Model rendering methods, apparatus, computer equipment and storage media

Abstract

This application relates to a model rendering method, apparatus, computer equipment, and storage medium. The method includes: acquiring the vibration and acoustic feature map and appearance model information of a gas-insulated switchgear (GIS); extracting saliency decoding features of the GIS based on the vibration and acoustic feature map using a saliency encoding / decoding network; extracting scene decoding features of the GIS based on the appearance model information using a scene encoding / decoding network; fusing the saliency decoding features and scene decoding features of the GIS to obtain fused features of the GIS; and rendering the original GIS model based on the fused features to obtain the target model of the GIS. This method can increase the efficiency of model rendering and reduce the consumption of manpower and material resources.

Description

Technical Field

[0001] This application relates to the field of neural network model technology, and in particular to a model rendering method, apparatus, computer device, and storage medium. Background Technology

[0002] With the development of neural network model technology, people can now use neural network models to model and render objects as auxiliary software for observing objects.

[0003] Normally, people render object models based on two-dimensional images and videos. However, when faced with complex scenes, a large number of two-dimensional images and videos are needed. Processing these images and videos requires a lot of manpower and resources. Therefore, how to render object models quickly and efficiently has become an urgent problem to be solved. Summary of the Invention

[0004] Therefore, it is necessary to provide a model rendering method, apparatus, computer equipment, and storage medium that can efficiently render object models in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a model rendering method. The method includes:

[0006] Obtain the vibration and acoustic characteristics and appearance model information of the gas-insulated switchgear (GIS);

[0007] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0008] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0009] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0010] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0011] In one embodiment, the saliency coding / decoding network includes a saliency coding network and a saliency decoding network;

[0012] Using a saliency encoding / decoding network, saliency decoding features of the GIS are extracted based on the vibration acoustic feature map, including:

[0013] The saliency coding network is used to extract features from the vibration acoustic feature map to obtain the saliency coding features of GIS.

[0014] The saliency decoding network is used to decode the saliency coding features of GIS to obtain the saliency decoding features of GIS.

[0015] In one embodiment, the saliency decoding network includes a first long short-term memory network and an attention mechanism network; the saliency decoding network decodes the saliency encoded features of the GIS to obtain the saliency decoded features of the GIS, including:

[0016] Based on the saliency coding features of GIS, determine the average vector and word embedding vector of the saliency coding features;

[0017] Using the first long short-term memory network, the hidden state vector of the saliency encoding feature is determined based on the average vector of the saliency encoding feature and the word embedding vector;

[0018] By using an attention mechanism network, saliency-weighted features are determined based on the hidden state vector and saliency encoding features, and these features are used as saliency decoding features for GIS.

[0019] In one embodiment, the scene encoding / decoding network includes a scene encoding network and a scene decoding network; through the scene encoding / decoding network, scene decoding features of the GIS are extracted based on appearance model information, including:

[0020] By using a scene coding network, feature extraction processing is performed on the appearance model information to obtain the scene coding features of GIS;

[0021] The scene decoding features of GIS are decoded by a scene decoding network to obtain the scene decoding features of GIS.

[0022] In one embodiment, the scene decoding network includes a second long short-term memory network; the scene decoding network decodes the scene coding features of the GIS to obtain the scene decoding features of the GIS, including:

[0023] Based on the scene coding characteristics of GIS, determine the word embedding vector of the scene coding characteristics;

[0024] By using a second long short-term memory network, the scene decoding features of GIS are determined based on scene encoding features and word embedding vectors of scene encoding features.

[0025] In one embodiment, the original GIS model is rendered based on fusion features to obtain the target GIS model, including:

[0026] Based on the reinforcement learning SCST algorithm, the original GIS model is rendered according to the fusion features to obtain the target GIS model.

[0027] Secondly, this application also provides a model rendering apparatus. The apparatus includes:

[0028] Obtain vibration and acoustic characteristic maps and appearance model information of gas-insulated switchgear (GIS);

[0029] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0030] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0031] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0032] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0033] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0034] Obtain vibration and acoustic characteristic maps and appearance model information of gas-insulated switchgear (GIS);

[0035] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0036] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0037] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0038] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0039] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0040] Obtain vibration and acoustic characteristic maps and appearance model information of gas-insulated switchgear (GIS);

[0041] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0042] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0043] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0044] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0045] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0046] Obtain vibration and acoustic characteristic maps and appearance model information of gas-insulated switchgear (GIS);

[0047] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0048] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0049] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0050] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0051] The aforementioned model rendering method, apparatus, computer equipment, and storage medium acquire the vibration and acoustic feature maps and appearance model information of the gas-insulated combined electrical equipment (GIS), then decode them to obtain decoded features, fuse the decoded features, and render the original GIS model based on the fused features to obtain the target model. This model rendering method does not require a large amount of two-dimensional images and video materials, nor does it require a large amount of manpower and resources, and can quickly and efficiently complete the rendering of the model. Attached Figure Description

[0052] Figure 1 This is a diagram illustrating the application environment of a model rendering method in one embodiment.

[0053] Figure 2 This is a flowchart illustrating a model rendering method in one embodiment;

[0054] Figure 3 This is a flowchart illustrating the process of obtaining GIS saliency encoding and decoding features in one embodiment;

[0055] Figure 4 This is a flowchart illustrating the process of obtaining GIS saliency decoding features in one embodiment;

[0056] Figure 5 This is a schematic diagram of decoding target salient features in a first long short-term memory network in one embodiment;

[0057] Figure 6 This is a schematic diagram of the process for obtaining GIS scene encoding and decoding features in one embodiment;

[0058] Figure 7 This is a schematic diagram of the process for obtaining GIS scene decoding features in one embodiment;

[0059] Figure 8 This is a schematic diagram illustrating the decoding of scene features in a second long short-term memory network in one embodiment.

[0060] Figure 9 This is a flowchart illustrating the model rendering method in another embodiment;

[0061] Figure 10 This is a structural block diagram of a model rendering apparatus in one embodiment;

[0062] Figure 11 This is a structural block diagram of the model rendering apparatus in another embodiment;

[0063] Figure 12 This is a structural block diagram of the model rendering apparatus in yet another embodiment;

[0064] Figure 13 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0066] The model rendering method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, in one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows. Figure 1 As shown. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data required for performance analysis of time-domain control elements. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements the model rendering method shown in any of the following embodiments.

[0067] In one embodiment, such as Figure 2 As shown, a model rendering method is provided, which can be applied to... Figure 1 Taking a computer device as an example, the explanation includes the following steps:

[0068] S201, Obtain the vibration acoustic characteristic map and appearance model information of the gas-insulated switchgear (GIS).

[0069] Among them, the vibration acoustic feature map is a map that characterizes the sound intensity value of each sound pressure point in the time-domain waveform of noise sound pressure; the appearance model information is a two-dimensional image obtained by taking pictures.

[0070] Optionally, a ring-shaped acoustic pressure sensor array is used to sequentially measure the vibration and noise sound pressure time-domain waveforms at various measuring points of the gas-insulated switchgear (GIS). The measured vibration and noise sound pressure time-domain waveforms of the GIS are then transmitted to a computer via a data acquisition card. The sound intensity value at each pair of sound pressure sensors is calculated using the sound pressure level of each pair of sensors (the formula for calculating the sound intensity value is: I = ρ × C / p). 2 Where I is sound intensity, p is sound pressure, ρ is medium density, and C is sound velocity, the sound intensity value of each sound pressure point in the time-domain waveform of the vibration noise sound pressure is used as the vibration characteristic map of gas-insulated combined electrical equipment (GIS).

[0071] The ring-shaped sound pressure sensor array consists of 10 sound pressure sensors forming 5 pairs of sound pressure probes. There is one sound pressure probe on the inner and outer sides of the arc-shaped support frame of the sensor array, forming a pair of sound pressure sensors. The measurement points are evenly arranged on the surface of the GIS shell, and a ring-shaped sound pressure sensor array is set every 20 centimeters.

[0072] Optionally, a two-dimensional image of the gas-insulated switchgear GIS can be captured by a camera as an appearance model information for the gas-insulated switchgear GIS.

[0073] S202 extracts saliency decoding features of GIS based on the vibration acoustic feature map through a saliency coding and decoding network.

[0074] Among them, the saliency encoding / decoding network is a neural network model used to extract saliency decoding features from GIS.

[0075] Optionally, the seismic feature map of the GIS can be input into a saliency coding network. The saliency coding network will perform feature encoding operations on the input seismic feature map of the GIS to obtain the saliency decoding features of the GIS.

[0076] S203 extracts scene decoding features from GIS based on appearance model information through a scene encoding / decoding network.

[0077] Among them, the scene encoding / decoding network is a neural network model used to extract scene decoding features from GIS.

[0078] Optionally, the appearance model information can be input into a scene encoding / decoding network, which will perform feature encoding on the input appearance model information to obtain the scene decoding features of the GIS.

[0079] S204, the saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS.

[0080] Optionally, the saliency decoding features and scene decoding features of GIS can be complemented. The saliency decoding features of GIS can display the sound intensity value of each sound pressure point, that is, the physical field information of GIS. The scene decoding features of GIS can display the appearance and shape of GIS. By fusing the two types of information, the physical field information of each appearance location of GIS can be obtained.

[0081] S205. Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0082] The target model is the original GIS model that has been rendered.

[0083] Optionally, the original GIS model can be rendered based on the fusion features using a reinforcement learning (Self-Critical Sequence Training, SCST) algorithm to obtain the target GIS model.

[0084] Specifically, for example, based on the similarity between the fusion features of the GIS and the features of the original GIS model, a Consensus-based Image Description Evaluation (CIDEr) score is calculated. Model optimization training is then performed based on the CIDEr score. During training, the expected negative reward of the trained model is calculated, and the trained model with the smallest expected negative reward value is selected as the rendered GIS model, i.e., the target model. The method for calculating the expected negative reward value is shown in the following formula (1):

[0085]

[0086] Where y1:T represents the descriptive statement generated by the model, r(.) represents the reinforcement learning reward function, and L RL This represents the expected value of the negative reward obtained through training in a learning manner, and E represents the function for calculating the expected value.

[0087] In the above model method, the saliency decoding features and scene decoding features of GIS are extracted using the model, and the two decoding features are fused. The original GIS model is then optimized based on the fused features. When dealing with objects in complex scenes, this method can also directly extract the two features from the model and render the original model of the object based on the fused features, which greatly increases the efficiency of model rendering and reduces the consumption of manpower and material resources.

[0088] The previous embodiment provided an overview of how to render the model. In this embodiment, the saliency encoding / decoding network is decomposed into a saliency encoding network and a saliency decoding network, such as... Figure 3 As shown, this paper elaborates on a specific method for extracting saliency decoding features of GIS based on vibration acoustic feature maps using a saliency encoding / decoding network, including:

[0089] S301 uses a saliency coding network to extract features from the vibration acoustic feature map, thereby obtaining the saliency coding features of GIS.

[0090] Among them, the saliency coding network is a neural network model used to extract saliency coding features; the saliency coding features of GIS are features obtained by saliency coding of vibration acoustic feature maps.

[0091] Optionally, the saliency coding network can be a Faster Region-Convolutional Neural Network (Faster R-CNN). The GIS feature map can be input into the Faster R-CNN network, and the features output by the last convolutional layer of the Faster R-CNN network can be used as the saliency coding features of the GIS. The formula for extracting the saliency coding features of GIS is as follows (2):

[0092] V=ffaster r-cnn(I) (2)

[0093] Where V = {v1, v2, v3, ..., v k}。 vi∈R D , represents the salient feature vector corresponding to the i-th physical quantity in the vibration acoustic feature map, where the physical quantities in this embodiment may include: sound intensity, sound pressure, and characteristic frequency. k represents the total number of physical quantities extracted from each image, faster r-cnn(.) represents the target display feature extraction method, I is the input vibration acoustic feature map, and R D Let vi be a D-dimensional vector, where the values ​​of each dimension of the vector vi are real numbers R.

[0094] S302, through the saliency decoding network, decodes the saliency coding features of GIS to obtain the saliency decoding features of GIS.

[0095] Among them, the saliency decoding network is a neural network model used to extract saliency decoding features; the saliency decoding features of GIS are the features after decoding the saliency encoding features of GIS.

[0096] Optionally, the saliency coding features of the GIS can be input into a saliency decoding network. The saliency decoding network will perform feature decoding operations on the input saliency coding features of the GIS and output the saliency decoding features of the GIS.

[0097] The above embodiments encode and decode GIS feature maps to obtain decoded GIS feature maps. This scheme makes the format of GIS feature maps more unified and facilitates subsequent feature fusion.

[0098] Based on the above embodiments, this embodiment further subdivides the saliency decoding network into a first long short-term memory network and an attention mechanism network, such as... Figure 4 As shown, the saliency coding features of GIS are decoded using a saliency decoding network to obtain the saliency decoded features of GIS, including:

[0099] S401, Based on the saliency coding features of GIS, determine the average vector and word embedding vector of the saliency coding features.

[0100] Specifically, based on formula (3), the average value of all the values ​​of V obtained from formula (2) is taken to obtain the average vector of the saliency coding features.

[0101]

[0102] in, is the average vector of saliency encoding features, k represents the target saliency features corresponding to k physical quantities, and vi represents the saliency feature vector corresponding to the i-th physical quantity of the vibration acoustic feature map.

[0103] When determining the word embedding vector, the saliency coding features corresponding to each physical quantity in the saliency coding features of GIS can be transformed in a preset way to obtain a vector of fixed length, which can be used as the word embedding vector corresponding to that physical quantity.

[0104] S402, through the first long short-term memory network, the hidden state vector of the saliency encoding feature is determined based on the average vector of the saliency encoding feature and the word embedding vector.

[0105] Among them, the first Long Short Term Memory (LSTM) network is a neural network model used to determine the hidden state vector of salient encoded features.

[0106] Optional, such as Figure 5 As shown, for each physical quantity, the saliency encoding feature is determined at time t in the first Long Short-Term Memory network. For each time t in the first Long Short-Term Memory network, the corresponding LSTM network layer is determined. et The average vector of saliency-encoding features The word embedding vector x of the physical quantity at this moment t The hidden transfer vector of the previous time step (i.e., time step t-1) and the hidden transfer vector of the previous time step. Input LSTM et The specific decoding process is shown in the following formula (4):

[0107]

[0108] in, Indicates LSTM at time t et The hidden state vector, x t This is the word embedding vector of the physical quantity corresponding to that moment. This is the hidden transfer vector of the previous time step (i.e., time step t-1).

[0109] S403 uses an attention mechanism network to determine saliency-weighted features based on the hidden state vector and saliency encoding features, which serve as saliency decoding features for GIS.

[0110] Among them, the attention mechanism network is used to extract salient weighted features based on the hidden state vector and salient encoded features.

[0111] Specifically, such as Figure 5 As shown, the attention mechanism network uses the LSTM network layer corresponding to each time step t in the first long short-term memory network. et Output hidden state vector and saliency coding feature v i As input, we obtain the significance-weighted features corresponding to each time t (i.e., each physical quantity). The specific calculation process is shown in the following formulas (5), (6), and (7):

[0112]

[0113] αi,t=Softmax(ai,t) (6)

[0114]

[0115] Among them, W a W va W ha It is the parameter matrix for learning, a i,t α represents the weight of the saliency encoding feature corresponding to the i-th physical quantity before normalization. i,t This represents the weight of the saliency encoding feature corresponding to the i-th physical quantity after normalization, where Softmax(.) represents the normalization process, and v i For saliency coding features, For LSTM et The hidden state vector, The significance-weighted feature at time t.

[0116] The above embodiments obtain GIS saliency decoding features by processing GIS saliency coding features. The encoding and decoding method using neural network models eliminates the need for manual analysis of a large number of two-dimensional images and videos, greatly reducing the consumption of manpower and material resources.

[0117] The above examples illustrate how to decode multiple GIS feature maps. Building upon this, we still need to decode the GIS appearance model information. The scene encoding / decoding network includes a scene encoding network and a scene decoding network, such as... Figure 6 As shown, the scene decoding features of GIS are extracted based on the appearance model information through a scene encoding / decoding network, including:

[0118] S601 uses a scene coding network to extract features from the appearance model information to obtain the scene coding features of the GIS.

[0119] The scene coding network is a neural network model used to extract scene coding features.

[0120] Optionally, the appearance model information is input into the scene coding network, and the features output by the last fully connected layer of the network are used as the scene coding features of the GIS. The extraction formula is as follows: (8)

[0121] G = f CNN (I) (8)

[0122] Where, G∈R D D represents the dimension of scene features, G represents the scene coding features in GIS, and f CNN (.) indicates the scene feature extraction process.

[0123] S602 uses a scene decoding network to decode the scene coding features of GIS to obtain the scene decoding features of GIS.

[0124] The scene decoding network is a neural network model used to extract scene decoding features.

[0125] Optionally, the scene encoding features of the GIS can be input into the scene decoding network, which will perform feature decoding operations on the input scene encoding features of the GIS and output the scene decoding features of the GIS.

[0126] In the above embodiments, scene decoding features are obtained by decoding appearance model information through neural network model, which provides scene decoding features in a unified format for subsequent feature fusion, and facilitates the fusion of the two features.

[0127] Based on the above embodiments, such as Figure 7 As shown, in this embodiment, the scene decoding network is specifically a second long short-term memory network. Through the scene decoding network, the scene encoding features of the GIS are decoded to obtain the scene decoding features of the GIS, including:

[0128] S701, Based on the scene coding features of GIS, determine the word embedding vector of the scene coding features.

[0129] Optionally, when determining the word embedding vector, the scene coding features of the GIS can be transformed in a preset way to obtain a vector of fixed length, which can be used as the word embedding vector corresponding to the physical quantity.

[0130] S702 uses a second long short-term memory network to determine the scene decoding features of GIS based on scene coding features and word embedding vectors of scene coding features.

[0131] Among them, the second long short-term memory network is a neural network model for determining scene decoding features in GIS.

[0132] Optional, such as Figure 8 As shown, for the scene encoding features, the corresponding time t in the second long short-term memory network is determined, and the network layer LSTM corresponding to each time t in the second long short-term memory network is... st The scene encoding feature vector G and the word embedding vector x corresponding to the scene at that moment are used to encode the scene feature vector G and the scene embedding vector x. t The hidden transfer vector of the previous time step (i.e., time step t-1) and the hidden transfer vector of the previous time step. Input LSTM st The specific decoding process is shown in the following formula (9):

[0133]

[0134] in, Indicates LSTM at time t st The hidden state vector, x t This is the word embedding vector of the physical quantity corresponding to that moment. This is the hidden transfer vector of the previous time step (i.e., time step t-1). [.] indicates a vector concatenation operation.

[0135] To more comprehensively demonstrate this solution, this embodiment presents an optional approach to energy dispatching, such as... Figure 9 As shown.

[0136] S901, acquire the vibration acoustic characteristic map and appearance model information of the gas-insulated switchgear (GIS).

[0137] S902 uses a saliency coding network to extract features from the vibration acoustic feature map, thus obtaining the saliency coding features of GIS.

[0138] S903, based on the saliency coding features of GIS, determine the average vector and word embedding vector of the saliency coding features.

[0139] S904, through the first long short-term memory network, determines the hidden state vector of the saliency encoding feature based on the average vector of the saliency encoding feature and the word embedding vector.

[0140] S905 uses an attention mechanism network to determine saliency-weighted features based on the hidden state vector and saliency encoding features, which serve as saliency decoding features for GIS.

[0141] S906 uses a scene coding network to extract features from the appearance model information to obtain the scene coding features of GIS.

[0142] S907, based on the scene coding characteristics of GIS, determine the word embedding vector of the scene coding characteristics.

[0143] S908 uses a second long short-term memory network to determine the scene decoding features of GIS based on scene encoding features and word embedding vectors of scene encoding features.

[0144] S909 integrates the saliency decoding features and scene decoding features of GIS to obtain the integrated features of GIS.

[0145] S910, based on the reinforcement learning SCST algorithm, renders the original GIS model according to the fusion features to obtain the target GIS model.

[0146] The specific processes of S901-S910 described above can be found in the description of the above method embodiments. Their implementation principles and technical effects are similar, and will not be repeated here.

[0147] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0148] Based on the same inventive concept, this application also provides a model rendering apparatus for implementing the model rendering method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more model rendering apparatus embodiments provided below can be found in the limitations of the model rendering method described above, and will not be repeated here.

[0149] In one embodiment, such as Figure 10 As shown, a model rendering device 10 is provided, including: an information acquisition module 100, a first extraction module 101, a second extraction module 102, a feature fusion module 103, and a model rendering module 104, wherein:

[0150] The information acquisition module 100 is used to acquire the vibration and acoustic characteristic map and appearance model information of gas-insulated combined electrical equipment (GIS).

[0151] The first extraction module 101 is used to extract saliency decoding features of GIS based on the vibration acoustic feature map through a saliency encoding and decoding network.

[0152] The second extraction module 102 is used to extract scene decoding features of GIS based on appearance model information through a scene encoding and decoding network;

[0153] The feature fusion module 103 is used to fuse the saliency decoding features and scene decoding features of GIS to obtain the fused features of GIS.

[0154] The model rendering module 104 is used to render the original GIS model based on the fusion features to obtain the target GIS model.

[0155] In one embodiment, such as Figure 11 As shown above, Figure 10 The first extraction module 101 may include:

[0156] The first extraction unit 1010 is used to perform feature extraction processing on the vibration acoustic feature map through a saliency coding network to obtain the saliency coding features of GIS.

[0157] The second extraction unit 1011 is used to decode the saliency coding features of GIS through a saliency decoding network to obtain the saliency decoding features of GIS.

[0158] In one embodiment, the above Figure 11 The second extraction unit 1011 is specifically used for:

[0159] Based on the saliency coding features of GIS, the average vector and word embedding vector of the saliency coding features are determined; through the first long short-term memory network, the hidden state vector of the saliency coding features is determined based on the average vector and word embedding vector of the saliency coding features; through the attention mechanism network, the saliency weighted features are determined based on the hidden state vector and the saliency coding features, which serve as the saliency decoding features of GIS.

[0160] In one embodiment, such as Figure 12 As shown above, Figure 10 The second extraction module 102 may include:

[0161] The third extraction unit 1020 is used to perform feature extraction processing on the appearance model information through the scene coding network to obtain the scene coding features of GIS.

[0162] The fourth extraction unit 1021 is used to decode the scene coding features of GIS through a scene decoding network to obtain the scene decoding features of GIS.

[0163] In one embodiment, the above Figure 12 The fourth extraction unit 1021 is specifically used for:

[0164] Based on the scene coding features of GIS, the word embedding vectors of the scene coding features are determined; through the second long short-term memory network, the scene decoding features of GIS are determined based on the scene coding features and the word embedding vectors of the scene coding features.

[0165] In one embodiment, the above Figure 10 The model rendering module 104 in the middle can also be used for:

[0166] Based on the reinforcement learning SCST algorithm, the original GIS model is rendered according to the fusion features to obtain the target GIS model.

[0167] Each module in the aforementioned model rendering device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0168] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 13 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices 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 the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a model rendering method. The display screen can be an LCD screen or an e-ink display screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0169] Those skilled in the art will understand that Figure 13 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0170] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0171] Obtain vibration and acoustic characteristic maps and appearance model information of gas-insulated switchgear (GIS);

[0172] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0173] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0174] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0175] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0176] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0177] The saliency coding network is used to extract features from the vibration acoustic feature map to obtain the saliency coding features of GIS; the saliency decoding network is used to decode the saliency coding features of GIS to obtain the saliency decoding features of GIS.

[0178] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0179] Based on the saliency coding features of GIS, the average vector and word embedding vector of the saliency coding features are determined; through the first long short-term memory network, the hidden state vector of the saliency coding features is determined based on the average vector and word embedding vector of the saliency coding features; through the attention mechanism network, the saliency weighted features are determined based on the hidden state vector and the saliency coding features, which serve as the saliency decoding features of GIS.

[0180] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0181] The scene coding network extracts features from the appearance model information to obtain the scene coding features of the GIS; the scene decoding network decodes the scene coding features of the GIS to obtain the scene decoding features of the GIS.

[0182] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0183] Based on the scene coding features of GIS, the word embedding vectors of the scene coding features are determined; through the second long short-term memory network, the scene decoding features of GIS are determined based on the scene coding features and the word embedding vectors of the scene coding features.

[0184] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0185] Based on the reinforcement learning SCST algorithm, the original GIS model is rendered according to the fusion features to obtain the target GIS model.

[0186] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0187] Obtain vibration and acoustic characteristic maps and appearance model information of gas-insulated switchgear (GIS);

[0188] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0189] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0190] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0191] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0192] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0193] The saliency coding network is used to extract features from the vibration acoustic feature map to obtain the saliency coding features of GIS; the saliency decoding network is used to decode the saliency coding features of GIS to obtain the saliency decoding features of GIS.

[0194] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0195] Based on the saliency coding features of GIS, the average vector and word embedding vector of the saliency coding features are determined; through the first long short-term memory network, the hidden state vector of the saliency coding features is determined based on the average vector and word embedding vector of the saliency coding features; through the attention mechanism network, the saliency weighted features are determined based on the hidden state vector and the saliency coding features, which serve as the saliency decoding features of GIS.

[0196] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0197] The scene coding network extracts features from the appearance model information to obtain the scene coding features of the GIS; the scene decoding network decodes the scene coding features of the GIS to obtain the scene decoding features of the GIS.

[0198] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0199] Based on the scene coding features of GIS, the word embedding vectors of the scene coding features are determined; through the second long short-term memory network, the scene decoding features of GIS are determined based on the scene coding features and the word embedding vectors of the scene coding features.

[0200] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0201] Based on the reinforcement learning SCST algorithm, the original GIS model is rendered according to the fusion features to obtain the target GIS model.

[0202] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0203] Obtain vibration and acoustic characteristic maps and appearance model information of gas-insulated switchgear (GIS);

[0204] Based on the vibration acoustic feature map, saliency decoding features of GIS are extracted using a saliency encoding / decoding network.

[0205] By using a scene encoding and decoding network, scene decoding features of GIS are extracted based on appearance model information;

[0206] The saliency decoding features and scene decoding features of GIS are fused to obtain the fused features of GIS;

[0207] Based on the fusion characteristics, the original GIS model is rendered to obtain the target GIS model.

[0208] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0209] The saliency coding network is used to extract features from the vibration acoustic feature map to obtain the saliency coding features of GIS; the saliency decoding network is used to decode the saliency coding features of GIS to obtain the saliency decoding features of GIS.

[0210] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0211] Based on the saliency coding features of GIS, the average vector and word embedding vector of the saliency coding features are determined; through the first long short-term memory network, the hidden state vector of the saliency coding features is determined based on the average vector and word embedding vector of the saliency coding features; through the attention mechanism network, the saliency weighted features are determined based on the hidden state vector and the saliency coding features, which serve as the saliency decoding features of GIS.

[0212] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0213] The scene coding network extracts features from the appearance model information to obtain the scene coding features of the GIS; the scene decoding network decodes the scene coding features of the GIS to obtain the scene decoding features of the GIS.

[0214] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0215] Based on the scene coding features of GIS, the word embedding vectors of the scene coding features are determined; through the second long short-term memory network, the scene decoding features of GIS are determined based on the scene coding features and the word embedding vectors of the scene coding features.

[0216] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0217] Based on the reinforcement learning SCST algorithm, the original GIS model is rendered according to the fusion features to obtain the target GIS model.

[0218] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0219] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0220] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A model rendering method, characterized by, The method comprises: acquiring a vibration acoustic feature map and appearance model information of a gas insulated switchgear (GIS); the vibration acoustic feature map is a map representing sound intensity values of each sound pressure point in a noise sound pressure time domain waveform; and the appearance model information is a two-dimensional image of the GIS; extracting saliency decoding features of the GIS according to the vibration acoustic feature map through a saliency coding and decoding network; the saliency coding and decoding network comprises a saliency coding network and a saliency decoding network; and the saliency decoding network comprises a first long short-term memory network and an attention mechanism network; extracting scene decoding features of the GIS according to the appearance model information through a scene coding and decoding network; fusing the saliency decoding features and the scene decoding features of the GIS to obtain fused features of the GIS; rendering an original model of the GIS according to the fused features to obtain a target model of the GIS; wherein the extracting the saliency decoding features of the GIS according to the vibration acoustic feature map through the saliency coding and decoding network comprises: performing feature extraction processing on the vibration acoustic feature map through the saliency coding network to obtain saliency coding features of the GIS; determining an average vector and a word embedding vector of the saliency coding features according to the saliency coding features of the GIS; determining a hidden state vector of the saliency coding features according to the average vector and the word embedding vector of the saliency coding features through the first long short-term memory network; determining a saliency weighted feature as the saliency decoding features of the GIS according to the hidden state vector and the saliency coding features through the attention mechanism network; wherein each saliency coding feature corresponding to a physical quantity in the saliency coding features of the GIS is converted in a preset manner to obtain a fixed-length vector; and the fixed-length vector is taken as a word embedding vector corresponding to the physical quantity.

2. The method of claim 1, wherein, the scene coding and decoding network comprises a scene coding network and a scene decoding network; and the extracting the scene decoding features of the GIS according to the appearance model information through the scene coding and decoding network comprises: performing feature extraction processing on the appearance model information through the scene coding network to obtain scene coding features of the GIS; performing decoding processing on the scene coding features of the GIS through the scene decoding network to obtain the scene decoding features of the GIS.

3. The method of claim 2, wherein, the scene decoding network comprises a second long short-term memory network; and the performing decoding processing on the scene coding features of the GIS through the scene decoding network to obtain the scene decoding features of the GIS comprises: determining a word embedding vector of the scene coding features according to the scene coding features of the GIS; determining the scene decoding features of the GIS according to the scene coding features and the word embedding vector of the scene coding features through the second long short-term memory network.

4. The method according to any one of claims 1 to 3, characterized in that, the rendering the original model of the GIS according to the fused features to obtain the target model of the GIS comprises: Based on a reinforcement learning SCST algorithm, the original model of the GIS is rendered according to the fusion feature to obtain a target model of the GIS.

5. A model rendering apparatus, characterized by comprising: The device comprises: An information acquisition module is configured to acquire a vibration acoustic feature map and appearance model information of a gas insulated switchgear (GIS); the vibration acoustic feature map is a graph representing sound intensity values of each sound pressure point in a noise sound pressure time domain waveform; and the appearance model information is a two-dimensional image of the GIS. A first extraction module is configured to extract saliency decoding features of the GIS from the vibration acoustic feature map through a saliency coding and decoding network; the saliency coding and decoding network comprises a saliency coding network and a saliency decoding network; the saliency decoding network comprises a first long short-term memory network and an attention mechanism network. A second extraction module is configured to extract scene decoding features of the GIS from the appearance model information through a scene coding and decoding network. A feature fusion module is configured to fuse the saliency decoding features and the scene decoding features of the GIS to obtain fusion features of the GIS. A model rendering module is configured to render an original model of the GIS according to the fusion features to obtain a target model of the GIS. The method comprises the following steps: The saliency coding network is used to perform feature extraction processing on the vibration acoustic feature map to obtain saliency coding features of the GIS; The average vector and the word embedding vector of the saliency coding features are determined according to the saliency coding features of the GIS; The hidden state vector of the saliency coding features is determined according to the average vector and the word embedding vector of the saliency coding features through the first long short-term memory network; The saliency weighted features are determined as the saliency decoding features of the GIS according to the hidden state vector and the saliency coding features through the attention mechanism network. The fixed-length vector is obtained by converting each saliency coding feature corresponding to a physical quantity in the saliency coding features according to a preset mode; and the fixed-length vector is used as the word embedding vector corresponding to the physical quantity.

6. The apparatus of claim 5, wherein, The scene coding and decoding network comprises a scene coding network and a scene decoding network; and the second extraction module comprises: A third extraction unit is configured to perform feature extraction processing on the appearance model information through the scene coding network to obtain scene coding features of the GIS; A fourth extraction unit is configured to perform decoding processing on the scene coding features of the GIS through the scene decoding network to obtain the scene decoding features of the GIS.

7. The apparatus of claim 5 or 6, wherein, The model rendering module is configured to render the original model of the GIS according to the fusion features based on a reinforcement learning SCST algorithm to obtain a target model of the GIS. The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 4.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 4.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, ​ 10. A computer program product comprising a computer program, characterized in that, The computer program, which when executed by the processor, implements the steps of the method of any one of claims 1 to 4.

Images