Millimeter wave SAR human body security check image contraband detection method and device

Abstract

The invention provides a millimeter wave SAR human body security check image contraband detection method and device, and belongs to the field of millimeter wave SAR detection and identification, and the method comprises the steps: 1, collecting a human body security check image, and constructing a data set; step 2, labeling targets in the data set to obtain a labeled data set; step 3, performing iterative training on an FSRCNN model based on the data set in the step 1 to obtain a trained FSRCNN model, and inputting the marked data set in the step 2 into the trained FSRCNN model to obtain a super-resolution reconstructed image; 4, training the super-resolution reconstructed image based on an improved YOLOv5s algorithm, and obtaining a contraband detection model when the loss function lMPDIOU is minimum; 5, inputting a to-be-detected human body security check image into the trained FSRCNN model for super-resolution reconstruction, and inputting a super-resolution reconstructed image into the contraband detection model to obtain a detection result; in combination with an FSRCNN super-resolution reconstruction algorithm and an improved YOLOv5s target detection framework, the image quality is enhanced, and the target detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of millimeter wave SAR detection and recognition, and in particular to a method and device for detecting contraband in millimeter wave SAR human body security inspection images. Background Art

[0002] Synthetic Aperture Radar (SAR) images have strong anti-interference properties. Millimeter-wave SAR technology can provide more detailed images than traditional metal detectors, but it still has the following problems: (1) The resolution of millimeter-wave SAR imaging is usually still lower than that of X-ray scanning technology. Low resolution may result in unclear image details and difficulty in identifying smaller objects or objects that are in close contact with each other. (2) Millimeter-wave SAR imaging technology is very sensitive to environmental conditions, especially temperature and humidity. Temperature fluctuations will change the dielectric constant and density of the air, which in turn affects the propagation speed and direction of the millimeter waves, resulting in attenuation or scattering of the millimeter waves. Increased humidity can also cause similar problems because the moisture in the air affects the dielectric constant of the air. Fluctuations in these environmental variables may cause the quality of millimeter-wave SAR images to be unstable, manifested as reduced resolution and loss of details, which impairs the imaging effect and ultimately leads to poor detection and identification of dangerous goods. Summary of the invention

[0003] The technical problem to be solved by the present invention is how to improve the efficiency and accuracy of security inspection.

[0004] The present invention solves the above technical problems through the following technical solutions: a method for detecting contraband in millimeter wave SAR human body security inspection images, comprising:

[0005] Step 1: Collect human security inspection images and build a data set;

[0006] Step 2: Label the targets in the data set to obtain a labeled data set;

[0007] Step 3: Iteratively train the FSRCNN model based on the data set in step 1 to obtain a trained FSRCNN model, and input the labeled data set in step 2 into the trained FSRCNN model to obtain a super-resolution reconstructed image;

[0008] Step 4: Train the super-resolution reconstructed image based on the improved YOLOv5s algorithm. When the loss function l MPDIoU When it is minimum, the contraband detection model is obtained;

[0009] Step 5: Input the human body security inspection image to be detected into the trained FSRCNN model for super-resolution reconstruction, and input the super-resolution reconstructed image into the contraband detection model to obtain the detection result.

[0010] The present invention combines the FSRCNN super-resolution reconstruction algorithm with the improved YOLOv5s target detection framework to enhance image quality while improving the accuracy of target detection. It can effectively detect dangerous goods carried by the human body, especially in human body security inspection scenarios that require high-precision millimeter-wave SAR. It improves the efficiency and accuracy of the contraband detection system and helps to better identify and prevent potential security threats.

[0011] Preferably, in step 2, LabelImg image annotation software is used to annotate the rectangular boundaries of the targets in the data set according to the PASCAL VOC data format.

[0012] Preferably, the structure of the FSRCNN model in step three includes a feature extraction layer, a contraction layer, a nonlinear feature mapping layer, an expansion layer, and a deconvolution layer. The feature extraction layer extracts useful features from the input data set. The convolution kernel size of the feature extraction layer is 5, and the number of output channels is d. The contraction layer reduces the feature channels from d to s through a 1×1 convolution kernel. Each nonlinear feature mapping layer is a 3×3 convolution kernel. The image features are refined and enhanced through the nonlinear feature mapping layer. The expansion layer is a 1×1 convolution layer for expanding the number of feature channels from s back to d. The deconvolution layer uses ConvTranspose2d convolution to reduce the resolution of the feature map and upsample it to high resolution.

[0013] Preferably, in the step 3, the FSRCNN model is iteratively trained based on the data set of step 1 using the Adam algorithm for optimization training. When the loss function MSE reaches the minimum or reaches the total number of iterations, the training is stopped. The loss function MSE is used to measure the error between the super-resolution reconstructed image and the original high-resolution image, and the calculation formula is:

[0014]

[0015] Among them, Y i is the pixel value in the original image, is the pixel value in the super-resolution reconstructed image predicted by the FSRCNN model, and n is the number of samples in the dataset.

[0016] Preferably, the loss function l in step 4 is MPDIoU The calculation formula is:

[0017] l MPDIoU =1-MPDIoU

[0018]

[0019] Among them, d1 and d2 are the distances of the corresponding vertices, and They are the real boxes B gt The upper left and lower right corners, and They are respectively predicted boxes B pre In the upper left corner and lower right corner, w and h are the width and height of the input super-resolution reconstructed image, respectively.

[0020] The present invention also provides a millimeter wave SAR human body security inspection image contraband detection device, comprising:

[0021] A data acquisition unit, used to collect human body security inspection images and construct a data set;

[0022] A data annotation unit is used to annotate the objects in the data set to obtain annotated data sets;

[0023] The super-resolution reconstruction unit is used to iteratively train the FSRCNN model based on the data set of the data acquisition unit to obtain a trained FSRCNN model, and input the labeled data set in the data labeling unit into the trained FSRCNN model to obtain a super-resolution reconstructed image;

[0024] The model training unit is used to train the super-resolution reconstructed image based on the improved YOLOv5s algorithm. When the loss function l MPDIoU When it is minimum, the contraband detection model is obtained;

[0025] The model detection unit is used to input the human body security inspection image to be detected into the trained FSRCNN model for super-resolution reconstruction, and input the super-resolution reconstructed image into the contraband detection model to obtain the detection result.

[0026] Preferably, the data annotation unit uses LabelImg image annotation software to perform rectangular boundary annotation on the targets in the data set according to the PASCAL VOC data format.

[0027] Preferably, the structure of the FSRCNN model in the super-resolution reconstruction unit includes a feature extraction layer, a contraction layer, a nonlinear feature mapping layer, an expansion layer, and a deconvolution layer. The feature extraction layer extracts useful features from the input data set. The convolution kernel size of the feature extraction layer is 5, and the number of output channels is d. The contraction layer reduces the feature channels from d to s through a 1×1 convolution kernel. Each nonlinear feature mapping layer is a 3×3 convolution kernel. The image features are refined and enhanced through the nonlinear feature mapping layer. The expansion layer is a 1×1 convolution layer for expanding the number of feature channels from s back to d. The deconvolution layer uses ConvTranspose2d convolution to reduce the resolution of the feature map and upsample it to high resolution.

[0028] Preferably, the super-resolution reconstruction unit performs iterative training on the FSRCNN model based on the data set of the data acquisition unit and uses the Adam algorithm for optimization training. When the loss function MSE reaches the minimum or reaches the number of iterations, the training is stopped. The loss function MSE is used to measure the error between the super-resolution reconstructed image and the original high-resolution image. The calculation formula is:

[0029]

[0030] Among them, Y i is the pixel value in the original image, is the pixel value in the super-resolution reconstructed image predicted by the FSRCNN model, and n is the number of samples in the dataset.

[0031] Preferably, the loss function l in the model training unit MPDIoU The calculation formula is:

[0032] l MPDIoU =1-MPDIoU

[0033]

[0034] Among them, d1 and d2 are the distances of the corresponding vertices, and They are the real boxes B gt The upper left and lower right corners, and They are respectively predicted boxes B pre In the upper left corner and lower right corner, w and h are the width and height of the input super-resolution reconstructed image, respectively.

[0035] The advantages provided by the present invention are:

[0036] (1) The present invention combines the FSRCNN super-resolution reconstruction algorithm with the improved YOLOv5s target detection framework to enhance image quality while improving the accuracy of target detection. It can effectively detect dangerous goods carried by the human body, especially in human body security inspection scenarios that require high-precision millimeter-wave SAR. It improves the efficiency and accuracy of the contraband detection system and helps to better identify and prevent potential security threats.

[0037] (2) The present invention improves the rated resolution and detail richness of images through super-resolution processing, and utilizes super-resolution reconstruction technology to recover high-resolution image details from low-resolution millimeter-wave SAR security inspection images, thereby significantly improving the clarity of images and the visibility of edges and details, making it easier to identify potentially dangerous items hidden under clothing, such as knives and explosives, and at the same time enhancing the ability to distinguish between non-prohibited items such as electronic devices and metal objects and dangerous goods.

[0038] (3) Millimeter-wave SAR human body security inspection images are usually accompanied by high noise and clutter interference. This noise may cause the boundaries of real objects to be unclear or covered by noise, making the positioning of the bounding box more difficult. The MPDIoU loss function used in the present invention directly optimizes the position and size errors of the bounding box by reducing the average distance between the predicted box and the real box, thereby optimizing the positioning accuracy of the bounding box. In particular, in environments with high noise and complex backgrounds, it can reduce false positives and negatives of contraband and optimize target detection performance. In particular, when dealing with overlapping or dense target areas, it can significantly improve the accuracy and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of a method for detecting contraband in millimeter wave SAR human body security inspection images provided by an embodiment of the present invention;

[0040] Figure 2 A structural diagram of the FSRCNN model in the millimeter wave SAR human body security inspection image contraband detection method provided by an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of the MPDIoU loss function in the millimeter wave SAR human body security inspection image contraband detection method provided by an embodiment of the present invention;

[0042] Figure 4 A simulation diagram of a method for detecting contraband in millimeter wave SAR human body security inspection images provided by an embodiment of the present invention;

[0043] Figure 5(a) is a physical picture of a contraband ceramic knife;

[0044] Figure 5(b) is the millimeter wave SAR image of a volunteer carrying a ceramic knife;

[0045] Figure 6 (a) is a physical picture of a contraband wrench;

[0046] Figure 6(b) is a millimeter wave SAR image of a volunteer carrying a wrench;

[0047] Figure 7(a) is a physical picture of a contraband pistol;

[0048] Figure 7(b) is a millimeter wave SAR image of a volunteer carrying a pistol;

[0049] Figure 8 (a) is a physical picture of a prohibited metal folding knife;

[0050] Figure 8(b) is a millimeter wave SAR image of a volunteer carrying a metal folding knife;

[0051] Figure 9(a) is a physical picture of the contraband hammer;

[0052] Figure 9(b) is the millimeter-wave SAR image of a volunteer carrying a hammer. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the technical solution of the present invention is clearly and completely described below in combination with specific embodiments and with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] Example 1

[0055] like Figure 1 As shown, this embodiment provides a method for detecting contraband in millimeter wave SAR human body security inspection images, including the following steps:

[0056] Step 1: Collect human body security inspection images and construct a data set. The present invention uses the millimeter wave SAR human body security inspection equipment built in the laboratory to collect images of volunteers of different body shapes (tall, short, fat, thin) carrying common contraband and placing them on different parts of the human body, as shown in Figures 5(b), 6(b), 7(b), 8(b), and 9(b). Figures 5(a), 6(a), 7(a), 8(a), and 9(a) respectively show five common contraband items, including a ceramic knife, a wrench, a pistol, a metal folding knife, and a hammer.

[0057] Step 2: Label the objects in the data set to obtain a labeled data set. The present invention uses LabelImg image labeling software to label the objects in the data set with rectangular boundaries according to the PASCAL VOC data format to obtain a labeled data set, and divides the labeled data set into a training set and a validation set in a ratio of 8:2.

[0058] Step 3: Iteratively train the FSRCNN model based on the data set in step 1 to obtain a trained FSRCNN model. Input the labeled data set in step 2 into the trained FSRCNN model to obtain a super-resolution reconstructed image.

[0059] like Figure 2 As shown in Figure 1, the structure of the FSRCNN model includes feature extraction layer, contraction layer, nonlinear feature mapping layer, expansion layer, and deconvolution layer. The first layer of the network is the feature extraction layer, which uses a convolution with a kernel size of 5 and an output channel number of d to extract useful features from the input millimeter wave SAR human body security inspection image, and maintains the image size through appropriate padding, and then uses a PReLU activation function to enable the model to capture nonlinear relationships.

[0060] The second layer is the contraction layer, which reduces the feature channel from d to s through a 1×1 convolution kernel to reduce the number of model parameters and computational complexity, which helps to improve the operation speed and use the PRELU activation function at the same time.

[0061] The third layer is the nonlinear feature mapping layer. There are m nonlinear feature mapping layers. Each nonlinear feature mapping layer uses a 3×3 convolution kernel. The size of the feature map is kept unchanged through appropriate padding, and the PReLU activation function is used to further refine and enhance the feature representation of the image through m nonlinear feature mapping layers.

[0062] The fourth layer is the expansion layer. The expansion layer is a 1×1 convolution layer after the nonlinear feature mapping layer. It is used to expand the number of feature channels from s back to d, which helps to recover important information lost in the contraction layer. The PReLU activation function is also used in the expansion layer.

[0063] The fifth layer is the deconvolution layer, which uses ConvTranspose2d convolution to upsample the reduced-resolution feature map to a high-resolution one.

[0064] In step 3, the FSRCNN model is iteratively trained based on the data set in step 1 and optimized using the Adam algorithm. When the loss function MSE reaches the minimum or reaches the total number of iterations, the training is stopped. The loss function MSE is used to measure the error between the super-resolution reconstructed image and the original high-resolution image. The calculation formula is:

[0065]

[0066] Among them, Y i is the pixel value in the original image, is the pixel value in the super-resolution reconstructed image predicted by the FSRCNN model, and n is the number of samples in the dataset.

[0067] In step 3, the performance of the trained FSRCNN model is evaluated by the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM). The peak signal-to-noise ratio (PSNR) measures the image reconstruction quality by comparing the difference between the original image and the distorted image. The structural similarity index (SSIM) is used to measure the similarity between the reconstructed image and the original high-resolution image. The calculation formula of the peak signal-to-noise ratio (PSNR) is:

[0068]

[0069] Among them, MAX represents the maximum value of the pixel, MAX = 2 n -1, from which we get:

[0070]

[0071] The structural similarity index SSIM is an indicator used to measure the similarity between two images. The calculation formula of the structural similarity index SSIM is:

[0072]

[0073] Among them, μ x , μ y are the means of image x and image y respectively, σ x , σ y are the standard deviations of image x and image y, respectively, xy is the covariance of image x and image y, are the variances of image x and image y respectively, c1 and c2 are constants used for stability, c1=(k1L) 2 , c2=(k2L) 2 , L is the dynamic range of pixel values, k1 and k2 are very small constants.

[0074] The present invention applies the trained FSRCNN model weights to the labeled data set in step 2, improves the rated resolution and detail richness of the image through super-resolution processing, and uses super-resolution reconstruction technology to restore high-resolution image details from low-resolution millimeter-wave SAR security inspection images, significantly improving the image clarity and the visibility of edges and details, making it easier to identify potential dangerous items hidden under clothing, such as knives, explosives, etc., and at the same time enhancing the ability to distinguish between non-prohibited items and dangerous items such as electronic equipment and metal objects, which is particularly critical in security inspection work.

[0075] Step 4: Train the super-resolution reconstructed image based on the improved YOLOv5s algorithm. When the loss function l MPDIoU When it is minimum, the contraband detection model is obtained; the loss function l MPDIoU The calculation formula is:

[0076] l MPDIoU =1-MPDIoU

[0077]

[0078] Among them, d1 and d2 are the distances of the corresponding vertices, and They are the real boxes B gt The upper left and lower right corners, and They are respectively predicted boxes B pre In the upper left corner and lower right corner, w and h are the width and height of the input super-resolution reconstructed image, respectively.

[0079] Millimeter-wave SAR human body inspection images are usually accompanied by high noise and clutter interference. This noise may cause the boundaries of real objects to be unclear or covered by noise, making it more difficult to locate the bounding box. Although the CIoU loss function takes into account the overlapping area, center distance and aspect ratio of the bounding box, its performance may be affected in a high-noise environment because the model may have difficulty accurately distinguishing objects from noise or background. The CIoU loss function takes into account the center point distance and aspect ratio of the bounding box, but for objects with smaller areas, even a small position deviation may result in a large CIoU loss, which may not be conducive to the model learning to accurately detect small objects.

[0080] The MPDIoU loss function directly optimizes the position and size errors of the bounding box by reducing the average distance between the predicted box and the true box, such as Figure 3 As shown in the figure, this method can directly and effectively reduce the prediction error more than the CIoU loss function, because the MPDIoU loss function directly focuses on the distance instead of optimizing through multiple indirect parameters (such as aspect ratio). By integrating the MPDIoU loss function into the YOLOv5s algorithm, the positioning accuracy of the bounding box is optimized, especially in environments with high noise and complex backgrounds, which can reduce false positives and false negatives of contraband, optimize the target detection performance, especially when dealing with overlapping or dense areas of targets, and can significantly improve the accuracy and reliability of the system.

[0081] Step 5: Input the human body security inspection image to be detected into the trained FSRCNN model for super-resolution reconstruction, and input the super-resolution reconstructed image into the contraband detection model to obtain the detection result.

[0082] The human body inspection image to be detected is input into the trained FSRCNN model for 4x super-resolution reconstruction, and the weight of the contraband detection model is used to detect contraband on the super-resolution reconstructed image. The detection results can be displayed on the user interface. Super-resolution technology can improve the quality of human body inspection images to be detected, help improve the accuracy of contraband detection, reduce misjudgments, and improve overall security inspection efficiency.

[0083] like Figure 4 As shown in the figure, the original YOLOv5s is compared with the millimeter wave SAR human body security inspection data map after 4 times super resolution and the improved YOLOv5s is compared with the millimeter wave SAR human body security inspection data after 4 times super resolution. The horizontal axis represents the number of iterations and the vertical axis represents the average precision. It can be seen from the figure that the data without super resolution has the worst effect on YOLOv5s, and the data with four times super resolution has better effect on YOLOv5s than that without super resolution. The super-resolved and improved YOLOv5s has the best effect.

[0084] The present invention combines the FSRCNN super-resolution reconstruction algorithm with the improved YOLOv5s target detection framework to enhance image quality while improving the accuracy of target detection. It can effectively detect dangerous goods carried by the human body and demonstrates superior processing capabilities than traditional single technologies in actual operations, especially in human body security inspection scenarios requiring high-precision millimeter-wave SAR. It improves the efficiency and accuracy of the contraband detection system, helps to better identify and prevent potential security threats, and not only reduces the inconvenience and cost caused by false alarms, but also significantly improves the performance and practical value of the overall system.

[0085] Example 2

[0086] This embodiment provides a millimeter wave SAR human body security inspection image contraband detection device, including:

[0087] The data acquisition unit is used to collect human body security inspection images and construct a data set.

[0088] The data annotation unit is used to annotate the objects in the data set to obtain an annotated data set. LabelImg image annotation software is used to annotate the rectangular boundaries of the objects in the data set according to the PASCAL VOC data format.

[0089] The super-resolution reconstruction unit is used to iteratively train the FSRCNN model based on the data set of the data acquisition unit to obtain a trained FSRCNN model, and input the labeled data set in the data labeling unit into the trained FSRCNN model to obtain a super-resolution reconstructed image.

[0090] The structure of the FSRCNN model includes feature extraction layer, contraction layer, nonlinear feature mapping layer, expansion layer, and deconvolution layer. The feature extraction layer extracts useful features from the input data set. The convolution kernel size of the feature extraction layer is 5, and the number of output channels is d. The contraction layer reduces the feature channels from d to s through a 1×1 convolution kernel. Each nonlinear feature mapping layer is a 3×3 convolution kernel. The image features are refined and enhanced through the nonlinear feature mapping layer. The expansion layer is a 1×1 convolution layer, which is used to expand the number of feature channels from s back to d. The deconvolution layer uses ConvTranspose2d convolution to reduce the resolution of the feature map and upsample it to high resolution.

[0091] Iterative training uses the Adam algorithm for optimization training. When the loss function MSE reaches the minimum or reaches the number of iterations, the training is stopped. The loss function MSE is used to measure the error between the super-resolution reconstructed image and the original high-resolution image. The calculation formula is:

[0092]

[0093] Among them, Y iis the pixel value in the original image, is the pixel value in the super-resolution reconstructed image predicted by the FSRCNN model, and n is the number of samples in the dataset.

[0094] The model training unit is used to train the super-resolution reconstructed image based on the improved YOLOv5s algorithm. When the loss function l MPDIoU When it is minimum, the contraband detection model is obtained; the loss function l MPDIoU The calculation formula is:

[0095] l MPDIoU =1-MPDIoU

[0096]

[0097] Among them, d1 and d2 are the distances of the corresponding vertices, and They are the real boxes B gt The upper left and lower right corners, and They are respectively predicted boxes B pre In the upper left corner and lower right corner, w and h are the width and height of the input super-resolution reconstructed image, respectively.

[0098] The model detection unit is used to input the human body security inspection image to be detected into the trained FSRCNN model for super-resolution reconstruction, and input the super-resolution reconstructed image into the contraband detection model to obtain the detection result.

[0099] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting contraband in millimeter wave SAR human body security inspection images, characterized by: include: Step 1: Collect human security inspection images and build a data set; Step 2: Label the targets in the data set to obtain a labeled data set; Step 3: Iteratively train the FSRCNN model based on the data set in step 1 to obtain a trained FSRCNN model, and input the labeled data set in step 2 into the trained FSRCNN model to obtain a super-resolution reconstructed image; Step 4: Train the super-resolution reconstructed image based on the improved YOLOv5s algorithm. When the loss function When it is minimum, the contraband detection model is obtained; Step 5: Input the human body security inspection image to be detected into the trained FSRCNN model for super-resolution reconstruction, and input the super-resolution reconstructed image into the contraband detection model to obtain the detection result.

2. The method for detecting contraband in millimeter wave SAR human body security inspection images according to claim 1, characterized in that: In the step 2, LabelImg image annotation software is used to annotate the rectangular boundaries of the targets in the data set according to the PASCAL VOC data format.

3. The method for detecting contraband in millimeter wave SAR human body security inspection images according to claim 1, characterized in that: The structure of the FSRCNN model in the step three includes a feature extraction layer, a contraction layer, a nonlinear feature mapping layer, an expansion layer, and a deconvolution layer. The feature extraction layer extracts useful features from the input data set. The convolution kernel size of the feature extraction layer is 5, and the number of output channels is d. The contraction layer reduces the feature channels from d to s through a 1×1 convolution kernel. Each nonlinear feature mapping layer is a 3×3 convolution kernel. The image features are refined and enhanced through the nonlinear feature mapping layer. The expansion layer is a 1×1 convolution layer for expanding the number of feature channels from s back to d. The deconvolution layer uses ConvTranspose2d convolution to reduce the resolution of the feature map and upsample it to high resolution.

4. The method for detecting contraband in millimeter wave SAR human body security inspection images according to claim 1, characterized in that: In the step 3, the FSRCNN model is iteratively trained based on the data set of step 1 using the Adam algorithm for optimization training. When the loss function MSE reaches the minimum or reaches the total number of iterations, the training is stopped. The loss function MSE is used to measure the error between the super-resolution reconstructed image and the original high-resolution image. The calculation formula is: Among them, Y i is the pixel value in the original image, is the pixel value in the super-resolution reconstructed image predicted by the FSRCNN model, and n is the number of samples in the dataset.

5. The method for detecting contraband in millimeter wave SAR human body security inspection images according to claim 1, characterized in that: The loss function in step 4 The calculation formula is: Among them, d1 and d2 are the distances of the corresponding vertices, and They are the real boxes B gt The upper left and lower right corners, and They are respectively predicted boxes B pre In the upper left corner and lower right corner, w and h are the width and height of the input super-resolution reconstructed image, respectively.

6. Millimeter wave SAR human body security inspection image contraband detection device, characterized by: include: A data acquisition unit, used to collect human body security inspection images and construct a data set; A data annotation unit is used to annotate the objects in the data set to obtain annotated data sets; The super-resolution reconstruction unit is used to iteratively train the FSRCNN model based on the data set of the data acquisition unit to obtain a trained FSRCNN model, and input the labeled data set in the data labeling unit into the trained FSRCNN model to obtain a super-resolution reconstructed image; Model training unit, used to train super-resolution reconstructed images based on the improved YOLOv5s algorithm. When it is minimum, the contraband detection model is obtained; The model detection unit is used to input the human body security inspection image to be detected into the trained FSRCNN model for super-resolution reconstruction, and input the super-resolution reconstructed image into the contraband detection model to obtain the detection result.

7. The millimeter wave SAR human body security inspection image contraband detection device according to claim 6, characterized in that: The data annotation unit uses LabelImg image annotation software to perform rectangular boundary annotation on the targets in the data set according to the PASCAL VOC data format.

8. The millimeter wave SAR human body security inspection image contraband detection device according to claim 1, characterized in that: The structure of the FSRCNN model in the super-resolution reconstruction unit includes a feature extraction layer, a contraction layer, a nonlinear feature mapping layer, an expansion layer, and a deconvolution layer. The feature extraction layer extracts useful features from the input data set. The convolution kernel size of the feature extraction layer is 5, and the number of output channels is d. The contraction layer reduces the feature channels from d to s through a 1×1 convolution kernel. Each nonlinear feature mapping layer is a 3×3 convolution kernel. The image features are refined and enhanced through the nonlinear feature mapping layer. The expansion layer is a 1×1 convolution layer for expanding the number of feature channels from s back to d. The deconvolution layer uses ConvTranspose2d convolution to reduce the resolution of the feature map and upsample it to high resolution.

9. The millimeter wave SAR human body security inspection image contraband detection device according to claim 1, characterized in that: The super-resolution reconstruction unit performs iterative training on the FSRCNN model based on the data set of the data acquisition unit and uses the Adam algorithm for optimization training. When the loss function MSE reaches the minimum or reaches the number of iterations, the training is stopped. The loss function MSE is used to measure the error between the super-resolution reconstructed image and the original high-resolution image. The calculation formula is: Among them, Y i is the pixel value in the original image, is the pixel value in the super-resolution reconstructed image predicted by the FSRCNN model, and n is the number of samples in the dataset.

10. The millimeter wave SAR human body security inspection image contraband detection device according to claim 1, characterized in that: The loss function in the model training unit The calculation formula is: Among them, d1 and d2 are the distances of the corresponding vertices, and They are the real boxes B gt The upper left and lower right corners, and They are respectively predicted boxes B pre In the upper left corner and lower right corner, w and h are the width and height of the input super-resolution reconstructed image, respectively.

Images