represents the pixel value in the horizontal direction of the millimeter-wave grayscale image, represents the pixel value in the vertical direction of the millimeter-wave grayscale image, represents the pixel value occupied by the width of the hidden object target box, represents the pixel value occupied by the height of the hidden object target box.

[0053] Statistically analyze the accumulation of the number of hidden objects of different categories in the dataset at each size, as Figure 2 shown, Figure 2 in which the abscissa is the pixel value occupied by the long side of the object annotation box, and the ordinate is the number of objects of this category at a specific size.

[0054] Step S2: Use a distribution function to fit the real size distribution curve and smooth the real size distribution curve; the fitting method for the real size distribution curve includes the following expression:

[0055] , (2)

[0056] 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 the value range is [1, n , n represents the maximum value in the value range of the pixel value corresponding to the size of the hidden object. The fitting of the size distribution curves of each category is shown in Table 1 below:

[0057]

[0058] In Table 1 above, laplace (Laplace distribution) is the name of a specific type of distribution function, loc (controlling the offset) is a parameter in the Laplace distribution function, scale (controlling the 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 results after fitting the distribution curve function are as Figure 3 shown.

[0059] Step S3: Use the YOLOv5 object detection model to detect the input image to obtain the position and size information [x, y, w, h] of the hidden object and the class probability based on deep learning P deep ; YOLOv5 (fully known as: You Only Look Once V5) is a classic object detection model. YOLOv5 can detect the position of the target from the picture and judge the class of the target. The YOLOv5 model includes three parts: the backbone, the neck, and the detection 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 judging the target position and class.

[0060] Step S4: Determine the class determination probability of the hidden object based on size, and calculate the class probability based on the size distribution as Figure 4 shown; The method for determining the class determination probability of the hidden object based on size includes the following expression:

[0061] , (3)

[0062] In the formula, represents the probability that when the size value of the hidden object is it belongs to a certain class l , represents the size value of the hidden object, Δ represents the small size offset, f l (x) represents the class l corresponding fitting function,f k (x) Indicates the category k The corresponding fitting function l Indicates the category l , k Indicates the category k , k The value range of is [1 ,N .

[0063] Step S5, the category determination probability based on deep learning P deep And the category determination probability based on size distribution P size , determine the joint category determination probability; the determination method of the joint category determination probability includes the following expression

[0064] , (4)

[0065] In the formula Indicates the joint category determination probability P deep Indicates the category determination probability based on deep learning P size Indicates the category determination probability based on size distribution

[0066] The final category prediction probability of the hidden object is the weighted mean of the joint category determination probability And the category determination probability based on deep learning P deep :

[0067] , (5)

[0068] In the formula Indicates the final category prediction probability of the hidden object Indicates the weighting coefficient Indicates the joint category determination probability P deep Indicates the category determination probability based on deep learning

[0069] To ensure that the category determination probability based on deep learning P deep Is the main judgment, the joint category determination probability Is only for reference, and the weighting coefficient is set to

[0070] ,

[0071] In the formula Indicates the weighting coefficient Represents the model learning parameters, Represents the natural constant.

[0072] The embodiments of the present application also provide a millimeter-wave image target detection system based on the size distribution of hidden objects, which may specifically include:

[0073] A statistical module for statistically analyzing the size distribution of hidden objects of the same category;

[0074] A fitting module for fitting the true size distribution curve using a distribution function and smoothing the true size distribution curve;

[0075] A detection module for detecting the input image using the YOLOv5 target detection model to obtain the position and size information [x, y, w, h] of the hidden object and the category probability based on deep learning P deep ;

[0076] A determination probability determination module for determining the category determination probability of the hidden object based on size;

[0077] A joint category determination probability determination module for determining the joint category determination probability based on the category determination probability based on deep learning P deep and the category determination probability based on the size distribution P size , to determine the joint category determination probability.

[0078] It should be understood that various modifications can be made to the embodiments applied herein. Therefore, the above description should not be regarded as a limitation, but only as an example of the embodiments. Those skilled in the art will think of other modifications within the scope and spirit of the present application.

[0079] The accompanying drawings, which are included in and constitute a part of this 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.

[0080] These and other features of the present application will become apparent from the following description of the preferred forms of the embodiments given by way of non-limiting example with reference to the accompanying drawings.

[0081] It should also be understood that although the present application has been described with reference to some specific examples, those skilled in the art can surely implement many other equivalent forms of the present application.

[0082] When combined with the accompanying drawings, the above and other aspects, features, and advantages of the present application will become more apparent in view of the following detailed description.

[0083] Specific embodiments of the present application will be described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments claimed are merely examples of the present application and that the present application may be implemented in various ways. Well-known and / or repetitive functions and structures are not described in detail to avoid obscuring the present application with unnecessary or redundant details. Thus, the specific structural and functional details claimed herein are not intended to be limiting but rather are merely a basis for the claims and a representative basis for teaching one of ordinary skill in the art to variously use the present application in substantially any suitable detailed structure.

[0084] This specification may use the phrases "in one embodiment", "in another embodiment", "in yet another embodiment", or "in other embodiments", each of which may refer to one or more of the same or different embodiments according to the present application.

[0085] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions within the spirit and protection scope of the present application, and such modifications or equivalent substitutions should also be regarded as falling within the protection scope of the present application.

Claims

1. A millimeter-wave image target detection method based on the size distribution of hidden objects, characterized in that, Comprising the following steps: Statistically analyze the size distribution of concealed objects of the same category; Use a distribution function to fit the true size distribution curve and smooth the true size distribution curve; Use an object detection model to detect the input image to obtain the position and size information of the concealed object and the category probability based on deep learning; Determine the category determination probability of the concealed object based on size; Based on the category determination probability based on deep learning and the category determination probability based on size distribution, determine the joint category determination probability; wherein: In the step of determining the category determination probability of the concealed object based on size, the method for the category determination probability of the concealed object based on size includes the following expression: ，(3) In the formula, represents the probability that when the size value of the hidden object is belongs to a certain category l ; represents the size value of the hidden object, Δ represents the small size offset, f l (x) represents the category l corresponding fitting function; f k (x) represents the category k corresponding fitting function; l represents the category l , k represents the category k , k The value range of ,N is [1 ,N ; The method for determining the joint category determination probability includes the following expression: ，(4) In the formula, represents the combined class determination probability, P deep represents the class determination probability based on deep learning, P size represents the class determination probability based on size distribution; The predicted probability of the final category of the hidden object is the joint category determination probability and the category determination probability based on deep learning P deep is the weighted mean value of: ， (5) Wherein, represents the final category prediction probability of the hidden object, represents the weighting coefficient, represents the joint category determination probability, P deep represents the category determination probability based on deep learning.

2. The millimeter-wave image target detection method based on the size distribution of hidden objects according to claim 1, wherein In the step of statistically analyzing the size distribution of concealed objects of the same category, the statistical method for the size distribution of concealed objects of the same category includes the following expression: ，(1) In the formula, represents the size of the hidden object, represents the normalized value of the width of the target box of the hidden object, represents the normalized value of the height of the target box of the hidden object, represents the pixel value in the horizontal direction of the millimeter-wave grayscale image, represents the pixel value in the vertical direction of the millimeter-wave grayscale image, represents the pixel value occupied by the width of the target box of the hidden object, represents the pixel value occupied by the height of the target box of the hidden object.

3. The millimeter-wave image target detection method based on the size distribution of hidden objects according to claim 1, wherein In the step of using a distribution function to fit the true size distribution curve and smooth the true size distribution curve, the fitting method for the true 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 the value range is [1, n , n represents the maximum value in the value range of the pixel value corresponding to the size of the hidden object.

4. The millimeter-wave image target detection method based on the size distribution of hidden objects according to claim 1, characterized in that Set the weighting coefficient to: ， In the formula, is expressed as a weighting coefficient, is expressed as a model learning parameter, is expressed as the natural constant.

5. A millimeter-wave image target detection system based on the size distribution of hidden objects, characterized in that, Including: A statistical module for statistically analyzing the size distribution of concealed objects of the same category; A fitting module for using a distribution function to fit the true size distribution curve and smooth the true size distribution curve; A detection module for using an object detection model to detect the input image to obtain the position and size information of the concealed object and the category probability based on deep learning; A determination probability determination module for determining the category determination probability of the concealed object based on size as in the method of claim 1; A joint category determination probability determination module for, based on the category determination probability based on deep learning and the category determination probability based on size distribution, determining the joint category determination probability as in the method of claim 1 to obtain the final category prediction probability of the concealed object.

Images