CN105023025B - An open-set trace image classification method and system - Google Patents

Abstract

The invention relates to a method and a system for classifying open-set trace images, which realize automatic classification of trace images. It adopts multilayer cascade mode to judge whether the image to be classified belongs to a certain category or new category in the trace image library. The invention has the characteristics of high speed, less manual participation and relatively accurate and comprehensive result, can realize automatic expansion of the image library and serial-parallel of cases by classifying field traces, and provides great help for police to handle cases. Therefore, the invention can be widely applied to the field of trace classification.

Description

An open-set trace image classification method and system

technical field

The invention relates to a classification method and system, in particular to an open-set trace image classification method and system.

Background technique

On-site traces are one of the important physical evidence in criminal investigation, and most of the extracted traces are stored in the form of images. How to efficiently classify and manage on-site trace images has become an urgent problem to be solved.

Trace image classification is a kind of image classification. At present, image classification methods are mainly used to: (1) determine whether the image contains objects of interest, such as people, airplanes, etc.; (2) determine the scene represented by the image, such as forest, Living room and factory, etc.; (3) Determine the morphological attributes of objects in the image, such as color, shape, etc. The basic framework of the image classification method is: (1) need to manually or automatically mark each image in the image library; (2) learn and train the classifier according to the marked samples; (3) extract the features of the image to be classified ; (4) Similarity measure; (5) Use classifier to classify.

The problems existing in the current image classification methods are: (1) The accuracy of the classifier depends on the accuracy of the markers. Since there are many trace images and their quality is poor, and most of the markers of the trace images are automatically marked, it is very normal for the markers to be wrong. It greatly affects the classification accuracy. (2) At present, most image classification methods are based on the similarity between the image to be classified and the common features of various types of images, and the trace images are mostly disturbed and incomplete images. If the common features within the class are obtained, it is possible The useful information of various images is weakened or lost. (3) In the current image classification method, if an image is added to the image library, retraining is required. Since there are many images in the trace image library, it is difficult to implement if each new image needs to be retrained.

Contents of the invention

In view of the above problems, the object of the present invention is to provide an efficient open-collection trace image classification method and system to realize the classification of traces extracted on the spot and the automatic expansion of the trace image library, and realize the serialization of cases through the classification of traces on the spot. And, provide great help for the police handling the case.

In order to achieve the above object, the present invention adopts the following technical solutions: an open-set trace image classification method, which includes the following steps: 1) preprocessing the image to be classified to form the specification of the trace image stored in the trace image library, Wherein the trace image library includes several trace images and the category of each trace image; 2) according to the similarity of each trace image in the preprocessed image to be classified and the trace image library, the candidate category of the image to be classified is screened; 3) according to The similarity between the image to be classified and the trace image ranked first in the candidate category and its corresponding representative image is used to determine whether the image to be classified belongs to the category or a new category. If the classification of the image to be classified cannot be directly given, then Go to the next step, otherwise, go to step 6); 4) According to the majority voting principle or the similarity gap judgment condition between the first and second trace images, it is judged that the image to be classified belongs to the category, if it cannot be directly given the The classification situation of the image to be classified, then enter the next step, otherwise, enter step 6); 5) provide the representative image in the candidate category of the image to be classified to the user, and interactively determine the category of the image to be classified by it, and enter the step 6); 6) The trace image library is updated, and the trace image stored in it is automatically expanded.

Described step 2) comprises the following steps: 1. calculate the feature of image to be classified, denoted as ②Calculate the similarity between the image to be classified and each trace image in the trace image library according to the characteristics of the image to be classified, recorded as S={s ₁ ,...,s _i ,...,s _N } where s _i =sim(T _x ,xi ₎ represents the similarity between the image to be classified and the trace image d _i ; _xi is the feature set of each trace image in the trace image library D, that is, the feature library is recorded as X={x ₁ ,…, _xi ,… , x _N }, where Represents the feature of the i-th trace image, and the feature is a d-dimensional real number; ③ Arrange the values in S in descending order, and get and from Sequentially select the first f types of images and their categories, as the candidate category of the image T to be classified, denoted as P.

Said step 3) includes the following steps: ① record the similarity between the image to be classified and the trace image ranked first in the candidate category P of the image to be classified as s _F , and record the image to be classified and the candidate category of the image to be classified The similarity of the representative image of the category of the trace image ranked first in P is recorded as s _R ; ②If s _F <s _New , then directly identify the image to be classified as a new category, and enter step 6); otherwise, enter the next step Step 1, where s _New represents the threshold value for the newly added category; ③ If s _F ≥ s _Dir and s _R ≥ s _Rth , then the image to be classified is directly identified as the category of the trace image that ranks first, and proceeds to step 6 ); otherwise, go to step 4), where s _Dir represents the similarity threshold between the image to be classified and the first trace image, and s _Rth represents the similarity threshold between the image to be classified and the class representative image of the first trace image , and s _Dir >s _Rth .

The acquisition process of the representative image in the step 3) is as follows: 1. extract the feature of each trace image in the trace image library to form a feature library X={x ₁ ,..., _xi ,...,x _N } where Represents the feature of the i-th trace image, and the feature is a d-dimensional real number; ② Calculate the similarity between members of each category in the trace image library, and obtain the similarity matrix between members of each category, denoted as W _{i ,j} (k), where l _i =l _j =k, and k=1,...,K; ③According to W _i,j (k) according to Calculate the representative image of each category That is, in each type of trace image in the trace image library, select the trace image with the largest similarity product with all other trace images of this type as the representative image.

Said step 4) comprises the following steps: 1. among the candidate categories P of images to be classified, if the number of trace images ranked in the top q number belongs to the same category, then the image to be classified belongs to this category, and enters step 6), Otherwise, go to the next step; ②In the candidate category P of the image to be classified, if the similarity with the first-ranked trace image is much greater than the similarity with the second-ranked trace image, that is, when the similarity gap is greater than Δs, then The category of the image to be classified is the category of the first trace image, go to step 6), otherwise go to step 5).

The step 5) includes the following steps: 1. present the representative images of the categories of the images in the candidate category P of the images to be classified to the user in order, and display the images to be classified at the same time; 2. The user checks the category according to each representative image The images and rankings that appear in the candidate category P of the image to be classified; ③The user judges the category of the image to be classified according to the information in ① and ②.

Described step 6) comprises the following steps: 1. the update of the trace image database, namely D=D∪{T}, the subscript index number of the trace image newly entered in the trace image database is N+1; the trace image database D includes several Trace images and the category to which each trace image belongs, wherein, D={d ₁ ,...,d _i ,...,d _N }, and represents the i-th trace image, the height and width of which are m and n respectively, The marks of the trace image library are marked as L={l ₁ ,…,l _i ,…,l _N }, which represents the tag category number of the trace image d _i , and there are K categories in the trace image library; ②The image to be classified belongs to the category l Class _t represents the update of the image R(l _t ). If the image to be classified is a new category, the image to be classified is directly a representative image, that is, R(l _t )=N+1; otherwise, proceed as follows: A, Calculate the similarity product between each non-to-be-classified image and all images of this class in class l _t , and find the maximum value R ₁ in all similarity products and the index number i ^* corresponding to the image with the maximum value: in B. Calculate the similarity product between the image to be classified and other images in the l _t class, that is, the representative image of the C, updated l _t class is: ③Mark update, ie L=L∪{l _t }; ④Feature library update, ie X=X∪{T _x }; ⑤Increase the number of trace images in the trace image library by 1, ie N=N+1.

An open-collection trace image classification system is characterized in that it includes a preprocessing module, an initial screening module, a trace image library, a direct identification module, a supplementary recording module, and a subjective judgment module, and the trace image library stores trace images, Each trace image belongs to the category; the preprocessing module preprocesses the image to be classified into the specification of the trace image stored in the trace image library, and transmits the preprocessed image to be classified to the initial screening module; The initial screening module screens the candidate category of the image to be classified according to the similarity between the preprocessed image to be classified and each trace image in the trace image library, and transmits the candidate category to the direct identification module; the direct The identification module determines whether the image to be classified belongs to the category or a new category according to the similarity between the image to be classified and the trace image ranked first in the candidate category and its corresponding representative image. If the module cannot directly give the image to be classified If the classification situation of the image is different, then the candidate category of the image to be classified is sent to the supplementary recording module, otherwise, the category of the image to be classified is transmitted to the trace image library for updating; the supplementary recording module is based on the principle of majority voting or The similarity gap judgment condition between the first and second trace images judges the category of the image to be classified. If the module cannot directly give the classification of the image to be classified, the candidate category of the image to be classified will be sent to the subjective judgment module, otherwise, the category of the image to be classified is sent to the trace image library for updating; the subjective judgment module provides the representative image of the candidate category of the image to be classified to the user, and the interactive Discriminate the category of the image to be classified, and transmit the category of the image to be classified to the trace image database for updating; the trace image database will supplement the image to be classified into the trace image database, if it is an existing category, then Add to the corresponding category, update the representative image of this category; otherwise, add a new category, set up a new category, update the representative image of this category; at the same time, update the number of feature databases, marks and trace images.

Because the present invention adopts the above technical scheme, it has the following advantages: 1. The method of the present invention includes: 1) performing preprocessing on the images to be classified to form the specifications of the trace images stored in the trace image database, wherein the trace image database includes several The trace image and the category of each trace image; 2) according to the similarity between the preprocessed image to be classified and each trace image in the trace image library, the candidate category of the image to be classified is screened; 3) according to the image to be classified and the candidate category The similarity between the first-ranked trace image and its corresponding representative image judges whether the image to be classified belongs to the category or a newly added category. If the classification of the image to be classified cannot be directly given, go to the next step; otherwise, the trace The image library is updated, and the trace images stored in it are automatically expanded; 4) According to the principle of majority voting or the similarity gap judgment condition between the first and second trace images, the category of the image to be classified is judged. If it cannot be directly given The classification situation of the image to be classified, then enter the next step, otherwise, the trace image database is updated, and the trace image stored in it is automatically expanded; 5) The representative image in the candidate category of the image to be classified is provided to the user, and the user is interactively judged The category of the image to be classified, the trace image library is updated, and the trace image stored in it is automatically expanded; the above method is used to realize the automatic expansion and marking of the trace image library. In addition, the above steps 3), 4) and 5) are multi-layer cascaded classification strategies, which are used to classify trace images from multiple levels, thereby greatly improving the classification accuracy; especially in step 5), the classification reliability When it is not high, representative images of candidate categories are also provided for human-computer interaction judgment, and only this step requires manual identification, thereby saving police time in handling cases. 2. The present invention uses the representative image as the representative of each category to carry out similarity comparison with the image to be classified. With the traditional image classification method that only utilizes each image to compare the similarity with the image to be classified, it solves the problem caused by the prior automatic The problem of low classification accuracy caused by labeling errors. 3. The present invention compares the image to be classified with the similarity of each trace image in the trace image database. Compared with the traditional image classification method that uses the common features of the image to be classified and all images, it is affected by various interferences. Smaller, so as to improve the classification accuracy, the result is relatively accurate and comprehensive. 4. In the case of adding new images, the present invention does not need to recalculate the trace images of the entire trace image database, but only updates the representative images in its category. Compared with the traditional method, training is required every time a new image is added. As far as the algorithm is concerned, the running time of the algorithm is greatly reduced, and the speed is relatively fast. Therefore, the present invention can be widely used in the field of trace classification.

Description of drawings

Fig. 1 is the overall flow chart diagram of the inventive method

Figure 2 is the generation process of representative images in the trace image library

Fig. 3 is the structural representation of the system of the present invention

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

As shown in Figure 1, an open-set trace image classification algorithm includes the following steps:

1) Preprocessing the image to be classified to form the specification of the trace image stored in the trace image library, which includes the following steps:

①The image to be classified The acquisition resolution and acquisition angle are adjusted to be on the same specification as the trace images stored in the trace image library to improve the classification accuracy, wherein the height and width of the image to be classified are m and n respectively;

② Extract the traces of interest in the image to be classified from the background in order to remove background interference;

Due to the complex background of the crime scene, there is no practical automatic extraction method at present. The more effective method is to extract the traces from the background through human-computer interaction;

③A mean value filter is used to remove Gaussian noise interference from the trace image, the filter window size is 6 times the standard deviation of Gaussian noise, and the standard deviation of Gaussian noise is approximated by the standard deviation of the smooth area in the image;

④Using the histogram specification to adjust the illumination conditions of the images to be classified to be consistent with the acquisition conditions of the images in the trace image database, and obtain the preprocessed images to be classified.

The above trace image database, denoted as D, includes several trace images and the category of each trace image, where D={d ₁ ,...,d _i ,...,d _N }, and Represents the i-th trace image, the height and width of which are m and n respectively, and the marks of the trace image library are marked as L={l ₁ ,…,l _i ,…,l _N }, where Indicates the tag category number of the trace image d _i , and there are K categories in the trace image library.

2) According to the similarity between the preprocessed image to be classified and each trace image in the trace image database, the candidate category of the image to be classified is screened, which includes the following steps:

① Calculate the features of the image to be classified, denoted as

According to the type of traces in the image to be classified, features are extracted in a corresponding way. The commonly used methods are SIFT (Scale-invariant feature transform, scale-invariant feature transform), SURF (Speeded up robust features, accelerated robust features) and wavelet Fourier leaf transformation coefficients, etc.;

②Calculate the similarity between the image to be classified and each trace image in the trace image library according to the characteristics of the image to be classified, recorded as S={s ₁ ,…,s _i ,…,s _N }, where s _i =sim(T _x , x _i ) represent the similarity between the image to be classified and the trace image d _i ; there are many definitions of the similarity function sim( ), and in this implementation, the normalized cross-correlation coefficient is used; _xi is the trace image The feature set of each trace image in library D, that is, the feature library is recorded as X={x ₁ ,…, _xi ,…,x _N }, where Represents the feature of the i-th trace image, and the feature is a d-dimensional real number;

③ Arrange the values in S in descending order to get and from Sequentially select the first f types of images and their categories, as the candidate category of the image T to be classified, denoted as P.

The above f generally takes 100.

3) According to the similarity between the image to be classified and the trace image ranked first in the candidate category and its corresponding representative image, it is judged whether the image to be classified belongs to the category or a new category, if the classification of the image to be classified cannot be directly given situation, then enter the next step, otherwise, enter step 6), which includes the following steps:

① Denote the similarity between the image to be classified and the trace image that ranks first in the candidate category P of the image to be classified as s _F , and the trace image that ranks first in the candidate category P of the image to be classified and the image to be classified The category of the image represents the similarity of the image and is recorded as s _R ;

② If s _F < s _New , the image to be classified is directly identified as a newly added category, and proceeds to step 6); otherwise, enters the next step, where s _New represents the threshold value for identifying as a newly added category, if the value is set too large, It is possible to identify a non-new image as a new image, and the threshold is generally 0.55, because the value range of image similarity is between 0 and 1, so it can be defined that the similarity below 0.55 is two different trace images, in line with human subjective thinking;

③If s _F ≥ s _Dir and s _R ≥ s _Rth , then the image to be classified is directly identified as the category of the first trace image, and enter step 6); otherwise, the classification of the image to be classified cannot be directly given , enter the next step, where s _Dir represents the similarity threshold between the image to be classified and the first-ranked trace image. The threshold is generally 0.71. If it is above 0.71, it is two similar images. Between 0.55 and 0.71, it is necessary to continue to judge the two images. Whether the two trace images are similar or not, below 0.55 is the two trace images that are not similar; s _Rth represents the similarity threshold between the image to be classified and the representative image of the category of the first trace image, the threshold is generally 0.63, and it is considered to be greater than 0.63 The trace image to be classified belongs to this category, otherwise it does not belong to this category, and s _Dir >s _Rth .

As shown in Figure 2, the above representative image acquisition process is as follows:

① Extract the features of each trace image in the trace image library to form a feature library;

Features need to be extracted in different ways according to the type of traces, such as SIFT, SURF, and wavelet Fourier transform coefficients. The feature library is recorded as X={x ₁ ,…,xi _i ,…,x _N }, where Represents the feature of the i-th trace image, and the feature is a d-dimensional real number;

②Calculate the similarity between members of each category in the trace image library, and obtain the similarity matrix between members of each category, which is denoted as W _i,j (k), where l _i =l _j =k, And k=1,...,K;

③ According to W _i,j (k) according to Calculate the representative image of each category That is, in each type of trace image in the trace image database, select the trace image with the largest similarity product with all other trace images of this type as the representative image.

4) Judging that the image to be classified belongs to the category according to the principle of majority voting or the similarity difference between the first and second trace images, if the classification of the image to be classified cannot be directly given, go to the next step , otherwise, enter step 6), which includes the following steps:

① Among the candidate categories P of images to be classified, if the number of trace images ranked in the top q is large and belong to the same category (majority voting principle), then the image to be classified belongs to this category, and then go to step 6), otherwise go to the next step step;

The above q generally takes 3.

②In the candidate category P of the image to be classified, if the similarity with the trace image ranked first is much greater than the similarity with the trace image ranked second, that is, when the similarity gap is greater than Δs, the category of the image to be classified For the category of the first trace image, go to step 6), otherwise go to the next step;

The above Δs is generally taken as 0.08.

5) The representative image in the candidate category of the image to be classified is provided to the user, and the category of the image to be classified is interactively judged by it, and enters step 6), which includes the following steps:

① Present the representative images of the categories of the images in the candidate category P of the images to be classified to the user in sequence, and display the images to be classified at the same time;

②According to each representative image, the user checks the images and rankings of its category in the candidate category P of the image to be classified;

③The user judges the category of the image to be classified according to the information in ① and ②.

6) The trace image library is updated and automatically expanded, which includes the following steps:

① The update of the trace image library, that is, D=D∪{T}, the subscript index number of the trace image newly entered into the trace image library is N+1, and the open set update is realized;

② The category l _t of the image to be classified represents the update of the image R(l _t ), if the image to be classified is a newly added class, the image to be classified is directly the representative image, that is, R(l _t )=N+1; otherwise, Proceed as follows:

A. Calculate the similarity product of each non-to-be-classified image and all images of this class in the l _t class, and find the index number i ^* corresponding to the maximum value _R in all similarity products and the trace image of the maximum value: in

B. Calculate the similarity product between the image to be classified and other trace images in the class l _t , that is

C. The representative image of the updated l _t class is:

③The update of the mark, that is, L=L∪{l _t };

④Update of feature database, that is, X＝X∪{T _x };

⑤ The number of trace images in the trace image library is increased by 1, that is, N=N+1.

It should be noted that, if the present invention is applied for the first time: if the trace image library is empty, the image to be classified is automatically added to the trace image library as a new category and marked; if there are trace images of the marked category in the trace image library , according to the similarity between the image to be classified and the trace image in the trace image database, it is divided into three cases for processing:

1) automatically identify the image to be classified as a category in the trace image library;

2) When the reliability of the automatic determination is not high, a list of representative images of the candidate categories is provided, and the judgment is made by human-computer subjective interaction;

3) Automatically identify the image to be classified as a newly added category, and automatically expand the category of the trace image library.

As shown in Figure 3, an open-set trace image classification system includes a preprocessing module 1, an initial screening module 2, a trace image library 3, a direct identification module 4, a supplementary recording module 5, and a subjective judgment module 6.

The preprocessing module 1 performs preprocessing on the images to be classified, and sends the preprocessed images to the initial screening module 2 .

The initial screening module 2 selects the candidate category of the image to be classified based on the similarity between the preprocessed image to be classified and the trace image in the trace image library 3 , and sends the candidate category to the direct identification module 4 .

The direct identification module 4 determines whether the image to be classified belongs to a category or a newly added category according to the similarity between the image to be classified and the trace image ranked first in the candidate category and its corresponding representative image, if the module cannot directly give the For the classification status of the image to be classified, the candidate category of the image to be classified is sent to the supplementary recording module 5, otherwise, the category of the image to be classified is sent to the trace image library 3 for updating.

The supplementary recording module 5 judges that the image to be classified belongs to the category according to the principle of majority voting or the similarity gap judgment condition between the first and second trace images, if the module cannot directly provide the classification of the image to be classified, The candidate category of the image to be classified is sent to the subjective decision module 6, otherwise, the category of the image to be classified is sent to the trace image database 3 for updating.

The subjective judgment module 6 provides representative images of the candidate categories of the images to be classified to the user, who interactively determines the categories of the images to be classified, and transmits the categories of the images to be classified to the trace image database 3 for updating.

The trace image library 3 supplements the image to be classified into the trace image library 3, if it is an existing category, it is added to the corresponding category, and the representative image of this category is updated; otherwise, it is a new category, a new category is set up, and the category is updated. Represents an image; at the same time, updates the number of feature databases, markers, and trace images.

Finally, it should be noted that the above embodiments are only used to illustrate rather than limit the technical solution of the present invention, and the present invention may also have other various embodiments. Those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and essence of the present invention. If these changes and deformations fall within the scope of the claims of the present invention and their equivalent technologies, then these changes and deformations shall all belong to the protection scope of the appended claims of the present invention.

Claims (7)

1. A kind of open set trace image classification method, it comprises the following steps: 1) Preprocessing the image to be classified to form the specification of the trace image stored in the trace image database, wherein the trace image database includes a number of trace images and the category to which each trace image belongs; 2) According to the similarity between the preprocessed image to be classified and each trace image in the trace image database, the candidate category of the image to be classified is screened, including the following steps: ① Calculate the features of the image to be classified, denoted as ②Calculate the similarity between the image to be classified and each trace image in the trace image library according to the characteristics of the image to be classified, recorded as S={s ₁ ,…,s _i ,…,s _N }, where s _i =sim(T _x , x _i ) represent the similarity between the image to be classified and the trace image d _i ; _xi is the feature set of each trace image in the trace image library D, that is, the feature library is recorded as X={x ₁ ,…, _xi , …,x _N }, where Represents the feature of the i-th trace image, and the feature is a d-dimensional real number; ③ Arrange the values in S in descending order to get and from Sequentially select the first f class images and their categories, as the candidate category of the image T to be classified, denoted as P; 3) According to the similarity between the image to be classified and the trace image ranked first in the candidate category and its corresponding representative image, it is judged whether the image to be classified belongs to the category or a new category, if the classification of the image to be classified cannot be directly given situation, then enter the next step, otherwise, enter step 6), including the following steps: ① Denote the similarity between the image to be classified and the trace image that ranks first in the candidate category P of the image to be classified as s _F , and the trace image that ranks first in the candidate category P of the image to be classified and the image to be classified The similarity of the representative image of the category of the image is recorded as s _R , where the acquisition process of the representative image is as follows: a. Extract the features of each trace image in the trace image library to form a feature library X={x ₁ ,…, _xi ,…,x _N }, where Represents the feature of the i-th trace image, and the feature is a d-dimensional real number; b. Calculate the similarity between members of each category in the trace image library, and obtain the similarity matrix between members of each category, which is denoted as W _i,j (k), where l _i =l _j =k , and k=1,...,K; c. According to W _i,j (k) according to Calculate the representative image of each category That is, in each type of trace image in the trace image library, select the trace image with the largest similarity product with all other trace images of this type as the representative image; ②If s _F <s _New , then directly identify the image to be classified as a new category, and enter step 6); otherwise, enter the next step, where s _New represents the threshold value for identifying as a new category; ③If s _F ≥ s _Dir and s _R ≥ s _Rth , then directly identify the image to be classified as the category of the first trace image, and enter step 6); otherwise, enter step 4), where s _Dir means to be classified The similarity threshold between the image and the first trace image, s _Rth represents the similarity threshold between the image to be classified and the class representative image of the first trace image, and s _Dir >s _Rth ; 4) Judging that the image to be classified belongs to the category according to the principle of majority voting or the similarity difference between the first and second trace images, if the classification of the image to be classified cannot be directly given, go to the next step , otherwise, go to step 6); 5) Provide the representative image in the candidate category of the image to be classified to the user, and interactively determine the category to which the image to be classified belongs, and enter step 6); 6) The trace image library is updated, and the trace image stored in it is automatically expanded. 2. A kind of open set trace image classification method as claimed in claim 1, is characterized in that: described step 4) comprises the following steps: 1. Among the candidate categories P of images to be classified, if the number of trace images ranked in the top q number belongs to the same category, then the image to be classified belongs to this category, and enters step 6), otherwise enters the next step; ② In the candidate category P of the image to be classified, calculate the similarity between the trace image ranked first and the trace image ranked second. When the similarity difference is greater than Δs, the category of the image to be classified is the first If it is the category of the name trace image, go to step 6), otherwise go to step 5), where the value of Δs is 0.08. 3. a kind of open set trace image classification method as claimed in claim 1, is characterized in that: described step 5) comprises the following steps: ① Present the representative images of the categories of the images in the candidate category P of the images to be classified to the user in sequence, and display the images to be classified at the same time; ②According to each representative image, the user checks the images and rankings of its category in the candidate category P of the image to be classified; ③ The user determines the category of the image to be classified according to the information in ① and ②. 4. a kind of open set trace image classification method as claimed in claim 2, is characterized in that: described step 5) comprises the following steps: ① Present the representative images of the categories of the images in the candidate category P of the images to be classified to the user in sequence, and display the images to be classified at the same time; ②According to each representative image, the user checks the images and rankings of its category in the candidate category P of the image to be classified; ③The user judges the category of the image to be classified according to the information in ① and ②. 5. a kind of open set trace image classification method as claimed in claim 1 or 4, is characterized in that: described step 6) comprises the following steps: ① The update of the trace image database, that is, D=D∪{T}, the subscript index number of the newly entered trace image database is N+1; The trace image library D includes several trace images and the category to which each trace image belongs, where D={d ₁ ,...,d _i ,...,d _N }, and Represents the i-th trace image, the height and width of which are m and n respectively, and the marks of the trace image library are marked as L={l ₁ ,…,l _i ,…,l _N }, where Indicates the tag category number of the trace image d _i , and there are K categories in the trace image library; ② The category l _t of the image to be classified represents the update of the image R(l _t ), if the image to be classified is a newly added category, the image to be classified is directly a representative image, that is, R(l _t )=N+1; otherwise, Proceed as follows: A. Calculate the similarity product of each non-to-be-classified image and all images of this class in the l _t class, and find the index number i ^* corresponding to the maximum value _R in all similarity products and the image of the maximum value: in B. Calculate the similarity product between the image to be classified and other images in the l _t class, that is C. The representative image of the updated l _t class is: ③The update of the mark, that is, L=L∪{l _t }; ④Update of feature database, that is, X=X∪{T _x }; ⑤ The number of trace images in the trace image library is increased by 1, that is, N=N+1. 6. A kind of open set trace image classification method as claimed in claim 2, is characterized in that: described step 6) comprises the following steps: ① The update of the trace image database, that is, D=D∪{T}, the subscript index number of the newly entered trace image database is N+1; The trace image library D includes several trace images and the category to which each trace image belongs, where D={d ₁ ,...,d _i ,...,d _N }, and Represents the i-th trace image, the height and width of which are m and n respectively, and the marks of the trace image library are marked as L={l ₁ ,…,l _i ,…,l _N }, where Indicates the tag category number of the trace image d _i , and there are K categories in the trace image library; ② The category l _t of the image to be classified represents the update of the image R(l _t ), if the image to be classified is a newly added class, the image to be classified is directly the representative image, that is, R(l _t )=N+1; otherwise, Proceed as follows: A. Calculate the similarity product of each non-to-be-classified image and all images of this class in the l _t class, and find the index number i ^* corresponding to the maximum value _R in all similarity products and the image of the maximum value: in B. Calculate the similarity product between the image to be classified and other images in the l _t class, that is C. The representative image of the updated l _t class is: ③The update of the mark, that is, L=L∪{l _t }; ④Update of feature database, that is, X＝X∪{T _x }; ⑤ The number of trace images in the trace image library is increased by 1, that is, N=N+1. 7. A system of an open-set trace image classification method according to any one of claims 1 to 6, characterized in that it includes a preprocessing module, an initial screening module, a trace image library, a direct identification module, and a supplementary recording module and a subjective judgment module, and trace images are stored in the trace image library, and the category to which each trace image belongs; The preprocessing module preprocesses the image to be classified into the specification of the trace image stored in the trace image library, and transmits the preprocessed image to be classified to the initial screening module; The initial screening module screens the candidate category of the image to be classified according to the similarity between the preprocessed image to be classified and each trace image in the trace image library, and transmits the candidate category to the direct identification module; The direct determination module judges whether the image to be classified belongs to a category or a newly added category according to the similarity between the image to be classified and the trace image ranked first in the candidate category and its corresponding representative image, if the module cannot directly give For the classification status of the image to be classified, the candidate category of the image to be classified is sent to the supplementary recording module, otherwise, the category of the image to be classified is sent to the trace image library for updating; The supplementary recording module judges the category of the image to be classified according to the majority voting principle or the similarity drop judgment condition between the first and second trace images, if the module cannot directly give the classification of the image to be classified , the candidate category of the image to be classified is sent to the subjective judgment module, otherwise, the category of the image to be classified is sent to the trace image library for updating; The subjective judgment module provides representative images of the candidate categories of the images to be classified to the user, and interactively determines the categories of the images to be classified, and transmits the categories of the images to be classified to the trace image library for updating; The trace image library supplements the image to be classified into the trace image library, if it is an existing category, it is added to the corresponding category, and the representative image of this category is updated; otherwise, it is a new category, a new category is set up, and the new category is updated. representative image of the category; at the same time, the number of feature databases, markers, and trace images are updated.