Method and apparatus for selecting optimal model for deep learning-based millimeter wave image detection

Abstract

The application relates to a kind of optimal model selection methods and devices based on deep learning millimeter wave image detection, the method comprises: collecting two-dimensional millimeter wave human body image;Concealed article labeling is carried out on two-dimensional millimeter wave human body image;Two-dimensional millimeter wave human body image and the corresponding labeled concealed article information are input to convolutional neural network for training to obtain a model sequence;Select optimal concealed article threshold value for each model in model sequence;Compare multiple models, determine the optimal detection model and the corresponding article threshold value.The application selects the optimal detection model and the threshold value of the detected article by selecting the model trained by neural network based on deep learning of millimeter wave security image, to achieve the optimal detection rate and false alarm rate of millimeter wave image, reduce human error, improve the detection efficiency of millimeter wave image human body carrying concealed article, improve the effectiveness of millimeter wave human body security equipment.

Description

Technical Field

[0001] This invention relates to the field of target detection technology, and specifically to a method and apparatus for selecting the optimal model for millimeter-wave image detection based on deep learning. Background Technology

[0002] Millimeter waves (3GHz-300GHz) and terahertz waves (300GHz-10THz) are electromagnetic waves that lie between light waves and radio waves. Millimeter waves can penetrate all clothing fabrics, and the energy of millimeter waves radiated by the human body is stronger than that of metals, ceramics, plastic explosives, powdered explosives, clothing, and insulating materials. Therefore, active / passive millimeter wave technology can detect various contraband items such as knives, guns, and explosives hidden on the surface of the human body. Millimeter waves have the following advantages: they can be used to detect metallic and non-metallic weapons, including explosives; they can penetrate ordinary clothing and be reflected by the human body and concealed objects; active millimeter wave images have good image contrast and spatial resolution, showing the location and shape of concealed objects; low-dose, moderate millimeter waves are non-ionizing and harmless to the human body, making them easily accepted by the public. Due to their superior capabilities compared to metal detection technology and their safety compared to X-ray technology, human body millimeter wave security inspection technology has developed rapidly in the past decade. Passive focal plane array scanning technology, multi-beam frequency scanning technology, and active three-dimensional holographic millimeter wave technology have been tested and applied successively.

[0003] After active millimeter-wave security inspection equipment achieves millimeter-wave imaging of the human body, the key to system application lies in how to analyze and process the images to detect, identify, and distinguish dangerous items. Achieving rapid and accurate detection of concealed items has significant practical implications. Similar to visible light images, target detection algorithms for millimeter-wave images have been extensively researched and applied. Millimeter-wave image target detection has evolved from traditional methods based on Haar features, histogram of Oriented Gradients (HOR) features, and support vector machines to deep learning methods based on convolutional neural networks (CNNs), such as R-CNN, Fast R-CNN, Faster R-CNN, and YOLO. For different types of deep learning methods, the general detection process involves: ① data acquisition; ② dividing the data into training and test sets; ③ data labeling; ④ training a specific neural network on the training set to obtain a training model; and ⑤ testing and validating the training model on the test set to obtain the optimal model. The number of training steps is usually determined based on personal experience, and the result may not be the optimal detection model. Therefore, how to obtain an optimal detection model and select the optimal detection threshold under that model is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this invention is to overcome the shortcomings of the prior art and provide a method and apparatus for selecting the optimal model for millimeter-wave image detection based on deep learning. This method is based on deep learning training of millimeter-wave security inspection images using neural networks, which can select the optimal detection model and the threshold for detected items, so as to achieve the optimal detection rate and false alarm rate of millimeter-wave images. The model selection is changed from manual to automatic, reducing the requirements for personnel, reducing human error, improving the detection efficiency of concealed items carried by the human body in millimeter-wave images, and improving the effectiveness of millimeter-wave human body security inspection equipment.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for selecting the optimal model for millimeter-wave image detection based on deep learning, comprising:

[0006] Acquire two-dimensional millimeter-wave human body images;

[0007] Hidden objects are marked on the two-dimensional millimeter-wave human body image;

[0008] The two-dimensional millimeter-wave human body image and the corresponding labeled hidden object information are input into a convolutional neural network for training to obtain a model sequence;

[0009] For each model in the model sequence, select the optimal hidden item threshold;

[0010] By comparing multiple models, the optimal detection model and its corresponding item threshold are determined.

[0011] Optionally, the acquisition of two-dimensional millimeter-wave human body images includes:

[0012] Human body imaging results are acquired by a millimeter-wave human body security inspection device, and the imaging results are projected onto a two-dimensional plane at a preset fixed angle to obtain several two-dimensional millimeter-wave human body images.

[0013] Optionally, for planar imaging systems, two images of the human body, one from the front and one from the back, can be obtained directly;

[0014] For cylindrical imaging systems, based on the rotational geometry between images, one image each from the front and back of the human body is used as a standard image, and the pixels of all images obtained in one scan are transformed in coordinates. Specifically, all front images are mapped to the front standard image, and all back images are mapped to the back standard image, resulting in several two-dimensional millimeter-wave human body images from the same front or back perspective.

[0015] Optionally, the step of marking concealed objects on the two-dimensional millimeter-wave human body image includes:

[0016] Through multiple data collections, a large number of two-dimensional images of human bodies carrying various suspicious items were obtained. Rectangular frames were used to mark the location and category of the suspicious items in the front or back standard images of the two-dimensional millimeter-wave human bodies.

[0017] Optionally, selecting the optimal concealed item threshold for each model in the model sequence includes:

[0018] The detection model is used to detect specific samples. The thresholds for different items are traversed from 0 to 1 in increments of 0.01. The detection rate and false alarm rate of the model under different threshold conditions are tested, and the optimal hidden item threshold for each model is determined based on the detection rate and false alarm rate obtained from the test.

[0019] Optionally, comparing multiple models to determine the optimal detection model and its corresponding item threshold includes:

[0020] The model is used to detect samples under different threshold conditions for different items;

[0021] After the detection is completed, the corresponding detection rate and false alarm rate are calculated, and the weighted sum of the detection rate and the false alarm rate is used as the measured value.

[0022] Select the optimal hidden item threshold sequence An when the measured value is the largest; generate S1 optimal detection models and their corresponding thresholds for different detection models 1 to S1.

[0023] Compare the measured values ​​X1 to XS1 under case S1, select the model and threshold corresponding to the maximum measured value, and determine the model corresponding to the maximum measured value as the optimal detection model, and determine the threshold corresponding to the maximum measured value as the item threshold corresponding to the optimal detection model.

[0024] The present invention also provides a deep learning-based device for selecting the optimal model for millimeter-wave image detection, comprising:

[0025] The acquisition module is used to acquire two-dimensional millimeter-wave human body images;

[0026] The annotation module is used to annotate concealed objects on the two-dimensional millimeter-wave human body image;

[0027] The training module is used to input the two-dimensional millimeter-wave human body image and the corresponding labeled hidden object information into the convolutional neural network for training, so as to obtain a model sequence;

[0028] The selection module is used to select the optimal hidden item threshold for each model in the model sequence.

[0029] The determination module is used to compare multiple models and determine the optimal detection model and its corresponding item threshold.

[0030] Optionally, selecting the optimal concealed item threshold for each model in the model sequence includes:

[0031] The detection model is used to detect specific samples. The thresholds for different items are traversed from 0 to 1 in increments of 0.01. The detection rate and false alarm rate of the model under different threshold conditions are tested, and the optimal hidden item threshold for each model is determined based on the detection rate and false alarm rate obtained from the test.

[0032] Optionally, comparing multiple models to determine the optimal detection model and its corresponding item threshold includes:

[0033] The model is used to detect samples under different threshold conditions for different items;

[0034] After the detection is completed, the corresponding detection rate and false alarm rate are calculated, and the weighted sum of the detection rate and the false alarm rate is used as the measured value.

[0035] Select the optimal hidden item threshold sequence An when the measured value is the largest; generate S1 optimal detection models and their corresponding thresholds for different detection models 1 to S1.

[0036] Compare the measured values ​​X1 to XS1 under case S1, select the model and threshold corresponding to the maximum measured value, and determine the model corresponding to the maximum measured value as the optimal detection model, and determine the threshold corresponding to the maximum measured value as the item threshold corresponding to the optimal detection model.

[0037] Furthermore, the present invention also provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the deep learning-based millimeter-wave image detection optimal model selection method as described in any of the preceding claims.

[0038] This invention employs the above technical solution, and the method for selecting the optimal model for millimeter-wave image detection based on deep learning includes: acquiring a two-dimensional millimeter-wave human body image; marking concealed objects on the two-dimensional millimeter-wave human body image; inputting the two-dimensional millimeter-wave human body image and the corresponding marked concealed object information into a convolutional neural network for training to obtain a model sequence; selecting the optimal concealed object threshold for each model in the model sequence; comparing multiple models to determine the optimal detection model and its corresponding object threshold. This method selects the optimal detection model and detection threshold for millimeter-wave security images based on a model trained using deep learning with a neural network, thereby achieving the optimal detection rate and false alarm rate for millimeter-wave images. The model selection is automated instead of manually, reducing the requirements for personnel, minimizing human error, improving the detection efficiency of concealed objects carried by the human body in millimeter-wave images, and enhancing the effectiveness of millimeter-wave human body security inspection equipment. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating an embodiment of the optimal model selection method for millimeter-wave image detection based on deep learning according to the present invention.

[0041] Figure 2 This is a schematic diagram illustrating the selection of the optimal detection model in an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the structure of an embodiment of the millimeter-wave image detection optimal model selection device based on deep learning according to the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0044] Deep learning can be used to train and detect millimeter-wave images, enabling the detection of various prohibited items such as knives, guns, and explosives hidden on the surface of the human body. However, there is still no good solution for how to obtain the optimal training model and the corresponding item detection threshold to achieve the best detection effect, the highest detection rate and the lowest false alarm rate.

[0045] Figure 1 This is a schematic diagram of the overall process of an embodiment of the optimal model selection method for millimeter-wave image detection based on deep learning according to the present invention.

[0046] like Figure 1 As shown, the method described in this embodiment of the invention includes:

[0047] S11: Acquire two-dimensional millimeter-wave human body images;

[0048] Furthermore, two-dimensional millimeter-wave human body images are obtained from millimeter-wave human body security inspection equipment. The specific acquisition process can be as follows: a person stands still within the imaging system; a broadband signal is emitted by a rotating or vertically scanning radar array; the signal reflected back from the human body is received; a series of signal processing operations are performed to synthesize a human body image. The imaging result is projected onto a two-dimensional plane at certain fixed angles to obtain several two-dimensional images. Due to the occlusion effect, half of the images capture the frontal information of the human body, while the other half captures the rear information.

[0049] For planar imaging systems, two images, one from the front and one from the back, can be obtained directly. For cylindrical imaging systems, based on the rotational geometry between the images, one image from the front and one from the back of the human body can be used as a standard image to perform coordinate transformation on the pixels of all images obtained in one scan. Specifically, all front images are mapped to the front standard image, and all back images are mapped to the back standard image to obtain several images from the same viewpoint on the front or back.

[0050] S12: Mark hidden objects on the two-dimensional millimeter-wave human body image;

[0051] Furthermore, through multiple data acquisitions, a large number of two-dimensional images of human bodies carrying various suspicious items were obtained. Rectangular frames were used to mark the location and category of the suspicious items in the front or back standard images of the two-dimensional millimeter-wave human bodies.

[0052] S13: Input the two-dimensional millimeter-wave human body image and the corresponding labeled hidden object information into a convolutional neural network for training to obtain a model sequence;

[0053] Two-dimensional millimeter-wave human images and their corresponding bounding boxes are input into a convolutional neural network for training. Typical networks used here include CNN, R-CNN, Fast R-CNN, Faster R-CNN, and YOLO. The object detection network model can include one or more of four basic modules: a feature extraction network, a region selection network, a region of interest pooling layer, and a classification and location detection network. During training, according to the model convergence procedure, a new model is generated every 50,000 steps initially, and after full convergence, a new model is generated every 10,000 steps, forming a model sequence.

[0054] S14: Select the optimal hidden item threshold for each model in the model sequence;

[0055] Furthermore, the detection model is used to detect specific selected samples. Thresholds for different items are iterated from 0 to 1 in increments of 0.01. The model's detection rate and false alarm rate are tested under different threshold conditions, and different weights m and n are assigned to the detection rate and false alarm rate, respectively. Typically, m=2 and n=1. The weight of the detection rate is higher than that of the false alarm rate. For each model, a set of optimal hidden item thresholds is obtained.

[0056] Step S14 is characterized by classifying various types of concealed items, using a trained model to detect test image samples, using different thresholds for all categories of items, weighting the detection rate and false alarm rate of the detection results differently, and selecting the optimal threshold group by calculating the highest score.

[0057] S15: Compare multiple models to determine the optimal detection model and its corresponding item threshold.

[0058] Furthermore, under different threshold conditions for different items, the model is used to detect the samples. After detection, the corresponding detection rate D and false alarm rate F are calculated, and then multiplied by the corresponding weights to obtain a measured value X1 = D*m + F*n. The threshold sequence An when X1 is maximized is selected. For different detection models 1 to S1, a total of S1 optimal detection models and their corresponding thresholds can be generated. The measured values ​​X1 to XS1 under S1 conditions are compared, and the model with the largest measured value and its corresponding threshold are selected. The model corresponding to the largest measured value is determined as the optimal detection model, and the threshold corresponding to the largest measured value is determined as the detection item threshold corresponding to the optimal detection model. For the process of selecting the optimal detection model, please refer to [link to relevant documentation]. Figure 2 As shown, the vertical axis R represents the measured values ​​X1 to XS1. Figure 2 It can be seen that when the 45th model sample is selected as the optimal detection model, its corresponding item threshold is the optimal item threshold.

[0059] Compared to traditional millimeter-wave image detection methods based on deep learning such as neural networks, this solution achieves better detection rates and false alarm rates, effectively improving the detection efficiency of millimeter-wave security inspection equipment and increasing the detection rate of concealed items.

[0060] Furthermore, the method also includes: expanding the sample set. When the number of sample images is insufficient, the sample images can be expanded by means of image flipping, brightness adjustment, contrast adjustment, etc., to increase the number of training samples.

[0061] Step S13 may specifically include: classifying the prepared sample set, for example, taking 90% as training samples and 10% as validation samples, training the samples using a neural network to obtain a detection model. Based on the model's training loss curve, before the curve stabilizes, a model is generated every 50,000 steps, and after the curve stabilizes, a model is generated every 10,000 steps, for a total of S1 models.

[0062] In practical applications, representative millimeter-wave images are selected, typically depicting five men and five women. 1000 images are acquired from each image, and the location and type of concealed objects are manually labeled. These images serve as detection samples to evaluate the performance of the detection model. The detection model is then used to detect these specific samples. Thresholds for different objects are applied in increments of 0.01 from 0 to 1. The model's detection rate and false alarm rate are tested under different threshold conditions, and different weights m and n are assigned to the detection rate and false alarm rate, respectively. Typically, m=2 and n=1.

[0063] Under different threshold conditions for different items, the model is used to detect samples. After detection, the detection rate D and false alarm rate F are calculated, and then multiplied by the corresponding weights to obtain a measured value X1 = D*m + F*n. The threshold sequence An when X1 is maximized is selected. For different detection models 1 to S1, a total of S1 optimal detection models and their corresponding thresholds can be generated. The measured values ​​X1 to XS1 under S1 conditions are compared, and the model with the largest measured value and its corresponding threshold are selected. The model corresponding to the largest measured value is determined as the optimal detection model, and the threshold corresponding to the largest measured value is determined as the item threshold corresponding to the optimal detection model.

[0064] The method described in this invention acquires a large number of two-dimensional millimeter-wave human body security inspection images, trains the sample images using various deep learning methods such as neural networks, generates training models in multiple stages during the training process, and evaluates the detection thresholds of various types of concealed items using a step-by-step approach in different models, obtains the optimal detection threshold set under the current model, compares multiple models, obtains the optimal training model, and finally obtains the optimal detection model and detection item threshold pair to obtain the optimal detection result. This method is beneficial for improving the detection rate of security inspection equipment, reducing the false alarm rate of security inspection equipment, reducing the requirements for personnel use, reducing human error, improving the accuracy of inspection of concealed dangerous items on the human body, reducing security inspection time, and enhancing the humanized application of millimeter-wave human body security inspection equipment.

[0065] Figure 3 This is a schematic diagram of the structure of an embodiment of the millimeter-wave image detection optimal model selection device based on deep learning according to the present invention.

[0066] like Figure 3 As shown, the apparatus described in this embodiment of the invention includes:

[0067] Acquisition module 31 is used to acquire two-dimensional millimeter-wave human body images;

[0068] The annotation module 32 is used to annotate concealed objects on the two-dimensional millimeter-wave human body image;

[0069] Training module 33 is used to input the two-dimensional millimeter-wave human body image and the corresponding labeled hidden object information into the convolutional neural network for training, so as to obtain a model sequence;

[0070] Selection module 34 is used to select the optimal hidden item threshold for each model in the model sequence;

[0071] The determination module 35 is used to compare multiple models and determine the optimal detection model and its corresponding detection item threshold.

[0072] Furthermore, selecting the optimal concealed item threshold for each model in the model sequence includes:

[0073] The detection model is used to detect specific samples. The thresholds for different items are traversed from 0 to 1 in increments of 0.01. The detection rate and false alarm rate of the model under different threshold conditions are tested, and the optimal hidden item threshold for each model is determined based on the detection rate and false alarm rate obtained from the test.

[0074] Furthermore, the comparison of multiple models to determine the optimal detection model and its corresponding detection item threshold includes:

[0075] The model is used to detect samples under different threshold conditions for different items;

[0076] After the detection is completed, the corresponding detection rate and false alarm rate are calculated, and the weighted sum of the detection rate and the false alarm rate is used as the measured value.

[0077] Select the optimal hidden item threshold sequence An when the measured value is the largest; generate S1 optimal detection models and their corresponding thresholds for different detection models 1 to S1.

[0078] Compare the measured values ​​X1 to XS1 under case S1, select the model and threshold corresponding to the maximum measured value, and determine the model corresponding to the maximum measured value as the optimal detection model, and determine the threshold corresponding to the maximum measured value as the detection item threshold corresponding to the optimal detection model.

[0079] The working principle of the deep learning-based millimeter-wave image detection optimal model selection device described in this embodiment is the same as that of the deep learning-based millimeter-wave image detection optimal model selection method described above, and will not be repeated here.

[0080] The device described in this invention improves the detection rate of concealed dangerous items on the human body by security inspection equipment, reduces the false alarm rate of security inspection equipment, reduces the requirements for personnel to use it, reduces human error, improves the accuracy of inspection of concealed dangerous items on the human body, reduces security inspection time, and enhances the humanized application of millimeter-wave human body security inspection equipment.

[0081] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the deep learning-based millimeter-wave image detection optimal model selection method as described in the preceding embodiments.

[0082] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0083] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0084] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0085] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0086] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0087] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0088] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0089] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0090] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for selecting the optimal model for millimeter-wave image detection based on deep learning, characterized in that, include: Acquire two-dimensional millimeter-wave human body images; Hidden objects are marked on the two-dimensional millimeter-wave human body image; The two-dimensional millimeter-wave human body image and the corresponding labeled hidden object information are input into a convolutional neural network for training to obtain a model sequence; For each model in the model sequence, select the optimal hidden item threshold; By comparing multiple models, the optimal detection model and its corresponding item threshold are determined; The step of selecting the optimal hidden item threshold for each model in the model sequence includes: The detection model is used to detect specific samples. The thresholds for different items are traversed from 0 to 1 in increments of 0.

01. The detection rate and false alarm rate of the model under different threshold conditions are tested, and the optimal hidden item threshold for each model is determined based on the detection rate and false alarm rate obtained from the test. Multiple types of concealed items are classified, and a trained model is used to detect test image samples. Different thresholds are used for detection of items in all categories. The detection rate and false alarm rate of the detection results are weighted differently, and the optimal threshold group is selected by calculating the highest score. The comparison of multiple models to determine the optimal detection model and its corresponding item threshold includes: The model is used to detect samples under different threshold conditions for different items; After the detection is completed, the corresponding detection rate and false alarm rate are calculated, and the weighted sum of the detection rate and the false alarm rate is used as the measured value. Select the optimal hidden item threshold sequence An when the measured value is the largest; generate S1 optimal detection models and their corresponding thresholds for different detection models 1 to S1. Compare the measured values ​​X1 to XS1 under case S1, select the model and threshold corresponding to the maximum measured value, and determine the model corresponding to the maximum measured value as the optimal detection model, and determine the threshold corresponding to the maximum measured value as the item threshold corresponding to the optimal detection model.

2. The method according to claim 1, characterized in that, The acquisition of two-dimensional millimeter-wave human body images includes: Human body imaging results are acquired by a millimeter-wave human body security inspection device, and the imaging results are projected onto a two-dimensional plane at a preset fixed angle to obtain several two-dimensional millimeter-wave human body images.

3. The method according to claim 2, characterized in that, For planar imaging systems, two images of the human body, one from the front and one from the back, can be obtained directly; For cylindrical imaging systems, based on the rotational geometry between images, one image each from the front and back of the human body is used as a standard image, and the pixels of all images obtained in one scan are transformed in coordinates. Specifically, all front images are mapped to the front standard image, and all back images are mapped to the back standard image, resulting in several two-dimensional millimeter-wave human body images from the same front or back perspective.

4. The method according to claim 1, characterized in that, The process of marking concealed objects on the two-dimensional millimeter-wave human body image includes: Through multiple data collections, a large number of two-dimensional images of human bodies carrying various suspicious items were obtained. Rectangular frames were used to mark the location and category of the suspicious items in the front or back standard images of the two-dimensional millimeter-wave human bodies.

5. An apparatus employing the deep learning-based millimeter-wave image detection optimal model selection method as described in any one of claims 1-4, characterized in that, include: The acquisition module is used to acquire two-dimensional millimeter-wave human body images; The annotation module is used to annotate concealed objects on the two-dimensional millimeter-wave human body image; The training module is used to input the two-dimensional millimeter-wave human body image and the corresponding labeled hidden object information into the convolutional neural network for training, so as to obtain a model sequence; The selection module is used to select the optimal hidden item threshold for each model in the model sequence. The determination module is used to compare multiple models and determine the optimal detection model and its corresponding item threshold.

6. The apparatus according to claim 5, characterized in that, The step of selecting the optimal hidden item threshold for each model in the model sequence includes: The detection model is used to detect specific samples. The thresholds for different items are traversed from 0 to 1 in increments of 0.

01. The detection rate and false alarm rate of the model under different threshold conditions are tested, and the optimal hidden item threshold for each model is determined based on the detection rate and false alarm rate obtained from the test.

7. The apparatus according to claim 5, characterized in that, The comparison of multiple models to determine the optimal detection model and its corresponding item threshold includes: The model is used to detect samples under different threshold conditions for different items; After the detection is completed, the corresponding detection rate and false alarm rate are calculated, and the weighted sum of the detection rate and the false alarm rate is used as the measured value. Select the optimal hidden item threshold sequence An when the measured value is the largest; generate S1 optimal detection models and their corresponding thresholds for different detection models 1 to S1. Compare the measured values ​​X1 to XS1 under case S1, select the model and threshold corresponding to the maximum measured value, and determine the model corresponding to the maximum measured value as the optimal detection model, and determine the threshold corresponding to the maximum measured value as the item threshold corresponding to the optimal detection model.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the deep learning-based millimeter-wave image detection optimal model selection method as described in any one of claims 1 to 4.

Images