AI bark stopper

Abstract

The invention provides an AI bark stopping device which comprises an MCU, a sound collector, a vibration data collector, a power module and a bark stopping module, the MCU is used for receiving signals of the sound collector and the vibration data collector, analyzing and identifying data through a built-in AI model, judging whether bark stopping is needed or not, and controlling the bark stopping module to execute corresponding bark stopping operation; the sound collector is used for collecting pet dog barking data and sending the collected data to the MCU in real time; the vibration data collector is used for collecting pet dog throat vibration data and sending the collected data to the MCU1 in real time; the power supply module is used for providing electric energy for the AI barking stopper; and the barking stopping module is used for triggering related intervention measures according to the instruction sent by the MCU1 so as to help to reduce or stop excessive barking of the pet dog. The method has the beneficial effects that when the AI model recognizes dog barks and detects throat vibration signals, the system immediately triggers the barking stopping action.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an AI barking control device. Background Art

[0002] Currently, pet bark control devices on the market rely primarily on two detection mechanisms: 1. Pet activity detection (e.g., motion sensors); 2. Sound intensity detection (e.g., microphones monitoring barking decibels). However, both methods have the following significant drawbacks:

[0003] Problem 1: Missed triggering (the barking control does not trigger when it should)

[0004] The intensity of different pets' barks varies greatly (small dogs and high-pitched dog breeds may not reach the trigger threshold);

[0005] The intensity of a pet's calls can vary depending on its emotions (e.g., anxiety, excitement), making it difficult for the system to accurately identify the pet.

[0006] Problem 2: False triggering (false triggering when barking should not be stopped)

[0007] When ambient noise (such as thunder, vehicle horns, and human voices) reaches the set intensity, the barking control function will be mistakenly triggered;

[0008] Non-barking pet sounds (such as purring while playing) may also be misinterpreted as barking. Summary of the Invention

[0009] In order to solve the problems in the prior art, the present invention provides an AI barking control device, including an MCU, a sound collector, a vibration data collector, a power module, and a barking control module, wherein:

[0010] The MCU is used to receive signals from the sound collector and the vibration data collector, analyze and identify the data through the built-in AI model, determine whether barking control is required, and control the barking control module to perform the corresponding barking control operation;

[0011] The sound collector is used to collect the barking data of the pet dog and send the collected data to the MCU in real time;

[0012] The vibration data collector is used to collect the vibration data of the pet dog's throat and send the collected data to the MCU in real time;

[0013] The power supply module is used to provide power to the AI barking control device;

[0014] The barking control module triggers relevant intervention measures according to the instructions sent by the MCU to help reduce or stop the excessive barking of pet dogs.

[0015] As a further improvement of the present invention, the AI model is a model based on a recurrent neural network, and the workflow of the AI model is as follows:

[0016] Model training steps: The collected pet dog barking data samples are transmitted to the training platform. The training platform extracts the sample features and passes them to the recurrent neural network. The recurrent neural network is trained based on the data features and results, and the training results are placed in the weight parameter table.

[0017] Target recognition step: The barking sounds of the pet dog collected by the sound collection module are sent to the recurrent neural network after feature extraction. The recurrent neural network uses the trained weight parameters to analyze and score. The scoring system will score the input sound and send the result. The final recognition result is obtained by comparing this result with the preset threshold.

[0018] As a further improvement of the present invention, the model training step specifically includes:

[0019] Step 1, data collection: Build a database of audio samples containing various dog barking sounds. The data sources include user-provided data samples and public data samples crawled from the Internet.

[0020] Step 2, data sorting: screening and segmenting the sample data of the audio sample database in step 1;

[0021] Step 3, data labeling: label each audio segment at each moment as whether it is a dog barking;

[0022] Step 4, data enhancement: Use noise addition, speed change, and pitch change methods to enhance the labeled data;

[0023] Step 5: Model training: Build a dog barking detection model and determine the appropriate loss function for model training.

[0024] Step 6, model testing: Test the trained model on a pre-recorded test set to screen out models with good performance.

[0025] As a further improvement of the present invention, the target identification step specifically includes:

[0026] Step S1, feature extraction: performing frame and window processing on the sound data collected by the sound collector to extract logfbank features;

[0027] Step S2, model prediction: the logfbank features extracted in step S1 are input into the model, and the last layer of the model uses a sigmoid function to output a probability value;

[0028] Step S3, result judgment: compare the probability value of step S2 with a preset threshold value. If the probability value is greater than the threshold value, the system is judged to be activated; if the probability value is less than the threshold value, the system is judged to be inactivated.

[0029] As a further improvement of the present invention, the MCU runs the following program:

[0030] Step Y1: receiving vibration signal data of the pet dog's throat detected by the vibration data collector, first converting the data from the time domain to the frequency domain, and then detecting the energy spectrum of the data to identify the vibration of the pet dog's throat;

[0031] Step Y2: Control the barking prevention module to stop the dog from barking based on the throat vibration of the dog identified in step Y1 and the barking characteristics of the dog determined in the target identification step.

[0032] As a further improvement of the present invention, the step Y1 further includes:

[0033] Step 1, data collection: collect pet dog throat vibration data;

[0034] Step 2: Band-pass filtering: Use band-pass filtering to remove the interference signal generated by the pet dog's movement;

[0035] Step 3, Fast Fourier Transform: Perform Fast Fourier Transform on the data filtered in Step 2 to transform the energy spectrum in the time domain into the frequency domain;

[0036] Step 4, normalization processing: normalize the energy spectrum data in the frequency domain of step 3, and then use convolution operation to identify the vibration of the pet dog's throat.

[0037] As a further improvement of the present invention, the AI barking control device further includes a human-computer interaction interface, and the MCU is connected to the human-computer interaction interface.

[0038] As a further improvement of the present invention, the human-computer interaction interface includes a button and a liquid crystal display, and the MCU is connected to the liquid crystal display and the button respectively, and the button is used to set the device parameters of the AI barking stopper.

[0039] As a further improvement of the present invention, the barking prevention module includes a ZVC boost circuit and / or a vibration motor and / or a buzzer, the sound collector is a microphone, and the vibration data collector is a gravity sensor or a six-axis gyroscope.

[0040] As a further improvement of the present invention, the power module includes a battery, a power management system and / or a charging management system.

[0041] The beneficial effects of the present invention are: 1. The AI bark stopper of the present invention accurately identifies dog barking through an AI model, and at the same time combines the detected throat vibration signal of the pet dog. When these two conditions are met at the same time, the MCU triggers a buzzer, vibration, electric shock and other actions to stop the barking; 2. The AI bark stopper of the present invention combines sound and vibration dual verification to ensure accurate barking cessation, solving the problem of missed triggering / false triggering of existing bark stoppers, and providing dog lovers with a quiet and harmonious dog-raising environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a functional block diagram of the AI barking control device of the present invention;

[0043] Figure 2 This is a flow chart of the AI barking control model training of the present invention;

[0044] Figure 3 This is a target recognition flow chart of the AI barking control device of the present invention;

[0045] Figure 4 This is a flow chart of vibration detection of the AI barking control device of the present invention. DETAILED DESCRIPTION

[0046] In response to the current technical defects, the present invention introduces AI technology to achieve accurate pet dog barking recognition and barking control, solving the defects of false triggering and missed triggering. The present invention adopts an AI algorithm model and introduces massive data samples to train the model. The trained model is run on the product platform. The product platform uses the collected pet dog barking and vibration data features as the input of the AI model. Through the analysis of the AI model, it can accurately identify the pet dog barking and trigger the barking control function.

[0047] like Figure 1 As shown, the present invention discloses an AI barking control device, including MCU1, a sound collector, a vibration data collector, a power module, and a barking control module, wherein:

[0048] MCU 1: Receives signals from the sound and vibration data collectors, analyzes and identifies the data using a built-in AI model, determines if barking suppression is necessary, and controls the barking suppression module to execute the corresponding suppression operation.

[0049] Sound collector: used to collect pet dog barking data and send the collected data to MCU1 in real time;

[0050] Vibration data collector: used to collect vibration data of the pet dog's throat and send the collected data to MCU 1 in real time; the vibration data collector of the present invention is a microphone;

[0051] Power module: used to provide power to the AI barking control device and support the stable operation of each component;

[0052] The power module includes a battery, a power management system 8, and a charging management system 9. The power management system 8 is responsible for efficiently distributing and managing battery energy, ensuring stable power supply to the various modules of the AI bark control device (microphone 3, MCU 1, ZVC boost circuit 5, vibration motor 6, buzzer 7, etc.), and optimizing power consumption to extend battery life. The charging management system is responsible for safely controlling the battery charging process (including charge and discharge protection, temperature monitoring and charging status indication), preventing overcharging / overdischarging, and extending battery life. The AI bark control device has a built-in battery.

[0053] Barking control module: triggers relevant intervention measures based on the instructions sent by MCU 1 to help reduce or stop the excessive barking of pet dogs.

[0054] The barking control module includes a ZVC boost circuit 5, a vibration motor 6, and a buzzer 7. The output voltage of the ZVC boost circuit 5 is as high as 3KV. The vibration motor 6 generates tactile feedback through vibration. The buzzer 7 is an electromagnetic buzzer for emitting an audible audio warning sound. The ZVC boost circuit 5, the vibration motor 6, and the buzzer 7 can be flexibly configured according to actual needs. As a preferred embodiment of the present invention, the AI barking control device is installed with a ZVC boost circuit 5, a vibration motor 6, and a buzzer 7. Once the barking control is triggered, the AI barking control device of the present invention performs the following actions: first, the buzzer 7 alarms for 1.5 seconds, then the vibration motor 6 vibrates for 1.5 seconds, and finally, the ZVC boost circuit 5 generates a 1.5-second electric shock (the electric shock is a non-harmful electrostatic stimulation that complies with the ISO10993 animal welfare standard).

[0055] The core of the AI model of the present invention is a model based on a recurrent neural network (RNN). In practical application, it is divided into two parts: the first is model training, and the second is target recognition. The workflow of the AI model is as follows:

[0056] Model training steps: The collected pet dog barking data samples are transmitted to the training platform. The training platform extracts the sample features and passes them to the recurrent neural network. The recurrent neural network is trained based on the data features and results, and the training results are placed in the weight parameter table.

[0057] Target recognition step: The barking sounds of the pet dog collected by the sound collection module are sent to the recurrent neural network after feature extraction. The recurrent neural network uses the trained weight parameters to analyze and score. The scoring system will score the input sound and send the result. The final recognition result is obtained by comparing this result with the preset threshold.

[0058] like Figure 2 As shown in the figure, the model training steps specifically include:

[0059] Step 1, data collection: Build a database of audio samples containing various dog barking sounds. The data sources include user-provided data samples and public data samples crawled from the Internet.

[0060] Step 2, data sorting: screening and segmenting the sample data of the audio sample database in step 1;

[0061] Step 3, data labeling: label each audio segment at each moment as whether it is a dog barking;

[0062] Step 4, data enhancement: Use noise addition, speed change, and pitch change methods to enhance the labeled data;

[0063] Step 5: Model training: Build a dog barking detection model and determine the appropriate loss function for model training.

[0064] Step 6, model testing: Test the trained model on a pre-recorded test set to screen out models with good performance.

[0065] like Figure 3 As shown in FIG, the target recognition step specifically includes:

[0066] Step S1, feature extraction: perform frame and window processing on the sound data collected by the sound collector to extract logfbank features;

[0067] Step S2, model prediction: The logfbank features extracted in step S1 are input into the model. The last layer of the model uses the sigmoid function to output a probability value.

[0068] Step S3, result judgment: compare the probability value of step S2 with a preset threshold. If it is greater than the threshold, it is determined to be activated, which means that the dog barking sound is recognized. If it is less than the threshold, it is determined to be inactivated, which means that the dog barking sound is not recognized.

[0069] MCU 1 runs the following program:

[0070] Step Y1: receiving vibration signal data of the pet dog's throat detected by the vibration data collector, first converting the data from the time domain to the frequency domain, and then detecting the energy spectrum of the data to identify the vibration of the pet dog's throat;

[0071] Step Y2: Based on the pet dog's throat vibration identified in step Y1 and the pet dog's barking characteristics determined in the target recognition step, the barking prevention module is controlled to stop the dog from barking.

[0072] like Figure 4 As shown, step Y1 also includes:

[0073] Step 1, data collection: collect pet dog throat vibration data;

[0074] Step 2: Band-pass filtering: Band-pass filtering is used to remove interference signals generated by the pet dog's movement, as well as invalid components such as walking and collision;

[0075] Step 3, Fast Fourier Transform: Perform Fast Fourier Transform on the data filtered in Step 2 to transform the energy spectrum in the time domain into the frequency domain;

[0076] Step 4, normalization processing: Normalize the energy spectrum data in the frequency domain in step 3, and then use correlation detection to identify the vibration of the pet dog's throat; the correlation detection here is a convolution operation, and the vibration data is converted to the frequency domain to obtain a set of sequences. Each data in this set of sequences represents the energy size at different frequencies. When a dog barks and vibrates, it has its own unique energy spectrum distribution. In order to identify this unique energy spectrum, this set of data can be used to perform a convolution operation with a unique window sequence. Because the window sequence and the dog barking vibration data have a strong correlation, the final convolution result will be large, and the result will be small if the correlation is small. The convolution value is used to determine whether the dog is barking.

[0077] To sum up, the core of the AI model consists of two parts: one is model training, and the other is target recognition. Model training runs on a PC platform. The collected big data is used to train a high-performance model, which is then run on a target board. The target board platform extracts features from the data collected by the microphone, and the extracted data features are fed into the model for model prediction to obtain a probability result. This result is then compared with the preset threshold to obtain the final recognition result.

[0078] An AI bark stopper of the present invention also includes a human-computer interaction interface, which includes a button 10 and a liquid crystal display 2. The MCU 1 is connected to the liquid crystal display 2 and the button 10 respectively. The button 10 is used to set the device parameters of the AI bark stopper, such as setting the scoring parameters of the AI model, setting the intensity of the buzzer 7, the vibration motor 6, and the ZVC boost circuit 5. The LCD 2 is mainly used to display the power level and the set parameters in real time.

[0079] The AI bark suppressor of the present invention only triggers the buzzer, vibration and electric shock actions to stop the dog from barking when the AI recognizes the dog's barking and detects the vibration of the pet dog's throat. This is the biggest difference between the present invention and the existing technology. Compared with the existing technology, the AI bark suppressor of the present invention can more accurately identify the dog's barking, solving the defects of missed triggering and false triggering of the current bark suppressor.

[0080] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. An AI barking control device, characterized in that: It includes MCU (1), sound collector, vibration data collector, power module, and barking control module, wherein: The MCU (1) is used to receive signals from the sound collector and the vibration data collector, analyze and identify the data through a built-in AI model, determine whether barking control is required, and control the barking control module to perform corresponding barking control operations; The sound collector is used to collect the barking data of the pet dog and send the collected data to the MCU (1) in real time; The vibration data collector is used to collect vibration data of the pet dog's throat and send the collected data to the MCU (1) in real time; The power supply module is used to provide power to the AI barking control device; The barking prevention module triggers relevant intervention measures according to the instructions sent by the MCU (1), helping to reduce or stop the excessive barking of the pet dog.

2. The AI bark suppressor according to claim 1, characterized in that: The AI model is based on a recurrent neural network. The workflow of the AI model is as follows: Model training steps: The collected pet dog barking data samples are transmitted to the training platform. The training platform extracts the sample features and passes them to the recurrent neural network. The recurrent neural network is trained based on the data features and results, and the training results are placed in the weight parameter table. Target recognition step: The barking sounds of the pet dog collected by the sound collection module are sent to the recurrent neural network after feature extraction. The recurrent neural network uses the trained weight parameters to analyze and score. The scoring system will score the input sound and send the result. The final recognition result is obtained by comparing this result with the preset threshold.

3. The AI bark suppressor according to claim 2, characterized in that: The model training steps specifically include: Step 1, data collection: Build a database of audio samples containing various dog barking sounds. The data sources include user-provided data samples and public data samples crawled from the Internet. Step 2, data sorting: screening and segmenting the sample data of the audio sample database in step 1; Step 3, data labeling: label each audio segment at each moment as whether it is a dog barking; Step 4, data enhancement: Use noise addition, speed change, and pitch change methods to enhance the labeled data; Step 5: Model training: Build a dog barking detection model and determine the appropriate loss function for model training. Step 6, model testing: Test the trained model on a pre-recorded test set to screen out models with good performance.

4. The AI bark suppressor according to claim 2, characterized in that: The target identification step specifically includes: Step S1, feature extraction: performing frame and window processing on the sound data collected by the sound collector to extract logfbank features; Step S2, model prediction: the logfbank features extracted in step S1 are input into the model, and the last layer of the model uses a sigmoid function to output a probability value; Step S3, result judgment: compare the probability value of step S2 with a preset threshold value. If the probability value is greater than the threshold value, the system is judged to be activated; if the probability value is less than the threshold value, the system is judged to be inactivated.

5. The AI bark suppressor according to claim 2, characterized in that: The MCU (1) runs the following program: Step Y1: receiving vibration signal data of the pet dog's throat detected by the vibration data collector, first converting the data from the time domain to the frequency domain, and then detecting the energy spectrum of the data to identify the vibration of the pet dog's throat; Step Y2: Control the barking prevention module to stop the dog from barking based on the throat vibration of the dog identified in step Y1 and the barking characteristics of the dog determined in the target identification step.

6. The AI bark suppressor according to claim 5, characterized in that: The step Y1 further comprises: Step 1, data collection: collect pet dog throat vibration data; Step 2: Band-pass filtering: Use band-pass filtering to remove the interference signal generated by the pet dog's movement; Step 3, Fast Fourier Transform: Perform Fast Fourier Transform on the data filtered in Step 2 to transform the energy spectrum in the time domain into the frequency domain; Step 4, normalization processing: normalize the energy spectrum data in the frequency domain of step 3, and then use convolution operation to identify the vibration of the pet dog's throat.

7. The AI bark suppressor according to claim 1, characterized in that: The AI barking control device further comprises a human-computer interaction interface (2), and the MCU (1) is connected to the human-computer interaction interface (2).

8. The AI bark suppressor according to claim 1, characterized in that: The human-computer interaction interface comprises a button (10) and a liquid crystal display (2); the MCU (1) is connected to the liquid crystal display (2) and the button (10) respectively; the button (10) is used to set device parameters of the AI barking control device.

9. The AI bark suppressor according to claim 1, characterized in that: The barking prevention module comprises a ZVC boost circuit (5) and / or a vibration motor (6) and / or a buzzer (7), the sound collector is a microphone (3), and the vibration data collector is a gravity sensor or a six-axis gyroscope.

10. The AI bark suppressor according to claim 1, characterized in that: The power module includes a battery, a power management system (8) and / or a charging management system (9).

Images