Active noise reduction system and active noise reduction method for carriage

Abstract

The invention provides a carriage active noise reduction system based on a semantic recognition model, and the system comprises a noise collection module, a noise preprocessing module, an audio feature extraction module, a rail transit noise recognition model, a mode matching module, a post-processing module, a main control unit, and a directional acoustic system. And then the noise recognition model analyzes the noise spectrum, and judges the noise source according to the model database. The method comprises the following steps of: receiving a noise spectrum, identifying the frequency of the noise for the noise spectrum, and then emitting frequency sound with opposite phases in real time through a directional acoustic system mounted in a vehicle body, thereby realizing environmental noise cancellation and active noise reduction of corresponding noise in a compartment.

Description

Technical Field

[0001] The present invention belongs to the technical field of rail vehicles, and in particular relates to an active noise reduction system and an active noise reduction method for a carriage. Background Art

[0002] Internationally, noise has been listed as one of the seven major environmental hazards. Research institutions in various countries have begun to study the pollution patterns, causes, transmission pathways and control methods of noise, as well as its harm to the human body.

[0003] In rail transit, as vehicles speed up and become lighter, noise inside the vehicle has become an important issue in train operation. However, the sound field environment of rail transit is complex, with many sound sources. The main source of noise is wheel-rail noise. In addition, air conditioning, transmission system, exhaust device, etc. will also generate noise.

[0004] The traditional noise reduction device is to install sound-absorbing cotton in the car body, but this noise reduction method cannot completely eliminate the noise. In addition, the spectrum of wheel-rail noise is different under different temperatures, different humidity, different loads, etc., and the sound-absorbing cotton is fixed and cannot be compatible with all noise spectra, which will greatly reduce the noise reduction effect. Summary of the invention

[0005] The present invention aims to solve the problem that traditional rail vehicle noise reduction devices cannot completely eliminate noise and are not compatible with all noise spectra, thereby providing an active noise reduction method for the carriage based on a semantic recognition model, which greatly reduces the noise in the carriage and improves the passenger experience.

[0006] To achieve the above-mentioned purpose of the invention, the present invention provides a carriage active noise reduction system based on a semantic recognition model, comprising a noise collection module, a noise preprocessing module, an audio feature extraction module, a rail transit noise recognition model, a pattern matching module, a post-processing module, a main control unit, and a directional acoustic system;

[0007] The noise collection module is used to collect the original noise spectrum in real time through the acoustic sensor, and convert the sound signal into an electrical signal and send it to the noise preprocessing module;

[0008] The noise preprocessing module is used to process the collected original noise spectrum, normalize and classify the sound through the preprocessing module, and send the preprocessed noise signal to the audio feature extraction module;

[0009] The audio feature extraction module is used to integrate the preprocessed noise signal with a function having good localization properties in both the time domain and the frequency domain, decompose the noise signal into components of different frequencies and times, and extract a noise feature sequence from the noise signal based on the auditory semantic recognition technology, and send it to the pattern matching module;

[0010] The pattern matching module includes a rail transit noise recognition model, which includes a variety of noise training sets for rail transit wheels, equipment, and transmission systems, and establishes training subsets for various noise sources for various vehicle models and environments. According to different types of data simulation algorithms, the data of the noise training subsets are enhanced and trained to form an optimized rail transit noise recognition model.

[0011] The pattern matching module is used to calculate and compare the matching distance between the input noise feature sequence to be identified and the model-related parameters, and finally obtain the best matching noise identification result, and send it to the post-processing module;

[0012] The post-processing module is used to identify the noise frequency, sound pressure level, and noise source device according to the matching recognition result, and transmit the result to the main control unit;

[0013] The main control unit is used to output an anti-phase signal frequency audio signal to a corresponding directional acoustic system according to different noise types such as direct sound, transmitted sound and vibration radiation sound;

[0014] The directional acoustic system is arranged between the outer wall and the inner wall of the vehicle compartment around the vehicle compartment, and is used for the power amplifier unit to drive the corresponding directional speakers to emit audio with a phase opposite to that of the noise for direct sound and transmitted sound.

[0015] Furthermore, the processing of the collected original noise spectrum includes AD conversion, filtering out frequency signals other than 20 Hz to 20 kHz, and performing endpoint detection and framing on the signal according to noise characteristic parameters.

[0016] Furthermore, the rail transit noise recognition model includes an input end, a model backbone network, a model neck, and a model output end. The model input end is used to extract frequency domain features of the noise signal for a set number of noise training sample images and to build a classifier to train the noise features; the model backbone network is used to extract and fuse features of the training noise; the model neck is used to use feature path aggregation to fuse the extracted noise image features at different levels; the model output end includes confidence, category probability, and bounding box position, and is used to output the size and category information of the noise.

[0017] An active noise reduction method using the above-mentioned vehicle cabin active noise reduction system based on semantic recognition model, wherein the noise acquisition module collects the original noise spectrum in real time through an acoustic sensor, and converts the sound signal into an electrical signal and sends it to the noise preprocessing module;

[0018] The noise preprocessing module processes the collected original noise spectrum and sends the preprocessed noise signal to the audio feature extraction module;

[0019] The audio feature extraction module integrates the noise signal with a function having good localization properties in both the time domain and the frequency domain, decomposes the noise signal into components of different frequencies and times, and extracts a noise feature sequence from the noise signal based on the auditory semantic recognition technology, and sends it to the pattern matching module;

[0020] The pattern matching module calculates and compares the matching distance between the input noise feature sequence to be identified and the parameters related to the rail transit noise identification model, and finally obtains the best matching noise identification result, and sends it to the post-processing module;

[0021] The post-processing module identifies the noise frequency, sound pressure level, and noise source device according to the matching identification result, and transmits the result to the main control unit;

[0022] The main control unit outputs an anti-phase signal frequency audio signal to the corresponding directional acoustic system according to different noise types such as direct sound, transmitted sound and vibration radiation sound;

[0023] The directional acoustic system is used for direct sound and transmitted sound, and the power amplifier unit drives the corresponding directional speaker to emit audio with a phase opposite to that of the noise.

[0024] Furthermore, the processing of the collected original noise spectrum includes AD conversion, filtering out frequency signals other than 20 Hz to 20 kHz, and performing endpoint detection and framing on the signal according to noise characteristic parameters.

[0025] Furthermore, the rail transit noise recognition model includes an input end, a model backbone network, a model neck, and a model output end. The model input end is a set number of noise training sample images, and frequency domain features of the noise signal are extracted and a classifier is constructed to train the noise features; the model backbone network extracts and fuses features of the training noise; the model neck uses feature path aggregation to fuse the extracted noise image features at different levels; the model output end includes confidence, category probability, and bounding box position, and outputs the size and category information of the noise.

[0026] The present invention uses semantic recognition noise reduction technology to identify the noise spectrum, and then the noise recognition model analyzes the noise spectrum and determines the noise source based on the model database. The noise frequency is identified from the noise spectrum, and then the directional acoustic system installed in the vehicle body emits frequency sounds with opposite phases in real time to cancel out the ambient noise and achieve active noise reduction of the corresponding noise in the vehicle compartment. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a block diagram of the active noise reduction method for a vehicle cabin of the present invention;

[0028] Figure 2Schematic diagram of the installation location of the directional acoustic system. DETAILED DESCRIPTION

[0029] In order to better understand the purpose, structure and function of the present invention, the following is a further detailed description of an active noise reduction system and an active noise reduction method for a vehicle compartment of the present invention in conjunction with the accompanying drawings.

[0030] The active noise reduction system for a carriage of the present invention comprises a noise collection module, a noise preprocessing module, an audio feature extraction module, a rail transit noise recognition model, a pattern matching module, a post-processing module, a main control unit, and a directional acoustic system 3.

[0031] Reference Figure 1 ,The noise collection module collects the original noise spectrum in real time through the acoustic sensor, and ,converts the sound signal into an electrical signal and sends it to the noise preprocessing module.

[0032] The noise preprocessing module processes the collected original noise spectrum, including AD conversion, filtering out frequency signals other than 20Hz~20kHz, endpoint detection and framing of the signal according to noise characteristic parameters, and sends the preprocessed noise signal to the audio feature extraction module.

[0033] The audio feature extraction module integrates the preprocessed noise signal with a function with good localization properties in both the time domain and the frequency domain, decomposes the noise signal into components of different frequencies and times, and extracts the noise feature sequence from the noise signal based on auditory semantic recognition technology and sends it to the pattern matching module.

[0034] The pattern matching module contains a rail transit noise recognition model, which includes a variety of noise training sets such as rail transit wheels, equipment, and transmission systems. Training subsets are established for various noise sources for various vehicle models and environments. According to different types of data simulation algorithms, the data of the noise training subsets are enhanced and trained to form an optimized rail transit noise recognition model.

[0035] The rail transit noise recognition model includes an input end, a model backbone network, a model neck, and a model output end. The model input end is a set number of noise training sample images. Compared with other scene noises, rail transit noise is a complex sound field composed of multiple noise sources superimposed together. The noise field is decomposed into multiple environmental noise sources. The noise signal is extracted from the frequency domain and a classifier is constructed to train the noise features. The model backbone network is the main structure of the model, which realizes the feature extraction and feature fusion of the training noise. It consists of a feature convolution layer, a batch processing module, and an activation function layer. Finally, the data is output to the fully connected layer to realize the fusion of multiple features in the model. Feature path aggregation is used in the model neck to fuse the extracted noise image features at different levels. The model output end mainly consists of confidence, category probability, and bounding box position. The size and category information of the noise are output to facilitate subsequent tasks such as noise recognition and tracking.

[0036] The pattern matching module calculates and compares the matching distance between the input noise feature sequence to be identified and the model-related parameters, and finally obtains the best matching noise recognition result.

[0037] The post-processing module identifies parameters such as noise frequency, sound pressure level, and noise source equipment based on the matching recognition results, and transmits the results to the main control unit.

[0038] The main control unit outputs anti-phase signal frequency audio signals to the corresponding directional acoustic system according to different noise types such as direct sound, transmitted sound and vibration radiation sound.

[0039] Reference Figure 2 The directional acoustic system 3 is arranged between the outer wall 1 and the inner wall 2 of the car around the car. For direct sound and transmitted sound, the directional acoustic system power amplifier unit drives the corresponding directional speaker to emit audio with a phase opposite to the noise. The noise audio and the anti-phase audio are superimposed on each other in space, thereby achieving noise elimination. For vibration radiation sound, the directional acoustic system eliminates the vibration radiation sound from the source through the reverse vibration of the acoustic system.

Claims

1. A vehicle cabin active noise reduction system based on a semantic recognition model, characterized in that: It includes noise collection module, noise preprocessing module, audio feature extraction module, rail transit noise recognition model, pattern matching module, post-processing module, main control unit, and directional acoustic system; The noise collection module is used to collect the original noise spectrum in real time through the acoustic sensor, and convert the sound signal into an electrical signal and send it to the noise preprocessing module; The noise preprocessing module is used to process the collected original noise spectrum, normalize and classify the sound through the preprocessing module, and send the preprocessed noise signal to the audio feature extraction module; The audio feature extraction module is used to integrate the preprocessed noise signal with a function having good localization properties in both the time domain and the frequency domain, decompose the noise signal into components of different frequencies and times, and extract a noise feature sequence from the noise signal based on the auditory semantic recognition technology, and send it to the pattern matching module; The pattern matching module includes a rail transit noise recognition model, which includes a variety of noise training sets for rail transit wheels, equipment, and transmission systems, and establishes training subsets for various noise sources for various vehicle models and environments. According to different types of data simulation algorithms, the data of the noise training subsets are enhanced and trained to form an optimized rail transit noise recognition model. The pattern matching module is used to calculate and compare the matching distance between the input noise feature sequence to be identified and the model-related parameters, and finally obtain the best matching noise identification result, and send it to the post-processing module; The post-processing module is used to identify the noise frequency, sound pressure level, and noise source device according to the matching recognition result, and transmit the result to the main control unit; The main control unit is used to output an anti-phase signal frequency audio signal to a corresponding directional acoustic system according to different noise types such as direct sound, transmitted sound and vibration radiation sound; The directional acoustic system is arranged between the outer wall and the inner wall of the vehicle compartment around the vehicle compartment, and is used for the power amplifier unit to drive the corresponding directional speakers to emit audio with a phase opposite to that of the noise for direct sound and transmitted sound.

2. The vehicle compartment active noise reduction system based on the semantic recognition model according to claim 1, characterized in that: The processing of the collected original noise spectrum includes AD conversion, filtering out frequency signals other than 20 Hz to 20 kHz, and performing endpoint detection and framing on the signal according to noise characteristic parameters.

3. The vehicle compartment active noise reduction system based on the semantic recognition model according to claim 1, characterized in that: The rail transit noise recognition model includes an input end, a model backbone network, a model neck, and a model output end. The model input end is used to extract frequency domain features of noise signals for a set number of noise training sample images and to construct a classifier to train noise features; the model backbone network is used to extract and fuse features of training noise; The neck of the model is used to fuse the extracted noise image features at different levels by using feature path aggregation; the output of the model includes confidence, category probability, and bounding box position, which are used to output the size and category information of the noise.

4. An active noise reduction method using the vehicle compartment active noise reduction system based on the semantic recognition model according to claim 1, characterized in that: The noise collection module collects the original noise spectrum in real time through the acoustic sensor, and converts the sound signal into an electrical signal and sends it to the noise preprocessing module; The noise preprocessing module processes the collected original noise spectrum, and sends the preprocessed noise signal to the audio feature extraction module; The audio feature extraction module integrates the noise signal with a function having good localization properties in both the time domain and the frequency domain, decomposes the noise signal into components of different frequencies and times, and extracts a noise feature sequence from the noise signal based on the auditory semantic recognition technology, and sends it to the pattern matching module; The pattern matching module calculates and compares the matching distance between the input noise feature sequence to be identified and the parameters related to the rail transit noise identification model, and finally obtains the best matching noise identification result, and sends it to the post-processing module; The post-processing module identifies the noise frequency, sound pressure level, and noise source device according to the matching identification result, and transmits the result to the main control unit; The main control unit outputs an anti-phase signal frequency audio signal to the corresponding directional acoustic system according to different noise types such as direct sound, transmitted sound and vibration radiation sound; The directional acoustic system is used for direct sound and transmitted sound, and the power amplifier unit drives the corresponding directional speaker to emit audio with a phase opposite to that of the noise.

5. An active noise reduction method according to claim 4, characterized in that: The processing of the collected original noise spectrum includes AD conversion, filtering out frequency signals other than 20 Hz to 20 kHz, and performing endpoint detection and framing on the signal according to noise characteristic parameters.

6. An active noise reduction method according to claim 5, characterized in that: The rail transit noise recognition model includes an input end, a model backbone network, a model neck, and a model output end. The model input end is a set number of noise training sample images, and the frequency domain feature extraction of the noise signal is performed and a classifier is constructed to train the noise feature; the model backbone network extracts and fuses the features of the training noise; The neck of the model uses feature path aggregation to fuse the extracted noise image features at different levels; the output end includes confidence, category probability, bounding box position, and outputs the size and category information of the noise.

Images