Edge ai-based city noise analysis method, device, equipment and storage medium

Abstract

The application provides an urban noise analysis method and device based on edge AI, equipment and a medium. The method comprises the following steps: sending target noise data collected by a sound collecting device to an edge AI device; inputting the target noise data into a pre-trained scene adaptation model by the edge AI device; determining a scene analysis result of the target noise data by the scene adaptation model, wherein the scene analysis result comprises audio features and scene features corresponding to the audio features; sending the scene analysis result to an intelligent analysis device by the edge AI device; determining a target prevention strategy based on the scene analysis result and a preset knowledge graph by the intelligent analysis device, wherein the target prevention strategy is a prevention strategy for the target noise data, and the knowledge graph comprises historical cases, and the historical cases record scene analysis results of historical noise data and corresponding historical prevention strategies. According to the technical scheme of the embodiment of the application, the analysis result of the target noise data can be output in real time.

Description

Urban noise analysis methods, devices, equipment, and storage media based on edge AI Technical Field

[0001] This invention relates to the field of noise analysis technology, and in particular to a method, apparatus, device, and storage medium for urban noise analysis based on edge AI. Background Technology

[0002] With the rapid development of society and the acceleration of urbanization, urban noise pollution has become one of the important factors affecting human health. How to analyze noise and obtain noise analysis results has become a key point of urban management.

[0003] Currently, multiple sound acquisition devices are installed at different locations within the city, and the cloud platform communicates with all of these devices. Noise data from various locations within the city is acquired through these devices and uploaded to the cloud. The cloud then analyzes the noise data to obtain the noise analysis results. However, the large number of sound acquisition devices in the city and the large volume of noise data that the cloud needs to process, coupled with the long processing time, prevent the real-time availability of noise analysis results. Summary of the Invention

[0004] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a method, apparatus, device, and storage medium for urban noise analysis based on edge AI. It can analyze target noise data in real time through edge AI devices and intelligent analysis devices to determine the scene analysis results and output target prevention and control strategies, which are noise prevention and control strategies specifically for target noise data.

[0005] In a first aspect, embodiments of the present invention provide a method for urban noise analysis based on edge AI, applied to an urban noise analysis system. The urban noise analysis system includes an intelligent analysis device and multiple noise acquisition devices. The noise acquisition devices include sound acquisition equipment and edge AI devices. The edge AI devices are communicatively connected to the intelligent analysis device. The method includes:

[0006] The sound acquisition device sends the acquired target noise data to the edge AI device. The edge AI device inputs the target noise data into a pre-trained scene adaptation model. The scene adaptation model determines the scene analysis result of the target noise data. The scene analysis result includes audio features and scene features corresponding to the audio features.

[0007] The edge AI device sends the scene analysis results to the intelligent analysis device. The intelligent analysis device determines the target prevention strategy based on the scene analysis results and a preset knowledge graph. The target prevention strategy is a prevention strategy for the target noise data. The knowledge graph includes historical cases, which record the scene analysis results of historical noise data and the corresponding historical prevention strategies.

[0008] According to some embodiments of the present invention, after the edge AI device inputs the target noise data into a pre-trained scene adaptation model, the process includes:

[0009] The edge AI device acquires target-related data associated with the target noise data, the target-related data including target meteorological data and target video feature data associated with the target noise data;

[0010] The edge AI device inputs the target noise data and the target association data into the scene adaptation model to obtain the scene features.

[0011] According to some embodiments of the present invention, the edge AI device is communicatively connected to both a weather instrument and a vision device;

[0012] The edge AI device acquires target association data associated with the target noise data, including:

[0013] The edge AI device acquires the target meteorological data uploaded by the meteorological instrument, and the edge AI device acquires the video data uploaded by the vision device;

[0014] The edge AI device performs image AI recognition on the video data to obtain the target video feature data.

[0015] According to some embodiments of the present invention, before sending the collected target noise data to the edge AI device via the sound acquisition device, the method further includes:

[0016] The noise query information is input into the pre-trained LLM model in the intelligent analysis device, and the LLM model obtains the query keywords of the noise query information.

[0017] When the query keyword includes the target location, the intelligent analysis device obtains the target sound acquisition device and obtains historical audio data based on the target sound acquisition device, wherein the target sound acquisition device is the sound acquisition device located at the target location;

[0018] When the query keyword includes a target time range, the target sound acquisition device extracts the target noise data based on the historical audio data, wherein the target noise data is the audio data of the historical audio data within the target time range.

[0019] According to some embodiments of the present invention, the knowledge graph includes a noise source feature library and a regulatory library, wherein the noise source feature library includes audio features of various noise sources, and the regulatory library includes noise pollution standards and noise pollution regulations;

[0020] After the scene analysis results are sent to the intelligent analysis device via the edge AI device, the method further includes:

[0021] The intelligent analysis device determines the noise source and occurrence scenario of the target noise data based on the audio features and the noise source feature library.

[0022] Based on the noise source, the occurrence scenario, and the regulatory database, the control requirements corresponding to the target noise data are determined.

[0023] According to some embodiments of the present invention, after determining the control requirements corresponding to the target noise data, the method further includes:

[0024] Obtain the control requirements and determine the initial prevention and control strategy based on the control requirements;

[0025] Analyze the changing trend of the target noise data to obtain noise trend analysis results;

[0026] Based on the scenario analysis results and the noise trend analysis results, the knowledge graph determines matching cases from the historical cases, and the matching cases include the historical prevention and control strategies.

[0027] The initial prevention and control strategy is modified based on the historical prevention and control strategies, and the target prevention and control strategy is obtained based on the modification results.

[0028] According to some embodiments of the present invention, before the edge AI device inputs the target noise data into the pre-trained scene adaptation model, the method further includes:

[0029] Acquire the historical noise data of all the sound acquisition devices, and perform vectorization processing on the historical noise data to obtain the main feature vector;

[0030] The historical noise data is obtained by acquiring historical associated data, and the historical associated data is vectorized to obtain subordinate feature vectors. The historical associated data includes historical meteorological data, historical video feature data, and scene data associated with the historical noise data. The scene data includes noise scenes that match the target noise data, the historical meteorological data, and the historical video feature data.

[0031] The primary feature vector and the subordinate feature vector are input into a deep learning model for training to obtain the scenario-based adaptation model.

[0032] Secondly, embodiments of the present invention provide an urban noise analysis device based on edge AI, including at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, the instructions being executed by the at least one control processor to enable the at least one control processor to perform the urban noise analysis method based on edge AI as described in the first aspect above.

[0033] Thirdly, embodiments of the present invention provide an electronic device including an urban noise analysis device based on edge AI as described in the second aspect above.

[0034] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions for performing the urban noise analysis method based on edge AI as described in the first aspect above.

[0035] The urban noise analysis method based on edge AI according to embodiments of the present invention has at least the following beneficial effects: The target noise data collected by the sound acquisition device is sent to the edge AI device; the edge AI device inputs the target noise data into a pre-trained scenario-based adaptation model; the scenario-based adaptation model determines the scenario analysis result of the target noise data, wherein the scenario analysis result includes audio features and scenario features corresponding to the audio features; the scenario analysis result is sent to the intelligent analysis device by the edge AI device; the intelligent analysis device determines the target prevention strategy based on the scenario analysis result and a preset knowledge graph; the target prevention strategy is the prevention strategy for the target noise data; wherein the knowledge graph includes historical cases, and the historical cases record the scenario analysis results of historical noise data and the corresponding historical prevention strategies. According to the technical solution of the present invention, the sound acquisition device sends target noise data to the edge AI device. The scene adaptation model of the edge AI device determines the scene analysis result based on the target noise data. The intelligent analysis device determines the target prevention and control strategy of the target noise data based on the scene analysis result. By performing audio analysis and scene analysis on the target noise data through the edge AI device and the intelligent analysis device, the edge AI device in the urban noise analysis system can output the scene analysis result of the target noise data, and the intelligent analysis device in the urban noise analysis system can output the prevention and control strategy of the target noise data in real time based on the scene analysis result. Attached Figure Description

[0036] Figure 1 is a schematic diagram of the urban noise analysis system provided in an embodiment of the present invention;

[0037] Figure 2 is a flowchart of an urban noise analysis method based on edge AI provided in another embodiment of the present invention;

[0038] Figure 3 is a structural diagram of an urban noise analysis device based on edge AI provided in another embodiment of the present invention. Detailed Implementation

[0039] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0040] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0041] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0042] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0043] The urban noise analysis method based on edge AI according to embodiments of the present invention has at least the following beneficial effects: The target noise data collected by the sound acquisition device is sent to the edge AI device; the edge AI device inputs the target noise data into a pre-trained scenario-based adaptation model; the scenario-based adaptation model determines the scenario analysis result of the target noise data, wherein the scenario analysis result includes audio features and scenario features corresponding to the audio features; the scenario analysis result is sent to the intelligent analysis device by the edge AI device; the intelligent analysis device determines the target prevention strategy based on the scenario analysis result and a preset knowledge graph; the target prevention strategy is the prevention strategy for the target noise data; wherein the knowledge graph includes historical cases, and the historical cases record the scenario analysis results of historical noise data and the corresponding historical prevention strategies. According to the technical solution of the present invention, the sound acquisition device sends target noise data to the edge AI device. The scene adaptation model of the edge AI device determines the scene analysis result based on the target noise data. The intelligent analysis device determines the target prevention and control strategy of the target noise data based on the scene analysis result. By performing audio analysis and scene analysis on the target noise data through the edge AI device and the intelligent analysis device, the edge AI device in the urban noise analysis system can output the scene analysis result of the target noise data, and the intelligent analysis device in the urban noise analysis system can output the prevention and control strategy of the target noise data in real time based on the scene analysis result.

[0044] First, refer to Figure 1, which is a schematic diagram of the urban noise analysis system provided in an embodiment of the present invention.

[0045] As shown in Figure 1, the urban noise analysis system of this embodiment includes an intelligent analysis device 10 and multiple noise acquisition devices 20. The noise acquisition device 20 includes a sound acquisition device 22 and an edge AI device 21. The edge AI device 21 is communicatively connected to the intelligent analysis device 10.

[0046] It should be noted that the edge AI device 21 applies edge AI technology and is a device that performs artificial intelligence calculations and processing near the data source. In this application, the edge AI device 21 is communicatively connected to the sound acquisition device 22 to perform artificial intelligence calculations and processing on the audio data of the sound acquisition device 22. The edge AI device 21 runs a pre-trained scene-based adaptation model. By executing scene-based adaptation model inference on the edge AI device 21, the target noise data of the sound acquisition device 22 is identified and classified, reducing the latency and bandwidth consumption of uploading the target noise data to the cloud, reducing the amount of data required for cloud computing, and realizing real-time, low-latency data processing and intelligent decision-making.

[0047] It should be noted that the edge AI device 21 is communicatively connected to both the sound acquisition device and the intelligent analysis device 10. The edge AI device 21 can be integrated with the sound acquisition device 22, or it can exist independently as a dedicated edge node. Through the edge AI device 21, source tracing and monitoring of target noise data are achieved. By analyzing and processing the target noise data acquired by the sound acquisition device, the edge AI device 21 can trace the source and propagation path of the target noise data, accurately identify and locate the noise source of the target noise data, thereby providing an analytical basis and scientific evidence for noise control of the target noise data.

[0048] It should be noted that an intelligent analysis device 10 is communicatively connected to multiple noise acquisition devices 20. The intelligent analysis device 10 acquires the audio data analysis results uploaded by multiple edge AI devices 21, performs secondary analysis, and obtains a prevention and control strategy for the target noise data. Generally, multiple noise acquisition devices 20 located at the same location are communicatively connected to an intelligent analysis device 10. The intelligent analysis device 10 can acquire the scene analysis results uploaded by the edge AI devices 21 and, based on the acquired scene analysis results from multiple devices located at the same location, obtain a prevention and control strategy for the target noise data, thereby enhancing the data processing capabilities and decision-making accuracy of the urban noise analysis system.

[0049] It should be noted that the intelligent analysis device 10 uses knowledge fusion technology to analyze the scene analysis results; knowledge fusion refers to combining knowledge from different fields with multiple sources and multiple modalities to form a comprehensive and complementary information system. In this application, knowledge fusion mainly involves the combination of noise data and noise knowledge graph in the field of noise monitoring.

[0050] Additionally, referring to Figure 1, the edge AI device 21 is communicatively connected to the weather instrument 23 and the vision device 24, respectively.

[0051] It should be noted that, in order to further enhance the accuracy of the scene analysis results output by the edge AI device 21, the edge AI device 21 is communicatively connected to the weather instrument 23 and the vision device 24 respectively. That is, a noise acquisition device 20 includes an edge AI device 21 connected to the intelligent analysis device 10, and a sound acquisition device, a weather instrument 23 and a vision device 24 are communicatively connected to the edge AI device 21 respectively. This ensures that the audio data, meteorological data and visual data acquired by a noise acquisition device 20 are all located in the same location, and the scene analysis results uploaded by the edge AI device 21 to the intelligent analysis device 10 are all analysis results located in the same location. This avoids the intelligent analysis device acquiring multiple unrelated scene analysis results, which could lead to the output of incorrect prevention and control strategies.

[0052] It should be noted that knowledge graphs serve as the foundation for knowledge fusion. By combining knowledge fusion with intelligent decision support, urban noise analysis systems can intelligently analyze the characteristics and potential hazards of noise sources in target noise data during noise monitoring, optimize noise prevention strategies, provide decision support for management departments, and further improve the accuracy of target noise data analysis.

[0053] It should be noted that the knowledge graph includes various sub-knowledge bases or function bases and model bases, which may include: noise source feature base, noise standard and regulation base, noise source knowledge graph base, business model base and historical noise database;

[0054] The noise source feature library contains location information and surrounding conditions, frequency range, intensity distribution, spatial distribution, and spatiotemporal distribution of various noise sources, which can be matched with audio features in the scene analysis results output by the edge AI device 21. The noise source feature library includes noise source feature information obtained by the edge AI device 21 through AI intelligent recognition, as well as feature data obtained by mature feature extraction methods;

[0055] The noise standards and regulations database, as defined in this application, includes relevant noise pollution standards and regulations for the cities where the urban noise analysis system is applied, providing a basis for determining whether noise levels exceed standards. Based on scenario analysis results and data from the noise source characteristic database, the noise standards and regulations database assesses the compliance of target noise data.

[0056] The noise source knowledge graph library includes a relational pairing data graph established by combining laws, regulations, standards and technical logic from the entire data process of sound source investigation, data collection, storage, analysis and management. It can extract key data element attributes of each part of the entire process.

[0057] The business model library includes proven and usable classic noise management cases, extracted from successful and uncontroversial noise management practices. These cases include audio characteristics, scene characteristics, compliance, and practical noise control strategies for the target noise data. The library retrieves matching noise management cases by searching the library using scene analysis results, target noise data compliance data, and noise source matching results. The actual noise control strategies from these output noise management cases are then used to refine existing noise control strategies for the target noise data.

[0058] The historical noise database includes historical monitoring data and trends from multiple noise acquisition devices 20 connected to the intelligent analysis device 10, thereby providing background information for noise analysis.

[0059] The technical solution of the present invention will be further described below based on the urban noise analysis system shown in Figure 1.

[0060] Referring to Figure 2, which is a flowchart of an urban noise analysis method based on edge AI provided by an embodiment of the present invention, the urban noise analysis method based on edge AI includes, but is not limited to, the following steps:

[0061] S10, the target noise data collected by the sound acquisition device 22 is sent to the edge AI device 21. The edge AI device 21 inputs the target noise data into the pre-trained scene adaptation model and determines the scene analysis result of the target noise data through the scene adaptation model. The scene analysis result includes audio features and scene features corresponding to the audio features.

[0062] S20, the scene analysis results are sent to the intelligent analysis device 10 through the edge AI device 21. The intelligent analysis device 10 determines the target prevention strategy based on the scene analysis results and the preset knowledge graph. The target prevention strategy is the prevention strategy for target noise data. The knowledge graph includes historical cases, which record the scene analysis results of historical noise data and the corresponding historical prevention strategies.

[0063] It should be noted that the sound acquisition device 22 can be a microphone, set at different distribution points in the same location, so that multiple sound acquisition devices 22 in the same location can obtain multiple target noise data. Multiple edge AI devices 21 can acquire multiple target noise data respectively to obtain scene analysis results. The intelligent analysis device 10 can acquire multiple scene analysis results and obtain a more accurate noise prevention strategy based on the multiple scene analysis results.

[0064] It should be noted that the scenario-based adaptation model is a pre-trained deep learning model. Based on the acquired target noise data, the scenario-based adaptation model can determine the audio features and scenario features of the target noise data. The audio features of the target noise data include relevant audio parameters such as the decibel level, frequency, and duration of the noise. The scenario features include environmental feature information of the target noise data, such as the sound of rain, snow, thunder, and other relevant scenario features in the environment.

[0065] It should be noted that the edge AI device 21 runs a pre-trained scenario-based adaptation model, which is a deep learning model trained on a more powerful computing node deployed in the cloud or locally. The scenario-based adaptation model has remote iterative upgrade capabilities. The trained scenario-based adaptation model can be executed in real time on the edge AI device 21 to classify and identify noise types, events, or scenarios, supporting comprehensive judgment by the subsequent intelligent analysis device 10. The scenario-based adaptation model can automatically detect and classify data uploaded from a data acquisition device (such as the sound acquisition device 22 of this application) that is connected to the edge AI device 21, and upload the recognition result sequence. It also supports remote or cloud-based model download upgrades, enabling automatic upgrades of the scenario-based adaptation model based on the edge AI device 21.

[0066] It should be noted that by using edge AI device 21 to achieve edge computing, the urban noise analysis system can realize real-time target noise data processing, reduce cloud computing latency and cloud computing data volume, improve the response speed of noise pollution control, and achieve real-time and efficient urban noise analysis.

[0067] It should be noted that the intelligent analysis device 10 realizes a comprehensive source tracing construction scheme that integrates multimodal noise monitoring data with a knowledge graph related to noise. This scheme supports efficient intelligent decision-making and reasoning, automatically generates intelligent noise monitoring data based on target noise data, and outputs control and management schemes and strategies for the target noise data, thereby greatly improving the effectiveness of noise prevention and control.

[0068] It should be noted that by transferring the processing of target noise data to the device end through edge AI device 21 and intelligent analysis device 10, the data transmission latency between the cloud and sound acquisition device 22 is greatly reduced, ensuring real-time analysis and feedback of target noise data, and improving the real-time performance and response speed of the prevention and control strategies obtained from noise data analysis.

[0069] It should be noted that by combining the edge AI device 21 and the intelligent analysis device 10 with a scenario-based adaptation model and knowledge graph, the urban noise analysis system can accurately identify noise sources and effectively identify and trace the noise sources of target noise data. This provides higher noise monitoring accuracy and more reliable decision support, offering intelligent, personalized, and targeted prevention and control strategies for noise governance. It also improves the effectiveness of noise monitoring result analysis and pollution prevention and control, addressing the shortcomings of traditional noise monitoring technologies in terms of real-time performance, monitoring accuracy, and intelligent decision-making. Furthermore, it can provide more efficient and accurate noise prevention and control strategy support for noise pollution prevention and control of target noise data.

[0070] It should be noted that the edge AI device 21 acquires target noise data sent by the sound acquisition device 22. The scene adaptation model running in the edge AI device 21 acquires the target noise data and determines the scene analysis result based on the target noise data. The intelligent analysis device 10 acquires the scene analysis result of the target noise data. Based on the audio features and scene features in the scene analysis result, it searches the pre-set knowledge graph in the intelligent analysis device 10 to obtain the target prevention and control strategy. By adding the edge AI device 21 and the intelligent analysis device 10, the amount of data required for cloud computing is shared. This not only reduces the time for uploading data to the cloud but also allows for simultaneous processing and analysis of multiple data segments to obtain scene analysis results. The intelligent analysis device 10 only needs to output the target prevention and control strategy based on multiple scene results, enabling the urban noise analysis system to output the prevention and control strategy of the target noise data in real time. Furthermore, since the target prevention and control strategy is generated based on the target noise data, the prevention and control effect is better.

[0071] In another embodiment, after the edge AI device 21 inputs the target noise data into the pre-trained scene adaptation model in step S10, the following steps are included, but are not limited to:

[0072] S11, after the edge AI device 21 inputs the target noise data into the pre-trained scene-based adaptation model, includes:

[0073] S12, the edge AI device 21 acquires target association data associated with the target noise data, the target association data including target meteorological data and target video feature data associated with the target noise data;

[0074] S13, the edge AI device 21 inputs the target noise data and target association data into the scene adaptation model to obtain scene features.

[0075] It should be noted that the target-related data, which is associated with the target noise data, is input together with the target noise data into the scene analysis results, thereby further improving the accuracy of the scene analysis results output by the scene adaptation model. For example, the target noise data is the construction noise of the first street from 9:00 AM to 10:00 AM on January 1, 2025. The edge AI device 21, which is connected to the sound acquisition device 22 that outputs the target noise data, acquires historical meteorological data and historical video feature data from 9:00 AM to 10:00 AM on January 1, 2025. The resulting data is the target meteorological data and the target video feature data.

[0076] It should be noted that the historical meteorological data refers to all meteorological data acquired by the meteorological instrument 23, including temperature, humidity, air pressure, wind speed, and precipitation at the location of the sound acquisition device 22. The historical meteorological data is based on the historical noise data timeline of the sound acquisition device, thereby achieving matching between historical meteorological data and historical noise data, facilitating the acquisition of target meteorological data along with target noise data. For example, the target noise data is the construction noise of the first street from 9:00 AM to 10:00 AM on January 1, 2025. The edge AI device 21, which is communicatively connected to the sound acquisition device 22 that outputs the target noise data, acquires historical meteorological data, extracts the historical meteorological data from 9:00 AM to 10:00 AM on January 1, 2025, to obtain the target meteorological data. The target meteorological data indicates that the temperature at the location of the sound acquisition device from 9:00 AM to 10:00 AM on January 1, 2025, is 25°C, the humidity is 50%, and the precipitation is 10 mm.

[0077] It should be noted that historical video data refers to all video data acquired by the vision device 24, and historical video feature data refers to the recognition parameters output by the edge AI device 21 when performing image AI recognition on the historical video data, such as rain and snow day recognition parameters or other characterization results strongly correlated with noise monitoring services; target video feature data refers to video surveillance image data associated with target noise data; for example, the target noise data is the construction noise of the first street from 9:00 AM to 10:00 AM on January 1, 2025, and the edge AI device 22, which is communicatively connected to the sound acquisition device 22 that outputs the target noise data, is... Example 21: Obtain historical video feature data, extract historical video feature data from 9:00 AM to 10:00 AM on January 1, 2025, to obtain target video feature data. The target video feature data indicates that the environment in which the sound acquisition device was located from 9:00 AM to 10:00 AM on January 1, 2025 was a rainy environment; or, if the user only inputs January 1, 2025, the target noise data and target association data for January 1, 2025 will be output; if the user inputs only January 1, 2025, the target noise data and target association data for January 1, 2025 will be output; if the user inputs 9:00 AM on January 1, 2025, the target noise data and target association data for January 1, 2025 from 9:00 AM to 10:00 AM on January 1, 2025 will be output.

[0078] It should be noted that target-related data may also include target meteorological data and target video feature data, as well as environmental survey data, noise measurement and monitoring data, sound source location data, traffic flow data, and any other general data that may be collected in noise monitoring.

[0079] In another embodiment, in step S12, the edge AI device 21 acquires target association data associated with the target noise data, which specifically includes, but is not limited to, the following steps:

[0080] S121, the edge AI device 21 acquires target association data associated with the target noise data, including:

[0081] S122, edge AI device 21 acquires target meteorological data uploaded by meteorological instrument 23, and edge AI device 21 acquires video data uploaded by vision device 24;

[0082] S123, the edge AI device 21 performs image AI recognition on the video data to obtain target video feature data.

[0083] It should be noted that the edge AI device 21 acquires target meteorological data and, based on the target meteorological data, acquires the noise environment meteorology when the target noise data occurs.

[0084] It should be noted that the video data uploaded by the vision device 24 is analog data. The edge AI device 21 acquires the video data, performs analog-to-digital conversion on the video data to obtain digital video data, performs image AI recognition on the digital video signal, determines the environment of the noisy environment, and outputs video feature signals, which are the scene of the noisy environment.

[0085] In another embodiment, before sending the collected target noise data to the edge AI device 21 via the sound acquisition device 22 in step S10, the following steps are included, but are not limited to:

[0086] S01, before sending the collected target noise data to the edge AI device 21 via the sound acquisition device 22, it also includes:

[0087] S02, input the noise query information into the pre-trained LLM model in the intelligent analysis device 10, and the LLM model obtains the query keywords of the noise query information;

[0088] S03, when the query keyword includes the target location, the intelligent analysis device 10 obtains the target sound acquisition device 22, and obtains historical audio data based on the target sound acquisition device 22, wherein the target sound acquisition device 22 is the sound acquisition device 22 located at the target location;

[0089] S04, when the query keyword includes the target time range, the target sound acquisition device 22 extracts the target noise data based on the historical audio data, wherein the target noise data is the audio data of the historical audio data within the target time range.

[0090] It should be noted that the intelligent analysis device 10 is installed at the integrated node of the edge AI device 21 or the cloud integrated node. The urban analysis system uses the noise query information input by the user as the trigger for the urban analysis system, and uses the scene analysis results of the edge AI device 21 as the basic query data source for the intelligent agent's comprehensive traceability and judgment.

[0091] It should be noted that the intelligent analysis device 10 selects an open-source Large Language Model (LLM model) for private or public deployment. When the intelligent analysis device can be connected to a network, the LLM model achieves better performance. This invention does not restrict the selection of the LLM model; a suitable LLM model can be selected based on actual needs and characteristics. The LLM model is used to obtain query keywords based on user-input noise query information, providing a query basis for subsequently obtaining target noise data from historical noise data, target meteorological data from historical meteorological data, and target video feature data from historical video feature data.

[0092] It should be noted that the noise query information is in text form and includes at least one piece of information about the noise environment or scene, such as noise type, source, and target time range. The more details of the noise query information input, the more accurate the target noise data obtained by the edge AI device 21, the higher the accuracy of the output scene analysis results, and the better the implementation effect of the prevention and control strategy output by the intelligent analysis device 10.

[0093] It should be noted that an intelligent analysis device 10 is communicatively connected to multiple noise acquisition devices 20 at the same location. For example, the noise acquisition devices 20 of the first community include noise acquisition devices 20 located on the first street, the second street, and the third street. In order to obtain the required target noise data more accurately, the required target noise data is precisely filtered by querying keywords to avoid interference from other low-relevance noise data.

[0094] For example, the intelligent analysis device 10 is communicatively connected to multiple noise collection devices 20 located on the first street, the second street, and the third street. The user inputs the construction noise on the first street from 9:00 AM to 10:00 AM on January 1, 2025. The LLM model captures the query keywords, with the target location being the first street and the target time range being from 9:00 AM to 10:00 AM on January 1, 2025. The intelligent analysis device 10 obtains the target noise collection device 20, which is the noise collection device 20 located on the first street. The target noise collection device includes the target sound collection device. The intelligent analysis device 10 obtains the historical audio data of the target sound collection device. Based on the target time range of 9:00 AM to 10:00 AM on January 1, 2025 in the query keywords, the device obtains the audio data of the historical audio data from 9:00 AM to 10:00 AM on January 1, 2025. This audio data is the target noise data. The edge AI device 21 outputs the scene analysis results based on the target noise data.

[0095] In another embodiment, after sending the scene analysis results to the intelligent analysis device 10 via the edge AI device 21 in step S20, the following steps are included, but are not limited to:

[0096] S21, the intelligent analysis device 10 determines the noise source and occurrence scenario of the target noise data based on audio features and a noise source feature library;

[0097] S22, based on the noise source, occurrence scenario, and regulatory database, determines the control requirements corresponding to the target noise data.

[0098] It should be noted that the intelligent analysis device 10 acquires scene analysis results, which include audio features and scene features. Based on the audio features, the intelligent analysis device 10 queries the noise source feature database to determine the noise source of the target noise data. The noise source must occur in a noise environment that meets certain conditions. Therefore, the occurrence scenario of the target noise data is determined based on the noise source. Based on the noise source, occurrence scenario, and regulatory database, the corresponding control requirements for the target noise data are determined.

[0099] It should be noted that the intelligent analysis device 10 controls the edge AI device 21 to perform data retrieval based on noise query information, and obtain target noise data and target-related data; the data retrieval includes historical noise data, environmental sensor data, scene-specific feature data and other external data sources, and the data retrieval range is selected according to the performance of the edge AI device 21.

[0100] For example, a user inputs the noise level of the first street from 9:00 AM to 10:00 AM on January 1, 2025. The edge AI device 21 acquires the target noise data and target-related data from the sound acquisition device 22 on the first street from 9:00 AM to 10:00 AM on January 1, 2025. The target noise data and target-related data are then input into the scene-based adaptation model to obtain scene analysis results. The intelligent analysis device 10 determines the scene of the target noise data as a construction site and the noise source as an excavator on the construction site based on audio features and a noise source feature library. The audio features include noise-related data such as the decibel level of the target noise data. Based on the noise source, the scene of occurrence, and the legal library, the daytime construction noise of the construction site in the city should not exceed 70 decibels. However, the audio features indicate that the target noise data is 85 decibels, which exceeds the standard for construction noise. Therefore, the target noise data can be controlled through prevention and control strategies such as setting up noise reduction devices, negotiating with the construction site manager, or penalizing the construction site.

[0101] In another embodiment, after determining the control requirements corresponding to the target noise data in step S22, the steps include, but are not limited to, the following:

[0102] S23, Obtain control requirements and determine initial prevention and control strategies based on control requirements;

[0103] S24, Analyze the changing trend of the target noise data to obtain the noise trend analysis results;

[0104] S25, the knowledge graph determines matching cases from historical cases based on the scene analysis results and noise trend analysis results, and the matching cases include historical prevention and control strategies;

[0105] S26, Based on historical prevention and control strategies, the initial prevention and control strategy is modified, and the target prevention and control strategy is obtained based on the modification results.

[0106] It should be noted that prevention strategies are determined based on control measures. For example, if the target noise data is not in violation, then a negotiation approach is adopted to prevent and control the target noise. If the target noise data is in violation, then penalties can be used to prevent and control the target noise.

[0107] It should be noted that, through big data analytics, trend analysis is performed on historical noise data, including target noise data, to predict the development trend of the target noise data.

[0108] It should be noted that the intelligent analysis device 10 acquires scene analysis results and has a preset matching comparison algorithm. When the matching comparison algorithm reaches a preset threshold, it outputs a matching case.

[0109] It should be noted that if there are no historical cases in the knowledge graph that match the scene analysis results and noise trend analysis results, there is no need to modify the initial prevention and control strategy. Instead, the initial prevention and control strategy is defined as the target prevention and control strategy and output.

[0110] The complete workflow of the intelligent analysis device 10 is demonstrated through the following embodiments:

[0111] S2001, the intelligent analysis device 10 is triggered by the noise query information input by the user. The intelligent analysis device 10 controls the edge AI device 21 to perform data retrieval based on the noise query information and obtain the target noise data and target related data.

[0112] S2002, the edge AI device 21 obtains the scene analysis results of the target noise data based on the target noise data, and the edge AI device 21 uploads the scene analysis results to the intelligent analysis device 10;

[0113] S2003, the intelligent analysis device 10 acquires analysis results from multiple scenarios, combines them with the noise source feature library and regulatory library in the preset knowledge graph, determines the noise source and compliance of the target noise data, and determines control requirements based on compliance;

[0114] S2004, Determine the initial prevention and control strategy for target noise data based on control requirements;

[0115] S2005 uses big data analytics to perform trend analysis on historical noise data, including target noise data, to obtain noise trend analysis results.

[0116] S2006 uses the noise trend analysis results, the noise sources of the target noise data, and their compliance as the query basis for the business model library of the knowledge graph to obtain matching cases, which include historical prevention and control results.

[0117] S2007, based on the matching cases, the initial prevention and control strategy is modified to obtain the target prevention and control strategy.

[0118] In another embodiment, in step S10, before the edge AI device 21 inputs the target noise data into the pre-trained scene adaptation model, the following steps are included, but are not limited to:

[0119] S31, acquire historical noise data from all sound acquisition devices 22, and perform vectorization processing on the historical noise data to obtain the main feature vector;

[0120] S32, acquire historical association data of historical noise data, perform vectorization processing on historical association data to obtain subordinate feature vectors, wherein historical association data includes historical meteorological data, historical video feature data and scene data associated with historical noise data, and scene data includes noise scenes matched with target noise data, historical meteorological data and historical video feature data;

[0121] S33, input the main feature vector and subordinate feature vector into the deep learning model for training, and obtain the scene-adaptive model.

[0122] It should be noted that more powerful computing nodes are deployed in the cloud or locally for training the deep learning model. The computing nodes acquire historical noise data, which is audio data. This historical noise data is vectorized using methods such as MFCC and Spectrogram to obtain the main feature vector. The computing nodes also acquire historical associated data, including historical meteorological data, historical video feature data, and scene-based data associated with the historical noise data. Scene-based data includes noise scenes matched with the target noise data, historical meteorological data, and historical video feature data. For textual data in the historical associated data, it is vectorized using methods such as TF-IDF, Word2Vec, and BERT. For images, it is vectorized using methods such as pixel values, HOG, SIFT, and CNN, thus obtaining subordinate feature vectors based on the historical associated data. The main and subordinate feature vectors are then input into a deep learning model such as a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), or transfer learning for training, resulting in a scene-adaptive model that can identify noise source types, events, or scenes.

[0123] It should be noted that contextualized data includes pairings between existing general noise source classification rules and common noise sources in specific scenarios, which can clearly indicate the classification information of common noise sources at specific monitoring points. For example, aerodynamic noise in traffic noise and aerodynamic noise from central air conditioning in residential noise; the contextualized adaptation model for acquiring audio features matches the audio features with existing contextualized data to determine the noise source and noise scenario of the target noise data.

[0124] It should be noted that both historical noise data and historical associated data can be vectorized by standardizing them using a 20Hz to 20kHz bandpass filter and normalization. Then, MFCC or Mel spectrum methods are used to extract features from the standardized historical noise data and historical associated data, obtaining primary and secondary feature vectors respectively. The primary and secondary feature vectors are then combined and normalized before being imported into the deep learning model for model training.

[0125] It should be noted that historical data with waveforms can be used to extract Mel spectrum features using the MFCC method, while historical data without waveform features can be directly used as training data for the scene-based adaptation model. For example, temperature parameters for stable temperatures and rainfall parameters for continuous rainfall are historical data without waveform features. Signal characterization parameters such as ZCR (zero crossing rate) and SNR (signal-to-noise ratio) can be used as subordinate feature vectors, and video feature data extracted by video feature extraction methods can be used as vectorized feature parameters.

[0126] It should be noted that, considering the limitations of computing resources, the imported subordinate feature vectors can be limited to three categories. Subordinate vector features not used in the scenario-based adaptation model can be added to the knowledge graph to provide an implementation basis for subsequent data matching and recognition.

[0127] To facilitate understanding of the complete technical solution of this application, the following embodiments are provided:

[0128] S401, the intelligent analysis device 10 is communicatively connected to multiple noise acquisition devices 20 located on the first street, the second street and the third street. The noise acquisition device 20 includes an edge AI device 21, a sound acquisition device, a weather instrument 23 and a vision device 24. The edge AI device 21 is communicatively connected to the intelligent analysis device 10, the sound acquisition device, the weather instrument 23 and the vision device 24 respectively.

[0129] S402, the historical noisy data and historical associated data are vectorized to obtain the main feature vector and the subordinate feature vector. The main feature vector and the subordinate feature vector are input into the deep learning model for training to obtain the scene-adaptive model. The scene-adaptive model is installed on the edge AI device 21.

[0130] S403, the user inputs noise query information, which is the construction noise of the First Street from 9:00 am to 10:00 am on January 1, 2025. The LLM model captures the query keywords in the noise query information.

[0131] S404, the query keywords include the target location "First Street", the intelligent analysis device 10 obtains the target noise acquisition device 20, the target noise acquisition device 20 is the noise acquisition device 20 located on First Street, the target noise acquisition device includes the target sound acquisition device; the intelligent analysis device 10 obtains the historical audio data of the target sound acquisition device;

[0132] S405, the query keywords include the target time range "9:00 AM to 10:00 AM on January 1, 2025", retrieve the historical audio data from 9:00 AM to 10:00 AM on January 1, 2025, and this audio data is the target noise data;

[0133] S406, the edge AI device 21 acquires target noise data, acquires target association data through weather instrument 23 and vision device 24, inputs target noise data and target association data into scene adaptation model, and obtains scene analysis results. The scene analysis results include audio features and scene features of target noise data.

[0134] S407, the intelligent analysis device 10 acquires the scene analysis results uploaded by multiple target edge AI devices 21, determines the noise source of the target noise data from the noise source feature library based on the audio features in the scene analysis results, and determines the compliance of the target noise data and the noise source from the legal library in the knowledge graph based on the noise source and audio features.

[0135] S408, based on the noise source and its compliance, determines the control requirements for the noise source, and based on the control requirements, determines the initial prevention and control strategy for the target noise data;

[0136] S409, the business model library of the knowledge graph determines matching cases from historical cases based on scenario analysis results and noise trend analysis results, and the matching cases include historical prevention and control strategies;

[0137] S410, the initial prevention and control strategy is modified based on the historical prevention and control strategy, and the target prevention and control strategy is obtained based on the modification result. The intelligent analysis device 10 outputs the target noise data and the target prevention and control strategy.

[0138] Through this application, the urban noise analysis method applied to the urban noise analysis system can analyze the acquired target noise data and target-related data in real time through the edge AI device 21, determine the scene analysis results of the target noise data, and the intelligent analysis device 10 acquires multiple scene analysis results. Based on the knowledge graph preset in the intelligent analysis device 10, the initial prevention and control strategy is determined based on the scene analysis results, and the initial prevention and control strategy is corrected based on the historical prevention and control strategies of matching cases that match the scene analysis results in historical cases, so as to obtain a target prevention and control strategy that is output in real time, which is highly feasible and has better prevention and control effect.

[0139] Figure 3 shows a structural diagram of an urban noise analysis device based on edge AI provided in an embodiment of the present invention. The present invention also provides an urban noise analysis device based on edge AI, comprising:

[0140] The processor 501 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0141] The memory 502 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 502 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 502 and is called and executed by the processor 501 to execute the urban noise analysis method based on edge AI of the embodiments of this application.

[0142] The input / output interface 503 is used to implement information input and output;

[0143] The communication interface 504 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0144] Bus 505 transmits information between various components of the device (e.g., processor 501, memory 502, input / output interface 503, and communication interface 504);

[0145] The processor 501, memory 502, input / output interface 503, and communication interface 504 are connected to each other within the device via bus 505.

[0146] This application also provides an electronic device, including the urban noise analysis device based on edge AI as described above.

[0147] This application embodiment also provides a storage medium, which is a computer-readable storage medium, storing a computer program that, when executed by a processor, implements the above-described urban noise analysis method based on edge AI.

[0148] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate, and may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0150] The above provides a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for urban noise analysis based on edge AI, characterized in that, An application is made in an urban noise analysis system, comprising an intelligent analysis device and multiple noise acquisition devices. The noise acquisition devices include a sound acquisition device and an edge AI device, which are communicatively connected to the intelligent analysis device. The method includes: sending target noise data acquired by the sound acquisition device to the edge AI device; the edge AI device inputting the target noise data into a pre-trained scenario-based adaptation model; determining a scenario analysis result for the target noise data using the scenario-based adaptation model, wherein the scenario analysis result includes audio features and corresponding scenario features; sending the scenario analysis result to the intelligent analysis device via the edge AI device; and the intelligent analysis device determining a target prevention strategy based on the scenario analysis result and a pre-set knowledge graph, wherein the target prevention strategy is a prevention strategy for the target noise data, and the knowledge graph includes historical cases, which record the scenario analysis results and corresponding historical prevention strategies for historical noise data.

2. The urban noise analysis method based on edge AI according to claim 1, characterized in that, After the edge AI device inputs the target noise data into a pre-trained scene adaptation model, the process includes: the edge AI device acquiring target association data associated with the target noise data, the target association data including target meteorological data and target video feature data associated with the target noise data; and the edge AI device inputting the target noise data and the target association data into the scene adaptation model to obtain the scene features.

3. The urban noise analysis method based on edge AI according to claim 2, characterized in that, The edge AI device is communicatively connected to both the weather instrument and the vision device. The edge AI device acquires target-related data associated with the target noise data, including: acquiring the target meteorological data uploaded by the weather instrument and acquiring video data uploaded by the vision device; and performing image AI recognition on the video data to obtain the target video feature data.

4. The urban noise analysis method based on edge AI according to claim 1, characterized in that, Before sending the collected target noise data to the edge AI device via the sound acquisition device, the method further includes: inputting noise query information into a pre-trained LLM model in the intelligent analysis device, wherein the LLM model obtains the query keywords of the noise query information; when the query keywords include a target location, the intelligent analysis device obtains a target sound acquisition device and acquires historical audio data based on the target sound acquisition device, wherein the target sound acquisition device is the sound acquisition device located at the target location; when the query keywords include a target time range, the target sound acquisition device extracts the target noise data based on the historical audio data, wherein the target noise data is the audio data of the historical audio data within the target time range.

5. The urban noise analysis method based on edge AI according to claim 4, characterized in that, The knowledge graph includes a noise source feature library and a regulatory library. The noise source feature library includes audio features of various noise sources, and the regulatory library includes noise pollution standards and regulations. After the scene analysis results are sent to the intelligent analysis device via the edge AI device, the process further includes: the intelligent analysis device determining the noise source and occurrence scenario of the target noise data based on the audio features and the noise source feature library; and determining the control requirements corresponding to the target noise data based on the noise source, the occurrence scenario, and the regulatory library.

6. The urban noise analysis method based on edge AI according to claim 5, characterized in that, After determining the control requirements corresponding to the target noise data, the method further includes: acquiring the control requirements and determining an initial prevention and control strategy based on the control requirements; analyzing the changing trend of the target noise data to obtain noise trend analysis results; the knowledge graph determining matching cases from the historical cases based on the scenario analysis results and the noise trend analysis results, the matching cases including the historical prevention and control strategies; modifying the initial prevention and control strategy based on the historical prevention and control strategies, and obtaining the target prevention and control strategy based on the modification results.

7. The urban noise analysis method based on edge AI according to claim 1, characterized in that, Before the edge AI device inputs the target noise data into the pre-trained scene adaptation model, the method further includes: acquiring historical noise data from all the sound acquisition devices; vectorizing the historical noise data to obtain a main feature vector; acquiring historical associated data of the historical noise data; vectorizing the historical associated data to obtain a subordinate feature vector, wherein the historical associated data includes historical meteorological data, historical video feature data, and scene data associated with the historical noise data; and the scene data includes noise scenes that match the target noise data, the historical meteorological data, and the historical video feature data; and inputting the main feature vector and the subordinate feature vector into a deep learning model for training to obtain the scene adaptation model.

8. A city noise analysis device based on edge AI, characterized in that, It includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor to enable the at least one control processor to perform the urban noise analysis method based on edge AI as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, Includes the urban noise analysis device based on edge AI as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the urban noise analysis method based on edge AI as described in any one of claims 1 to 7.

Images