Fire smoke detector, test platform and fire smoke detection method

Abstract

The invention discloses a fire smoke detector, a test platform and a fire smoke detection method, and relates to the technical field of fire detection. An optical cavity in the fire smoke detector is used for fixing an included angle between an incident optical axis of a first light source module and an optical axis of the detector as a target angle; the first receiving module is used for collecting scattered spectrum signals of a plurality of discrete channels at a target angle; the first processing unit is used for processing the scattered spectrum signal based on a pre-trained target algorithm and outputting a fire probability; the target angle is determined by the measuring platform; a second receiving module in the measurement platform is used for collecting test scattering spectrum signals of a plurality of discrete channels under various environmental aerosols within a preset test angle range; and the second processing unit is used for screening out a target angle according to the recognition accuracy of a plurality of preset machine learning algorithms under different test angles. By adopting the method, the optimal scattering angle can be selected so as to realize high-sensitivity identification and quick response to fire smoke.

Description

Technical Field

[0001] The present invention relates to the technical field of fire detection, in particular to a fire smoke detector, a test platform and a fire smoke detection method. Background Art

[0002] Photoelectric smoke detectors typically use a single-wavelength LED light source, coupled with a fixed-angle photosensor, to determine the presence of a fire by monitoring the intensity of incident light scattered by smoke particles. These detectors typically achieve high sensitivity under standard testing conditions. However, in complex real-world scenarios, such as the presence of non-fire particles like dust, water mist, and soot, they are prone to false alarms or missed alarms, significantly impacting their reliability and practicality.

[0003] The fundamental reason for this problem is that the light scattering and absorption behavior of aerosol particles is influenced by multiple factors, including the wavelength of the light source, the scattering angle, the chemical composition of the particles, their size distribution, and their concentration. Traditional detectors, however, only provide low-dimensional signal input based on a single wavelength and angle. This limited information dimension makes it difficult to accurately distinguish the optical responses of fire smoke from non-fire interference sources. This low-dimensional data structure lacks sufficient discriminative power in practical applications, leading to a high false alarm rate, especially in complex environments such as high humidity, dusty conditions, and kitchens.

[0004] To improve detection accuracy, attempts are underway to introduce multi-wavelength, multi-angle scatterometry techniques to increase the information dimension. For example, using a light source with two or more wavelengths and measuring the scattering intensity at different angles, this approach enhances recognition capabilities by leveraging the combined spectral and angular features. Related technologies, guided by Mie scattering theory, select typical angle combinations such as 45° and 135°. However, this angle selection generally relies on theoretical analysis or empirical settings, resulting in limited effectiveness in practical applications. Alternatively, the relative deviation of the spectral response is calculated based on a pre-set aerosol model (e.g., a set particle size distribution or refractive index parameter), and the optimal angle is selected based on minimizing this deviation. However, fire smoke in real-world environments is highly dynamic, and its particle size and optical properties are affected by multiple factors, such as the burning material and ventilation conditions. Pre-set models are often difficult to accurately fit, resulting in insufficient generalization performance for angle selection. Furthermore, most of these methods focus on spectral response consistency rather than directly optimizing classification accuracy, failing to fully reflect the true performance of fire identification in practical detection applications.

[0005] In summary, although existing technologies have attempted to improve detection accuracy by increasing the sensing dimension, there are still many limitations in scattering angle selection, signal discrimination ability and adaptability to complex environments. Further research is urgently needed to develop more practical and accurate solutions. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems in the related art to a certain extent. To this end, the present invention aims to provide a fire smoke detector, a test platform and a fire smoke detection method to achieve high-sensitivity recognition and rapid response to fire smoke.

[0007] In order to achieve the above objectives, a first embodiment of the present invention provides a fire smoke detector, comprising: an optical cavity, used to fix the angle between the incident light axis of the first light source module and the detector optical axis to a target angle; a first receiving module, configured to collect scattered spectrum signals of a plurality of discrete channels at the target angle, and output the scattered spectrum signals to a first processing unit; a first processing unit, configured to process the scattered spectrum signal based on a pre-trained target algorithm and output a fire probability; Wherein, the target angle is determined by a measuring platform; the measuring platform includes a second light source module and a second receiving module installed on an optical motion turntable; The second light source module is used to output a test light source; The optical motion turntable rotates counterclockwise within a preset test angle range at a set step length; The second receiving module is used to collect test scattering spectrum signals of multiple discrete channels under multiple environmental aerosols within a preset test angle range as the optical motion turntable rotates to form a data set; The second processing unit is used to train the preset multiple machine learning algorithm data based on the data set, and screen out the target angle in the preset test angle range according to the recognition accuracy of the preset multiple machine learning algorithms at different test angles.

[0008] In addition, the method of the above embodiment of the present invention may also have the following additional technical features: According to one embodiment of the present invention, the target algorithm is determined by the measurement platform; The second processing unit is further used to determine the target algorithm from among the preset multiple machine learning algorithms based on the recognition effects of the preset multiple machine learning algorithms in the fire smoke and interference source binary classification task.

[0009] According to an embodiment of the present invention, the first processing unit is further configured to perform normalization processing on the scattering spectrum signals corresponding to a plurality of preset discrete channels.

[0010] According to one embodiment of the present invention, the wavelengths of the plurality of discrete channels include 415 nm, 445 nm, 480 nm, 515 nm, 555 nm, 590 nm, 630 nm and 680 nm; The preset test angle range is 40° to 85° when the optical motion turntable rotates counterclockwise; the setting step is 5°; The environmental aerosols include fire smoke and non-fire interference sources; The preset multiple machine learning algorithms include random forest algorithm, logistic regression algorithm, support vector machine algorithm, XGBoost algorithm, and K nearest neighbor algorithm.

[0011] According to one embodiment of the present invention, in an initial state, the second light source module and the second processing unit are on the same horizontal line; The second light source module emits a light beam when the optical motion turntable rotates counterclockwise to a starting angle of the preset test angle range, and the second receiving module is used to collect background scattering spectrum signals; If the difference between the test scattering spectrum signal and the background scattering spectrum signal is not greater than 0, re-collection of the background scattering spectrum signal is triggered.

[0012] According to one embodiment of the present invention, the first light source module and the second light source module respectively include an LED light source; the LED light source sequentially outputs a parallel light beam through a collimator, a beam expander, and a lens; The first receiving module and the second receiving module respectively include a spectrum sensor chip and an aperture; the front end of the spectrum sensor chip is closely adjacent to the rear end of the aperture.

[0013] According to one embodiment of the present invention, the first processing unit is further configured to: If the fire probability is greater than a first set threshold, an audible and visual alarm is triggered and pushed to the cloud; If the fire probability is less than the first set threshold and greater than the second set threshold, a local buzzer alarm is triggered.

[0014] To achieve the above objectives, a second embodiment of the present invention provides a test platform, comprising: A second light source module, used for outputting a test light source; A second receiving module mounted on an optical motion turntable; the optical motion turntable rotates counterclockwise within a preset test angle range at a set step size; the second receiving module is configured to collect test scattering spectral signals of multiple discrete channels under multiple ambient aerosols within the preset test angle range as the optical motion turntable rotates, thereby forming a training set; The second processing unit is used to train the preset multiple machine learning algorithm data based on the training set, determine the target algorithm according to the recognition effect of the preset multiple machine learning algorithms in the fire smoke and interference source binary classification task; and screen out the target angle within the preset test angle range according to the recognition accuracy of the preset multiple machine learning algorithms at different test angles; the target algorithm represents the algorithm model used to process the scattered spectrum signal collected by the fire smoke detector; the target angle represents the scattering angle corresponding to the fire smoke detector when collecting the scattered spectrum signal.

[0015] To achieve the above-mentioned object, a third embodiment of the present invention provides a fire smoke detection method, which is applied to the fire smoke detector provided in the first embodiment of the present invention. The method includes: Controlling the first light source module to emit a light beam to illuminate the aerosol to be measured, and collecting scattered spectrum signals of the aerosol to be measured in multiple discrete channels at a target angle based on the first receiving module; Normalizing the scattered spectrum signal, inputting the normalized scattered spectrum signal into a target algorithm for processing, and outputting a fire probability; wherein, if the fire probability is greater than a first set threshold, triggering an audible and visual alarm and sending the information to the cloud; If the fire probability is less than the first set threshold and greater than the second set threshold, a local buzzer alarm is triggered.

[0016] To achieve the above objectives, a fourth embodiment of the present invention provides a fire smoke detection method, which is applied to the test platform provided in the second embodiment of the present invention. The method includes: Controlling the optical motion turntable to rotate counterclockwise to a starting angle of a preset test angle range; wherein, when the optical motion turntable is in an initial state, the second light source module and the second receiving module are on the same horizontal line; controlling the second light source module to emit a light beam, and collecting a background scattering spectrum signal based on the second receiving module; introducing different ambient aerosols into the measurement platform, controlling the optical motion turntable to rotate counterclockwise within the preset test angle range with a set step length, and collecting test scattering spectrum signals of multiple discrete channels under multiple ambient aerosols based on the second receiving module to form a data set; If the difference between the test scattering spectrum signal and the background scattering spectrum signal is not greater than 0, triggering re-collection of the background scattering spectrum signal; If the difference between the test scattering spectrum signal and the background scattering spectrum signal is greater than 0, the preset multiple machine learning algorithm data are trained based on the data set, and the target algorithm is determined among the preset multiple machine learning algorithms according to the recognition effect of the preset multiple machine learning algorithms in the fire smoke and interference source binary classification task; and according to the recognition accuracy of the preset multiple machine learning algorithms at different test angles, the target angle is screened out within the preset test angle range; the target algorithm represents the algorithm used to process the scattering spectrum signal collected by the fire smoke detector; the target angle represents the scattering angle at which the fire smoke detector is detecting.

[0017] The fire smoke detector, test platform, and fire smoke detection method of the present invention utilize a first light source module to output a parallel light beam. Combined with an optical cavity, the angle between the incident light axis and the detector optical axis is fixed at a target angle determined by the measurement platform. This provides a stable geometric configuration for scattered spectrum acquisition, enhancing the consistency and comparability of data acquisition. Furthermore, the use of a single-angle detector effectively reduces hardware cost and size. The first receiving module collects scattered spectrum signals from multiple discrete channels at the target angle, and the first processing unit processes them using a pre-trained target algorithm, thereby improving the accuracy and response speed of fire smoke recognition. Furthermore, the target angle and target algorithm are not manually set. Instead, they rely on training data collected by the measurement platform under multiple environmental conditions and at different angles. The second processing unit trains and evaluates multiple machine learning algorithms, automatically selecting the optimal recognition model and measurement angle. This helps adapt to complex and changing fire smoke and interference source environments, significantly enhancing the system's intelligence and adaptability. Overall, the detector boasts enhanced environmental adaptability, sensitivity, and discrimination capabilities, enabling more reliable and accurate early fire warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of a fire smoke detector according to one embodiment; Figure 2 A schematic diagram of a detection principle of a test platform in one embodiment; Figure 3 1 is a flow chart of a fire smoke detection method applied to a fire smoke detector in one embodiment; Figure 4 1 is a flow chart of a fire and smoke detection method applied to a measurement platform in one embodiment; Figure 5 This is a schematic diagram showing the comparison of the accuracy of the binary classification of fire smoke and interference sources in one embodiment.

[0019] Description of the drawings: fire smoke detector housing 1, first light source module 10, LED light source 2, collimator 3, beam expander 4, lens 5, first receiving module 20, spectrum sensor chip 6, aperture 7, second light source module 30, second receiving module 40. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0021] The following describes in detail the implementation details of the technical solutions of the embodiments of the present application.

[0022] Figure 1 The fire smoke detector is mainly composed of a first light source module 10, an optical cavity, a first receiving module 20 and a first processing unit. Figure 1 And different embodiments, the working principle of the fire smoke detector is described in detail.

[0023] The first light source module 10 emits a parallel beam of light toward the aerosol to be measured. This parallel beam has a substantially consistent direction and a small divergence angle, ensuring that aerosol particles at different spatial locations receive incident light from the same direction. This ensures that the subsequent scattered signals are angularly comparable and analytically meaningful. The aerosol to be measured refers to the suspended aerosol particles in the area to be measured, which are the detection targets of fire smoke detectors. In actual applications, these aerosols may be smoke particles generated in the early stages of a fire, or they may contain other interference sources such as non-fire particles such as steam and dust.

[0024] To enhance the differences in scattering directions between different aerosol particles and improve the ability of subsequent algorithms to distinguish these different particles, fire smoke detectors select specific scattering angles for spectral signal acquisition. To this end, the optical cavity is designed to precisely adjust the angle between the incident light beam and the detector's receiving path, limiting the angle between the incident optical axis and the detector optical axis to a preset target angle. This target angle is the specific angle used by fire delay detectors to collect scattered spectral signals.

[0025] The first receiving module 20 is responsible for acquiring scattered spectrum signals of multiple discrete channels at a target angle and transmitting the acquired scattered spectrum signals to the first processing unit.

[0026] The first processing unit has a built-in target algorithm, which is a pre-trained machine learning model. The target algorithm uses the collected multi-channel scattered spectrum signal as input, extracts the spectral characteristics and angular response characteristics, and combines the discrimination boundary learned during the training process to accurately classify fire smoke and common interference sources (such as water vapor, dust, etc.), thereby outputting the corresponding fire probability. .

[0027] It's important to note that this target algorithm is a classification algorithm. It analyzes the input multi-channel scattering spectrum signal to determine whether the aerosol being measured is fire smoke or an interference, and outputs the corresponding classification probability. This classification probability essentially represents the probability that the aerosol being measured is fire smoke and is often used as an important indicator of fire, also known as the fire probability.

[0028] In one embodiment, the first processing unit is further configured to perform normalization processing on the scattered spectrum signals corresponding to the preset multiple discrete channels. For example, taking 8 discrete channels as an example, each channel outputs an original spectrum intensity value, and the first receiving module 20 synchronously collects the scattered spectrum intensity signals of the multiple channels at the target angle and records them as to .

[0029] In order to reduce the impact of overall light intensity fluctuations or ambient light changes on subsequent recognition and judgment, the first processing unit performs channel normalization processing on the scattered spectrum intensity signal of each channel. The calculation method is:

[0030] in, Represents the normalized The value of each channel indicates the proportion of the channel to the overall light intensity; Indicates the The scattered spectrum intensity signal of each channel; Represents the sum of the scattered spectrum intensity signals of all channels.

[0031] The normalized structure forms a scattering feature vector , which is used to eliminate the absolute intensity difference and retain only the relative distribution characteristics between different bands. The scattering feature vector is used as the input of the target algorithm for further fire probability prediction.

[0032] In one embodiment, in order to achieve intelligent response to fire conditions of different levels, the first processing unit makes a graded judgment on the fire probability and controls different alarm strategies based on set thresholds.

[0033] Specifically, when the calculated fire probability value When the value exceeds the first set threshold (such as 0.94), it is considered that there is a high fire risk. The first processing unit immediately drives the sound and light alarm device to send an alarm signal, and synchronously uploads the alarm information to the cloud platform through the wireless communication module, so that the remote monitoring center or user terminal can receive fire notification in real time and take emergency measures in time.

[0034] If the fire probability value If the value falls between the first set threshold (e.g., 0.94) and the second set threshold (e.g., 0.86), the current situation is identified as a potential risk. Instead of triggering the high-priority audible and visual alarms, the first processing unit controls a local buzzer to sound a warning, alerting nearby personnel to the potential abnormality. This hierarchical alarm mechanism allows the detector to balance false alarm suppression with timely response, preventing frequent remote alarm triggering due to minor interference while enabling a rapid response to suspicious smoke conditions, thus enhancing the practicality and intelligence of early fire warnings.

[0035] In one embodiment, Figure 3 As shown, Figure 3 A flow chart of a fire smoke detection method applied to a fire smoke detector is shown, which may include the following steps: Step S101 : controlling the first light source module to emit a light beam to illuminate the aerosol to be measured, and collecting scattered spectrum signals of the aerosol to be measured in multiple discrete channels at a target angle based on the first receiving module.

[0036] The first light source module 10 emits a light beam at a set time interval and target angle, and irradiates the aerosol particles in the current air environment to be measured. The irradiated particles generate scattered light at different angles. The angle between the incident light axis of the first light source module 10 and the detector optical axis is fixed at the target angle, that is, the angle between the first light source module 10 and the first receiving module 20 is the target angle. The first receiving module 20 collects the scattered light intensity under multiple discrete channels at the target angle to form a corresponding multi-channel scattered light spectrum signal.

[0037] Step S102 : normalize the scattered spectrum signal, input the normalized scattered spectrum signal into the target algorithm for processing, and output the fire probability.

[0038] Since the scattering spectrum data collected by the first receiving module 20 is easily interfered by non-specific factors such as light intensity fluctuations and particle concentration changes, the signals of each channel are normalized before data processing. Assume that the scattering spectrum signal of each discrete channel is , then the normalized signal is expressed as:

[0039] in, This normalization method can weaken the impact of overall intensity changes, allowing subsequent algorithm processing to focus more on the relative scattering characteristics between channels, thereby enhancing the ability to identify different aerosol types.

[0040] After the normalization process is completed, the normalized scattered spectrum signal is input into the pre-trained target algorithm as a feature vector. The target algorithm outputs the fire probability value corresponding to the current sampling data.

[0041] Step S103: If the fire probability is greater than the first set threshold, an audible and visual alarm is triggered and pushed to the cloud.

[0042] Based on the comparison between the fire probability and the set threshold, the corresponding alarm operation is executed. If the fire probability is greater than the first set threshold (such as 0.94), the sound and light alarm device is immediately triggered, and the alarm information is pushed to the cloud platform via the wireless communication module, enabling remote alarm and event recording.

[0043] Step S104: If the fire probability is less than the first set threshold and greater than the second set threshold, a local buzzer alarm is triggered.

[0044] If the fire probability is between the first set threshold (such as 0.94) and the second set threshold (such as 0.86), only the local buzzer will be triggered to issue an early warning, reminding the user that there is a fire risk but it has not yet reached the alarm level.

[0045] The fire smoke detection method of this embodiment uses multi-channel scattered spectrum signals collected at the target angle and introduces normalization processing, which can effectively eliminate the influence of factors such as light source fluctuations, device errors or aerosol concentration changes on the feature extraction results, thereby improving the consistency of input data and the stability of model judgment; based on the target algorithm to output the fire probability and set a multi-level response mechanism, not only can fast and accurate identification and early warning be achieved in the early stages of the fire, but also low-intensity local alarms can be provided when there is a certain degree of uncertainty, reminding users to pay attention to potential risks, avoiding both missed reports and false alarms, and realizing a more flexible and intelligent fire response strategy.

[0046] Through the above design, the fire smoke detector achieves high-precision identification of smoke and interference sources while using only a single target angle structure, which not only effectively improves the accuracy of fire detection, but also effectively avoids the high cost and large volume problems brought by multi-angle systems.

[0047] It's understandable that the target angle, a key parameter for collecting scattered spectrum signals in fire smoke detectors, directly impacts signal distinguishability and the classification performance of the recognition algorithm, ultimately determining the accuracy of fire detection. To this end, this embodiment utilizes a turntable-type multi-angle testing system built on a measurement platform. Classification performance at different angles is quantitatively evaluated to determine the optimal detection angle.

[0048] Figure 2 The schematic diagram of the detection principle of the test platform is shown in FIG. Figure 2 As shown, the test platform includes a second light source, a second receiving module 40 mounted on an optical motion turntable, and a second processing unit for algorithm training and evaluation. The second light source module 30 is used to output a stable and repeatable test light source to ensure consistent incident light conditions during training.

[0049] The optical motion dial rotates counterclockwise within a preset test angle range in set increments (e.g., 5°), enabling the second receiving module 40 to acquire test light signals at different receiving angles. Specifically, as the dial rotates, the second receiving module 40 collects scattered light spectral signals from multiple discrete channels under various ambient aerosols at multiple scattering angles, thereby constructing a dataset rich in characteristic information.

[0050] The second processing unit trains multiple pre-set machine learning algorithms based on the constructed dataset. After training, the recognition accuracy of the different algorithms at various test angles is compared to identify the target angle with the best performance in the binary classification task of fire smoke and interference sources. In practical applications, continuous experimentation ultimately confirmed that 75° is the optimal detection angle. The resulting target angle will be incorporated into the optical path design of fire smoke detectors to achieve high-precision fire identification under single-angle conditions.

[0051] The measurement platform in this embodiment utilizes a second receiving module 40 mounted on an optical motion turntable to collect scattered spectral signals under different aerosol conditions at multiple discrete test angles. This allows for the construction of a comprehensive training dataset covering multiple angles and environments, effectively improving the machine learning model's ability to distinguish between fire smoke and non-fire interference sources. Simultaneously, the second processing unit trains and evaluates multiple preset algorithm models, combining recognition accuracy at different angles to ultimately select the optimal target algorithm and target angle. This helps improve the recognition accuracy and stability of fire smoke detectors under actual deployment conditions, reducing false alarm and missed alarm rates. This provides data support and model assurance for the fire smoke detector's detection angles and deployed learning algorithms, enabling highly reliable intelligent fire identification.

[0052] In one embodiment, the measurement platform is not only used to collect scattered spectrum signals, but also used to evaluate the algorithm effects among multiple candidate machine learning algorithms based on training data, and determine the target algorithm accordingly.

[0053] In actual use, the measurement platform collects multiple sets of scattered spectrum signals, including fire smoke samples and interference source samples, through the second receiving module 40. The second processing unit standardizes the collected data sets according to a unified data preprocessing process. Subsequently, multiple pre-set machine learning algorithms are trained and tested based on the same training and validation sets.

[0054] The second processing unit compares and evaluates the performance of each candidate algorithm in the binary classification recognition task of fire smoke and interference sources. The evaluation indicators include but are not limited to accuracy, recall rate, F1 value and model inference efficiency. Finally, based on the comprehensive performance of each algorithm on multiple indicators, the algorithm that is most suitable for the current acquisition environment and recognition target is determined as the target algorithm. In actual application, after continuous experiments, it was finally verified that the random forest algorithm is the optimal classification model. The random forest algorithm contains 50 decision trees, each tree has a maximum depth of 5 layers, and the model inference delay is less than 3 milliseconds. The target algorithm is parameterized and recorded or loaded into the fire smoke detector for classification and discrimination in the subsequent actual operation process.

[0055] In one embodiment, the wavelengths of multiple discrete channels can be set to 415nm, 445nm, 480nm, 515nm, 555nm, 590nm, 630nm and 680nm, covering the key scattering response areas in the visible light and near-infrared bands, helping to enhance the optical response differences to different types of aerosols, thereby improving the accuracy of classification and identification.

[0056] To achieve multi-angle spectral acquisition, the optical motion turntable on the measurement platform rotates counterclockwise from 40° to 85°, fully covering the characteristic variations of typical aerosols at medium forward scattering angles. To achieve a balance between angular resolution and acquisition efficiency, the turntable's step size is set to 5°, meaning that multi-channel scattering signals are collected every 5°. This allows the second receiving module 40 to acquire test scattering spectral signals at a total of 10 different scattering angles, ranging from 40° to 80°.

[0057] Multiple machine learning algorithms are preset for training and testing. The environmental aerosol samples selected in the experiment include fire smoke (such as n-heptane open flame, polyurethane open flame, smoldering wood fire, smoldering cotton rope fire, burning newspaper fire, etc.) and various non-fire interference sources (such as water mist, loess dust, cement dust, Arizona dust, cooking fumes, etc.) to simulate the complex detection background that may appear in actual application scenarios.

[0058] During operation, the measurement platform collects multi-channel scattering spectral signals at different scattering angles for the 10 different types of ambient aerosol samples described above. Specifically, for each aerosol scattering test, the second receiving module 40 acquires the corresponding 8-channel scattering spectral signals at 10 scattering angles, thereby constructing an 80-dimensional feature matrix (10 angles x 8 channels). Based on these acquisition results, a scattering spectral dataset was constructed, including five standard fire smoke samples and five typical interference sources.

[0059] The measurement platform has multiple preset machine learning algorithms, including random forest algorithm, logistic regression algorithm, support vector machine algorithm, XGBoost algorithm, K nearest neighbor algorithm and other commonly used classification models.

[0060] In one embodiment, reference Figure 2 As shown, when the optical motion turntable is in the initial state, that is, the optical motion turntable is at the 0° position, the second light source module 30 and the second receiving module 40 are located on the same horizontal line.

[0061] During the test, the optical motion dial rotates counterclockwise to the starting angle of the preset test angle range (40° for example). Without aerosol injection, the second light source module 30 is turned on, emitting a beam of light at a preset wavelength into the test space. At this point, since there are no particulate matter in the test space, the second receiving module 40 captures a background scattering spectrum signal, denoted as PB. This background signal reflects the baseline scattering level in the absence of particulate matter in the test environment.

[0062] On this basis, an ambient aerosol is introduced into the test space to ensure that the aerosol is fully distributed in the detection space. The optical motion turntable then rotates gradually within a preset test angle range (e.g., 40° to 85°) from a starting angle of 40° in set steps (e.g., 5°), completing the sampling process within the entire preset test angle range.

[0063] At each angle position, the second light source module 30 emits a light beam, and the second receiving module 40 collects the test scattering spectrum signals of multiple discrete channels (such as 8) corresponding to the angle, which are recorded as .

[0064] The second processing unit determines whether the difference between the particle scattering spectrum signal P and the background scattering spectrum signal PB is greater than 0. If the difference is not greater than 0, it indicates that the current background signal may be offset or interfered with. The measurement platform will trigger the re-collection of PB to ensure the accuracy and consistency of subsequent sampling data. If the difference is greater than 0, the next angle sampling process or subsequent normalization and feature analysis will be carried out.

[0065] After collecting scattering spectrum data for one aerosol across all angles, the test space is cleared of aerosol particles and another aerosol is introduced. The above process is repeated, starting from the preset angle starting position, to collect scattering spectrum data for this aerosol at each angle.

[0066] Based on this, in one embodiment, Figure 4 As shown, Figure 4 A flow chart of a fire and smoke detection method applied to a measurement platform is shown, which may include the following steps: Step S201 : controlling the optical motion turntable to rotate counterclockwise to a starting angle of a preset test angle range.

[0067] When the optical motion turntable is in its initial position (0°), the second light source module 30 and the second receiving module 40 are aligned horizontally, serving as a calibration reference for the measurement platform. Subsequently, the optical motion turntable is controlled to rotate counterclockwise to the starting angle of a preset test angle range, for example, 40°, which serves as the starting point for the measurement platform's test.

[0068] Step S202: Control the second light source module to emit a light beam, and collect a background scattering spectrum signal based on the second receiving module.

[0069] When the optical motion dial is adjusted to the starting angle, the second light source module 30 is controlled to emit a light beam, and the second receiving module 40 collects the background scattering spectrum signal (denoted as PB) when no aerosol is introduced into the current test space. This background signal reflects the scattering characteristics of the test space in the absence of particulate matter interference.

[0070] In step S203, different ambient aerosols are introduced into the measurement platform, the optical motion turntable is controlled to rotate counterclockwise within a preset test angle range with a set step size, and the test scattering spectrum signals of multiple discrete channels under various ambient aerosols are collected based on the second receiving module to form a data set.

[0071] After obtaining the background signal PB, various known types of ambient aerosol samples were introduced into the measurement platform in sequence, including five types of typical fire smoke (such as n-heptane open flame, polyurethane open flame, smoldering wood, smoldering cotton rope and burning newspaper) and five types of non-fire interference sources (such as water mist, loess powder, cement powder, Arizona dust and kitchen fume).

[0072] To ensure comprehensive coverage of scattering angle information, the optical motion turntable is controlled to rotate counterclockwise within a preset test angle range (e.g., 40° to 85°) at a fixed step size (e.g., every 5°), and at each angle, the second receiving module 40 collects test scattering spectrum signals under multiple discrete channels (e.g., 8 channels), which are denoted as P, to ultimately form a multidimensional training data set containing angle and channel information.

[0073] Step S204 : If the difference between the test scattering spectrum signal and the background scattering spectrum signal is not greater than 0, re-collection of the background scattering spectrum signal is triggered.

[0074] During each acquisition, the difference between the test scattering spectrum signal P and the background signal PB is compared. If the difference is not greater than 0, it indicates that there may be background drift or signal interference, triggering the re-acquisition of the background signal PB to ensure the accuracy and reliability of the training data.

[0075] In step S205, if the difference between the test scattered spectrum signal and the background scattered spectrum signal is greater than 0, the preset multiple machine learning algorithm data is trained based on the data set, and the target algorithm is determined from the preset multiple machine learning algorithms according to the recognition effect of the preset multiple machine learning algorithms in the fire smoke and interference source binary classification task; and the target angle is screened out in the preset test angle range according to the recognition accuracy of the preset multiple machine learning algorithms at different test angles.

[0076] When the difference is greater than 0, the sampling is determined to be valid and subsequent data processing begins.

[0077] In subsequent data processing, a high-dimensional feature matrix (such as 10 angles × 8 channels, forming an 80-dimensional feature vector) is constructed based on the collected training set, and is input into a variety of preset machine learning algorithms for training, including but not limited to random forest algorithm, logistic regression algorithm, support vector machine algorithm, XGBoost algorithm, K nearest neighbor algorithm, etc., to train the classification model to complete the binary classification recognition task between fire smoke and interference source.

[0078] By comparing the recognition accuracy, stability, and generalization ability of different algorithms in the binary classification task of fire smoke and interference sources, the classification algorithm with the best recognition effect is determined as the target algorithm (for example, the random forest algorithm was selected in the experiment); at the same time, based on the recognition accuracy at different scattering angles, the test angle with the best recognition effect is selected as the target angle (for example, the angle with the highest recognition accuracy is 75°).

[0079] like Figure 5 As shown, Figure 5 This figure shows a comparison of the accuracy of different machine learning algorithms for binary classification of fire smoke and interference sources at different test angles. The horizontal axis represents the test angle of the optical motion turntable, and the vertical axis represents the aerosol classification accuracy of the corresponding machine learning algorithm.

[0080] Depend on Figure 5As can be seen, the Random Forest algorithm demonstrated high classification accuracy at all test angles. Therefore, Random Forest was selected as the final classification model and integrated into the first processing unit of the fire smoke detector. Furthermore, a comparison of the classification accuracy of different machine learning algorithms at the same test angle revealed that, at a test angle of 75°, the classification accuracy of each algorithm was generally high, with Random Forest achieving the highest accuracy. Therefore, 75° was selected as the target detection angle for this invention and applied to the optical cavity design of the fire smoke detector.

[0081] Ultimately, the determined target algorithm and target angle will serve as the basis for embedding models and configuration parameters in fire smoke detectors to achieve high-precision classification and identification of unknown aerosol samples in subsequent online detection tasks.

[0082] In this embodiment, by controlling the rotation of the optical motion turntable and collecting the scattered spectral signals of different environmental aerosols at multiple angles, and performing difference judgment in combination with the background scattered spectral signals, invalid background data can be effectively eliminated and the quality of the training data set can be improved; further, the data set is trained using multiple preset machine learning algorithms, and the optimal target algorithm and target angle are screened out based on the comprehensive recognition accuracy, which helps to achieve high-precision classification and recognition of fire smoke and interference sources, significantly improve the accuracy and stability of subsequent fire smoke detection, and provide data support and decision-making basis for the angle optimization and algorithm selection of the detector.

[0083] In one embodiment, Figure 1 As shown, the first light source module 10 uses an LED light source 2 as the emission unit. This LED light source 2 offers stable and adjustable output, providing continuous spectral radiation over a wide wavelength range (e.g., 400nm to 700nm), meeting the requirements of multi-channel spectral detection. To achieve a high-quality incident beam with excellent parallelism and controllable divergence, the light emitted by the LED light source 2 is sequentially processed by a collimator 3, a beam expander 4, and a lens 5.

[0084] Among them, the collimator 3 is responsible for initially shaping the divergent light guided by the optical fiber bundle into collimated light, the beam expander 4 further adjusts the beam diameter and divergence angle to improve the spatial coverage of the beam, and the lens 5 is used to converge and correct the beam to ensure that the light intensity incident on the aerosol area is uniform and the direction is stable, which helps to improve the efficiency and accuracy of obtaining subsequent scattered signals.

[0085] although Figure 2 A specific schematic diagram of the second light source module 30 is not shown in detail, but the configuration of the second light source module 30 is similar to that of the first light source module 10 .

[0086] In practice, the LED light source 2 features a front adjustment knob that allows for exponential adjustment of the output light intensity from 0 to 100%. It also accepts an included fiber bundle. This bundle is clamped in a fixed V-mount, with its output end precisely positioned behind the collimator 3 using screws. The beam expander 4, with a 3X beam expansion ratio, is secured at both ends by an auto-centering barrel. The lens 5 can be mounted directly in a standard threaded cage mounting plate.

[0087] like Figure 1 As shown, the first receiving module 20 integrates a spectral sensor chip 6 and an aperture 7 positioned in front of the spectral sensor chip 6. The spectral sensor chip 6 can collect scattered light signals at multiple wavelengths with high sensitivity. To improve the signal-to-noise ratio of the received signal and suppress stray light interference, the adjustable aperture 7 is used to limit the incident light flux. The aperture can be adjusted according to actual testing requirements to optimize angular resolution.

[0088] The aperture 7 and spectral sensor chip in the first receiving module 20 are integrally mounted on the fire smoke detector housing 1, with the receiving optical axis forming a target angle with the incident optical axis emitted by the first light source module 10. The aperture and spectral sensor chip in the second receiving module 40 are integrally mounted on an optical rotating disk, enabling spectral signal acquisition at various scattering angles as the disk moves.

[0089] It should be noted that although Figure 2 The specific schematic diagram of the second receiving module 40 is not shown in detail, but the configuration of the second receiving module 40 is similar to that of the first receiving module 20 .

[0090] In the above embodiment, the fire smoke detector introduces a target angle determined by the measurement platform so that the angle between the incident direction of the light source in the detector and the receiving direction is maintained at the optimal angle that has been optimized and screened, thereby significantly enhancing the detector's response to the scattering characteristics of different types of aerosols. At the same time, the use of a receiving module with multi-channel spectral resolution capability helps to capture richer spectral information, providing a solid data foundation for subsequent feature extraction and classification judgment. Furthermore, the processing unit can more accurately identify fire smoke and assess its probability of occurrence by calling a target algorithm trained based on a large-scale measured data set, effectively reducing the false alarm rate and missed alarm rate, and improving the reliability and practicality of the detector in complex environments. This solution not only achieves directional optimization of the optical structure, uses a single-angle detector to realize the classification of fire smoke and interference sources, effectively reduces hardware costs and reduces size, but also introduces a data-driven optimization strategy of the machine learning algorithm, taking into account the synergistic improvement of optical design and intelligent discrimination.

[0091] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0092] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0093] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A fire smoke detector, characterized in that: include: A first light source module is used to output a parallel light beam to the aerosol to be measured; an optical cavity, used to fix the angle between the incident light axis of the first light source module and the detector optical axis to a target angle; a first receiving module, configured to collect scattered spectrum signals of a plurality of discrete channels at the target angle, and output the scattered spectrum signals to a first processing unit; a first processing unit, configured to process the scattered spectrum signal based on a pre-trained target algorithm and output a fire probability; Wherein, the target angle is determined by a measuring platform; the measuring platform includes a second light source module and a second receiving module installed on an optical motion turntable; The second light source module is used to output a test light source; The optical motion turntable rotates counterclockwise within a preset test angle range at a set step length; The second receiving module is used to collect test scattering spectrum signals of multiple discrete channels under multiple environmental aerosols within a preset test angle range as the optical motion turntable rotates to form a data set; The second processing unit is used to train the preset multiple machine learning algorithm data based on the data set, and screen out the target angle in the preset test angle range according to the recognition accuracy of the preset multiple machine learning algorithms at different test angles.

2. The fire smoke detector according to claim 1, characterized in that: The target algorithm is determined by the measurement platform; The second processing unit is further used to determine the target algorithm from among the preset multiple machine learning algorithms based on the recognition effects of the preset multiple machine learning algorithms in the fire smoke and interference source binary classification task.

3. The fire smoke detector according to claim 1, characterized in that: The first processing unit is further configured to perform normalization processing on the scattering spectrum signals corresponding to a plurality of preset discrete channels.

4. The fire smoke detector according to claim 1, characterized in that: The wavelengths of the plurality of discrete channels include 415 nm, 445 nm, 480 nm, 515 nm, 555 nm, 590 nm, 630 nm, and 680 nm; The preset test angle range is 40° to 85° when the optical motion turntable rotates counterclockwise; the setting step is 5°; The environmental aerosols include fire smoke and non-fire interference sources; The preset multiple machine learning algorithms include random forest algorithm, logistic regression algorithm, support vector machine algorithm, XGBoost algorithm, and K nearest neighbor algorithm.

5. The fire smoke detector according to claim 1, characterized in that: In an initial state, the second light source module and the second processing unit are on the same horizontal line; The second light source module emits a light beam when the optical motion turntable rotates counterclockwise to a starting angle of the preset test angle range, and the second receiving module is used to collect background scattering spectrum signals; If the difference between the test scattering spectrum signal and the background scattering spectrum signal is not greater than 0, re-collection of the background scattering spectrum signal is triggered.

6. The fire smoke detector according to claim 1, characterized in that: The first light source module and the second light source module respectively include an LED light source; the LED light source sequentially outputs a parallel light beam through a collimator, a beam expander and a lens; The first receiving module and the second receiving module respectively include a spectrum sensor chip and an aperture; the front end of the spectrum sensor chip is closely adjacent to the rear end of the aperture.

7. The fire smoke detector according to claim 1, characterized in that: The first processing unit is further configured to: If the fire probability is greater than a first set threshold, an audible and visual alarm is triggered and pushed to the cloud; If the fire probability is less than the first set threshold and greater than the second set threshold, a local buzzer alarm is triggered.

8. A measuring platform, characterized in that: The measurement platform includes: A second light source module, used for outputting a test light source; A second receiving module mounted on an optical motion turntable; the optical motion turntable rotates counterclockwise within a preset test angle range at a set step size; the second receiving module is configured to collect test scattering spectral signals of multiple discrete channels under multiple ambient aerosols within the preset test angle range as the optical motion turntable rotates, thereby forming a training set; The second processing unit is used to train the preset multiple machine learning algorithm data based on the training set, determine the target algorithm according to the recognition effect of the preset multiple machine learning algorithms in the fire smoke and interference source binary classification task; and screen out the target angle within the preset test angle range according to the recognition accuracy of the preset multiple machine learning algorithms at different test angles; the target algorithm represents the algorithm model used to process the scattered spectrum signal collected by the fire smoke detector; the target angle represents the scattering angle corresponding to the fire smoke detector when collecting the scattered spectrum signal.

9. A fire smoke detection method, characterized in that: Applied to the fire smoke detector according to any one of claims 1 to 7, the method comprises: Controlling the first light source module to emit a light beam to illuminate the aerosol to be measured, and collecting scattered spectrum signals of the aerosol to be measured in multiple discrete channels at a target angle based on the first receiving module; Normalizing the scattered spectrum signal, inputting the normalized scattered spectrum signal into a target algorithm for processing, and outputting a fire probability; wherein, if the fire probability is greater than a first set threshold, triggering an audible and visual alarm and sending the information to the cloud; If the fire probability is less than the first set threshold and greater than the second set threshold, a local buzzer alarm is triggered.

10. A fire smoke detection method, characterized in that: Applied to the measurement platform of claim 8, the method comprises: Controlling the optical motion turntable to rotate counterclockwise to a starting angle of a preset test angle range; wherein, when the optical motion turntable is in an initial state, the second light source module and the second receiving module are on the same horizontal line; controlling the second light source module to emit a light beam, and collecting a background scattered spectrum signal based on the second receiving module; introducing different ambient aerosols into the measurement platform, controlling the optical motion turntable to rotate counterclockwise within the preset test angle range with a set step length, and collecting test scattering spectrum signals of multiple discrete channels under multiple ambient aerosols based on the second receiving module to form a data set; If the difference between the test scattering spectrum signal and the background scattering spectrum signal is not greater than 0, triggering re-collection of the background scattering spectrum signal; If the difference between the test scattering spectrum signal and the background scattering spectrum signal is greater than 0, the preset multiple machine learning algorithm data are trained based on the data set, and the target algorithm is determined among the preset multiple machine learning algorithms according to the recognition effect of the preset multiple machine learning algorithms in the fire smoke and interference source binary classification task; and according to the recognition accuracy of the preset multiple machine learning algorithms at different test angles, the target angle is screened out within the preset test angle range; the target algorithm represents the algorithm used to process the scattering spectrum signal collected by the fire smoke detector; the target angle represents the scattering angle at which the fire smoke detector is detecting.

Images