Smoking monitoring and early warning method and system based on AI identification

Abstract

The invention discloses a smoking monitoring and early warning method and system based on AI identification, and relates to the technical field of intelligent safety monitoring, and the method comprises the steps: collecting intelligent sensor data, and carrying out the preprocessing; constructing a Smoke-Unet model to perform image segmentation, calculating a probability value of a smoke region through a Sigmoid activation function to obtain a high-confidence smoke region, determining a smoking candidate region in combination with the comprehensive confidence of temperature and sound wave reflection signals, and predicting a smoke diffusion trend based on a TCNN model; fuzzy logic is used for dividing smoking threat levels, dynamic voice combination is used for voice broadcasting, and rectification period data are monitored in real time; the sensor data generated by collection and analysis is stored. According to the method, data collected by the intelligent sensor and image segmentation of the Smoke-Unet model are combined, the accuracy of smoking behavior detection is improved, and the response speed and early warning accuracy of a monitoring system are improved through combination of the smoking candidate area and smoke diffusion prediction of the TCNN model.

Description

Technical Field

[0001] The present invention relates to the field of intelligent safety monitoring technology, and in particular to a smoking monitoring and early warning method and system based on AI recognition. Background Art

[0002] With the rapid development of artificial intelligence and sensor technology, intelligent monitoring systems have gradually penetrated into various industries, especially in the fields of public safety, environmental monitoring and health. In terms of smoking monitoring, traditional methods mainly rely on smoke detectors, cameras and manual inspections to identify and monitor smoking behavior. Computer vision technology and deep learning models are constantly improving, which can improve the efficiency and accuracy of monitoring, making smoking monitoring systems more intelligent and comprehensive, avoiding the limitations of traditional methods.

[0003] However, existing technologies can only make judgments through a single smoke sensor, making it difficult to distinguish the source of smoke and other environmental factors. Existing methods for predicting smoke diffusion trends lack the laws of smoke propagation in different environments and lack real-time emergency response capabilities. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a smoking monitoring and early warning method and system based on AI recognition, which solves the problem that the above-mentioned existing technology can only make judgments through a single smoke sensor, making it difficult to distinguish the source of smoke and other environmental factors. The existing method of predicting the smoke diffusion trend lacks the propagation rules of smoke in different environments and has no real-time emergency response capabilities.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a smoking monitoring and early warning method based on AI recognition, which includes collecting intelligent sensor data and performing preprocessing; constructing a Smoke-Unet model for image segmentation, calculating the probability value of the smoke area through the Sigmoid activation function, obtaining a high-confidence smoke area, combining the comprehensive confidence of temperature and sound wave reflection signals to determine the smoking candidate area, and predicting the smoke diffusion trend based on the TCNN model; using fuzzy logic to divide the smoking threat level, using dynamic voice combination for voice broadcasting, and real-time monitoring of rectification period data; storing, collecting and analyzing the generated sensor data.

[0008] As a preferred solution of the smoking monitoring and early warning method based on AI recognition described in the present invention, wherein: collecting smart sensor data and preprocessing it refers to installing smart sensors in the monitoring area to collect sensor data and preprocess it;

[0009] The intelligent sensors include high-definition cameras, smoke sensors, infrared thermal imagers, wind speed and direction sensors, and acoustic wave sensors;

[0010] The sensor data includes RGB images, smoke concentration, temperature map, wind speed and direction, and sound wave reflection signal data;

[0011] The preprocessing includes using linear interpolation to time align the sensor data, using wavelet transform to enhance the RGB image, and using a linear normalization method to normalize the smart sensor data.

[0012] As a preferred embodiment of the AI-based smoking monitoring and early warning method of the present invention, the method comprises: constructing a Smoke-Unet model for image segmentation, calculating the probability value of the smoke area using a Sigmoid activation function to obtain a high-confidence smoke area, and combining the comprehensive confidence of the temperature and the acoustic reflection signal to determine the candidate smoking area, which comprises extracting the preprocessed RGB image, the temperature map, and the acoustic reflection signal;

[0013] The time domain signal of the acoustic wave reflection signal is converted into a frequency domain signal by short-time Fourier transform to generate a spectrum diagram;

[0014] Use channel stacking to combine the RGB image, temperature map, and spectrum map to generate a five-channel tensor;

[0015] Build the Smoke-Unet model, including the input layer, encoder part, bottleneck layer, decoder part and output layer;

[0016] Set the input layer to a five-channel tensor;

[0017] Through the encoder part, the convolution formula, ReLU activation function and maximum pooling formula are used to extract features from the image;

[0018] Feature compression is used in the bottleneck layer to generate feature maps. The decoder part uses the deconvolution formula and jump connection to restore the spatial resolution of the compressed image to the size before input.

[0019] In the output layer, convolution operation and Sigmoid activation function are used to calculate the probability value of each pixel in the image belonging to the smoke area and non-smoke area, and a binary segmentation map is generated by fixed threshold method to obtain the smoke area;

[0020] The five-channel tensor is input into the Smoke-Unet model for model training, and the model parameters are optimized using the loss function and Adam optimizer;

[0021] The RGB image, temperature map, and acoustic reflection signal collected in real time by the smart sensor are converted into a five-channel tensor and input into the Smoke-Unet model to obtain the probability value of each pixel in the image belonging to the smoke area and non-smoke area. A binary segmentation map is generated using a fixed threshold method to obtain the smoke area.

[0022] In the smoke area, the probability threshold is set by the maximum entropy threshold method, the probability value is compared with the probability threshold, and the probability value greater than the probability threshold is screened to obtain the high-confidence smoke area;

[0023] Extract the corresponding spectrum in the high-confidence smoke area and use peak extraction to obtain the intensity of the sound wave reflection signal;

[0024] In high-confidence smoke areas, the thresholds of temperature and acoustic wave reflection signal intensity of each pixel are set separately through the empirical threshold method, and the confidence of temperature and acoustic wave reflection intensity of each pixel is obtained by binarization.

[0025] By setting the classification threshold based on statistical analysis, the probability value, temperature confidence, and spectrum confidence of each pixel are weighted and fused to obtain the fused confidence. The fused confidence is then compared with the classification threshold, and the pixels corresponding to the fused confidence that is higher than the classification threshold are integrated into the smoking candidate area.

[0026] Perform minimum bounding rectangle processing on the candidate smoking area and output the coordinates of the candidate smoking area.

[0027] As a preferred embodiment of the AI-based smoking monitoring and early warning method of the present invention, the smoke diffusion trend prediction based on the TCNN model comprises integrating the coordinates of the candidate smoking area and the wind speed and direction to generate a data set;

[0028] Set up the TCNN model, including the input layer, convolutional layer, fully connected layer, and output layer;

[0029] The input format of the input layer is set to the dataset. The convolution layer uses the ReLU activation function to extract the smoke diffusion features at different times and outputs the smoke concentration field through the fully connected layer.

[0030] Use the dataset to train the TCNN model and use the Adam optimizer and loss function to optimize the model parameters;

[0031] The coordinates and wind speed and direction of the newly generated smoking candidate area are integrated to generate a new dataset, which is input into the trained TCNN model to predict the smoke concentration field within time t.

[0032] Use linear regression to calculate the smoke diffusion speed, sort the wind speed and direction data in chronological order, generate a trend chart, and use time series analysis methods to predict the direction of smoke diffusion;

[0033] Combine the smoking candidate area with the smoke diffusion area and output the corresponding smoking point coordinates.

[0034] As a preferred embodiment of the AI-based smoking monitoring and early warning method of the present invention, the method of using fuzzy logic to classify the smoking threat level refers to fuzzifying the smoke diffusion speed and smoke concentration through a membership function;

[0035] Establish fuzzy rules and use Mamdani fuzzy reasoning method to calculate the triggering strength of each rule;

[0036] The trigger strength of all rules and the corresponding threat level center value are weighted averaged, and the result is defuzzified using the center of gravity method to obtain the threat level value.

[0037] As a preferred embodiment of the AI-based smoking monitoring and early warning method of the present invention, the use of a dynamic voice combination for voice broadcasting and real-time monitoring of rectification period data refer to setting a monitoring threshold using empirical rules, comparing the threat level value with the monitoring threshold, and determining the intervention level:

[0038] If the threat level exceeds the monitoring threshold, primary intervention is required. A personalized warning audio is generated through dynamic voice combination, and content about smoking bans and smoking hazards is played. The system then enters a time-limited rectification monitoring phase. If smoking behavior persists during the rectification phase, advanced intervention is implemented. The coordinates of the smoking point, smoke distribution map, and predicted smoke diffusion map are transmitted to the monitoring center in real time. Intercom equipment is used to warn smokers, and staff are notified to stop them on the spot.

[0039] The storing of the sensor data collected and analyzed refers to storing the collected sensor data and the monitoring results generated by the analysis in a database according to timestamps, setting security access measures, and regularly performing integrity checks on the data in the database.

[0040] In a second aspect, the present invention provides a smoking monitoring and early warning system based on AI recognition, comprising:

[0041] Collection and processing module, used to collect smart sensor data and perform preprocessing;

[0042] The monitoring and evaluation module is used to construct a Smoke-Unet model for image segmentation. The Sigmoid activation function is used to calculate the probability value of the smoke area to obtain the high-confidence smoke area. The combined confidence of temperature and acoustic reflection signals is combined to determine the candidate smoking area. The smoke diffusion trend is predicted based on the TCNN model.

[0043] The early warning and intervention module uses fuzzy logic to classify smoking threat levels, uses dynamic voice combinations for voice broadcasting, and monitors rectification period data in real time;

[0044] The storage management module is used to store the sensor data collected and analyzed.

[0045] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the smoking monitoring and early warning method based on AI recognition as described in the first aspect of the present invention is implemented.

[0046] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the smoking monitoring and early warning method based on AI recognition as described in the first aspect of the present invention is implemented.

[0047] As a preferred solution of the computer device of the present invention, it further includes a timing module and a detection module. The detection module is a detection telescopic rod, and an image recognition device and an infrared temperature detector are provided at the end of the detection telescopic rod.

[0048] The beneficial effects of the present invention are as follows: the present invention improves the accuracy of detecting smoking behavior by combining data collected by intelligent sensors with image segmentation of the Smoke-Unet model, and improves the response speed and warning accuracy of the monitoring system by combining the smoking candidate area with the smoke diffusion prediction of the TCNN model. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 any creative work.

[0050] Figure 1 This is a flow chart of the smoking monitoring and early warning method based on AI recognition in Example 1.

[0051] Figure 2 This is a schematic diagram of the smoking monitoring and early warning system based on AI recognition in Example 1.

[0052] Figure 3 This is a flow chart for predicting smoke diffusion trends based on the TCNN model in Example 1.

[0053] Figure 4This is a flow chart of using fuzzy logic to classify smoking threat levels in Example 1. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0057] Example 1, with reference to Figures 1 to 4 , which is the first embodiment of the present invention, provides a smoking monitoring and early warning method based on AI recognition, comprising the following steps:

[0058] S1, collect smart sensor data and preprocess;

[0059] Specifically, the smart sensors include high-definition cameras, smoke sensors, infrared thermal imagers, wind speed and direction sensors, and acoustic wave sensors;

[0060] The sensor data includes RGB images, smoke concentration, temperature map, wind speed and direction, and sound wave reflection signal data;

[0061] The preprocessing includes using linear interpolation to time align the sensor data, using wavelet transform to enhance the RGB image, and using a linear normalization method to normalize the smart sensor data.

[0062] By combining multimodal sensor data acquisition with preprocessing such as linear interpolation, wavelet transform, and linear normalization, comprehensive capture of smoking characteristics is achieved, which improves data consistency and quality, and enhances the robustness, real-time performance, and recognition accuracy of the monitoring system, laying a solid foundation for subsequent AI analysis. This approach is particularly suitable for complex no-smoking environments.

[0063] S2. Construct a Smoke-Unet model for image segmentation. Calculate the probability of smoke regions using the Sigmoid activation function to obtain high-confidence smoke regions. Combined with the confidence of temperature and acoustic reflection signals, determine candidate smoking areas. Predict smoke diffusion trends based on the TCNN model.

[0064] Specifically, the pre-processed RGB image, temperature map and acoustic wave reflection signal are extracted;

[0065] The time domain signal of the acoustic wave reflection signal is converted into a frequency domain signal by short-time Fourier transform to generate a spectrum diagram;

[0066] Use channel stacking to combine the RGB image, temperature map, and spectrum map to generate a five-channel tensor;

[0067] Build the Smoke-Unet model, including the input layer, encoder part, bottleneck layer, decoder part and output layer;

[0068] Set the input layer to a five-channel tensor;

[0069] Through the encoder part, the convolution formula, ReLU activation function and maximum pooling formula are used to extract features from the image;

[0070] Feature compression is used in the bottleneck layer to generate feature maps. The decoder part uses the deconvolution formula and jump connection to restore the spatial resolution of the compressed image to the size before input.

[0071] In the output layer, convolution operation and Sigmoid activation function are used to calculate the probability value of each pixel in the image belonging to the smoke area and non-smoke area, and a binary segmentation map is generated by fixed threshold method to obtain the smoke area;

[0072] The five-channel tensor is input into the Smoke-Unet model for model training, and the model parameters are optimized using the loss function and Adam optimizer;

[0073] The RGB image, temperature map, and acoustic reflection signal collected in real time by the smart sensor are converted into a five-channel tensor and input into the Smoke-Unet model to obtain the probability value of each pixel in the image belonging to the smoke area and non-smoke area. A binary segmentation map is generated using a fixed threshold method to obtain the smoke area.

[0074] In the smoke area, the probability threshold is set by the maximum entropy threshold method, the probability value is compared with the probability threshold, and the probability value greater than the probability threshold is screened to obtain the high-confidence smoke area;

[0075] Extract the corresponding spectrum in the high-confidence smoke area and use peak extraction to obtain the intensity of the sound wave reflection signal;

[0076] In high-confidence smoke areas, the thresholds of temperature and acoustic wave reflection signal intensity of each pixel are set separately through the empirical threshold method, and the confidence of temperature and acoustic wave reflection intensity of each pixel is obtained by binarization.

[0077] By setting the classification threshold based on statistical analysis, the probability value, temperature confidence, and spectrum confidence of each pixel are weighted and fused to obtain the fused confidence. The fused confidence is then compared with the classification threshold, and the pixels corresponding to the fused confidence that is higher than the classification threshold are integrated into the smoking candidate area.

[0078] Perform minimum bounding rectangle processing on the candidate smoking area and output the coordinates of the candidate smoking area.

[0079] The frequency characteristics of the acoustic signal are extracted through short-time Fourier transform, which enhances the recognition ability of smoking actions. The five-channel input fully utilizes the multi-dimensional features of RGB, temperature and acoustic spectrum. The encoder extracts deep features, and the jump connection retains spatial details. It outputs an accurate smoke segmentation map, and weighted fusion generates high-confidence candidate regions, which improves the specificity of the candidate regions and integrates scattered pixel points into regular rectangles, thereby improving the efficiency of real-time warning.

[0080] Furthermore, the coordinates of the candidate smoking areas and the wind speed and direction are integrated to generate a data set;

[0081] Set up the TCNN model, including the input layer, convolutional layer, fully connected layer, and output layer;

[0082] The input format of the input layer is set to the dataset. The convolution layer uses the ReLU activation function to extract the smoke diffusion features at different times and outputs the smoke concentration field through the fully connected layer.

[0083] Use the dataset to train the TCNN model and use the Adam optimizer and loss function to optimize the model parameters;

[0084] The coordinates and wind speed and direction of the newly generated smoking candidate area are integrated to generate a new dataset, which is input into the trained TCNN model to predict the smoke concentration field within time t.

[0085] Use linear regression to calculate the smoke diffusion speed, sort the wind speed and direction data in chronological order, generate a trend chart, and use time series analysis methods to predict the direction of smoke diffusion;

[0086] Combine the smoking candidate area with the smoke diffusion area and output the corresponding smoking point coordinates.

[0087] By combining regional coordinates with wind speed and direction data, a smoking behavior dataset in a dynamic environment was constructed, breaking through the limitations of traditional static coordinate analysis and laying the foundation for smoke diffusion prediction. Using the TCNN model, the dynamic nature of smoke diffusion was modeled in the time dimension, which improved the accuracy of the prediction. The ReLU activation function was used to enhance the TCNN's ability to extract smoke diffusion features. Combined with the fully connected layer to output the concentration field, the prediction error was reduced. The smoke speed was quantified through a linear regression formula, combined with the wind direction prediction direction, to enhance the accuracy of the diffusion behavior.

[0088] S3. Use fuzzy logic to classify smoking threat levels, use dynamic voice combinations for voice broadcasting, and monitor rectification period data in real time;

[0089] Specifically, the smoke diffusion speed and smoke concentration are fuzzy processed through membership functions;

[0090] Establish fuzzy rules and use Mamdani fuzzy reasoning method to calculate the triggering strength of each rule;

[0091] The trigger strength of all rules and the corresponding threat level center value are weighted averaged, and the result is defuzzified using the center of gravity method to obtain the threat level value.

[0092] By fuzzifying continuous variables into discrete fuzzy sets, the traditional method's reliance on precise thresholds is overcome, and the uncertainty problem of multidimensional data in complex scenarios is solved. The Mamdani fuzzy inference method fuses multiple variables through the minimum value method, surpassing the linear judgment based on a single parameter in existing technologies and enhancing the system's real-time response capability to smoking behavior. The weighted average and center of gravity methods convert fuzzy outputs into precise values, improving the flexibility, accuracy, and real-time performance of detection.

[0093] Furthermore, we use empirical rules to set monitoring thresholds, compare the threat level value with the monitoring threshold, and determine the intervention level:

[0094] If the threat level value is greater than the monitoring threshold, primary intervention is required. Personalized warning audio is generated through dynamic voice combination, smoking ban regulations and smoking hazard education content are played, and the time-limited rectification monitoring stage is entered. If smoking behavior still exists during the rectification stage, advanced intervention is carried out. The coordinates of the smoking point, smoke distribution map, and predicted smoke diffusion map are transmitted to the monitoring center in real time. The intercom equipment is used to warn the smoker and the staff are notified to stop the smoker on the spot.

[0095] The monitoring threshold is set through empirical rules to provide a scientific basis for subsequent threat comparison and ensure the rationality of intervention. The threat level value is compared with the monitoring threshold to determine the intervention level, quickly screen events that require intervention, improve real-time performance, and generate personalized warning audio through dynamic voice combination, which resolves the contradiction between real-time performance and personalization, enhances smokers' compliance awareness, sets a rectification period to solve the problem of excessive intervention, and uses advanced intervention and intercom warnings to improve the efficiency of stopping smoking behavior.

[0096] S4, store, collect and analyze the generated sensor data;

[0097] Specifically, the collected sensor data and the monitoring results generated by the analysis are stored in the database according to the timestamp, security access measures are set, and the integrity of the data in the database is regularly checked.

[0098] Through timestamp storage, the timeline integration of data is achieved, which not only facilitates post-analysis, but also supports real-time dynamic monitoring, provides an accurate basis for subsequent intervention, sets security access measures, solves data security and compliance issues, reduces the risk of internal leakage, improves the practicality and legal compliance of the system, performs integrity testing, and solves data credibility and long-term storage reliability issues.

[0099] This embodiment also provides a smoking monitoring and warning system based on AI recognition, including:

[0100] Collection and processing module, used to collect smart sensor data and perform preprocessing;

[0101] The monitoring and evaluation module is used to construct a Smoke-Unet model for image segmentation. The Sigmoid activation function is used to calculate the probability value of the smoke area to obtain the high-confidence smoke area. The combined confidence of temperature and acoustic reflection signals is combined to determine the candidate smoking area. The smoke diffusion trend is predicted based on the TCNN model.

[0102] The early warning and intervention module uses fuzzy logic to classify smoking threat levels, uses dynamic voice combinations for voice broadcasting, and monitors rectification period data in real time;

[0103] The storage management module is used to store the sensor data collected and analyzed.

[0104] This embodiment also provides a computer device, which is suitable for the smoking monitoring and early warning method based on AI recognition, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the smoking monitoring and early warning method based on AI recognition proposed in the above embodiment.

[0105] Specifically, it also includes a timing module and a detection module. The detection module is a detection telescopic rod, and an image recognition device and an infrared temperature detector are provided at the end of the detection telescopic rod.

[0106] Image recognition device: Identifies and filters cigarette images (cylindrical shape with a length of 60-120mm and a diameter of 5-8mm) and the top shape of lighters (such as push-to-ignite device, hinged metal cover, windproof flame nozzle, etc.), and issues a "no smoking" warning to smokers. At the same time, it plays the detailed rules and regulations on indoor smoking prohibition, penalties, and the dangers of smoking to stop them from smoking.

[0107] Infrared detection system: detects the temperature of cigarette butts and lighter flames, monitors PM2.5 and nicotine content, determines the specific location of smokers, issues warnings through dynamic voice combinations, informs smokers of the specific consequences of smoking, requires immediate cessation of smoking behavior, and provides a 5-second rectification period. If smoking is not stopped within the rectification period, the specific location and image will be transmitted to the main monitoring room, and the specific location of the smoker will be announced by voice in the main control room. The main computer room will intercom to call out to the smoker and contact staff for on-site intervention.

[0108] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0109] This embodiment further provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the program implements the smoking monitoring and early warning method based on AI recognition as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0110] In summary, the present invention improves the accuracy of detecting smoking behavior by combining data collected by intelligent sensors with image segmentation of the Smoke-Unet model, and improves the response speed and warning accuracy of the monitoring system by combining the smoking candidate area with the smoke diffusion prediction of the TCNN model.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A smoking monitoring and early warning method based on AI recognition, characterized by: include, Collect smart sensor data and pre-process it; A Smoke-Unet model was constructed for image segmentation. The probability of smoke regions was calculated using the Sigmoid activation function to obtain high-confidence smoke regions. The combined confidence of temperature and acoustic reflection signals was combined to identify candidate smoking areas. The TCNN model was then used to predict smoke diffusion trends. Use fuzzy logic to classify smoking threat levels, use dynamic voice combinations for voice broadcasting, and monitor rectification period data in real time; Stores collected and analyzed sensor data.

2. The smoking monitoring and early warning method based on AI recognition according to claim 1, characterized in that: The collecting of smart sensor data and pre-processing thereof refers to installing smart sensors in the monitoring area to collect sensor data and pre-processing thereof; The intelligent sensors include high-definition cameras, smoke sensors, infrared thermal imagers, wind speed and direction sensors, and acoustic wave sensors; The sensor data includes RGB images, smoke concentration, temperature map, wind speed and direction, and sound wave reflection signal data; The preprocessing includes using linear interpolation to time align the sensor data, using wavelet transform to enhance the RGB image, and using a linear normalization method to normalize the smart sensor data.

3. The smoking monitoring and early warning method based on AI recognition according to claim 2, characterized in that: The Smoke-Unet model is constructed to perform image segmentation, and the probability value of the smoke area is calculated by the Sigmoid activation function to obtain the high-confidence smoke area. The combined confidence of the temperature and the acoustic reflection signal is combined to determine the candidate smoking area, which includes extracting the preprocessed RGB image, the temperature map, and the acoustic reflection signal; The time domain signal of the acoustic wave reflection signal is converted into a frequency domain signal by short-time Fourier transform to generate a spectrum diagram; Use channel stacking to combine the RGB image, temperature map, and spectrum map to generate a five-channel tensor; Build the Smoke-Unet model, including the input layer, encoder part, bottleneck layer, decoder part and output layer; Set the input layer to a five-channel tensor; Through the encoder part, the convolution formula, ReLU activation function and maximum pooling formula are used to extract features from the image; Feature compression is used in the bottleneck layer to generate feature maps. The decoder part uses the deconvolution formula and jump connection to restore the spatial resolution of the compressed image to the size before input. In the output layer, convolution operation and Sigmoid activation function are used to calculate the probability value of each pixel in the image belonging to the smoke area and non-smoke area, and a binary segmentation map is generated by fixed threshold method to obtain the smoke area; The five-channel tensor is input into the Smoke-Unet model for model training, and the model parameters are optimized using the loss function and Adam optimizer; The RGB image, temperature map, and acoustic reflection signal collected in real time by the smart sensor are converted into a five-channel tensor and input into the Smoke-Unet model to obtain the probability value of each pixel in the image belonging to the smoke area and non-smoke area. A binary segmentation map is generated using a fixed threshold method to obtain the smoke area. In the smoke area, the probability threshold is set by the maximum entropy threshold method, the probability value is compared with the probability threshold, and the probability value greater than the probability threshold is screened to obtain the high-confidence smoke area; Extract the corresponding spectrum in the high-confidence smoke area and use peak extraction to obtain the intensity of the sound wave reflection signal; In high-confidence smoke areas, the thresholds of temperature and acoustic wave reflection signal intensity of each pixel are set separately through the empirical threshold method, and the confidence of temperature and acoustic wave reflection intensity of each pixel is obtained by binarization. By setting the classification threshold based on statistical analysis, the probability value, temperature confidence, and spectrum confidence of each pixel are weighted and fused to obtain the fused confidence. The fused confidence is then compared with the classification threshold, and the pixels corresponding to the fused confidence that is higher than the classification threshold are integrated into the smoking candidate area. Perform minimum bounding rectangle processing on the candidate smoking area and output the coordinates of the candidate smoking area.

4. The smoking monitoring and early warning method based on AI recognition according to claim 3, characterized in that: The prediction of smoke diffusion trend based on the TCNN model refers to integrating the coordinates of the candidate smoking area and the wind speed and direction to generate a data set; Set up the TCNN model, including the input layer, convolutional layer, fully connected layer, and output layer; The input format of the input layer is set to the dataset. The convolution layer uses the ReLU activation function to extract the smoke diffusion features at different times and outputs the smoke concentration field through the fully connected layer. Use the dataset to train the TCNN model and use the Adam optimizer and loss function to optimize the model parameters; The coordinates and wind speed and direction of the newly generated smoking candidate area are integrated to generate a new dataset, which is input into the trained TCNN model to predict the smoke concentration field within time t. Use linear regression to calculate the smoke diffusion speed, sort the wind speed and direction data in chronological order, generate a trend chart, and use time series analysis methods to predict the direction of smoke diffusion; Combine the smoking candidate area with the smoke diffusion area and output the corresponding smoking point coordinates.

5. The smoking monitoring and early warning method based on AI recognition according to claim 4, characterized in that: The use of fuzzy logic to classify the smoking threat level refers to fuzzifying the smoke diffusion speed and smoke concentration through a membership function; Establish fuzzy rules and use Mamdani fuzzy reasoning method to calculate the triggering strength of each rule; The trigger strength of all rules and the corresponding threat level center value are weighted averaged, and the result is defuzzified using the center of gravity method to obtain the threat level value.

6. The smoking monitoring and early warning method based on AI recognition according to claim 5, characterized in that: The use of dynamic voice combinations for voice broadcasting and real-time monitoring of rectification period data refers to setting monitoring thresholds using empirical rules, comparing the threat level value with the monitoring threshold, and determining the intervention level: If the threat level exceeds the monitoring threshold, primary intervention is required. A personalized warning audio is generated through dynamic voice combination, and content about smoking bans and smoking hazards is played. The system then enters a time-limited rectification monitoring phase. If smoking behavior persists during the rectification phase, advanced intervention is implemented. The coordinates of the smoking point, smoke distribution map, and predicted smoke diffusion map are transmitted to the monitoring center in real time. Intercom equipment is used to warn smokers, and staff are notified to stop them on the spot. The storing of the sensor data collected and analyzed refers to storing the collected sensor data and the monitoring results generated by the analysis in a database according to timestamps, setting security access measures, and regularly performing integrity checks on the data in the database.

7. A smoking monitoring and early warning system based on AI recognition, based on the smoking monitoring and early warning method based on AI recognition according to any one of claims 1 to 6, characterized in that: include, Collection and processing module, used to collect smart sensor data and perform preprocessing; The monitoring and evaluation module is used to construct a Smoke-Unet model for image segmentation. The Sigmoid activation function is used to calculate the probability value of the smoke area to obtain the high-confidence smoke area. The combined confidence of temperature and acoustic reflection signals is combined to determine the candidate smoking area. The smoke diffusion trend is predicted based on the TCNN model. The early warning and intervention module uses fuzzy logic to classify smoking threat levels, uses dynamic voice combinations for voice broadcasting, and monitors rectification period data in real time; The storage management module is used to store the sensor data collected and analyzed.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the smoking monitoring and early warning method based on AI recognition according to any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the smoking monitoring and early warning method based on AI recognition are implemented in any one of claims 1 to 6.

10. The computer device according to claim 9, wherein: It also includes a timing module and a detection module. The detection module is a detection telescopic rod, and an image recognition device and an infrared temperature detector are provided at the end of the detection telescopic rod.

Images