Method and system for verifying that enterprises have not implemented suspension measures based on online monitoring data

Abstract

The present application discloses a method and system for verifying whether an enterprise has not implemented suspension measures based on online monitoring data. The method uses the ambient sound signal collected in real time by the first sound sensor and the sound signal near the monitored enterprise object collected in real time by the second sound sensor, and uses the signal processing and analysis algorithm based on the deep learning neural network to perform time-frequency feature analysis on the ambient sound signal and the sound signal near the monitored enterprise object, so as to intelligently judge whether the monitored enterprise object has not implemented suspension measures based on the time-frequency features of the nearby sound signal after filtering out the ambient sound. In this way, the sound signal of the monitored enterprise can be monitored online, and the production and operation status of the enterprise can be understood in a timely manner, reducing the workload and subjectivity of manual verification, thereby improving the accuracy and efficiency of the verification results.

Description

Technical Field

[0001] This application relates to the field of intelligent management, and more specifically, to a method and system for verifying that an enterprise has not implemented production suspension measures based on online monitoring data. Background Art

[0002] Verifying that an enterprise has not implemented production suspension measures refers to the inspection by relevant regulatory agencies of whether an enterprise has implemented emergency emission reduction measures under specific circumstances, such as during a heavy pollution weather warning. If an enterprise fails to comply with these regulations, it may pose a risk to public health and bear corresponding legal consequences. Such verification is an important measure in environmental management, aiming to ensure that enterprises take appropriate countermeasures when facing environmental risks, reduce the impact on the environment and protect public health, and at the same time maintain product and service quality standards when resuming production.

[0003] However, the traditional verification of whether an enterprise has implemented production suspension measures usually relies on manual verification and supervision, which is easily affected by human factors, such as the subjective judgment of supervisors and the negligence of employees, resulting in situations such as missed inspections and reports, thus affecting the accuracy and reliability of verification results. Moreover, traditional methods usually require a large amount of manpower and time for verification and supervision, with low efficiency, which is not conducive to the enterprise's rapid response and problem handling.

[0004] Therefore, there is a need for a method for verifying that an enterprise has not implemented production suspension measures based on online monitoring data. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide a method and system for verifying that an enterprise has not implemented production suspension measures based on online monitoring data. It collects the environmental sound signal collected by the first sound sensor in real time and the sound signal near the monitored enterprise object collected by the second sound sensor in real time, and uses a signal processing and analysis algorithm based on a deep learning neural network to perform time-frequency feature analysis on the environmental sound signal and the sound signal near the monitored enterprise object. Based on this, it intelligently judges whether the monitored enterprise object has not implemented production suspension measures based on the time-frequency features of the nearby sound signal after filtering out the environmental sound. In this way, the sound signal of the monitored enterprise can be monitored online, and the production and operation status of the enterprise can be understood in a timely manner, reducing the workload and subjectivity of manual verification, thereby improving the accuracy and efficiency of verification results.

[0006] According to one aspect of this application, there is provided a method for verifying that an enterprise has not implemented production suspension measures based on online monitoring data, which includes:

[0007] Obtaining the environmental sound signal collected by the first sound sensor and the sound signal near the monitored enterprise object collected by the second sound sensor;

[0008] Perform wavelet transform on the nearby acoustic signal and the ambient acoustic signal of the monitored enterprise object to obtain the time-frequency image of the nearby acoustic signal and the time-frequency image of the ambient acoustic signal;

[0009] Extract the time-frequency features of the acoustic signal from the time-frequency image of the nearby acoustic signal and the time-frequency image of the ambient acoustic signal respectively to obtain the time-frequency feature matrix of the nearby acoustic signal and the time-frequency feature matrix of the ambient acoustic signal;

[0010] Calculate the difference between the time-frequency feature matrix of the nearby acoustic signal and the time-frequency feature matrix of the ambient acoustic signal to obtain the time-frequency feature matrix of the nearby acoustic signal filtered by the ambient sound;

[0011] Pass the time-frequency feature matrix of the nearby acoustic signal filtered by the ambient sound through the important component saliency module based on the class foreground attention mechanism to obtain the saliency time-frequency feature matrix of the nearby acoustic signal filtered by the ambient sound as the saliency time-frequency feature of the nearby acoustic signal filtered by the ambient sound;

[0012] Obtain the inspection result based on the saliency time-frequency feature of the nearby acoustic signal filtered by the ambient sound.

[0013] In the above verification method for enterprises not implementing production suspension measures based on online monitoring data, extracting the time-frequency features of the acoustic signal from the time-frequency image of the nearby acoustic signal and the time-frequency image of the ambient acoustic signal respectively to obtain the time-frequency feature matrix of the nearby acoustic signal and the time-frequency feature matrix of the ambient acoustic signal includes: passing the time-frequency image of the nearby acoustic signal and the time-frequency image of the ambient acoustic signal through the acoustic signal time-frequency feature extractor based on the GoogLeNet model respectively to obtain the time-frequency feature matrix of the nearby acoustic signal and the time-frequency feature matrix of the ambient acoustic signal.

[0014] In the above verification method for enterprises not implementing production suspension measures based on online monitoring data, calculating the difference between the time-frequency feature matrix of the nearby acoustic signal and the time-frequency feature matrix of the ambient acoustic signal to obtain the time-frequency feature matrix of the nearby acoustic signal filtered by the ambient sound includes: calculating the position-wise difference between the time-frequency feature matrix of the nearby acoustic signal and the time-frequency feature matrix of the ambient acoustic signal to obtain the time-frequency feature matrix of the nearby acoustic signal filtered by the ambient sound.

[0015] In the above verification method for enterprises not implementing production suspension measures based on online monitoring data, the time-frequency feature matrix of the ambient sound filtered nearby sound signals is passed through an important component saliency module based on a class foreground attention mechanism to obtain a saliency time-frequency feature matrix of the ambient sound filtered nearby sound signals as the saliency time-frequency feature of the ambient sound filtered nearby sound signals, including: performing feature representation on the time-frequency feature matrix of the ambient sound filtered nearby sound signals to obtain a time-frequency representation matrix of the ambient sound filtered nearby sound signals; performing masking processing on the time-frequency representation matrix of the ambient sound filtered nearby sound signals to obtain a time-frequency mask weight matrix of the ambient sound filtered nearby sound signals; and multiplying the time-frequency mask weight matrix of the ambient sound filtered nearby sound signals and the time-frequency feature matrix of the ambient sound filtered nearby sound signals at each position point to obtain the saliency time-frequency feature matrix of the ambient sound filtered nearby sound signals.

[0016] In the above verification method for enterprises not implementing production suspension measures based on online monitoring data, performing feature representation on the time-frequency feature matrix of the ambient sound filtered nearby sound signals to obtain a time-frequency representation matrix of the ambient sound filtered nearby sound signals, including: taking the negative of each position eigenvalue of the time-frequency feature matrix of the ambient sound filtered nearby sound signals as the exponent of the natural constant to calculate the exponential function value with the natural constant as the base at each position to obtain a time-frequency class support feature matrix of the ambient sound filtered nearby sound signals; and calculating the reciprocal of the sum of each position eigenvalue and the constant one in the time-frequency class support feature matrix of the ambient sound filtered nearby sound signals to obtain the time-frequency representation matrix of the ambient sound filtered nearby sound signals.

[0017] In the above verification method for enterprises not implementing production suspension measures based on online monitoring data, performing masking processing on the time-frequency representation matrix of the ambient sound filtered nearby sound signals to obtain a time-frequency mask weight matrix of the ambient sound filtered nearby sound signals, including: setting the eigenvalues greater than or equal to a predetermined threshold at each position of the time-frequency representation matrix of the ambient sound filtered nearby sound signals to one, and the rest to zero to obtain the time-frequency mask weight matrix of the ambient sound filtered nearby sound signals.

[0018] In the above verification method for enterprises not implementing production suspension measures based on online monitoring data, based on the saliency time-frequency feature of the ambient sound filtered nearby sound signals, obtaining an inspection result, including: passing the saliency time-frequency feature matrix of the ambient sound filtered nearby sound signals through an inspector based on a classifier to obtain an inspection result, and the inspection result is used to indicate whether the monitored enterprise object has not implemented the production suspension measures.

[0019] In the above verification method for enterprises not implementing production suspension measures based on online monitoring data, it also includes training for the time-frequency feature extractor of the sound signal based on the GoogLeNet model, the important component saliency module based on the class foreground attention mechanism, and the inspector based on the classifier.

[0020] In the above verification method for the enterprise's failure to implement production suspension measures based on online monitoring data, the training step includes: obtaining training data, where the training data includes the training environmental sound signals collected by the first sound sensor and the sound signals near the monitored enterprise object collected by the second sound sensor, and the true value of whether the monitored enterprise fails to implement production suspension measures; performing wavelet transform on the sound signals near the monitored enterprise object and the training environmental sound signals to obtain the time-frequency images of the sound signals near the training and the time-frequency images of the training environmental sound signals; respectively passing the time-frequency images of the sound signals near the training and the time-frequency images of the training environmental sound signals through the sound signal time-frequency feature extractor based on the GoogLeNet model to obtain the time-frequency feature matrix of the sound signals near the training and the time-frequency feature matrix of the training environmental sound signals; calculating the position-wise difference between the time-frequency feature matrix of the sound signals near the training and the time-frequency feature matrix of the training environmental sound signals to obtain the time-frequency feature matrix of the environmental sound filtering out the sound signals near the training; passing the time-frequency feature matrix of the environmental sound filtering out the sound signals near the training through the important component saliency module based on the class foreground attention mechanism to obtain the time-frequency feature matrix of the training significant environmental sound filtering out the sound signals near the training; passing the time-frequency feature matrix of the training significant environmental sound filtering out the sound signals near the training through the inspector based on the classifier to obtain the classification loss function value; and training the sound signal time-frequency feature extractor based on the GoogLeNet model, the important component saliency module based on the class foreground attention mechanism, and the inspector based on the classifier based on the classification loss function value and through the backpropagation of gradient descent.

[0021] According to another aspect of the present application, there is provided a verification system for an enterprise's failure to implement production suspension measures based on online monitoring data, which includes:

[0022] A sound signal acquisition module, configured to acquire environmental sound signals collected by the first sound sensor and sound signals near the monitored enterprise object collected by the second sound sensor;

[0023] A wavelet transform module, configured to perform wavelet transform on the sound signals near the monitored enterprise object and the environmental sound signals to obtain the time-frequency images of the sound signals near the training and the time-frequency images of the environmental sound signals;

[0024] A sound signal time-frequency feature extraction module, configured to respectively perform sound signal time-frequency feature extraction on the time-frequency images of the sound signals near the training and the time-frequency images of the environmental sound signals to obtain the time-frequency feature matrix of the sound signals near the training and the time-frequency feature matrix of the environmental sound signals;

[0025] A sound signal difference calculation module, configured to calculate the difference between the time-frequency feature matrix of the sound signals near the training and the time-frequency feature matrix of the environmental sound signals to obtain the time-frequency feature matrix of the environmental sound filtering out the sound signals near the training;

[0026] An important component saliency module, which is used to obtain a saliency environmental sound filtered nearby sound signal time-frequency feature matrix as the saliency environmental sound filtered nearby sound signal time-frequency feature by passing the time-frequency feature matrix of the environmental sound filtered nearby sound signal through the important component saliency module based on the class foreground attention mechanism;

[0027] An inspection result generation module, which is used to obtain an inspection result based on the saliency environmental sound filtered nearby sound signal time-frequency feature.

[0028] Compared with the prior art, an enterprise non-implementation of production suspension measure verification method and system based on online monitoring data provided by the present application collects, through an environmental sound signal collected in real time by a first sound sensor and a nearby sound signal of a monitored enterprise object collected in real time by a second sound sensor, and uses a signal processing and analysis algorithm based on a deep learning neural network to perform time-frequency feature analysis on the environmental sound signal and the nearby sound signal of the monitored enterprise object, so as to intelligently judge whether the monitored enterprise object fails to implement the production suspension measure based on the time-frequency feature of the nearby sound signal filtering out the environmental sound. In this way, the sound signal of the monitored enterprise can be monitored online, and the production and operation status of the enterprise can be understood in time, reducing the workload and subjectivity of manual verification, thereby improving the accuracy and efficiency of the verification result. Description of the Drawings

[0029] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0030] Figure 1 It is a flowchart of an enterprise non-implementation of production suspension measure verification method based on online monitoring data according to an embodiment of the present application.

[0031] Figure 2 It is a schematic structural diagram of an enterprise non-implementation of production suspension measure verification method based on online monitoring data according to an embodiment of the present application.

[0032] Figure 3 It is a flowchart of obtaining a saliency environmental sound filtered nearby sound signal time-frequency feature matrix as the saliency environmental sound filtered nearby sound signal time-frequency feature by passing the time-frequency feature matrix of the environmental sound filtered nearby sound signal through the important component saliency module based on the class foreground attention mechanism in the enterprise non-implementation of production suspension measure verification method based on online monitoring data according to an embodiment of the present application.

[0033] Figure 4It is a flowchart for training the time-frequency feature extractor of acoustic signals based on the GoogLeNet model, the significant component saliency module based on the class foreground attention mechanism, and the inspector based on the classifier in the enterprise's failure to implement production suspension measures verification method according to the embodiments of the present application.

[0034] Figure 5 It is a block diagram of the enterprise's failure to implement production suspension measures verification system based on online monitoring data according to the embodiments of the present application. Detailed implementation manners

[0035] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Instead, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0036] It should be understood that the various steps recorded in the method embodiments of the present disclosure can be executed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this regard.

[0037] In the description of the embodiments of the present disclosure, the term "including" and its similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. There may also be other explicit and implicit definitions below.

[0038] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly specified in the context, it should be understood as "one or more".

[0039] The verification of whether an enterprise has implemented production suspension measures refers to the inspection by relevant regulatory agencies of whether an enterprise has implemented emergency emission reduction measures under specific circumstances, such as during heavy pollution weather warnings. For example, during heavy pollution weather, in order to protect the environment and public health, regulatory agencies will require certain industrial enterprises to take measures to limit production or completely suspend production. If an enterprise fails to comply with these regulations, it may pose a risk to public health and bear corresponding legal consequences. This verification is an important measure in environmental management, aiming to ensure that enterprises take appropriate response measures when facing environmental risks, reduce the impact on the environment and protect public health, and at the same time maintain the quality standards of products and services when resuming production.

[0040] However, the traditional verification of whether an enterprise has implemented production suspension measures usually relies on manual verification and supervision, which is easily affected by human factors, such as the subjective judgment of supervisors and the negligence of employees, resulting in situations such as missed inspections and reports, thus affecting the accuracy and reliability of verification results. Moreover, traditional methods usually require a large amount of manpower and time for verification and supervision, with low efficiency, which is not conducive to the enterprise's quick response and problem handling.

[0041] Therefore, to address the above technical problems, the technical concept of this application is to collect the ambient sound signal in real time by the first sound sensor and the sound signal near the monitored enterprise object in real time by the second sound sensor, and use the signal processing and analysis algorithm based on the deep learning neural network to perform time-frequency feature analysis on the ambient sound signal and the sound signal near the monitored enterprise object, so as to intelligently judge whether the monitored enterprise object has failed to implement production suspension measures based on the time-frequency features of the nearby sound signal after filtering out the ambient sound. In this way, the sound signal of the monitored enterprise can be monitored online, and the production and operation status of the enterprise can be understood in a timely manner, reducing the workload and subjectivity of manual verification, thereby improving the accuracy and efficiency of verification results.

[0042] Figure 1 It is a flowchart of the method for verifying whether an enterprise has failed to implement production suspension measures based on online monitoring data according to an embodiment of the present application. Figure 2 It is a schematic diagram of the architecture of the method for verifying whether an enterprise has failed to implement production suspension measures based on online monitoring data according to an embodiment of the present application. As Figure 1 and Figure 2As shown in the figure, the verification method for enterprises not implementing production suspension measures based on online monitoring data according to an embodiment of the present application includes: S110, obtaining an environmental sound signal collected by a first sound sensor and a sound signal near a monitored enterprise object collected by a second sound sensor; S120, performing wavelet transform on the sound signal near the monitored enterprise object and the environmental sound signal to obtain a time-frequency image of the sound signal near the object and a time-frequency image of the environmental sound signal; S130, respectively performing sound signal time-frequency feature extraction on the time-frequency image of the sound signal near the object and the time-frequency image of the environmental sound signal to obtain a time-frequency feature matrix of the sound signal near the object and a time-frequency feature matrix of the environmental sound signal; S140, calculating the difference between the time-frequency feature matrix of the sound signal near the object and the time-frequency feature matrix of the environmental sound signal to obtain a time-frequency feature matrix of the environmental sound-filtered near-object sound signal; S150, passing the time-frequency feature matrix of the environmental sound-filtered near-object sound signal through a significant component saliency module based on a class foreground attention mechanism to obtain a saliency time-frequency feature matrix of the environmental sound-filtered near-object sound signal as the saliency time-frequency feature of the environmental sound-filtered near-object sound signal; and S160, obtaining an inspection result based on the saliency time-frequency feature of the environmental sound-filtered near-object sound signal.

[0043] In step S110, an environmental sound signal collected by a first sound sensor and a sound signal near a monitored enterprise object collected by a second sound sensor are obtained. It should be understood that the environmental sound signal refers to the sound data collected in real time by the first sound sensor, which represents the overall sound environment within the monitoring area, and this may include natural environmental noises (such as wind sounds, rain sounds), traffic noises, and other sounds from non-specific sources. The sound signal near the monitored enterprise object refers to the sound data collected by the second sound sensor within a specific enterprise or factory area, and these sounds may be directly related to the production and operation activities of the enterprise, such as mechanical operation sounds, operation sounds, etc. Based on this, in order to distinguish the sounds generated by the enterprise's production activities from the background noise of the environment, so as to improve the accuracy of the inspection result, in the technical solution of the present application, obtaining the environmental sound signal collected by the first sound sensor and the sound signal near the monitored enterprise object collected by the second sound sensor, and performing signal analysis and processing on them can intelligently identify the sound patterns related to production activities from the noise, thereby improving the credibility of the inspection result.

[0044] In step S120, wavelet transform is performed on the acoustic signal near the monitored enterprise object and the environmental acoustic signal to obtain the time-frequency image of the acoustic signal near the monitored enterprise object and the time-frequency image of the environmental acoustic signal. Correspondingly, considering that both the acoustic signal near the monitored enterprise object and the environmental acoustic signal are time-domain signals, their time-varying characteristics can be revealed through time-frequency analysis, and there are local time-frequency characteristic information about the acoustic signal near the monitored enterprise object and the environmental acoustic signal respectively. Based on this, in order to more clearly capture the information of the acoustic signal near the monitored enterprise object and the environmental acoustic signal in terms of local characteristics, in the technical solution of the present application, wavelet transform is performed on the acoustic signal near the monitored enterprise object and the environmental acoustic signal to obtain the time-frequency image of the acoustic signal near the monitored enterprise object and the time-frequency image of the environmental acoustic signal. It is worth mentioning that wavelet transform is a mathematical tool that can provide both time-domain and frequency-domain information, and is suitable for processing non-stationary signals and signals with strong locality, such as sudden changes and anomalies that may exist in sound signals. It can perform local analysis of signals in the time-frequency domain to better capture the instantaneous characteristics of signals. That is to say, through wavelet transform, it is possible to better capture the local time-frequency characteristics and instantaneous change information in the acoustic signal near the monitored enterprise object and the environmental acoustic signal, providing support for the subsequent judgment of whether the monitored enterprise object has not implemented the production suspension measures.

[0045] In step S130, time-frequency feature extraction is respectively performed on the time-frequency image of the nearby sound signal and the time-frequency image of the ambient sound signal to obtain a time-frequency feature matrix of the nearby sound signal and a time-frequency feature matrix of the ambient sound signal. Specifically, in the embodiment of the present application, performing time-frequency feature extraction on the time-frequency image of the nearby sound signal and the time-frequency image of the ambient sound signal to obtain a time-frequency feature matrix of the nearby sound signal and a time-frequency feature matrix of the ambient sound signal includes: respectively passing the time-frequency image of the nearby sound signal and the time-frequency image of the ambient sound signal through a time-frequency feature extractor of the sound signal based on the GoogLeNet model to obtain the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal. It should be understood that considering that both the time-frequency image of the nearby sound signal and the time-frequency image of the ambient sound signal contain more high-level feature information hidden in the sound near the enterprise and the ambient sound, such as the distribution of different frequency components in the sound signal, including information such as the main frequency, frequency range, and frequency change, and the time-domain features of the sound signal include information such as the waveform, amplitude, and duration of the sound signal. These hidden high-level feature information plays an important role in the processing and analysis of sound signals. Considering that GoogLeNet is a deep convolutional neural network model with strong feature extraction ability, it can learn hidden higher-level and more abstract feature representations from complex time-frequency images of sound signals, which helps to distinguish different types of signals. Therefore, in the technical solution of the present application, the time-frequency image of the nearby sound signal and the time-frequency image of the ambient sound signal are respectively passed through a time-frequency feature extractor of the sound signal based on the GoogLeNet model to respectively capture and mine the time-frequency high-level features of the hidden nearby sound signal and ambient sound signal, so as to obtain a time-frequency feature matrix of the nearby sound signal and a time-frequency feature matrix of the ambient sound signal that can better represent the internal structure and feature information of the data.

[0046] In step S140, the difference between the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal is calculated to obtain the time-frequency feature matrix of the ambient sound-filtered nearby sound signal. Specifically, in the embodiment of the present application, calculating the difference between the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal to obtain the time-frequency feature matrix of the ambient sound-filtered nearby sound signal includes: calculating the position-wise difference between the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal to obtain the time-frequency feature matrix of the ambient sound-filtered nearby sound signal. Correspondingly, considering that the time-frequency feature matrix of the nearby sound signal reflects the time-domain and frequency-domain characteristics of the sound signals generated around the monitored enterprise, these characteristics can reflect the operation status of the enterprise, the working conditions of the equipment, and the possible abnormal sounds. The time-frequency feature matrix of the ambient sound signal reflects the sound characteristics of the surrounding environment, including natural sounds, urban noises, etc., and these characteristics can help identify the background noise level and characteristics of the environment. Therefore, in order to distinguish the ambient sound from the sound near the monitored enterprise so as to extract the target sound from the complex sound background, in the technical solution of the present application, the position-wise difference between the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal is calculated to obtain the time-frequency feature matrix of the ambient sound-filtered nearby sound signal. That is to say, there may be some common time-frequency characteristics between the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal, such as frequency range, amplitude, etc. By calculating the position-wise difference between the two, the sound signal components that are similar to the ambient noise or masked by the ambient noise can be effectively filtered out, so as to better highlight and retain the characteristics of the nearby sound signal.

[0047] In step S150, the time-frequency feature matrix of the ambient sound-filtered nearby sound signal is passed through an important component saliency module based on a class foreground attention mechanism to obtain a saliency ambient sound-filtered nearby sound signal time-frequency feature matrix as the saliency ambient sound-filtered nearby sound signal time-frequency feature. It should be understood that considering that the time-frequency feature matrix of the ambient sound-filtered nearby sound signal is obtained by the position-wise difference between the nearby sound signal time-frequency feature matrix and the ambient sound signal time-frequency feature matrix, although the interference of noise in the ambient sound is reduced, there is still a small amount of interference from background ambient information in the nearby sound. Therefore, in order to more significantly distinguish the target nearby sound and the background sound in the time-frequency feature matrix of the ambient sound-filtered nearby sound signal, so as to more accurately highlight the important salient feature information in the nearby sound, in the technical solution of this application, the time-frequency feature matrix of the ambient sound-filtered nearby sound signal is passed through an important component saliency module based on a class foreground attention mechanism to obtain a saliency ambient sound-filtered nearby sound signal time-frequency feature matrix. It is worth mentioning that the class foreground attention mechanism can help the model better distinguish the foreground (target) and background (environment) in the signal, thereby highlighting the important foreground components. That is to say, through the important component saliency module based on the class foreground attention mechanism, the difference between the nearby sound signal and the ambient noise can be enhanced, so that the model is more focused on extracting and retaining the important salient features in the nearby sound signal related to the target signal, reducing the interference of the ambient sound on the analysis of the nearby sound signal, thereby improving the effect of ambient sound filtering, so as to effectively improve the discrimination and performance of the saliency ambient sound-filtered nearby sound signal time-frequency feature.

[0048] Figure 3 A flowchart showing that in the enterprise non-implementation of production suspension measure verification method based on online monitoring data according to an embodiment of the present application, the time-frequency feature matrix of the ambient sound-filtered nearby sound signal is passed through an important component saliency module based on a class foreground attention mechanism to obtain a saliency ambient sound-filtered nearby sound signal time-frequency feature matrix as the saliency ambient sound-filtered nearby sound signal time-frequency feature. Specifically, in the embodiment of the present application, as Figure 3As shown, the time-frequency feature matrix of the nearby sound signal after filtering the environmental sound is passed through the important component saliency module based on the class foreground attention mechanism to obtain the saliency time-frequency feature matrix of the nearby sound signal after filtering the environmental sound as the saliency time-frequency feature of the nearby sound signal after filtering the environmental sound, including: S210, performing feature representation on the time-frequency feature matrix of the nearby sound signal after filtering the environmental sound to obtain the time-frequency representation matrix of the nearby sound signal after filtering the environmental sound; S220, performing masking processing on the time-frequency representation matrix of the nearby sound signal after filtering the environmental sound to obtain the time-frequency mask weight matrix of the nearby sound signal after filtering the environmental sound; and, S230, multiplying the time-frequency mask weight matrix of the nearby sound signal after filtering the environmental sound and the time-frequency feature matrix of the nearby sound signal after filtering the environmental sound at each position point to obtain the saliency time-frequency feature matrix of the nearby sound signal after filtering the environmental sound.

[0049] Specifically, in the embodiment of the present application, performing feature representation on the time-frequency feature matrix of the nearby sound signal after filtering the environmental sound to obtain the time-frequency representation matrix of the nearby sound signal after filtering the environmental sound includes: taking the negative of each position eigenvalue of the time-frequency feature matrix of the nearby sound signal after filtering the environmental sound as the exponent of the natural constant to calculate the exponential function value with the natural constant as the base at each position to obtain the time-frequency class support feature matrix of the nearby sound signal after filtering the environmental sound; and, calculating the reciprocal of the sum of each position eigenvalue in the time-frequency class support feature matrix of the nearby sound signal after filtering the environmental sound and the constant one to obtain the time-frequency representation matrix of the nearby sound signal after filtering the environmental sound.

[0050] Specifically, in the embodiment of the present application, performing masking processing on the time-frequency representation matrix of the nearby sound signal after filtering the environmental sound to obtain the time-frequency mask weight matrix of the nearby sound signal after filtering the environmental sound includes: setting the eigenvalues greater than or equal to a predetermined threshold at each position of the time-frequency representation matrix of the nearby sound signal after filtering the environmental sound to one, and the rest to zero to obtain the time-frequency mask weight matrix of the nearby sound signal after filtering the environmental sound.

[0051] In the embodiment of the present application, preferably, passing the time-frequency feature matrix of the nearby sound signal after filtering the environmental sound through the important component saliency module based on the class foreground attention mechanism to obtain the saliency time-frequency feature matrix of the nearby sound signal after filtering the environmental sound includes: using the important component saliency module based on the class foreground attention mechanism to process the time-frequency feature matrix of the nearby sound signal after filtering the environmental sound with the following class foreground attention formula to obtain the saliency time-frequency feature matrix of the nearby sound signal after filtering the environmental sound; where the class foreground attention formula is:

[0052]

[0053] M c =S⊙M

[0054] Among them, M(i,j) represents the eigenvalue at the (i,j) position of the time-frequency feature matrix of the nearby sound signals after ambient sound filtering, S(i,j) represents the eigenvalue at the (i,j) position of the time-frequency mask weight matrix of the nearby sound signals after ambient sound filtering, ε is a hyperparameter, exp(·) represents the exponential function with the natural constant e as the base, mask(·) represents the masking process, M represents the time-frequency feature matrix of the nearby sound signals after ambient sound filtering, S represents the time-frequency mask weight matrix of the nearby sound signals after ambient sound filtering, ⊙ represents element-wise multiplication, and M c is the time-frequency feature matrix of the nearby sound signals after significant ambient sound filtering.

[0055] In step S160, based on the time-frequency features of the nearby sound signals after significant ambient sound filtering, an inspection result is obtained. Specifically, in the embodiment of the present application, obtaining the inspection result based on the time-frequency features of the nearby sound signals after significant ambient sound filtering includes: passing the time-frequency feature matrix of the nearby sound signals after significant ambient sound filtering through an inspector based on a classifier to obtain an inspection result, and the inspection result is used to indicate whether the monitored enterprise object has not implemented the production suspension measure. That is, classification processing is performed using the time-frequency features of the nearby sound signals after significant ambient sound filtering obtained by significant component saliency of the time-frequency feature matrix of the nearby sound signals after ambient sound filtering, so as to intelligently determine whether the monitored enterprise object has not implemented the production suspension measure. Through this method, the sound signals of the monitored enterprise can be monitored online, and the production and operation status of the enterprise can be understood in a timely manner, reducing the workload and subjectivity of manual verification, thereby improving the accuracy and efficiency of the verification result.

[0056] It is worth mentioning that those of ordinary skill in the art should be aware that before applying the deep neural network model for inference, the deep neural network model needs to be trained first so that the deep neural network can implement specific function capabilities.

[0057] Specifically, in the embodiment of the present application, it further includes training for the time-frequency feature extractor of the sound signal based on the GoogLeNet model, the important component saliency module based on the class foreground attention mechanism, and the inspector based on the classifier.

[0058] Figure 4 is a flowchart for training the time-frequency feature extractor of the sound signal based on the GoogLeNet model, the important component saliency module based on the class foreground attention mechanism, and the inspector based on the classifier in the method for verifying that an enterprise has not implemented the production suspension measure based on online monitoring data according to the embodiment of the present application. As Figure 4As shown, the training steps include: S310, obtaining training data, where the training data includes the training environmental sound signals collected by the first sound sensor, the sound signals near the training of the monitored enterprise object collected by the second sound sensor, and the true value of whether the monitored enterprise fails to implement the production suspension measure; S320, performing wavelet transform on the sound signals near the training of the monitored enterprise object and the training environmental sound signals to obtain the time-frequency image of the sound signals near the training and the time-frequency image of the training environmental sound signals; S330, respectively passing the time-frequency image of the sound signals near the training and the time-frequency image of the training environmental sound signals through the sound signal time-frequency feature extractor based on the GoogLeNet model to obtain the time-frequency feature matrix of the sound signals near the training and the time-frequency feature matrix of the training environmental sound signals; S340, calculating the position-wise difference between the time-frequency feature matrix of the sound signals near the training and the time-frequency feature matrix of the training environmental sound signals to obtain the time-frequency feature matrix of the environmental sound filtering out the sound signals near the training; S350, passing the time-frequency feature matrix of the environmental sound filtering out the sound signals near the training through the important component saliency module based on the class foreground attention mechanism to obtain the time-frequency feature matrix of the training salient environmental sound filtering out the sound signals near the training; S360, passing the time-frequency feature matrix of the training salient environmental sound filtering out the sound signals near the training through the inspector based on the classifier to obtain the classification loss function value; and S370, training the sound signal time-frequency feature extractor based on the GoogLeNet model, the important component saliency module based on the class foreground attention mechanism, and the inspector based on the classifier based on the classification loss function value and through the backpropagation of gradient descent.

[0059] Specifically, in a preferred example, when updating the model parameters through gradient backpropagation based on the classification loss function, a predetermined loss function for promoting classification understanding of the complex feature expression of the time-frequency feature matrix of the training salient environmental sound filtering out the sound signals near the training is further introduced. The calculation of the predetermined loss function is as follows:

[0060] First, calculate the mean and variance between the feature values at any two positions of the time-frequency feature vector of the training salient environmental sound filtering out the sound signals near the training obtained after expanding the time-frequency feature matrix of the training salient environmental sound filtering out the sound signals near the training, for example, denoted as V, to obtain the mean weight matrix M μ and the variance weight matrix M σ , and then multiply the time-frequency feature vector V of the training salient environmental sound filtering out the sound signals near the training by the mean weight matrix M μ and the variance weight matrix M σAfter performing query-based multiplication, calculate the Frobenius norm of the autocorrelation matrix of the two obtained eigenvectors, and subtract the product of the Frobenius norm of the autocorrelation matrix of the training significant environmental sound filtering nearby sound signal time-frequency eigenvector V and the weight as a hyperparameter to obtain the predetermined loss function.

[0061] For example, it is expressed as

[0062]

[0063] where V represents the training significant environmental sound filtering nearby sound signal time-frequency eigenvector obtained after expanding the training significant environmental sound filtering nearby sound signal time-frequency feature matrix, M μ is the mean weight matrix, M σ is the variance weight matrix, T represents the transpose of the vector, represents matrix multiplication operation, α is the weight as a hyperparameter, ‖·‖ F represents the Frobenius norm of the matrix, Loss represents the value of the predetermined loss function, and as described above:

[0064]

[0065] where, v i represents the i-th eigenvalue of the training significant environmental sound filtering nearby sound signal time-frequency eigenvector, v j represents the j-th eigenvalue of the training significant environmental sound filtering nearby sound signal time-frequency eigenvector, M μ (i,j) represents the eigenvalue at the (i,j) position of the mean weight matrix, M σ (i,j) represents the eigenvalue at the (i,j) position of the variance weight matrix.

[0066] That is, due to the enhancement of the local image differential semantic spatial distribution caused by the significant component saliency of the class foreground attention mechanism for the image differential semantics between the training environmental sound signal and the nearby sound signal of the monitored enterprise object, the generated local enhanced outer distribution noise will increase the class distribution complexity of the overall training significant environmental sound filtering nearby sound signal time-frequency feature matrix, resulting in an obvious problem of insufficient classification regression understanding of the complex feature noise distribution expression of the training significant environmental sound filtering nearby sound signal time-frequency feature matrix, affecting the accuracy of the classification result obtained by its classifier.

[0067] The predetermined loss function determines the overall distributed structural bottleneck of the time-frequency feature matrix of the training saliency environmental sound filtering nearby sound signals based on the multi-dimensional interaction of the detail groups of the time-frequency feature matrix of the training saliency environmental sound filtering nearby sound signals. On this basis, through the low-rank structured inference of the time-frequency feature matrix of the training saliency environmental sound filtering nearby sound signals, the correlation between the structural details and the macroscopic structured feature representation behavior of the time-frequency feature matrix of the training saliency environmental sound filtering nearby sound signals is associated and simulated. In this way, by training the classifier with the predetermined loss function, the structural detail dependence relative to the macroscopic complexity can be modeled in the classification scenario, thereby promoting the understanding of the macroscopic complex feature expression based on the structural details of the time-frequency feature matrix of the training saliency environmental sound filtering nearby sound signals, and improving the accuracy of the inspection results obtained by the inspector based on the classifier for the time-frequency feature matrix of the training saliency environmental sound filtering nearby sound signals. Through this method, the sound signals of the monitored enterprise can be monitored online, and the production and operation status of the enterprise can be understood in a timely manner, reducing the workload and subjectivity of manual verification, thereby improving the accuracy and efficiency of the verification results.

[0068] In summary, the method for verifying that an enterprise has not implemented a production suspension measure based on online monitoring data according to an embodiment of the present application is elucidated. It collects the environmental sound signals collected in real time by the first sound sensor and the nearby sound signals of the monitored enterprise object collected in real time by the second sound sensor, and uses the signal processing and analysis algorithm based on the deep learning neural network to perform time-frequency feature analysis on the environmental sound signals and the nearby sound signals of the monitored enterprise object, so as to intelligently judge whether the monitored enterprise object has not implemented a production suspension measure based on the time-frequency features of the nearby sound signals filtering out the environmental sound. Through this method, the sound signals of the monitored enterprise can be monitored online, and the production and operation status of the enterprise can be understood in a timely manner, reducing the workload and subjectivity of manual verification, thereby improving the accuracy and efficiency of the verification results.

[0069] Figure 5 FIG. is a block diagram of a system for verifying that an enterprise has not implemented a production suspension measure based on online monitoring data according to an embodiment of the present application. As Figure 5As shown, the verification system 100 for enterprises that have not implemented production suspension measures based on online monitoring data according to an embodiment of the present application includes: a sound signal acquisition module 110, configured to obtain the ambient sound signal collected by the first sound sensor and the sound signal near the enterprise object to be monitored collected by the second sound sensor; a wavelet transform module 120, configured to perform wavelet transform on the sound signal near the enterprise object to be monitored and the ambient sound signal to obtain a time-frequency image of the sound signal near the enterprise object and a time-frequency image of the ambient sound signal; a sound signal time-frequency feature extraction module 130, configured to perform sound signal time-frequency feature extraction on the time-frequency image of the sound signal near the enterprise object and the time-frequency image of the ambient sound signal respectively to obtain a time-frequency feature matrix of the sound signal near the enterprise object and a time-frequency feature matrix of the ambient sound signal; a sound signal difference calculation module 140, configured to calculate the difference between the time-frequency feature matrix of the sound signal near the enterprise object and the time-frequency feature matrix of the ambient sound signal to obtain a time-frequency feature matrix of the ambient sound-filtered sound signal near the enterprise object; a significant component saliency module 150, configured to pass the time-frequency feature matrix of the ambient sound-filtered sound signal near the enterprise object through a significant component saliency module based on a class foreground attention mechanism to obtain a time-frequency feature matrix of the significant ambient sound-filtered sound signal near the enterprise object as the time-frequency feature of the significant ambient sound-filtered sound signal near the enterprise object; and an inspection result generation module 160, configured to obtain an inspection result based on the time-frequency feature of the significant ambient sound-filtered sound signal near the enterprise object.

[0070] Here, those skilled in the art can understand that the specific operations of each step in the above verification system for enterprises that have not implemented production suspension measures based on online monitoring data have been introduced in detail in the description of the Figures 1 to 4 verification method for enterprises that have not implemented production suspension measures based on online monitoring data above, and thus, the repeated description thereof will be omitted.

[0071] As described above, the verification system 100 for enterprises that have not implemented production suspension measures based on online monitoring data according to an embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with a verification algorithm for enterprises that have not implemented production suspension measures based on online monitoring data. In a possible implementation manner, the verification system 100 for enterprises that have not implemented production suspension measures based on online monitoring data according to an embodiment of the present disclosure can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the verification system 100 for enterprises that have not implemented production suspension measures based on online monitoring data can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the verification system 100 for enterprises that have not implemented production suspension measures based on online monitoring data can also be one of the many hardware modules of the wireless terminal.

[0072] Alternatively, in another example, the enterprise's verification system 100 for unimplemented production suspension measures based on online monitoring data and the wireless terminal can also be separate devices, and the enterprise's verification system 100 for unimplemented production suspension measures based on online monitoring data can be connected to the wireless terminal through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0073] The foregoing are merely examples of the principles of the present disclosure, and those skilled in the art can make various modifications without departing from the scope of the present disclosure. The above embodiments are presented for purposes of illustration rather than limitation. The present disclosure can also take many forms other than those explicitly described herein. Therefore, it is emphasized that the present disclosure is not limited to the explicitly disclosed methods, systems, and devices, but is intended to include variations and modifications within the spirit scope of the appended claims.

Claims

1. A method for verifying that an enterprise has not implemented suspension measures based on online monitoring data, characterized in that: include: Acquire an ambient sound signal collected by a first sound sensor and a sound signal near a monitored enterprise object collected by a second sound sensor; Performing wavelet transformation on the nearby acoustic signals of the monitored enterprise object and the ambient acoustic signals to obtain a time-frequency image of the nearby acoustic signals and a time-frequency image of the ambient acoustic signals; Extracting time-frequency features of acoustic signals from the time-frequency image of nearby acoustic signals and the time-frequency image of ambient acoustic signals respectively to obtain a time-frequency feature matrix of nearby acoustic signals and a time-frequency feature matrix of ambient acoustic signals; Calculating the difference between the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal to obtain the time-frequency feature matrix of the ambient sound-filtered nearby sound signal; The time-frequency feature matrix of the ambient sound-filtered nearby sound signal is passed through an important component saliency module based on a foreground-like attention mechanism to obtain a salient ambient sound-filtered nearby sound signal time-frequency feature matrix as a salient ambient sound-filtered nearby sound signal time-frequency feature; Based on the significant environmental sound, the time-frequency characteristics of nearby sound signals are filtered out to obtain the audit result.

2. The method for verifying that an enterprise has not implemented shutdown measures based on online monitoring data according to claim 1 is characterized in that: The time-frequency features of the nearby sound signals and the time-frequency images of the ambient sound signals are respectively extracted to obtain a time-frequency feature matrix of the nearby sound signals and a time-frequency feature matrix of the ambient sound signals, including: respectively passing the time-frequency features of the nearby sound signals and the time-frequency images of the ambient sound signals through a time-frequency feature extractor based on a GoogLeNet model to obtain a time-frequency feature matrix of the nearby sound signals and a time-frequency feature matrix of the ambient sound signals.

3. The method for verifying that an enterprise has not implemented suspension measures based on online monitoring data according to claim 2 is characterized in that: Calculating the difference between the time-frequency feature matrix of the nearby sound signals and the time-frequency feature matrix of the ambient sound signals to obtain the time-frequency feature matrix of the ambient sound-filtered nearby sound signals, including: calculating the positional difference between the time-frequency feature matrix of the nearby sound signals and the time-frequency feature matrix of the ambient sound signals to obtain the time-frequency feature matrix of the ambient sound-filtered nearby sound signals.

4. The method for verifying that an enterprise has not implemented suspension measures based on online monitoring data according to claim 3 is characterized in that: The time-frequency feature matrix of the ambient sound-filtered nearby sound signal is passed through an important component saliency module based on a foreground-like attention mechanism to obtain a salient ambient sound-filtered nearby sound signal time-frequency feature matrix as a salient ambient sound-filtered nearby sound signal time-frequency feature, including: Characterizing the time-frequency characteristic matrix of the ambient sound-filtered nearby sound signals to obtain a time-frequency characteristic matrix of the ambient sound-filtered nearby sound signals; Masking the ambient sound-filtered nearby sound signal time-frequency representation matrix to obtain an ambient sound-filtered nearby sound signal time-frequency mask weight matrix; The time-frequency mask weight matrix of the ambient sound-filtered nearby sound signal and the time-frequency feature matrix of the ambient sound-filtered nearby sound signal are multiplied point by point to obtain the time-frequency feature matrix of the significant ambient sound-filtered nearby sound signal.

5. The method for verifying that an enterprise has not implemented shutdown measures based on online monitoring data according to claim 4 is characterized in that: Characterizing the time-frequency characteristic matrix of the ambient sound-filtered nearby sound signal to obtain the time-frequency characteristic matrix of the ambient sound-filtered nearby sound signal, including: Using the negative number of the characteristic value of each position of the time-frequency characteristic matrix of the ambient sound filtering nearby sound signal as the exponent of the natural constant to calculate the exponential function value based on the natural constant according to the position to obtain the time-frequency class support characteristic matrix of the ambient sound filtering nearby sound signal; The inverse of the sum of the eigenvalues ​​at each position in the time-frequency class support feature matrix of the ambient sound-filtered nearby sound signal and a constant one is calculated to obtain the time-frequency representation matrix of the ambient sound-filtered nearby sound signal.

6. The method for verifying that an enterprise has not implemented shutdown measures based on online monitoring data according to claim 5 is characterized in that: The time-frequency characterization matrix of the ambient sound filtered nearby sound signal is masked to obtain the time-frequency mask weight matrix of the ambient sound filtered nearby sound signal, including: setting the eigenvalues ​​greater than or equal to a predetermined threshold in each position of the time-frequency characterization matrix of the ambient sound filtered nearby sound signal to one, and setting the rest to zero to obtain the time-frequency mask weight matrix of the ambient sound filtered nearby sound signal.

7. The method for verifying that an enterprise has not implemented suspension measures based on online monitoring data according to claim 6 is characterized in that: Based on the significant environmental sound filtering out the time-frequency characteristics of nearby sound signals, an audit result is obtained, including: passing the time-frequency characteristic matrix of the significant environmental sound filtering out nearby sound signals through a classifier-based auditor to obtain an audit result, and the audit result is used to indicate whether the monitored enterprise object has not implemented the production suspension measures.

8. The method for verifying that an enterprise has not implemented suspension measures based on online monitoring data according to claim 7 is characterized in that: It also includes a method for training the acoustic signal time-frequency feature extractor based on the GoogLeNet model, the important component saliency module based on the foreground-like attention mechanism, and the classifier-based auditor.

9. The method for verifying that an enterprise has not implemented production suspension measures based on online monitoring data according to claim 8 is characterized in that: The training step comprises: Acquire training data, the training data including a training environment sound signal collected by a first sound sensor and a training vicinity sound signal of a monitored enterprise object collected by a second sound sensor, and a true value of whether the monitored enterprise has not implemented the production suspension measures; Performing wavelet transform on the monitored enterprise object's training vicinity sound signal and the training environment sound signal to obtain a training vicinity sound signal time-frequency image and a training environment sound signal time-frequency image; The training nearby sound signal time-frequency image and the training environment sound signal time-frequency image are respectively passed through the sound signal time-frequency feature extractor based on the GoogLeNet model to obtain a training nearby sound signal time-frequency feature matrix and a training environment sound signal time-frequency feature matrix; Calculating the position difference between the training nearby sound signal time-frequency feature matrix and the training environment sound signal time-frequency feature matrix to obtain the training environment sound filtered nearby sound signal time-frequency feature matrix; The training environment sound filtering nearby sound signal time-frequency feature matrix is ​​passed through the important component saliency module based on the foreground-like attention mechanism to obtain the training salient environment sound filtering nearby sound signal time-frequency feature matrix; Passing the training significant ambient sound filtering nearby sound signal time-frequency feature matrix through the classifier-based auditor to obtain a classification loss function value; The acoustic signal time-frequency feature extractor based on the GoogLeNet model, the important component saliency module based on the foreground-like attention mechanism, and the classifier-based auditor are trained based on the classification loss function value and through back propagation of gradient descent.

10. A verification system for enterprises that have not implemented suspension measures based on online monitoring data, characterized in that: include: A sound signal acquisition module, used to acquire the environmental sound signal acquired by the first sound sensor and the sound signal near the monitored enterprise object acquired by the second sound sensor; A wavelet transform module, used for performing wavelet transform on the nearby acoustic signals of the monitored enterprise object and the ambient acoustic signals to obtain a time-frequency image of the nearby acoustic signals and a time-frequency image of the ambient acoustic signals; An acoustic signal time-frequency feature extraction module is used to extract acoustic signal time-frequency features from the nearby acoustic signal time-frequency image and the ambient acoustic signal time-frequency image to obtain a nearby acoustic signal time-frequency feature matrix and an ambient acoustic signal time-frequency feature matrix; A sound signal difference calculation module, used for calculating the difference between the time-frequency feature matrix of the nearby sound signal and the time-frequency feature matrix of the ambient sound signal to obtain the time-frequency feature matrix of the nearby sound signal after the ambient sound is filtered out; An important component saliency module is used to pass the time-frequency feature matrix of the ambient sound-filtered nearby sound signal through the important component saliency module based on the foreground attention mechanism to obtain a salient ambient sound-filtered nearby sound signal time-frequency feature matrix as a salient ambient sound-filtered nearby sound signal time-frequency feature; The audit result generation module is used to filter out the time-frequency characteristics of nearby sound signals based on the significant environmental sound to obtain the audit result.

Images