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

Abstract

The invention discloses a hidden object position distribution-based millimeter wave image target detection method and a hidden object position distribution-based millimeter wave image target detection system. The hidden object position distribution-based millimeter wave image target detection method comprises the following steps of carrying out statistics on a distribution region of a central position of a hidden object; determining a high-probability distribution area of the hidden objects; the high probability distribution area is sliced; carrying out super-resolution enhancement on the high-probability distribution region slices through a discrete wavelet transform super-resolution module; performing target detection on the original millimeter wave image to obtain a first hidden object detection result; performing target detection on the high-probability distribution region slice to obtain a second hidden object detection result; and combining the first hidden object detection result and the second hidden object detection result. The method has the advantages of effectively relieving the contradiction between the millimeter wave human body security check accuracy and the real-time performance and the like.

Description

Technical Field

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

[0002] Millimeter-wave holographic imaging technology is an advanced technology in the current security field. It 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. Although its transmission power is less than one thousandth of the electromagnetic radiation of a mobile phone, it can effectively improve the objectivity, accuracy and pertinence of the inspection, reduce the labor intensity of security inspectors, and improve security inspection efficiency.

[0003] Compared with visible light, millimeter wave imaging technology has relatively low imaging resolution and is easily affected by the surrounding environment, forming clutter and reducing imaging quality. At the same time, hidden objects are usually small in size and have limited texture information, making them difficult to detect. Existing millimeter wave target detection technology has a low detection rate for small targets, poses a risk of missed detection, and poses a safety hazard.

[0004] At the same time, due to the particularity of security inspection tasks, the detection results need to be output on the spot, and the real-time requirements are very high. These limit the use of complex models with large parameters. Any solution to improve the model detection performance must take into account its real-time requirements. Summary of the invention

[0005] On the one hand, an embodiment of the present application provides a millimeter wave image target detection method based on hidden object position distribution, which may include the following steps: Count the distribution areas of the central locations of hidden objects; Determine the areas where hidden objects are likely to be distributed; Slicing the high probability distribution area; Performing super-resolution enhancement on the high probability distribution area slices through a discrete wavelet transform super-resolution module; Performing target detection on the original millimeter wave image to obtain a first hidden object detection result; Performing target detection on the high probability distribution area slice to obtain a second hidden object detection result; The first hidden object detection result and the second hidden object detection result are combined.

[0006] The present application is a millimeter wave image target detection method based on the hidden object position distribution. By statistically analyzing the hidden object position distribution and based on the extremely uneven distribution of hidden object positions in millimeter wave security inspection tasks, differentiated detection methods are adopted for different distribution situations. In the millimeter wave image target detection task, the hidden object position distribution is not uniform, and most hidden objects are concentrated in specific areas. Therefore, the areas where these hidden objects are concentrated can be sliced, and the slices can be enhanced with high resolution. Then, the target detector can be used to perform target detection on the original image and the slices respectively, and finally the detection results of the two can be fused.

[0007] The high-probability areas where hidden objects appear are sliced ​​and enhanced to improve the detection capability of the model; at the same time, high-resolution enhancement is performed only on key areas, which saves calculations and compresses model processing time.

[0008] Preferably, the step of performing super-resolution enhancement on the high probability distribution area slice by using a discrete wavelet transform super-resolution module comprises the following steps: The input image is subjected to wavelet transform; The wavelet components are concatenated as new feature vectors; The wavelet transform component of the low-resolution image is fed into the convolutional neural network model as input to obtain the output convolutional neural network prediction difference; Get the enhanced wavelet component; The high-resolution image is obtained by inverse wavelet transform.

[0009] Through the above technical scheme, the discrete wavelet transform super-resolution module (DWSR module) is divided into the following steps: first, the input image is subjected to a two-dimensional discrete wavelet transform to obtain a wavelet component, and then the wavelet component is input as a whole and sent to the neural network model. The output of the neural network is the difference between the high-resolution image and the low-resolution image; the difference between the high-resolution image and the low-resolution image is added to the sub-image of the original input image to obtain an enhanced sub-image; finally, the enhanced wavelet component is subjected to an inverse wavelet transform to obtain a super-resolution enhanced slice.

[0010] Preferably, the step of performing wavelet transform on the input image comprises the following steps: Perform a two-dimensional discrete wavelet transform on the input original image to obtain a wavelet component of low-frequency information, a wavelet component of horizontal high-frequency information, a wavelet component of vertical high-frequency information, and a wavelet component of diagonal high-frequency information; By means of the two-dimensional discrete wavelet transform, the original image is transformed into a wavelet component of low-frequency information, a wavelet component of horizontal high-frequency information, a wavelet component of vertical high-frequency information and a wavelet component of diagonal high-frequency information, and the original image and the wavelet components are mutually converted by means of transform and inverse transform; The two-dimensional discrete wavelet transformation method includes formula (1): , (1) The inverse transform method of the two-dimensional discrete wavelet includes formula (2): , (2) In formula (1) and formula (2), A represents the first adjacent pixel value in the original image, B represents the second adjacent pixel value in the original image, C represents the third adjacent pixel value in the original image, and D represents the fourth adjacent pixel value in the original image. a represents the pixel value at the corresponding position in the wavelet component of the low-frequency information, b represents the pixel value at the corresponding position in the wavelet component of the horizontal high-frequency information, c represents the pixel value at the corresponding position in the wavelet component of the vertical high-frequency information, and d represents the pixel value at the corresponding position in the wavelet component of the diagonal high-frequency information.

[0011] Preferably, the method for splicing the wavelet components as a new feature vector comprises: The low-frequency information wavelet component, the horizontal high-frequency information wavelet component, the vertical high-frequency information wavelet component and the diagonal high-frequency information wavelet component are concatenated in the channel dimension to obtain a new feature vector. The method for obtaining the new feature vector is as shown in formula (3): , (3) In the formula, represents the new feature vector, LL represents the low-frequency information wavelet component, LH Represents horizontal high-frequency information, HL represents the vertical high-frequency information wavelet component, HH represents the diagonal high-frequency information wavelet component, Represents the 2D discrete wavelet transform of the low-resolution original image.

[0012] Preferably, the step of inputting the wavelet transform component of the low-resolution image into the convolutional neural network model as input to obtain the output convolutional neural network prediction difference comprises the following steps: A discrete wavelet transform super-resolution neural network model (ie, DWSR neural network model) is used, and the difference between the high-resolution image and the low-resolution image is used as the true value for training the discrete wavelet transform super-resolution neural network model; The wavelet transform component of the low-resolution image is used as input, and a convolutional neural network is used to predict the difference between the high-resolution image and the low-resolution image. The difference between the high-resolution image and the low-resolution image is as shown in formula (4): , (4) In the formula, Represents the difference between the high-resolution image and the low-resolution image. They respectively represent the difference in the low-frequency dimension, the difference in the horizontal high-frequency dimension, the difference in the vertical high-frequency dimension, and the difference in the diagonal high-frequency dimension.

[0013] Preferably, the method for obtaining the enhanced wavelet component comprises: The new feature vector is added to the difference between the high-resolution image and the low-resolution image output by the convolutional neural network to obtain the enhanced wavelet component: , (5) In the formula, represents the new feature vector, Represents the difference between the high-resolution image and the low-resolution image. They represent the difference in low-frequency dimension, horizontal high-frequency dimension, vertical high-frequency dimension, and diagonal high-frequency dimension, respectively. LL represents the low-frequency information wavelet component, LH Represents horizontal high-frequency information, HL represents the vertical high-frequency information wavelet component, HH Represents the diagonal high-frequency information wavelet component.

[0014] Preferably, the method for obtaining a high-resolution image by inverse wavelet transform comprises formula (6): , (6) In the formula, HR represents a high-resolution image, 2 DWT {.} represents the two-dimensional discrete wavelet inverse transform function, represents the new feature vector, Represents the difference between a high-resolution image and a low-resolution image.

[0015] Preferably, the step of merging the first hidden object detection result and the second hidden object detection result comprises the following steps: Restoring the position coordinates of the high probability distribution area slice detection result to the coordinate system of the original millimeter wave image; The non-maximum suppression algorithm NMS is used to merge the first hidden object detection result and the second hidden object detection result.

[0016] Preferably, the step of calculating the distribution area of ​​the center position of the hidden object comprises the following steps: Divide the hidden object distribution map into a grid of a number of preset pixels; Traverse the data set and count the number of center points of hidden objects in the labeled samples that appear in each grid.

[0017] Another aspect of the present application embodiment provides a millimeter wave image target detection system based on hidden object position distribution, which may include: The statistical module is used to count the distribution area of ​​the central location of the hidden object; A determination module, used to determine the high probability distribution area of ​​hidden objects; A slicing module, used for slicing the high probability distribution area; A super-resolution enhancement module, used for performing super-resolution enhancement on the high probability distribution area slices through a discrete wavelet transform super-resolution module; A first hidden object detection module, used to perform target detection on the original millimeter wave image to obtain a first hidden object detection result; A second hidden object detection module, used to perform target detection on the high probability distribution area slice to obtain a second hidden object detection result; The detection result merging module is used to merge the first hidden object detection result and the second hidden object detection result.

[0018] The embodiments of the present application have at least the following beneficial effects: 1. In the detection task of millimeter wave image targets, the position distribution of hidden objects is not uniform, and most hidden objects appear in specific areas. Therefore, these areas can be sliced ​​and the slices can be enhanced with high resolution. Then, the target detector can be used to perform target detection on the original image and the slices respectively, and finally the detection results of the original image and the slices are fused. The present application is based on a millimeter wave image target detection method based on the position distribution of hidden objects. The high probability area where hidden objects appear is sliced ​​and enhanced, which improves the detection of hidden objects, especially the high detection rate of small targets. At the same time, only the key areas are enhanced with high resolution, which saves the amount of calculation and compresses the processing time of the model.

[0019] 2. This application is based on the millimeter wave image target detection method of the position distribution of hidden objects. According to the characteristics of the huge difference in the distribution probability of hidden objects in different positions in the millimeter wave security inspection task, the computing resources are allocated in a targeted manner. By statistically analyzing the position distribution of the center point of the hidden object, the high probability area of ​​the hidden object is determined, and super-resolution enhancement and detection are carried out in a targeted manner. This application not only effectively improves the detection rate of millimeter wave human body security inspection, but also reduces the overall calculation amount, saves time, improves real-time performance, and effectively alleviates the contradiction between the accuracy and real-time performance of millimeter wave human body security inspection. It has great practical significance for the widespread promotion and application of millimeter wave security inspection technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of the millimeter wave image target detection method based on the location distribution of hidden objects in this application.

[0021] Figure 2 This is a schematic diagram of the principle of the millimeter wave image target detection method based on the location distribution of hidden objects in this application.

[0022] Figure 3 This is a distribution statistical map of the central location of hidden objects in this application.

[0023] Figure 4a This is a schematic diagram of a specific location where most objects in this application are concentrated.

[0024] Figure 4b Schematic diagram of the four areas of high probability distribution of objects defined for this application.

[0025] Figure 5 This is a slice map of the high probability distribution area for this application.

[0026] Figure 6 This is a schematic diagram of the DWSR module structure of this application.

[0027] Figure 7 This is a schematic diagram of the mutual conversion between discrete wavelet transform and inverse discrete wavelet transform in this application. DETAILED DESCRIPTION

[0028] Example Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0029] The millimeter wave image target detection method based on the hidden object position distribution of this embodiment is as follows: Figure 1 As shown, the following steps may be specifically included: S1. Count the distribution area of ​​the central location of the hidden objects; S2. Determine the area where hidden objects are likely to be distributed; S3, slicing the high probability distribution area; S4, performing super-resolution enhancement on the high probability distribution area slices through a discrete wavelet transform super-resolution module; S5. Perform target detection on the original millimeter wave image to obtain a first hidden object detection result; S6, performing target detection on the high probability distribution area slice to obtain a second hidden object detection result; S7: Merge the first hidden object detection result and the second hidden object detection result.

[0030] In the millimeter wave image target detection task, the hidden object position distribution is not uniform, and most of the hidden objects appear in specific areas. The millimeter wave image target detection method based on the hidden object position distribution of this embodiment slices the specific areas where most of these hidden objects appear, and performs high-resolution enhancement on the slices. Then, the target detector is used to perform target detection on the original image and the slices respectively, and finally the detection results of the original image and the slices are fused, as shown in FIG. Figure 2 shown.

[0031] The specific plan is as follows: Step S1: Count the distribution of the center positions of hidden objects.

[0032] like Figure 3 As shown, the distribution map is divided into 10×10 pixel grids, the data set is traversed, the number of center points of hidden objects in the labeled samples appearing in each grid is recorded, and then the distribution map is visualized. The more hidden objects each grid corresponds to in the distribution map, the higher the brightness.

[0033] Step S2: Determine high probability distribution areas Analyzing the distribution statistics of hidden objects, most of the hidden objects are concentrated in several specific locations, namely the sides of the chest and abdomen and the sides of the thighs, such as Figure 4a As shown, the areas where most hidden objects are concentrated are divided into four areas, such as Figure 4b The four boxes in the figure are called high probability distribution areas. According to statistics, more than 70% of hidden objects fall in this high probability distribution area.

[0034] Step S3: Slice the high probability area According to the location of the high-probability area, the original input image is sliced ​​to obtain four local images, such as Figure 5 shown.

[0035] Step S4: Use the Discrete Wavelet Transform Super-Resolution module (DWSR) to perform super-resolution enhancement on the high probability area slices.

[0036] DWSR is divided into the following steps: First, the input image is subjected to a two-dimensional discrete wavelet transform (2dDWT) to obtain four wavelet components. The input image is represented by LR (LowResolution, LR), and the four wavelet components are represented by LL, HL, LH, and HH respectively; then the four wavelet components are input as a whole and sent to the neural network model. The output of the neural network is the difference between high and low resolution; the difference is added to the original sub-image to obtain an enhanced sub-image; finally, the enhanced wavelet component is subjected to an inverse wavelet transform (2dIDWT) to obtain a super-resolution enhanced slice. The DWSR module workflow is as follows: Figure 6 shown.

[0037] Step S41: Input image for wavelet transformation The input image LR is subjected to Haar two-dimensional discrete wavelet transform (ie: 2dDWT) to obtain four wavelet components: low-frequency information wavelet component LL, horizontal high-frequency information wavelet component LH, vertical high-frequency information wavelet component HL and diagonal high-frequency information wavelet component HH.

[0038] Haar two-dimensional discrete wavelet transform Figure 7 As shown, Figure 7 The left side is the original input image. Figure 7 On the right side of are the transformed four wavelet components, namely, low-frequency information wavelet component LL, horizontal high-frequency information wavelet component LH, vertical high-frequency information wavelet component HL and diagonal high-frequency information wavelet component HH; the input original image and components can be converted to each other through transformation and inverse transformation; among them, A, B, C and D represent the four adjacent pixel values ​​in the input original image, and a, b, c and d represent the pixel values ​​of the corresponding positions in the four wavelet components, namely, low-frequency information wavelet component LL, horizontal high-frequency information wavelet component LH, vertical high-frequency information wavelet component HL and diagonal high-frequency information wavelet component HH.

[0039] The two-dimensional discrete wavelet transformation method includes formula (1): , (1) The inverse transform method of the two-dimensional discrete wavelet includes formula (2): , (2) In formula (1) and formula (2), A represents the first adjacent pixel value in the original image, B represents the second adjacent pixel value in the original image, C represents the third adjacent pixel value in the original image, and D represents the fourth adjacent pixel value in the original image. a represents the pixel value at the corresponding position in the wavelet component of the low-frequency information, b represents the pixel value at the corresponding position in the wavelet component of the horizontal high-frequency information, c represents the pixel value at the corresponding position in the wavelet component of the vertical high-frequency information, and d represents the pixel value at the corresponding position in the wavelet component of the diagonal high-frequency information.

[0040] Step S42: concatenate the wavelet components as a new feature vector The four wavelet components of low-frequency information wavelet component LL, horizontal high-frequency information wavelet component LH, vertical high-frequency information wavelet component HL and diagonal high-frequency information wavelet component HH are concatenated in the channel dimension to obtain a new eigenvector (Low Resolution Sub Bands, LRSB for short), the new eigenvector The method of obtaining is as follows: , (3) In the formula, represents the new feature vector, LL represents the low-frequency information wavelet component, LH Represents horizontal high-frequency information, HL represents the vertical high-frequency information wavelet component, HH represents the diagonal high-frequency information wavelet component, Represents the 2D discrete wavelet transform of the low-resolution original image.

[0041] Step S43: Send the LRSB as input to the convolutional neural network model, and the convolutional neural network model outputs the difference between the high-resolution image and the low-resolution image .

[0042] The DWSR neural network model converts the difference between the high-resolution image and the low-resolution image into As the true value for model training, the wavelet transform component LRSB of the low-resolution image LR is used as input, and the convolutional neural network is used to predict the difference between the high-resolution image and the low-resolution image. .

[0043] The DWSR neural network model converts the difference between the high-resolution image and the low-resolution image into As the true value for model training, the wavelet transform component LRSB of the low-resolution image LR is used as input, and the convolutional neural network is used to predict the difference between the high-resolution image and the low-resolution image. , the difference between the high-resolution image and the low-resolution image predicted by the convolutional neural network As shown in formula (4): , (4) In the formula, Represents the difference between the high-resolution image and the low-resolution image. They respectively represent the difference in the low-frequency dimension, the difference in the horizontal high-frequency dimension, the difference in the vertical high-frequency dimension, and the difference in the diagonal high-frequency dimension.

[0044] This embodiment uses a trained DWSR convolutional neural network model.

[0045] Step S44: Obtain enhanced wavelet components The new feature vector LRSB is combined with the difference between the high-resolution image and the low-resolution image output by the convolutional neural network Add and sum to get the enhanced wavelet component: , (5) In the formula, represents the new feature vector, Represents the difference between the high-resolution image and the low-resolution image. They represent the difference in low-frequency dimension, horizontal high-frequency dimension, vertical high-frequency dimension, and diagonal high-frequency dimension, respectively. LL represents the low-frequency information wavelet component, LH Represents horizontal high-frequency information, HL represents the vertical high-frequency information wavelet component, HH Represents the diagonal high-frequency information wavelet component.

[0046] Step S45: Obtain a high-resolution image by using a two-dimensional discrete wavelet inverse transform method The inverse transform method of the two-dimensional discrete wavelet is shown in formula (2). The method of obtaining the high-resolution image HR by the inverse transform of the two-dimensional discrete wavelet includes formula (6): , (6) In the formula, HR represents a high-resolution image, 2 DWT {.} represents the two-dimensional discrete wavelet inverse transform function, represents the new feature vector, Represents the difference between a high-resolution image and a low-resolution image.

[0047] Step S5: Perform target detection on the original millimeter wave image to obtain a first hidden object detection result.

[0048] Step S6: performing target detection on the high probability distribution area slice to obtain a second hidden object detection result; Step S7: merging the first hidden object detection result and the second hidden object detection result.

[0049] The step of merging the first hidden object detection result and the second hidden object detection result comprises the following steps: Restoring the position coordinates of the high probability distribution area slice detection result to the coordinate system of the original millimeter wave image; The non-maximum suppression algorithm NMS is used to merge the first hidden object detection result and the second hidden object detection result.

[0050] 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.

[0051] 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.

[0052] 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.

[0053] 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.

[0054] 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.

[0055] 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.

[0056] 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.

[0057] 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 position distribution, characterized in that: The following steps are involved: Count the distribution areas of the central locations of hidden objects; Determine the areas with high probability of hidden objects being distributed; Slicing the high probability distribution area; Performing super-resolution enhancement on the high probability distribution area slices through a discrete wavelet transform super-resolution module; Performing target detection on the original millimeter wave image to obtain a first hidden object detection result; Performing target detection on the high probability distribution area slice to obtain a second hidden object detection result; The first hidden object detection result and the second hidden object detection result are combined.

2. The method according to claim 1, characterized in that The step of performing super-resolution enhancement on the high probability distribution area slice by using a discrete wavelet transform super-resolution module comprises the following steps: The input image is subjected to wavelet transform; The wavelet components are concatenated as new feature vectors; The wavelet transform component of the low-resolution image is fed into the convolutional neural network model as input to obtain the output convolutional neural network prediction difference; Get the enhanced wavelet component; The high-resolution image is obtained by inverse wavelet transform.

3. The method according to claim 2, characterized in that The step of performing wavelet transform on the input image comprises the following steps: Perform a two-dimensional discrete wavelet transform on the input original image to obtain a wavelet component of low-frequency information, a wavelet component of horizontal high-frequency information, a wavelet component of vertical high-frequency information, and a wavelet component of diagonal high-frequency information; By means of the two-dimensional discrete wavelet transform, the original image is transformed into a wavelet component of low-frequency information, a wavelet component of horizontal high-frequency information, a wavelet component of vertical high-frequency information and a wavelet component of diagonal high-frequency information, and the original image and the wavelet components are mutually converted by means of transform and inverse transform; The two-dimensional discrete wavelet transformation method includes formula (1): ，（1） The inverse transform method of the two-dimensional discrete wavelet includes formula (2): ， （2） In formula (1) and formula (2), A represents the first adjacent pixel value in the original image, B represents the second adjacent pixel value in the original image, C represents the third adjacent pixel value in the original image, and D represents the fourth adjacent pixel value in the original image. a represents the pixel value at the corresponding position in the wavelet component of the low-frequency information, b represents the pixel value at the corresponding position in the wavelet component of the horizontal high-frequency information, c represents the pixel value at the corresponding position in the wavelet component of the vertical high-frequency information, and d represents the pixel value at the corresponding position in the wavelet component of the diagonal high-frequency information.

4. The method according to claim 2, characterized in that: The method for splicing the wavelet components as a new feature vector includes: The low-frequency information wavelet component, the horizontal high-frequency information wavelet component, the vertical high-frequency information wavelet component and the diagonal high-frequency information wavelet component are concatenated in the channel dimension to obtain a new feature vector. The method for obtaining the new feature vector is as shown in formula (3): ，（3） In the formula, represents the new feature vector, LL represents the low-frequency information wavelet component, LH Represents horizontal high-frequency information, HL represents the vertical high-frequency information wavelet component, HH represents the diagonal high-frequency information wavelet component, Represents the 2D discrete wavelet transform of the low-resolution original image.

5. The method according to claim 2, characterized in that: The step of sending the wavelet transform component of the low-resolution image as input into the convolutional neural network model to obtain the output convolutional neural network prediction difference includes the following steps: Using a discrete wavelet transform super-resolution neural network model, the difference between the high-resolution image and the low-resolution image is used as the true value for training the discrete wavelet transform super-resolution neural network model; The wavelet transform component of the low-resolution image is used as input, and the convolutional neural network is used to predict the difference between the high-resolution image and the low-resolution image. The difference between the high-resolution image and the low-resolution image predicted by the convolutional neural network is as shown in formula (4): ，（4） In the formula, Represents the difference between the high-resolution image and the low-resolution image. They respectively represent the difference in the low-frequency dimension, the difference in the horizontal high-frequency dimension, the difference in the vertical high-frequency dimension, and the difference in the diagonal high-frequency dimension.

6. The method according to claim 2, characterized in that The method for obtaining the enhanced wavelet component comprises: The new feature vector is added to the difference between the high-resolution image and the low-resolution image output by the convolutional neural network to obtain the enhanced wavelet component: ，（5） In the formula, represents the new feature vector, Represents the difference between the high-resolution image and the low-resolution image. They represent the difference in low-frequency dimension, horizontal high-frequency dimension, vertical high-frequency dimension, and diagonal high-frequency dimension, respectively. LL represents the low-frequency information wavelet component, LH Represents horizontal high-frequency information, HL represents the vertical high-frequency information wavelet component, HH Represents the diagonal high-frequency information wavelet component.

7. The method according to claim 2, characterized in that The method for obtaining a high-resolution image by inverse wavelet transform includes formula (6): ，（6） In the formula, HR represents a high-resolution image, 2 DWT {.} represents the two-dimensional discrete wavelet inverse transform function, represents the new feature vector, Represents the difference between a high-resolution image and a low-resolution image.

8. The method according to claim 1, characterized in that The step of merging the first hidden object detection result and the second hidden object detection result comprises the following steps: Restoring the position coordinates of the high probability distribution area slice detection result to the coordinate system of the original millimeter wave image; The first hidden object detection result and the second hidden object detection result are combined by using a non-maximum suppression algorithm.

9. The method according to claim 1, characterized in that: The step of counting the distribution area of ​​the central position of the hidden object comprises the following steps: Divide the hidden object distribution map into a grid of a number of preset pixels; Traverse the data set and count the number of center points of hidden objects in the labeled samples that appear in each grid.

10. A millimeter wave image target detection system based on hidden object location distribution, characterized in that: include: The statistical module is used to count the distribution area of ​​the central location of the hidden object; A determination module, used to determine the high probability distribution area of ​​hidden objects; A slicing module, used for slicing the high probability distribution area; A super-resolution enhancement module, used for performing super-resolution enhancement on the high probability distribution area slices through a discrete wavelet transform super-resolution module; A first hidden object detection module, used to perform target detection on the original millimeter wave image to obtain a first hidden object detection result; A second hidden object detection module, used to perform target detection on the high probability distribution area slice to obtain a second hidden object detection result; The detection result merging module is used to merge the first hidden object detection result and the second hidden object detection result.

Images