A universal target detection and recognition method and system for CT security inspection equipment

Abstract

The present invention discloses a universal target detection and recognition method and system for CT security inspection equipment, the method comprising: obtaining at least one historical projection image, the historical projection image containing the target object; inputting at least one historical projection image into a pre-constructed feature extraction model in sequence, and outputting feature information corresponding to at least one historical projection image; compressing at least one feature information according to a preset mathematical model to obtain at least one compressed feature information, and constructing a feature information library according to at least one compressed feature information; obtaining a real-time projection image, and comparing the target feature information in the real-time projection image with each feature information in the feature information library based on a feature retrieval algorithm to obtain a recognition result with the real-time projection image. A data feature dimensionality reduction algorithm is established and can efficiently solve the optimal projection matrix, further compress features, and can quickly achieve data dimensionality reduction and retrieval, thereby improving the efficiency of feature retrieval.

Description

Technical Field

[0001] The present invention belongs to the technical field of target detection, and in particular relates to a universal target detection and recognition method and system for CT security inspection equipment. Background Art

[0002] At present, security inspection machines are a type of equipment used for security inspections, and are mainly used in public places such as airports, stations, and subways. Security inspection machines use technologies such as X-ray scanning or metal detectors to check people and objects to ensure the safety of public places. The application of security inspection machines not only improves the safety of public places, but also provides more convenient services for people to travel. In the context of the widespread application of artificial intelligence-related technologies, the intelligent identification technology of dangerous goods has gradually been applied to the traditional security inspection industry. The main application of artificial intelligence in security inspection machines is to use artificial intelligence technology to intelligently identify prohibited items in security inspection pictures. Target recognition can effectively solve the "pain point" of manual image judgment of security inspection machines. After the efforts of R&D personnel, artificial intelligence technology has begun to be applied. While reducing labor costs, it can improve the detection rate and reduce the missed detection rate. There is no need to replace security inspection equipment on a large scale, nor does it change the existing security inspection mode. Therefore, applying deep learning technology in the field of artificial intelligence to security inspection instruments is of great significance in reducing labor costs, improving security inspection efficiency, and reducing missed detection rates.

[0003] The intelligent recognition algorithm in traditional CT security inspection equipment generally uses projection images from multiple angles as the only input source, and then performs target detection in two-dimensional images. However, the detection of contraband by traditional intelligent recognition algorithms is based on known training data, and the trained model can only detect target objects in the training data. The algorithm cannot effectively identify items outside the training data. Therefore, due to the diversity of contraband in the inspected packages in security inspection CT equipment, it is almost impossible for the training data to cover all the contraband objects to be detected. Therefore, traditional intelligent recognition algorithms can only detect specific contraband objects. This will lead to a large number of missed detections in the actual use of CT equipment, resulting in low recognition accuracy. Summary of the invention

[0004] The present invention provides a universal target detection and recognition method and system for CT security inspection equipment, which is used to solve the technical problem that traditional intelligent recognition algorithms can only detect specific contraband objects, thereby causing a large number of missed detections in the actual use of CT equipment.

[0005] In a first aspect, the present invention provides a universal target detection and recognition method for CT security inspection equipment, comprising:

[0006] Acquire at least one historical projection image, where the historical projection image contains a target object;

[0007] The at least one historical projection image is sequentially input into a pre-built feature extraction model, and the feature extraction model outputs feature information corresponding to the at least one historical projection image, wherein the expression of the loss function in the feature extraction model is:

[0008] ,

[0009] In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector;

[0010] At least one feature information is compressed according to a preset mathematical model to obtain at least one compressed feature information, and a feature information library is constructed according to the at least one compressed feature information, wherein the mathematical model is expressed as:

[0011] ,

[0012] In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category;

[0013] A real-time projection image is acquired, and target feature information in the real-time projection image is compared with each feature information in the feature information library based on a feature retrieval algorithm to obtain a recognition result with the real-time projection image.

[0014] In a second aspect, the present invention provides a universal target detection and recognition system for CT security inspection equipment, comprising:

[0015] An acquisition module is configured to acquire at least one historical projection image, wherein the historical projection image contains a target object;

[0016] The output module is configured to sequentially input the at least one historical projection image into a pre-built feature extraction model, wherein the feature extraction model outputs feature information corresponding to the at least one historical projection image, wherein the expression of the loss function in the feature extraction model is:

[0017] ,

[0018] In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector;

[0019] The compression module is configured to compress at least one feature information according to a preset mathematical model to obtain at least one compressed feature information, and construct a feature information library according to the at least one compressed feature information, wherein the expression of the mathematical model is:

[0020] ,

[0021] In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category;

[0022] The recognition module is configured to obtain a real-time projection image, compare the target feature information in the real-time projection image with each feature information in the feature information library based on a feature retrieval algorithm, and obtain a recognition result with the real-time projection image.

[0023] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the universal target detection and recognition method for CT security inspection equipment of any embodiment of the present invention.

[0024] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor executes the steps of the universal target detection and recognition method for CT security inspection equipment of any embodiment of the present invention.

[0025] The universal target detection and recognition method and system for CT security inspection equipment of the present application designs a subject detection model based on target detection, then establishes a suitable data model to train the subject detection model according to the existing data, and then crops the target object according to the target frame detected by the subject detection model, and then uses a deep learning algorithm to roughly extract features from the cropped target object image, and then uses a preset mathematical model to extract features to accelerate the efficiency of feature retrieval, and finally uses feature retrieval and comparison based on the extracted features and the existing feature library, and uses a feature retrieval algorithm to perform feature comparison to determine the specific category of the target object, thereby ultimately achieving the purpose of identifying the target object. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 A flowchart of a universal target detection and recognition method for CT security inspection equipment provided by an embodiment of the present invention;

[0028] Figure 2 A structural block diagram of a universal target detection and recognition system for CT security inspection equipment provided by an embodiment of the present invention;

[0029] Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. 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.

[0031] See also Figure 1 , which shows a flow chart of a universal target detection and recognition method for CT security inspection equipment of the present application.

[0032] like Figure 1 As shown, the universal target detection and recognition method for CT security inspection equipment specifically includes the following steps:

[0033] Step S101: acquiring at least one historical projection image, wherein the historical projection image contains a target object.

[0034] In this step, we select training data suitable for the subject detection algorithm. Assume that the amount of data selected is N, and classify all the data labels into one category for model training. The network structure of the subject detection model mainly includes a 101-layer ResNet (residual network), a feature pyramid net (FPN), a class subnet prediction category subnet, and a box subnet (prediction box subnet) structure.

[0035] The loss function of the detection model uses the Focal loss function to improve the detection rate of difficult samples:

[0036] ,

[0037] ,

[0038] In the formula, y is the real label information of the target object, As parameters, is the confidence level of the target object category when y=1, is the value of p when the prediction is correct.

[0039] Step S102: inputting the at least one historical projection image into a pre-built feature extraction model in sequence, and the feature extraction model outputs feature information corresponding to the at least one historical projection image.

[0040] In this step, feature extraction is performed on the historical projection image, and the main network structure of the feature extraction model is divided into Backbone, Neck, Head and Loss parts.

[0041] Among them, Backbone (backbone network): a backbone network used to extract preliminary features of the input image, Neck (neck network): used for feature enhancement and feature dimension transformation. It can be a simple FC Layer (fully connected layer) for feature dimension transformation; it can also be a more complex FPN (feature pyramid network) structure for feature enhancement. Head (detection head) is used to transform the output feature of Neck through the logits function (logic function), so that the model can be trained in the form of classification tasks during the training phase. In addition to the commonly used FC Layer, it can also be replaced with other function modules. Loss specifies the Loss function used. We design Loss as a combination of loss, and select Classification Loss and Similarity preserving Loss to combine together. The Similarity preserving Loss is an improvement on the original TripletLoss (difficult triplet loss), which changes the optimization target from L2 Euclidean space to cosine space and adds a hard distance constraint between the anchor and the positive / negative (positive and negative samples), making the training and testing targets closer and improving the generalization ability of the model. The combination of (positive and negative samples) and center loss is expressed as:

[0042] ,

[0043] In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector;

[0044] Step S103: compress at least one feature information according to a preset mathematical model to obtain at least one compressed feature information, and construct a feature information library according to the at least one compressed feature information.

[0045] In this step, the extracted feature information is further compressed to accelerate the efficiency of feature library retrieval, the data features are further extracted and the similarities between different feature data points are considered to improve the performance of feature extraction, and the following mathematical model is established:

[0046] ,

[0047] In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category;

[0048] The expression for calculating the total scatter matrix of the data sample is:

[0049] ,

[0050] In the formula, is the mean of the sample;

[0051] The process of solving the model is as follows:

[0052] Initialize a similarity matrix , then fix the similarity matrix , by generalized eigenvalue decomposition to obtain the dimension reduction projection matrix , and rewrite the expression of the mathematical model into the target expression:

[0053] ,

[0054] In the formula, is a matrix, X is the matrix composed of all samples, L=DS, D is the column sum of S matrix, S is the similarity matrix, and L can be called Laplace matrix; where, equal It is composed of the eigenvectors corresponding to the smallest k eigenvalues;

[0055] According to the dimension reduction projection matrix Obtaining the similarity matrix , design similarity matrix Directly depends on the dimension reduction projection matrix ,Right now:

[0056] ,

[0057] For the dimension reduction projection matrix and the similarity matrix Perform alternating iterations to solve until the optimal projection dimension reduction matrix is ​​obtained. ;

[0058] According to the optimal projection dimension reduction matrix The feature information is reduced in dimension to obtain compressed feature information, where the expression of the compressed feature information is:

[0059] ,

[0060] In the formula, To compress feature information, It is feature information.

[0061] It should be noted that a feature information library is constructed based on at least one compressed feature information, which is used for later feature comparison of the detected target object to determine its category. The method for establishing the feature library is as follows:

[0062] Since the same item in the luggage package, even if it is of the same type, can have different features due to different placements and orientations, images of items at multiple angles should be selected for registration when establishing a feature library. Suppose that images of target objects at m angles and orientations are selected for registration.

[0063] The feature extraction model is used to extract features from the target object images at the selected m angles and directions, and the dimension of feature extraction is set to D dimension.

[0064] The extracted m D-dimensional features are further compressed and stored in the feature library, and the categories are identified for later feature comparison and recognition.

[0065] Step S104 , obtaining a real-time projection image, and comparing the target feature information in the real-time projection image with each feature information in the feature information library based on a feature retrieval algorithm to obtain a recognition result with the real-time projection image.

[0066] In this step, after feature detection is completed and the feature library of the object to be detected has been established, feature retrieval and matching algorithms are used to perform feature matching to identify the category of the object to be detected. The feature retrieval algorithm mainly includes input data, encoding, dimensionality reduction, composition, indexing and other modules.

[0067] The conventional indexing method is to search for vectors with similar distances through a given vector, which is a vector-to-vector search. That is, the first step is to encode the incoming image into a vector, and then use this vector to quickly find the topK (the top K after sorting from small to large) vectors with the closest distance in the given vector pool, and then map these topK vectors back to the image's label to find similar images. This process is relatively cumbersome, and it is also necessary to add database-like content to record the corresponding information between labels and vectors. In order to solve the problem of back-and-forth mapping, we introduced a bidirectional hash structure in the model's graph indexing algorithm, which supports the incoming label to be directly mapped into a vector in the model, and also supports the searched neighbor vector to be mapped back to the corresponding label. It greatly simplifies the engineering link.

[0068] Specifically, it is assumed that the dimension of the target feature information M in the real-time projection image is dimension, and the feature dimension of a feature information N in the feature information database is Dimension, where equal ; Calculate the similarity between the target feature information M and a certain feature information N, and determine whether the similarity is greater than a preset threshold ; If it is greater than the preset threshold , then the category corresponding to a certain feature information N is assigned to the target feature information M, that is, the category of the target feature information M is obtained; if it is not greater than the preset threshold , then calculate the similarity between the target feature information M and other feature information in the feature information database.

[0069] Among them, the expression for calculating similarity is:

[0070] ,

[0071] In the formula, is the i-th eigenvalue in the eigenvector M, is the i-th eigenvalue in the eigenvector N, is the dimension of the feature vector M.

[0072] In summary, the method of the present application solves the problem of being able to identify and detect objects even when they are not trained, thus improving the versatility and generalization of contraband detection in CT security inspection equipment; by registering object features in different directions and angles, an efficient feature retrieval and matching algorithm is used for matching and identification, thus improving the stability and accuracy of algorithm recognition. In order to improve the efficiency of feature retrieval, a data feature dimensionality reduction algorithm is established for the same-screen comparison algorithm, and the optimal projection matrix can be efficiently solved, and features can be further compressed, so that data dimensionality reduction can be quickly achieved and retrieved, thereby improving the efficiency of feature retrieval.

[0073] In a specific embodiment, a training data model is first established based on existing data, and all training data categories are classified into one category. After the data model is established, the subject detection model is used for model training, and then the trained model is used to detect the target frame of prohibited items. Secondly, the CT security inspection machine needs to establish a feature library of various items before performing target detection. The feature library is formed by processing a single object image through a feature extraction module, and each object in the feature library can have multiple feature vectors, that is, when the image of the same object has multiple directions and angles, each angle and direction can be collected and processed into a feature vector to improve the stability of the feature library. When the feature library is established, when performing contraband identification, when processing a package, the image data is first detected by the trained model for the target item frame. After obtaining all the target frames of the image data, the feature extraction module extracts features from each target frame to obtain the feature vector of each target frame. After obtaining the feature vector of the target frame, the established data feature dimension reduction is used to further extract features to accelerate the efficiency of feature retrieval. Finally, the obtained feature vector is compared with the feature in the feature library to determine the correspondence between the feature in the target frame and the feature in the feature library. Finally, the similarity of the two feature vectors is calculated. When the similarity is greater than the set threshold, it is determined that the recognition is successful, and the feature category corresponding to the feature vector in the feature library is set to the true category of the target frame. Otherwise, it is determined that the recognition fails. Finally, the above steps are repeated until all target frames in the current image data are processed and recognized.

[0074] See also Figure 2 , which shows a structural block diagram of a universal target detection and recognition system for CT security inspection equipment of the present application.

[0075] like Figure 2 As shown, the target detection and recognition system 200 includes an acquisition module 210 , an output module 220 , a compression module 230 and a recognition module 240 .

[0076] The acquisition module 210 is configured to acquire at least one historical projection image, wherein the historical projection image contains a target object;

[0077] The output module 220 is configured to sequentially input the at least one historical projection image into a pre-built feature extraction model, and the feature extraction model outputs feature information corresponding to the at least one historical projection image, wherein the expression of the loss function in the feature extraction model is:

[0078] ,

[0079] In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector;

[0080] The compression module 230 is configured to compress at least one feature information according to a preset mathematical model to obtain at least one compressed feature information, and construct a feature information library according to the at least one compressed feature information, wherein the mathematical model is expressed as:

[0081] ,

[0082] In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category;

[0083] The recognition module 240 is configured to obtain a real-time projection image, compare the target feature information in the real-time projection image with each feature information in the feature information library based on a feature retrieval algorithm, and obtain a recognition result with the real-time projection image.

[0084] It should be understood that Figure 2 Modules and references documented in Figure 1 Therefore, the operations and features described above for the method and the corresponding technical effects are also applicable to Figure 2 The modules in it will not be described in detail here.

[0085] In other embodiments, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program instructions are executed by a processor, the processor executes the universal target detection and recognition method for CT security inspection equipment in any of the above method embodiments;

[0086] As an implementation mode, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are configured as follows:

[0087] Acquire at least one historical projection image, where the historical projection image contains a target object;

[0088] The at least one historical projection image is sequentially input into a pre-built feature extraction model, and the feature extraction model outputs feature information corresponding to the at least one historical projection image, wherein the expression of the loss function in the feature extraction model is:

[0089] ,

[0090] In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector;

[0091] At least one feature information is compressed according to a preset mathematical model to obtain at least one compressed feature information, and a feature information library is constructed according to the at least one compressed feature information, wherein the mathematical model is expressed as:

[0092] ,

[0093] In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category;

[0094] A real-time projection image is acquired, and target feature information in the real-time projection image is compared with each feature information in the feature information library based on a feature retrieval algorithm to obtain a recognition result with the real-time projection image.

[0095] The computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the universal target detection and recognition system for CT security inspection equipment, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the universal target detection and recognition system for CT security inspection equipment via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0096] Figure 3 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 3 As shown, the device includes: a processor 310 and a memory 320. The electronic device may also include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 3 In the example, the bus connection is used. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, the universal target detection and recognition method for CT security inspection equipment in the above-mentioned method embodiment is realized. The input device 330 can receive input digital or character information, and generate key signal input related to user settings and function control of the universal target detection and recognition system for CT security inspection equipment. The output device 340 may include display devices such as display screens.

[0097] The electronic device can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0098] As an implementation mode, the electronic device is applied to a universal target detection and recognition system for CT security inspection equipment, and is used for a client, and includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can:

[0099] Acquire at least one historical projection image, where the historical projection image contains a target object;

[0100] The at least one historical projection image is sequentially input into a pre-built feature extraction model, and the feature extraction model outputs feature information corresponding to the at least one historical projection image, wherein the expression of the loss function in the feature extraction model is:

[0101] ,

[0102] In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector;

[0103] At least one feature information is compressed according to a preset mathematical model to obtain at least one compressed feature information, and a feature information library is constructed according to the at least one compressed feature information, wherein the mathematical model is expressed as:

[0104] ,

[0105] In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category;

[0106] A real-time projection image is acquired, and target feature information in the real-time projection image is compared with each feature information in the feature information library based on a feature retrieval algorithm to obtain a recognition result with the real-time projection image.

[0107] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 universal target detection and recognition method for CT security inspection equipment, characterized in that: include: Acquire at least one historical projection image, where the historical projection image contains a target object; The at least one historical projection image is sequentially input into a pre-built feature extraction model, and the feature extraction model outputs feature information corresponding to the at least one historical projection image, wherein the expression of the loss function in the feature extraction model is: ， In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector; At least one feature information is compressed according to a preset mathematical model to obtain at least one compressed feature information, and a feature information library is constructed according to the at least one compressed feature information, wherein the mathematical model is expressed as: ， In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category; A real-time projection image is acquired, and target feature information in the real-time projection image is compared with each feature information in the feature information library based on a feature retrieval algorithm to obtain a recognition result with the real-time projection image.

2. The universal target detection and recognition method for CT security inspection equipment according to claim 1 is characterized in that: The step of compressing at least one feature information according to a preset mathematical model to obtain at least one compressed feature information comprises: Initialize a similarity matrix , then fix the similarity matrix , by generalized eigenvalue decomposition to obtain the dimension reduction projection matrix , and rewrite the expression of the mathematical model into the target expression: ， In the formula, is a matrix, X is the matrix composed of all samples, L=DS, D is the column sum of S matrix, S is the similarity matrix, L can be called Laplace matrix, is the trace of the matrix; According to the dimension reduction projection matrix Obtaining the similarity matrix , design similarity matrix Directly depends on the dimension reduction projection matrix ,Right now: ， For the dimension reduction projection matrix and the similarity matrix Perform alternating iterations to solve until the optimal projection dimension reduction matrix is ​​obtained. ; According to the optimal projection dimension reduction matrix The feature information is reduced in dimension to obtain compressed feature information, where the expression of the compressed feature information is: ， In the formula, To compress feature information, It is feature information.

3. The universal target detection and recognition method for CT security inspection equipment according to claim 1 is characterized in that: in, The expression for calculating the total scatter matrix of the data sample is: ， In the formula, is the sample mean, is the total number of samples.

4. The universal target detection and recognition method for CT security inspection equipment according to claim 1 is characterized in that: The comparing the target feature information in the real-time projection image with each feature information in the feature information library based on the feature retrieval algorithm includes: Assume that the dimension of the target feature information M in the real-time projection image is dimension, and the feature dimension of a feature information N in the feature information database is dimension; Calculate the similarity between the target feature information M and a certain feature information N, and determine whether the similarity is greater than a preset threshold ; If it is greater than the preset threshold , then the category corresponding to a certain feature information N is assigned to the target feature information M, that is, the category of the target feature information M is obtained; If it is not greater than the preset threshold , then calculate the similarity between the target feature information M and other feature information in the feature information database.

5. A universal target detection and recognition system for CT security inspection equipment, characterized in that: include: An acquisition module is configured to acquire at least one historical projection image, wherein the historical projection image contains a target object; The output module is configured to sequentially input the at least one historical projection image into a pre-built feature extraction model, wherein the feature extraction model outputs feature information corresponding to the at least one historical projection image, wherein the expression of the loss function in the feature extraction model is: ， In the formula, is the loss function, is the normalized exponential loss function, is a hyperparameter, is the center loss function, is the number of clusters, is the number of data samples, is the dimension reduction projection matrix, is the i-th sample, is the bias term of the i-th sample, is the i-th cluster center, is the matrix transpose, is the jth weight vector, For the A weight vector; The compression module is configured to compress at least one feature information according to a preset mathematical model to obtain at least one compressed feature information, and construct a feature information library according to the at least one compressed feature information, wherein the expression of the mathematical model is: ， In the formula, is the number of categories of data samples, is the number of samples in the kth category, is the similarity between the i-th sample and the j-th sample in the k-th class sample, is the total scatter matrix of the data sample, is the i-th feature information in the k-th category, is the jth feature information in the kth category; The recognition module is configured to obtain a real-time projection image, compare the target feature information in the real-time projection image with each feature information in the feature information library based on a feature retrieval algorithm, and obtain a recognition result with the real-time projection image.

6. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images