Target detection method, target detection model training method and related device

By extracting image features, preliminary segmentation detection and attention processing on the target image, the hidden attention characteristics are generated, and the problem of low accuracy of hidden object detection is solved, and higher detection accuracy and adaptability are achieved.

CN118552719BActive Publication Date: 2024-11-08ZHEJIANG DAHUA TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411032138.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-08
Estimated Expiration
2044-07-30

AI Technical Summary

Technical Problem

The existing object detection methods have low accuracy in the detection of hidden objects in images, making it difficult to effectively identify and segment hidden objects.

Method used

By acquiring the image features of the target image, the first segmentation detection is obtained, the preliminary segmentation characteristics and the degree of concealment evaluation value are obtained, attention processing is performed to obtain the degree of concealment attention characteristics, and the second segmentation detection is performed, and the final segmentation diagram is performed in combination with the degree of concealment evaluation value. Generation of .

Benefits of technology

It improves the detection accuracy of hidden targets and is more adaptable, and can improve the detection accuracy of hidden targets in various scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118552719B_ABST
    Figure CN118552719B_ABST
Patent Text Reader

Abstract

The present application discloses a target detection method, a training method for a target detection model, and related devices. The target detection method comprises: obtaining image features of a target image; performing a first segmentation detection using the image features to obtain preliminary segmentation features and a concealment evaluation value of the target; performing attention processing on the preliminary segmentation features and the concealment evaluation value of the target to obtain a concealment attention feature; performing a second segmentation detection based on the concealment attention feature to obtain a final segmentation map of the target. The above scheme can improve the accuracy of detecting concealed targets.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of target detection technology, and in particular to a target detection method, a target detection model training method and related devices. Background Art

[0002] With the development of computer intelligence technology, the requirements for target detection tasks in the field of image processing are gradually increasing.

[0003] Object detection can identify and detect objects in an image and obtain information such as the location of the object in the image. At present, the object detection method mainly detects objects by performing image processing on the collected images. However, when there are hidden objects in the image, the accuracy of object detection is low. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a target detection method, a target detection model training method and related devices, which can improve the accuracy of hidden target detection.

[0005] In order to solve the above problems, the first aspect of the present application provides a target detection method, which includes: obtaining image features of the target image; performing a first segmentation detection using the image features to obtain preliminary segmentation features and a concealment level assessment value of the target; performing attention processing on the preliminary segmentation features and the concealment level assessment value of the target to obtain a concealment level attention feature; performing a second segmentation detection based on the concealment level attention feature to obtain a final segmentation map of the target.

[0006] In order to solve the above problems, the second aspect of the present application provides a method for training a target detection model, which method includes: obtaining a sample image; using the sample image as the target image, and adopting a target detection model to implement the above target detection method to obtain a sample detection result; using the sample detection result to determine a loss value; training the target detection model based on the loss value to obtain a trained target detection model.

[0007] In order to solve the above problem, the third aspect of the present application provides a computer device, which includes a memory and a processor coupled to each other, the memory stores program data, and the processor is used to execute the program data to implement any step of any of the above methods.

[0008] In order to solve the above problems, the fourth aspect of the present application provides a computer-readable storage medium, which stores program data that can be executed by a processor, and the program data is used to implement any step of any of the above methods.

[0009] The above scheme obtains the image features of the target image, uses the image features to perform the first segmentation detection, obtains the preliminary segmentation features and concealment evaluation values ​​of the target, performs attention processing on the preliminary segmentation features and concealment evaluation values ​​of the target, obtains the concealment attention features, performs the second segmentation detection based on the concealment attention features, obtains the final segmentation map of the target, and performs target detection in combination with the concealment evaluation values. It can perform target detection on targets with different concealment levels. In addition, different concealment attention features can be obtained for targets with different concealment levels to obtain different attention distributions, and then perform segmentation detection to obtain a more accurate final segmentation map, which can improve the accuracy of hidden target detection and can improve the accuracy of hidden target detection in a variety of scenarios, with greater adaptability.

[0010] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in this application, the following briefly introduces the drawings required in the description of the embodiments. Obviously, the drawings described below are only some embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0012] Figure 1 It is a flowchart of an embodiment of the target detection method of the present application;

[0013] Figure 2 This application Figure 1 A schematic diagram of a flow chart of an embodiment of step S13;

[0014] Figure 3 This is a schematic diagram of the structure of an embodiment of a hidden degree attention network of the present application;

[0015] Figure 4 This application Figure 1 A schematic diagram of a flow chart of an embodiment of step S14;

[0016] Figure 5 is a schematic diagram of the structure of an embodiment of a second segmentation detection network of the present application;

[0017] Figure 6 This application Figure 5 A schematic diagram of the structure of an embodiment of a multi-level fusion layer;

[0018] Figure 7 It is a flowchart of an embodiment of a training method for a target detection model of the present application;

[0019] Figure 8 is a schematic diagram of an example of an embodiment of a sample image of the present application;

[0020] Fig. 9 This is the true value segmentation of this application Figure 1 Schematic diagram of an example of an embodiment;

[0021] Fig.10 It is a structural diagram of an embodiment of the target detection model of the present application;

[0022] Fig.11 This is the preliminary segmentation diagram of the sample in this application and the final segmentation of the sample Figure 1 Schematic diagram of an example of an embodiment;

[0023] Fig.12 This application Figure 7 A schematic diagram of a flow chart of an embodiment of step S23;

[0024] Fig.13 It is a structural schematic diagram of an embodiment of the target detection device of the present application;

[0025] Fig.14 It is a structural schematic diagram of an embodiment of a training device for a target detection model of the present application;

[0026] Fig.15 It is a structural schematic diagram of an embodiment of a computer device of the present application;

[0027] Fig.16 It is a structural diagram of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] The terms "first" and "second" in this application are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0030] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0031] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the objects associated before and after are in an "or" relationship. In addition, "many" in this article means two or more than two. In addition, the term "at least one" in this article means any combination of at least two of any one or more of a plurality of, for example, including at least one of A, B, and C, can mean including any one or more elements selected from the set consisting of A, B, and C.

[0032] The present application provides the following embodiments, and each embodiment is described in detail below.

[0033] It is understandable that the target detection method and the target detection model training method in this application can be executed by a computer device, and the computer device can be any device with processing capabilities, such as a mobile phone, a computer, a server, etc., and this application does not limit this. In some possible implementations, the above method can be implemented by a processor calling program data stored in a memory.

[0034] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of the target detection method of the present application. The method may include the following steps:

[0035] S11: Obtain image features of the target image.

[0036] The target image may be an image that needs to be detected to detect the target in the target image. The target image may be an image acquired by using a camera to capture the target. The target may be a target object, such as an active or stationary entity. This application does not limit the target image, target, etc.

[0037] In some embodiments, the target may be a hidden target or a camouflaged target. For example, the target is visually similar to the background. For example, the target actively camouflages itself, the target entity has a protective color, the target and the surrounding environment have similar colors, etc. This application does not limit the target.

[0038] A feature extraction network may be used to obtain image features of the target image. For example, feature extraction networks include VGGNet, ResNet, InceptionNet, MobileNet, DenseNet, FPN (Feature Pyramid Networks), etc. This application does not limit the method of extracting image features.

[0039] In some embodiments, the image feature includes hierarchical features of multiple levels, and the scales of hierarchical features of different levels are different. Among them, different scales can represent different resolutions. That is, the image feature can include hierarchical features of multiple different scales or different resolutions. Hierarchical features of different scales or different resolutions also have different expressive capabilities for features.

[0040] Exemplarily, an encoder may be used to extract hierarchical features of multiple levels of a target image, where hierarchical features of different levels have different resolutions.

[0041] Exemplarily, the hierarchical features of multiple levels of the target image can be extracted using a feature pyramid network, and the hierarchical features of different levels have different resolutions. The resolution of the hierarchical features increases layer by layer, which can also be called a feature pyramid. Among the hierarchical features of multiple levels, the features with lower resolution represent the features of the lower level, such as color, geometry, etc., and the features with higher resolution represent the features of the higher level, such as shape, structure, and other more abstract and semantic features. In the feature pyramid network, as the depth of the network increases, the size of the hierarchical features becomes smaller and smaller, and the semantic information becomes more and more abstract. The hierarchical features of each level can represent the features from shallow features to deep features respectively. Among them, the shallow features can represent the features extracted by the shallow network of the feature extraction network, which are closer to the input layer, contain more pixel information, and fine-grained information (such as color, texture, edge or angular information, etc.). The shallow feature receptive field is smaller and can capture more details. The deep features can represent the features extracted by the deep network of the feature extraction network, which are closer to the output layer, and can contain some coarse-grained information and more abstract semantic information. The deep features have a larger receptive field and can obtain some information about the overall image, with stronger semantic information.

[0042] S12: Perform a first segmentation detection using image features to obtain preliminary segmentation features and a concealment level assessment value of the target.

[0043] The image features are input into the first segmentation detection network, and the first segmentation detection is performed on the image features to obtain the preliminary segmentation features and the concealment evaluation value of the target. In this process, the features of each level can be fused and the preliminary features of the target can be extracted. The preliminary features can include the preliminary segmentation map and the concealment evaluation value of the target.

[0044] The concealment level evaluation value of the target can indicate the concealment level of the target. The concealment level evaluation value can be divided into multiple levels, and different levels of concealment level evaluation values ​​indicate different concealment levels of the target. The larger the concealment level evaluation value, the more concealed the target is. Exemplarily, each level of the concealment level evaluation value can be expressed in the form of a numerical value, for example, it can be divided into five levels {1, 2, 3, 4, 5}, and the smaller the target concealment level evaluation value, the smaller the concealment level of the target.

[0045] The preliminary segmentation map and the concealment evaluation value are obtained based on the preliminary segmentation features. The preliminary segmentation map can be obtained by detection in the form of target classification. The detection of the preliminary segmentation map and the concealment evaluation value can be combined to process the image features using the first segmentation detection network to obtain the preliminary segmentation map and the concealment evaluation value.

[0046] In some embodiments, the first segmentation detection network includes a feature extraction layer, a segmentation head for segmentation detection, and a classification head for target classification, wherein the segmentation head and the classification map may be located after the feature extraction layer, respectively.

[0047] The feature extraction layer processes the image features, for example, by fusing the features of each level to obtain preliminary segmentation features. Then, the preliminary segmentation features are processed by the segmentation head to obtain a preliminary segmentation map. The preliminary segmentation features are processed by the classification head to obtain a concealment level assessment value. In this way, the preliminary segmentation features and concealment level assessment values ​​of the target can be obtained. In some application scenarios, the preliminary segmentation features may be features output by the penultimate layer in the first segmentation detection network, that is, the preliminary segmentation features may be features of the previous layer of the network that obtains the preliminary segmentation map and / or concealment level assessment value. This application does not impose any restrictions on this.

[0048] The Segmentation Head refers to the part of the network model used to perform image segmentation tasks. It is usually located at the end of the network model, immediately after the feature extraction layer, and is responsible for converting the preliminary segmentation features of the feature extraction layer into pixel-level classification predictions.

[0049] Specifically, the segmentation head receives the preliminary segmentation features from the feature extraction layer in front of the network as input, and then outputs the category prediction of each pixel through a series of convolutional layers, upsampling layers (or transposed convolutional layers), activation functions, etc. The segmentation head converts the preliminary segmentation features into a classification map (i.e., the category prediction of each pixel) with the same spatial dimensions as the input image, that is, the preliminary segmentation map is obtained.

[0050] The classification head is the part of the network model that maps the learned feature representation to the category prediction. It is usually located at the end of the network model, immediately after the feature extraction layer, and maps the feature space of the preliminary segmentation features to the predetermined category space for the final classification task.

[0051] Specifically, the preliminary segmentation features output by the feature extraction layer are input into the classification head. The classification head contains a small number of layers, such as a fully connected layer (Linear Layer), an activation function (such as SoftMax), etc., which can map the preliminary segmentation features to the category space and determine their generation probability distribution, that is, the concealment degree evaluation value can be obtained.

[0052] For example, taking the first segmentation detection network including the segmentation algorithm Mask R-CNN as an example, the segmentation head of the segmentation algorithm Mask R-CNN can be used to obtain a preliminary segmentation map of the target, and the output of the classification head is used as the concealment evaluation value of the target.

[0053] S13: Perform attention processing on the preliminary segmentation features and the concealment level evaluation value of the target to obtain concealment level attention features.

[0054] By performing attention processing on the preliminary segmentation features and the concealment evaluation value of the target, the interactive information of the preliminary segmentation features and the concealment evaluation value can be obtained to obtain the concealment attention features.

[0055] In some embodiments, a hidden degree attention network may be used to perform attention processing on the preliminary segmentation features and hidden degree evaluation values ​​of the target to obtain hidden degree attention features.

[0056] In some embodiments, see Figure 2 , step S13 of the above embodiment can be further expanded. Attention processing is performed on the preliminary segmentation features and the concealment evaluation value of the target to obtain the concealment attention features. This embodiment may include the following steps:

[0057] S131: Perform convolution processing on the preliminary segmentation features to obtain segmentation convolution features.

[0058] The dimension of the initial segmentation feature is , the initial segmentation features are convolved to obtain segmentation convolution features. Indicates the number of channels, Indicates characteristics, Indicates height, Indicates width.

[0059] See also Figure 3 , the hidden degree attention network includes the first convolution layer, the embedding layer and the multi-head attention layer. Among them, the input of the multi-head attention layer includes the output of the first convolution layer and the output of the embedding layer. The input of the first convolution layer is the preliminary segmentation feature. The input of the embedding layer is the hidden degree evaluation value.

[0060] In some implementations, the first convolution layer may be used to perform convolution processing on the preliminary segmentation features to obtain segmentation convolution features. For example, the first convolution layer is a two-dimensional convolution, the convolution kernel of the first convolution layer is 1, and the number of output channels of the first convolution layer is (embedding dimension, ), where the number of output channels is related to the embedding dimension of the embedding layer, The embedding dimension represents the convolution output dimension, that is, the number of channels. The initial segmentation feature is convolved through the first two-dimensional convolution layer, and the dimension of the segmentation convolution feature obtained is .

[0061] In some implementations, the segmentation convolution feature may be transformed in dimension to obtain a segmentation transformation feature. For example, the dimension of the segmentation convolution feature may be transformed into a segmentation transformation feature by performing a dimensional flattening and transposing process. , and obtain the segmentation transformation features.

[0062] In some embodiments, the segmentation transformation features can be used as the query tensor (Q, Query) input to the multi-head attention layer.

[0063] S132: Embedding the concealment degree evaluation value to obtain a concealed embedding feature.

[0064] The dimension of the concealment evaluation value is , the embedding layer is used to embed the hidden degree evaluation value, and the hidden embedding feature can be obtained, and the dimension is .

[0065] In some implementations, the latent embedding feature is dimensionally transformed to obtain a latent degree transformation feature; wherein the dimension of the latent degree transformation feature is the same as the dimension of the segmentation transformation feature.

[0066] The dimensional transformation of the latent embedded features includes: dimensionality increase processing, expansion and transposition processing, etc. For example, the dimensionality increase processing is expansion processing, etc.

[0067] Specifically, the hidden embedded features are processed by dimensionality increase and converted into a high-dimensional tensor with the dimension of Then, the high-dimensional tensor is flattened and transposed to convert its dimensions to , that is, the concealment degree transformation feature is obtained.

[0068] Among them, the dimension of the concealment degree transformation feature is the same as the dimension of the segmentation transformation feature.

[0069] In some implementations, the hidden degree transformation features may be used as the key tensor (Key) and value tensor (Value) of the input multi-head attention layer.

[0070] S133: Use the segmented convolutional features and the hidden embedded features to perform attention processing to obtain the hidden degree attention features.

[0071] By using a multi-head attention layer to pay attention to the segmentation convolutional features and the latent embedding features, the latent degree attention features are obtained.

[0072] In some implementations, attention processing may be performed on the segmented convolution features and latent embedded features after dimension conversion to obtain latent degree attention features.

[0073] For example, a multi-head attention layer can be used to process the segmentation transformation features and the hidden degree transformation features to obtain the hidden degree attention features. Specifically, the hidden degree transformation features are used as the key tensor (Key) and the value tensor (Value), and the segmentation transformation features are used as the query tensor (Q, Query), and input into the multi-head attention layer. Thus, the hidden degree attention feature is obtained about the target's hidden representation, and the hidden degree attention feature is obtained. Among them, the dimension of the hidden degree attention feature is .

[0074] The above scheme performs attention processing based on the preliminary segmentation features and the concealment degree evaluation value, obtains the attention distribution of targets with different degrees of concealment, and improves the accuracy of subsequent target detection.

[0075] Please continue reading Figure 1 After the above step S13, the following steps may be performed:

[0076] S14: Perform a second segmentation detection based on the hidden degree attention feature to obtain the final segmentation map of the target.

[0077] The second segmentation detection can be performed based on the hidden degree attention feature, so that the hidden degree attention feature is used to guide the second segmentation detection process to obtain the final segmentation map of the target.

[0078] In some implementations, hidden attention features and image features may be used to perform a second segmentation detection to obtain a final segmentation map of the target.

[0079] In some implementations, attention processing may be performed on hidden attention features and image features, and then segmentation detection may be performed on the results obtained after the attention processing to obtain a final segmentation map of the target.

[0080] In some embodiments, attention processing can be performed on the hidden attention features and the image features, and then the results obtained after the attention processing are fused with the image features, and then segmentation detection is performed to obtain a final segmentation map of the target.

[0081] It is understandable that the present application can select the corresponding segmentation detection method according to the specific application scenario to perform the second segmentation detection based on the concealment degree attention feature to obtain the final segmentation map of the target, and the present application does not impose any restrictions on this.

[0082] In some embodiments, a second segmentation detection network may be used to perform a second segmentation detection based on the hidden degree attention feature to obtain a final segmentation map of the target. For example, the second segmentation detection network is used to achieve a refined segmentation of the hidden target using hierarchical features of multiple levels of image features under the guidance of the hidden degree attention feature to obtain a more accurate final segmentation map.

[0083] In some embodiments, see Figure 4 , step S14 of the above embodiment can be further expanded. Based on the concealment degree attention feature, a second segmentation detection is performed to obtain a final segmentation map of the target. This embodiment may include the following steps:

[0084] S141: Use the hidden degree attention feature and the image feature to perform attention processing to obtain the global attention feature.

[0085] Attention processing can be performed on the hidden degree attention features and the image features to obtain the interactive information of the hidden degree attention features and the image features and obtain the global attention features.

[0086] In some embodiments, since the image features include hierarchical features of multiple levels, and the scales of hierarchical features of different levels are different, hierarchical features of a preset level can be selected and the hidden degree attention can be processed to obtain a global attention feature. The hierarchical features of the preset level can be shallow features or low-level hierarchical features.

[0087] For example, shallow features can be used as low-level hierarchical features, and the low level can be a low level in the image features (such as feature pyramids). The low level and the high level are relative, and the low level can be determined according to the hierarchical situation of the image features, and this application does not limit this.

[0088] Exemplarily, the feature pyramid includes hierarchical features of 1 to 4 levels from low to high, and levels 1 to 2 can be used as the low level. Hierarchical features of any level from levels 1 to 2 can be used as hierarchical features of the preset level.

[0089] See also Figure 5 ,The second segmentation detection network includes the second convolutional layer, the third convolutional layer, the global attention layer, the multi-level fusion layer and the segmentation head.

[0090] In some embodiments, the second convolution layer can be used to perform convolution processing on the hidden degree attention feature to obtain the hidden convolution feature. And, the third convolution layer is used to perform convolution processing on the hierarchical features of the preset level to obtain the hierarchical convolution features. And the hidden convolution features and the hierarchical convolution features are dimensionally transformed to be converted into the same dimension. Among them, the hidden degree attention feature represents the deep semantic information, and the hierarchical features of the preset level are the low-level hierarchical features in the feature pyramid, that is, the low-level hierarchical features in the multi-level image features. The low-level hierarchical features that need to be processed with attention are shallow features, which aim to achieve the fusion of shallow features and deep semantic information.

[0091] Then, the global attention layer (Holistic Attention) is used to process the hidden convolutional features and hierarchical convolutional features to obtain global attention features. This global attention layer can be used to realize the interaction between shallow features and deep semantic information.

[0092] S142: Fuse the global attention feature and the image feature to obtain a fused feature.

[0093] By fusing the global attention features and image features (such as feature pyramids), the interactive information of shallow features and deep semantic information can be fused in the image features. The fusion method can be concatenation, dot multiplication, or addition, etc. The fusion method can be determined according to the application scenario, and this application does not limit the fusion method.

[0094] In some implementations, a multi-level fusion layer may be used to fuse the global attention feature and the image feature to obtain a fused feature.

[0095] In some embodiments, since the image features include hierarchical features of multiple levels, the scales of hierarchical features of different levels are different. According to the order of the levels of the image features, the global attention features can be fused with the hierarchical features of each level in sequence to obtain fused features. For example, the global attention features can be fused with the hierarchical features from low to high levels in the feature pyramid in sequence to obtain fused features.

[0096] In some embodiments, see Figure 6 The multi-level fusion layer at least includes a splicing module, a fusion module and an upsampling module. For fusing the global attention feature and the image feature to obtain the fusion feature, the following steps may be included:

[0097] The global attention features and the hierarchical features of the current level are spliced ​​to obtain the spliced ​​features of the current level. For example, a splicing module can be used to splice the global attention features and the hierarchical features of the current level i (recorded as the i-th level features) to obtain the spliced ​​features of the current level i. Among them, when the current level i is the first level, the global attention features and the hierarchical features of the first level (such as the 1st level features) are spliced ​​to obtain the 1st level spliced ​​features. When the current level i is not the first level, the i-th level features and the spliced ​​features of the previous level i-1 are spliced ​​to obtain the spliced ​​features of the current level i. This embodiment is illustrated by taking the current level as the first level as an example. It can be understood that the splicing method of the global attention features and the hierarchical features of the current level in this application is not limited to this.

[0098] A fusion module can be used to perform feature fusion processing on the spliced ​​features to obtain the fused features of the current level i. The fusion module can perform fusion processing through modules such as convolution processing, dilated convolution or residual channel attention block. Exemplarily, taking the fusion module using the residual channel attention block for fusion processing as an example, the spliced ​​features of the current level i are convolved using a convolution layer to obtain a first convolution feature, and then the first convolution feature is processed using an activation layer (such as a Sigmoid function) to obtain an activation feature. The first convolution feature and the activation feature are point-multiplied to obtain a convolution activation feature. Finally, the convolution activation feature and the spliced ​​features of the current level i are summed (or added) to obtain the fused features of the current level i.

[0099] The fused features of the current level i are dimensionally adjusted to obtain dimensionally adjusted features, wherein the dimension of the dimensionally adjusted features is the same as the dimension of the hierarchical features of the next level. Specifically, an upsampling module may be used to dimensionally adjust the fused features of the current level i so that after the adjustment, features having the same dimension as the hierarchical features of the next level i+1 (recorded as i+1th level features) are obtained. For example, the upsampling module may use a deconvolution layer to perform deconvolution processing on the fused features of the current level i to obtain dimensionally adjusted features of the current level i.

[0100] Then, the above steps are repeated until the current level is the last level, and the final fusion feature is obtained. For example, the dimension adjustment feature of the current level i can be used as a new global attention feature, and the above steps are repeated for the next level i+1.

[0101] Specifically, a splicing module can be used to splice the dimension adjustment features of the current level i with the level features of the next level i+1 to obtain a new splicing feature of the current level, so as to repeat the above-mentioned steps of performing feature fusion processing on the splicing features to obtain the fusion features of the current level and subsequent steps until the current level is the last level and the final fusion features are obtained.

[0102] In some embodiments, in the process of splicing the global attention feature and the hierarchical feature of the current level to obtain the splicing feature of the current level, when splicing the hierarchical feature of the lowest level (that is, the first hierarchical feature), the splicing feature can be obtained according to the above embodiment. Alternatively, the consistency feature of the global attention feature can be obtained from the feature pyramid, such as the same dimension or scale. After convolution and activation processing (SoftMax) are performed on the consistency feature, the convolution activation feature is obtained. The bottom-level hierarchical feature is deconvoluted to obtain the bottom-level deconvolution feature. The convolution activation feature is dot-multiplied with the bottom-level deconvolution feature to obtain a consistent dot-multiplication feature. The consistent dot-multiplication feature and the global attention feature are then dot-multiplied to obtain a global dot-multiplication feature. The global dot product features and the bottom-level deconvolution features are added together to obtain the concatenated features of the bottom-level and global attention features. When concatenating with the hierarchical features of the next level, no dimension adjustment is required. Subsequently, the above-mentioned step of performing feature fusion processing on the concatenated features to obtain the fused features of the current level and subsequent steps are repeated according to the above-mentioned embodiment until the current level is the last level and the final fused features are obtained.

[0103] S143: Use the fusion features to perform segmentation conversion to obtain the final segmentation map of the target.

[0104] The fused features are segmented and transformed to obtain the final segmentation map of the target, where the segmentation transformation includes converting the fused features into pixel-level classification predictions.

[0105] In some implementations, the final fusion feature can be input into a segmentation head. The segmentation head is used to segment and transform the fusion feature to obtain a final segmentation map of the target. The specific implementation process of the segmentation head can refer to the description of the segmentation head in the above embodiment, which will not be repeated here.

[0106] The above scheme obtains the image features of the target image, uses the image features to perform the first segmentation detection, obtains the preliminary segmentation features and concealment evaluation values ​​of the target, performs attention processing on the preliminary segmentation features and concealment evaluation values ​​of the target, obtains the concealment attention features, performs the second segmentation detection based on the concealment attention features, obtains the final segmentation map of the target, and performs target detection in combination with the concealment evaluation values. It can perform target detection on targets with different concealment levels. In addition, different concealment attention features can be obtained for targets with different concealment levels to obtain different attention distributions, and then perform segmentation detection to obtain a more accurate final segmentation map, which can improve the accuracy of hidden target detection and can improve the accuracy of hidden target detection in a variety of scenarios, with greater adaptability.

[0107] In addition, by performing attention processing on the preliminary segmentation features and concealment evaluation values ​​of the target, the concealment attention features are obtained, and then the concealment attention features are used to guide the further extraction of image features, which can improve the accuracy of the final segmentation map.

[0108] In some embodiments, the target detection method described above in the present application can be implemented using a target detection model, that is, the target detection model is used to implement the above steps S11 to S14 to obtain the final segmentation map of the target. In some application scenarios, the target detection model can be trained before using it. The training process can be specifically referred to in the following embodiments.

[0109] See also Figure 7 , Figure 7 1 is a flow chart of an embodiment of a training method for a target detection model of the present application. The method may include the following steps:

[0110] S21: Acquire a sample image.

[0111] A sample image is obtained, where the sample image may be an image containing a target. There may be multiple sample images, so that each sample image can be processed separately, and the target detection model is trained using multiple sample images.

[0112] For example, see Figure 8 , the sample image contains a target. For example, the target is a hidden target, and the similarity between the hidden target and the surrounding environment is high.

[0113] In some implementations, the target detection model may be trained based on the true values ​​of the concealment levels of the hidden targets marked in the sample images.

[0114] After obtaining the sample image, a category label can be set for the sample image. The category label can represent the true value of the concealment degree of the target. Therefore, the category label annotated by the annotator on the sample image can be obtained to obtain the true value of the concealment degree.

[0115] In some embodiments, the time it takes for the target in the sample image to be discovered can be used to perform grade classification to obtain the true value of the degree of concealment. For example, the time it takes for the annotator to discover the target in the sample image is obtained to obtain the time it takes for the target to be discovered. Then, the discovery time is graded. For example, it can be divided into multiple levels from short to long according to the discovery time, and each level can be labeled in the form of an integer. Exemplarily, it can be divided into five levels {1, 2, 3, 4, 5}, and the value of the level can be used as the true value of the degree of concealment of the target in the sample image. Among them, the smaller the value of the true value of the degree of concealment, the shorter the time it takes for the target to be discovered by the annotator, which means that the degree of concealment of the target is smaller.

[0116] In some embodiments, the similarity between the target hidden area and the background area in the sample image can be used to perform grade division to obtain the true value of the degree of concealment. For example, the annotator marks the area where the target is located in the sample image to obtain the target hidden area, and then the similarity between the target hidden area and the background area (that is, the area other than the target hidden area) is calculated to obtain the similarity. The true value of the degree of concealment can be obtained by performing grade division from small to large according to the similarity value. The smaller the similarity, the smaller the degree of concealment of the target.

[0117] In some embodiments, the time it takes for the target to be discovered in the sample image and the similarity between the target hidden area and the background area can be comprehensively utilized to perform level classification to determine the true value of the concealment degree. It is understandable that the method for obtaining the true value of the concealment degree can be determined according to the specific application scenario, and this application does not limit the method for obtaining the true value of the concealment degree.

[0118] In the above method, by digitizing the target's concealment evaluation value and obtaining its annotated category label to add to the target detection model for calculation, the target's concealment evaluation value can be used to guide the segmentation of the hidden target, thereby improving the accuracy of the segmentation of the final segmentation map.

[0119] In some implementations, a target hidden region where the target in the sample image is located may be determined, and a true value segmentation map of the target may be obtained based on the target hidden region.

[0120] For example, see Fig. 9 , a segmentation map can be obtained for the target contained in the sample image, for example, segmentation can be performed according to the hidden area of ​​the target to obtain a true value segmentation map of the target. The true value segmentation map can be a binary image. This application does not limit this.

[0121] S22: Using the sample image as the target image, and using the target detection model to implement the above target detection method, to obtain a sample detection result.

[0122] The sample image is input into the target detection model so that the sample image can be used as the target image, the sample image is processed by the target detection model, and the target detection model is used to implement the above target detection method to obtain the sample detection result. The specific implementation of this step can refer to the implementation process of the embodiment of the above target detection method, which will not be repeated here.

[0123] In some embodiments, the sample detection result includes a preliminary sample segmentation map, a sample concealment level, and a final sample segmentation map.

[0124] See also Fig.10The target detection model includes a feature extraction network, a first segmentation detection network, a hidden degree attention network and a second segmentation detection network. The sample detection results can be obtained in the following way.

[0125] The sample image is input into the feature extraction network. The sample image is processed by the feature extraction network to obtain the image features of the sample image.

[0126] The first segmentation detection network is used to perform the first segmentation detection on the image features to obtain the target sample preliminary segmentation map and the sample concealment degree.

[0127] The hidden degree attention network is used to pay attention to the sample preliminary segmentation map and sample hidden degree of the target to obtain the hidden degree attention feature.

[0128] The second segmentation detection network is used to perform the second segmentation detection based on the hidden degree attention feature to obtain the final segmentation map of the target sample.

[0129] It can be understood that the specific methods for obtaining the sample preliminary segmentation map, the sample concealment level and the sample final segmentation map can refer to the implementation process of obtaining the preliminary segmentation map, the concealment level evaluation value and the final segmentation map in the above-mentioned target detection method embodiment, and will not be repeated here.

[0130] For example, see Fig.11 The sample preliminary segmentation map and the sample final segmentation map can segment the target in the sample image, and indicate the region and contour of the target. The sample preliminary segmentation map and the sample final segmentation map can be binary images. This application does not limit this.

[0131] S23: Determine the loss value using the sample test results.

[0132] The loss value can be determined through the sample detection results of the target detection model.

[0133] In some embodiments, the sample detection result includes a sample preliminary segmentation map, a sample concealment degree, and a sample final segmentation map. The loss value can be determined using at least one of the sample preliminary segmentation map, the sample concealment degree, and the sample final segmentation map.

[0134] For example, the final segmentation map of the sample can be used to determine the loss value.

[0135] For example, the sample concealment degree and the sample final segmentation map can be used to respectively determine the corresponding concealment degree loss value and final segmentation loss value, and then the concealment degree loss value and the final segmentation loss value are combined to obtain the loss value.

[0136] For example, the sample preliminary segmentation map, sample concealment degree and sample final segmentation map can be combined to obtain their corresponding preliminary segmentation loss value, concealment degree loss value and final segmentation loss value respectively, and then the preliminary segmentation loss value, concealment degree loss value and final segmentation loss value can be combined to obtain the loss value.

[0137] It is understandable that the present application may determine the loss value according to the specific application scenario, and the present application does not impose any limitation on this.

[0138] The following is an example of determining the loss value by integrating the sample preliminary segmentation map, the sample concealment level and the sample final segmentation map.

[0139] In some embodiments, see Fig.12 , step S23 of the above embodiment can be further expanded. Using the sample detection results to determine the loss value, this embodiment may include the following steps:

[0140] S231: Using the sample preliminary segmentation map and the true value segmentation map, obtain a preliminary segmentation loss value.

[0141] The difference between the sample preliminary segmentation map and the true value segmentation map can be used to obtain the preliminary segmentation loss value. For example, the binary cross entropy loss (BCE Loss) can be used to calculate the loss of the sample preliminary segmentation map and the true value segmentation map to obtain the preliminary segmentation loss value.

[0142] S232: Using the sample concealment degree and the true value of the concealment degree, obtain the concealment degree loss value.

[0143] The difference between the sample concealment degree and the true concealment degree value can be used to obtain the concealment degree loss value. For example, the cross entropy loss (CE Loss) can be used to calculate the loss between the sample concealment degree and the true concealment degree value to obtain the concealment degree loss value.

[0144] S233: Using the sample final segmentation map and the true value segmentation map, obtain the final segmentation loss value.

[0145] The final segmentation loss value can be obtained by using the difference between the final segmentation map of the sample and the true value segmentation map. For example, the structure similarity index measure (SSIM) algorithm can be used to calculate the loss of the final segmentation map of the sample and the true value segmentation map to obtain the final segmentation loss value.

[0146] It is understandable that the method and loss function for determining each loss value in the present application are not limited to this, and other loss functions may also be used to determine the loss value, and the present application does not impose any restrictions on this.

[0147] S234: The loss value is obtained by combining the initial segmentation loss value, the hidden degree loss value and the final segmentation loss value.

[0148] By combining the above preliminary segmentation loss value, hidden degree loss value and final segmentation loss value, the sum of the three can be obtained to get the loss value.

[0149] In some implementations, the loss value may be expressed as follows:

[0150] .

[0151] Among them, L represents the loss value. represents the initial segmentation loss value, represents the hidden degree loss value, Represents the final segmentation loss value. represents the sample preliminary segmentation map, Represents the true value segmentation map. Indicates the degree of sample concealment. Represents the true value of the concealment degree. Represents the final segmentation map of the sample.

[0152] Through the above method, the loss value of the sample preliminary segmentation map, sample concealment degree and sample final segmentation map output by the target detection model can be obtained, so that after the target detection model is trained, the detection of the above preliminary segmentation map, concealment degree evaluation value and final segmentation map is more accurate.

[0153] Continue to refer to the above Figure 7 After the above step S23, the following steps may be performed:

[0154] S24: Train the target detection model based on the loss value to obtain a trained target detection model.

[0155] The target detection model is trained through the loss value, and the network parameters of the target detection model are adjusted to obtain the trained target detection model.

[0156] In some embodiments, the target detection model may be used to output or not output the preliminary segmentation map. During the training process, the preliminary segmentation map is output to calculate the preliminary segmentation loss. If the preliminary segmentation loss is not needed, the preliminary segmentation map may not be output.

[0157] The above scheme obtains a sample image, takes the sample image as the target image, adopts the target detection model to implement the above target detection method, obtains the sample detection result, then uses the sample detection result to determine the loss value, and trains the target detection model based on the loss value to obtain the trained target detection model, which can improve the accuracy of hidden target detection using the trained target detection model.

[0158] This application only provides technology. The data required for specific applications shall be based on actual conditions. However, users are reminded that when actually using this technology, the collection and processing of data should comply with laws and regulations related to data and personal information protection.

[0159] It is understandable that in the above method of the specific implementation mode, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0160] For the above embodiments, the present application also provides a target detection device for implementing the target detection method of any of the above embodiments.

[0161] See also Fig.13 , Fig.13 3 is a schematic diagram of the structure of an embodiment of the target detection device of the present application. The target detection device 30 includes an image acquisition module 31, a first segmentation module 32, an attention module 33 and a second segmentation module 34.

[0162] The image acquisition module 31 is used to acquire image features of a target image.

[0163] The first segmentation module 32 is used to perform a first segmentation detection using image features to obtain preliminary segmentation features and a concealment level evaluation value of the target.

[0164] The attention module 33 is used to perform attention processing on the preliminary segmentation features and the concealment level evaluation value of the target to obtain concealment level attention features.

[0165] The second segmentation module 34 is used to perform a second segmentation detection based on the concealment degree attention feature to obtain a final segmentation map of the target.

[0166] It should be noted that the target detection device provided in the above embodiment and the target detection method provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs the operation has been described in detail in the method embodiment and will not be repeated here. In practical applications, the target detection device provided in the above embodiment can distribute the above functions to different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and this application does not limit this.

[0167] For the above embodiments, the present application also provides a target detection model training device, which is used to implement the training of the target detection model of any of the above embodiments.

[0168] See also Fig.14 , Fig.14 4 is a schematic diagram of a structure of an embodiment of a training device for a target detection model of the present application. The training device 40 for a target detection model includes a sample module 41, a segmentation module 42, a loss module 43 and a training module 44.

[0169] The sample module 41 is used to obtain a sample image.

[0170] The segmentation module 42 is used to use the sample image as the target image and adopt the target detection model to implement the above target detection method to obtain the sample detection result.

[0171] The loss module 43 is used to determine the loss value using the sample detection result.

[0172] The training module 44 is used to train the target detection model based on the loss value to obtain a trained target detection model.

[0173] It should be noted that the training device for the target detection model provided in the above embodiment and the training method for the target detection model provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs the operation has been described in detail in the method embodiment and will not be repeated here. In practical applications, the training device for the target detection model provided in the above embodiment can distribute the above functions to different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and this application does not limit this.

[0174] For the above embodiment, the present application provides a computer device, see Fig.15 , Fig.15 1 is a schematic diagram of the structure of an embodiment of a computer device of the present application. The computer device 50 includes a memory 51 and a processor 52, wherein the memory 51 and the processor 52 are coupled to each other, the memory 51 stores program data, and the processor 52 is used to execute the program data to implement the steps of any embodiment of the above-mentioned target detection method and the target detection model training method.

[0175] In this embodiment, the processor 52 may also be referred to as a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip having the ability to process signals. The processor 52 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or the processor 52 may also be any conventional processor, etc.

[0176] The method of the above embodiment may be implemented in the form of a computer program, and thus the present application proposes a computer-readable storage medium.

[0177] See also Fig.16 , Fig.16 6 is a schematic diagram of a computer-readable storage medium of the present application. The computer-readable storage medium 60 stores program data 61 that can be run by a processor, and the program data 61 can be executed by the processor to implement the steps of any of the above-mentioned target detection methods and target detection model training methods.

[0178] The computer-readable storage medium 60 of this embodiment may be a medium that can store the program data 61, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or may be a server storing the program data 61. The server may send the stored program data 61 to other devices for execution, or may execute the stored program data 61 by itself.

[0179] In some embodiments, the functions or modules included in the device provided in the above embodiments of the present application can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, the present application will not go into details here.

[0180] The above description of the various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other. For the sake of brevity, this application will not go into details here.

[0181] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0182] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0183] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0184] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium, which is a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a computer, server, or network device, etc.) or a processor to execute all or part of the steps of each implementation method of the present application.

[0185] Obviously, a person skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a computer-readable storage medium and executed by a computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0186] The above descriptions are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A target detection method, characterized in that: include: Obtaining image features of a target image; Performing a first segmentation detection using the image features to obtain preliminary segmentation features and a concealment level assessment value of the target; wherein the concealment level assessment value is divided into multiple levels, and each level of the concealment level assessment value is expressed in the form of a numerical value; Performing attention processing on the preliminary segmentation features and the concealment level evaluation value of the target to obtain concealment level attention features; Performing a second segmentation detection based on the concealment degree attention feature to obtain a final segmentation map of the target; The step of performing attention processing on the preliminary segmentation features and the concealment evaluation value of the target to obtain the concealment attention features includes: Performing convolution processing on the preliminary segmentation features to obtain segmentation convolution features; and Embedding the concealment degree evaluation value to obtain a concealed embedding feature; The segmented convolutional features and the hidden embedded features are used to perform attention processing to obtain the hidden degree attention features.

2. The method according to claim 1, characterized in that Before performing attention processing using the segmented convolutional features and the hidden embedded features to obtain the hidden degree attention features, the method includes: Performing dimension transformation on the segmentation convolution feature to obtain a segmentation transformation feature; Performing a dimension transformation on the hidden embedding feature to obtain a hidden degree transformation feature; wherein the dimension of the hidden degree transformation feature is the same as the dimension of the segmentation transformation feature; The step of performing attention processing using the segmented convolutional features and the hidden embedded features to obtain the hidden degree attention features includes: A multi-head attention layer is used to perform attention processing on the segmentation transformation feature and the concealment degree transformation feature to obtain the concealment degree attention feature.

3. The method according to claim 1, characterized in that: The second segmentation detection is performed based on the concealment degree attention feature to obtain a final segmentation map of the target, including: Performing attention processing using the concealment degree attention feature and the image feature to obtain a global attention feature; Fusing the global attention feature and the image feature to obtain a fused feature; The fusion features are used to perform segmentation conversion to obtain a final segmentation map of the target.

4. The method according to claim 3, characterized in that The image features include hierarchical features of multiple levels, and the scales of hierarchical features of different levels are different; the use of the hidden degree attention features and the image features to perform attention processing to obtain a global attention feature includes: Performing convolution processing on the hidden degree attention feature to obtain a hidden convolution feature; Perform convolution processing on the hierarchical features of the preset level to obtain hierarchical convolution features; The hidden convolutional features and the hierarchical convolutional features are used to perform attention processing to obtain the global attention features.

5. The method according to claim 3, characterized in that: The image features include hierarchical features of multiple levels, and the scales of hierarchical features of different levels are different; The fusing of the global attention feature and the image feature comprises: fusing the global attention feature with the hierarchical features of each level in sequence according to the hierarchical order of the image features; Alternatively, the fusing the global attention feature and the image feature to obtain a fused feature includes: Concatenating the global attention feature and the hierarchical feature of the current level to obtain a concatenated feature of the current level; Performing feature fusion processing on the spliced ​​features to obtain fusion features of the current level; Performing dimension adjustment on the fusion feature of the current level to obtain dimension adjustment features, wherein the dimension of the dimension adjustment features is the same as the dimension of the level features of the next level; The dimension adjustment feature is spliced ​​with the hierarchical feature of the next level to obtain a new spliced ​​feature of the current level, and the step of performing feature fusion processing on the spliced ​​feature to obtain the fused feature of the current level and subsequent steps are repeatedly performed until the current level is the last level and the final fused feature is obtained.

6. A method for training a target detection model, characterized in that: include: Get a sample image; Using the sample image as a target image, and using a target detection model to implement the target detection method according to any one of claims 1 to 5, to obtain a sample detection result; Determine the loss value using the sample test results; The target detection model is trained based on the loss value to obtain a trained target detection model.

7. The method according to claim 6, characterized in that The sample detection results include a sample preliminary segmentation map, a sample concealment level, and a sample final segmentation map; The method of using the sample detection result to determine the loss value includes: Using the sample preliminary segmentation map and the true value segmentation map, a preliminary segmentation loss value is obtained; Using the sample concealment degree and the true concealment degree value, a concealment degree loss value is obtained; Using the final segmentation map of the sample and the true value segmentation map, a final segmentation loss value is obtained; The loss value is obtained by combining the preliminary segmentation loss value, the hidden degree loss value and the final segmentation loss value.

8. The method according to claim 7, characterized in that The method further comprises: The target in the sample image is discovered for a certain time period, and the target is classified into different levels to obtain the true value of the concealment degree; and / or, The similarity between the target hidden area and the background area in the sample image is used to perform level division to obtain the true value of the concealment degree.

9. A computer device, characterized in that: The invention comprises a memory and a processor coupled to each other, wherein the memory stores program data, and the processor is used to execute the program data to implement the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: Program data that can be executed by a processor are stored, and the program data is used to implement the steps of the method according to any one of claims 1 to 8.