An image recognition-based intelligent exhaled gas alcohol content detection system

Abstract

The application provides an image recognition-based intelligent breath alcohol content detection system, and relates to the technical field of intelligent detection, and aims to solve the problems of traditional sensors, such as being easily disturbed by the environment, needing frequent calibration and insufficient dynamic response. The system comprises a gas-sensitive color development unit, a multispectral dynamic imaging unit, a data processing unit and a closed-loop control unit. The gas-sensitive color development unit adopts a reversible gas-sensitive film, generates a color change signal through an alcohol-specific color development reaction, the multispectral dynamic imaging unit combines ambient light suppression and dynamic area positioning technology to capture spectral reflectance time series data of the color development process, the data processing unit maps the chroma characteristics into alcohol concentration values based on a Gaussian process regression model, and triggers near-infrared auxiliary verification through confidence evaluation, and the closed-loop control unit maintains a constant detection environment through a temperature control and airflow stabilization module. The scheme can realize non-invasive high-precision detection of alcohol concentration in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection technology, and in particular to an intelligent system for detecting the alcohol content in exhaled breath based on image recognition. Background Technology

[0002] Traditional breath alcohol detection technologies primarily rely on electrochemical sensors or semiconductor gas-sensitive elements. Their accuracy is easily affected by factors such as ambient temperature and humidity, airflow fluctuations, and sensor aging. For example, electrochemical sensors require regular electrolyte replacement and are prone to baseline drift in high-humidity environments; while semiconductor sensors are less expensive, they suffer from poor selectivity and are susceptible to interference from other volatile organic compounds (such as acetone), leading to a high false alarm rate. Furthermore, existing technologies mostly employ single-point static detection, failing to capture the dynamic changes in alcohol concentration in real time, and lack self-calibration mechanisms, requiring periodic manual calibration, resulting in high maintenance costs. In complex environments (such as extreme temperatures and light variations), the stability of traditional sensors significantly decreases, making it difficult to meet the high-reliability detection requirements of law enforcement, medical, and other fields. Summary of the Invention

[0003] This invention provides an intelligent breath alcohol content detection system based on image recognition to solve the problems of existing technologies, such as the susceptibility of detection accuracy to environmental interference, the need for frequent manual calibration, and insufficient dynamic response.

[0004] This invention provides an intelligent system for detecting the alcohol content in exhaled breath based on image recognition, comprising: a gas-sensitive colorimetric unit, configured with a reversible gas-sensitive film that undergoes a specific colorimetric reaction with alcohol, the color change of which is related to the alcohol concentration by a preset function; A multispectral dynamic imaging unit, equipped with a multispectral light source and an image sensor, is used to capture the time-series data of the spectral reflectance of the color-developing film during the dynamic reaction process; The data processing unit maps the chromaticity features in the time series data of spectral reflectance to alcohol concentration values ​​based on a machine learning model and generates confidence assessment results. The closed-loop control unit integrates temperature regulation and airflow control devices, and maintains a constant temperature environment and stable expiratory flow rate in the air chamber through feedback control algorithms.

[0005] Furthermore, the multispectral dynamic imaging unit includes: The ambient light suppression module extracts ambient light interference signals from continuously acquired spectral images using a time-difference algorithm. The dynamic region positioning module, based on the edge gradient characteristics of the color developing film, divides the effective reaction region in real time. The effective reaction region is a continuous region in the color developing film with uniform color change and meeting the preset morphological conditions, and excludes non-uniform regions caused by airflow fluctuations. The spectral data generation module aggregates multi-band reflectance data within the effective reaction area according to a time series to generate a dynamic spectral feature matrix.

[0006] Furthermore, the data processing unit includes: The chromaticity feature extraction module converts the dynamic spectral feature matrix to a preset color space and extracts a joint feature vector composed of saturation and hue. The concentration mapping module performs nonlinear fitting on the joint feature vector using a Gaussian process regression model, and outputs the predicted alcohol concentration and confidence interval. The cross-validation module triggers the near-infrared transmission spectroscopy channel to acquire auxiliary data when the confidence interval width exceeds the threshold, and then recalculates the concentration value after weighted fusion with the colorimetric features.

[0007] Furthermore, the training method for the Gaussian process regression model includes: collecting historical colorimetric reaction data, including time-series data of spectral reflectance at different alcohol concentrations and the corresponding standard concentration true values; fitting the nonlinear relationship between chromaticity features and concentration values ​​through a kernel function optimization algorithm, and dynamically adjusting the model hyperparameters based on the Bayesian criterion; performing robustness verification on the trained model, and if the error rate on the interference test set exceeds a preset upper limit, then re-performing feature selection and kernel function matching.

[0008] Furthermore, the closed-loop control unit includes: The temperature control module monitors the temperature of the air chamber in real time and compares it with the preset target value. If a temperature deviation is detected, dynamic compensation is performed through a semiconductor temperature control device. The airflow stabilization module adjusts the output power of the micro air pump using a proportional-integral-derivative algorithm based on the fluctuation data fed back by the expiratory flow rate sensor, so that the flow rate is maintained within a preset range. The anomaly handling module triggers system self-locking and generates a hardware fault alarm signal when the temperature or flow rate continuously exceeds the set limit for a set period of time.

[0009] Furthermore, the system includes an adaptive calibration unit, whose execution logic includes: activating the built-in reference sample before each detection and collecting the colorimetric feature data of the reference sample; calculating the feature deviation between the real-time colorimetric feature and the reference sample; if the deviation exceeds the tolerance threshold, initiating the model parameter correction program; and injecting the deviation compensation amount into the machine learning model through an incremental learning algorithm to update the concentration mapping relationship until the feature deviation amount is restored to the tolerance range.

[0010] Furthermore, the reset method of the gas-sensitive colorimetric unit includes: after the detection is completed, starting the thin film heating device to apply a preset temperature field to the colorimetric film to trigger the reverse reaction to restore the initial color; verifying the color state after reset through an optical sensor, if the color difference with the initial state exceeds the allowable range, marking the film as faulty and prohibiting subsequent detection; counting the cumulative number of times the film is used, and automatically generating a replacement reminder signal when the number of uses reaches a preset lifespan threshold.

[0011] Furthermore, the system includes a user interaction unit, which includes: dynamically plotting an alcohol concentration change curve and overlaying and displaying the confidence interval boundary and the statutory threshold reference line; automatically generating a verifiable digital certificate containing a timestamp, concentration value, and calibration status when the detected value exceeds the statutory threshold; uploading the digital certificate to the regulatory platform through an encrypted channel and receiving the compliance certification result from the platform.

[0012] Furthermore, the execution logic of the incremental learning algorithm is as follows: calculate the model weights and correct the gradient based on the feature deviation, and limit the gradient update magnitude to prevent overfitting; retain the statistical distribution characteristics of historical training data, and balance the weight ratio of new and old data through a sliding window mechanism; perform local verification on the corrected model after calibration, and if the verification error does not decrease, roll back to the previous stable version.

[0013] This solution achieves dynamic colorimetric reaction of alcohol concentration through the reversible thin film of the gas-sensitive colorimetric unit. Combined with multispectral dynamic imaging and Gaussian process regression model, it significantly improves detection accuracy and anti-interference capability. The closed-loop control unit maintains a constant detection environment through temperature control and airflow stabilization modules, solving the temperature and humidity sensitivity problem of traditional sensors. The adaptive calibration unit uses an incremental learning algorithm to dynamically correct model parameters, extending the calibration cycle and reducing maintenance costs. The dynamic region positioning module eliminates airflow fluctuation interference areas, ensuring high detection stability even in complex environments. Overall, this solution addresses the shortcomings of existing technologies, such as susceptibility to environmental interference, the need for frequent manual calibration, and insufficient dynamic response. Attached Figure Description

[0014] Figure 1 This is a diagram illustrating the architecture of an intelligent system for detecting exhaled breath alcohol content based on image recognition, as described in this invention. Detailed Implementation

[0015] This invention relates to an intelligent system for detecting alcohol content in exhaled breath based on image recognition, which aims to achieve accurate detection of alcohol content in exhaled breath by combining image recognition technology with multi-feature analysis.

[0016] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0017] Example 1 like Figure 1 The architecture diagram of an intelligent system for detecting breath alcohol content based on image recognition is shown, specifically including: The gas-sensitive colorimetric unit is equipped with a reversible gas-sensitive membrane that undergoes a specific colorimetric reaction with alcohol. The color change of this membrane is related to the alcohol concentration according to a preset function. The gas-sensitive membrane is supported on a specific redox catalyst (such as CrO2 / SiO2) using nanomaterials. When alcohol molecules come into contact with the membrane surface, a redox reaction is triggered, causing the membrane color to gradually change from an initial color (e.g., orange-red) to a target color (e.g., green). The dynamic process of color change exhibits a non-linear mapping relationship with the alcohol concentration. A "color-concentration" conversion model is established through pre-experimental calibration. For example, when the alcohol concentration increases from 0.01 mg / L to 0.1 mg / L, the membrane hue (H value) changes linearly from 120° to 60°, and the saturation (S value) increases from 30% to 80%. The membrane is reversible; after detection, it can be reset to its initial color state by heating, ensuring reusability.

[0018] A multispectral dynamic imaging unit, equipped with a multispectral light source and image sensor, is used to capture the temporal data of the spectral reflectance of the color-developing film during the dynamic reaction process. Specifically, a narrowband multispectral light source (e.g., 520nm±5nm, 600nm±5nm) is used to cover the color-developing reaction sensitive wavelength band, avoiding ambient light interference. A high frame rate CMOS sensor (e.g., 30 frames per second) is configured to acquire the dynamic spectral image sequence of the color-developing film in real time. Based on edge detection algorithms (e.g., the Canny operator), continuous regions with uniform color changes in the film (defined as effective reaction regions) are segmented to eliminate edge blurring or non-uniform regions caused by airflow disturbances. Through a temporal difference algorithm, the static ambient light component is separated from the continuous image sequence, preserving the dynamic color-developing signal. For example, the difference between two adjacent frames is calculated to eliminate fixed background noise.

[0019] The data processing unit maps chromaticity features from time-series spectral reflectance data to alcohol concentration values ​​based on a machine learning model and generates confidence assessment results. Specifically, it converts the spectral image of the effective reaction region to the HSV color space and extracts a joint feature vector composed of saturation (S) and hue (H). For example, H values ​​are sensitive to changes at low alcohol concentrations, while S values ​​change significantly at high concentrations. The Matérn kernel function is used to fit the nonlinear relationship between chromaticity features and alcohol concentration. Training data includes historical colorimetric reaction data (such as H / S values ​​at 100 different concentrations) and standard concentration values ​​from gas chromatography (GC). When the confidence interval width of the model output exceeds a threshold (such as ±5%), the near-infrared transmission spectrum (1450nm) auxiliary detection channel is triggered, and the accuracy is improved by weighted fusion of dual-modal data.

[0020] The closed-loop control unit integrates temperature regulation and airflow control devices, maintaining a constant temperature environment and stable expiratory flow rate within the air chamber through a feedback control algorithm. Specifically, it monitors the air chamber temperature in real time; if it deviates from the preset range (e.g., 25±0.5℃), it dynamically compensates by heating or cooling using a Peltier element; it predicts trends based on historical temperature fluctuation data and dynamically adjusts the integral term weight of the PID algorithm to prevent overshoot; it monitors the expiratory flow rate through a MEMS flow sensor; if the detected flow rate fluctuation exceeds a threshold (e.g., ±5%), it adjusts the power output of the micro-pump; and it employs a feedforward-feedback composite control algorithm to quickly respond to changes in expiratory pressure.

[0021] Furthermore, the multispectral dynamic imaging unit includes: The ambient light suppression module extracts ambient light interference signals from continuously acquired spectral images using a time-difference algorithm. The dynamic region positioning module, based on the edge gradient characteristics of the color developing film, divides the effective reaction region in real time. The effective reaction region is a continuous region in the color developing film with uniform color change and meeting the preset morphological conditions, and excludes non-uniform regions caused by airflow fluctuations. The spectral data generation module aggregates multi-band reflectance data within the effective reaction area according to a time series to generate a dynamic spectral feature matrix.

[0022] Specifically, in a continuously acquired sequence of spectral images (e.g., 30 frames per second), pixel-level difference operations are performed on adjacent frames to extract dynamically changing colorimetric signals; high-frequency noise is eliminated by sliding window mean filtering while retaining low-frequency colorimetric components. Ambient light interference is also eliminated.

[0023] The Canny edge detection algorithm is used to identify the boundaries of the color-developing film and generate a binary mask. Based on morphological closing operations, the internal pores of the film are filled, and continuous regions with uniform color change (defined as the effective reaction region) are extracted. The average chromaticity of pixels within the effective reaction region is calculated as the basis for subsequent analysis. Edge blurring caused by airflow disturbances is excluded.

[0024] Time series alignment is performed on multi-band (e.g., 520nm, 600nm) reflectance data within the effective reaction area. A dynamic spectral feature matrix is ​​constructed, where rows represent time points and columns represent band reflectance intensity; standardized input data is generated to meet the processing requirements of machine learning models.

[0025] Furthermore, the data processing unit includes: The chromaticity feature extraction module converts the dynamic spectral feature matrix to a preset color space and extracts a joint feature vector composed of saturation and hue. The concentration mapping module performs nonlinear fitting on the joint feature vector using a Gaussian process regression model, and outputs the predicted alcohol concentration and confidence interval. The cross-validation module triggers the near-infrared transmission spectroscopy channel to acquire auxiliary data when the confidence interval width exceeds the threshold, and then recalculates the concentration value after weighted fusion with the colorimetric features.

[0026] Specifically, the dynamic spectral feature matrix is ​​transformed from RGB space to HSV space, and the joint feature vector of saturation (S) and hue (H) is extracted; the H and S values ​​are normalized to eliminate the influence of light intensity differences; and the feature dimension is compressed to 2-dimensional.

[0027] A Gaussian process regression (GPR) model was adopted, and the nonlinear relationship between chromaticity features and alcohol concentration was fitted with a Matérn5 / 2 kernel function. The hyperparameters of the kernel function (such as length scale and variance) were automatically adjusted by Bayesian optimization, which reduced the prediction error of the model in the concentration range of 0.01-2.0 mg / L.

[0028] When the confidence interval width output by the GPR model exceeds the threshold (e.g., ±5%), near-infrared transmission spectroscopy (1450nm) is triggered for auxiliary detection; the colorimetric data and transmission data are weighted and fused (the weights are dynamically allocated according to the confidence level), and the alcohol concentration is recalculated.

[0029] Furthermore, the training method for the Gaussian process regression model includes: collecting historical colorimetric reaction data, including time-series data of spectral reflectance at different alcohol concentrations and the corresponding standard concentration true values; fitting the nonlinear relationship between chromaticity features and concentration values ​​through a kernel function optimization algorithm, and dynamically adjusting the model hyperparameters based on the Bayesian criterion; performing robustness verification on the trained model, and if the error rate on the interference test set exceeds a preset upper limit, then re-performing feature selection and kernel function matching.

[0030] Specifically, historical colorimetric reaction data are collected, covering different temperature, humidity, and alcohol concentration conditions; abnormal data (such as sensor failure frames) are removed, and missing values ​​are filled by interpolation.

[0031] The grid search method is used to evaluate the fitting effect of different kernel functions (such as RBF and Matérn) and select the optimal kernel type; the hyperparameters of the kernel function are dynamically adjusted based on the marginal likelihood maximization criterion.

[0032] Verify the model's generalization ability on a disturbance test set (including simulated ambient light abrupt changes and airflow disturbance data); if the error rate exceeds the threshold (e.g., 5%), re-perform feature selection or introduce regularization terms.

[0033] Furthermore, the closed-loop control unit includes: The temperature control module monitors the temperature of the air chamber in real time and compares it with the preset target value. If a temperature deviation is detected, dynamic compensation is performed through a semiconductor temperature control device. The airflow stabilization module adjusts the output power of the micro air pump using a proportional-integral-derivative algorithm based on the fluctuation data fed back by the expiratory flow rate sensor, so that the flow rate is maintained within a preset range. The anomaly handling module triggers system self-locking and generates a hardware fault alarm signal when the temperature or flow rate continuously exceeds the set limit for a set period of time.

[0034] Specifically, for example, the temperature of the air chamber is monitored in real time by a DS18B20 temperature sensor with a sampling frequency of 10Hz; if the temperature deviates from the preset range (e.g., 25±0.5℃), the Peltier element is triggered to perform dynamic compensation (the heating / cooling power is adjusted by PID according to the deviation).

[0035] The MEMS flow sensor provides real-time feedback on the expiratory flow rate. If the flow rate exceeds the limit (e.g., 1.5L / min ± 5%), the pump speed is adjusted via PWM. Feedforward control is used to predict changes in expiratory pressure and adjust the pump output in advance, thereby improving flow rate stability and avoiding local overload of the membrane.

[0036] If the temperature or flow rate continuously exceeds the set time (e.g., 10 seconds), the system will trigger a self-lock and generate a fault code (e.g., E01: abnormal temperature, E02: abnormal flow rate); an alarm will be triggered synchronously via a buzzer and LED indicator, and a fault log will be recorded.

[0037] Furthermore, the adaptive calibration unit: activates the built-in reference sample before each detection, collects the colorimetric feature data of the reference sample; calculates the feature deviation between the real-time colorimetric feature and the reference sample; if the deviation exceeds the tolerance threshold, the model parameter correction program is initiated; and injects the deviation compensation amount into the machine learning model through an incremental learning algorithm to update the concentration mapping relationship until the feature deviation amount is restored to the tolerance range.

[0038] Specifically, before testing, the built-in reference colorimetric block (sealed and containing 0.1 mg / L alcohol gas) is activated, and its colorimetric characteristic data are collected. , ); Calculate real-time colorimetric features ( , Deviation from reference value: , .

[0039] like >2° or If the value is greater than 5%, initiate the incremental learning algorithm to update the GPR model parameters.

[0040] The bias is converted into a gradient to correct the model weights, and the magnitude of a single update is limited to no more than 10% of the original weights.

[0041] A sliding window of historical data (such as the most recent 100 detections) is retained to prevent model drift. Furthermore, the reset method of the gas-sensitive colorimetric unit includes: after the detection is completed, starting the thin film heating device to apply a preset temperature field to the colorimetric film to trigger the reverse reaction to restore the initial color; verifying the color state after reset through an optical sensor, if the color difference with the initial state exceeds the allowable range, marking the film as faulty and prohibiting subsequent detection; counting the cumulative number of times the film is used, and automatically generating a replacement reminder signal when the number of uses reaches a preset lifespan threshold.

[0042] Specifically, after the test is completed, the thin film heating device (50±2℃) is activated and the reaction of CrO2 is triggered for 30 seconds to restore the initial orange-red color.

[0043] The LAB color difference value of the film after reset was measured using a color sensor. ,like >5 (compared to the initial state), the reset is determined to have failed.

[0044] The system tracks the cumulative number of times the film has been used. When the number of uses exceeds a threshold (e.g., 500 times), the system prompts the user to replace the film via an interactive interface.

[0045] Furthermore, the user interaction unit includes: dynamically plotting an alcohol concentration change curve and overlaying it with the confidence interval boundary and the legal threshold reference line; automatically generating a verifiable digital certificate containing a timestamp, concentration value, and calibration status when the detected value exceeds the legal threshold; uploading the digital certificate to the regulatory platform through an encrypted channel and receiving the compliance certification result from the platform.

[0046] Specifically, the alcohol concentration-time curve is plotted in real time, and confidence intervals are displayed overlaid (e.g., shaded areas represent 95% confidence); when the concentration approaches the legal threshold (0.15 mg / L), the curve color gradually changes from green to red.

[0047] Generate a test report containing a timestamp, device serial number, and SHA-256 signature, and upload it to the monitoring platform via HTTPS protocol; if the network is interrupted, store the encrypted report locally until the connection is restored.

[0048] Furthermore, the execution logic of the incremental learning algorithm is as follows: calculate the model weights and correct the gradient based on the feature deviation, and limit the gradient update magnitude to prevent overfitting; retain the statistical distribution characteristics of historical training data, and balance the weight ratio of new and old data through a sliding window mechanism; perform local verification on the corrected model after calibration, and if the verification error does not decrease, roll back to the previous stable version.

[0049] Specifically, the optimization logic of the incremental learning algorithm: Calculate the model weight correction based on the characteristic deviation. Where α is the learning rate (default 0.01), and X is the input feature. To account for the expected output bias, further, the most recent N sets of training data (N=100) are retained, and the weight of the new data is 1.2 times that of the old data to prevent dilution of historical data. After the update, use the validation set (20% of historical data) to test the model error. If the error increases by more than 1%, roll back to the previous stable version.

[0050] This system achieves non-invasive, high-precision alcohol detection through a combination of technologies including gas-sensitive colorimetric reaction, multispectral dynamic imaging, closed-loop environmental control, and adaptive calibration, solving the problems of traditional sensors being greatly affected by environmental interference and requiring frequent calibration.

[0051] It should be understood that the embodiments disclosed in this invention and the above description enable those skilled in the art to implement this invention. However, this invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this invention.

Claims

1. An intelligent system for detecting exhaled breath alcohol content based on image recognition, characterized in that, include: The gas-sensitive colorimetric unit is equipped with a reversible gas-sensitive film that undergoes a specific colorimetric reaction with alcohol, and its color change is related to the alcohol concentration in a preset functional relationship. A multispectral dynamic imaging unit, equipped with a multispectral light source and an image sensor, is used to capture the time-series data of the spectral reflectance of the color-developing film during the dynamic reaction process; The data processing unit maps the chromaticity features in the time series data of spectral reflectance to alcohol concentration values ​​based on a machine learning model and generates confidence assessment results. The closed-loop control unit integrates a temperature regulation device and an airflow control device, and maintains a constant temperature environment and stable expiratory flow rate in the air chamber through a feedback control algorithm. The data processing unit includes: The chromaticity feature extraction module converts the dynamic spectral feature matrix to a preset color space and extracts a joint feature vector composed of saturation and hue. The concentration mapping module performs nonlinear fitting on the joint feature vector using a Gaussian process regression model, and outputs the predicted alcohol concentration and confidence interval. The cross-validation module triggers the near-infrared transmission spectroscopy channel to acquire auxiliary data when the confidence interval width exceeds the threshold, and then recalculates the concentration value after weighted fusion with the colorimetric features. The closed-loop control unit includes: The temperature control module monitors the temperature of the air chamber in real time and compares it with the preset target value. If a temperature deviation is detected, dynamic compensation is performed through a semiconductor temperature control device. The airflow stabilization module adjusts the output power of the micro air pump using a proportional-integral-derivative algorithm based on the fluctuation data fed back by the expiratory flow rate sensor, so that the flow rate is maintained within a preset range. The anomaly handling module triggers system self-locking and generates a hardware fault alarm signal when the temperature or flow rate continuously exceeds the set time. The system also includes an adaptive calibration unit, whose execution logic includes: activating the built-in reference sample before each detection and collecting the colorimetric feature data of the reference sample; calculating the feature deviation between the real-time colorimetric feature and the reference sample; if the deviation exceeds the tolerance threshold, starting the model parameter correction program; injecting the deviation compensation amount into the machine learning model through an incremental learning algorithm, and updating the concentration mapping relationship until the feature deviation amount is restored to the tolerance range. The reset method of the gas-sensitive colorimetric unit includes: after the detection is completed, the thin film heating device is activated to apply a preset temperature field to the colorimetric thin film to trigger the reverse reaction to restore the initial color; the color state after reset is verified by an optical sensor, and if the color difference with the initial state exceeds the allowable range, the thin film is marked as faulty and subsequent detection is prohibited; the cumulative number of times the thin film is used is counted, and when the number of times reaches the preset lifespan threshold, a replacement reminder signal is automatically generated.

2. The intelligent breath alcohol content detection system based on image recognition as described in claim 1, characterized in that, The multispectral dynamic imaging unit includes: The ambient light suppression module extracts ambient light interference signals from continuously acquired spectral images using a time-difference algorithm. The dynamic region positioning module, based on the edge gradient characteristics of the color developing film, divides the effective reaction region in real time. The effective reaction region is a continuous region in the color developing film with uniform color change and meeting the preset morphological conditions, and excludes non-uniform regions caused by airflow fluctuations. The spectral data generation module aggregates multi-band reflectance data within the effective reaction area according to a time series to generate a dynamic spectral feature matrix.

3. The intelligent breath alcohol content detection system based on image recognition as described in claim 1, characterized in that, The training method for the Gaussian process regression model includes: collecting historical colorimetric reaction data, including time-series data of spectral reflectance at different alcohol concentrations and the corresponding standard concentration true values; fitting the nonlinear relationship between colorimetric features and concentration values ​​using a kernel function optimization algorithm, and dynamically adjusting the model hyperparameters based on the Bayesian criterion; and performing robustness verification on the trained model. If the error rate on the interference test set exceeds a preset upper limit, feature selection and kernel function matching are re-executed.

4. The intelligent breath alcohol content detection system based on image recognition as described in claim 1, characterized in that, The system further includes a user interaction unit, which includes: dynamically plotting an alcohol concentration change curve and overlaying and displaying the confidence interval boundary and the statutory threshold reference line; automatically generating a verifiable digital certificate containing a timestamp, concentration value, and calibration status when the detected value exceeds the statutory threshold; uploading the digital certificate to the regulatory platform through an encrypted channel and receiving the compliance certification result from the platform.

5. The intelligent breath alcohol content detection system based on image recognition as described in claim 1, characterized in that, The execution logic of the incremental learning algorithm is as follows: calculate the model weights and correct the gradient based on the feature deviation, and limit the gradient update magnitude to prevent overfitting; retain the statistical distribution characteristics of historical training data, and balance the weight ratio of new and old data through a sliding window mechanism; after calibration, perform local verification on the corrected model, and if the verification error does not decrease, roll back to the previous stable version.

Images