Method, device, electronic equipment and medium for noise recognition

Abstract

The application discloses a noise recognition method and device, electronic equipment and medium. In the application, the optimal slope can be dynamically calculated from a plurality of candidate slopes and used as the slope of the deep residual shrinkage network, so that different optimal slopes of the soft threshold corresponding to the residual network can be selected to remove noise redundancy in the noise reduction process according to different noise signals. Thus, the response of the entire network model to external noise signals is improved, and the accuracy and precision of noise fault diagnosis are improved.

Description

Technical Field

[0001] This application relates to data processing technology, and in particular to a method, apparatus, electronic device, and medium for noise identification. Background Technology

[0002] With the development of technology, the proportion of high-quality products in the overall national production is increasing. Whether it is manufactured goods or consumer products, as products are used more, they will inevitably suffer wear and tear. Therefore, an accurate fault diagnosis system directly determines the quality of a product.

[0003] Furthermore, judging the operating sound of many products is an important method for fault diagnosis in related technologies. Currently, related technologies often employ soft thresholding for noise reduction. Among them, the soft thresholding function, as a classic method, is particularly practical in the field of signal denoising. The inherent non-linear properties of soft thresholding are well-suited for use in the computation and transmission processes of deep neural networks.

[0004] However, how to use the soft thresholding function for noise identification has become a problem that needs to be solved by those in the field. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and medium for noise recognition. This application addresses the problem in related technologies that the optimal slope depth residual shrinkage network cannot be deployed in noise recognition devices to improve noise recognition performance.

[0006] According to one aspect of the embodiments of this application, a noise identification method is provided, comprising:

[0007] Obtain the target slope value set and the target soft thresholding function;

[0008] According to a preset order strategy, each target slope value in the target slope value set is combined with the target soft thresholding function to generate the corresponding slope soft thresholding function.

[0009] Noise data is input into multiple soft-threshold residual networks to obtain the noise recognition result output by each soft-threshold residual network. The soft-threshold residual network is a different soft-threshold residual network constructed by each slope soft-thresholding function and the residual network.

[0010] The target soft threshold residual network is determined to be deployed on the noise recognition device. The target soft threshold residual network is the soft threshold residual network corresponding to the noise recognition result with the highest noise recognition rate.

[0011] Optionally, in another embodiment based on the method described above in this application, before obtaining the target slope value set, the method further includes:

[0012] Obtain the set of slope values ​​to be selected;

[0013] Multiple target slope values ​​in the set of slope values ​​to be selected are selected at preset intervals.

[0014] The target slope value set is generated based on the plurality of target slope values.

[0015] Optionally, in another embodiment based on the method described above in this application, the preset order strategy combines each target slope value in the target slope value set with the target soft thresholding function to generate a corresponding slope soft thresholding function, including:

[0016] In order of value size, each target slope value is sequentially paired with the target soft thresholding function to generate a corresponding slope soft thresholding function.

[0017] Optionally, in another embodiment based on the method described above in this application, obtaining the target soft thresholding function includes:

[0018] Obtain the first input feature, and perform at least two convolution operations on the first input feature, as well as batch regularization and linear rectified function operations to obtain the first output result;

[0019] The absolute value algorithm and global average pooling operation are performed on the first output result to obtain the second output result;

[0020] Based on the second output result, the target soft thresholding function is obtained.

[0021] Optionally, in another embodiment based on the method described above in this application, obtaining the target soft thresholding function based on the second output result includes:

[0022] The second output result is subjected to convolution, batch regularization, fully connected operation, and linear rectified function operation to obtain the third output result.

[0023] The third output result is sigmoidized to obtain the target soft thresholding function.

[0024] Optionally, in another embodiment based on the method described above in this application, the step of inputting noise data into multiple soft-threshold residual networks to obtain the noise identification result output by each soft-threshold residual network includes:

[0025] The noise data is acquired, and a convolution operation is performed on the noise data to extract semantic information from the noise data, thereby obtaining the output result of the first dimension.

[0026] The slope soft thresholding function is used to remove noise redundancy from the output of the first dimension;

[0027] Batch regularization, global average pooling, and linear rectified function transformation are performed on the first dimension output to remove noise redundancy, resulting in the second dimension output.

[0028] The noise recognition result is obtained based on the output of the second dimension.

[0029] Optionally, in another embodiment based on the method described above, obtaining the noise recognition result based on the output result of the second dimension includes:

[0030] Perform a transformation operation to convert the output of the second dimension into an N-dimensional vector;

[0031] The noise recognition result is obtained by performing a fully connected operation and a softmax function operation on the N-dimensional vector.

[0032] According to another aspect of the embodiments of this application, a noise identification apparatus is provided, comprising:

[0033] The acquisition module is configured to acquire the target slope value set and the target soft thresholding function;

[0034] The generation module is configured to combine each target slope value in the target slope value set with the target soft thresholding function according to a preset order strategy to generate the corresponding slope soft thresholding function.

[0035] The generation module is configured to input noise data into multiple soft threshold residual networks to obtain the noise recognition result output by each soft threshold residual network. The soft threshold residual network is a different soft threshold residual network constructed by each slope soft thresholding function and the residual network.

[0036] The determination module is configured to determine the deployment of a target soft threshold residual network to the noise recognition device, wherein the target soft threshold residual network is the soft threshold residual network corresponding to the noise recognition result with the highest noise recognition rate.

[0037] According to another aspect of the embodiments of this application, an electronic device is provided, comprising:

[0038] Memory, used to store executable instructions;

[0039] A display is provided for displaying the executable instructions to the memory to perform the operation of any of the noise recognition methods described above.

[0040] According to another aspect of the embodiments of this application, a computer-readable storage medium is provided for storing computer-readable instructions, which, when executed, perform the operation of any of the noise recognition methods described above.

[0041] In this application, after obtaining the target slope value set and the target soft thresholding function, a corresponding slope soft thresholding function can be generated for each target slope value in the target slope value set and the target soft thresholding function according to a preset order strategy. Noise data is then input into multiple soft thresholding residual networks. After obtaining the noise recognition result output by each soft thresholding residual network, the soft thresholding residual network corresponding to the noise recognition result with the highest noise recognition rate is deployed to the noise recognition device. By applying the technical solution of this application, the optimal slope can be dynamically calculated from multiple selectable slopes and used as the slope of the deep residual shrinkage network. This allows for the selection of residual networks corresponding to different optimal slopes during the noise reduction process to remove noise redundancy based on different noise signals. This improves the response of the entire network model to external noise signals and enhances the accuracy and precision of noise fault diagnosis.

[0042] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0043] The accompanying drawings, which form part of this specification, illustrate embodiments of this application and, together with the description, serve to explain the principles of this application.

[0044] This application can be more clearly understood with reference to the accompanying drawings and the following detailed description, wherein:

[0045] Figure 1 This is a schematic diagram of a noise identification method proposed in this application;

[0046] Figures 2-8 This is a schematic diagram of the slope-based soft threshold residual network proposed in this application;

[0047] Figure 9 This is a schematic diagram of the noise recognition device of this application;

[0048] Figure 10 This is a schematic diagram of the electronic device structure for noise recognition in this application. Detailed Implementation

[0049] Various exemplary embodiments of this application will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this application.

[0050] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0051] The following description of at least one exemplary embodiment is merely illustrative and is not intended to limit the scope of this application or its application or use.

[0052] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0053] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0054] Furthermore, the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0055] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0056] In one embodiment, this application also proposes a method, apparatus, target terminal, and medium for noise identification.

[0057] Figure 1 A schematic flowchart illustrating a noise identification method according to an embodiment of this application is shown. Figure 1 As shown, the method includes:

[0058] S101, obtain the target slope value set and the target soft thresholding function.

[0059] Furthermore, with the development of technology, mechanization and productization are playing an increasingly important role in the country's overall production, whether for industrial or consumer products. High-quality products inevitably rely on high-quality production processes, which in turn drive continuous improvement in mechanical production techniques. And whether it's production tools or consumer goods, as products are used more, wear and tear is inevitable. Therefore, a precise fault diagnosis system directly determines the quality of a product.

[0060] For example, rotating bearings constitute a significant proportion of many mechanical manufacturing products and everyday consumer goods. Bearing wear is also one of the most common phenomena in both manufacturing and everyday appliance wear. Judging the sound of a rotating bearing is an important method for fault diagnosis. However, the rotation of bearings, whether in production or daily life, usually involves a large amount of noise and redundant signals. Therefore, directly judging the extent of damage by sound can lead to significant errors.

[0061] Currently, the commonly used traditional method employs statistical learning to analyze audio signal noise. However, traditional methods typically involve many parameters that need to be manually set, which is a complex decision-making process usually requiring extensive statistical experiments. However, different environments will determine different parameters; for example, the internal and external differences of the machine itself, the material of the bearing, and even the ambient temperature and humidity can all affect the settings of these hyperparameters. Therefore, these manually set parameters are difficult to standardize. Soft thresholding, as a classic method, is very practical, especially in the field of signal denoising. However, as mentioned earlier, the threshold in the soft thresholding function is a hyperparameter, and setting a reasonable value is a very challenging problem.

[0062] The widespread adoption of the Internet of Things (IoT), big data, and mobile devices in recent years, especially the explosive development of deep learning, has made intelligent detection and recognition technologies based on deep learning possible. Unlike traditional methods, deep learning can automatically learn the parametric features of perturbed signals and automatically derive the correct and appropriate parameters, thus possessing extremely high practical value. Furthermore, the inherent nonlinear properties of soft thresholding are well-suited for use in the computation and propagation processes of deep neural networks. Deep residual networks (ResNet), as a classic deep learning network, have been successfully applied in many fields. Networks combining deep residual networks with nonlinear soft thresholding functions, namely deep residual shrinkage networks (RSNs), have also proven their practicality in signal denoising. RSNs employ an attention mechanism (similar to Squeeze-and-Excitation Networks) to automatically set the threshold, avoiding the hassle of manual threshold setting.

[0063] Therefore, since the aforementioned deep residual shrinkage network using soft thresholding removes all features within the threshold range, it inevitably deletes some effective information features within that range. In this solution, we employ a soft thresholding function with a slope. Networks with this function can learn signal features within the threshold range, which will significantly improve the accuracy and precision of fault diagnosis.

[0064] Furthermore, this application first describes the process of obtaining the target soft thresholding function as follows:

[0065] In deep residual networks (ResNet), the rectified linear unit (ReLU) is a common, general, and efficient nonlinear activation function, mathematically expressed as:

[0066]

[0067] Here, x and y are real numbers, representing the input and output respectively; its derivative is...

[0068]

[0069] Similarly, x and y above are both real numbers, representing the input and output respectively. Furthermore, the function shape and derivative shape of ReLU are as follows: Figure 2 As shown. The corresponding deep residual shrinkage network uses a soft thresholding activation function, mathematically expressed as:

[0070]

[0071] In the above expression, τ is the threshold, and x and y are real numbers representing the input and output, respectively. The shape of the function is as follows: Figure 3 As shown. Its derivative is

[0072]

[0073] Where τ is the threshold, and the derivative has the following shape: Figure 4 As shown. Furthermore, since the soft thresholding function is a function where the value is 0 within the threshold range, and outside the threshold, similar to ReLU, it maintains a slope of 1. This suppresses noise interference within the threshold range and can eliminate easily obtrusive features.

[0074] Furthermore, for slope-based attention-based depth residual shrinkage network design, the entire network design is essentially a series of stacked attention-based slope-based soft thresholding modules (Attention Same Slope Units), abbreviated as ASSU, such as... Figures 5-6 As shown, the generation architecture for a slope-based depth residual shrinkage network is as follows:

[0075] Step one: This application first obtains the input features as CxWx1. It should be noted that in this feature, C corresponds to the output channel, W refers to the feature width, and 1 refers to the feature height, since the feature height in a noisy signal is 1. Since the arrangement of channels may vary depending on the deep learning framework (some frameworks may use Wx1xC), this application places the channel in the first dimension. This input is then saved separately and used for the final residual summation.

[0076] Step 2: This application can perform CBN operation on CxWx1 features. Here, CBN is an abbreviation for Convolution (Con) + Batch Normalization + ReLU operation.

[0077] Step 3: Perform another CBN operation on the above output to extract higher-level semantic information.

[0078] Step four involves applying an attention mechanism that automatically derives a threshold from the output of step two. This part first uses absolute value plus global average pooling (GAP) to obtain the output.

[0079] Step 5: This application can perform a fully connected (FC) + batch normalization + ReLU + fully connected (FC) operation on the output obtained in step 3, and then perform the sigmoid function to obtain the output of the threshold τ.

[0080] Step six involves applying an attention mechanism based on slope-based automatic derivation to the output of step three. It should be noted that this part first uses absolute value + global average pooling (GAP) to obtain the output.

[0081] Step 7: This application can perform a fully connected (FC) + batch normalization + ReLU + fully connected (FC) operation on the output of Step 6 to directly obtain the slope value α.

[0082] Step eight: Based on the threshold τ and slope value α obtained in steps five and seven respectively, this application constructs a slope-based soft thresholding function. The output of this slope-based soft thresholding function is then combined with the original input signal (the input signal in step one), and multiple residual units (IRSBUs) are superimposed to obtain the summed result as the classification output softmax. This completes the design description of a complete slope-based soft thresholding module (ASSU).

[0083] Furthermore, this application allows the slope α automatically derived by the network to be substituted into the soft thresholding function to replace the original soft thresholding function with a slope of 1. The expression for this new function is as follows:

[0084]

[0085] Where τ is the threshold, x and y are real numbers representing the input and output respectively, α is the slope, and the shape of the function is as follows: Figure 7 As shown. The mathematical expression for the corresponding derivative is:

[0086]

[0087] Here, τ is the threshold, x and y are real numbers representing the input and output respectively, α is the slope, and the shape of the function is as follows: Figure 8 As shown.

[0088] S102, following a preset order strategy, each target slope value in the target slope value set is combined with the target soft thresholding function to generate the corresponding slope soft thresholding function.

[0089] Furthermore, for example, if the target slope value set includes slope value 1, slope value 2, and slope value 3, this application can generate a corresponding slope soft thresholding function, i.e., slope soft thresholding function 1, by sequentially generating a corresponding slope soft thresholding function, i.e., slope soft thresholding function 2, by generating a corresponding slope soft thresholding function, i.e., slope soft thresholding function 2, and by generating a corresponding slope soft thresholding function, i.e., slope soft thresholding function 3, ...

[0090] S103, input the noise data into multiple soft threshold residual networks to obtain the noise recognition result output by each soft threshold residual network. The soft threshold residual network is a different soft threshold residual network constructed by each slope soft thresholding function and the residual network.

[0091] Using the same example as above, after generating slope soft thresholding function 1, slope soft thresholding function 2, and slope soft thresholding function 3, this application can use these slope soft thresholding functions and residual networks to generate multiple corresponding soft thresholding residual networks (e.g., soft thresholding residual network 1, soft thresholding residual network 2, and soft thresholding residual network 3). Then, noise data is input into each soft thresholding residual network to generate multiple corresponding noise recognition results.

[0092] Furthermore, the specific steps of generating the corresponding depth residual shrinkage network based on the slope soft thresholding function in this application are as follows:

[0093] Step 1: The input is the original signal with some noise, with dimensions CxWx1. First, a convolution operation is performed to extract semantic information, still with dimensions CxWx1.

[0094] Step two involves several slope-based soft thresholding modules (ASSU) mentioned above, used to remove noise redundancy.

[0095] Step 3: After a batch normalization operation, ReLU, and global average pooling (GAP) are performed, the output is transformed into an Nx1x1 dimension, where N is a hyperparameter representing the channel dimension.

[0096] Step four involves reshaping the Nx1x1 features into a vector of length N. A fully connected (FC) operation is then performed on the N-dimensional vector, followed by a softmax operation, directly outputting the noise recognition result.

[0097] Furthermore, for slope-based deep residual shrinkage networks, after training, these networks can provide accurate fault diagnosis outputs even in noisy environments. Because the network design can automatically derive the same slope based on different noise signals, its accuracy is also higher.

[0098] S104, Determine to deploy the target soft threshold residual network to the noise recognition device. The target soft threshold residual network is the soft threshold residual network corresponding to the noise recognition result with the highest noise recognition rate.

[0099] Understandably, after obtaining all noise recognition results, this application can statistically analyze the test accuracy of all slopes, select the slope with the highest accuracy as the optimal slope value, and then deploy the depth residual shrinkage network corresponding to the optimal slope into the noise recognition device.

[0100] To address the aforementioned issues, this application employs a method that automatically selects the optimal slope from multiple slopes. This dynamically calculates the optimal slope from a pool of candidate slopes and uses it as the slope for the deep residual shrinkage network. This allows for the selection of residual networks corresponding to different optimal slope soft thresholds during the denoising process, based on varying noise signals, to remove noise redundancy. This improves the overall network model's response to external noise signals and enhances the accuracy and precision of noise fault diagnosis.

[0101] Optionally, in one possible implementation of this application, before obtaining the target slope value set, the method further includes:

[0102] Obtain the set of slope values ​​to be selected;

[0103] Multiple target slope values ​​in the set of slope values ​​to be selected are selected at preset intervals.

[0104] The target slope value set is generated based on the plurality of target slope values.

[0105] Optionally, in one possible implementation of this application, the preset order strategy combines each target slope value in the target slope value set with the target soft thresholding function to generate a corresponding slope soft thresholding function, including:

[0106] In order of value size, each target slope value is sequentially paired with the target soft thresholding function to generate a corresponding slope soft thresholding function.

[0107] For example, when the set of slope values ​​to be selected corresponds to the closed interval [0.001, 0.2], this application can perform equal-interval sampling at preset interval units (e.g., 0.05 intervals) and form a sequence of the obtained sampling points as the target slope value set.

[0108] Furthermore, in this embodiment, a slope can be extracted sequentially from smallest to largest, thus determining the soft thresholding function with the slope. These multiple slope soft thresholding functions are then used with the residual network to generate multiple corresponding soft thresholding residual networks (e.g., soft thresholding residual network 1, soft thresholding residual network 2, and soft thresholding residual network 3). Noise data is then input into each soft thresholding residual network to generate multiple corresponding noise recognition results.

[0109] Optionally, in one possible implementation of this application, before S101 (acquiring noise data), the following step is further included:

[0110] Obtain the first input feature, and perform at least two convolution operations on the first input feature, as well as batch regularization and linear rectified function operations to obtain the first output result;

[0111] The absolute value algorithm and global average pooling operation are performed on the first output result to obtain the second output result;

[0112] Based on the second output, the first slope and the second slope are obtained.

[0113] The first input feature can be CxWx1. Here, C refers to the output channel, W refers to the width of the feature, and 1 corresponds to the height of the feature, since the feature height in a noise signal is 1.

[0114] Furthermore, this application can also perform CBN operation on the features of CxWx1, where CBN refers to the operation of convolution (Con) + batch normalization (Batch Normalization) + ReLU linear rectified function operation to obtain the corresponding first output result.

[0115] Optionally, this application may also re-perform the above-mentioned convolution operation (Con) + batch normalization operation + ReLU linear rectified function operation on the first output result, thereby achieving the purpose of extracting higher-level semantic information.

[0116] The next step, after obtaining the first output, is to apply an attention mechanism that automatically derives a threshold from the output. Specifically, this can be done by performing an absolute value algorithm and a global average pooling operation to obtain the corresponding second output.

[0117] Optionally, in one possible implementation of this application, obtaining the first slope and the second slope based on the second output result characteristic includes:

[0118] The second output is processed by convolution, batch normalization, fully connected operation, and linear rectified function operation to obtain the third output.

[0119] The third output result is sigmoidized to obtain the target soft thresholding function.

[0120] Furthermore, after obtaining the corresponding second output result, this application can perform a fully connected (FC) + batch normalization + ReLU linear rectified function transformation + fully connected (FC) operation on the output result to obtain the corresponding third output result. This allows subsequent sigmoid function transformation of the third output result to obtain the target soft thresholding function τ.

[0121] Optionally, in one possible implementation of this application, noise data is input into multiple soft-threshold residual networks to obtain the noise identification result output by each soft-threshold residual network, including:

[0122] Acquire noisy data, perform convolution on the noisy data, extract semantic information from the noisy data, and obtain the output result of the first dimension;

[0123] The slope soft thresholding function is used to remove noise redundancy from the output of the first dimension;

[0124] Batch regularization, global average pooling, and linear rectified function transformation are performed on the first dimension output to remove noise redundancy, resulting in the second dimension output.

[0125] The noise recognition result is obtained based on the output of the second dimension.

[0126] Optionally, in one possible implementation of this application, the noise identification result is obtained based on the output result of the second dimension, including:

[0127] Perform a transformation operation to convert the output of the second dimension into an N-dimensional vector;

[0128] Perform a fully connected operation on the N-dimensional vector, and then perform a softmax function operation to obtain the noise recognition result.

[0129] Furthermore, this application can acquire noisy data and perform convolution operations on the noisy data to extract semantic information from the noisy data, obtaining the first-dimensional output result. For example, the input is the original signal with noise, with a dimension of CxWx1. First, a convolution operation is performed to extract semantic information once, still with a dimension of CxWx1. Then, it passes through several soft thresholding modules (ADSU) with different slopes mentioned above to remove noise redundancy.

[0130] Furthermore, this application transforms the output of the first dimension (after removing noise redundancy) into an Nx1x1 dimension through batch normalization, ReLU, and global average pooling (GAP), where N is a hyperparameter representing the channel dimension. This yields the output of the second dimension.

[0131] Furthermore, this application can also transform the Nx1x1 feature of the second dimension output into a vector of length N through a reshape operation, perform a fully connected (FC) operation on the N-dimensional vector, and then perform a softmax operation to directly output the final noise recognition result.

[0132] In this application, after obtaining the target slope value set and the target soft thresholding function, a corresponding slope soft thresholding function can be generated for each target slope value in the target slope value set and the target soft thresholding function according to a preset order strategy. Noise data is then input into multiple soft thresholding residual networks. After obtaining the noise recognition result output by each soft thresholding residual network, the soft thresholding residual network corresponding to the noise recognition result with the highest noise recognition rate is deployed to the noise recognition device. By applying the technical solution of this application, the optimal slope can be dynamically calculated from multiple selectable slopes and used as the slope of the deep residual shrinkage network. This allows for the selection of residual networks corresponding to different optimal slopes during the noise reduction process to remove noise redundancy based on different noise signals. This improves the response of the entire network model to external noise signals and enhances the accuracy and precision of noise fault diagnosis.

[0133] In another embodiment of this application, such as Figure 9 As shown, this application also provides a noise recognition device. The device includes an acquisition module 201, a generation module 202, and a determination module 203, wherein...

[0134] The acquisition module 201 is configured to acquire the target slope value set and the target soft thresholding function;

[0135] The generation module 202 is configured to combine each target slope value in the target slope value set with the target soft thresholding function according to a preset order strategy to generate the corresponding slope soft thresholding function.

[0136] The generation module 202 is configured to input noise data into multiple soft threshold residual networks to obtain the noise recognition result output by each soft threshold residual network. The soft threshold residual network is a different soft threshold residual network constructed by each slope soft thresholding function and the residual network.

[0137] The determining module 203 is configured to determine to deploy a target soft threshold residual network to the noise recognition device, wherein the target soft threshold residual network is the soft threshold residual network corresponding to the noise recognition result with the highest noise recognition rate.

[0138] By applying the technical solution of this application, the optimal slope can be dynamically calculated from multiple selectable slopes and used as the slope of the deep residual shrinkage network. This allows for the selection of residual networks corresponding to different optimal slope soft thresholds during the denoising process, based on different noise signals, to remove noise redundancy. This improves the overall network model's response to external noise signals and enhances the accuracy and precision of noise fault diagnosis.

[0139] In another embodiment of this application, the acquisition module 201 further includes:

[0140] Module 201 is configured to acquire a set of slope values ​​to be selected.

[0141] The acquisition module 201 is configured to select multiple target slope values ​​from the set of slope values ​​to be selected at preset intervals.

[0142] The acquisition module 201 is configured to generate the target slope value set based on the plurality of target slope values.

[0143] In another embodiment of this application, the acquisition module 201 further includes:

[0144] The acquisition module 201 is configured to sequentially generate a corresponding slope soft thresholding function by matching each target slope value with the target soft thresholding function in order of value size.

[0145] In another embodiment of this application, the acquisition module 201 is configured to further include:

[0146] The acquisition module 201 is configured to acquire a first input feature, and perform at least two convolution operations on the first input feature, as well as perform batch regularization and linear rectified function operations to obtain a first output result;

[0147] The acquisition module 201 is configured to perform an absolute value algorithm on the first output result and to perform a global average pooling operation to obtain a second output result;

[0148] The acquisition module 201 is configured to acquire the target soft thresholding function based on the second output result.

[0149] In another embodiment of this application, the acquisition module 201 further includes:

[0150] The acquisition module 201 is configured to perform convolution operation, batch regularization operation, fully connected operation, and linear rectified function operation on the second output result to obtain the third output result;

[0151] The acquisition module 201 is configured to perform sigmoid functionalization on the third output result to obtain the target soft thresholding function.

[0152] In another embodiment of this application, the acquisition module 201 further includes:

[0153] The acquisition module 201 is configured to acquire the noise data, perform a convolution operation on the noise data, extract semantic information from the noise data, and obtain the output result of the first dimension.

[0154] The acquisition module 201 is configured to use the slope soft thresholding function to remove noise redundancy from the output result of the first dimension;

[0155] The acquisition module 201 is configured to perform batch regularization, global average pooling, and linear rectified functionization on the output of the first dimension after removing noise redundancy, to obtain the output of the second dimension.

[0156] The acquisition module 201 is configured to obtain the noise recognition result based on the output result of the second dimension.

[0157] In another embodiment of this application, the acquisition module 201 further includes:

[0158] The acquisition module 201 is configured to perform a transformation operation, converting the output of the second dimension into an N-dimensional vector;

[0159] The acquisition module 201 is configured to perform a fully connected operation and a softmax function operation on the N-dimensional vector to obtain the noise recognition result.

[0160] Figure 10This is a logical structure block diagram of an electronic device according to an exemplary embodiment. For example, the electronic device 300 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0161] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory including instructions. These instructions can be executed by a processor of an electronic device to complete the network monitoring method described above. The method includes: acquiring a set of target slope values ​​and a target soft thresholding function; combining each target slope value in the set of target slope values ​​with the target soft thresholding function according to a preset order strategy to generate a corresponding slope soft thresholding function; inputting noise data into multiple soft thresholding residual networks to obtain a noise recognition result output by each soft thresholding residual network, wherein the soft thresholding residual network is a different soft thresholding residual network constructed by each slope soft thresholding function and the residual network; and determining to deploy a target soft thresholding residual network to a noise recognition device, wherein the target soft thresholding residual network is the soft thresholding residual network corresponding to the noise recognition result with the highest noise recognition rate. Optionally, the instructions can also be executed by a processor of an electronic device to complete other steps involved in the exemplary embodiment described above. For example, the non-transitory computer-readable storage medium can be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0162] In an exemplary embodiment, an application / computer program product is also provided, including one or more instructions that can be executed by a processor of an electronic device to complete the aforementioned network monitoring method. The method includes: acquiring a set of target slope values ​​and a target soft thresholding function; combining each target slope value in the set of target slope values ​​with the target soft thresholding function according to a preset order strategy to generate a corresponding slope soft thresholding function; inputting noise data into multiple soft thresholding residual networks to obtain a noise recognition result output by each soft thresholding residual network, wherein the soft thresholding residual network is a different soft thresholding residual network constructed by each slope soft thresholding function and the residual network; and determining to deploy a target soft thresholding residual network to a noise recognition device, wherein the target soft thresholding residual network is the soft thresholding residual network corresponding to the noise recognition result with the highest noise recognition rate. Optionally, the above instructions can also be executed by a processor of an electronic device to complete other steps involved in the above exemplary embodiment.

[0163] Figure 10 This is an example diagram of computer device 30. Those skilled in the art will understand that it is illustrative. Figure 10This is merely an example of computer device 30 and does not constitute a limitation on computer device 30. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device 30 may also include input / output devices, network access devices, buses, etc.

[0164] The processor 302 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or processor 302 may be any conventional processor. Processor 302 is the control center of the computer device 30, connecting all parts of the computer device 30 via various interfaces and lines.

[0165] The memory 301 can be used to store computer-readable instructions 303. The processor 302 implements various functions of the computer device 30 by running or executing the computer-readable instructions or modules stored in the memory 301 and calling the data stored in the memory 301. The memory 301 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device 30, etc. In addition, the memory 301 may include a hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, read-only memory (ROM), random access memory (RAM), or other non-volatile / volatile storage devices.

[0166] If the modules integrated in the computer device 30 are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by instructing related hardware through computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium, and when executed by a processor, the computer-readable instructions can implement the steps of the various method embodiments described above.

[0167] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0168] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for noise identification, characterized in that, include: Obtain the target slope value set and the target soft thresholding function; According to a preset order strategy, each target slope value in the target slope value set is combined with the target soft thresholding function to generate the corresponding slope soft thresholding function. Noise data is input into multiple soft-threshold residual networks to obtain the noise recognition result output by each soft-threshold residual network. The soft-threshold residual network is a different soft-threshold residual network constructed by each slope soft-thresholding function and the residual network. The target soft threshold residual network is determined to be deployed on the noise recognition device, wherein the target soft threshold residual network is the soft threshold residual network corresponding to the noise recognition result with the highest noise recognition rate; The process involves inputting noisy data into multiple soft-threshold residual networks to obtain a noise recognition result output by each network. This includes: acquiring noisy data and performing a convolution operation on the data to extract semantic information and obtain a first-dimensional output; removing noise redundancy from the first-dimensional output using the slope soft-thresholding function; performing batch regularization, global average pooling, and linear rectified function operations on the noise-redundant-removed first-dimensional output to obtain a second-dimensional output; and obtaining the noise recognition result based on the second-dimensional output.

2. The method as described in claim 1, characterized in that, Before obtaining the target slope value set, the method further includes: Obtain the set of slope values ​​to be selected; Multiple target slope values ​​in the set of slope values ​​to be selected are selected at preset intervals. The target slope value set is generated based on the plurality of target slope values.

3. The method as described in claim 1, characterized in that, The step of combining each target slope value in the target slope value set with the target soft thresholding function according to a preset order strategy to generate a corresponding slope soft thresholding function includes: In order of value size, each target slope value is sequentially combined with the target soft thresholding function to generate the corresponding slope soft thresholding function.

4. The method as described in claim 1, characterized in that, The process of obtaining the target soft thresholding function includes: Obtain the first input feature, and perform at least two convolution operations on the first input feature, as well as batch regularization and linear rectified function operations to obtain the first output result; The absolute value algorithm and global average pooling operation are performed on the first output result to obtain the second output result; Based on the second output result, the target soft thresholding function is obtained.

5. The method as described in claim 4, characterized in that, Based on the second output result, the target soft thresholding function is obtained, including: The second output result is subjected to convolution, batch regularization, fully connected operation, and linear rectified function operation to obtain the third output result. The third output result is sigmoidized to obtain the target soft thresholding function.

6. The method as described in claim 5, characterized in that, The noise recognition result obtained based on the output result of the second dimension includes: Perform a transformation operation to convert the output of the second dimension into an N-dimensional vector; The noise recognition result is obtained by performing a fully connected operation and a softmax function operation on the N-dimensional vector.

7. A noise identification device, characterized in that, include: The acquisition module is configured to acquire the target slope value set and the target soft thresholding function; The generation module is configured to combine each target slope value in the target slope value set with the target soft thresholding function according to a preset order strategy to generate the corresponding slope soft thresholding function. The generation module is configured to input noise data into multiple soft threshold residual networks to obtain the noise recognition result output by each soft threshold residual network. The soft threshold residual network is a different soft threshold residual network constructed by each slope soft thresholding function and the residual network. In addition, noise data is acquired, and a convolution operation is performed on the noise data to extract semantic information from the noise data, thereby obtaining the output result of the first dimension; the noise redundancy of the output result of the first dimension is removed using the slope soft thresholding function. Batch regularization, global average pooling, and linear rectified function transformation are performed on the first dimension output to remove noise redundancy, resulting in the second dimension output. Based on the output of the second dimension, the noise recognition result is obtained; The determination module is configured to determine the deployment of a target soft threshold residual network to the noise recognition device, wherein the target soft threshold residual network is the soft threshold residual network corresponding to the noise recognition result with the highest noise recognition rate.

8. An electronic device, characterized in that, include: Memory, used to store executable instructions; A processor configured to interact with the memory to execute the executable instructions to perform the operation of the noise identification method according to any one of claims 1-6.

9. A computer-readable storage medium for storing computer-readable instructions, characterized in that, When the instruction is executed, it performs the operation of the noise recognition method according to any one of claims 1-6.

Images