Chewing energy efficiency evaluation system and method based on artificial intelligence

Abstract

The invention relates to the technical field of chewing energy efficiency evaluation, in particular to a chewing energy efficiency evaluation system and method based on artificial intelligence. The system comprises a data acquisition unit, a data processing unit and a model export unit. The chewing information of a subject is obtained by acquiring a chewing face image of the subject, environment sound during chewing and jaw vibration sound through the data acquisition unit, and the data acquired by the data acquisition unit is divided into a training set, a verification set and a test set through the data processing unit. A deep learning module is used for processing data in a training set to establish a chewing energy efficiency regression model for a model establishment module, and then an optimization module is used for importing a verification set and a test set into the chewing energy efficiency regression model to optimize and improve the chewing energy efficiency regression model, so that the precision and robustness of the chewing energy efficiency regression model are improved; and the accuracy of chewing energy efficiency evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of chewing energy efficiency evaluation, and in particular to a chewing energy efficiency evaluation system and method based on artificial intelligence. Background Art

[0002] The traditional way to evaluate chewing efficiency is to judge chewing ability through questionnaires or clinical manual observation. This self-reporting method is not only inaccurate, but also brings time and psychological burden to patients or participants. With the rapid development of science and technology, more and more hardware and software equipment are used in food intake and chewing tests. The traditional method of measuring chewing ability has gradually been replaced. For example, the literature (Farooq M, Sazonov E. Automatic measurement of chew count and chewing rate during food intake[J]. Electronics, 2016, 5(4):62.) proposed to observe food intake and monitor eating habits by automatically and quantitatively capturing chewing events through a piezoelectric sensor system. This method relies on chewing as an indicator of food intake and uses a micro microphone in the ear, an accelerometer, a monitoring camera and a piezoelectric sensor on the mandible to monitor the chewing process.

[0003] In addition, there are literatures (Bakke M, Tuxen A, Vilmann P, Jensen BR, Vilmann A, ToftM. Ultrasound image of human masseter muscle related to bite force, electromyography, facial morphology, and occlusal factors. Scand J Dent Res1992; 100: 164–171.) that evaluate chewing efficiency by investigating the tension of masticatory muscles. The specific methods include using real-time ultrasonic scanning to measure the thickness of masticatory muscles, or using electromyography to measure the contraction of masticatory muscles, and thus concluded that chewing ability is significantly correlated with the thickness of masticatory muscles.

[0004] Although the above two invasive methods measure data more accurately than traditional methods, their disadvantages are also obvious, including high cost, low acceptance of participants and potential health risks.

[0005] Subsequently, non-invasive monitoring gradually came into the public eye. Early researchers used in-ear microphones to capture chewing sounds and processed the audio to monitor diet. However, the sounds collected by the microphones were easily disturbed by environmental and background sounds. Therefore, subsequent studies tended to extract sounds based on bone conduction devices and process them as eating behavior data. For example, the literature (Fontana JM, Sazonov E S. A robust classification scheme for detection of offood intake through non-invasive monitoring of chewing[C] / / 2012 AnnualInternational Conference of the IEEE Engineering in Medicine and BiologySociety. IEEE, 2012: 4891-4894.) used a mandibular motion sensor with a bone conduction sound system to monitor chewing behavior, and then used SVM to analyze food intake, but did not evaluate chewing ability.

[0006] In addition, there is a paper (Jockusch J, Hahnel S, Nitschke I. Use of handgripstrength measurement as an alternative for assessing chewing function in people with dementia[J]. BMC geriatrics, 2022, 22(1): 769.) that proposes using handgrip strength measurement as an alternative for measuring chewing function. This method indirectly assesses chewing efficiency by measuring grip strength rather than directly measuring it, so the error of this method is relatively large.

[0007] With the rapid development of artificial intelligence technology, in order to solve the problem of the blank of artificial intelligence in chewing efficiency evaluation, the present invention proposes a chewing energy efficiency evaluation system and method based on artificial intelligence. Summary of the invention

[0008] The purpose of the present invention is to provide a chewing energy efficiency evaluation system and method based on artificial intelligence to solve the problems raised in the above background technology.

[0009] To achieve the above-mentioned object, one of the objects of the present invention is to provide a chewing energy efficiency evaluation system based on artificial intelligence, comprising a data acquisition unit, a data processing unit and a model derivation unit; The data acquisition unit captures basic information of the tester, a facial image of the tester chewing an object, and the sound of the tester chewing the object; The data processing unit processes the image data and audio data collected by the data collection unit, and differentiates the data after the processing, differentiates the data collected by the data collection unit into multiple parts, and then trains the differentiated multiple parts, and establishes a regression model according to the training results. After the regression model is established, the regression model is optimized with the help of the data of the differentiated multiple parts; The model exporting unit exports the model after the data processing unit optimizes the model.

[0010] As a further improvement of the technical solution, the data acquisition unit includes a basic data acquisition module, an image acquisition module and an audio acquisition module; The basic data acquisition module obtains basic information of multiple subjects and the physical health conditions of the subjects, and classifies subjects with different conditions so that subjects with the same health condition are divided into the same group; The image acquisition module acquires images of the front face and the side face of the subject by means of a camera; The audio collection module uses a microphone to collect the chewing sound of the subject and the sound of the bones when the subject chews an object, and pre-processes the collected sound.

[0011] As a further improvement of the technical solution, the data processing unit includes an audio processing module and an image processing module; The audio processing module uses the ESResNet network structure to process the audio data collected by the audio collection module to obtain clear audio data; The image processing module processes the image of the chewing action of the subject acquired by the image acquisition module to acquire clear jaw movement image data.

[0012] As a further improvement of the technical solution, the audio processing module adopts the ESResNet network structure including: ① Input layer: receives the audio signal preprocessed by the audio acquisition module and converts the one-dimensional audio data into a two-dimensional feature map using short-time Fourier transform; ②, Convolution layer: The initial convolution layer uses a smaller convolution kernel to extract features from the two-dimensional feature map, focusing on capturing the local frequency and time pattern of the audio signal to obtain clear audio frequencies; ③. Residual block: This step consists of multiple residual blocks, each of which contains two to three layers of convolution operations. This step is used to assist with batch normalization and activation functions; ④. Global pooling layer: At the back end of the network, the two-dimensional feature map is compressed in the time dimension through global average pooling, and the global feature vector is formed through the compressed two-dimensional feature map; ⑤. Fully connected layer: The global feature vector in step ④ is input into the fully connected layer.

[0013] As a further improvement of the technical solution, after obtaining clear jaw movement image data, the image processing module performs image extraction on the captured image information, obtains the subject's chewing action image, and establishes a chewing action model through the chewing action image.

[0014] As a further improvement of the technical solution, the data processing unit also includes a data differentiation module, a deep learning module and a model building module; The data differentiation module receives the audio and image data processed by the audio processing module and the image processing module, and differentiates the acquired data into a training set, a validation set and a test set according to the time axis, wherein the data acquired at the beginning, when the subject is chewing an object, is used as the training set, the data of the object chewed by the subject in the second half is used as the validation set, and the data of the object chewed by the subject at the end is used as the test set; The deep learning module receives the data of the training set differentiated by the data differentiation module, and trains the data in the training set by means of deep learning simulation; The model building module builds a chewing energy efficiency regression model according to the training results of the deep learning module on the training set.

[0015] As a further improvement of the technical solution, the steps of training the training set of the deep learning module are: ①. Receive image data and audio data in the training set; ② Comprehensively analyze the received data and extract the significant features of chewing behavior.

[0016] As a further improvement of the present technical solution, the data processing unit also includes an optimization module. After the model building module establishes the chewing energy efficiency regression model, the optimization module receives the verification set and the test set differentiated by the data differentiation module, and imports the audio data and image data obtained in the verification set and the test set into the chewing energy efficiency regression model to determine the accuracy of the established chewing energy efficiency regression model. If an error occurs, the chewing energy efficiency regression model is optimized with the help of the data in the verification set, and the test set is used to evaluate the generalization ability of the model after optimization.

[0017] The second object of the present invention is to provide a method for operating the above-mentioned artificial intelligence-based chewing energy efficiency evaluation system, comprising the following method steps: S1. Before the experiment begins, the basic data collection module obtains the basic information of the subjects; S2. At the beginning of the experiment, the subjects were asked to stand at a designated position. The experimenter fixed two microphones on the forehead and collar. The subjects were instructed to look at the front camera in a normal posture and chew gum and bread cubes according to their daily habits under the instructions of the experimenter. S3. In the data collection stage, the image acquisition module records the complete dynamic picture of the subject's chewing action, and the audio acquisition module records the jaw vibration and chewing audio signals in the environment; S4. After the collection is completed, the data processing unit divides all the data into training set, validation set and test set, and inputs them into the deep learning module for training. The model conducts a comprehensive analysis of the data collected by the data collection unit, learns and extracts the significant features in the chewing behavior, and establishes an accurate chewing energy efficiency regression model through the model building module. After the chewing energy efficiency regression model is established, the optimization module uses the data of the validation set to optimize the chewing energy efficiency regression model, and uses the test set to evaluate the generalization ability of the model.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. In the chewing energy efficiency evaluation system and method based on artificial intelligence, the data acquisition unit collects facial images of the subject chewing, the sound of the environment during chewing, and the sound of the jaw vibration to obtain the subject's chewing information, and the data processing unit divides the data collected by the data acquisition unit into a training set, a validation set, and a test set. The deep learning module is used to process the data in the training set to establish a chewing energy efficiency regression model in the model building module, and then the validation set and the test set are imported into the chewing energy efficiency regression model with the help of the optimization module to optimize and improve the chewing energy efficiency regression model, so as to improve the accuracy and robustness of the chewing energy efficiency regression model and improve the accuracy of the chewing energy efficiency evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is an overall block diagram of the present invention; Figure 2 This is a schematic diagram of collecting chewing data of subjects according to the present invention.

[0020] The meaning of each number in the figure is: 1. Data acquisition unit; 11. Basic data acquisition module; 12. Image acquisition module; 13. Audio acquisition module; 2. Data processing unit; 21. Audio processing module; 22. Image processing module; 23. Data differentiation module; 24. Deep learning module; 25. Model building module; 26. Optimization module; 3. Model export unit. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example

[0022] See also Figure 1-Figure 2 As shown, one of the purposes of this embodiment is to provide a chewing energy efficiency evaluation system based on artificial intelligence, including a data acquisition unit 1, a data processing unit 2 and a model export unit 3; The data collection unit 1 captures the basic information of the tester, the facial image of the tester chewing the object, and the sound of the tester chewing the object; The data acquisition unit 1 includes a basic data acquisition module 11, an image acquisition module 12 and an audio acquisition module 13; The basic data acquisition module 11 obtains basic information of multiple subjects, including age, gender, height, weight and eating habits, and obtains the physical health of the subjects, which mainly refers to the condition of the head, such as oral health, gum wear and jaw bite strength, etc. This information can provide personalized background data for model analysis, improve the adaptability and accuracy of the model, and classify subjects in different conditions so that subjects with the same health condition are divided into the same group; The image acquisition module 12 uses a camera to acquire images of the front face and side face of the subject. When acquiring images of the front face and side face of the subject, a front frame camera, a right side frame camera and an event camera are used respectively. The event camera captures the dynamic features of the front chewing action of the subject, and the frame camera obtains high-resolution frame information of the chewing action from the front and right sides respectively. The audio collection module 13 uses a microphone to collect the chewing sound of the subject and the sound of the bones when the subject chews an object, and pre-processes the collected sound; the chewing sound and the sound of the bones when the subject chews an object are collected using an air microphone and a bone conduction microphone, the air microphone is fixed at the collar position to capture the chewing sound in the environment, and the bone conduction microphone is fixed on the forehead of the subject to record the sound signal transmitted by the vibration of the jaw during the chewing process; At the beginning of the experiment, the subjects were asked to stand or sit in a designated position. The experimenter would fix the bone conduction microphone on the forehead and the air microphone on the collar. The subjects were instructed to look at the front-frame camera in a normal posture and chew gum and bread according to their daily habits under the instruction of the experimenter. In order to ensure the scientific nature of the experiment, the chewing time of each food was uniformly controlled by the experimenter through verbal instructions. During the data collection stage, the front frame camera and the right face frame camera recorded the complete dynamic picture of the subject's chewing action, the event camera captured the timing details of the front action in a high-frequency manner, and the bone conduction microphone and the air microphone recorded the jaw vibration and the chewing audio signal in the environment, respectively, to obtain the data set. The data set contains multimodal information of both chewing gum and bread cubes, providing rich sample data for subsequent modeling and analysis.

[0023] By collecting basic information and health conditions of different types of subjects, the chewing data of different types of people are obtained, so that the data processing unit 2 can obtain diversified data through the chewing data of different types of people, thereby improving the accuracy of subsequent evaluation of chewing energy efficiency.

[0024] The data processing unit 2 processes the image data and audio data collected by the data collection unit 1, and differentiates the data after processing, differentiates the data collected by the data collection unit 1 into multiple parts, and then trains the differentiated multiple parts, and establishes a regression model according to the training results. At the same time, after the regression model is established, the regression model is optimized with the help of the data of the differentiated multiple parts; The data processing unit 2 includes an audio processing module 21 and an image processing module 22; The audio processing module 21 uses the ESResNet network structure to process the audio data collected by the audio collection module 13 to obtain clear audio data; The audio processing module 21 adopts the ESResNet network structure including: ① Input layer: receives the audio signal preprocessed by the audio acquisition module 13, and uses short-time Fourier transform to convert the one-dimensional audio data into a two-dimensional feature map. These feature maps can effectively capture the time-frequency information of the audio signal and provide the basic input required by the network; ②, Convolution layer: The initial convolution layer uses a smaller convolution kernel such as 3×3 or 7×7 to extract features from the two-dimensional feature map, focusing on capturing the local frequency and time pattern of the audio signal and obtaining clear audio frequencies; the convolution operation gradually extracts higher-level features by increasing the channel depth while maintaining spatial resolution. The convolution layer is followed by batch normalization and activation function operations. The activation function can use ReLU, which helps to accelerate training and stabilize the network; ③. Residual block: This step consists of multiple residual blocks, each of which contains two to three layers of convolution operations. This step is used to assist with batch normalization and activation functions. The residual connection of the residual block enables the network to learn the residual mapping of features, significantly alleviating the gradient vanishing problem in deep networks and improving the training efficiency of the network. This modular design not only helps to learn features efficiently, but also allows deeper network structures without significantly increasing the difficulty of training; ④. Global pooling layer: At the back end of the network, the two-dimensional feature map is compressed in the time dimension through global average pooling, and the global feature vector is formed through the compressed two-dimensional feature map; this process converts the multi-dimensional feature map into a single global feature vector, retaining the most discriminative information, significantly reducing the number of parameters, and improving the generalization ability of the model; ⑤. Fully connected layer: The global feature vector in step ④ is input into the fully connected layer. The fully connected layer usually includes one or more layers of neurons for further nonlinear combination of the extracted features. The last layer is the output layer. According to the task requirements, it can be a classification head Softmax outputting multi-class probabilities or a regression head directly predicting continuous values. In the energy efficiency value prediction task of chewing audio signals, the output is a regression value, which represents the energy efficiency level of the audio signal; Through this network architecture, the chewing audio signal captured by the bone conduction microphone can be learned and analyzed by the ESResNet network after preprocessing and feature extraction. Compared with traditional methods, this network can automatically discover implicit patterns in the audio, providing a guarantee for the accuracy and robustness of the model.

[0025] The image processing module 22 processes the image of the subject's chewing action acquired by the image acquisition module 12 to acquire clear jaw movement image data; After obtaining clear jaw movement image data, the image processing module 22 performs image extraction on the captured image information to obtain the subject's chewing action image, and establishes a chewing action model through the chewing action image.

[0026] The data processing unit 2 also includes a data differentiation module 23, a deep learning module 24 and a model building module 25; The data differentiation module 23 receives the audio and image data processed by the audio processing module 21 and the image processing module 22, and differentiates the acquired data into a training set, a validation set and a test set according to the time axis. The data acquired at the beginning is used as the training set, when the subject is chewing the object, the data of the object chewed by the subject in the second half is used as the validation set, and the data of the object chewed by the subject at the end is used as the test set, that is, the data collected when a subject starts to chew the object is mostly the training set, while the validation set and the test set are small parts of the second half; The deep learning module 24 receives the data of the training set differentiated by the data differentiation module 23, and trains the data in the training set by means of deep learning simulation; The steps of deep learning module 24 training the training set are: ① Receive image data and audio data in the training set, including images taken by the frame camera and images taken by the event camera, and audio data including sound information collected by the bone conduction microphone and the air microphone; ② Comprehensively analyze the received data and extract the significant features of chewing behavior, including the sound of teeth biting objects, the sound of jaw vibration when teeth bite objects, the size of jaw opening and closing, and the positional relationship between the upper and lower jaws when the jaws bite; The model building module 25 builds a chewing energy efficiency regression model according to the training result of the deep learning module 24 on the training set; The data processing unit 2 also includes an optimization module 26. After the model building module 25 builds the chewing energy efficiency regression model, the optimization module 26 receives the verification set and the test set differentiated by the data differentiation module 23, and imports the audio data and image data obtained in the verification set and the test set into the chewing energy efficiency regression model to determine the accuracy of the established chewing energy efficiency regression model. If an error occurs, the chewing energy efficiency regression model is optimized with the help of the data in the verification set, and the test set is used to evaluate the generalization ability of the model after optimization to ensure that the model can accurately predict the chewing energy efficiency in unseen samples.

[0027] By collecting information from a certain number of subjects, the data collection unit 1 can obtain chewing data of different disease conditions. In the process of the data processing unit 2 processing the data collected by the data collection unit 1, different chewing energy efficiency regression models are established according to different diseases, so that the system can cope with patients with different diseases, thereby further improving the accuracy of the chewing energy efficiency regression model in evaluating chewing energy efficiency.

[0028] The model exporting unit 3 exports the model after the data processing unit 2 optimizes the model.

[0029] The facial image of the subject chewing, the sound of the environment during chewing, and the sound of the jaw vibration are collected by the data acquisition unit 1 to obtain the subject's chewing information, and the data processing unit 2 divides the data collected by the data acquisition unit 1 into a training set, a validation set, and a test set, and the deep learning module 24 is used to process the data in the training set so that the model building module 25 establishes a chewing energy efficiency regression model, and then the validation set and the test set are imported into the chewing energy efficiency regression model with the help of the optimization module 26 to optimize and improve the chewing energy efficiency regression model, so as to improve the accuracy and robustness of the chewing energy efficiency regression model and improve the accuracy of the chewing energy efficiency evaluation.

[0030] Once the chewing energy efficiency regression model is established, when the user evaluates the chewing energy efficiency of patients with the disease, he only needs to match the patient's condition with the patient's condition in the database to obtain a chewing energy efficiency regression model that matches the patient's condition. This improves the accuracy of the chewing energy efficiency evaluation of different patients, making it easier for doctors to provide corresponding treatment to patients based on the results of the chewing energy efficiency regression model.

[0031] The second object of the present invention is to provide a method for operating the above-mentioned chewing energy efficiency evaluation system based on artificial intelligence, comprising the following method steps: S1. Before the experiment begins, the basic data collection module 11 obtains basic information of the subject; It is necessary to collect the basic information of the subjects, including age, gender, height, weight and their current oral health status. This information can provide personalized background data for model analysis and improve the adaptability and accuracy of the model; S2. At the beginning of the experiment, the subjects were asked to stand at a designated position. The experimenter fixed two microphones on the forehead and collar. The subjects were instructed to look at the front camera in a normal posture and chew gum and bread cubes according to their daily habits under the instructions of the experimenter. The experimenter will fix the bone conduction microphone on the forehead and the air microphone on the collar. The subjects are instructed to look at the front camera in a normal posture; S3, in the data collection stage, the image collection module 12 records the complete dynamic picture of the subject's chewing action, and the audio collection module 13 records the jaw vibration and the chewing audio signal in the environment; The front and right face-frame cameras recorded the full dynamic picture of the subject's chewing action, the event camera captured the temporal details of the frontal action in a high-frequency manner, while the bone conduction microphone and the air microphone recorded the jaw vibration and the chewing audio signal in the environment, respectively; S4. After the collection is completed, the data processing unit 2 divides all the data into a training set, a validation set and a test set, and inputs them into the deep learning module 24 for training. The model conducts a comprehensive analysis of the data collected by the data collection unit 1, learns and extracts the significant features in the chewing behavior, and establishes an accurate chewing energy efficiency regression model through the model building module 25. After the chewing energy efficiency regression model is established, the optimization module 26 uses the data of the validation set to optimize the chewing energy efficiency regression model, and uses the test set to evaluate the generalization ability of the model.

[0032] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A chewing energy efficiency evaluation system based on artificial intelligence, characterized in that: It includes a data acquisition unit (1), a data processing unit (2) and a model export unit (3); The data acquisition unit (1) captures basic information of the tester, a facial image of the tester chewing an object, and the sound of the tester chewing the object; The data processing unit (2) processes the image data and audio data collected by the data collection unit (1), and differentiates the data after the processing, differentiates the data collected by the data collection unit (1) into multiple parts, and then trains the differentiated multiple parts, and establishes a regression model based on the training results. After the regression model is established, the regression model is optimized with the help of the data of the differentiated multiple parts; The model exporting unit (3) exports the model after the data processing unit (2) optimizes the model.

2. The chewing energy efficiency evaluation system based on artificial intelligence according to claim 1, characterized in that: The data acquisition unit (1) comprises a basic data acquisition module (11), an image acquisition module (12) and an audio acquisition module (13); The basic data acquisition module (11) acquires basic information of a plurality of subjects and obtains the physical health conditions of the subjects, and classifies subjects with different conditions so that subjects with the same health condition are grouped in the same group; The image acquisition module (12) acquires images of the front face and the side face of the subject by means of a camera; The audio collection module (13) uses a microphone to collect the chewing sound of the subject and the sound of the bones when the subject chews an object, and pre-processes the collected sound.

3. The chewing energy efficiency evaluation system based on artificial intelligence according to claim 2, characterized in that: The data processing unit (2) comprises an audio processing module (21) and an image processing module (22); The audio processing module (21) uses an ESResNet network structure to process the audio data collected by the audio collection module (13) to obtain clear audio data; The image processing module (22) processes the image of the chewing action of the subject acquired by the image acquisition module (12) to acquire clear jaw movement image data.

4. The chewing energy efficiency evaluation system based on artificial intelligence according to claim 3 is characterized in that: The audio processing module (21) adopts an ESResNet network structure including: ① Input layer: receiving the audio signal preprocessed by the audio acquisition module (13), and converting the one-dimensional audio data into a two-dimensional feature map using short-time Fourier transform; ② Convolutional layer: The initial convolutional layer uses a smaller convolution kernel to extract features from the two-dimensional feature map, focusing on capturing the local frequency and time pattern of the audio signal to obtain clear audio frequencies; ③. Residual block: This step consists of multiple residual blocks, each of which contains two to three layers of convolution operations. This step is used to assist with batch normalization and activation functions; ④. Global pooling layer: At the back end of the network, the two-dimensional feature map is compressed in the time dimension through global average pooling, and the global feature vector is formed through the compressed two-dimensional feature map; ⑤. Fully connected layer: The global feature vector in step ④ is input into the fully connected layer.

5. The chewing energy efficiency evaluation system based on artificial intelligence according to claim 4 is characterized in that: After obtaining clear jaw movement image data, the image processing module (22) performs image extraction on the captured image information to obtain a chewing action image of the subject, and establishes a chewing action model through the chewing action image.

6. The chewing energy efficiency evaluation system based on artificial intelligence according to claim 5, characterized in that: The data processing unit (2) further comprises a data differentiation module (23), a deep learning module (24) and a model building module (25); The data differentiation module (23) receives the audio and image data processed by the audio processing module (21) and the image processing module (22), and differentiates the acquired data into a training set, a validation set, and a test set according to a time axis, wherein the data acquired at the beginning, when the subject is chewing an object, is used as the training set, the data of the object chewed by the subject in the second half is used as the validation set, and the data of the object chewed by the subject at the end is used as the test set; The deep learning module (24) receives the data of the training set differentiated by the data differentiation module (23), and trains the data in the training set by means of deep learning simulation; The model building module (25) builds a chewing energy efficiency regression model based on the training results of the deep learning module (24) on the training set.

7. The chewing energy efficiency evaluation system based on artificial intelligence according to claim 6 is characterized in that: The steps of training the training set in the deep learning module (24) are as follows: ①. Receive image data and audio data in the training set; ② Comprehensively analyze the received data and extract the significant features of chewing behavior.

8. The chewing energy efficiency evaluation system based on artificial intelligence according to claim 6, characterized in that: The data processing unit (2) further comprises an optimization module (26). After the model building module (25) builds the chewing energy efficiency regression model, the optimization module (26) receives the validation set and the test set differentiated by the data differentiation module (23), and imports the audio data and the image data obtained in the validation set and the test set into the chewing energy efficiency regression model to determine the accuracy of the established chewing energy efficiency regression model. If an error occurs, the chewing energy efficiency regression model is optimized with the help of the data in the validation set, and after the optimization, the generalization ability of the model is evaluated using the test set.

9. A method for operating the chewing energy efficiency evaluation system based on artificial intelligence according to any one of claims 1 to 8, characterized in that: The method comprises the following steps: S1. Before the experiment begins, the basic data collection module (11) obtains the basic information of the subject; S2. At the beginning of the experiment, the subjects were asked to stand at a designated position. The experimenter fixed two microphones on the forehead and collar. The subjects were instructed to look at the front camera in a normal posture and chew gum and bread according to their daily habits under the instructions of the experimenter. S3, in the data collection stage, the image collection module (12) records the complete dynamic picture of the subject's chewing action, and the audio collection module (13) records the jaw vibration and the chewing audio signal in the environment; S4. After the collection is completed, the data processing unit (2) divides all the data into a training set, a validation set and a test set, and inputs them into the deep learning module (24) for training. The model comprehensively analyzes the data collected by the data collection unit (1), learns and extracts significant features in the chewing behavior, and establishes an accurate chewing energy efficiency regression model through the model building module (25). After the chewing energy efficiency regression model is established, the optimization module (26) uses the data of the validation set to optimize the chewing energy efficiency regression model, and uses the test set to evaluate the generalization ability of the model.

Images