A method, apparatus, medium, and equipment for classifying categories
By employing the Online Category Mining (OCD) method, and utilizing supervised learning and hashing projectors, the limitations of existing visual recognition models in classifying new categories are addressed. This enables instant classification of unseen categories, enhancing the model's openness and generalization ability.
Patent Information
- Application Number
- CN202310076588.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-01-16
- Publication Date
- 2025-12-02
- Estimated Expiration
- 2043-01-16
AI Technical Summary
Existing visual recognition models can only predict categories that have already been seen and cannot effectively classify new categories. Furthermore, existing technologies rely on auxiliary information or offline reasoning and cannot achieve real-time classification.
The Online Category Mining (OCD) method is adopted. By defining a data structure, the data is divided into a support set and a query set. A classification model is trained using supervised learning, and feature projection is performed through a projector with hashing and sign-amplitude decoupling to achieve instant classification of unseen categories.
It enables effective classification of unseen categories without the need for auxiliary information, possesses instant reasoning capabilities, and enhances the openness and generalization ability of visual recognition models.
Smart Images
Figure CN117115497B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of image target detection technology, and more specifically, to a method, apparatus, medium, and device for classifying categories. Background Technology
[0002] Deep learning models have been widely used in visual recognition. However, existing visual recognition models can still only predict for a specific set of categories. Deep models trained on annotated data can easily handle closed-set recognition tasks—that is, the categories to be identified have all appeared in the training data annotations—due to their powerful fitting capabilities. However, these models can be said to be learning by memorizing data, because they are limited to the categories seen in training and lack the ability to model new categories. This is a significant gap compared to human visual recognition capabilities. For humans, given samples containing new categories, users can not only distinguish which are new but also which samples might belong to a new category. For example, even if you have never seen a "hedgehog," you can easily recognize that they are different from other creatures you have seen before and recognize that multiple hedgehog images belong to the same category, even if you don't know the name of that category.
[0003] To address this issue, this disclosure proposes a novel online new category discovery (OCD) method. Given a support set containing visible categories, the algorithm model learns how to output a consistent and unique category identifier for images belonging to each category, and can generalize this ability to new category samples that have never been seen before. Summary of the Invention
[0004] This addresses the technical problem that existing category classification methods have limitations and cannot meet the actual needs of users.
[0005] To achieve the above-mentioned technical objectives, this disclosure provides a category classification method, including:
[0006] Define the data structure for online category mining;
[0007] Construct a classification model, wherein the classification model includes: an encoder for feature extraction. and a projector for projecting the extracted features into a discriminative representation space.
[0008] The classification model is trained based on the data used for training in the data structure.
[0009] The trained classification model is tested using the data used for testing in the aforementioned data structure.
[0010] Furthermore, the defined online category mining data structure specifically includes:
[0011] All collected data is divided into a support set D for training. S and the query set D used for testing Q ;
[0012]
[0013] Where, x i Represents the sample y in the dataset i This represents the corresponding label, and N and M represent the number of samples in the support set and query set, respectively. It's worth noting that the query set D... Q It includes both seen and unseen categories, i.e., Y S ∈Y Q .
[0014] Furthermore, training the classification model based on the data used for training in the data structure specifically includes:
[0015] The classification model is trained using supervised learning based on the data used for training in the data structure.
[0016] Furthermore, the process of training the classification model using supervised learning specifically includes:
[0017] The classification model is subjected to supervised contrastive learning using the following formula;
[0018] Its meaning is x i Feature representation;
[0019] The optimization process can be represented by the following formula:
[0020]
[0021] Where B is the multi-angle enhancement including x i Batch data, P i Representing B and x i Other corresponding positive samples.
[0022] Furthermore, the method also includes:
[0023] By hashing the sample representations, each category corresponds to a specific hash code.
[0024] Furthermore, the step of hashing the sample representations to assign a specific hash code to each category specifically includes:
[0025] For any sample xi Its hash identifier is obtained using the following formula:
[0026]
[0027] Where L is the length of the hash value, and hash(·) is the hash function for any vector a = [a1, ..., a2]. l ,…,a L The hash value of ];
[0028]
[0029] further,
[0030] The classification model is trained based on a two-branch structure, wherein the projector in the baseline model... It was replaced in parallel with a symbol projector. and amplitude projector
[0031] For symbolic projectors, a symbolic activation function is used to remove symbolic information:
[0032]
[0033] For amplitude projectors, the absolute value of the output representation is used to remove the sign information:
[0034]
[0035] Final characteristics in, This indicates element-wise multiplication.
[0036] To achieve the above-mentioned technical objectives, this disclosure also provides a category classification device, comprising:
[0037] The data processing module is used to define the data structure for online category mining;
[0038] A model building module is used to construct a classification model, wherein the classification model includes: an encoder for feature extraction. and a projector for projecting the extracted features into a discriminative representation space.
[0039] The model training module is used to train the classification model based on the training data in the data structure.
[0040] The testing module is used to perform classification tests on the trained classification model based on the test data in the data structure.
[0041] To achieve the above-mentioned technical objectives, this disclosure also provides a computer storage medium storing a computer program thereon, which, when executed by a processor, is used to implement the steps of the above-mentioned category classification method.
[0042] To achieve the above-mentioned technical objectives, this disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned category classification method.
[0043] The beneficial effects of this disclosure are as follows:
[0044] This disclosure proposes an innovative visual recognition method called Online Category Mining (OCD). Compared with previous techniques, the proposed method achieves instance-level prediction of unseen categories without relying on auxiliary information, is applicable to open data, and enables instantaneous category mining. The superiority of the proposed method is verified on six commonly used classification datasets. Attached Figure Description
[0045] Figure 1 A flowchart illustrating Embodiment 1 of this disclosure is shown;
[0046] Figure 2 A schematic diagram of Embodiment 1 of this disclosure is shown;
[0047] Figure 3 A schematic diagram of the structure of Embodiment 2 of this disclosure is shown;
[0048] Figure 4 A schematic diagram of the structure of Embodiment 4 of this disclosure is shown. Detailed Implementation
[0049] Embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.
[0050] The accompanying drawings illustrate various structural schematics according to embodiments of the present disclosure. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.
[0051] Currently, both one-stage and two-stage object detection algorithms show a significant performance degradation when detecting long, narrow objects with large aspect ratios, such as skis and pencils, compared to detecting objects with smaller aspect ratios, such as soccer balls and cars. In some cases, the network cannot detect objects with large aspect ratios, resulting in a high false negative rate. In other cases, although the network can detect objects with large aspect ratios, the target location is not accurate enough.
[0052] One reason for this phenomenon is that when the aspect ratio of the target object is large, the intersection and union ratio between the candidate boxes and the target box is low. These candidate boxes are very likely to be set as ignored samples or negative samples, which results in a small number of candidate boxes for targets with large aspect ratios. This causes the model to be unable to learn the features of objects with large aspect ratios well, and thus cannot correctly classify the target or accurately locate its position.
[0053] Existing technology 1: Open set recognition
[0054] In real-world applications, open-set recognition (OSR) has been around for nearly a decade. OSR assumes that new classes only appear in the test set, and its goal is to identify and reject them during inference. Related research mainly follows two trends: generative models and discriminative models. For generative model-based OSR methods, generative models are typically used to synthesize samples of unseen classes, allowing the open-set decision boundary to be learned in a supervised manner. As for discriminative model-based methods, they directly model the open-set boundary using methods such as SVM, sparse representation, and distance measurements to adjust the decision space.
[0055] Existing technology 2: Zero-shot learning
[0056] Another similar approach to OCD is zero-shot learning (ZSL), which focuses directly on recognizing unseen categories. However, it requires additional auxiliary information such as attributes and word vectors to identify categories that have never been seen before.
[0057] Existing technology 3: New category mining
[0058] Traditional novel category discovery (NCD) tasks consist of a support set containing labeled known categories and an unlabeled query set containing unlabeled unknown categories. The NCD algorithm aims to identify unseen categories by segmenting the query set into groups with the same latent category. Existing NCD algorithms are based on inductive learning and offline inference. Specifically, the model is first trained on the support set using supervised learning, and then trained on the query set using unsupervised or semi-supervised learning. Afterward, the model performs unsupervised clustering on the features extracted from the query set to discover new categories.
[0059] Regarding the existing technology 1, the current OSR model can only distinguish between seen categories and unseen categories, but cannot distinguish between unseen categories, while the Online Category Mining (OCD) method can classify unseen categories.
[0060] For existing technology 2, the ZSL method relies on auxiliary information (such as word vectors, attributes, etc.) shared between seen and unseen categories for classification, while OCD can implicitly transfer the concept of category from seen category to unseen category, while additionally relying on shared auxiliary information.
[0061] For the prior art 3, the NCD model relies on (1) inductive learning, that is, the model can use data of unseen categories during the training process, which makes the model unable to handle open sets of unseen categories. (2) offline inference, that is, the model needs to cluster a set of unseen category data, which makes the model unable to perform on-the-fly inference on a single sample. In this work, the present disclosure uses inductive learning and on-the-fly inference methods.
[0062] To solve the above technical problems:
[0063] like Figure 1 As shown:
[0064] This disclosure provides a category classification method, including:
[0065] S101: Define the data structure for online category mining;
[0066] S102: Construct a classification model, wherein the classification model includes: an encoder for feature extraction. and a projector for projecting the extracted features into a discriminative representation space.
[0067] S103: Train the classification model based on the data used for training in the data structure;
[0068] S104: Perform classification tests on the trained classification model based on the data used for testing in the data structure.
[0069] Furthermore, the defined online category mining data structure specifically includes:
[0070] All collected data is divided into a support set D for training. S and the query set D used for testing Q ;
[0071]
[0072] Where, x i Represents the sample y in the dataset i This represents the corresponding label, and N and M represent the number of samples in the support set and query set, respectively. It's worth noting that the query set D... Q It includes both seen and unseen categories, i.e., Y S ∈Y Q .
[0073] Furthermore, training the classification model based on the data used for training in the data structure specifically includes:
[0074] The classification model is trained using supervised learning based on the data used for training in the data structure.
[0075] Furthermore, the process of training the classification model using supervised learning specifically includes:
[0076] The classification model is subjected to supervised contrastive learning using the following formula;
[0077] Its meaning is x i Feature representation;
[0078] The optimization process can be represented by the following formula:
[0079]
[0080] Where B is the multi-angle enhancement including x i Batch data, P i Representing B and x i Other corresponding positive samples.
[0081] Furthermore, the method also includes:
[0082] By hashing the sample representations, each category corresponds to a specific hash code.
[0083] Furthermore, the step of hashing the sample representations to assign a specific hash code to each category specifically includes:
[0084] For any sample x iIts hash identifier is obtained using the following formula:
[0085]
[0086] Where L is the length of the hash value, and hash(·) is the hash function for any vector a = [a1, ..., a2]. l ,…,a L The hash value of ];
[0087]
[0088] further,
[0089] The classification model is trained based on a two-branch structure, wherein the projector in the baseline model... It was replaced in parallel with a symbol projector. and amplitude projector
[0090] For symbolic projectors, a symbolic activation function is used to remove symbolic information:
[0091]
[0092] For amplitude projectors, the absolute value of the output representation is used to remove the sign information:
[0093]
[0094] Final characteristics in, This indicates element-wise multiplication.
[0095] During training, we used To calculate supervised contrast loss And only the output of the symbolic projector is used for reasoning, i.e., the hash descriptor is In actual training, we use the smoothed sign and absolute value functions sign(·) and abs(·) to help the gradient propagate better, that is...
[0096]
[0097] and
[0098]
[0099] Where a is an arbitrary vector, and τ is a hyperparameter that controls the smoothness of the two functions. The symbols and ⊙ represent element-wise division and element-wise multiplication of matrices, respectively. It is worth noting that if... and Shared weights It can degenerate into f i.
[0100] In addition, we used the sign smoothing function sign. * (·), we propose an amplitude regularization term. Messenger Approaching 1 or -1 reduces the symbolic projector's sensitivity to inter-class differences.
[0101] The overall optimization objective can be expressed as
[0102]
[0103] α is a hyperparameter used to balance the influence of the two loss functions.
[0104] This disclosure defines the data structure for the OCD task as follows:
[0105] The total data D includes a support set D used for training. S And a query set for testing. Where x i Represents the sample y in the dataset i This represents the corresponding label, and N and M represent the number of samples in the support set and query set, respectively. It's worth noting that the query set D... Q It includes both seen and unseen categories, i.e., Y S ∈Y Q .
[0106] Based on the hash-coded baseline model.
[0107] The proposed OCD model consists of two parts: an encoder for feature extraction. A projector for projecting the extracted features into a discriminative representation space. Labels can be obtained by clustering samples from the query set. This decision-making process allows the model to obtain results beyond the known categories, so this disclosure designs its baseline method according to this type of approach.
[0108] During training, this disclosure uses only the labeled support set D. S Come to learn and This disclosure uses supervised comparative learning. It is x i The characteristic representation of , the optimization process can be expressed by the following formula.
[0109]
[0110] Where B is the multi-angle enhancement including x i Batch data, P i Representing B and x i Other corresponding positive samples.
[0111] To achieve real-time feedback, this invention normalizes the feature space into a structured hash space, that is, by hashing the sample representation, each category corresponds to a specific hash code. For any sample x i The hash identifier of this disclosure can be obtained in the following ways.
[0112]
[0113] Where L is the length of the hash value, and hash(·) is the hash function for any vector a = [a1, ..., a2]. l ,…,a L The hash value of ].
[0114]
[0115] During testing, this disclosure treats samples with the same hash code as belonging to the same category. Unlike existing clustering-based solutions, this algorithm allows for on-the-fly reasoning because each x... i The inference does not depend on the distribution of the entire query set data.
[0116] Symbol-amplitude decoupling
[0117] This disclosure proposes a two-branch structure, in which the projector in the baseline model It was replaced in parallel with a symbol projector. and amplitude projector For symbolic projectors, this disclosure uses a symbolic activation function to remove symbolic information, i.e. For amplitude projectors, this disclosure uses the absolute value of the output representation to remove symbol information, i.e. As the final feature, among which This indicates element-wise multiplication.
[0118] During training, this disclosure uses To calculate supervised contrast loss And only the output of the symbolic projector is used for reasoning, i.e., the hash descriptor is In actual training, this disclosure uses smoothed sign and absolute value functions sign(·) and abs(·) to help gradients propagate better, i.e.
[0119]
[0120] and
[0121]
[0122] Where a is an arbitrary vector, and τ is a hyperparameter that controls the smoothness of the two functions. The symbols and ⊙ represent element-wise division and element-wise multiplication of matrices, respectively. It is worth noting that if... and Shared weights It can degenerate into f i .
[0123] In addition, this disclosure employs the sign smoothing function sign. * (·), this disclosure proposes an amplitude regularization term. Messenger Approaching 1 or -1 reduces the symbolic projector's sensitivity to inter-class differences.
[0124] The overall optimization objective can be expressed as
[0125]
[0126] α is a hyperparameter used to balance the influence of the two loss functions.
[0127] Example 2:
[0128] like Figure 2 As shown:
[0129] This disclosure also provides a category classification device, including:
[0130] Data processing module 201 is used to define the data structure for online category mining;
[0131] Model building module 202 is used to construct a classification model, wherein the classification model includes: an encoder for feature extraction. and a projector for projecting the extracted features into a discriminative representation space.
[0132] The model training module 203 is used to train the classification model based on the training data in the data structure.
[0133] The testing module 204 is used to perform classification tests on the trained classification model based on the test data in the data structure.
[0134] The data processing module 201 is connected in sequence to the model building module 202, the model training module 203 and the testing module 204.
[0135] Example 3:
[0136] This disclosure also provides a computer storage medium storing a computer program thereon, which, when executed by a processor, is used to implement the steps of the above-described category classification method.
[0137] The storage medium may be non-volatile or non-volatile. The storage medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).
[0138] Furthermore, the storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, applications required for at least one function, etc.; and the data storage area may store data created based on the use of blockchain nodes, etc.
[0139] Example 4:
[0140] This disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described category classification method.
[0141] Figure 4 This is a schematic diagram of the internal structure of an electronic device in one embodiment. For example... Figure 4 As shown, the electronic device includes a processor, storage medium, memory, and network interface connected via a system bus. The storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When executed by the processor, the computer-readable instructions enable the processor to implement a category classification method. The processor provides computing and control capabilities, supporting the operation of the entire computer device. The memory stores computer-readable instructions, which, when executed by the processor, enable the processor to implement a category classification method. The network interface of the computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0142] This electronic device includes, but is not limited to, smartphones, computers, tablets, wearable smart devices, artificial intelligence devices, and power banks.
[0143] In some embodiments, the processor may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory (e.g., executing remote data read / write programs) and calls data stored in the memory to perform various functions of the electronic device and process data.
[0144] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory and at least one processor, etc.
[0145] Figure 4 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 4 The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.
[0146] For example, although not shown, the electronic device may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor via a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be elaborated further here.
[0147] Furthermore, the electronic device may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device and other electronic devices.
[0148] Optionally, the electronic device may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.
[0149] Furthermore, the computer's usable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, applications required for at least one function, etc.; and the data storage area may store data created based on the use of blockchain nodes, etc.
[0150] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.
[0151] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0152] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.
[0153] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.
Claims
1. A category classification method, characterized in that, Used for image classification, including: Define the data structure for online category mining; Construct a classification model, wherein the classification model includes: an encoder for feature extraction. and a projector for projecting the extracted features into a discriminative representation space. The classification model is trained based on the data used for training in the data structure. The trained classification model is tested based on the data used for testing in the data structure described above. The defined online category mining data structure specifically includes: All collected data is divided into a support set D for training. S and the query set D used for testing Q ; Where, x i y represents a sample in the dataset. i The labels represent the corresponding tags, and N and M represent the number of samples in the support set and query set, respectively. It's worth noting that the query set D... Q It includes both seen and unseen categories, i.e., Y S ∈Y Q ; The classification model is trained based on a two-branch structure, wherein the projector in the baseline model... It was replaced in parallel with a symbol projector. and amplitude projector For symbolic projectors, a symbolic activation function is used to remove symbolic information: For amplitude projectors, the absolute value of the output representation is used to remove the sign information: Final characteristics in, This indicates element-wise multiplication.
2. The method according to claim 1, characterized in that, The specific steps of training the classification model based on the data used for training in the data structure include: The classification model is trained using supervised learning based on the data used for training in the data structure.
3. The method according to claim 2, characterized in that, The process of training the classification model using supervised learning specifically includes: The classification model is subjected to supervised contrastive learning using the following formula; Its meaning is x i Feature representation; The optimization process can be represented by the following formula: Where B is the multi-angle enhancement including x i Batch data, P i Representing B and x i Other corresponding positive samples.
4. The method according to claim 3, characterized in that, The method further includes: By hashing the sample representations, each category corresponds to a specific hash code.
5. The method according to claim 4, characterized in that, The process of hashing the sample representations to assign a specific hash code to each category specifically includes: For any sample x i Its hash identifier is obtained using the following formula: Where L is the length of the hash value, and hash(·) is the hash function for any vector a = [a1, ..., a2]. l ,…,a L The hash value of ]; 6. A category classification device, characterized in that, Used for image classification, including: The data processing module is used to define the data structure for online category mining; A model building module is used to construct a classification model, wherein the classification model includes: an encoder for feature extraction. and a projector for projecting the extracted features into a discriminative representation space. The model training module is used to train the classification model based on the training data in the data structure. The testing module is used to perform classification tests on the trained classification model based on the test data in the data structure. The defined online category mining data structure specifically includes: All collected data is divided into a support set D for training. S and the query set D used for testing Q ; Where, x i y represents a sample in the dataset. i The labels represent the corresponding tags, and N and M represent the number of samples in the support set and query set, respectively. It's worth noting that the query set D... Q It includes both seen and unseen categories, i.e., Y S ∈Y Q ; The classification model is trained based on a two-branch structure, wherein the projector in the baseline model... It was replaced in parallel with a symbol projector. and amplitude projector For symbolic projectors, a symbolic activation function is used to remove symbolic information: For amplitude projectors, the absolute value of the output representation is used to remove the sign information: Final characteristics in, This indicates element-wise multiplication.
7. An electronic device, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.
8. A computer storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it is used to implement the steps corresponding to the method described in any one of claims 1 to 5.