Digital smoke generation method and system based on image recognition and deep learning

Through image recognition and deep learning technology, combining the matching of mouth shape and smoking flow, a realistic smoke effect is generated, which solves the problem of insufficient authenticity and interactivity of smoke effects in the existing technology, and achieves high realistic and diverse smoke. Experience.

CN119006670APending Publication Date: 2024-11-22WUHAN YANCHENG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411088267.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The prior art lacks authenticity and interaction in simulating smoke effects, cannot provide a realistic smoke experience, and the equipment is complex and costly.

Method used

Using a digital smoke generation method based on image recognition and deep learning, a realistic smoke effect is generated through the matching of mouth shape and smoking flow, and two modes are provided: one mode is precisely matched by mouth shape and smoke shape, and the other mode is Use deep learning model GANs to quickly generate smoke effects to adapt to changes in facial expressions of smokers.

Benefits of technology

It achieves a more realistic smoke dynamic effect, enhances the realism and immersion of simulated smoke, provides better interaction and entertainment, and meets the diverse needs of different smokers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006670A_ABST
    Figure CN119006670A_ABST
Patent Text Reader

Abstract

The invention discloses a digital smoke generation method based on image recognition and deep learning, and the method comprises the steps: firstly building a smoke effect library, classifying the mouth shape of a smoker, enabling the mouth shape to correspond to a simulated smoke shape, dividing the smoking flow into range segments, and enabling the range segments to correspond to a simulated smoke amount; a mouth shape image and a facial expression image of a smoker are shot through a camera, and the smoking flow is collected through a flow sensor; in the first mode, the mouth type category and the smoking flow are recognized, and a dynamic smoke effect picture is generated in real time and displayed; in the second mode, inputting the facial expression image into the trained GANs deep learning model, and outputting and displaying a corresponding simulated smoke effect picture; according to the invention, a more vivid simulation smoke dynamic effect is realized, and good interactivity and entertainment are achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of virtual reality technology, and in particular to a digital smoke generation method and system based on image recognition and deep learning. Background Art

[0002] Traditional simulated cigarette products mainly focus on simulating the appearance of cigarettes and the feel of smoking, but they are obviously insufficient in simulating the dynamic effects of smoke during real smoking. Such products often fail to provide smokers with a realistic smoke experience, thus affecting the realism and immersion of simulated smoke. In order to make up for this shortcoming, some products have appeared on the market that attempt to simulate smoke effects through physical means, such as using smoke generators to generate smoke-like gases. However, these methods usually have problems such as a single smoke effect, the inability to dynamically adjust the smoke effect according to the actual smoking behavior of smokers, and complex equipment and high costs.

[0003] In recent years, with the development of virtual reality technology, a new way has been provided for simulating the smoke effect of simulated cigarettes. For example, the patent with publication number CN114503060B discloses a smoking system and a control method thereof, in which a reproduction device is configured to reproduce sound and images in response to a user wearing the reproduction device sucking or ingesting a product, thereby providing a virtual smoking experience for the user.

[0004] However, although the existing technology has achieved certain results, there are still obvious defects in the authenticity and interactivity of simulated smoke images. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the prior art and provide a digital smoke generation method and system based on image recognition and deep learning to achieve a more realistic simulated smoke effect with better interactivity and entertainment.

[0006] Two modes are provided for smokers to choose from. When the first mode is selected, smokers can obtain a smoke effect that is closer to the real smoking experience. This mode can generate a more realistic smoke dynamic effect through the precise matching of the mouth shape and the simulated smoke shape, and the precise matching of the flow and the simulated smoke volume. When the second mode is selected, smokers can generate smoke effects more quickly. This mode only needs to capture facial expression images, without paying attention to the precise matching of mouth shape and smoking flow. The deep learning model GANs is used in conjunction with high-performance computing equipment to ensure that the time for generating simulated smoke effect images is very short, so that it can change in real time with the changes in the smoker's facial expression, with better interactivity and entertainment.

[0007] To achieve the above object, the technical solution adopted by the present invention is:

[0008] A digital smoke generation method based on image recognition and deep learning comprises: classifying the mouth shape of a smoker so that each mouth shape corresponds to a simulated smoke shape; dividing the smoking flow into a plurality of range segments, each flow range segment corresponds to a simulated smoke volume; establishing a smoke effect library according to each simulated smoke shape and simulated smoke volume; photographing the mouth shape image and facial expression image of the smoker when smoking simulated smoke by a camera, and collecting the smoking flow of the smoker by a simulated smoke flow sensor; when the smoker selects a first mode, identifying the mouth shape category of the smoker by image recognition, matching the corresponding simulated smoke shape in the smoke effect library, and matching the corresponding simulated smoke volume in the smoke effect library according to the range segment where the smoking flow of the smoker is located; generating a dynamic smoke effect picture of the smoker in real time according to the simulated smoke shape and the simulated smoke volume, and displaying it using a display terminal; when the smoker selects a second mode, inputting the facial expression image of the smoker when smoking simulated smoke into a trained deep learning model GANs, outputting the corresponding simulated smoke effect picture, and displaying it using a display terminal.

[0009] The method of classifying the smoker's mouth shape so that each mouth shape corresponds to a simulated smoke shape includes: classifying the smoker's mouth shape into a fully open mouth shape, a narrow mouth shape and an O-shaped mouth shape; setting the fully open mouth shape to correspond to a ball-shaped smoke shape, the narrow mouth shape to correspond to a thin strip smoke shape, and the O-shaped mouth shape to correspond to a ring-shaped smoke shape.

[0010] The smoking flow rate is divided into several range segments, each flow range segment corresponds to a simulated smoke volume, including: dividing the smoking flow rate range of 0-50 ml / s into a low flow range segment, the range of 50-150 ml / s into a medium flow range segment, and the range above 150 ml / s into a high flow range segment; setting the low flow range segment to correspond to the smoke volume displayed as a lighter color; setting the medium flow range segment to correspond to the smoke volume displayed as a moderate color; and setting the high flow range segment to correspond to the smoke volume with a darker color.

[0011] The method of identifying the smoker's mouth shape category by using image recognition includes: extracting the area containing the smoker's mouth shape from the captured image; preprocessing the extracted mouth shape image; inputting the preprocessed mouth shape image into a trained convolutional neural network CNN, and outputting the mouth shape category.

[0012] The method generates a dynamic smoke effect picture of a smoker in real time according to the simulated smoke shape and the simulated smoke volume, and displays it using a display terminal, including: calling the corresponding simulated smoke shape and simulated smoke volume from a preset smoke effect resource library; using three-dimensional modeling software to create a basic model of smoke, and adjusting the shape, size and transparency of the model according to the different smoke shapes and smoke volumes; designing the dynamic behavior of the smoke, and using animation software to add animation effects to the smoke model; and transmitting the dynamic smoke effect to the display terminal in real time for display.

[0013] A simulated burning lamp is arranged at the end of the simulated cigarette, and the light intensity of the simulated burning lamp changes with the flow rate detected by the flow sensor, simulating the lighting effect of a burning cigarette.

[0014] The artificial cigarette mouthpiece is filled with substances with different flavors and tastes that satisfy sensory stimulation, adding a new sense of experience. Each flavor and taste corresponds to a unique identification code, and each flavor and taste corresponds to a simulated smoke color. The identification code of the artificial cigarette flavor and taste is read through the Bluetooth module of the artificial cigarette, and the corresponding simulated smoke color can be matched in the smoke effect library.

[0015] The training process of the deep learning model GANs includes: collecting real smoke images containing various smoke shapes, densities, and motion states, collecting different facial expression images of smokers when smoking simulated smoke, and pairing the real smoke images with the corresponding facial expression images of smokers; using random noise vectors and the facial expression images of smokers as inputs of the generator, and the generator outputs a simulated smoke image corresponding to the input facial expression image; using the simulated smoke image or real smoke image generated by the generator and the corresponding facial expression image of the smoker as inputs of the discriminator, and the discriminator outputs a probability value between 0 and 1, indicating the confidence that the input image is real smoke; setting the generator loss function to encourage it to generate simulated smoke images that can deceive the discriminator while maintaining the correlation with the input facial expression image; alternately training the generator and the discriminator, and judging that the model has reached a convergence state by observing the quality of the generated simulated smoke images.

[0016] A digital smoke generation system based on image recognition and deep learning comprises: simulated smoke, including a flow sensor for detecting the smoking flow of a smoker; a camera for photographing a mouth shape image and a facial expression image of a smoker when smoking simulated smoke; a mouth shape classification unit for classifying the mouth shape of the smoker; a flow division unit for dividing the smoking flow into a plurality of range segments; a smoke effect library establishment unit for establishing a smoke effect library according to the classified mouth shapes and the divided flow ranges, wherein the smoke effect library contains various simulated smoke shapes and simulated smoke volumes; a mode selection module for providing a user interface for selecting a first mode or a second mode; a mouth shape recognition and flow matching unit for recognizing the mouth shape of the smoker by image recognition technology and matching the corresponding simulated smoke shape in the smoke effect library; and at the same time, matching the corresponding simulated smoke volume according to the smoking flow; a deep learning model GANs unit for receiving a facial expression image as input in the second mode and outputting a corresponding simulated smoke effect picture; and a display terminal for displaying the generated dynamic smoke effect picture in real time.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] By matching the smoker's mouth shape and smoking flow with the preset simulated smoke shape and simulated smoke volume, a high degree of simulation of the dynamic effect of smoke in the real smoking process is achieved, which enhances the realism and immersion of using simulated cigarettes, satisfies the sensory stimulation brought by smoking, and plays a role in smoking control;

[0019] Two modes are provided for smokers to choose from. When the first mode is selected, smokers can obtain a smoke effect that is closer to the real smoking experience. This mode can generate a more realistic smoke dynamic effect through the precise matching of the mouth shape and the simulated smoke shape, and the precise matching of the flow and the simulated smoke volume. When the second mode is selected, smokers can generate smoke effects more quickly. This mode only needs to capture facial expression images, without paying attention to the precise matching of mouth shape and smoking flow. The deep learning model GANs is used in conjunction with high-performance computing equipment to ensure that the time for generating simulated smoke effect images is very short, so that it can change in real time with the changes in the smoker's facial expression, with better interactivity and entertainment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 This is a schematic diagram of a simulated smoke and a display terminal in an embodiment of the present application;

[0022] In the figure: 1. Harmless simulated smoke; 2. Power supply; 3. Wireless transmission module; 4. Flow sensor; 5. Airflow sensor; 6. Simulated combustion lamp; 7. Terminal display. DETAILED DESCRIPTION

[0023] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings 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.

[0024] In a first aspect of the present application, a method for generating digital smoke based on image recognition and deep learning is provided, comprising the following steps:

[0025] S1. Classify the mouth shapes of smokers so that each mouth shape corresponds to a simulated smoke shape; divide the smoking flow rate into several range segments, each flow rate range segment corresponds to a simulated smoke volume; establish a smoke effect library according to each simulated smoke shape and simulated smoke volume;

[0026] S2, using a camera to capture the mouth shape and facial expression of the smoker when he / she smokes the simulated cigarette, and using a flow sensor of the simulated cigarette to collect the smoker's smoking flow;

[0027] S3. When the smoker selects the first mode, the mouth shape of the smoker is identified by image recognition, and the corresponding simulated smoke shape is matched in the smoke effect library. According to the range of the smoker's smoking flow rate, the corresponding simulated smoke volume is matched in the smoke effect library;

[0028] S4, generating a dynamic smoke effect picture of a smoker in real time according to the simulated smoke shape and the simulated smoke volume, and displaying it using the display terminal 7;

[0029] S5. When the smoker selects the second mode, the facial expression image of the smoker when inhaling the simulated smoke is input into the trained deep learning model GANs, and the corresponding simulated smoke effect picture is output and displayed using the display terminal 7.

[0030] In this embodiment, by matching the smoker's mouth shape and smoking flow with the preset simulated smoke shape and simulated smoke volume, a high degree of simulation of the dynamic effects of smoke in the actual smoking process is achieved, thereby enhancing the realism and immersion of the use of simulated smoke, satisfying the sensory stimulation brought by smoking, and thus playing a role in smoking control.

[0031] Two modes are provided for smokers to choose from. When the first mode is selected, smokers can obtain a smoke effect that is closer to the real smoking experience. This mode can generate a more realistic smoke dynamic effect through the precise matching of the mouth shape and the simulated smoke shape, and the precise matching of the flow and the simulated smoke volume. When the second mode is selected, smokers can generate smoke effects more quickly. This mode only needs to capture facial expression images, without paying attention to the precise matching of mouth shape and smoking flow. The deep learning model GANs is used in conjunction with high-performance computing equipment to ensure that the time for generating simulated smoke effect images is very short, so that it can change in real time with the changes in the smoker's facial expression, with better interactivity and entertainment.

[0032] In some embodiments, the smoker's mouth shape is classified so that each mouth shape corresponds to a simulated smoke shape, including: dividing the smoker's mouth shape into a fully open mouth shape, a narrow mouth shape and an O-shaped mouth shape; setting the fully open mouth shape to correspond to a ball-shaped smoke shape, the narrow mouth shape to correspond to a thin strip smoke shape, and the O-shaped mouth shape to correspond to a ring-shaped smoke shape.

[0033] Through the present embodiment, the system dynamically adjusts the shape of the simulated smoke according to the actual mouth shape of the smoker, achieving a more realistic and personalized smoke effect, improving the realism of using simulated cigarettes, and meeting the diverse needs of different smokers.

[0034] In some embodiments, the smoking flow rate is divided into several range segments, each flow range segment corresponds to a simulated smoke volume, including: dividing the smoking flow rate range of 0-50 ml / s into a low flow range segment, the range of 50-150 ml / s into a medium flow range segment, and the range above 150 ml / s into a high flow range segment; setting the low flow range segment to correspond to the amount of smoke displayed as a lighter color; setting the medium flow range segment to correspond to the amount of smoke displayed as a moderate color; and setting the high flow range segment to correspond to the amount of smoke displayed as a darker color.

[0035] Through this embodiment, the system can dynamically adjust the color and concentration of the simulated smoke according to the actual smoking flow of the smoker, thereby achieving a more realistic smoke effect, improving the realism of the use of simulated smoke, and meeting the diverse needs of different smokers. At the same time, smokers can perceive their smoking intensity by observing the color and concentration of the smoke, so as to better control their smoking habits in a targeted manner.

[0036] In some embodiments, the method of identifying the smoker's mouth shape category by using image recognition includes: extracting the area containing the smoker's mouth shape from the captured image; preprocessing the extracted mouth shape image; inputting the preprocessed mouth shape image into a trained convolutional neural network CNN, and outputting the mouth shape category.

[0037] Specifically, firstly, an image containing the smoker's face is captured, and the area containing the smoker's mouth shape is extracted using image segmentation technology; the extracted mouth shape image is preprocessed, such as grayscale, normalization, filtering, etc., to improve image quality and reduce noise; the preprocessed mouth shape image is input into a trained convolutional neural network CNN. CNN is particularly suitable for image recognition tasks. Through training, it can be identified that the smoker's mouth shape belongs to which preset category, such as fully open mouth, narrow mouth, O-shaped mouth, etc.

[0038] In some embodiments, the real-time generation of a dynamic smoke effect image of a smoker based on a simulated smoke shape and a simulated smoke volume, and displaying it using a display terminal, includes: calling the corresponding simulated smoke shape and simulated smoke volume from a preset smoke effect resource library; using three-dimensional modeling software to create a basic smoke model, and based on the known smoke shape and smoke volume, modifying the parameters of the basic smoke model to make it conform to a predetermined shape, size and transparency; designing the dynamic behavior of the smoke, and using animation software to add animation effects to the smoke model, and the animation effects may include smoke drifting, diffusion, rising and other actions; transmitting the dynamic smoke effect to the display terminal 7 in real time for display.

[0039] Through this embodiment, the system modifies the basic smoke model according to the simulated smoke shape and simulated smoke volume, realizes the generation of simulated smoke images, and displays dynamic smoke effect pictures, providing smokers with a more realistic and personalized smoking experience.

[0040] In some embodiments, Figure 1 As shown, the simulated cigarette includes: a cigarette rod 1, a power source 2, a wireless transmission module 3, a flow sensor 4, and an airflow sensor 5;

[0041] In addition, a simulated burning lamp 6 is provided at the end of the simulated cigarette, and the light intensity of the simulated burning lamp changes with the flow rate detected by the flow sensor, simulating the lighting effect of a burning cigarette.

[0042] When the smoker starts to smoke, the flow sensor transmits the detected flow information to the control system of the simulated smoke, and the simulated smoke control system will intelligently adjust the light intensity of the simulated burning lamp according to the received flow data. If the smoker inhales a large amount of air flow, the light intensity of the simulated burning lamp will be increased accordingly to simulate a brighter flame when the cigarette is burning; conversely, if the inhaled air flow is small, the light intensity will be reduced accordingly to simulate a weaker flame when the cigarette is burning.

[0043] Through this design, the simulated cigarette can adjust the lighting effect at its end in real time according to the smoker's behavior, thereby more realistically simulating the burning process of a real cigarette and improving the user experience of the simulated cigarette.

[0044] In some embodiments, the mouthpiece of the simulated cigarette is filled with substances with different flavors that satisfy sensory stimulation, each flavor corresponds to a unique identification code, and each flavor corresponds to a simulated smoke color. The identification code of the simulated smoke flavor is read through the Bluetooth module of the simulated cigarette, and the corresponding simulated smoke color can be matched in the smoke effect library.

[0045] Specifically, different flavors such as mint, apple, strawberry, lemon, etc., these flavor substances are guaranteed to be harmless to the human body and can add a new sense of experience; the identification code is encoded on the internal chip of the simulated smoke and can be read through the Bluetooth module. When connected to a paired device, it can automatically or manually trigger the reading of the flavor identification code. After receiving the identification code, the system searches and matches the corresponding simulated smoke color in the smoke effect library. According to the matched smoke color, the system generates and displays a dynamic smoke effect screen with the corresponding color in real time.

[0046] In some embodiments, the training process of the deep learning model GANs includes: collecting real smoke images containing various smoke shapes, densities, and motion states, collecting different facial expression images of smokers when smoking simulated smoke, and pairing real smoke images with corresponding facial expression images of smokers; using random noise vectors and facial expression images of smokers as inputs of the generator, and the generator outputs simulated smoke images corresponding to the input facial expression images; using the simulated smoke image or real smoke image generated by the generator, and the corresponding facial expression image of the smoker as inputs of the discriminator, and the discriminator outputs a probability value between 0 and 1, indicating the confidence that the input image is real smoke; setting the generator loss function to encourage it to generate simulated smoke images that can deceive the discriminator while maintaining the correlation with the input facial expression image; alternately training the generator and the discriminator, and judging that the model has reached a convergence state by observing the quality of the generated simulated smoke images. Through this training, the GANs model can learn the complex relationship between real smoke images and smokers' facial expressions, and generate simulated smoke images that are highly correlated with facial expressions.

[0047] The second aspect of the present application provides a digital smoke generation system based on image recognition and deep learning, comprising:

[0048] The simulated smoke includes a flow sensor for detecting the smoke flow of the smoker;

[0049] A camera, used for capturing images of the smoker's mouth shape and facial expression when the smoker inhales the simulated cigarette;

[0050] Mouth shape classification unit, used to classify the mouth shape of smokers;

[0051] A flow division unit, used for dividing the smoking flow into a plurality of range segments;

[0052] The smoke effect library establishment unit establishes a smoke effect library according to the classified mouth shapes and divided flow ranges, which contains various simulated smoke shapes and simulated smoke volumes;

[0053] A mode selection module, used for providing a user interface to select a first mode or a second mode;

[0054] The mouth shape recognition and flow matching unit recognizes the smoker's mouth shape through image recognition technology, and matches the corresponding simulated smoke shape in the smoke effect library; at the same time, it matches the corresponding simulated smoke volume according to the smoking flow;

[0055] The deep learning model GANs unit, in the second mode, receives the facial expression image as input and outputs the corresponding simulated smoke effect picture;

[0056] The display terminal 7 is used to display the generated dynamic smoke effect picture in real time.

[0057] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0058] In the description of the present specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish descriptions and cannot be understood as indicating or suggesting relative importance. It should also be understood that although the terms "first", "second", etc. are used to describe various elements in some embodiments of the present application in the text, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, the first table can be named as the second table, and similarly, the second table can be named as the first table without departing from the scope of the various described embodiments. The first table and the second table are both tables, but they are not the same table.

[0059] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0060] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A digital smoke generation method based on image recognition and deep learning, characterized in that: include: Classify the mouth shapes of smokers so that each mouth shape corresponds to a simulated smoke shape; divide the smoking flow rate into several range segments, each flow range segment corresponds to a simulated smoke volume; establish a smoke effect library based on each simulated smoke shape and simulated smoke volume; The camera is used to capture the mouth shape and facial expression of the smoker when he / she smokes the simulated cigarette, and the flow rate of the smoker is collected through the simulated smoke flow sensor; When the smoker selects the first mode, the mouth type of the smoker is identified by image recognition, and the corresponding simulated smoke shape is matched in the smoke effect library. According to the range of the smoker's smoking flow, the corresponding simulated smoke volume is matched in the smoke effect library; According to the simulated smoke shape and simulated smoke volume, a dynamic smoke effect picture of the smoker is generated in real time and displayed using a display terminal; When the smoker selects the second mode, the facial expression image of the smoker when inhaling the simulated smoke is input into the trained deep learning model GANs, and the corresponding simulated smoke effect picture is output and displayed using the display terminal.

2. The digital smoke generation method based on image recognition and deep learning according to claim 1, characterized in that: The method of classifying the mouth shapes of smokers so that each mouth shape corresponds to a simulated smoke shape includes: The smoker's mouth type is divided into fully open mouth type, narrow mouth type and O-shaped mouth type; The fully open mouth corresponds to the ball-shaped smoke, the narrow mouth corresponds to the thin strip smoke, and the O-shaped mouth corresponds to the circle-shaped smoke.

3. The digital smoke generation method based on image recognition and deep learning according to claim 1, characterized in that: The smoking flow rate is divided into several range segments, each flow range segment corresponds to a simulated smoke volume, including: Divide the smoking flow rate into a low flow range segment, a medium flow range segment and a high flow range segment; The low flow range corresponds to the lighter smoke volume; the medium flow range corresponds to the medium smoke volume; the high flow range corresponds to the darker smoke volume.

4. The digital smoke generation method based on image recognition and deep learning according to claim 1, characterized in that: The method of identifying the mouth shape of the smoker by image recognition includes: extracting a region containing the smoker's mouth shape from the captured image; Preprocessing the extracted mouth shape image; The preprocessed mouth shape image is input into the trained convolutional neural network CNN to output the mouth shape category.

5. The digital smoke generation method based on image recognition and deep learning according to claim 1, characterized in that: The method of generating a dynamic smoke effect picture of a smoker in real time according to the simulated smoke shape and the simulated smoke volume, and displaying it using a display terminal, comprises: Call the corresponding simulated smoke shape and simulated smoke volume from the preset smoke effect resource library; Use 3D modeling software to create a basic model of smoke, and adjust the shape, size and transparency of the model according to the shape and amount of smoke; Design the dynamic behavior of smoke and use animation software to add animation effects to the smoke model; The dynamic smoke effect is transmitted to the display terminal in real time for display.

6. The digital smoke generation method based on image recognition and deep learning according to claim 1, characterized in that: A simulated burning lamp is arranged at the end of the simulated cigarette, and the light intensity of the simulated burning lamp changes with the flow rate detected by the flow sensor, simulating the lighting effect of a burning cigarette.

7. The digital smoke generation method based on image recognition and deep learning according to claim 1, characterized in that: The training process of the deep learning model GANs includes: Collect real smoke images with various smoke shapes, densities, and motion states, collect different facial expression images of smokers when they smoke simulated smoke, and pair the real smoke images with the corresponding facial expression images of smokers; The random noise vector and the facial expression image of the smoker are taken as inputs of the generator, and the generator outputs a simulated smoke image corresponding to the input facial expression image; The simulated smoke image or real smoke image generated by the generator and the corresponding smoker's facial expression image are used as the input of the discriminator, and the discriminator outputs a probability value between 0 and 1, indicating the confidence that the input image is real smoke; Set the generator loss function to encourage it to generate simulated smoke images that can deceive the discriminator while maintaining relevance to the input facial expression image; The generator and discriminator are trained alternately, and the model reaches convergence by observing the quality of the generated simulated smoke images.

8. The digital smoke generation method based on image recognition and deep learning according to claim 1, characterized in that: The artificial cigarette mouthpiece is filled with substances with different flavors and tastes that satisfy sensory stimulation, adding a new sense of experience. Each flavor and taste corresponds to a unique identification code, and each flavor and taste corresponds to a simulated smoke color. The identification code of the artificial cigarette flavor and taste is read through the Bluetooth module of the artificial cigarette, and the corresponding simulated smoke color can be matched in the smoke effect library.

9. A digital smoke generation system based on image recognition and deep learning, characterized in that: include: The simulated smoke includes a flow sensor for detecting the smoke flow of the smoker; A camera, used for capturing images of the smoker's mouth shape and facial expression when the smoker inhales the simulated cigarette; Mouth shape classification unit, used to classify the mouth shape of smokers; A flow division unit, used for dividing the smoking flow into a plurality of range segments; The smoke effect library establishment unit establishes a smoke effect library according to the classified mouth shapes and divided flow ranges, which contains various simulated smoke shapes and simulated smoke volumes; A mode selection module, used for providing a user interface to select a first mode or a second mode; The mouth shape recognition and flow matching unit recognizes the smoker's mouth shape through image recognition technology, and matches the corresponding simulated smoke shape in the smoke effect library; at the same time, it matches the corresponding simulated smoke volume according to the smoking flow; The deep learning model GANs unit, in the second mode, receives the facial expression image as input and outputs the corresponding simulated smoke effect picture; The display terminal is used to display the generated dynamic smoke effect picture in real time.