CN115545148A - Body sound auscultation device and body sound data diagnosis method - Google Patents
Body sound auscultation device and body sound data diagnosis method Download PDFInfo
- Publication number
- CN115545148A CN115545148A CN202211020188.7A CN202211020188A CN115545148A CN 115545148 A CN115545148 A CN 115545148A CN 202211020188 A CN202211020188 A CN 202211020188A CN 115545148 A CN115545148 A CN 115545148A
- Authority
- CN
- China
- Prior art keywords
- body sound
- target
- module
- sound
- auscultation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B7/00—Instruments for auscultation
- A61B7/02—Stethoscopes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Theoretical Computer Science (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- Animal Behavior & Ethology (AREA)
- Medical Informatics (AREA)
- Heart & Thoracic Surgery (AREA)
- Veterinary Medicine (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Surgery (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Acoustics & Sound (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention provides a body sound auscultation device and a body sound data diagnosis method, wherein the body sound auscultation device comprises an SoC chip, and a data acquisition module and a body sound diagnosis module which are integrated on the SoC chip, wherein the data acquisition module is connected with the body sound diagnosis module; the data acquisition module is used for acquiring a target volume tone digital signal; the body sound diagnosis module is used for determining a target auscultation result of the target body sound digital signal based on the target body sound digital signal and the target identification model; the target recognition model is obtained by training an LSTM model and a full connection layer based on sample body sound digital signals and body sound labels, and the target auscultation result comprises body sound normality and body sound abnormity. The invention can realize the purpose of auscultation aiming at various different types of sound signals on the premise of not limiting the auscultation environment, and improves the convenient flexibility, timeliness and high efficiency of auscultation.
Description
Technical Field
The invention relates to the technical field of voice processing, in particular to a body sound auscultation device and a body sound data diagnosis method.
Background
In recent two years, the first two diseases with the highest death rate in China, whether in rural areas or cities and towns, are malignant tumors and heart diseases, and even common malignant tumors such as lung cancer, gastric cancer, colorectal cancer and the like, jointly cause one fourth of the total burden of cancer morbidity and mortality. Therefore, the early diagnosis and prevention of organ diseases have very important functions, and the auscultation, which is an important auxiliary diagnostic means in western medicine, is used for effectively predicting the diseases and carrying out subsequent targeted treatment.
In the prior art, an artificial intelligence based auscultation function is generally deployed in a terminal or a server, and then a body sound signal to be auscultated is transmitted to the terminal or the server through WiFi or bluetooth for diagnosis.
However, the existing auscultation functions based on artificial intelligence are mainly concentrated on the server and the terminal, so that auscultation is limited under the limited auscultation environments without bluetooth or WiFi, and only one body sound can be input for diagnosis at a time, and thus the body sound auscultation is single and the auscultation limitation is large.
Disclosure of Invention
The invention provides a body sound auscultation device and a body sound data diagnosis method, which are used for overcoming the defects of single body sound auscultation and larger auscultation limitation in the prior art, achieving the aim of auscultation aiming at various different types of body sound signals on the premise of not limiting an auscultation environment, and improving the convenience, flexibility, timeliness and high efficiency of auscultation.
The invention provides a body sound auscultation device, which comprises an SoC chip, and a data acquisition module and a body sound diagnosis module which are integrated on the SoC chip, wherein the data acquisition module is connected with the body sound diagnosis module;
the data acquisition module is used for acquiring a target volume tone digital signal;
the body sound diagnosis module is used for determining a target auscultation result of the target body sound digital signal based on the target body sound digital signal and a target identification model;
the target recognition model is obtained by training an LSTM model and a full connection layer based on sample body sound digital signals and body sound labels, and the target auscultation result comprises body sound normality and body sound abnormality.
According to the body sound auscultation device provided by the invention, the body sound diagnosis module comprises a feature extraction module, a feature identification module and a result correction module which are sequentially connected, wherein:
the characteristic extraction module is used for extracting the characteristics of the target volume sound digital signal and determining the time-frequency domain volume sound characteristics obtained by characteristic extraction;
the characteristic identification module is used for identifying the time-frequency domain body sound characteristic based on the target identification model and determining a target identification result of the target body sound digital signal;
and the result correction module is used for carrying out noise filtration on the target identification result and determining a target auscultation result of the target body sound digital signal based on a result obtained by the noise filtration.
According to the body sound auscultation device provided by the invention, the characteristic extraction module comprises a short-time Fourier transform module, a Mel filtering module, a logarithm operation module, a discrete cosine transform module and a normalization mapping module which are sequentially connected, wherein:
the short-time Fourier transform module is used for framing the target volume tone digital signal and determining spectrogram characteristics corresponding to frame data obtained by framing based on windowing and frequency domain conversion operation;
the Mel filtering module is used for carrying out Mel filtering on the spectrogram characteristics to determine the Mel spectrogram characteristics;
the logarithm operation module is used for carrying out logarithm operation conversion on the Mel frequency spectrogram characteristics to determine logarithm Mel frequency spectrogram characteristics;
the discrete cosine transform module is used for performing discrete cosine transform on the logarithmic Mel frequency spectrum diagram characteristics to determine a Mel cepstrum coefficient;
and the normalization mapping module is used for normalizing the Mel cepstrum coefficient and determining the time-frequency domain body sound characteristic of the target body sound digital signal.
According to the body sound auscultation device provided by the invention, the data acquisition module comprises a digital microphone module, a down-sampling filtering module and a noise reduction module which are sequentially connected, wherein:
the digital microphone module is used for acquiring an original body sound signal and converting the original body sound signal into a digital body sound signal;
the down-sampling filtering module is used for down-sampling the digital volume sound signal;
and the noise reduction module is used for reducing noise of the digital volume sound signal after down sampling and determining a target volume sound digital signal.
According to the body sound auscultation device provided by the invention, the body sound diagnosis module further comprises a configuration scheduling module, wherein the output end of the configuration scheduling module is connected with the input end of the characteristic extraction module;
the configuration scheduling module is configured to configure a control signal matched with the volume class information for the original volume sound signal based on the volume class information corresponding to the original volume sound signal, and schedule a target recognition model based on the control signal.
According to the body sound auscultation device provided by the invention, the body sound diagnosis module further comprises a storage module, and the storage module is connected with the feature identification module;
the storage module is used for storing different weight parameters of different target recognition models, and each weight parameter corresponds to different control signals respectively.
According to the body sound auscultation device provided by the invention, the training process of the target recognition model comprises the following steps:
acquiring a sample body sound digital signal set carrying a body sound label, and dividing the sample body sound digital signal set into a training sample and a test sample, wherein the body sound label comprises a body sound category label, a normal body sound label and an abnormal body sound label;
training an initial recognition model containing an LSTM model and a full connection layer by using the training sample to obtain a recognition model obtained by training;
and testing the recognition model obtained by training by using the test sample to determine a target recognition model.
According to the body sound auscultation device provided by the invention, the device further comprises a buzzer and a communication module, wherein the buzzer and the communication module are respectively connected with the body sound diagnosis module, and the body sound auscultation device comprises:
the buzzer is used for carrying out early warning based on the target auscultation result; and the communication module is used for transmitting the target auscultation result to an external terminal.
The present invention also provides a method for diagnosing volume sound data, which is based on the volume sound auscultation device of the first aspect, and comprises:
when the fact that the body sound auscultation device and the target part of the target object meet the preset distance relation is determined, a target diagnosis instruction is generated;
controlling the body sound data to be diagnosed corresponding to the target part to enter the body sound auscultation device based on the target diagnosis instruction, and acquiring a target diagnosis result of the body sound data to be diagnosed; the body sound data to be diagnosed comprises original body sound signals of different body sound categories;
and outputting the target diagnosis result.
According to a method for diagnosing volume sound data provided by the present invention, the outputting the target diagnosis result includes:
outputting the target diagnosis result through a built-in buzzer of the body sound auscultation device; and/or outputting the target diagnosis result to an external terminal in communication connection with the body sound auscultation device.
The invention provides a body sound auscultation device and a diagnosis method of body sound data, wherein the body sound auscultation device acquires a target body sound digital signal through a data acquisition module, and the body sound diagnosis module determines a target auscultation result of the target body sound digital signal based on the target body sound digital signal and a target recognition model, so that the body sound auscultation device can quickly and reliably carry out body sound auscultation without a terminal and a server, the auscultation environment is not limited, the body sound category of the target body sound digital signal is not limited, and the convenience, flexibility, timeliness and high efficiency of the auscultation are greatly improved. Furthermore, the target recognition model used in the body sound diagnosis module is obtained by training the LSTM model and the full connection layer based on the sample body sound digital signal and the body sound label, so that the accuracy and reliability of body sound diagnosis can be improved by combining the mode of training the recurrent neural network and the full connection layer, the aims of effective prediction and timely and targeted treatment can be fulfilled for organ diseases, and the method has wide applicability and popularization value.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
Fig. 1 is a schematic structural diagram of a body sound auscultation device provided by the invention;
FIG. 2 is a schematic structural diagram of a body sound diagnostic module provided in the present invention;
FIG. 3 is a schematic structural diagram of a feature extraction module provided in the present invention;
FIG. 4 is a schematic diagram of a signal acquisition module according to the present invention;
FIG. 5 is a schematic diagram of a configuration scheduling process provided by the present invention;
fig. 6 is a flow chart of a method for diagnosing volume sound data according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
According to the national published annual book of health statistics of 2021 China, lung cancer, gastric cancer, colon cancer and other common malignant tumors in rural areas or cities jointly cause one fourth of the total burden of cancer morbidity and mortality, and meanwhile, heart disease is also one of the main causes of death of the population in China. Therefore, the early diagnosis and prevention of organ diseases have very important functions.
Auscultation is an important auxiliary diagnosis means for western medicine, so that great help is provided for diagnosing the disease condition, and the experienced meaning can make effective prediction on the disease condition through auscultation, thereby being more targeted for treatment. However, the conventional auscultation technology has some disadvantages, for example, particularly weak pathological sounds, pathological sounds covered by noisy environmental noises, and auscultation environmental noises can cause serious auscultation effect; moreover, auscultation has higher requirements on the experience and hearing of doctors, for example, doctors have abundant experience and very sensitive hearing, otherwise auscultation becomes a setup due to missing important information.
Currently, auscultation is usually performed by deploying an artificial intelligence based auscultation function in a terminal or a server, that is, the artificial intelligence based auscultation function is deployed in the terminal or the server, and then a body sound signal to be auscultated is transmitted to the terminal or the server through WiFi or bluetooth for diagnosis. However, because the existing auscultation function based on artificial intelligence is mainly centralized and deployed at a server and a terminal, auscultation is limited under the limited auscultation environments without bluetooth or WiFi, and only one body sound can be input for diagnosis at a time, so that the body sound auscultation is single and the auscultation limitation is large. In addition, the artificial intelligent neural network structure has higher requirements for realizing auscultation at the edge side due to the problems of power consumption, size and the like.
Based on the above problems, the present invention provides a body sound auscultation device and a method for diagnosing body sound data, wherein the body sound auscultation device may be embodied as a body sound auscultation chip, which can perform body sound auscultation without a terminal and a server, and does not limit an auscultation environment, i.e., can perform auscultation anytime and anywhere. For example, the target body sound digital signal of the target object can be input to the body sound auscultation device, that is, the auscultation process for the target body sound digital signal can be performed, and does not need to be sent to the terminal or the server for processing. The body sound auscultation apparatus and the diagnosis method of body sound data of the present invention will be described with reference to fig. 1 to 6.
Referring to fig. 1, which is a schematic structural view of a body sound auscultation device provided by the present invention, as shown in fig. 1, the body sound auscultation device includes: the system comprises an SoC chip, and a data acquisition module and a body sound diagnosis module which are integrated on the SoC chip, wherein the data acquisition module is connected with the body sound diagnosis module;
the data acquisition module is used for acquiring a target volume tone digital signal;
the body sound diagnosis module is used for determining a target auscultation result of the target body sound digital signal based on the target body sound digital signal and the target identification model;
the target recognition model is obtained by training an LSTM model and a full connection layer based on sample body sound digital signals and body sound labels, and the target auscultation result comprises body sound normality and body sound abnormality. The target volume sound digital signal may be a clean volume sound digital signal without noise, and the volume sound category of the target volume sound digital signal may be multiple, for example, the volume sound category of the target volume sound digital signal may include bowel sound, breath sound, heart sound, and the like. And is not particularly limited herein.
It can be understood that, in addition to the data acquisition module and the body sound diagnosis module integrated on the SoC chip, the SoC chip further includes a main controller, a Read-Only Memory (ROM), an Advanced High Performance Bus (AHB), and an Advanced Peripheral Bus (APB), where the main controller may be specifically an ARM Cortex-M0+ microprocessor. Based on this, when the main controller determines that the body sound auscultation device is powered on, the start auscultation program can be read from the ROM through the AHB, and the target body sound digital signal can be acquired by the control data acquisition module by analyzing the start auscultation instruction obtained by starting the auscultation program, and the acquired target body sound digital signal is input to the body sound diagnosis module for auscultation, so as to determine the target auscultation result of the target body sound digital signal.
It should be noted that, considering that a Long-Short Term Memory model (LSTM) is an optimized recurrent neural network, and is mainly used for prediction and classification of time series data, the prediction and classification can be realized by adopting a parallel calculation mode with 4 multipliers. In order to reduce power consumption and complexity of design, the activation functions sigmoid and tanh are implemented by using a linear lookup table. Moreover, since the hidden layer unit in the LSTM needs to calculate the value of the hidden unit in the historical world, a Random Access Memory (RAM) for caching the hidden unit needs to be additionally integrated on the SoC chip; the full connection layer is used for mapping the features extracted by the LSTM from the target volume tone digital signal to a sample mark space so as to realize the diagnosis of the target volume tone digital signal, and comprises a multiply-accumulate and diagnosis module, wherein the full connection layer is mainly used for multiply-accumulate of the weight parameters and the corresponding input features, and a diagnosis module based on a comparator is carried out on the aspect of an algorithm.
According to the body sound auscultation device provided by the invention, the data acquisition module is used for acquiring the target body sound digital signal, and the body sound diagnosis module is used for determining the target auscultation result of the target body sound digital signal based on the target body sound digital signal and the target identification model, so that the body sound auscultation device can quickly and reliably carry out body sound auscultation without a terminal and a server, the auscultation environment is not limited, the body sound category of the target body sound digital signal is not limited, and the convenience, flexibility, timeliness and high efficiency of auscultation are greatly improved. Furthermore, the target recognition model used in the body sound diagnosis module is obtained by training the LSTM model and the full connection layer based on the sample body sound digital signal and the body sound label, so that the accuracy and the reliability of body sound diagnosis can be improved by combining a mode of training a recurrent neural network and the full connection layer, the aims of effective prediction and timely and targeted treatment can be realized for organ diseases, and the method has wide applicability and popularization value.
Optionally, the body sound diagnosis module includes a feature extraction module, a feature recognition module and a result correction module, which are connected in sequence, wherein:
the characteristic extraction module is used for extracting the characteristics of the target volume tone digital signal and determining the time-frequency domain volume tone characteristics obtained by characteristic extraction; the characteristic identification module is used for identifying the time-frequency domain volume sound characteristic based on the target identification model and determining a target identification result of the target volume sound digital signal; and the result correction module is used for carrying out noise filtration on the target identification result and determining a target auscultation result of the target volume sound digital signal based on the result obtained by the noise filtration.
It can be understood that, as shown in fig. 2, the body sound diagnosis module includes, in addition to the feature extraction module, the feature recognition module and the result correction module, a main controller for controlling the feature extraction module, the feature recognition module and the result correction module, where the main controller can control the sequential flow of the target body sound digital signal entering the feature extraction module, the feature recognition module and the result correction module, that is, when the target body sound digital signal enters the body sound diagnosis module, the feature extraction is performed by the feature extraction module, then the time-frequency domain body sound feature obtained by the feature extraction is recognized by the feature recognition module using the target recognition model, then the target recognition result obtained by the recognition is subjected to noise filtering by the result correction module, and based on the result obtained by the noise filtering, the target auscultation result of the target body sound digital signal is determined. If the result obtained by noise filtering is a plurality of recognition results of a plurality of frame data aiming at each target volume sound digital signal, judging the magnitude relation between the recognition result with normal volume sound and the recognition result with abnormal volume sound in the plurality of recognition results, and if the number of the recognition results with normal volume sound is larger than that of the recognition results with abnormal volume sound, determining that the target auscultation result corresponding to the target volume sound digital signal is abnormal volume sound; otherwise, if the number of the recognition results of the normal body sounds is smaller than the number of the recognition results of the abnormal body sounds, the target auscultation result corresponding to the target body sound digital signal is determined to be normal body sounds.
According to the body sound auscultation device provided by the invention, the target auscultation result of the target body sound digital signal is determined in a mode that the target body sound digital signal enters the body sound diagnosis module to be subjected to feature extraction firstly, then the time-frequency domain body sound feature obtained by feature extraction is identified, and then the target identification result is subjected to noise filtration, so that the accuracy and the reliability of the determined target auscultation result are improved.
Optionally, the feature extraction module includes a short-time fourier transform module, a mel filtering module, a logarithm operation module, a discrete cosine transform module and a normalization mapping module, which are connected in sequence, wherein:
the short-time Fourier transform module is used for framing the target volume tone digital signal and determining spectrogram characteristics corresponding to frame data obtained by framing based on windowing and frequency domain conversion operation; the Mel filtering module is used for carrying out Mel filtering on the spectrogram characteristics and determining the Mel spectrogram characteristics; the logarithm operation module is used for carrying out logarithm operation conversion on the characteristics of the Mel frequency spectrogram and determining the characteristics of the logarithm Mel frequency spectrogram; the discrete cosine transform module is used for performing discrete cosine transform on the characteristics of the logarithmic Mel frequency spectrogram to determine a Mel cepstrum coefficient; and the normalization mapping module is used for normalizing the Mel cepstrum coefficients and determining the time-frequency domain volume sound characteristics of the target volume sound digital signals.
It can be understood that, as shown in fig. 3, the target volume sound digital signal is subjected to short-time fourier transform processing, and under the condition that the short-time fourier transform module includes framing, windowing and fast fourier transform operations, the target volume sound digital signal may be segmented in a frame form, that is, framed, then the frame data obtained by the segmentation is subjected to windowing operation by using a hanning window, then the windowed frame data is converted to a frequency domain by using the fast fourier transform operation, and all the frame data corresponding to each target volume sound digital signal are spliced in a sequence manner, and then spectrogram features are determined; the Mel filtering module is composed of a plurality of Mel filter banks, and conducts Mel filtering on the spectrogram characteristics to convert the spectrogram characteristics into Mel spectrogram characteristics, so that the body sound signals in the hearing frequency range of human ears are concerned by simulating the hearing characteristics of human ears; the logarithm operation module performs logarithm operation conversion on the characteristics of the Mel frequency spectrogram to convert the characteristics of the Mel frequency spectrogram into the characteristics of the logarithm Mel frequency spectrogram, so that the aim of presenting the perception of the human ear to the frequency in a logarithmic relation is fulfilled, and meanwhile, low-energy signals can be observed more favorably; the discrete cosine transform module is used for performing discrete cosine transform on the logarithmic Mel frequency spectrum map characteristics so as to transform the logarithmic Mel frequency spectrum map characteristics into Mel cepstrum coefficients, namely performing inverse transform on logarithmic energy spectrums of the logarithmic Mel frequency spectrum map characteristics in a time domain to determine the Mel cepstrum coefficients; the normalization mapping module normalizes the Mel cepstrum coefficient to reduce the range of characteristic distribution and increase the utilization rate of hardware quantization bit width, and the normalization mapping module achieves the purpose of normalizing the Mel cepstrum coefficient in a translation and scaling mode, and finally scales the characteristic range of the Mel cepstrum coefficient to a preset range, so that the time-frequency domain body voice characteristic of the target body voice digital signal is determined. The predetermined range may be [ -1,1], and the execution order and the execution number of the panning and the zooming are not particularly limited. And the output end of the normalization mapping module is connected with the input end of the feature extraction module.
It should be noted that, the window coefficient of the predetermined hanning window participates in the operation in the form of the lookup table, so that the power consumption is reduced; the coefficient of the Mel filter in the Mel filter module is predetermined to participate in the operation in the form of a lookup table, so that the power consumption is reduced, and when the Mel filter module is designed in advance, the Mel filter coefficient of the Mel filter group is also predetermined to participate in the operation in the form of the lookup table, so that the effect of reducing the power consumption is achieved; the logarithm operation in the logarithm operation module is determined in advance to participate in the operation in a lookup table mode, so that the power consumption is reduced; the cosine calculation in the predetermined discrete cosine transform module participates in the operation in the form of a lookup table, so that the power consumption is reduced.
In addition, it should be noted that, because the time-frequency domain body sound feature of the target body sound digital signal is a feature output in a frame form, the target recognition result may also be considered as a recognition result output in a frame form, and considering that the recognition result output in a frame form is influenced by environment and other noises, so as to cause interference to an actual result, therefore, it is reasonable to perform noise filtering by the result correction module that the recognition result of the preset frame number is counted from the target recognition result between the current time as a starting point and the preset historical time, and then the result with the largest occurrence number is selected as the target auscultation result of the target body sound digital signal. For example, a preset number of frames is counted by using 31 frames, a counter is used as an index, the identification result of each frame is stored to a corresponding bit through a shift register, finally, each bit of the shift register is added, and a target auscultation result of the target body sound digital signal is determined based on the result of comparison with 16.
The body sound auscultation device provided by the invention determines the time-frequency domain body sound characteristics of the target body sound digital signal by means of framing, windowing, fast Fourier transform framing, mel filtering, logarithmic operation conversion, discrete cosine conversion and normalization on the target body sound digital signal, thereby realizing the purpose of accurately and quickly extracting the characteristics on the premise of low power consumption and high utilization rate.
Optionally, the data acquisition module includes a digital microphone module, a down-sampling filtering module and a noise reduction module that are connected in sequence, wherein:
the digital microphone module is used for acquiring an original body sound signal and converting the original body sound signal into a digital body sound signal; the down-sampling filtering module is used for down-sampling the digital voice signal; and the noise reduction module is used for reducing noise of the digital volume sound signal after down sampling and determining a target volume sound digital signal.
Specifically, as can be understood with reference to the digital microphone module shown in fig. 4, when the body sound auscultation device is powered on and started, the main controller may control the data acquisition module to acquire data, that is, the digital microphone module in the data acquisition module acquires the original body sound signal first, and may acquire the original body sound signal by collecting the original body sound signal from the human body, or may acquire the original body sound signal collected from the human body by manually inputting the original body sound signal collected from the human body to the digital microphone module. Since the original body sound signal is a body sound signal directly coming from the human body, the original body sound signal is an impure body sound signal containing noise, and the body sound categories of the original body sound signal are the same as and in one-to-one correspondence with the body sound categories of the target body sound digital signal, that is, the body sound categories of the original body sound signal may also be multiple, for example, the body sound categories of the original body sound signal may include bowel sound, breath sound, heart sound, and the like.
At this time, the original body sound signal may be further audio-digitally converted using the word microphone module to convert the original body sound signal into a digital body sound signal. Then, down-sampling the digital voice signal by using a down-sampling filtering module, where the down-sampling rate can be predetermined, for example, 4kHz; and the down-sampling filter may use a half-band Finite long unit Impulse Response (FIR) filter for filtering. And finally, the noise reduction module is used for reducing noise of the digital body sound signal after the down sampling, two noise reduction algorithms of a spectral subtraction method and an empirical mode decomposition method can be adopted for reducing noise, the spectral subtraction method is used for eliminating background noise and hardware background noise in the digital body sound signal after the down sampling, the empirical mode decomposition method is used for decomposing the digital body sound signal after the noise reduction through the spectral subtraction method, and a body sound part which can be used for auscultation is reserved, so that pure body sound without noise, namely the target body sound digital signal, is determined.
According to the body sound auscultation device provided by the invention, the data acquisition module determines the target body sound digital signal in a mode of firstly acquiring the original body sound signal, then performing frequency-digital conversion on the original body sound signal, and then performing down sampling and noise reduction on the digital body sound signal, so that the reliability and the efficiency of acquiring the target body sound digital signal are improved.
Optionally, the body-tone diagnosis module further includes a configuration scheduling module, and an output end of the configuration scheduling module is connected to an input end of the feature extraction module;
and the configuration scheduling module is used for configuring a control signal matched with the body sound type information for the original body sound signal based on the body sound type information corresponding to the original body sound signal, and scheduling the target identification model based on the control signal.
It can be understood that, as shown in fig. 5, the configuration scheduling module may specifically be composed of an Electrically Erasable Programmable Read Only Memory (EEPROM), a Random Access Memory (RAM), a register set, and an AHB bus, where weight parameters of different body tone diagnoses and EEPROM control programs corresponding to the weight parameters are stored in the EEPROM in advance, and a control signal of the body tone diagnosis is configured in each EEPROM control program, and the RAM and the register set may be configured correspondingly based on the weight parameters configured in the EEPROM. Based on this, when the main controller determines that the body sound auscultation device is powered on, the main controller may read the start auscultation program from the ROM through the AHB based on the body sound category information corresponding to the original body sound signal, and read the control signal of the corresponding address from the EEPROM through the AHB-to-APB bridge and the two-wire serial bus (I2C) protocol again by analyzing the start auscultation instruction obtained by starting the auscultation program, and then read the corresponding weight parameter based on the read control signal, thereby scheduling the target identification model corresponding to the read weight parameter.
According to the body sound auscultation device provided by the invention, the configuration scheduling module is arranged to configure different control signals and scheduling target identification models for different original body sound signals, so that the pertinence and flexibility of the scheduling target identification models are improved.
Optionally, the body sound diagnosis module further includes a storage module, and the storage module is connected with the feature recognition module; and the storage module is used for storing different weight parameters of different target recognition models, and each weight parameter corresponds to different control signals respectively.
It can be understood that the storage module may specifically be an EEPROM, different weight parameters of different target identification models may be stored in the EEPROM in advance, and when the weight parameters are different, the neural network structures in the corresponding target identification models are different, that is, the target identification models having different neural network structures are configured with corresponding weight parameters in advance, and meanwhile, each weight parameter is also configured with a corresponding control signal in advance, so that the subsequent main controller performs scheduling based on the control signal. It should be noted that, although the target recognition models may have different neural network structures due to different weight parameters, the training and testing methods of each target recognition model are the same.
According to the body sound auscultation device provided by the invention, the storage module is arranged to store different weight parameters of different target identification models in advance, and each weight parameter corresponds to different control signals respectively, so that the convenience, high efficiency and reliable stability of the subsequent configuration and scheduling of the target identification models are improved, and a foundation is laid for the accuracy of the subsequent acquisition of target auscultation results.
Optionally, the training process of the target recognition model includes:
acquiring a sample body sound digital signal set carrying a body sound label, and dividing the sample body sound digital signal set into a training sample and a testing sample, wherein the body sound label comprises a body sound category label, a normal body sound label and an abnormal body sound label; training an initial recognition model containing an LSTM model and a full connection layer by using a training sample to obtain a trained recognition model; and testing the trained recognition model by using the test sample to determine a target recognition model.
It can be understood that the sample body sound digital signal set includes a large number of sample body sound digital signals, each sample body sound digital signal is marked with a body sound category label, such as an bowel sound label, a heart sound label, a breath sound label, and the like, and each sample body sound digital signal is also marked with a normal body sound label or an abnormal body sound label, and then the sample body sound digital signal set is divided into two subsets, wherein one subset is a training sample and the other subset is a testing sample. Based on the method, training samples are input into an initial recognition model containing an LSTM model and a full connection layer for training, after sample body tone digital signals contained in the training samples all participate in the training once, the initial recognition model is determined to be subjected to one round of training, the intermediate recognition model is tested by using the test samples according to the intermediate recognition model obtained through one round of training, whether the accuracy of the result obtained through the test is matched with the actual accuracy of the test samples or not is judged, the accuracy of the test result is determined according to the ratio of the number of correct body tones in the result obtained through the test to the total number of the test samples, and the actual accuracy is determined according to the ratio of the number of correct body tones in the test samples to the total number of the test samples. If the accuracy of the result obtained by the test is matched with the actual accuracy of the test sample, determining that the corresponding intermediate recognition model is the trained target recognition model during matching; otherwise, if the accuracy of the result obtained by the test is determined to be not matched with the actual accuracy of the test sample, the intermediate recognition model after the parameter update is obtained according to the corresponding intermediate recognition model when the result is not matched with the actual accuracy of the test sample, and then the training sample is used for carrying out the next round of training and testing on the intermediate recognition model after the parameter update. Until the accuracy of the result matches the actual accuracy of the test sample.
According to the body sound auscultation device provided by the invention, the initial recognition model containing the LSTM model and the full connection layer is trained through the training sample, and the trained recognition model is tested by using the test sample, so that the problems of low stability and reliability caused by the fact that a network is trained only by using the training sample image in the traditional method are solved, and the stability and reliability of the training process can be greatly improved by combining the training sample and the test sample to train the recurrent neural network and the full connection layer.
Optionally, the body sound auscultation device of the present invention may further comprise a buzzer and a communication module, the buzzer and the communication module are respectively connected to the body sound diagnosis module, wherein:
the buzzer is used for carrying out early warning based on a target auscultation result; and the communication module is used for transmitting the target auscultation result to an external terminal.
The external terminal may be a mobile terminal such as a Personal Computer (PC), a portable device, a notebook Computer, a smart phone, a tablet Computer, and a portable wearable device, or may be a desktop Computer, and a server that are not convenient to move.
It can be understood that when the main controller determines that the target auscultation result represents that the corresponding target body sound digital signal is abnormal body sound, the buzzer can be controlled to give an early warning prompt, so that the purposes of wearable and portable health monitoring are achieved; can also transmit target auscultation result to external terminal under the effect of wiFi or bluetooth through communication module to this realizes the purpose of long-range wisdom medical treatment and online doctor-seeing. And under the condition that the target auscultation result represents that the corresponding target body sound digital signal is abnormal body sound, the buzzer is controlled to give an early warning, and the target auscultation result can be transmitted to an external terminal through the communication module. And is not particularly limited herein.
According to the body sound auscultation device provided by the invention, the buzzer and the communication module are arranged to output the target auscultation result, so that the purposes of remote intelligent medical treatment and online medical observation can be realized, and the purposes of wearable and portable health monitoring can also be realized, thereby improving the timeliness and flexibility of organ disease monitoring.
Referring to fig. 6, a schematic flow chart of the method for diagnosing body sound data according to the present invention is shown, wherein the method for diagnosing body sound data is based on the body sound auscultation device described in the foregoing embodiment, and an execution main body of the method for diagnosing body sound data may be a controller, and the controller may be a main controller built in the body sound auscultation device or a controller communicatively connected to the body sound auscultation device. The specific form of the controller is not limited herein. As shown in fig. 6, the method for diagnosing volume sound data may include the following steps:
and step 610, generating a target diagnosis instruction when the fact that the body sound auscultation device and the target part of the target object meet the preset distance relation is determined.
The preset distance relation can represent that the body sound auscultation device is in close contact with the target part of the target object.
Specifically, when the controller determines that the body sound auscultation device is a body sound auscultation chip and determines that the body sound auscultation chip is in close contact with the target portion of the target object, the target diagnosis instruction can be automatically generated, the close contact can be that the body sound auscultation chip is tightly attached to the target portion, the target portion can be a plurality of portions on the body of the target object, which is on clothes, where original body sound signals of different body sound categories can be collected, and the target object can be other human bodies to be diagnosed, such as the aged, the young and the middle-aged, children and the like.
And step 620, controlling the body sound data to be diagnosed corresponding to the target part to enter the body sound auscultation device based on the target diagnosis instruction, and acquiring a target diagnosis result of the body sound data to be diagnosed.
The body sound data to be diagnosed comprise original body sound signals of different body sound categories, and the target diagnosis result comprises body sound normal and body sound abnormal body sound data to be diagnosed.
Specifically, the controller may control the body sound auscultation device to collect data from the target portion and control the collected body sound data to be diagnosed to enter the body sound auscultation device based on the target diagnosis instruction, so as to process the body sound data to be diagnosed, thereby obtaining a target diagnosis result of the body sound data to be diagnosed.
And step 630, outputting a target diagnosis result.
Specifically, when it is determined that the body sound auscultation device obtains the target diagnosis result of the body sound data to be diagnosed, the output instruction of the target object can be obtained within the preset time period, and when the output instruction is received within the preset time period, the target diagnosis result can be output based on the output instruction; when the output instruction is not received within the preset time length, the target diagnosis result obtained this time can be stored.
According to the body sound data diagnosis method provided by the invention, based on the target diagnosis instruction generated when the body sound auscultation device and the target part of the target object meet the preset distance relationship, the target body sound auscultation device is controlled to execute the diagnosis operation aiming at the body sound data to be diagnosed and output the target diagnosis result, so that the purposes of wearable and portable health detection are achieved under the condition of not depending on a terminal or a server, and the convenience, rapidness and accuracy and reliability of organ disease diagnosis are greatly improved.
Optionally, the specific implementation process of step 630 may include:
outputting the target diagnosis result through a built-in buzzer of the body sound auscultation device; and/or outputting the target diagnosis result to an external terminal in communication connection with the body sound auscultation device.
It can be understood that, when the target object is an adult or the current scene of the target object is convenient to answer the voice information, the controller can instruct the target diagnosis result to be output through a built-in buzzer of the body sound auscultation device, so that the target object can know the self health condition in time; when the target object is a child or the current scene of the target object is inconvenient to answer the voice information, the controller can instruct to send the target diagnosis result to an external terminal in communication connection with the body sound auscultation device; when the target object is an adult and the current scene of the target object is convenient to answer voice information, the controller can instruct the target diagnosis result to be sent to an external terminal in communication connection with the body sound auscultation device under the action of WiFi or Bluetooth while instructing the target diagnosis result to be output through a built-in buzzer of the body sound auscultation device. The specific manner and scenario for outputting the target diagnosis result are not particularly limited herein.
According to the body sound data diagnosis method provided by the invention, the flexibility and the reliability of the output target diagnosis result are improved in a mode of outputting the target diagnosis result through the built-in buzzer of the body sound auscultation device and/or the external terminal connected with the body sound auscultation device, so that the purpose of effectively preventing organ diseases can be realized.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A body sound auscultation device is characterized by comprising an SoC chip, and a data acquisition module and a body sound diagnosis module which are integrated on the SoC chip, wherein the data acquisition module is connected with the body sound diagnosis module;
the data acquisition module is used for acquiring a target volume tone digital signal;
the body sound diagnosis module is used for determining a target auscultation result of the target body sound digital signal based on the target body sound digital signal and a target identification model;
the target recognition model is obtained by training an LSTM model and a full connection layer based on sample body sound digital signals and body sound labels, and the target auscultation result comprises body sound normality and body sound abnormality.
2. The body sound auscultation device of claim 1, wherein the body sound diagnosis module comprises a feature extraction module, a feature recognition module and a result modification module, which are connected in sequence, wherein:
the characteristic extraction module is used for carrying out characteristic extraction on the target body sound digital signal and determining the time-frequency domain body sound characteristic obtained by the characteristic extraction;
the characteristic identification module is used for identifying the time-frequency domain volume sound characteristic based on the target identification model and determining a target identification result of the target volume sound digital signal;
and the result correction module is used for carrying out noise filtration on the target identification result and determining a target auscultation result of the target body sound digital signal based on a result obtained by the noise filtration.
3. The auscultation device of claim 2, wherein the feature extraction module comprises a short-time fourier transform module, a mel filtering module, a logarithm operation module, a discrete cosine transform module and a normalization mapping module, which are connected in sequence, wherein:
the short-time Fourier transform module is used for framing the target volume tone digital signal and determining spectrogram characteristics corresponding to frame data obtained by framing based on windowing and frequency domain conversion operation;
the Mel filtering module is used for carrying out Mel filtering on the spectrogram characteristics to determine the Mel spectrogram characteristics;
the logarithm operation module is used for carrying out logarithm operation conversion on the Mel frequency spectrogram characteristics to determine logarithm Mel frequency spectrogram characteristics;
the discrete cosine transform module is used for performing discrete cosine transform on the logarithmic Mel frequency spectrum diagram characteristics to determine a Mel cepstrum coefficient;
and the normalization mapping module is used for normalizing the Mel cepstrum coefficients and determining the time-frequency domain body sound characteristics of the target body sound digital signals.
4. The body sound auscultation device of claim 2, wherein the data acquisition module comprises a digital microphone module, a down-sampling filter module and a noise reduction module connected in sequence, wherein:
the digital microphone module is used for acquiring an original body sound signal and converting the original body sound signal into a digital body sound signal;
the down-sampling filtering module is used for down-sampling the digital volume sound signal;
and the noise reduction module is used for reducing noise of the digital volume sound signal after down sampling and determining a target volume sound digital signal.
5. The body sound auscultation device of claim 4, wherein the body sound diagnosis module further comprises a configuration scheduling module, an output terminal of the configuration scheduling module is connected to an input terminal of the feature extraction module;
the configuration scheduling module is configured to configure a control signal matched with the volume class information for the original volume sound signal based on the volume class information corresponding to the original volume sound signal, and schedule a target recognition model based on the control signal.
6. The body sound auscultation device of claim 4, wherein the body sound diagnosis module further comprises a storage module, the storage module is connected with the feature recognition module;
the storage module is used for storing different weight parameters of different target recognition models, and each weight parameter corresponds to different control signals respectively.
7. The body sound auscultation device of any one of claims 2-6, wherein the training process of the target recognition model comprises:
acquiring a sample body sound digital signal set carrying a body sound label, and dividing the sample body sound digital signal set into a training sample and a testing sample, wherein the body sound label comprises a body sound category label, a normal body sound label and an abnormal body sound label;
training an initial recognition model containing an LSTM model and a full connection layer by using the training sample to obtain a trained recognition model;
and testing the recognition model obtained by training by using the test sample to determine a target recognition model.
8. The body sound auscultation device of any one of claims 2-6, further comprising a buzzer and a communication module, the buzzer and the communication module being respectively connected with the body sound diagnosis module, wherein:
the buzzer is used for carrying out early warning based on the target auscultation result; and the communication module is used for transmitting the target auscultation result to an external terminal.
9. A method for diagnosing volume sound data based on the volume sound auscultation device of any one of claims 1 to 7, comprising:
when the fact that the body sound auscultation device and the target part of the target object meet the preset distance relation is determined, a target diagnosis instruction is generated;
controlling the body sound data to be diagnosed corresponding to the target part to enter the body sound auscultation device based on the target diagnosis instruction, and acquiring a target diagnosis result of the body sound data to be diagnosed; the body sound data to be diagnosed comprises original body sound signals of different body sound categories;
and outputting the target diagnosis result.
10. The method for diagnosing voice data according to claim 9, wherein the outputting the target diagnosis result includes:
outputting the target diagnosis result through a buzzer arranged in the body sound auscultation device; and/or outputting the target diagnosis result to an external terminal in communication connection with the body sound auscultation device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211020188.7A CN115545148A (en) | 2022-08-24 | 2022-08-24 | Body sound auscultation device and body sound data diagnosis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211020188.7A CN115545148A (en) | 2022-08-24 | 2022-08-24 | Body sound auscultation device and body sound data diagnosis method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115545148A true CN115545148A (en) | 2022-12-30 |
Family
ID=84726171
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211020188.7A Pending CN115545148A (en) | 2022-08-24 | 2022-08-24 | Body sound auscultation device and body sound data diagnosis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115545148A (en) |
-
2022
- 2022-08-24 CN CN202211020188.7A patent/CN115545148A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110123367B (en) | Computer device, heart sound recognition method, model training device, and storage medium | |
| CN101023469B (en) | Digital filtering method, digital filtering equipment | |
| CN110880329B (en) | Audio identification method and equipment and storage medium | |
| US20210030390A1 (en) | Electronic stethoscope | |
| CN108899037A (en) | Animal vocal print feature extracting method, device and electronic equipment | |
| CN107622797A (en) | A kind of health based on sound determines system and method | |
| CN110444202B (en) | Composite voice recognition method, device, equipment and computer readable storage medium | |
| WO2019023879A1 (en) | Cough sound recognition method and device, and storage medium | |
| CN113436726B (en) | Automatic lung pathological sound analysis method based on multi-task classification | |
| CN109631104A (en) | Air quantity Automatic adjustment method, device, equipment and the storage medium of kitchen ventilator | |
| CN202801659U (en) | Electronic stethoscope based on intelligent identification function | |
| CN113539294A (en) | Method for collecting and identifying sound of abnormal state of live pig | |
| CN110720946A (en) | Intelligent auscultation system based on deep learning | |
| Sharan et al. | Cough sound analysis for diagnosing croup in pediatric patients using biologically inspired features | |
| CN115346561B (en) | Depression emotion assessment and prediction method and system based on voice characteristics | |
| CN110415824B (en) | Cerebral apoplexy disease risk assessment device and equipment | |
| CN115910097A (en) | Audible signal identification method and system for latent fault of high-voltage circuit breaker | |
| CN201683910U (en) | Intelligent cardiopulmonary analyzing instrument | |
| CN111789629A (en) | Breath sound intelligent diagnosis and treatment system and method based on deep learning | |
| CN117762372A (en) | Multi-mode man-machine interaction system | |
| CN110956949B (en) | Buccal type silence communication method and system | |
| CN117542373A (en) | Non-air conduction voice recovery system and method | |
| CN115545148A (en) | Body sound auscultation device and body sound data diagnosis method | |
| US20150190091A1 (en) | Device, system and method for detection of fluid accumulation | |
| CN215349053U (en) | Congenital heart disease intelligent screening robot |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |