A noise monitoring system for urban noise pollution based on intelligent sensing

Abstract

The application provides a kind of noise monitoring system of urban noise pollution based on intelligent sensing, it is related to data processing technical field.In the present application, first, first noise audio and second noise audio are subjected to semantic mining respectively, to form first noise semantic vector and second noise semantic vector;Second, noise type semantic vector is mined;Then, noise type semantic vector and first noise semantic vector are subjected to noise type correlation analysis, and noise type correlation data is output;Further, noise type correlation data is used as the basis for semantic enhancement, and second noise semantic vector is subjected to semantic enhancement to form enhanced noise semantic vector;Finally, noise type prediction is carried out according to enhanced noise semantic vector, to form target noise type.Based on the above method, the problem of relatively low reliability of noise monitoring in the prior art can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a noise monitoring system for urban noise pollution based on intelligent sensing. BACKGROUND

[0002] With the acceleration of urbanization and the improvement of industrialization, noise pollution has become one of the important environmental problems in modern society, which affects the quality of people's life. Among them, noise identification is an important basis for noise pollution control. Specifically, based on the results of noise identification, various means and technologies can be used to reduce, control or eliminate noise in the environment to improve the quality of living environment. However, the traditional noise pollution control mainly adopts passive measures such as setting soundproof walls and vehicle speed limits, and the effect is relatively poor. Therefore, in the prior art, noise identification is performed, and then targeted noise pollution control is performed based on the results of noise identification, and the effect is improved.

[0003] However, the inventors have found that in the prior art, different noise categories cannot be effectively and reliably identified, so effective control measures cannot be taken according to the characteristics of different noise sources, which affects the effect of noise pollution control. Therefore, it is necessary to provide a scheme capable of reliably identifying noise types. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a noise monitoring system for urban noise pollution based on intelligent sensing to improve the problem of relatively low reliability of noise monitoring in the prior art.

[0005] To achieve the above-mentioned purpose, the embodiments of the present application adopt the following technical solutions:

[0006] A noise monitoring system for urban noise pollution based on intelligent sensing, comprising a processor and a memory, the processor being configured to execute a computer program in the memory to implement a noise monitoring method, wherein the noise monitoring method comprises:

[0007] respectively performing semantic mining on a first noise audio and a second noise audio to form a first noise semantic vector and a second noise semantic vector, wherein the first noise audio and the second noise audio are audio collected by two different types of noise monitoring sensors at the same time on a target space to form;

[0008] mining a noise type semantic vector, wherein the noise type semantic vector is used to represent semantic information of each noise type;

[0009] Noise type correlation analysis is performed on the noise type semantic vector and the first noise semantic vector to output noise type correlation data, wherein the noise type correlation data is used to reflect the correlation between each of the noise types and the first noise semantic vector;

[0010] Using the noise type-related data as the basis for semantic enhancement, the second noise semantic vector is semantically enhanced to form an enhanced noise semantic vector.

[0011] Based on the enhanced noise semantic vector, noise type prediction is performed to form the target noise type.

[0012] In some preferred embodiments, in the above-mentioned noise monitoring system for urban noise pollution based on intelligent sensing, the step of performing semantic mining on the first noise audio and the second noise audio respectively to form a first noise semantic vector and a second noise semantic vector includes:

[0013] The first noise audio is subjected to Fourier transform to form a first noise spectrum, and the first noise spectrum is subjected to global harmonic semantic mining to form a first noise semantic vector.

[0014] The second noise audio is subjected to Fourier transform to form a second noise spectrum, and global harmonic semantic mining is performed on the second noise spectrum to form a second noise semantic vector.

[0015] In some preferred embodiments, in the above-mentioned noise monitoring system for urban noise pollution based on intelligent sensing, the steps of performing Fourier transform on the first noise audio to form a first noise spectrum, and performing global harmonic semantic mining on the first noise spectrum to form a first noise semantic vector, include:

[0016] Perform a Fourier transform on the first noise audio to form a first noise spectrum.

[0017] The first noise spectrum is subjected to higher harmonic identification to obtain multiple higher harmonic frequencies in the first noise spectrum.

[0018] Based on each of the plurality of higher harmonic frequencies, local frequency hiding processing is performed on the first noise spectrum to form a plurality of hidden noise spectrums.

[0019] The first noise spectrum map and each of the hidden noise spectrum maps are semantically mapped to form a first noise spectrum map vector corresponding to the first noise spectrum map and a hidden noise spectrum map vector corresponding to each of the hidden noise spectrum maps.

[0020] Each of the hidden noise spectrogram vectors is merged into the first noise spectrogram vector to form the first noise semantic vector.

[0021] In some preferred embodiments, in the above-described noise monitoring system for urban noise pollution based on intelligent sensing, the step of fusing each of the hidden noise spectrum vectors into the first noise spectrum vector to form a first noise semantic vector includes:

[0022] Based on the order of frequencies retained in the corresponding hidden noise spectrum maps from largest to smallest, the multiple hidden noise spectrum map vectors are sorted to form a target vector sequence;

[0023] In multiple fusion stages, multiple hidden noise spectrogram vectors are sequentially fused into the first noise spectrogram vector according to the order in the target vector sequence to form a first noise semantic vector, wherein the number of fusion stages is equal to the number of hidden noise spectrogram vectors.

[0024] In some preferred embodiments, in the above-described noise monitoring system for urban noise pollution based on intelligent sensing, the step of sequentially fusing multiple hidden noise spectrum vectors into the first noise spectrum vector according to the order in the target vector sequence during multiple fusion stages to form a first noise semantic vector includes:

[0025] For the first fusion stage among the multiple fusion stages, the first hidden noise spectrogram vector in the target vector sequence is mapped to the gating mapping parameter distribution of the first noise spectrogram vector, and based on the gating mapping parameter distribution, the first noise spectrogram vector is weighted and mapped to form the fusion output vector of the first fusion stage;

[0026] For the last fusion stage among the multiple fusion stages, the last hidden noise spectrogram vector in the target vector sequence is mapped to the gating mapping parameter distribution of the fusion output vector of the previous fusion stage. Based on the attention parameter distribution between the hidden noise spectrogram vector and the fusion output vector, the gating mapping parameter distribution is adjusted to form an adjusted gating mapping parameter distribution. Based on the adjusted gating mapping parameter distribution, the fusion output vector is weighted and mapped to form a first noise semantic vector.

[0027] In some preferred embodiments, in the above-described noise monitoring system for urban noise pollution based on intelligent sensing, the step of performing noise type correlation analysis on the noise type semantic vector and the first noise semantic vector, and outputting noise type correlation data, includes:

[0028] The noise type semantic vector is transformed using the first semantic space to form a noise type transformation vector;

[0029] The first noise semantic vector is transformed into a second semantic space to form a first noise transformed vector.

[0030] Attention analysis is performed on the noise type conversion vector and the first noise conversion vector to form an attention parameter distribution between the noise type semantic vector and the first noise semantic vector, wherein the attention parameter distribution serves as noise type-related data.

[0031] In some preferred embodiments, in the above-mentioned noise monitoring system for urban noise pollution based on intelligent sensing, the step of performing attention analysis on the noise type conversion vector and the first noise conversion vector to form the attention parameter distribution between the noise type semantic vector and the first noise semantic vector includes:

[0032] The vector formed by transposing the first noise conversion vector is multiplied with the noise type conversion vector to form the initial correlation parameter distribution;

[0033] By utilizing the importance parameter distribution corresponding to each enhancement stage in the semantic enhancement process, an intermediate relevant parameter distribution corresponding to each enhancement stage is mapped to be formed. The importance parameter distribution serves as the network parameters of the corresponding neural network model and is formed during the learning and training process.

[0034] The intermediate relevant parameter distributions and the initial relevant parameter distributions corresponding to each enhancement stage are respectively fused to form the target relevant parameter distributions corresponding to each enhancement stage;

[0035] The distribution of target-related parameters corresponding to each enhancement stage is normalized and mapped to form the distribution of attention parameters corresponding to each enhancement stage.

[0036] The step of using the noise type-related data as a basis for semantic enhancement to semantically enhance the second noise semantic vector to form an enhanced noise semantic vector includes:

[0037] For each enhancement stage, the attention parameter distribution corresponding to the enhancement stage is used as the basis for semantic enhancement. The noise semantic vector to be processed in the enhancement stage is semantically enhanced. The output vector of the last enhancement stage is determined as the enhanced noise semantic vector. The noise semantic vector to be processed in the first enhancement stage is the second noise semantic vector. The noise semantic vector to be processed in the second and subsequent enhancement stages is the output vector of the previous enhancement stage.

[0038] In some preferred embodiments, in the above-described noise monitoring system for urban noise pollution based on intelligent sensing, the step of mapping the distribution of importance parameters corresponding to each enhancement stage in the semantic enhancement process to form the distribution of intermediate relevant parameters corresponding to each enhancement stage includes:

[0039] Obtain the first and second correlation parameters from the initial correlation parameter distribution, wherein the first correlation parameter is greater than the second correlation parameter;

[0040] The first correlation parameter and each correlation parameter in the initial correlation parameter distribution are subtracted to form the first difference parameter distribution;

[0041] The difference between each relevant parameter in the initial relevant parameter distribution and the second relevant parameter is calculated to form the second difference parameter distribution;

[0042] Determine the distribution of importance parameters for each enhancement stage;

[0043] The first and second difference parameter distributions are used as parameter intervals for the importance parameter distributions corresponding to each enhancement stage, and parameter mapping is performed to obtain the intermediate relevant parameter distributions corresponding to each enhancement stage.

[0044] In some preferred embodiments, in the above-described noise monitoring system for urban noise pollution based on intelligent sensing, the step of fusing the intermediate correlation parameter distribution and the initial correlation parameter distribution corresponding to each enhancement stage to form the target correlation parameter distribution corresponding to each enhancement stage includes:

[0045] For each enhancement stage, the intermediate relevant parameter distribution and the initial relevant parameter distribution are summed to form the target relevant parameter distribution for that enhancement stage.

[0046] In some preferred embodiments, in the above-described noise monitoring system for urban noise pollution based on intelligent sensing, the step of predicting the noise type based on the enhanced noise semantic vector to form a target noise type includes:

[0047] The enhanced noise semantic vector is fully connected to obtain a noise fully connected vector, wherein the dimension of the noise fully connected vector is equal to the number of each noise type;

[0048] The noise fully connected vector is classified and mapped to form a noise type probability distribution, wherein each probability value in the noise type probability distribution corresponds to a noise type;

[0049] The noise type corresponding to the highest probability value in the noise type probability distribution is determined as the target noise type.

[0050] In a noise monitoring system for urban noise pollution based on intelligent sensing provided in this embodiment of the invention, firstly, semantic mining is performed on a first noise audio and a second noise audio to form a first noise semantic vector and a second noise semantic vector, respectively; secondly, a noise type semantic vector is mined; then, noise type correlation analysis is performed on the noise type semantic vector and the first noise semantic vector to output noise type correlation data; further, the noise type correlation data is used as the basis for semantic enhancement to semantically enhance the second noise semantic vector to form an enhanced noise semantic vector; finally, noise type prediction is performed based on the enhanced noise semantic vector to form the target noise type. Based on the above method, since two different types of noise monitoring sensors are used to collect noise audio for semantic mining and prediction, the basis is more sufficient, which can effectively improve the reliability of noise type prediction. Furthermore, since the semantic information mined from the noise audio is also analyzed based on the semantic information of each noise type, it is easier to mine or capture semantic information related to noise type prediction, thereby reducing semantic information unrelated to noise type prediction. This improves the accuracy of the basis for noise type prediction (i.e., the accuracy of semantic representation), and further improves the reliability of prediction, effectively addressing the problem of relatively low reliability of noise monitoring in existing technologies.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0052] Figure 1 This is a structural block diagram of an urban noise pollution monitoring system based on intelligent sensing, provided in an embodiment of the present invention.

[0053] Figure 2 This is a flowchart illustrating the steps included in the noise monitoring method provided in this embodiment of the invention.

[0054] Figure 3 This is a schematic diagram of semantic mining provided in an embodiment of the present invention.

[0055] Figure 4 This is a schematic diagram of multi-stage fusion provided in an embodiment of the present invention.

[0056] Figure 5 This is a schematic diagram of attention analysis provided in an embodiment of the present invention.

[0057] Figure 6 This is a schematic diagram illustrating semantic enhancement provided in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0059] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0060] like Figure 1 As shown, this embodiment of the invention provides a noise monitoring system for urban noise pollution based on intelligent sensing. The noise monitoring system for urban noise pollution based on intelligent sensing may include a memory and a processor.

[0061] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, they can be electrically connected via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that exists in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby implementing the noise monitoring method provided in the embodiments of the present invention (as described below).

[0062] Optionally, the memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. Furthermore, the processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0063] and, Figure 1 The structure shown is for illustrative purposes only. The noise monitoring system for urban noise pollution based on intelligent sensing may also include components such as... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices. In an alternative example, the urban noise pollution monitoring system based on intelligent sensing may be a server with data processing capabilities.

[0064] Combination Figure 2 This invention also provides a noise monitoring method applicable to the aforementioned noise monitoring system for urban noise pollution based on intelligent sensors. The method steps defined in the relevant process of the noise monitoring method can be implemented by the noise monitoring system for urban noise pollution based on intelligent sensors (hereinafter referred to as the noise monitoring system). The following will describe... Figure 2 The specific process shown will be explained in detail.

[0065] Step S110: Semantic mining is performed on the first noisy audio and the second noisy audio respectively to form the first noise semantic vector and the second noise semantic vector.

[0066] In this embodiment of the invention, the noise monitoring system can perform semantic mining on a first noise audio and a second noise audio, respectively, to form a first noise semantic vector and a second noise semantic vector. The first and second noise audios are formed by simultaneously acquiring audio data from a target space using two different types of noise monitoring sensors. That is, semantic mining can be performed on the first noise audio to form a first noise semantic vector, and semantic mining can be performed on the second noise audio to form a second noise semantic vector. Semantic mining refers to extracting potential semantic information from the first and second noise audios and representing it as a vector (or matrix), thus obtaining the corresponding semantic vector. In one example, the noise monitoring sensor corresponding to the first noise audio can be a condenser microphone, and the noise monitoring sensor corresponding to the second noise audio can be a MEMS microphone (Micro-Electro-Mechanical Systems Microphone). While condenser microphones offer higher accuracy, they are susceptible to humidity effects, while MEMS microphones offer relatively stable performance. Therefore, in subsequent steps, correlation analysis can be performed based on the semantic information representing each noise type to constrain the semantic information of the first noise audio acquired by the condenser microphone, thereby improving the accuracy of semantic representation.

[0067] Step S120: Extract the semantic vector of the noise type.

[0068] In this embodiment of the invention, the noise monitoring system can mine noise type semantic vectors. These noise type semantic vectors are used to characterize the semantic information of each noise type. Furthermore, noise types may include, but are not limited to, traffic noise, industrial noise, construction noise, human activity noise, and animal activity noise.

[0069] Step S130: Perform noise type correlation analysis on the noise type semantic vector and the first noise semantic vector, and output noise type correlation data.

[0070] In this embodiment of the invention, after obtaining the noise type semantic vector and the first noise semantic vector, the noise monitoring system can perform noise type correlation analysis on the noise type semantic vector and the first noise semantic vector, and output noise type correlation data. The noise type correlation data reflects the correlation between each of the noise types and the first noise semantic vector.

[0071] Step S140: Using the noise type-related data as the basis for semantic enhancement, the second noise semantic vector is semantically enhanced to form an enhanced noise semantic vector.

[0072] In this embodiment of the invention, after obtaining the noise type-related data and the second noise semantic vector, the noise monitoring system can use the noise type-related data as a basis for semantic enhancement to semantically enhance the second noise semantic vector, forming an enhanced noise semantic vector. It is understood that the noise type-related data may include the first noise semantic vector and the semantic information carried by the noise type semantic vector. Thus, after semantically enhancing the second noise semantic vector, the fusion of three dimensions of semantic information can be achieved, resulting in an enhanced noise semantic vector carrying three dimensions of semantic information, thereby giving the enhanced noise semantic vector better semantic representation capabilities.

[0073] Step S150: Based on the enhanced noise semantic vector, noise type prediction is performed to form the target noise type.

[0074] In this embodiment of the invention, after obtaining the enhanced noise semantic vector, the noise monitoring system can predict the noise type based on the enhanced noise semantic vector to form a target noise type. The target noise type can be one of the noise types corresponding to the noise type semantic vector.

[0075] Based on the above method, since two different types of noise monitoring sensors are used to collect noise audio for semantic mining and prediction, the basis is more sufficient, which can effectively improve the reliability of noise type prediction. Furthermore, since the semantic information mined from the noise audio is also analyzed based on the semantic information of each noise type, it is easier to mine or capture semantic information related to noise type prediction, thereby reducing semantic information unrelated to noise type prediction. This improves the accuracy of the basis for noise type prediction (i.e., the accuracy of semantic representation), and further improves the reliability of prediction, effectively addressing the problem of relatively low reliability of noise monitoring in existing technologies.

[0076] In the first part, regarding step S110, it should be noted that the specific method for semantic mining of the first noisy audio and the second noisy audio is not limited and can be selected according to actual needs.

[0077] For example, in one feasible implementation, the first noisy audio and the second noisy audio in the time domain can be sampled and processed to form corresponding first audio signal sequences and second audio signal sequences. Then, the first audio signal sequences and the second audio signal sequences can be preprocessed (such as normalization and standardization) to obtain a first preprocessed sequence and a second preprocessed sequence. Finally, the first preprocessed sequence and the second preprocessed sequence can be convolved to obtain a first noise semantic vector and a second noise semantic vector. The convolution process can be implemented through a corresponding convolutional network layer.

[0078] For example, in another possible implementation, in order to ensure that the formed first noise semantic vector and second noise semantic vector have better semantic representation capabilities, such as taking into account the global semantic information of the first noise audio and the second noise audio, it is also possible to represent the harmonic semantic information. Since the harmonic semantic information has a high representational effect on the identification of noise type, in order to capture semantic information that can effectively identify the noise type through semantic mining, the above step S110 can further include steps S111 and S112, the specific contents of which are as follows.

[0079] Step S111: Perform Fourier transform on the first noise audio to form a first noise spectrum, and perform global harmonic semantic mining on the first noise spectrum to form a first noise semantic vector.

[0080] In this embodiment of the invention, the first noise audio can be subjected to Fourier transform to form a first noise spectrum, and the first noise spectrum can be subjected to global and harmonic semantic mining to form a first noise semantic vector. That is, after obtaining the first noise spectrum by performing Fourier transform, the first noise spectrum can be subjected to global semantic information mining and harmonic semantic information mining to obtain a first noise semantic vector carrying global semantic information and harmonic semantic information.

[0081] Step S112: Perform Fourier transform on the second noise audio to form a second noise spectrum, and perform global harmonic semantic mining on the second noise spectrum to form a second noise semantic vector.

[0082] In this embodiment of the invention, the second noise audio can be subjected to Fourier transform to form a second noise spectrum, and global harmonic semantic mining can be performed on the second noise spectrum to form a second noise semantic vector. That is, after obtaining the second noise spectrum through Fourier transform, global semantic information mining and harmonic semantic information mining can be performed on the second noise spectrum to obtain a second noise semantic vector carrying global and harmonic semantic information. Furthermore, it should be noted that the method for performing global harmonic semantic mining on the second noise spectrum can be the same as the method for performing global harmonic semantic mining on the first noise spectrum; the specific method can be referred to the relevant description below.

[0083] It should be further noted that the specific method for forming the first noise semantic vector in step S111 is not limited. For example, in one feasible implementation, in order to ensure that the formed first noise semantic vector can fully represent the harmonic semantic information, step S111 may further include steps S111a, S111b, S111c, S111d, and S111e. The specific contents of each step are as follows (in conjunction with...). Figure 3 (As shown).

[0084] Step S111a: Perform Fourier transform on the first noise audio to form a first noise spectrum.

[0085] In this embodiment of the invention, the first noise audio can be subjected to Fourier transform to form a first noise spectrum, wherein the horizontal axis of the first noise spectrum can be frequency, and the vertical axis of the first noise spectrum can be amplitude (which can be expressed in decibels, etc.).

[0086] Step S111b: Identify higher harmonics in the first noise spectrum to obtain multiple higher harmonic frequencies in the first noise spectrum.

[0087] In this embodiment of the invention, after obtaining the first noise spectrum, higher harmonics can be identified in the first noise spectrum to obtain multiple higher harmonic frequencies. It should be noted that the identification of higher harmonics may include: first determining the fundamental frequency from the first noise spectrum; then, frequencies such as 2 times, 3 times, 4 times, 5 times, 6 times, and 7 times the fundamental frequency can be used as higher harmonic frequencies to obtain multiple higher harmonic frequencies.

[0088] Step S111c: Based on each of the plurality of higher harmonic frequencies, perform local frequency hiding processing on the first noise spectrum to form a plurality of hidden noise spectrums.

[0089] In this embodiment of the invention, after obtaining the plurality of higher harmonic frequencies, local frequency hiding processing can be performed on the first noise spectrum map based on each of the plurality of higher harmonic frequencies to form a plurality of hidden noise spectrum maps. It should be noted that performing local frequency hiding processing on the first noise spectrum map based on twice the fundamental frequency forms the first hidden noise spectrum map; performing local frequency hiding processing on the first noise spectrum map based on three times the fundamental frequency forms the second hidden noise spectrum map; performing local frequency hiding processing on the first noise spectrum map based on four times the fundamental frequency forms the third hidden noise spectrum map, and so on. Specifically, in the first hidden noise spectrum map, only the region corresponding to twice the fundamental frequency is retained (other regions are hidden), or in other words, only the region corresponding to twice the fundamental frequency and the region corresponding to its nearest neighboring frequency are retained. Correspondingly, in the second hidden noise spectrum map, only the region corresponding to three times the fundamental frequency is retained (other regions are hidden), or in other words, only the region corresponding to three times the fundamental frequency and the region corresponding to its nearest neighboring frequency are retained. Accordingly, in the third hidden noise spectrum image, only the region corresponding to 4 times the fundamental frequency is preserved (all other regions are hidden), or in other words, only the region corresponding to 4 times the fundamental frequency and the region corresponding to the nearest frequency of that fundamental frequency are preserved. Furthermore, it should be noted that the size of the multiple hidden noise spectrum images is the same as the size of the first noise spectrum image; that is, the hiding operation does not change the size of the spectrum image. For example, the first noise spectrum image can be a binary image, represented by 0s and 1s. Thus, during the hiding process, it is only necessary to update the 1s in the regions to be hidden to 0s.

[0090] Step S111d: Semantic space mapping is performed on the first noise spectrum map and each of the hidden noise spectrum maps to form a first noise spectrum map vector corresponding to the first noise spectrum map and a hidden noise spectrum map vector corresponding to each of the hidden noise spectrum maps.

[0091] In this embodiment of the invention, after obtaining the first noise spectrum map and each of the hidden noise spectrum maps, the first noise spectrum map and each of the hidden noise spectrum maps can be semantically mapped to form a first noise spectrum map vector corresponding to the first noise spectrum map and a hidden noise spectrum map vector corresponding to each of the hidden noise spectrum maps. That is, the first noise spectrum map can be mapped into the semantic space to obtain the first noise spectrum map vector, and the hidden noise spectrum map can be mapped into the semantic space to form the hidden noise spectrum map vector. Furthermore, the mapping to the semantic space can be implemented using a convolutional network layer. For example, the architecture of this convolutional network layer can be:

[0092] Convolutional layer 1: The kernel size is (3, 3), the stride is 1, "same" padding is used to make the output size the same as the input, the activation function is ReLU (Rectified Linear Unit) to add non-linearity, and the number of output channels is 64;

[0093] Convolutional layer 2, convolutional layer 3, etc.: As the convolutional depth increases, the number and size of the convolutional kernels can be gradually increased to capture more complex spectral features, from 64 to 128 to 256 convolutional kernels;

[0094] Pooling layer: The pooling layer is used to reduce the dimension of the convolutional feature map. MaxPooling is used, the pooling window is (3,3), and the stride is set to 2.

[0095] Fully connected layer: Maps the features extracted by the convolutional pooling layer to a fixed semantic vector space: flattens the pooled feature map into a one-dimensional vector, which is used as the input of the fully connected layer; the number of units is set to 512, and the activation function is ReLU.

[0096] Step S111e: Each of the hidden noise spectrogram vectors is fused into the first noise spectrogram vector to form a first noise semantic vector.

[0097] In this embodiment of the invention, after obtaining each of the audio noise spectrum vectors and the first noise spectrum vector, each of the hidden noise spectrum vectors can be fused into the first noise spectrum vector to form a first noise semantic vector. In this way, it can be ensured that the first noise semantic vector can represent the global semantic information in the first noise audio while also focusing on representing the harmonic semantic information that contributes to the identification of noise type.

[0098] It should be further explained that, for step S111e above, the specific way in which each hidden noise spectrogram vector is fused into the first noise spectrogram vector is not limited. For example, in one feasible implementation, in order to ensure that the formed first noise semantic vector has a high semantic representation capability, it can be processed according to the importance of semantic information during the fusion process. Specifically, step S111e above can further include step e1 and step e2, and the specific contents of each step are as follows.

[0099] Step e1: Based on the order of the frequencies retained in the corresponding hidden noise spectrum maps from largest to smallest, sort the multiple hidden noise spectrum map vectors to form a target vector sequence.

[0100] In this embodiment of the invention, multiple hidden noise spectrum vectors can be sorted according to the order of the frequencies retained in the corresponding hidden noise spectrum vectors from largest to smallest to form a target vector sequence. That is, in the target vector sequence, the hidden noise spectrum vector corresponding to the first hidden noise spectrum vector has the largest retained frequency, and the hidden noise spectrum vector corresponding to the last hidden noise spectrum vector has the smallest retained frequency. For example, when the target vector sequence includes a first hidden noise spectrum vector, a second hidden noise spectrum vector, a third hidden noise spectrum vector, a fourth hidden noise spectrum vector, a fifth hidden noise spectrum vector, and a sixth hidden noise spectrum vector, the frequency retained in the hidden noise spectrum corresponding to the first hidden noise spectrum vector is 7 times the fundamental frequency, the frequency retained in the hidden noise spectrum corresponding to the second hidden noise spectrum vector is 6 times the fundamental frequency, the frequency retained in the hidden noise spectrum corresponding to the third hidden noise spectrum vector is 5 times the fundamental frequency, the frequency retained in the hidden noise spectrum corresponding to the fourth hidden noise spectrum vector is 4 times the fundamental frequency, the frequency retained in the hidden noise spectrum corresponding to the fifth hidden noise spectrum vector is 3 times the fundamental frequency, and the frequency retained in the hidden noise spectrum corresponding to the sixth hidden noise spectrum vector is 2 times the fundamental frequency.

[0101] Step e2: In multiple fusion stages, according to the order in the target vector sequence, multiple hidden noise spectrogram vectors are sequentially fused into the first noise spectrogram vector to form a first noise semantic vector.

[0102] In this embodiment of the invention, after forming the target vector sequence, multiple hidden noise spectrogram vectors can be sequentially fused into the first noise spectrogram vector in multiple fusion stages, according to the order in the target vector sequence, to form a first noise semantic vector. The number of fusion stages is equal to the number of hidden noise spectrogram vectors. For example, when the target vector sequence includes 6 hidden noise spectrograms, there can be 6 fusion stages. That is, the first fusion stage is used to fuse the first hidden noise spectrogram vector into the first noise spectrogram vector, the second fusion stage is used to fuse the second hidden noise spectrogram vector into the first noise spectrogram vector, the third fusion stage is used to fuse the third hidden noise spectrogram vector into the first noise spectrogram vector, the fourth fusion stage is used to fuse the first hidden noise spectrogram vector into the fourth noise spectrogram vector, the fifth fusion stage is used to fuse the fifth hidden noise spectrogram vector into the first noise spectrogram vector, and the sixth fusion stage is used to fuse the sixth hidden noise spectrogram vector into the first noise spectrogram vector.

[0103] It should be noted that, generally speaking, the 2nd and 3rd harmonics are usually more effective in noise identification, the 4th and 5th harmonics can further help distinguish different noise sources, but their impact is usually smaller, and the 6th and higher harmonics have a weaker effect, usually contributing only in special cases. Therefore, to avoid the loss of important semantic information due to the loss of earlier semantic information as the fusion stage progresses, the semantic information of higher frequency harmonics can be fused in the early stages of the fusion stage, the semantic information of harmonics of general frequencies can be fused in the middle stages, and the semantic information of lower frequency harmonics can be fused in the later stages. This avoids the loss of important semantic information of lower frequency harmonics during the fusion process.

[0104] It should be further noted that the specific method of multi-stage fusion in step e2 above is not limited. For example, in one feasible implementation, in order to balance the reliability and computational cost of semantic fusion, that is, to ensure that the semantic accuracy of the resulting first noisy semantic vector is high and to reduce computational cost, step e2 above may further include the following (in conjunction with...). Figure 4 ):

[0105] The first step involves mapping the first hidden noise spectrum vector (corresponding to the highest harmonic frequency, such as 7 times the fundamental frequency) in the target vector sequence to a gating mapping parameter distribution of the first noise spectrum vector. Based on this gating mapping parameter distribution, a weighted mapping is performed on the first noise spectrum vector (i.e., the gating mapping parameter distribution and the first noise spectrum vector are multiplied bitwise to filter or weight the semantic information of each position in the first noise spectrum vector by representing the importance of each position). This forms the fusion output vector for the penultimate fusion stage. It should be noted that a fully connected operation can be performed on the hidden noise spectrum vector to achieve a linear mapping. Then, the result of the linear mapping is processed using activation functions such as SigmaDose to obtain a gating mapping parameter distribution that can represent importance. The fully connected operation does not change the size of the vector; that is, the hidden noise spectrum vector, the gating mapping parameter distribution, the first noise spectrum vector, and the fusion output vector all have the same size.

[0106] The second step involves mapping the last hidden noise spectrum vector (with the smallest corresponding harmonic frequency, such as twice the fundamental frequency) in the target vector sequence to the gating mapping parameter distribution of the fusion output vector of the previous fusion stage. Then, based on the attention parameter distribution between the hidden noise spectrum vector and the fusion output vector, the gating mapping parameter distribution is adjusted (for example, by performing a weighted summation calculation based on the attention parameter distribution to achieve adjustment), forming an adjusted gating mapping parameter distribution. Finally, based on the adjusted gating mapping parameter distribution, the fusion output vector is weighted and mapped to form a first noise semantic vector.

[0107] For the fusion stages other than the first and last fusion stages (i.e., fusing the corresponding hidden noise spectrogram vector with the fusion output vector of the previous fusion stage), either the fusion method in "Step 1" or the fusion method in "Step 2" can be used, and the configuration can be adjusted according to actual needs. It should be noted that the gating and weighted mapping in "Step 1" can achieve fusion of two semantic vectors with less computation. While the gating, attention adjustment, and weighted mapping in "Step 2" increase computation due to the attention mechanism, it achieves higher accuracy fusion. Therefore, this scheme can be used for important semantic information, achieving a balance between stability and expressive power.

[0108] For example, for the six hidden noise spectrum vectors corresponding to the 2nd harmonic frequency to the 7th harmonic frequency, the two hidden noise spectrum vectors corresponding to the 7th harmonic frequency and the 6th harmonic frequency can be fused using the method described in "Step 1" above; the two hidden noise spectrum vectors corresponding to the 3rd harmonic frequency and the 2nd harmonic frequency can be fused using the method described in "Step 2" above; and the two hidden noise spectrum vectors corresponding to the 5th harmonic frequency and the 4th harmonic frequency can be fused using a cross-attention mechanism.

[0109] Furthermore, regarding the "second step" mentioned above, it should be noted that the size of the last hidden noise spectrogram vector can be n*1 (i.e., one column with multiple rows), and the size of the fusion output vector from the previous fusion stage can also be n*1. Thus, the last hidden noise spectrogram vector and the transpose (1*n) of the fusion output vector from the previous fusion stage can be multiplied to obtain an n*n matrix. Then, this matrix can be normalized using functions such as softmax to form an n*n attention parameter distribution. Furthermore, the last hidden noise spectrogram vector can be mapped to an n*1 gating parameter distribution. Then, based on the n*n attention parameter distribution, the n*1 gating parameter distribution can be weighted and summed to obtain an n*1 adjusted gating parameter distribution (thus achieving full integration of the gating and attention mechanisms). Finally, this n*1 adjusted gating parameter distribution is multiplied bitwise with the n*1 fusion output vector from the previous fusion stage to obtain the n*1 fusion output vector of the current fusion stage (or the n*1 first noise semantic vector).

[0110] In the second part, regarding step S120, it should be noted that the specific method for mining the semantic vector of the noise type is not restricted and can be selected according to actual needs.

[0111] For example, in one feasible implementation, the noise type semantic vector can be obtained by word embedding processing of the target text, wherein the target text contains a description of a noise type.

[0112] For example, in another feasible implementation, audio of various noise types can be collected, and then semantic mining can be performed on each audio file. The results of the semantic mining are then concatenated to obtain a noise type semantic vector. The semantic mining process can be referred to the explanation of step S110 above.

[0113] The third part, regarding step S130, should be noted that the specific method for performing noise type correlation analysis on the noise type semantic vector and the first noise semantic vector is not limited and can be selected according to actual needs.

[0114] For example, in one feasible implementation, the noise type semantic vector and the first noise semantic vector can be concatenated. Then, the concatenated vector can be convolved, and the result of the convolution can be non-linearly activated to obtain noise type related data. This data can then be used to filter semantic information from the second noise semantic vector.

[0115] For example, in another possible implementation, in order to ensure that the determined noise type related data can fully characterize the correlation between the noise type semantic vector and the first noise semantic vector, that is, to achieve sufficient constraint on the first noise semantic vector by the noise type semantic vector, the above step S130 can further include steps S131, S132 and S133, the specific contents of each step are as follows.

[0116] Step S131: Perform a first semantic space transformation on the noise type semantic vector to form a noise type transformation vector.

[0117] In this embodiment of the invention, the noise type semantic vector can be transformed using a first semantic space to form a noise type transformation vector. For example, the noise type semantic vector can be multiplied by a first weight matrix in the attention network to obtain the noise type transformation vector.

[0118] Step S132: Perform a second semantic space transformation on the first noise semantic vector to form a first noise transformation vector.

[0119] In this embodiment of the invention, the first noise semantic vector can be transformed using a second semantic space to form a first noise transformation vector. For example, the first noise semantic vector can be multiplied by a second weight matrix in the attention network to obtain the first noise transformation vector.

[0120] Step S133: Perform attention analysis on the noise type conversion vector and the first noise conversion vector to form the attention parameter distribution between the noise type semantic vector and the first noise semantic vector.

[0121] In this embodiment of the invention, attention analysis can be performed on the noise type conversion vector and the first noise conversion vector to form an attention parameter distribution between the noise type semantic vector and the first noise semantic vector. This attention parameter distribution serves as noise type-related data, thus identifying the noise type-related data. In other words, the correlation between the noise type semantic vector and the first noise semantic vector can be mined based on an attention mechanism.

[0122] It should be further noted that in step S133 above, the specific method of performing attention analysis on the noise type conversion vector and the first noise conversion vector is not limited. For example, in one feasible implementation, in order to ensure that the formed attention parameter distribution has high quality, step S133 above may further include steps S133a, S133b, S133c, and S133d, the specific contents of each step of which are as follows (in conjunction with...). Figure 5 ).

[0123] Step S133a: Multiply the vector formed by transposing the first noise conversion vector with the noise type conversion vector to form an initial correlation parameter distribution.

[0124] In this embodiment of the invention, the vector formed by transposing the first noise conversion vector and the noise type conversion vector can be multiplied to form an initial correlation parameter distribution.

[0125] Step S133b: Using the importance parameter distribution corresponding to each enhancement stage in the semantic enhancement process, map to form the intermediate relevant parameter distribution corresponding to each enhancement stage.

[0126] In this embodiment of the invention, the importance parameter distribution corresponding to each enhancement stage in the semantic enhancement process can be used to map and form the intermediate relevant parameter distribution corresponding to each enhancement stage. The importance parameter distribution serves as the network parameters of the corresponding neural network model and is formed during the learning and training process. (That is, the steps included in the noise monitoring method provided in this embodiment of the invention can be implemented based on this neural network model. This neural network model can be trained based on a first noise audio sample, a second noise audio sample, text samples of various noise types, and the actual noise type labels corresponding to the first and second noise audio samples. The training and learning process can refer to relevant existing technologies and is not specifically limited here.) For example, the importance parameter distribution corresponding to the first enhancement stage is used to map and form the intermediate relevant parameter distribution corresponding to the first enhancement stage; the importance parameter distribution corresponding to the second enhancement stage is used to map and form the intermediate relevant parameter distribution corresponding to the second enhancement stage; and the importance parameter distribution corresponding to the third enhancement stage is used to map and form the intermediate relevant parameter distribution corresponding to the third enhancement stage.

[0127] Step S133c: The intermediate correlation parameter distribution and the initial correlation parameter distribution corresponding to each enhancement stage are fused to form the target correlation parameter distribution corresponding to each enhancement stage.

[0128] In this embodiment of the invention, after forming the intermediate correlation parameter distribution and the initial correlation parameter distribution, the intermediate correlation parameter distribution and the initial correlation parameter distribution corresponding to each enhancement stage can be fused to form the target correlation parameter distribution corresponding to each enhancement stage. That is, not only is the correlation between the noise type semantic vector and the first noise semantic vector considered, but the correlation learned from the samples can also be fused, thus ensuring that the formed target correlation parameter distribution has high reliability.

[0129] Step S133d: Normalize and map the distribution of target-related parameters corresponding to each enhancement stage to form the distribution of attention parameters corresponding to each enhancement stage.

[0130] In this embodiment of the invention, after the target-related parameter distribution is formed, the target-related parameter distribution corresponding to each enhancement stage can be normalized and mapped (e.g., by using a function such as softmax) to form the attention parameter distribution corresponding to each enhancement stage.

[0131] It should be further explained that in step S133b above, the specific method of mapping to form the intermediate relevant parameter distribution corresponding to each enhancement stage is not limited. For example, in an alternative implementation, in order to ensure the high accuracy of the final attention parameter distribution, so as to better guide semantic enhancement in subsequent steps, the importance parameter distribution can be constrained based on the initial relevant parameter distribution to avoid the problem of introducing too much error due to a large deviation between the importance parameter distribution and the actual situation. Step S133b above can further include the following implementable contents:

[0132] The first step is to obtain the first correlation parameter and the second correlation parameter in the initial correlation parameter distribution, wherein the first correlation parameter is greater than the second correlation parameter. For example, the first correlation parameter may be the maximum value in the initial correlation parameter distribution, and the second correlation parameter may be the minimum value in the initial correlation parameter distribution.

[0133] The second step is to calculate the difference between the first relevant parameter and each relevant parameter in the initial relevant parameter distribution to form a first difference parameter distribution; that is, for any parameter in the first difference parameter distribution, the parameter is equal to the difference between the first relevant parameter and the parameter whose position is in the corresponding position in the initial relevant parameter distribution.

[0134] The second step is to calculate the difference between each relevant parameter in the initial relevant parameter distribution and the second relevant parameter to form a second difference parameter distribution; that is, for any parameter in the second difference parameter distribution, the parameter is equal to the difference between the parameter whose position is in the initial relevant parameter distribution and the second relevant parameter.

[0135] The third step is to determine the distribution of importance parameters for each enhancement stage, as described above;

[0136] Fourth, the first and second difference parameter distributions can be used as parameter intervals for the importance parameter distributions corresponding to each enhancement stage, to perform parameter mapping and obtain the intermediate relevant parameter distributions corresponding to each enhancement stage. For example, for each importance parameter in the importance parameter distribution, if the importance parameter is greater than the larger value between the corresponding parameter in the first and second difference parameter distributions, then the parameter at the corresponding position in the intermediate relevant parameter distribution is determined to be the larger value. If the importance parameter is less than the smaller value between the corresponding parameter in the first and second difference parameter distributions, then the parameter at the corresponding position in the intermediate relevant parameter distribution is determined to be the smaller value. If the importance parameter is greater than the smaller value between the corresponding parameter in the first and second difference parameter distributions, and less than the larger value between the corresponding parameter in the first and second difference parameter distributions, then the parameter at the corresponding position in the intermediate relevant parameter distribution is determined to be the importance parameter.

[0137] It should be further explained that in step S133c above, the specific method of fusing the intermediate relevant parameter distribution and the initial relevant parameter distribution corresponding to each enhancement stage is not limited. For example, in one feasible implementation, for the intermediate relevant parameter distribution corresponding to each enhancement stage, the intermediate relevant parameter distribution and the initial relevant parameter distribution corresponding to the enhancement stage are summed to form the target relevant parameter distribution corresponding to the enhancement stage.

[0138] In the fourth part, regarding step S140, it should be noted that the noise type-related data is used as the basis for semantic enhancement. The specific method of semantic enhancement of the second noise semantic vector is not limited and can be selected according to actual needs.

[0139] For example, in an alternative implementation, corresponding to the first implementation of step S130, the noise type-related data and the second noise semantic vector can be multiplied bitwise to form an enhanced noise semantic vector;

[0140] For example, in another achievable implementation, corresponding to the second implementation of step S130, the formed attention parameter distribution is used as noise type-related data. For each enhancement stage, the attention parameter distribution corresponding to the enhancement stage can be used as the basis for semantic enhancement. Semantic enhancement is performed on the noise semantic vector to be processed in the enhancement stage (i.e., based on the attention parameter distribution, a weighted summation is performed on the noise semantic vector to be processed; in some implementations, the result of the weighted summation can be added to the noise semantic vector to be processed to obtain an output vector). The output vector of the last enhancement stage is determined as the enhanced noise semantic vector, where the noise semantic vector to be processed in the first enhancement stage is the second noise semantic vector, and the noise semantic vector to be processed in each of the second and subsequent enhancement stages is the output vector of the previous enhancement stage. Figure 6 As shown.

[0141] In the fifth part, regarding step S150, it should be noted that the specific method for predicting the noise type based on the enhanced noise semantic vector is not limited and can be selected according to actual needs.

[0142] For example, in one feasible implementation, firstly, a fully connected mapping can be performed on the enhanced noise semantic vector to obtain a noise fully connected vector, wherein the dimension of the noise fully connected vector is equal to the number of each noise type; secondly, a classification mapping can be performed on the noise fully connected vector (e.g., using a function such as softmax) to form a noise type probability distribution, wherein each probability value in the noise type probability distribution corresponds to a noise type; finally, the noise type corresponding to the largest probability value in the noise type probability distribution can be determined as the target noise type, thus completing the prediction of the noise type.

[0143] In summary, in the noise monitoring system for urban noise pollution based on intelligent sensing provided by this invention, firstly, semantic mining can be performed on the first noise audio and the second noise audio respectively to form a first noise semantic vector and a second noise semantic vector; secondly, a noise type semantic vector can be mined; then, noise type correlation analysis can be performed on the noise type semantic vector and the first noise semantic vector to output noise type correlation data; furthermore, the noise type correlation data can be used as the basis for semantic enhancement to semantically enhance the second noise semantic vector to form an enhanced noise semantic vector; finally, noise type prediction can be performed based on the enhanced noise semantic vector to form the target noise type. Based on the above method, since two different types of noise monitoring sensors are used to collect noise audio for semantic mining and prediction, the basis is more sufficient, which can effectively improve the reliability of noise type prediction. Furthermore, since the semantic information mined from the noise audio is also analyzed based on the semantic information of each noise type, it is easier to mine or capture semantic information related to noise type prediction, thereby reducing semantic information unrelated to noise type prediction. This improves the accuracy of the basis for noise type prediction (i.e., the accuracy of semantic representation), and further improves the reliability of prediction, effectively addressing the problem of relatively low reliability of noise monitoring in existing technologies.

[0144] In the several embodiments provided in this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0145] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0146] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A noise monitoring system for smart sensor based urban noise pollution, characterized in that, The device comprises a processor and a memory, the processor is used to execute the computer program in the memory to realize the noise monitoring method, wherein the noise monitoring method comprises: respectively performing semantic mining on the first noise audio and the second noise audio to form a first noise semantic vector and a second noise semantic vector, wherein the first noise audio and the second noise audio are audio collected by two different types of noise monitoring sensors at the same time on a target space to form; mining a noise type semantic vector, wherein the noise type semantic vector is used to represent semantic information of each noise type; performing noise type correlation analysis on the noise type semantic vector and the first noise semantic vector to output noise type correlation data, wherein the noise type correlation data is used to reflect the correlation between each noise type and the first noise semantic vector; performing semantic enhancement on the second noise semantic vector according to the noise type correlation data as the semantic enhancement basis to form an enhanced noise semantic vector; performing noise type prediction according to the enhanced noise semantic vector to form a target noise type.

2. The smart-sensor based urban noise pollution noise monitoring system as claimed in claim 1, wherein, The step of performing semantic mining on the first noise audio and the second noise audio to form a first noise semantic vector and a second noise semantic vector comprises: performing Fourier transform on the first noise audio to form a first noise spectrum graph, and performing global and harmonic semantic mining on the first noise spectrum graph to form a first noise semantic vector; performing Fourier transform on the second noise audio to form a second noise spectrum graph, and performing global and harmonic semantic mining on the second noise spectrum graph to form a second noise semantic vector.

3. The smart-sensor based urban noise pollution noise monitoring system as claimed in claim 2, wherein, The step of performing Fourier transform on the first noise audio to form a first noise spectrum graph, and performing global and harmonic semantic mining on the first noise spectrum graph to form a first noise semantic vector comprises: performing Fourier transform on the first noise audio to form a first noise spectrum graph; identifying high-order harmonics of the first noise spectrum graph to obtain a plurality of high-order harmonic frequencies in the first noise spectrum graph; respectively performing local frequency hiding processing on the first noise spectrum graph based on each high-order harmonic frequency in the plurality of high-order harmonic frequencies to form a plurality of hidden noise spectrum graphs; respectively performing semantic space mapping on the first noise spectrum graph and each hidden noise spectrum graph to form a first noise spectrum graph vector corresponding to the first noise spectrum graph and a hidden noise spectrum graph vector corresponding to each hidden noise spectrum graph; fusing each hidden noise spectrum graph vector into the first noise spectrum graph vector to form a first noise semantic vector.

4. The smart-sensor based urban noise pollution noise monitoring system as claimed in claim 3, wherein, The step of fusing each hidden noise spectrum graph vector into the first noise spectrum graph vector to form a first noise semantic vector comprises: sorting a plurality of hidden noise spectrum graph vectors according to the descending order of the frequencies retained in the corresponding hidden noise spectrum graph to form a target vector sequence; In a plurality of fusion stages, sequentially fuse a plurality of the hidden noise spectrum graph vectors into the first noise spectrum graph vector according to the order in the target vector sequence to form a first noise semantic vector, wherein the number of fusion stages is equal to the number of hidden noise spectrum graph vectors.

5. The smart-sensor based urban noise pollution noise monitoring system as claimed in claim 4, wherein, The step of sequentially fusing a plurality of the hidden noise spectrum graph vectors into the first noise spectrum graph vector according to the order in the target vector sequence to form a first noise semantic vector in a plurality of fusion stages comprises: For a first fusion stage in the plurality of fusion stages, map a first hidden noise spectrum graph vector in the target vector sequence to a gating mapping parameter distribution of the first noise spectrum graph vector, and perform weighted mapping on the first noise spectrum graph vector based on the gating mapping parameter distribution to form a fusion output vector of the first fusion stage; For a last fusion stage in the plurality of fusion stages, map a last hidden noise spectrum graph vector in the target vector sequence to a gating mapping parameter distribution of a fusion output vector of a previous fusion stage, adjust the gating mapping parameter distribution based on an attention parameter distribution between the hidden noise spectrum graph vector and the fusion output vector to form an adjusted gating mapping parameter distribution, and perform weighted mapping on the fusion output vector based on the adjusted gating mapping parameter distribution to form a first noise semantic vector.

6. The smart-sensing based urban noise pollution noise monitoring system as claimed in claim 1, wherein, The step of performing noise type related analysis on the noise type semantic vector and the first noise semantic vector to output noise type related data comprises: performing first semantic space conversion on the noise type semantic vector to form a noise type conversion vector; performing second semantic space conversion on the first noise semantic vector to form a first noise conversion vector; performing attention analysis on the noise type conversion vector and the first noise conversion vector to form an attention parameter distribution between the noise type semantic vector and the first noise semantic vector, wherein the attention parameter distribution is taken as the noise type related data.

7. The smart-sensor based urban noise pollution noise monitoring system as claimed in claim 6, wherein, The step of performing attention analysis on the noise type conversion vector and the first noise conversion vector to form an attention parameter distribution between the noise type semantic vector and the first noise semantic vector comprises: performing multiplication operation on a vector formed by transposing the first noise conversion vector and the noise type conversion vector to form an initial related parameter distribution; mapping an intermediate related parameter distribution corresponding to each enhancement stage in the semantic enhancement process using an importance parameter distribution corresponding to each enhancement stage, wherein the importance parameter distribution is taken as network parameters of a corresponding neural network model and is formed in a learning and training process; fuse the intermediate related parameter distribution corresponding to each enhancement stage and the initial related parameter distribution respectively to form a target related parameter distribution corresponding to each enhancement stage; perform normalization mapping on the target related parameter distribution corresponding to each enhancement stage respectively to form an attention parameter distribution corresponding to each enhancement stage; and performing multiplication operation on a vector formed by transposing the first noise conversion vector and the noise type conversion vector to form an initial related parameter distribution; mapping an intermediate related parameter distribution corresponding to each enhancement stage in the semantic enhancement process using an importance parameter distribution corresponding to each enhancement stage, wherein the importance parameter distribution is taken as network parameters of a corresponding neural network model and is formed in a learning and training process; fuse the intermediate related parameter distribution corresponding to each enhancement stage and the initial related parameter distribution respectively to form a target related parameter distribution corresponding to each enhancement stage; perform normalization mapping on the target related parameter distribution corresponding to each enhancement stage respectively to form an attention parameter distribution corresponding to each enhancement stage; and performing multiplication operation on a vector formed by transposing the first noise conversion vector and the noise type conversion vector to form an initial related parameter distribution; The step of performing semantic enhancement on the second noise semantic vector according to the noise type related data as semantic enhancement basis to form an enhanced noise semantic vector comprises: For each enhancement stage, taking the attention parameter distribution corresponding to the enhancement stage as semantic enhancement basis to perform semantic enhancement on the noise semantic vector to be processed of the enhancement stage, and determining the output vector of the last enhancement stage as the enhanced noise semantic vector, wherein the noise semantic vector to be processed of the first enhancement stage is the second noise semantic vector, and the noise semantic vector to be processed of each second and subsequent enhancement stage is the output vector of the semantic enhancement of the previous enhancement stage.

8. The smart-sensor based urban noise pollution noise monitoring system as claimed in claim 7, wherein, The step of mapping to form the intermediate related parameter distribution corresponding to each enhancement stage according to the importance parameter distribution corresponding to each enhancement stage in the semantic enhancement process comprises: obtaining a first related parameter and a second related parameter in the initial related parameter distribution, wherein the first related parameter is greater than the second related parameter; performing difference calculation on the first related parameter and each related parameter in the initial related parameter distribution respectively to form a first difference parameter distribution; performing difference calculation on each related parameter in the initial related parameter distribution and the second related parameter respectively to form a second difference parameter distribution; determining the importance parameter distribution corresponding to each enhancement stage; taking the first difference parameter distribution and the second difference parameter distribution as the parameter interval of the importance parameter distribution corresponding to each enhancement stage to perform parameter mapping to obtain the intermediate related parameter distribution corresponding to each enhancement stage.

9. The smart-sensor based urban noise pollution noise monitoring system as claimed in claim 7, wherein, The step of fusing the intermediate related parameter distribution corresponding to each enhancement stage and the initial related parameter distribution respectively to form the target related parameter distribution corresponding to each enhancement stage comprises: performing sum calculation on the intermediate related parameter distribution corresponding to each enhancement stage and the initial related parameter distribution to form the target related parameter distribution corresponding to the enhancement stage.

10. The smart-sensor based urban noise pollution noise monitoring system as claimed in any one of the claims 1 to 9, wherein, The step of performing noise type prediction according to the enhanced noise semantic vector to form a target noise type comprises: performing full connection mapping on the enhanced noise semantic vector to obtain a noise full connection vector, wherein the dimension of the noise full connection vector is equal to the number of noise types; performing classification mapping on the noise full connection vector to form a noise type probability distribution, wherein each probability value in the noise type probability distribution corresponds to a noise type; determining the noise type corresponding to the maximum probability value in the noise type probability distribution as the target noise type.

Images