Intelligent sensing method, system and device based on graph structure construction and medium
Through the dual-branch graph structure and multiple edge weight strategies, the graph convolution problem is solved by insufficient utilization of differential between nodes of graph convolution network, and the accuracy and robustness of signal classification perception are improved, and it is suitable for equipment status monitoring in intelligent manufacturing.
Patent Information
- Application Number
- CN202510596283.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-09
- Publication Date
- 2025-09-02
AI Technical Summary
When building graph convolutional networks, existing methods cannot fully utilize semantic or structural differences between nodes, resulting in poor data perception.
The dual-branch graph structure is adopted, and the time-frequency image and time-domain graph structure are constructed through continuous wavelet transformation, and the graph convolution is carried out in combination with K-dimensional tree and multiple edge weight strategies, and the edge weight is dynamically adjusted to improve the feature extraction effect.
It improves the feature extraction effect and expression ability of graph convolution, improves the classification perception accuracy of multi-view signals, and is suitable for equipment operation status perception in intelligent manufacturing.
Smart Images
Figure CN120579013A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of deep learning technology, and in particular to an intelligent perception method, system, device and medium based on graph structure construction. Background Art
[0002] In actual manufacturing systems, vibration signals collected by equipment exhibit rich temporal and frequency characteristics. Although a single physical signal is monomodal, it can be transformed into multiple feature perspectives (such as the original time domain, frequency domain, and time-frequency diagrams) through various mathematical transformations. Joint modeling of these features helps to more comprehensively characterize the equipment's operating status. However, these heterogeneous representations differ in dimensionality, structure, and spatial distribution, making it difficult for traditional feature extraction methods to account for the inherent connections between these various features.
[0003] To address these issues, graph neural networks (GNNs) have become a crucial tool for complex data perception in recent years, owing to their strengths in processing non-Euclidean data. However, in practical applications, the quality of graph construction directly determines the effectiveness of GCN (Graph Convolutional Network) feature extraction. Existing methods, which mostly rely on fixed adjacency or simple distance rules, fail to fully exploit semantic or structural differences between nodes, leaving much room for improvement in data perception. Summary of the Invention
[0004] The main purpose of the present invention is to provide an intelligent perception method, system, device and medium based on a graph structure, aiming to solve the technical problem in the prior art that the effect of data perception needs to be improved.
[0005] To achieve the above-mentioned objectives, the first aspect of the present invention provides an intelligent perception method for graph structure construction, comprising: obtaining a one-dimensional original vibration signal; performing a continuous wavelet transform on the original vibration signal to obtain a two-dimensional time-frequency image; and establishing a time domain graph structure for the original vibration signal using a k-dimensional tree; performing feature extraction on the two-dimensional time-frequency image to obtain a first eigenvector, and performing feature extraction on the time domain graph structure to obtain a second eigenvector; fusing the first eigenvector and the second eigenvector to obtain a fused vector; and classifying the fused vector to obtain a classification judgment of the original vibration signal.
[0006] Furthermore, the feature extraction of the two-dimensional time-frequency image to obtain the first feature vector includes: taking the two-dimensional time-frequency image as a grid graph, taking each pixel point in the grid graph as a node in the graph, establishing edge connections between adjacent pixel points, and constructing an undirected graph G = (V, E), wherein V is a set of pixel points, E is a set of connecting edges, and the feature of the node is X i =(Ri , G i , B i ), represents the RGB three-channel pixel value, and the edge weight w between nodes ij =||X i -X j ||; Perform a two-layer ChebNet convolution on the undirected graph to obtain the first eigenvector.
[0007] Furthermore, the k-dimensional tree is used to establish a time domain graph structure for the original vibration signal, including: taking each sampling point of the graph structure as a node in the graph, and establishing edge links between the node and its five nearest neighbors to obtain a time domain graph structure.
[0008] Furthermore, the extracting features from the time-domain graph structure to obtain a second feature vector includes: performing a 2-layer GraphConv convolution on the time-domain graph structure to extract time-domain structural features to obtain a second feature vector.
[0009] Furthermore, the feature extraction of the time domain graph structure to obtain the second feature vector also includes: in the process of performing 2-layer GraphConv convolution on the time domain graph structure to extract the time domain structure features, uneven edge weighting is performed on the convolution result to dynamically adjust the edge weights according to the feature differences and distances between different nodes, wherein the uneven edge weighting includes: at least one of: 0-1 weighting, distance negative exponential weighting, cosine similarity weighting, heat kernel weighting, and Gaussian kernel weighting.
[0010] Furthermore, after performing uneven edge weighting on the convolution result, in the process of performing two-layer GraphConv convolution on the time domain graph structure to extract the time domain structure features, the convolution calculation formula includes:
[0011]
[0012] in, is the adjacency matrix calculated based on the weighted strategy, is the degree matrix corresponding to the adjacency matrix, W (l) is the learnable weight parameter matrix.
[0013] Furthermore, the first feature vector and the second feature vector are both 64-dimensional graph feature vectors; the fused vector is a 128-dimensional feature vector; and when classifying the fused vector, a fully connected layer is used for classification.
[0014] The second aspect of the present invention also provides an intelligent perception system constructed based on a graph structure, including: a signal acquisition module for acquiring a one-dimensional original vibration signal; a signal processing module for performing a continuous wavelet transform on the original vibration signal to obtain a two-dimensional time-frequency image; and using a k-dimensional tree to establish a time domain graph structure for the original vibration signal; a feature extraction module for performing feature extraction on the two-dimensional time-frequency image to obtain a first feature vector, and performing feature extraction on the time domain graph structure to obtain a second feature vector; a feature fusion module for fusing the first feature vector and the second feature vector to obtain a fusion vector; and a feature classification module for classifying the fusion vector to obtain a classification judgment of the original vibration signal.
[0015] The third aspect of the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any one of the above-mentioned intelligent perception methods based on graph structure.
[0016] A fourth aspect of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements any one of the above-mentioned intelligent perception methods based on graph structure.
[0017] The intelligent perception method, system, device and medium based on graph structure provided by the present invention have the beneficial effects of: by constructing a dual-branch graph structure, graph models are constructed for the original time domain signal and its time-frequency image after continuous wavelet transform, and the graph structure is optimized and modeled using a K-dimensional tree, thereby improving the feature extraction effect and expression ability of graph convolution, and ultimately improving the effect of feature fusion in completing classification perception tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.
[0019] Figure 1 A flowchart of an intelligent perception method based on a graph structure provided in an embodiment of the present application;
[0020] Figure 2 A model architecture diagram of the intelligent perception method based on graph structure provided in an embodiment of the present application;
[0021] Figure 3A graph showing the training loss and training accuracy of the intelligent perception method based on the graph structure provided in the embodiment of the present application;
[0022] Figure 4 A classification confusion matrix diagram of the intelligent perception method based on the graph structure provided in the embodiment of the present application;
[0023] Figure 5 A feature distribution diagram of the intelligent perception method based on the graph structure provided in the embodiment of the present application, which is embedded in the t-distributed random neighborhood and visualized by dimensionality reduction after 0-1 weighting;
[0024] Figure 6 A feature distribution diagram of the intelligent perception method based on the graph structure provided in the embodiment of the present application, which is weighted by the Euclidean distance and embedded in the t-distributed random neighborhood with dimensionality reduction visualization;
[0025] Figure 7 The feature distribution diagram of the intelligent perception method based on the graph structure provided in the embodiment of the present application is visualized by embedding the dimension reduction through the t-distributed random neighborhood after the distance negative exponential weighting;
[0026] Figure 8 The feature distribution graph of the intelligent perception method based on the graph structure provided in the embodiment of the present application is visualized by embedding the dimensionality reduction through the t-distributed random neighborhood after the cosine similarity is weighted;
[0027] Figure 9 The feature distribution graph of the intelligent perception method based on the graph structure provided in the embodiment of the present application is visualized by embedding the dimensionality reduction through the t-distributed random neighborhood after the heat kernel weighting;
[0028] Figure 10 A feature distribution diagram visualized by Gaussian weighting and t-distributed random neighborhood embedding dimensionality reduction of the intelligent perception method based on graph structure provided in an embodiment of the present application;
[0029] Figure 11 A framework diagram of an intelligent perception system based on a graph structure provided in an embodiment of the present application;
[0030] Figure 12 This is a schematic block diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0031] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.
[0032] In actual manufacturing systems, vibration signals collected by equipment exhibit rich temporal and frequency characteristics. Although a single physical signal is monomodal, it can be transformed into multiple feature perspectives (such as the original time domain, frequency domain, and time-frequency diagrams) through various mathematical transformations. Joint modeling of these features helps to more comprehensively characterize the equipment's operating status. However, these heterogeneous representations differ in dimensionality, structure, and spatial distribution, making it difficult for traditional feature extraction methods to account for the inherent connections between these various features.
[0033] In recent years, graph neural networks (GNNs) have become an important tool for complex data perception due to their advantages in processing non-Euclidean data. However, in practical applications, the quality of graph construction directly determines the effectiveness of GCN feature extraction. Existing methods mostly use fixed adjacency or simple distance rules, which cannot fully exploit the semantic or structural differences between nodes.
[0034] Based on the above problems, embodiments of the present invention provide an intelligent perception method, system, device and medium based on a graph structure, which can improve data perception effects.
[0035] See also Figure 1 , is an intelligent perception method based on graph structure, including:
[0036] S101, obtaining a one-dimensional original vibration signal;
[0037] S102, performing continuous wavelet transform on the original vibration signal to obtain a two-dimensional time-frequency image; and using a k-dimensional tree to establish a time domain graph structure for the original vibration signal;
[0038] S103, performing feature extraction on the two-dimensional time-frequency image to obtain a first feature vector, and performing feature extraction on the time domain graph structure to obtain a second feature vector;
[0039] S104, fusing the first eigenvector and the second eigenvector to obtain a fused vector;
[0040] S105: Classify the fused vector to obtain the classification judgment of the original vibration signal.
[0041] In step S101 , the original vibration signal is a vibration signal collected by a device in an actual manufacturing system.
[0042] In steps S102-S104, the time-domain graph structure represents a time series signal. To fully exploit the deep features within the modal information, this embodiment employs a dual-branch graph structure: one branch processes the two-dimensional time-frequency image generated by the continuous wavelet transform, while the other processes the original one-dimensional vibration signal. Both branches employ graph convolution using a graph neural network to extract features, resulting in a first eigenvector and a second eigenvector. Both the first and second eigenvectors are 64-dimensional graph eigenvectors, while the fused vector is a 128-dimensional eigenvector.
[0043] Therefore, the intelligent perception method based on graph structure provided in this embodiment constructs a dual-branch graph structure, constructs graph models for the original time domain signal and its time-frequency image after continuous wavelet transform, and uses K-dimensional tree to optimize the modeling of the graph structure, thereby improving the feature extraction effect and expression ability of graph convolution, and ultimately improving the effect of feature fusion to complete the classification perception task.
[0044] In some embodiments, extracting features from the two-dimensional time-frequency image to obtain a first feature vector includes: treating the two-dimensional time-frequency image as a grid graph, treating each pixel in the grid graph as a node in the graph, establishing edge connections between adjacent pixels, and constructing an undirected graph G=(V, E), where V is a set of pixels, E is a set of connecting edges, and the feature of the node is X i =(R i , G i , B i ), represents the RGB three-channel pixel value, and the edge weight w between nodes ij =||X i -X j ||; Perform a two-layer ChebNet convolution on the undirected graph to obtain the first eigenvector.
[0045] In this embodiment, the original vibration signal is transformed into a two-dimensional time-frequency image through continuous wavelet transform. The image can be viewed as a regular grid graph, with each pixel as a node in the graph. Edges are established between adjacent pixels to construct an undirected graph G = (V, E), where V is the set of pixels and E is the set of connecting edges. Node feature X i =(R i , G i , B i ), expressed as RGB three-channel pixel values, and the edge weight between nodes is w ij =||X i -X j ||, defined based on the similarity between pixels. This branch passes through 2 layers of ChebNet convolution, with feature dimensions of 3→128→64.
[0046] In some embodiments, the step of establishing a time domain graph structure for the original vibration signal using a k-dimensional tree includes taking each sampling point of the graph structure as a node in the graph, and establishing edge links between the node and its five nearest neighbors to obtain a time domain graph structure.
[0047] In some embodiments, extracting features from the time-domain graph structure to obtain a second feature vector includes: performing a 2-layer GraphConv convolution on the time-domain graph structure to extract time-domain structural features to obtain a second feature vector.
[0048] In this example, a KD-Tree (k-dimensional tree) graph is constructed for the original one-dimensional vibration signal. Each sampling point is treated as a node in the graph, and edges are connected to its five nearest neighbors. This constructed graph is input into the improved GCN model, where time-domain structural features are extracted through two layers of GraphConv convolution, with the convolution layer configuration shifted from 1024 to 256 to 64.
[0049] In some embodiments, the feature extraction of the time domain graph structure to obtain the second feature vector also includes: in the process of performing 2-layer GraphConv convolution on the time domain graph structure to extract the time domain structure features, uneven edge weighting is performed on the convolution result to dynamically adjust the edge weights according to the feature differences and distances between different nodes, wherein the uneven edge weighting includes: at least one of: 0-1 weighting, distance negative exponential weighting, cosine similarity weighting, heat kernel weighting, and Gaussian kernel weighting.
[0050] In order to improve the performance of graph convolution, this embodiment introduces an uneven edge weighting strategy, which enables GCN to dynamically adjust edge weights based on the feature differences and distances between nodes, thereby improving the modeling capabilities of heterogeneous graph structures. Common edge weighting methods are as follows:
[0051] (1) 0-1 weighted: only considers whether the edge exists, which is simple to calculate but ignores the distance.
[0052] (2) Euclidean distance weighting: The geometric distance between nodes is used to represent similarity. The expression is:
[0053]
[0054] (3) Negative distance exponential weighting: The weight of distant nodes is attenuated. The expression is:
[0055]
[0056] (4) Cosine similarity weighting: Considering the similarity of node features, it is suitable for high-dimensional data and the expression is:
[0057]
[0058] (5) Heat kernel weighting: smooths distance changes and reduces noise interference. The expression is:
[0059]
[0060] (6) Gaussian kernel weighting: Considering the Gaussian attenuation characteristics of distance, the modeling capability is improved. The expression is:
[0061]
[0062] In some embodiments, after performing uneven edge weighting on the convolution result, in the process of performing two-layer GraphConv convolution on the time domain graph structure to extract time domain structural features, the convolution calculation formula includes:
[0063]
[0064] in, is the adjacency matrix calculated based on the weighted strategy, is the degree matrix corresponding to the adjacency matrix, W (l) is the learnable weight parameter matrix.
[0065] The 64-dimensional graph feature vectors output by the two branches are concatenated to form a 128-dimensional fused feature. The fused feature is input into the fully connected layer, ultimately achieving classification and discrimination for multi-view intelligent perception tasks.
[0066] Therefore, this paper introduces the KD-Tree neighbor structure and multiple edge weight design strategies (such as Euclidean distance weighting, cosine similarity, heat kernel function, etc.) to build a more expressive graph structure, thereby improving the modeling performance of graph convolution under heterogeneous feature expression. Although this method does not change the core computing framework of graph convolution, it achieves an improvement in the overall perception effect through graph modeling optimization, and is suitable for multi-perspective modeling and intelligent recognition scenarios for homologous signals.
[0067] In addition, the method provided in this embodiment achieves better accuracy than traditional feature extraction methods in multiple typical working condition recognition experiments, has good engineering practicality, and is suitable for perception and analysis scenarios of equipment operating status in intelligent manufacturing.
[0068] In order to verify the above statement, the embodiments of the present application also carried out actual verification, as follows:
[0069] (1) Dataset division: divided into training set, validation set and test set in a ratio of 7:2:1;
[0070] (2) GCN model training: Train two GCN branch models separately to learn multi-view features;
[0071] (3) Feature fusion and classification: concatenate the features of the two branches and input them into the classifier to complete multi-class discrimination;
[0072] (4) Model evaluation: Use the test set for verification and output the recognition accuracy, confusion matrix and weighted average performance index.
[0073] Models such as Figure 2 shown.
[0074] In order to verify the effectiveness of the intelligent perception method based on graph structure provided by the present invention in multi-working condition bearing fault diagnosis, a public bearing vibration dataset was selected for functional verification.
[0075] The data set used is from the Electronic Engineering Laboratory of Case Western Reserve University in the United States, and includes the following four typical data subsets: normal state, drive end bearing fault, fan end bearing fault, and operating data at different speeds. In this invention, "12k Drive End Bearing Fault Data" is selected as the research object, and three different speed working conditions are set (respectively 1797rpm, 1772rpm, and 1750rpm, recorded as working conditions A, B, and C). Each working condition contains four bearing states: inner ring fault, outer ring fault, rolling element fault, and normal state. Different fault degrees are set for each fault state (fault diameter 0mils, 7mils, 14mils, and 21mils), forming a sample set of multiple fault levels, as shown in Table 1:
[0076] Table 1 Description of bearing fault dataset
[0077]
[0078] The original vibration signal is copied into two branches and input into the GCN diagnosis model according to the above method. The features are obtained and then fused. The number of training samples, validation samples and test samples are 1631, 466 and 233 respectively. The model training loss, training accuracy, validation loss and validation accuracy curves are calculated as follows: Figure 3 shown.
[0079] As can be seen from the figure, after 100 epochs of training, the training loss is close to 0, the training accuracy is close to 100%, and there is no underfitting. After the training is completed, in order to evaluate the fault classification ability of the GCN model, the test set is input into the trained model, and the classification results and the classification accuracy confusion matrix are as follows: Figure 4 As shown, C1-C10 represents ball 1-3 / inner 1-3 / health / outer 1-3. According to the accuracy:
[0080] Precision = TP / (TP+FP)
[0081] Where TP represents the number of samples correctly predicted by the model, and FP represents the number of samples predicted incorrectly. The recall rate can be calculated as:
[0082] Recall = TP / (TP+FN)
[0083] FP represents the sample row with incorrect prediction, and the weighted average of each category is calculated as:
[0084] F1-score=2*Precision / Recall
[0085] The classification accuracy in the figure can be calculated to be 95.5%. It can be seen from the figure that the diagnosis model basically predicts the 10 types of fault standards correctly.
[0086] Different weighting methods were used to determine the model classification accuracy. Each method was performed three times, divided into groups A, B, and C. The three groups were then averaged to obtain the final classification accuracy, as shown in Table 2. Weighted measurement methods include 0-1 weighting, Euclidean distance, negative distance exponential, cosine similarity, and Gaussian kernel function. In one-dimensional bearing data, each one-dimensional signal represents a node, where a and b represent two different nodes. Different edge weighting methods are used to measure the relationship between the two nodes. As can be seen from the table, compared with the original edge structure Euclidean distance, 0-1 weighting, Gaussian kernel function, heat kernel, and negative distance exponential weighting methods have a negative impact on the diagnosis results. Cosine similarity improves the diagnosis accuracy. This simplifies the composition to a certain extent, eliminates the need for advanced prior knowledge, reduces the influence of human experience, and achieves more efficient end-to-end fault diagnosis.
[0087] Table 2 Fault classification accuracy with different weights
[0088]
[0089] The results show that the graph structure modeled by Euclidean distance outperforms other methods in classification accuracy, with an average accuracy of 94.41%. In addition, the feature distribution extracted by the model under different weighting methods is visualized by T-SNE dimensionality reduction. Figure 5-10 ,Euclidean distance and cosine similarity make the categories more distinguishable and the ,feature distribution more clustered, which further shows that improving the graph ,structure can significantly improve the classification performance.
[0090] In summary, this embodiment verifies the robustness and effectiveness of the graph structure construction method described in the present invention in multi-condition fault data scenarios, and is particularly suitable for multi-physical domain signal modeling and intelligent diagnosis tasks based on graph structures.
[0091] See also Figure 11, an embodiment of the present application also provides an intelligent perception system constructed based on a graph structure, including: a signal acquisition module 1, used to obtain a one-dimensional original vibration signal; a signal processing module 2, used to perform a continuous wavelet transform on the original vibration signal to obtain a two-dimensional time-frequency image; and use a k-dimensional tree to establish a time domain graph structure for the original vibration signal; a feature extraction module 3, used to perform feature extraction on the two-dimensional time-frequency image to obtain a first feature vector, and perform feature extraction on the time domain graph structure to obtain a second feature vector; a feature fusion module 4, used to fuse the first feature vector and the second feature vector to obtain a fusion vector; a feature classification module 5, used to classify the fusion vector to obtain a classification judgment of the original vibration signal.
[0092] The intelligent perception system based on graph structure provided in this embodiment constructs a dual-branch graph structure, constructs graph models for the original time domain signal and its time-frequency image after continuous wavelet transform, and uses a K-dimensional tree to optimize the modeling of the graph structure, thereby improving the feature extraction effect and expression ability of graph convolution, and ultimately improving the effect of feature fusion in completing classification perception tasks.
[0093] In some embodiments, the feature extraction module 3 includes: an undirected graph construction unit, which is used to treat the two-dimensional time-frequency image as a grid graph, treat each pixel point in the grid graph as a node in the graph, establish edge connections between adjacent pixel points, and construct an undirected graph G = (V, E), where V is a set of pixel points, E is a set of connecting edges, and the feature of the node is X i =(R i , G i , B i ), represents the RGB three-channel pixel value, and the edge weight w between nodes ij =||X i -X j ||; A first convolution unit is used to perform a two-layer ChebNet convolution on the undirected graph to obtain a first eigenvector.
[0094] In some embodiments, the signal processing module 2 is specifically configured to treat each sampling point of the graph structure as a node in the graph, and establish edge links between the node and its five nearest neighboring points to obtain a time domain graph structure.
[0095] In some embodiments, the feature extraction module 3 further includes: a second convolution unit, configured to perform a two-layer GraphConv convolution on the time-domain graph structure to extract time-domain structural features and obtain a second feature vector.
[0096] In some embodiments, the feature extraction module 3 also includes: an uneven edge weighting unit, which is used to perform uneven edge weighting on the convolution result during the process of the second convolution unit performing a two-layer GraphConv convolution on the time domain graph structure to extract the time domain structural features, so as to dynamically adjust the edge weights according to the feature differences and distances between different nodes, wherein the uneven edge weighting includes: at least one of: 0-1 weighting, distance negative exponential weighting, cosine similarity weighting, heat kernel weighting, and Gaussian kernel weighting.
[0097] In some embodiments, after performing uneven edge weighting on the convolution result, in the process of performing two-layer GraphConv convolution on the time domain graph structure to extract time domain structural features, the convolution calculation formula includes:
[0098]
[0099] in, is the adjacency matrix calculated based on the weighted strategy, is the degree matrix corresponding to the adjacency matrix, W (l) is the learnable weight parameter matrix.
[0100] In some embodiments, the first feature vector and the second feature vector are both 64-dimensional graph feature vectors; the fused vector is a 128-dimensional feature vector; and when classifying the fused vector, a fully connected layer is used for classification.
[0101] The present application also provides an electronic device. Figure 12 The electronic device includes: a memory 601, a processor 602, and a computer program stored in the memory 601 and executable on the processor 602. When the processor 602 executes the computer program, the intelligent perception method based on the graph structure described above is implemented.
[0102] Furthermore, the electronic device also includes: at least one input device 603 and at least one output device 604 .
[0103] The memory 601 , processor 602 , input device 603 , and output device 604 are connected via a bus 605 .
[0104] The input device 603 may be a camera, a touch panel, a physical button, a mouse, etc. The output device 604 may be a display screen.
[0105] The memory 601 can be a high-speed random access memory (RAM) memory or a non-volatile memory such as a disk memory. The memory 601 is used to store a set of executable program codes. The processor 602 is coupled to the memory 601.
[0106] Furthermore, embodiments of the present application also provide a computer-readable storage medium, which may be provided in the electronic device described in each of the above embodiments, and may be the aforementioned memory 601. The computer-readable storage medium stores a computer program, which, when executed by the processor 602, implements the graph-structure-based intelligent perception method described in the above embodiments.
[0107] Furthermore, the computer storable medium may also be various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory 601 (ROM), a RAM, a magnetic disk, or an optical disk.
[0108] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.
[0109] Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of these modules may be selected to achieve the purpose of this embodiment based on actual needs.
[0110] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or software functional modules.
[0111] If the integrated module is implemented in the form of a software function module 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 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, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present invention. It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily all necessary for the present invention.
[0112] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.
[0113] The above is a description of the intelligent perception method, system, electronic device and storage medium based on the graph structure provided by the present invention. For those skilled in the art, according to the ideas of the embodiments of the present invention, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
Claims
1. An intelligent perception method based on graph structure, characterized in that: include: Obtain one-dimensional original vibration signal; Performing a continuous wavelet transform on the original vibration signal to obtain a two-dimensional time-frequency image; and establishing a time domain graph structure for the original vibration signal using a k-dimensional tree; Performing feature extraction on the two-dimensional time-frequency image to obtain a first feature vector, and performing feature extraction on the time domain graph structure to obtain a second feature vector; Fusing the first eigenvector and the second eigenvector to obtain a fused vector; The fusion vector is classified to obtain a classification judgment of the original vibration signal.
2. The intelligent perception method based on graph structure construction according to claim 1 is characterized in that: The step of extracting features from the two-dimensional time-frequency image to obtain a first feature vector includes: The two-dimensional time-frequency image is used as a grid graph, each pixel in the grid graph is used as a node in the graph, and edges are established between adjacent pixels to construct an undirected graph G = (V, E), where V is the set of pixels, E is the set of connecting edges, and the feature of the node is X i =(R i , G i , B i ), represents the RGB three-channel pixel value, and the edge weight w between nodes ij =||X i -X j ||; Perform a 2-layer ChebNet convolution on the undirected graph to obtain the first eigenvector.
3. The intelligent perception method based on graph structure construction according to claim 1 is characterized in that: The method of establishing a time domain graph structure for the original vibration signal by using a k-dimensional tree includes: Each sampling point of the graph structure is regarded as a node in the graph, and edge links are established between the node and its five nearest neighboring points to obtain a time domain graph structure.
4. The intelligent perception method based on graph structure construction according to claim 1 is characterized in that: The extracting features of the time domain graph structure to obtain a second feature vector includes: Perform a two-layer GraphConv convolution on the time-domain graph structure to extract the time-domain structural features and obtain a second feature vector.
5. The intelligent perception method based on graph structure construction according to claim 4 is characterized in that: The extracting features of the time domain graph structure to obtain a second feature vector further includes: In the process of performing two-layer GraphConv convolution on the time domain graph structure to extract time domain structural features, uneven edge weighting is performed on the convolution results to dynamically adjust the edge weights based on feature differences and distances between different nodes, wherein the uneven edge weighting includes: at least one of: 0-1 weighting, negative distance exponential weighting, cosine similarity weighting, heat kernel weighting, and Gaussian kernel weighting.
6. The intelligent perception method based on graph structure construction according to claim 5 is characterized in that: After performing uneven edge weighting on the convolution result, in the process of performing two-layer GraphConv convolution on the time domain graph structure to extract time domain structural features, the convolution calculation formula includes: in, is the adjacency matrix calculated based on the weighted strategy, is the degree matrix corresponding to the adjacency matrix, W (1) is the learnable weight parameter matrix.
7. The intelligent perception method based on graph structure construction according to claim 1 is characterized in that: The first eigenvector and the second eigenvector are both 64-dimensional graph eigenvectors; The fusion vector is a 128-dimensional feature vector; When classifying the fused vector, a fully connected layer is used for classification.
8. An intelligent perception system based on a graph structure, characterized in that: include: A signal acquisition module is used to obtain a one-dimensional original vibration signal; A signal processing module is used to perform a continuous wavelet transform on the original vibration signal to obtain a two-dimensional time-frequency image; and to establish a time domain graph structure for the original vibration signal using a k-dimensional tree; a feature extraction module, configured to perform feature extraction on the two-dimensional time-frequency image to obtain a first feature vector, and perform feature extraction on the time domain graph structure to obtain a second feature vector; a feature fusion module, configured to fuse the first feature vector and the second feature vector to obtain a fused vector; The feature classification module is used to classify the fusion vector to obtain the classification judgment of the original vibration signal.
9. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.