Intelligent acoustic positioning system for sick pigs in hog house

Abstract

The invention discloses an intelligent acoustic positioning system for sick pigs in a hog house, and relates to the field of sound recognition and positioning. The problems that an existing sick pig acoustic positioning method is poor in voice recognition accuracy and large in positioning error in a complex acoustic environment are solved. According to the invention, a time difference and phase difference complementary positioning principle is adopted, and an adaptive beam forming technology is combined, so that the problem of inaccurate sound source positioning in a pig house environment with strong reflection and high noise is solved, multi-feature extraction and intelligent diagnosis are carried out on enhanced sound data through the sick pig sound recognition unit, and the recognition accuracy is improved. The method is mainly used for positioning sick pigs in a complex breeding environment.

Description

Technical Field

[0001] This invention relates to the field of sound recognition and positioning. Background Technology

[0002] With the development of large-scale and intensive pig farming, disease prevention and control has become a core issue restricting the industry's efficiency. Traditional methods of detecting sick pigs rely on manual inspections, which are inefficient, have a high rate of missed detections, and are prone to cross-infection. There is an urgent need for intelligent technologies to improve disease monitoring capabilities.

[0003] In the field of animal monitoring in livestock farms, existing technologies mainly include machine vision monitoring, wireless sensor network (WSN) monitoring, and acoustic monitoring. Among them, acoustic monitoring technology, due to its unique advantages such as being non-contact, highly real-time, unaffected by lighting conditions, and having a wide monitoring range with virtually no blind spots, shows great application potential in animal health and behavior analysis. In particular, combining artificial intelligence (AI) and the Internet of Things (IoT) technologies, through the continuous collection and analysis of pig sound signals, is expected to become a key technological path for achieving early identification and warning of sick pigs. However, in the complex actual farming environment, simple sound recognition technology is insufficient to meet the needs of precise prevention and control. Therefore, in automated farming processes, it is of practical significance to use automated technology to determine the health status of pigs and quickly locate the pigs making the sounds.

[0004] The main problems with existing acoustic localization technology for sick pigs are summarized as follows:

[0005] Insufficient positioning accuracy and reliability: In complex acoustic environments such as pigsties with strong reflections and multiple noises (such as fan noise, other pig noises, and equipment operation noise), existing sound source localization technologies (such as technologies based on single or simple arrays) are easily affected by reverberation and interference, making it difficult to stably and accurately estimate the direction of the sound source, resulting in large positioning errors.

[0006] The system has weak anti-interference ability: the background noise in pig houses is diverse and varies in intensity. Existing systems often lack effective adaptive noise reduction mechanisms, which causes the sound signal of the target sick pig to be submerged, resulting in a low signal-to-noise ratio (SNR) that seriously affects the accuracy of subsequent identification and positioning.

[0007] The specificity and robustness of feature extraction need to be improved: Feature extraction for pathological sounds in pigs (such as coughing and wheezing) often directly uses general speech features (such as MFCC), lacking in-depth modeling of the physiological mechanism of pig vocalization and the changes in acoustic characteristics caused by diseases. This results in insufficient discriminativeness and robustness of features in complex environments, leading to poor accuracy of model sound recognition.

[0008] In summary, existing technologies have not yet effectively solved the comprehensive challenge of high-precision, high-reliability, and real-time individualized localization and recognition of the sounds of sick pigs in complex acoustic environments. These problems urgently need to be addressed. Summary of the Invention

[0009] The purpose of this invention is to solve the problems of poor sound recognition accuracy and large positioning error in existing acoustic localization methods for sick pigs in complex acoustic environments. This invention provides an intelligent acoustic localization system for sick pigs in pig houses.

[0010] An intelligent acoustic positioning system for sick pigs in a pigsty includes:

[0011] The acquisition unit is used to acquire mixed sounds within the pigsty via a microphone array. The sound source localization and noise reduction unit is used to calculate the arrival time difference and phase difference between any two microphones for each target sound source based on the mixed sounds acquired by each microphone in the array. Based on all arrival time differences, phase differences, and the geometry of the microphone array, it uses a direction-of-arrival estimation algorithm to deduce the directional angle of the target sound source. It is also used to employ adaptive beamforming technology to denoise and enhance the sound emitted by the target sound source at each directional angle, obtaining enhanced sound data.

[0012] The sick pig sound recognition unit is used to extract multiple features and perform intelligent diagnosis on the enhanced sound data to determine whether the sound emitted by the target sound source corresponding to the enhanced sound data comes from a sick pig.

[0013] Preferably, the microphone array is an array of six microphones in a specific geometric shape.

[0014] Preferably, the specific geometry is circular or square.

[0015] Preferably, the time difference and phase difference between any two microphones at which each target sound source arrives are calculated using a cross-correlation function.

[0016] Preferably, the sick pig sound recognition unit includes:

[0017] The feature extraction module is used to extract Mel frequency cepstral coefficients (timbre fingerprint features) and linear predictive coding coefficients as feature parameters from the enhanced sound data; among them, the Mel frequency cepstral coefficients are used as timbre fingerprint features;

[0018] The intelligent diagnosis module uses a pre-trained three-channel convolutional neural network model to predict the enhanced sound data input to it, as well as the Mel frequency cepstral coefficients and linear predictive coding coefficients corresponding to the enhanced sound data, and obtains the diagnosis result, which is either a sick pig or a non-sick pig.

[0019] Preferably, the method for extracting Mel-frequency cepstral coefficients from the enhanced sound data is as follows:

[0020] The enhanced audio data is converted into a spectrum by Fourier transform. The Mel spectral energy in the spectrum is extracted by Mel filter bank. The logarithm of the Mel spectral energy is then taken and subjected to discrete cosine transform to obtain the Mel frequency cepstral coefficients.

[0021] Preferably, the implementation method for extracting linear prediction coding coefficients from the enhanced audio data is as follows:

[0022] The enhanced audio data is pre-emphasized.

[0023] The enhanced audio data after pre-emphasis is processed by frame segmentation, and divided into frame sequences with a frame length of 20 to 30 milliseconds and a frame overlap rate of 50%; a Hamming window is applied to each frame signal after segmentation.

[0024] Based on the previous frame of each frame of signal after applying the Hamming window Using audio signal sample values ​​at each time point, a linear prediction equation is constructed;

[0025] With the goal of minimizing the mean square value of the prediction error, the linear prediction equation is solved using the autocorrelation method to obtain a set of prediction coding coefficients. This set of prediction coding coefficients includes... Each predictive coding coefficient.

[0026] Preferably, the expression for the linear prediction equation is: ;in,

[0027] The current time in each frame of the signal after applying the Hamming window. The audio signal sample value, ;

[0028] For the first One linear prediction coding coefficient;

[0029] For the prediction order;

[0030] This represents the prediction error;

[0031] For the front The audio signal sample value at each moment.

[0032] Preferably, the intelligent diagnostic module includes an input layer, a flattened layer, a first fully connected layer, a second fully connected layer, and an output layer arranged sequentially in the propagation direction;

[0033] The input layer is used to fuse the received three-channel data into a three-channel fused feature map by employing a feature concatenation and attention mechanism fusion strategy.

[0034] A flattening layer is used to transform the three-channel fused feature map into a one-dimensional vector;

[0035] The first fully connected layer is used to integrate global information and learn nonlinear combination of one-dimensional vectors, and outputs the first feature vector.

[0036] The first fully connected layer is used to integrate global information from the first feature vector and output the second feature vector.

[0037] The output layer is used to perform linear transformation and activation classification on the second feature vector, outputting the probability of each category, and taking the category corresponding to the maximum probability as the diagnostic result.

[0038] An intelligent acoustic positioning system for sick pigs in a pigsty also includes a management platform for receiving and processing the output results of the sound source localization and noise reduction unit and the sick pig sound recognition unit, and for visual display and alarm.

[0039] The beneficial effects of this invention are:

[0040] This invention discloses an intelligent acoustic positioning system for sick pigs in pigsties, which systematically integrates and innovates three cutting-edge technologies: microphone array signal processing, adaptive beamforming, and deep learning. This integration is not a simple aggregation, but rather a synergistic effect at the algorithm level, achieving improved system performance. Specifically, this is reflected in the following aspects:

[0041] (1) High-precision sound source localization technology: The positioning principle of complementary time difference and phase difference is adopted and combined with adaptive beamforming technology to solve the problem of inaccurate sound source localization in pig house environment with strong reflection and high noise. Even in strong noise and reverberation environment, the sound source direction can still be locked with an accuracy of ±5°, realizing the qualitative change from "area alarm" to "individual positioning".

[0042] (2) High accuracy of sick pig identification: The sick pig sound recognition unit is used to extract features and make intelligent diagnoses on the enhanced sound data. It can provide early warnings by detecting abnormal coughing, panting and other sounds of sick pigs before they show obvious clinical symptoms.

[0043] Furthermore, the intelligent diagnostic module in the sick pig sound recognition unit learns sound features from different dimensions: one channel focuses on the overall spectral structure (enhanced sound data), one channel focuses on MFCC features (Mel frequency cepstral coefficients) that mimic human ear perception, and another channel focuses on LPC features (linear predictive coding coefficients) related to physiological structure.

[0044] The intelligent diagnostic module employs a three-channel convolutional neural network model to process audio features. This model can learn multiple feature representations of sound in parallel (perceptual features of MFCC, physiological structural features of LPC, etc.). Through multi-feature fusion, it greatly improves the model's generalization ability and robustness in complex scenarios. The recognition accuracy exceeds 90%, a figure derived from the mining of specific pathological acoustic features. This indicates that the model is not only recognizing "pig noises" but also diagnosing "pathological pig noises," demonstrating high technical barriers and improved recognition accuracy.

[0045] This invention enables early detection, precise location, and efficient warning of diseased pigs in complex breeding environments. Attached Figure Description

[0046] Figure 1 This is a schematic diagram illustrating the principle of an intelligent acoustic positioning system for sick pigs in a pigsty, as described in this invention.

[0047] Figure 2 This is a physical diagram of the PCB layout for a six-microphone array;

[0048] Figure 3 This is a schematic diagram of a demonstration machine integrating the acquisition unit, the sound source localization and noise reduction unit, and the sick pig sound recognition unit in this invention;

[0049] Figure 4 This is a demonstration image of voice recognition using a demo unit. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0052] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0053] Specific Implementation Method 1: Combination Figure 1 This embodiment describes an intelligent acoustic positioning system for sick pigs in a pigsty, comprising:

[0054] The acquisition unit is used to collect mixed sound within the pigsty via a microphone array; the microphone array is an array of six microphones in a specific geometric shape, which can be circular or square. Figure 2 In application, multiple microphone arrays can be optimally arranged according to the different sizes and specifications of the pigsty, and the multiple arrays can cross-scan to achieve accurate location of sick pigs.

[0055] The sound source localization and noise reduction unit is used to calculate the arrival time difference and phase difference between any two microphones for each target sound source based on the mixed sound collected by each microphone in the array, and to deduce the directional angle of the target sound source using a direction-of-arrival estimation algorithm based on all arrival time differences, phase differences, and the geometry of the microphone array; it is also used to denoise and enhance the sound emitted by the target sound source at each directional angle using adaptive beamforming technology to obtain enhanced sound data.

[0056] The sick pig sound recognition unit is used to extract multiple features and make intelligent diagnoses on the enhanced sound data to determine whether the sound emitted by the target sound source corresponding to the enhanced sound data comes from a sick pig.

[0057] The management platform receives and processes the output results of the sound source localization and noise reduction unit and the sick pig sound recognition unit, and provides visual display and alarms. The output results of the sound source localization and noise reduction unit include the directional angle of each target sound source and the enhanced sound data, while the output results of the sick pig sound recognition unit indicate whether the pig is sick or not.

[0058] This system forms a complete closed loop of "precise positioning → signal enhancement → intelligent recognition → visual alarm," effectively solving the problems of low efficiency and high missed detection rate in traditional manual inspections. It provides farms with a data-driven solution for precise disease prevention and control and efficient management. The system integrates the acquisition unit, sound source localization and noise reduction unit, and sick pig sound recognition unit from this invention into a demonstration machine, which was used in [the demonstration]. Figure 3 ; Use the demo unit for voice recognition, see [link / reference] Figure 4 .

[0059] When a pig vocalizes, the sound waves propagate outwards as spherical waves. Due to the different spatial positions of the microphones, the sound waves arrive at each microphone at slightly different times; this is known as the Time Difference of Arrival (TDOA). Simultaneously, this also results in a phase difference when the sound waves reach different microphones. The system uses algorithms such as cross-correlation functions to accurately calculate the TDOA and phase difference between each pair of microphones. Based on these time / phase difference data and the geometry of the microphone array, the direction of arrival (DOA) estimation algorithm (such as GCC-PHAT, MUSIC, etc.) can be used to deduce the azimuth angle (horizontal azimuth) of the sound source.

[0060] To address the issues of noisy pigsty environments, including fan noise and the sounds of other healthy pigs, this invention employs adaptive beamforming technology. It performs real-time digital weighting processing on the signals received by the array, automatically adjusting parameters to form a "pickup beam" pointing towards the target sound source (sick pig), maximizing signal gain in that direction. Simultaneously, it creates "nulls" in other interference directions (such as noise sources), significantly attenuating signals from these directions. This substantially improves the signal-to-noise ratio (SNR) of the target sound, laying a solid foundation for accurate identification.

[0061] This invention utilizes adaptive beamforming technology to create a high-gain sound pickup beam in the positioning direction while suppressing interference from other directions, significantly improving the signal-to-noise ratio of the target sound and ensuring the purity of subsequent recognition input. A sick pig sound recognition unit is used to extract features and perform intelligent diagnosis on the enhanced sound data, enabling early warning through abnormal coughing, wheezing, and other sounds in sick pigs before obvious clinical symptoms appear.

[0062] Furthermore, the sick pig sound recognition unit includes a feature extraction module and an intelligent diagnosis module;

[0063] The feature extraction module is used to extract Mel-frequency cepstral coefficients (timbre fingerprint features) and linear predictive coding coefficients as feature parameters from the enhanced sound data; among them,

[0064] The method for extracting Mel-frequency cepstral coefficients from enhanced audio data is as follows:

[0065] The enhanced audio data is converted into a spectrum by Fourier transform. The Mel spectral energy in the spectrum is extracted by Mel filter bank. The logarithm of the Mel spectral energy is then taken and subjected to discrete cosine transform to obtain the Mel frequency cepstral coefficients.

[0066] Mel Frequency Cepstral Coefficients (MFCC): These features are designed to mimic the characteristics of human hearing and are excellent at describing the spectral shape (timbre) of sound. They are crucial for distinguishing different types of sounds (such as coughing, wheezing, and normal calls). The process involves converting a sound wave signal into a spectrum via Fourier transform, extracting the perceptually relevant frequency bands using a Mel filter bank, and then obtaining the cepstral coefficients through a discrete cosine transform (DCT). This process highlights the advantages of MFCC in simulating the characteristics of human hearing. MFCC demonstrates superior performance compared to traditional spectral features in terms of noise suppression, frequency resolution, and computational efficiency, and exhibits extremely high robustness in complex environments.

[0067] The method for extracting linear predictive coding coefficients from enhanced audio data is as follows:

[0068] S1. Pre-emphasize the enhanced audio data to compensate for high-frequency attenuation of the signal during propagation;

[0069] S2. The enhanced audio data after pre-emphasis is processed by dividing it into frames, which are then divided into a frame sequence with a frame length of 20 to 30 milliseconds and a frame overlap rate of 50%, in order to avoid the influence of signal non-stationarity.

[0070] S3. Apply a Hamming window to each frame of signal after framing to reduce spectral leakage caused by frame truncation.

[0071] S4. Based on the previous frame signal after applying the Hamming window... Using audio signal sample values ​​at each time point, a linear prediction equation is constructed;

[0072] The expression for the linear prediction equation is: ;in,

[0073] The current time in each frame of the signal after applying the Hamming window. The audio signal sample value, ;

[0074] For the first One linear prediction coding coefficient;

[0075] For the prediction order;

[0076] This represents the prediction error;

[0077] For the front The audio signal sample value at each moment.

[0078] S5. With the goal of minimizing the mean square value of the prediction error, the linear prediction equation is solved using the autocorrelation method to obtain a set of prediction coding coefficients. This set of prediction coding coefficients includes... Each predictive coding coefficient.

[0079] Linear Predictive Coding (LPC) coefficients: This feature is primarily used to characterize the formant structure of sound. Respiratory diseases (such as pneumonia and bronchitis) alter the morphology of a pig's respiratory tract, directly affecting the formant frequencies and bandwidth of its vocalizations. Therefore, LPC coefficients are a key and sensitive feature for identifying respiratory diseases. The acquired sound signals undergo pre-emphasis, framing, and windowing processing.

[0080] The intelligent diagnosis module uses a pre-trained three-channel convolutional neural network model to predict the enhanced sound data input to it, as well as the Mel frequency cepstral coefficients and linear predictive coding coefficients corresponding to the enhanced sound data, and obtains the diagnosis result, which is either a sick pig or a non-sick pig.

[0081] Furthermore, the intelligent diagnostic module includes an input layer, a flattened layer, a first fully connected layer, a second fully connected layer, and an output layer arranged sequentially in the propagation direction;

[0082] The input layer is used to fuse the received three-channel data into a three-channel fused feature map by employing a feature concatenation and attention mechanism fusion strategy; that is, to transform the three-channel features (each 64×64×1) into a 64×64×3 three-channel fused feature map.

[0083] The flattening layer is used to transform the three-channel fused features into a one-dimensional vector (converting 64×64×3 features into a 12288-dimensional vector).

[0084] The first fully connected layer is used to integrate global information and learn nonlinear combination of one-dimensional vectors, and outputs the first feature vector.

[0085] The first fully connected layer is used to integrate global information from the first feature vector and output the second feature vector.

[0086] The output layer is used to perform linear transformation and activation classification on the second feature vector, outputting the probability of each category, and taking the category corresponding to the maximum probability as the diagnostic result.

[0087] The intelligent diagnostic module learns sound features from different dimensions simultaneously: one channel focuses on the overall spectral structure, another on MFCC features mimicking human auditory perception, and a third on LPC features related to physiological structure. The model then fuses these high-level features extracted from the three channels in the backend to make a more comprehensive and accurate judgment. By fusing the high-level features from the three channels, sound recognition and diagnosis are completed through classification.

[0088] By utilizing a three-channel convolutional neural network model that integrates Mel frequency cepstral coefficients and linear predictive coding coefficients, the enhanced sound is subjected to in-depth feature analysis and intelligent diagnosis, with an accuracy rate exceeding 90%. It can provide early warning by detecting abnormal coughing, panting, and other sounds in sick pigs before obvious clinical symptoms appear.

[0089] Before application, the three-channel convolutional neural network model is pre-trained using a large amount of labeled "healthy pig sounds" and "sick pig sounds". After pre-training, the relevant features extracted from the new sounds are classified, and the probability that they are the sounds of sick pigs is output, thus improving the recognition accuracy.

[0090] The pre-training process can be implemented using existing technologies, and the technical tools employed are as follows:

[0091] 1. The Adam optimizer is used, with an initial learning rate of 0.001, which decays to 0.1 every 10 rounds; the cross-entropy loss function is used; the training batch size is set to 32, the number of iterations is 50 rounds, and an early stopping strategy is used to prevent overfitting.

[0092] 2. Technical tools: Signal processing uses Python + Librosa library and SciPy library; the three-channel convolutional neural network model is built and trained using TensorFlow / PyTorch framework; the data annotation tool uses LabelStudio, which supports the annotation of sound categories and feature regions.

[0093] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. An intelligent acoustic positioning system for sick pigs in a pigsty, characterized in that, include: The acquisition unit is used to collect mixed sounds inside the pigsty via a microphone array; The sound source localization and noise reduction unit is used to calculate the arrival time difference and phase difference of each target sound source to any two microphones based on the mixed sound collected by each microphone in the array, and to deduce the direction angle of the target sound source by using a direction-of-arrival estimation algorithm based on all arrival time differences, phase differences and the geometry of the microphone array. It is also used to denoise and enhance the sound emitted by the target sound source at various directional angles using adaptive beamforming technology, so as to obtain enhanced sound data; The sick pig sound recognition unit is used to extract multiple features and perform intelligent diagnosis on the enhanced sound data to determine whether the sound emitted by the target sound source corresponding to the enhanced sound data comes from a sick pig.

2. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 1, characterized in that, A microphone array is an array of six microphones in a specific geometric shape.

3. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 1, characterized in that, The specific geometric shape is either a circle or a square.

4. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 1, characterized in that, The arrival time difference and phase difference of each target sound source to any two microphones are calculated using a cross-correlation function.

5. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 1, characterized in that, The sick pig sound recognition unit includes: The feature extraction module is used to extract Mel-frequency cepstral coefficients and linear predictive coding coefficients from the enhanced sound data as feature parameters; among them, the Mel-frequency cepstral coefficients serve as timbre fingerprint features. The intelligent diagnosis module uses a pre-trained three-channel convolutional neural network model to predict the enhanced sound data input to it, as well as the Mel frequency cepstral coefficients and linear predictive coding coefficients corresponding to the enhanced sound data, and obtains the diagnosis result, which is either a sick pig or a non-sick pig.

6. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 5, characterized in that, The method for extracting Mel-frequency cepstral coefficients from enhanced audio data is as follows: The enhanced audio data is converted into a spectrum by Fourier transform. The Mel spectral energy in the spectrum is extracted by Mel filter bank. The logarithm of the Mel spectral energy is then taken and subjected to discrete cosine transform to obtain the Mel frequency cepstral coefficients.

7. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 5, characterized in that, The method for extracting linear predictive coding coefficients from enhanced audio data is as follows: The enhanced audio data is pre-emphasized. The enhanced audio data after pre-emphasis is processed in frames and divided into frame sequences with a frame length of 20 to 30 milliseconds and a frame overlap rate of 50%. Apply a Hamming window to each frame of the signal after framing; Based on the previous frame of each frame of signal after applying the Hamming window Using audio signal sample values ​​at each time point, a linear prediction equation is constructed; With the goal of minimizing the mean square value of the prediction error, the linear prediction equation is solved using the autocorrelation method to obtain a set of prediction coding coefficients. This set of prediction coding coefficients includes... Each predictive coding coefficient.

8. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 7, characterized in that, The expression for the linear prediction equation is: ;in, The current time in each frame of the signal after applying the Hamming window. The audio signal sample value, ; For the first One linear prediction coding coefficient; For the prediction order; This represents the prediction error; For the front The audio signal sample value at each moment.

9. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 1, characterized in that, The intelligent diagnostic module includes an input layer, a flattened layer, a first fully connected layer, a second fully connected layer, and an output layer arranged sequentially in the propagation direction; The input layer is used to fuse the received three-channel data into a three-channel fused feature map by employing a feature concatenation and attention mechanism fusion strategy. A flattening layer is used to transform the three-channel fused feature map into a one-dimensional vector; The first fully connected layer is used to integrate global information and learn nonlinear combination of one-dimensional vectors, and outputs the first feature vector. The first fully connected layer is used to integrate global information from the first feature vector and output the second feature vector. The output layer is used to perform linear transformation and activation classification on the second feature vector, outputting the probability of each category, and taking the category corresponding to the maximum probability as the diagnostic result.

10. The intelligent acoustic positioning system for sick pigs in a pigsty according to claim 1, characterized in that, It also includes a management platform for receiving and processing the output results of the sound source localization and noise reduction unit and the sick pig sound recognition unit, and for visual display and alarm.

Images