Artificial intelligence data annotation method
Through the dynamic neighborhood weighting mechanism and multi-layer convolutional structure optimization graph neural network, combined with label consistency constraints and adaptive convolution kernels, the problems of insufficient adaptive capabilities of graph neural networks and poor cross-scale fusion of convolutional networks in traditional methods are solved, and more efficient and accurate data annotation is achieved.
Patent Information
- Application Number
- CN202510529267.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-25
- Publication Date
- 2025-08-08
AI Technical Summary
Traditional graph neural networks lack adaptability when processing local features, and convolutional neural networks perform poorly when processing multi-scale information, resulting in low data labeling efficiency and unstable quality.
The dynamic neighborhood weighting mechanism is used to optimize the graph neural network, combining multi-layer convolution and pooling structure, through the coordinated optimization of graph neural network and convolutional neural network, local feature adaptability and cross-scale information fusion capabilities are improved, label consistency constraints and adaptive convolution kernels are introduced, and the annotation process is optimized.
It significantly improves the accuracy and efficiency of image data annotation, especially when processing complex or irregular input data, it can capture key features more accurately, solving the shortcomings of cross-scale information fusion in traditional methods.
Smart Images
Figure CN120451705A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence data labeling method. Background Art
[0002] With the rapid development of artificial intelligence (AI), data labeling has become a crucial and important step in machine learning. In particular, accurate data labeling improves model generalization and robustness in complex scenarios, particularly in tasks such as image recognition, time series analysis, and natural language processing. However, traditional manual data labeling methods suffer from low efficiency and inconsistent quality. Consequently, automated data labeling has become a hot topic in AI research in recent years.
[0003] 1. In existing technologies, although traditional graph neural networks (GNNs) can effectively propagate information in graph structures, they often lack sufficient adaptive capabilities when processing local features; 2. In existing technologies, convolutional neural networks (CNNs) are widely used for image data annotation, but they have shortcomings in processing complex multi-scale information. Traditional CNNs mainly extract features through local convolution operations, but their performance in multi-scale or cross-scale feature fusion is not ideal. Summary of the Invention
[0004] In view of the shortcomings of the existing technology, the present invention provides an artificial intelligence data labeling method to solve the problems raised in the above background technology.
[0005] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides an artificial intelligence data labeling method, comprising the following steps: S1. Data preprocessing and feature extraction: remove noise, format data, fill missing values, and perform dimensionality reduction to obtain data features; S2. Define the labeling task using data features: Cluster the data into different categories according to its features, clarify the labeling goals, and further divide the labeling tasks according to the data features to obtain data categories; S3. Perform preliminary label assignment based on data category: assign a preliminary label to each data point based on the clustering results; S4. Manual verification of preliminary labels: assign preliminary labels to labelers for manual verification and correction to obtain corrected data; S5. Perform labeling consistency check on the corrected data: Use a graph neural network algorithm to build a correlation model between data points to check whether the labeling conforms to the predetermined rules, complete the consistency check, and obtain the test results; S6. Label unification based on consistency test results: Based on the consistency test results, the data labels are unified and revised to obtain unified labels; S7. Use unified labels to train and validate data feature prediction models.
[0006] To further optimize this technical solution, in step S5, the graph neural network algorithm updates the features of each node using the following update formula: ; in: : No. The node feature matrix of the layer; : No. The weight matrix of the layer is used to linearly transform node features; : adjacency matrix, which represents the similarity relationship between nodes in the graph; : Non-linear activation function.
[0007] To further optimize this technical solution, in step S5, the graph neural network algorithm optimizes label consistency through the following loss function: ; in: and : No. data points and labels for the data points; and : No. data points and Feature representation of data points; : A hyperparameter used to adjust the effect of label consistency on the loss; : No. data points and The Euclidean distance of the data point features is used to measure the similarity between two nodes in the feature space; : Elements in the adjacency matrix, representing the data points and The similarity between data points.
[0008] To further optimize this technical solution, in step S5, the final optimization goal of the graph neural network algorithm is: ; in, Represents the final optimized label vector.
[0009] To further optimize this technical solution, in step S5, the process of using the formula model in the graph neural network algorithm includes: Training phase; Reasoning stage; Label consistency optimization.
[0010] Further optimizing this technical solution, in step S7, the first The formula for the layer convolution operation is: ; in: Represents the convolution operation; No. The convolution kernel of the layer; is the bias term; is a nonlinear activation function; : Input dataset, where Represents input data, is the number of data points, is the feature dimension of each data point.
[0011] To further optimize this technical solution, the adaptive convolution kernel is introduced into the convolutional neural network in step S7, and the size of each convolution kernel is By an adaptive function Generate, the formula is: ; in, is the input feature in dimension The dimension with larger variance corresponds to a larger convolution kernel in order to capture more complex features.
[0012] To further optimize this technical solution, in step S7, features at different levels are combined through a weighted fusion mechanism, and the fusion process is as follows: ; in: is the weight coefficient, which indicates the importance of features at different levels in the final feature fusion; Represents the number of layers in the graph neural network.
[0013] To further optimize this technical solution, in step S7, the convolutional neural network is optimized by combining the graph neural network in step S5 and propagating labels, and the adjacency matrix is introduced into the output features of each layer. Perform consistency detection and propagate label consistency using the following formula: ; To further optimize this technical solution, in step S7, the convolutional neural network model is used as follows: Training phase; Reasoning stage; Label consistency optimization.
[0014] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of an artificial intelligence data labeling method as described in the first aspect of the present invention are implemented.
[0015] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of an artificial intelligence data labeling method as described in the first aspect of the present invention are implemented.
[0016] Compared with the existing technology, the present invention provides an artificial intelligence data annotation method with the following beneficial effects: This artificial intelligence data labeling method sets up a dynamic neighborhood weighting mechanism and improves the adaptability of graph neural networks to local features by adjusting the influence between nodes. By introducing local neighborhood feature weighting and label consistency constraints, the graph neural network can more accurately capture key features when dealing with data imbalance or complex node relationships; the multi-layer convolution and pooling structure enables the network to simultaneously process and fuse features from different scales at different levels, allowing the convolutional neural network to better identify and process cross-scale information, thereby improving the accuracy and efficiency of image data labeling. In labeling tasks, especially when facing complex image data or irregular input data, this method can significantly improve the model's integration capabilities in processing different feature scales, solving the defects of cross-scale information fusion in traditional convolutional neural network methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0018] Figure 1 This is a structural diagram of an artificial intelligence data annotation method proposed by the present invention; Figure 2 A schematic diagram of the training and verification process of a data feature prediction model for an artificial intelligence data annotation method proposed in the present invention; DETAILED DESCRIPTION
[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0021] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it designate a separate or selective embodiment that is mutually exclusive with other embodiments.
[0022] Example 1: Reference Figures 1 and 2 , which is the first embodiment of the present invention, provides an artificial intelligence data labeling method, comprising the following steps: S1. Data preprocessing and feature extraction: remove noise, format data, fill missing values, and perform dimensionality reduction to obtain data features; S2. Define the labeling task using data features: Cluster the data into different categories according to its features, clarify the labeling goals, and further divide the labeling tasks according to the data features to obtain data categories; S3. Perform preliminary label assignment based on data category: assign a preliminary label to each data point based on the clustering results; This process is achieved through the support vector machine (SVM) algorithm: : Input dataset, is the number of data samples, is the characteristic dimension of each sample; : target label, It is The labels of samples. We assume this is a binary classification problem; : weight vector of the hyperplane; : bias term of the hyperplane; :sample Lagrange multipliers of ; :sample and The kernel function between is used to calculate the nonlinear mapping; : Adjacency matrix, which represents the relationship between samples and is usually used in graph structures; :sample The weighting coefficient of is adjusted according to the relationship between the sample and its neighborhood; The dynamic boundary adjustment and local feature weighting mechanism in the support vector machine is expressed by the following formula: Dynamic Boundary Adjustment: The dynamic adjustment function is , automatically adjusted according to data distribution: ; in, is a hyperparameter that controls the flexibility of the boundary. is the adjustment factor used to control the magnitude of boundary changes; Local feature weighting mechanism: For each sample , weighted according to its relationship with neighboring samples, weight It can be calculated by the following formula: ; in, Represents a sample Neighborhood, is the value in the adjacency matrix, representing the sample and The similarity between is the distance measure between samples; Label propagation and consistency constraints: Label consistency constraints are introduced during the training process. The goal of label consistency propagation is to minimize the following loss function: ; in: The first term is the standard SVM loss function, which is used to maximize the classification margin; The second term is the label consistency loss term, where Represents a sample and The relationship between is the importance parameter that controls the consistency constraint; The purpose of this loss function is to: if two samples are adjacent in the graph, their labels should be as consistent as possible. With this constraint, the support vector machine can better handle the relationship between samples, especially when the sample categories are unevenly distributed, which can effectively improve the classification accuracy. The process of using this model is as follows: Training phase: During training, the support vector machine learns the optimal hyperplane by optimizing the aforementioned loss function. A dynamic boundary adjustment mechanism enables the model to adapt to data complexity, a local feature weighting mechanism helps the model better handle the peculiarities of local data, and label consistency propagation further optimizes label predictions. Inference phase: In the inference phase, given new input data , the support vector machine is trained based on the hyperplane parameters and , combining dynamically adjusted boundaries and local weighted features to perform label prediction: ; Among them, label prediction not only depends on the hyperplane parameters, but also considers the local relations and consistency constraints of data points.
[0023] S4. Manual verification of preliminary labels: assign preliminary labels to labelers for manual verification and correction to obtain corrected data; S5. Perform labeling consistency check on the corrected data: Use a graph neural network algorithm to build a correlation model between data points to check whether the labeling conforms to the predetermined rules, complete the consistency check, and obtain the test results; In step S5, the graph neural network algorithm updates the features of each node using the following update formula: ; in: : No. The node feature matrix of the layer; : No. The weight matrix of the layer is used to linearly transform node features; : adjacency matrix, which represents the similarity relationship between nodes in the graph; : nonlinear activation function; In step S5, the graph neural network algorithm optimizes label consistency through the following loss function: ; in: and : No. data points and labels for the data points; and : No. data points and Feature representation of data points; : A hyperparameter used to adjust the effect of label consistency on the loss; : No. data points and The Euclidean distance of the data point features is used to measure the similarity between two nodes in the feature space; : Elements in the adjacency matrix, representing the data points and The similarity between data points; The meaning of this loss function is: if two nodes and In the figure similar (i.e. values are larger), their labels and Should be as consistent as possible. If their eigenvectors and If the values are also similar, the penalty for label consistency will be reduced; In step S5, the final optimization goal of the graph neural network algorithm is: ; in, Represents the final optimized label vector; The process of using this model is as follows: Training Phase: During the training phase, we first train the graph neural network using a graph with labeled data. By updating node features and minimizing the loss function, the model learns the label consistency between nodes. When the node feature and label consistency constraints are satisfied, the model converges and generates accurate labels.
[0024] Inference phase: During the inference phase, given a new dataset, the graph neural network propagates node features and predicts labels based on the label relationships between adjacent nodes. By leveraging the similarity of nodes in the graph structure, the model can efficiently perform consistency checks and correct potential labeling errors.
[0025] Label consistency optimization: Using graph neural networks to detect label consistency can effectively reduce errors and conflicts in the labeling process. Especially when dealing with complex labeling tasks (such as multidimensional data with complex correlations), graph neural networks can better capture the dependencies between data points, thereby improving the accuracy of labeling.
[0026] S6. Label unification based on consistency test results: Based on the consistency test results, the data labels are unified and revised to obtain unified labels; S7. Use unified labels to train and validate data feature prediction models; In step S7, the convolutional neural network The formula for the layer convolution operation is: ; in: Represents the convolution operation; No. The convolution kernel of the layer; is the bias term; is a nonlinear activation function; : Input dataset, where Represents input data, is the number of data points, is the characteristic dimension of each data point; The adaptive convolution kernel is introduced into the convolutional neural network in step S7, and the size of each convolution kernel is By an adaptive function Generate, the formula is: ; in, is the input feature in dimension The dimension with larger variance corresponds to a larger convolution kernel in order to capture more complex features; In step S7, features at different levels are combined through a weighted fusion mechanism, and the fusion process is as follows: ; in: is the weight coefficient, which indicates the importance of features at different levels in the final feature fusion; Represents the number of layers in the graph neural network; In step S7, the convolutional neural network is optimized by combining the graph neural network in step S5 and label consistency propagation, and the adjacency matrix is introduced into the output features of each layer. Perform consistency detection and propagate label consistency using the following formula: ; The function of this formula is: if two data points and In the graph, the relationship is strong (i.e. The value of is large), then their labels should be as consistent as possible. Through label consistency propagation, convolutional neural networks can automatically adjust labels, thereby improving the accuracy of annotation; The process of using this model is as follows: Training Phase: During the training phase, convolutional neural networks continuously learn the deep features of the input data through adaptive convolution kernels and multi-level feature fusion. With the help of the label consistency propagation mechanism, the network can correct label conflicts and reduce human errors.
[0027] Inference phase: In the inference phase, the network performs convolution operations based on the features of the input data and corrects the output labels through the label consistency propagation mechanism to ensure the accuracy and consistency of the prediction results.
[0028] Label consistency optimization: Combining graph structure and convolution operations, the model optimizes the label of each data point to further improve the accuracy of the labeling task.
[0029] Example 2: This embodiment also provides a computer device suitable for an artificial intelligence data labeling method, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement an artificial intelligence data labeling method proposed in the above embodiment.
[0030] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, an artificial intelligence data labeling method proposed in the above embodiment is implemented.
[0031] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.
[0032] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.
[0033] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0034] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.
[0035] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.
[0036] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. An artificial intelligence data annotation method, characterized in that: The following steps are involved: S1. Data preprocessing and feature extraction: remove noise, format data, fill missing values, and perform dimensionality reduction to obtain data features; S2. Define the labeling task using data features: Cluster the data into different categories according to its features, clarify the labeling goals, and further divide the labeling tasks according to the data features to obtain data categories; S3. Perform preliminary label assignment based on data category: assign a preliminary label to each data point based on the clustering results; S4. Manual verification of preliminary labels: assign preliminary labels to labelers for manual verification and correction to obtain corrected data; S5. Perform labeling consistency check on the corrected data: Use a graph neural network algorithm to build a correlation model between data points to check whether the labeling conforms to the predetermined rules, complete the consistency check, and obtain the test results; S6. Label unification based on consistency test results: Based on the consistency test results, the data labels are unified and revised to obtain unified labels; S7. Use unified labels to train and validate data feature prediction models.
2. The artificial intelligence data annotation method according to claim 1, characterized in that: In step S5, the graph neural network algorithm updates the features of each node using the following update formula: ； in: : No. The node feature matrix of the layer; : No. The weight matrix of the layer is used to linearly transform node features; : adjacency matrix, which represents the similarity relationship between nodes in the graph; : Non-linear activation function.
3. The artificial intelligence data annotation method according to claim 2, characterized in that: In step S5, the graph neural network algorithm optimizes label consistency through the following loss function: ； in: and : No. data points and labels for the data points; and : No. data points and Feature representation of data points; : A hyperparameter used to adjust the effect of label consistency on the loss; : No. data points and The Euclidean distance of the data point features is used to measure the similarity between two nodes in the feature space; : Elements in the adjacency matrix, representing the data points and The similarity between data points.
4. The artificial intelligence data annotation method according to claim 4, characterized in that: In step S5, the final optimization goal of the graph neural network algorithm is: ； in, Represents the final optimized label vector.
5. The artificial intelligence data annotation method according to claim 4, characterized in that: In step S5, the process of using the formula model in the graph neural network algorithm includes: Training phase; Reasoning stage; Label consistency optimization.
6. The artificial intelligence data annotation method according to claim 1, characterized in that: In step S7, the data feature prediction model specifically adopts the convolutional neural network in deep learning. The formula for the layer convolution operation is: ； in: Represents the convolution operation; No. The convolution kernel of the layer; is the bias term; is a nonlinear activation function; : Input dataset, where Represents input data, is the number of data points, is the feature dimension of each data point.
7. The artificial intelligence data annotation method according to claim 6, characterized in that: The adaptive convolution kernel is introduced into the convolutional neural network in step S7, and the size of each convolution kernel is By an adaptive function Generate, the formula is: ； in, is the input feature in dimension The variance on .
8. The artificial intelligence data annotation method according to claim 6, characterized in that: In step S7, features at different levels are combined through a weighted fusion mechanism, and the fusion process is as follows: ； in: is the weight coefficient, which indicates the importance of features at different levels in the final feature fusion; Represents the number of layers in the graph neural network.
9. The artificial intelligence data annotation method according to claim 6, characterized in that: In step S7, the convolutional neural network is optimized by combining the graph neural network in step S5 and label consistency propagation, and the adjacency matrix is introduced into the output features of each layer. Perform consistency detection and propagate label consistency using the following formula: 。 10. The artificial intelligence data annotation method according to claim 6, characterized in that: In step S7, the convolutional neural network model is used as follows: Training phase; Reasoning stage; Label consistency optimization.