A method and system for detecting a flame of an explosive source

Abstract

The embodiment of the application discloses an explosion source flame detection method and system, the method comprises the following steps: collecting multi-modal signals of the explosion source flame, the multi-modal signals comprise acoustic vibration signals and optical signals; preprocessing the acoustic vibration signals and the optical signals to generate acoustic vibration feature sets and optical feature sets; using a persistent homology algorithm to perform topological analysis on the acoustic vibration feature sets to identify dynamic behavior patterns of the explosion source flame, and obtaining dynamic behavior pattern recognition results of the explosion source flame; using a symbolic dynamics algorithm to convert the dynamic behavior pattern recognition results and the optical feature sets into symbolic sequences, analyzing cross-modal symbolic patterns through a transition probability matrix, and generating comprehensive feature representations of the explosion flame; based on the comprehensive feature representations, determining whether the explosion source flame exists through a classification algorithm, and outputting a detection result. The application realizes high-precision explosion source flame detection suitable for complex environments.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flame detection, and in particular to a method and system for detecting explosion source flame. BACKGROUND

[0002] Explosion source flame has the characteristics of strong instantaneity, high energy, and interference with smoke or strong light. Traditional flame detection methods (such as infrared detection, ultraviolet detection, or YOLO algorithm based on image processing) have limitations in complex environments (such as low visibility and high noise). For example, infrared or ultraviolet detection is easily disturbed by background light, and image processing methods fail under smoke obstruction. In addition, existing methods rely on a single modality signal, making it difficult to capture the dynamic behavior pattern of explosion flame, resulting in high false detection rate or slow response speed.

[0003] Therefore, there is an urgent need for a high-precision explosion source flame detection method that can integrate multi-modal signals and adapt to complex environments. SUMMARY

[0004] The present application provides a method for detecting explosion source flame, which realizes the recognition and comprehensive feature representation of the dynamic behavior pattern of explosion flame through the phased processing of multi-modal signals (acoustic-vibration and optical), combined with the persistent homology algorithm and the symbolic dynamics algorithm, significantly improving the detection accuracy and environmental adaptability.

[0005] The present application provides the following solutions:

[0006] According to a first aspect, a method for detecting explosion source flame is provided, the method comprising: collecting multi-modal signals of explosion source flame, the multi-modal signals including acoustic-vibration signals and optical signals; preprocessing the acoustic-vibration signals and the optical signals to generate an acoustic-vibration feature set and an optical feature set; using a persistent homology algorithm to perform topological analysis on the acoustic-vibration feature set to identify the dynamic behavior pattern of explosion source flame, and obtaining a dynamic behavior pattern recognition result of explosion source flame; using a symbolic dynamics algorithm to convert the dynamic behavior pattern recognition result and the optical feature set into a symbolic sequence, analyzing the cross-modal symbolic pattern through a transition probability matrix, and generating a comprehensive feature representation of explosion flame; based on the comprehensive feature representation, determining whether there is an explosion source flame through a classification algorithm, and outputting a detection result.

[0007] According to an implementable manner in the embodiments of the present application, the pre-processing of the acoustic vibration type signal and the optical type signal to generate an acoustic vibration feature set and an optical feature set comprises: performing multi-scale decomposition on the acoustic wave signal and the micro-vibration signal by using a discrete wavelet transform, extracting time-frequency features at each scale, and removing environmental noise interference by adaptive threshold filtering to generate the acoustic vibration feature set; and performing frequency distribution analysis on the multi-spectral signal by using a fast Fourier transform, extracting spectral line features in a wavelength range of 400 nanometers to 2500 nanometers, and generating the optical feature set.

[0008] According to an implementable manner in the embodiments of the present application, the topological analysis of the acoustic vibration feature set by using a persistent homology algorithm to identify a dynamic behavior mode of an explosion source flame comprises: mapping the acoustic vibration feature set to a high-dimensional point cloud representation, constructing a persistent homology barcode, and extracting topological features of the dynamic behavior mode of the explosion flame, the topological features comprising a duration of 0-dimensional connected components and 1-dimensional ring structures.

[0009] According to an implementable manner in the embodiments of the present application, the conversion of the dynamic behavior mode identification result and the optical feature set into a symbolic sequence by using a symbolic dynamics algorithm, the analysis of a cross-modal symbolic mode by using a transition probability matrix, and the generation of a comprehensive feature representation of the explosion flame comprise: discretizing the dynamic behavior mode identification result and the optical feature set into a finite symbolic sequence respectively, and constructing a joint transition probability matrix; calculating a conditional entropy of the cross-modal symbolic sequence, and generating the comprehensive feature representation of the explosion flame, wherein the conditional entropy is used to quantify the relevance of the two types of signals in the instantaneous dynamics of the explosion flame.

[0010] According to an implementable manner in the embodiments of the present application, the discretization of the dynamic behavior mode identification result and the optical feature set into a finite symbolic sequence respectively, and the construction of a joint transition probability matrix comprise: discretizing the dynamic behavior mode identification result and the optical feature set into a finite symbolic sequence by using an adaptive partition coding method, wherein the adaptive partition coding dynamically adjusts a symbol division threshold according to the instantaneous intensity of the explosion flame.

[0011] According to an implementable manner in the embodiments of the present application, the judgment of whether there is an explosion source flame by using a classification algorithm comprises: performing sparse coding on the comprehensive feature representation by using a sparse representation based classifier, constructing a shared dictionary of explosion flame features, and judging whether there is an explosion source flame by using a dictionary reconstruction error.

[0012] According to an implementable manner in the embodiments of the present application, the determining whether the explosion source flame exists by the classification algorithm comprises: adopting a classification method based on rough set theory, performing attribute reduction on the comprehensive feature representation, extracting a core feature subset of the explosion flame, and determining whether the explosion source flame exists by a rough set decision rule, wherein the decision rule is adapted to the detection requirement in a high-noise environment by dynamic updating.

[0013] According to a second aspect, an explosion source flame detection system is provided, comprising: a signal acquisition unit configured to acquire multi-modal signals of an explosion source flame, the multi-modal signals comprising acoustic vibration type signals and optical type signals; a feature set generation unit configured to pre-process the acoustic vibration type signals and the optical type signals, and generate an acoustic vibration feature set and an optical feature set; a pattern recognition unit configured to perform topological analysis on the acoustic vibration feature set by using a persistent homology algorithm to identify a dynamic behavior pattern of the explosion source flame, and obtain a dynamic behavior pattern recognition result of the explosion source flame; a comprehensive feature generation unit configured to convert the dynamic behavior pattern recognition result and the optical feature set into a symbolic sequence by using a symbolic dynamics algorithm, analyze a cross-modal symbolic pattern by using a transition probability matrix, and generate a comprehensive feature representation of the explosion flame; and a detection result generation unit configured to determine whether the explosion source flame exists by a classification algorithm based on the comprehensive feature representation, and output a detection result.

[0014] According to a third aspect, a computer readable storage medium is provided, which stores a computer program, and the program is executed by a processor to implement the steps of the method of any one of the first aspect.

[0015] According to a fourth aspect, an electronic device is provided, comprising: one or more processors; and a memory associated with the one or more processors, the memory being configured to store program instructions, the program instructions being configured to perform the steps of the method of any one of the first aspect when executed by the one or more processors.

[0016] According to the specific embodiments provided in the present application, the following technical effects are disclosed in the present application:

[0017] The application realizes high-precision and real-time detection through multi-modal signal fusion. The method combines acoustic-vibration signals and optical signals, uses persistent homology algorithm to identify the dynamic behavior pattern of explosion flame, captures its nonlinear topological characteristics, and then converts the two types of signals into symbolic sequences through symbolic dynamics algorithm, analyzes the cross-modal correlation based on transition probability matrix, generates comprehensive feature representation, and finally determines the existence of flame through classification algorithm. The method breaks through the limitations of traditional single-mode detection and significantly improves the robustness in low-visibility or high-interference environment. The adaptive partition coding and conditional entropy calculation of symbolic dynamics enhance the dynamic adaptability of feature fusion, which is suitable for the instantaneousness and complexity of explosion flame.

[0018] Of course, implementing any product of the present application does not necessarily require all the advantages described above to be achieved at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0020] Figure 1 System architecture diagram suitable for embodiments of the present application;

[0021] Figure 2 Flowchart of an explosion source flame detection method provided by the embodiments of the present application;

[0022] Figure 3 Structural block diagram of an explosion source flame detection system provided by the embodiments of the present application;

[0023] Figure 4 Schematic block diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0025] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0026] It should be understood that the term "and / or" used herein is merely an association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent three cases of A existing alone, A and B existing together, and B existing alone. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after it.

[0027] Depending on the context, the word "if" as used herein can be interpreted as meaning "when" or "upon" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if it is determined" or "if (a stated condition or event) is detected" can be interpreted as meaning "when it is determined" or "in response to determining" or "when (a stated condition or event) is detected" or "in response to detecting (a stated condition or event)".

[0028] In recent years, multi-modal signal fusion technology (such as infrared and visible light combination) has been applied to flame detection to some extent, but there are still the following problems: the fusion method is mostly simple feature splicing or linear weighting, which is difficult to handle the nonlinear dynamic characteristics of explosion flame; the use of acoustic or vibration signals is insufficient, which limits the detection ability in low visibility environment; there is a lack of adaptive processing mechanism for the instantaneous nature of explosion flame.

[0029] In view of this, the present application provides a new idea. In order to facilitate the understanding of the present application, first, the system architecture based on the present application is described. Figure 1 An exemplary system architecture to which embodiments of the present application can be applied is shown, as shown in Figure 1 The system architecture can include a user device and an explosion source flame detection system located at the server end.

[0030] The user can input multi-modal signals through the user device, and the user device sends them to the explosion source flame detection system at the server end. The explosion source flame detection system can use the method provided in the embodiments of the present application to obtain the detection result. The server end can send the detection result to the user terminal, and the user terminal uses the detection result for subsequent operation.

[0031] Among them, the user device can include but is not limited to, such as: smart mobile terminal, smart home device, wearable device, PC (Personal Computer, personal computer), etc. Among them, the smart mobile device can include, such as mobile phone, tablet computer, notebook computer, PDA (Personal Digital Assistant, personal digital assistant), Internet car, etc. The smart home device can include smart TV, smart refrigerator, etc. The wearable device can include, such as smart watch, smart glasses, virtual reality device, augmented reality device, mixed reality device, etc.

[0032] The explosion source flame detection system can be set as an independent server, a server group, or a cloud server. The cloud server, also known as a cloud computing server or a cloud host, is a host product in the cloud computing service system, which solves the defects of large management difficulty and weak service scalability in traditional physical hosts and virtual private server (VPS, Virtual Private Server) services. In addition to the architecture shown in the figure, the explosion source flame detection system can also be set in a computer terminal with strong computing power. Figure 1

[0033] It should be understood that Figure 1 The user equipment and the explosion source flame detection system in the figure are only illustrative. According to the implementation needs, there can be any number of user equipment and explosion source flame detection systems.

[0034] Figure 2 The explosion source flame detection method flow chart provided for the embodiments of the present application. As shown in the figure, the method can include the following steps: Figure 2

[0035] Step 201: Collecting multi-modal signals of explosion source flame, the multi-modal signals including acoustic vibration signals and optical signals.

[0036] Step 202: Preprocessing the acoustic vibration signals and the optical signals to generate acoustic vibration feature set and optical feature set.

[0037] Step 203: Using persistent homology algorithm to analyze the acoustic vibration feature set to identify the dynamic behavior mode of the explosion source flame, and obtaining the dynamic behavior mode identification result of the explosion source flame.

[0038] Step 204: Using symbolic dynamics algorithm to convert the dynamic behavior mode identification result and the optical feature set into symbolic sequence, analyzing cross-modal symbolic mode through transition probability matrix, and generating comprehensive feature representation of explosion flame.

[0039] Step 205: Based on the comprehensive feature representation, determining whether there is an explosion source flame through a classification algorithm, and outputting the detection result.

[0040] ​​As can be seen from the above process, the present application realizes high-precision and real-time detection through multi-modal signal fusion. The method combines acoustic vibration signals and optical signals, uses the persistent homology algorithm to identify the dynamic behavior patterns of explosion flames, captures their nonlinear topological properties, and then converts the two types of signals into symbolic sequences through the symbolic dynamics algorithm, analyzes the cross-modal correlation based on the transition probability matrix, generates comprehensive feature representations, and finally determines the presence or absence of flames through a classification algorithm. This method breaks through the limitations of traditional single-mode detection and significantly improves the robustness in low-visibility or high-interference environments. The adaptive partition coding and conditional entropy calculation of symbolic dynamics enhance the dynamic adaptability of feature fusion, which is suitable for the instantaneousness and complexity of explosion flames.

[0041] The steps in the above process and the effects that can be further produced will be described in detail below in conjunction with embodiments. First, the step 201, i.e., "collecting multi-modal signals of explosion source flames, the multi-modal signals including acoustic vibration signals and optical signals", will be described in detail in conjunction with embodiments.

[0042] Explosion source flames have the characteristics of strong instantaneousness, high energy, and complex environmental interference, and a single signal type is difficult to fully characterize their behavior. Therefore, the present application uses a multi-modal signal acquisition strategy, combining acoustic vibration signals and optical signals, to achieve comprehensive perception of explosion flames. This multi-modal method can effectively deal with low-visibility or high-noise environments, ensuring the robustness and reliability of detection.

[0043] Acoustic vibration signals can include specific frequency acoustic signals generated by explosion flames and ground or air micro-vibration signals caused by explosion shock waves. The specific frequency acoustic signals are derived from the unique acoustic characteristics produced during the explosion flame combustion process, such as shock wave sound or sound pressure fluctuations of combustion reactions formed by the rapid expansion of high-temperature gases. These acoustic signals are usually concentrated in a specific frequency range and are significantly different from conventional flames or other environmental noise (such as wind or mechanical vibration), providing key clues for the identification of explosion flames. Micro-vibration signals are derived from the disturbance of explosion shock waves to the surrounding medium, such as ground vibration or air pressure changes. These vibration signals have high sensitivity and instantaneousness, and can provide supplementary information in scenes where optical signals are limited (such as smoke obstruction), enhancing the stability of detection.

[0044] The optical signals can include visible light, near-infrared, and mid-infrared multi-spectral signals. Explosion flames release a wide range of spectral energy during the combustion process, covering various features from visible light to infrared bands. Visible light signals can capture the shape and color features of the flame, such as bright white or orange-red light, which are closely related to the high temperature characteristics of explosion flames. Near-infrared and mid-infrared signals can reflect the thermal radiation characteristics of the flame, especially in smog or dust environments, where the penetration ability of mid-infrared signals is stronger, effectively capturing the obscured flame spectral features. Through the comprehensive collection of multi-spectral signals, the invention can extract wavelength-specific features of explosion flames, thereby improving the discrimination of detection.

[0045] The collection of multi-modal signals is achieved by deploying high-sensitivity sensor arrays, such as high-frequency acoustic wave sensors for capturing acoustic signals, micro-vibration sensors for detecting ground or air vibrations, and multi-spectral cameras for acquiring optical signals. These sensors work together to form a multi-dimensional representation of explosion flames.

[0046] The above step 202, i.e., "preprocessing the acoustic vibration signals and the optical signals to generate acoustic vibration feature sets and optical feature sets," is described in detail below in conjunction with an embodiment.

[0047] The preprocessing of acoustic vibration signals and optical signals to generate acoustic vibration feature sets and optical feature sets can extract features that can represent explosion flame characteristics from raw multi-modal signals, providing high-quality input for subsequent dynamic behavior pattern recognition and comprehensive analysis. This preprocessing process converts complex acoustic vibration signals and optical signals into structured feature sets through specific signal processing techniques, to adapt to the instantaneous, nonlinear, and complex environmental interference characteristics of explosion flames. The preprocessing step can apply time-frequency analysis and spectral analysis techniques to acoustic vibration signals and optical signals respectively, extract key features and remove environmental noise interference, to generate acoustic vibration feature sets and optical feature sets.

[0048] As an implementable way, preprocessing the acoustic vibration signals and the optical signals to generate acoustic vibration feature sets and optical feature sets includes: using discrete wavelet transform method to perform multi-scale decomposition on acoustic wave signals and micro-vibration signals, extracting time-frequency features at each scale, and removing environmental noise interference through adaptive threshold filtering to generate acoustic vibration feature sets; using fast Fourier transform to analyze the frequency distribution of multi-spectral signals, extracting spectral line features in the wavelength range of 400 nanometers to 2500 nanometers, and generating optical feature sets.

[0049] The preprocessing of acoustic vibration class signals uses a discrete wavelet transform method to perform multi-scale decomposition on acoustic signals and micro-vibration signals to extract time-frequency features at each scale. Acoustic signals of explosion flames usually contain instantaneous shock wave sounds of specific frequencies, and micro-vibration signals reflect ground or air disturbances caused by explosion shock waves. These signals have non-stationary and high dynamic characteristics. Discrete wavelet transform can accurately capture these transient changes, such as frequency peaks of acoustic waves or energy distribution of vibrations, by decomposing signals into sub-bands of different frequencies and time scales. To further improve feature quality, adaptive threshold filtering technology is introduced in the preprocessing process, which dynamically adjusts the filtering threshold according to the energy level of the signal, effectively removing environmental noise interference such as wind noise or mechanical vibration, thereby generating a high-quality acoustic vibration feature set. This method is more adaptable to complex noise environments in explosion scenes compared to traditional filtering techniques.

[0050] For the preprocessing of optical signals, a fast Fourier transform method is used to analyze the frequency distribution of multi-spectral signals in the visible light (400-700 nanometers), near-infrared (700-1100 nanometers), and mid-infrared (1100-2500 nanometers) wavelength bands, and to extract spectral line features in the 400 nanometer to 2500 nanometer wavelength range. This range covers typical spectral characteristics of explosion flames, such as visible light characteristics of high-temperature combustion and infrared characteristics of thermal radiation. The spectral signal of explosion flame contains unique wavelength characteristics, such as 780 nm or 1600 nm spectral lines produced by high-temperature combustion, which reflect the thermal radiation and chemical composition of the flame in different wavelength bands. Fast Fourier transform accurately extracts these spectral line features by converting the spectral signal from the time domain to the frequency domain, generating an optical feature set. Compared with traditional spectral analysis methods, this method focuses on the wide spectral characteristics of explosion flames and can maintain the effectiveness of features under smoke or strong light interference.

[0051] The following describes the step 203, i.e., "topological analysis of the acoustic vibration feature set using the persistent homology algorithm to identify the dynamic behavior pattern of the explosion source flame, to obtain the dynamic behavior pattern recognition result of the explosion source flame", in detail in combination with embodiments.

[0052] Acoustic vibration signals of explosion flames usually contain instantaneous shock wave sounds and ground or air micro-vibrations, with high nonlinearity and non-stationarity, making it difficult for traditional time-frequency analysis to fully characterize their complex dynamics. Persistent homology algorithm, as a topological data analysis tool, can reveal the topological structure of signals and is suitable for processing the dynamic characteristics of explosion flames, significantly improving the discrimination and robustness of detection.

[0053] Specifically, the acoustic-vibration feature set consists of time-frequency features of acoustic signals and energy distribution of micro-vibration signals, which reflect the instantaneous shock wave and vibration modes of explosion flame. The persistent homology algorithm first maps these feature sets to a high-dimensional point cloud representation, forming an abstract geometric structure where each point corresponds to a data point in the feature set, such as the frequency peak of acoustic waves or the amplitude of vibration. By constructing the RIPS complex of the point cloud, the algorithm generates a series of topological structures varying with scale, records the persistence of these structures at different scales, and generates the so-called persistent homology barcode. Topological features in the barcode include 0-dimensional connected components and 1-dimensional loop structures, representing the connectivity and periodicity patterns of the signal, respectively. For example, 0-dimensional connected components can reflect the instantaneous propagation range of explosion shock waves, and 1-dimensional loop structures may correspond to periodic fluctuations of acoustic waves or vibrations.

[0054] The duration information of the persistent homology barcode is used to quantify the dynamic behavior patterns of explosion flame. For example, the acoustic signal of explosion flame may exhibit a short high-intensity peak in a certain frequency range, which can be captured by the algorithm through a shorter 0-dimensional connected component duration; the micro-vibration signal may contain periodic shock wave patterns, characterized by the duration of 1-dimensional loop structures. These topological features can effectively distinguish the dynamic patterns of explosion flame from regular flames or other interference sources (such as mechanical vibration or environmental noise), generating dynamic behavior pattern recognition results and providing high-discrimination inputs for subsequent steps.

[0055] The acoustic-vibration feature set consists of time-frequency features of acoustic signals and energy distribution of micro-vibration signals, represented as a set of feature vectors where each is a dimensional feature vector containing time-frequency features such as frequency peak, energy intensity, or vibration amplitude. First, the feature set is standardized to eliminate dimensional differences and enhance the stability of topological analysis. The standardization formula is:

[0056] (1)

[0057] where is the mean of each dimension of the feature set, is the standard deviation. The standardized feature set forms a high-dimensional point cloud representation, where each point represents a feature vector of the explosion flame acoustic-vibration signal, mapped to space, forming a point cloud structure.

[0058] Based on the point cloud , the RIPS complex is constructed to capture the topological structure. The RIPS complex is a distance-based simplex complex, constructed by a scale parameter Definition. For any pair of points in the point cloud , the Euclidean distance is calculated:

[0059] (2)

[0060] When , the points and are connected to form an edge, constructing 0-dimensional simplices (points), 1-dimensional simplices (edges), and higher-dimensional simplices. The RIPS complex is defined as:

[0061] (3)

[0062] By gradually increasing from 0 to the maximum value (such as the maximum distance of the point cloud), a series of RIPS complexes are generated, forming a filtering process .

[0063] For the filtered , the persistent homology is calculated, extracting topological features. Persistent homology describes the topological structure of the point cloud through homology groups , where represents connected components, and represents loop structures. For each dimensional homology group, its generator is calculated with the birth and death times of :

[0064] Birth time : the topological feature (such as a connected component or a loop) appears at .

[0065] Death time : the topological feature disappears at . The persistent homology barcode is composed of all the birth-death pairs of topological features , represented as:

[0066] (4)

[0067] where is the index set of dimensional topological features. The 0-dimensional barcode represents the duration of connected components, reflecting the instantaneous propagation range of the blast shock wave; the 1-dimensional barcode represents the duration of loop structures, representing the periodic pattern of sound waves or vibrations.

[0068] From the persistent homology barcode, the dynamic behavior pattern features of the blast flame are extracted. For 0-dimensional connected components, the duration is calculated:

[0069] (5)

[0070] The long-duration 0D features reflect how the instantaneous blast wave of the explosion flame quickly connects the point cloud, characterizing its propagation range.

[0071] For 1D ring structures, the duration is calculated as:

[0072] (6)

[0073] The 1D feature duration reflects the periodic dynamics of the explosion flame, such as frequency fluctuations of acoustic signals or periodic impacts of micro-vibrations. The 0D and 1D durations are combined into a topological feature vector:

[0074] (7)

[0075] This vector characterizes the dynamic behavior patterns of the explosion flame, such as the instantaneous propagation of the blast wave and the periodic vibration characteristics.

[0076] The topological feature vector is post-processed to filter out features with durations exceeding a threshold to remove noise effects:

[0077] (8)

[0078] where is dynamically set according to the signal intensity dynamics of the explosion flame, such as based on acoustic peak values or vibration amplitudes. The final output dynamic behavior pattern recognition result is , which is used for subsequent symbolic dynamics analysis and comprehensive feature representation.

[0079] The following describes step 204, "adopting a symbolic dynamics algorithm to convert the dynamic behavior pattern recognition result and the optical feature set into a symbolic sequence, analyzing the cross-modal symbolic pattern through a transition probability matrix, and generating a comprehensive feature representation of the explosion flame", in detail with examples.

[0080] This step uses a symbolic dynamics algorithm to nonlinearly fuse the dynamic behavior pattern recognition result of acoustic vibration signals and the feature set of optical signals, generating a comprehensive feature representation that can fully characterize the characteristics of explosion flames. Explosion flames have strong instantaneousness, nonlinear dynamics, and high environmental interference characteristics, making it difficult for traditional feature fusion methods such as linear concatenation or weighted averaging to effectively capture their complex patterns. Symbolic dynamics provides a unique and efficient feature fusion method by discretizing continuous signals into symbolic sequences and analyzing their dynamic transition patterns, making it suitable for the instantaneousness and complexity of explosion flames.

[0081] Specifically, the dynamic behavior pattern recognition results are derived from the topological analysis of the acoustic vibration feature set by persistent homology algorithm, containing the duration of 0-dimensional connected components and 1-dimensional loop structures, representing the instantaneous shock wave propagation and periodic vibration patterns of the explosion flame. The optical feature set contains spectral line features in visible, near-infrared, and mid-infrared bands, reflecting the thermal radiation and chemical composition characteristics of the explosion flame. The symbolic dynamics algorithm first converts these two types of features into symbolic sequences, discretizes the continuous feature vectors into a finite set of symbols by partition encoding, for example, dividing the feature values into several intervals, each interval corresponds to a symbol such as A, B, C. This discretization process simplifies high-dimensional features into discrete symbolic sequences, retains the dynamic patterns of explosion flame, and reduces computational complexity, suitable for real-time detection requirements.

[0082] After generating the symbolic sequences, the cross-modal symbolic patterns are analyzed by constructing a joint transition probability matrix. The joint transition probability matrix records the transition probabilities between acoustic vibration symbolic sequences and optical symbolic sequences, for example, the probability of transitioning from acoustic vibration symbol A to optical symbol B. This matrix captures the dynamic association of the two types of signals in time or feature space, reflecting the instantaneous behavior of the explosion flame, such as the synchronicity of shock waves and spectral peaks. Through the Markov model based on information entropy optimization, the conditional entropy of cross-modal symbolic sequences is calculated, quantifying the correlation between acoustic vibration and optical features. The conditional entropy value, as a core part of the comprehensive feature representation, can effectively distinguish explosion flames from other flame types or environmental disturbances, ensuring high discrimination of the feature representation.

[0083] As an implementable way, the dynamic behavior pattern recognition results and the optical feature set are discretized into finite symbolic sequences respectively, and the joint transition probability matrix is constructed by: discretizing the dynamic behavior pattern recognition results and the optical feature set into finite symbolic sequences through an adaptive partition encoding method, wherein the adaptive partition encoding dynamically adjusts the symbol division threshold according to the instantaneous intensity of the explosion flame.

[0084] By monitoring the instantaneous intensity of the explosion flame in real time, such as the peak sound pressure or spectral amplitude, the algorithm dynamically adjusts the symbol division threshold. For example, in high-intensity explosion scenarios, the acoustic signal may exhibit rapidly changing frequency peaks, and the algorithm will increase the number of symbol intervals to capture the details; in low-visibility environments, such as smoke obstruction leading to optical signal weakening, the algorithm will reduce the threshold sensitivity to ensure the stability of the symbolic sequence. This adaptability significantly improves the method's ability to adapt to complex environments such as smoke, strong light, or high noise, outperforming traditional fixed partition encoding methods.

[0085] Preferably, the dynamic behavior pattern recognition results T' and the optical feature set S' are normalized feature vectors, representing topological features such as duration and spectral features such as intensity, respectively. The symbol set is defined as , the number of symbols Instantaneous intensity Calculated by peak pressure and spectrum amplitude weighting:

[0086] (9)

[0087] where, is the maximum amplitude of acoustic signal, is the maximum intensity of spectrum signal. Partition threshold According to Dynamic adjustment:

[0088] (10)

[0089] For eigenvalue or , assign symbol: if , then ; if , then Symbol ; if , then . Similarly, assign by same threshold, generate symbol sequence

[0090] Based on symbol sequence and , construct joint transition probability matrix M, dimension . Matrix element represents the transition probability from acoustic vibration symbol to optical symbol :

[0091] (11)

[0092] is the number of times that symbol is followed by , and is the total number of times that symbol occurs, calculated using a 10 ms sliding window to capture the instantaneous dynamic correlation of explosion flame.

[0093] Calculate the entropy value of the symbol sequence to verify the amount of information:

[0094] (12)

[0095] where, and are the symbol occurrence probabilities. If or is lower than 0.5, increase or adjust the scaling parameter such as 0.8 to 0.9 and re-encode. The final output and , for conditional entropy calculation.

[0096] The step 205, i.e., "judging whether there is an explosion source flame based on the comprehensive feature representation and outputting a detection result" will be described in detail below in combination with an embodiment.

[0097] This step uses the comprehensive feature representation generated by the previous step to distinguish the characteristics of explosion flame through a classification algorithm, determines whether it is an explosion source flame, and outputs a clear detection result. The comprehensive feature representation combines the dynamic behavior patterns of acoustic vibration signals and the spectral characteristics of optical signals, and contains the instantaneous, nonlinear, and cross-modal correlation characteristics of explosion flame. The classification algorithm distinguishes explosion flame from other types of flame or environmental interference by analyzing these characteristics, provides high-precision detection results, and is suitable for real-time monitoring of complex scenes such as chemical explosion and industrial blasting.

[0098] The comprehensive feature representation is derived from the symbolic dynamics algorithm, which captures the dynamic correlation of acoustic vibration signals and optical signals through adaptive partition coding and joint transition probability matrix, generates conditional entropy values and transition probability features. These features are represented in the form of a vector, which contains the unique patterns of explosion flame, such as the topological characteristics of instantaneous shock waves and the synchronicity of spectral peak values. The task of the classification algorithm is to map this high-dimensional feature vector to a binary or multi-class decision space to determine whether there is an explosion source flame. The output result is usually "detected explosion flame" or "not detected", and can include a confidence score to support real-time decision-making and subsequent warning systems.

[0099] In practical implementation, the classification algorithm can employ a sparse representation-based classifier. A sparse representation classifier is a machine learning method based on sparse signal decomposition, particularly suitable for handling high-dimensional, noisy data. The core idea is to represent the input feature vector as a linear combination of a predefined dictionary, where the combination coefficients are as sparse as possible, i.e., most coefficients are zero. This sparsity highlights the core patterns of the data and suppresses noise interference. In explosion source flame detection, the comprehensive feature representation includes the topological features of acoustic vibration signals and the spectral features of optical signals, exhibiting high dimensionality and nonlinearity. The sparse representation classifier constructs a shared dictionary of explosion flame features, decomposing the comprehensive feature vector into sparse coefficients, and determines whether it is an explosion flame based on the reconstruction error. The shared dictionary can be optimized using training data, containing typical feature patterns of explosion flames, such as instantaneous sound wave frequency changes or spectral intensity distributions. By calculating the reconstruction error between the comprehensive feature vector and the dictionary, the algorithm determines whether the input features match the explosion flame pattern. If the reconstruction error is below a set threshold, it is determined to be an explosion flame; otherwise, it is determined to be a non-target signal. This method is robust to high-noise environments and can effectively distinguish between explosion flames and conventional flames or environmental disturbances.

[0100] In implementation, the sparse representation classifier first constructs a shared dictionary using the training data. Each column of the dictionary, called an atom, represents a typical characteristic pattern of an explosion flame or non-flame signal. The training data includes known explosion flame feature vectors such as conditional entropy, topological duration, and spectral peaks. The dictionary can be constructed using the K-SVD algorithm, with iterative optimization ensuring that the dictionary atoms effectively represent the training data. For the input comprehensive feature vector... The classifier solves for the sparse coefficients. , making ,in It is sparse. The solution process uses the orthogonal matching pursuit algorithm, and the optimization objective is:

[0101] (13)

[0102] in, For sparsity constraints, reconstruction error Used for classification. If the error is below a set threshold, it indicates... Match the dictionary pattern of the explosion flame and determine it as an explosion flame; otherwise, determine it as a non-target signal.

[0103] The advantage in explosion flame detection is that sparse representation can effectively handle high-noise environments such as smoke or strong light interference. The shared dictionary captures the explosiveness and nonlinearity of the explosion flame through training, such as the topological pattern of the shock wave and the instantaneous peak of the spectrum, enhancing the discriminability of classification. In addition, the computational complexity of sparse representation is low, suitable for real-time detection requirements, such as completing classification within 100 milliseconds. Compared with traditional classifiers such as support vector machines, sparse representation makes more full use of the sparsity of high-dimensional data, suitable for the complex features of explosion flame.

[0104] Another implementation is a classification method based on rough set theory. Rough set theory is a mathematical tool for handling uncertainty and incomplete data, with the core idea of extracting core features from complex data through attribute reduction and decision rule generation, and building a simple classification model. In explosion source flame detection, the comprehensive feature representation contains multiple dimensions such as conditional entropy, topological features, and spectral features, which may have redundancy or noise. Rough set theory reduces redundant attributes, extracts a core feature subset, and generates dynamic decision rules to determine whether there is an explosion flame. Rough set reduces redundant information by attribute reduction to extract a core feature subset from the comprehensive feature representation, improving computational efficiency. The core feature subset may include conditional entropy values, topological features, or spectral peak values, etc. The algorithm determines whether there is an explosion flame based on these feature subsets by constructing rough set decision rules. The decision rules are dynamically updated to adapt to different noise levels in different scenarios, such as adjusting the rule threshold in a smoky environment to reduce false positives. This method is particularly suitable for handling the uncertainty of explosion flame features and complex environmental interference.

[0105] In implementation, the rough set method first organizes the comprehensive feature representation into an information table, with rows representing samples and columns representing features such as conditional entropy values, 0-dimensional connected component duration, spectral peak values, etc. The goal is to classify the label, i.e., explosion flame or non-explosion flame. Attribute reduction removes redundant features that are irrelevant to classification through discernibility matrix or information entropy calculation, retaining a core feature subset. For example, conditional entropy values may be crucial in distinguishing explosion flames, while certain spectral features may be redundant. After reduction, decision rules are generated based on the core feature subset, in the form of:

[0106] (14)

[0107] where, and are threshold values optimized through training data. The rules are dynamically updated to adapt to environmental changes, such as adjusting the threshold in a smoky environment to reduce false positives. During classification, the input feature vector is matched with the rules, and the detection result is output.

[0108] The advantage in explosion flame detection is that rough set theory can handle feature uncertainty and noise without assuming data distribution, suitable for complex scenarios. The dynamic rule updating mechanism enhances the adaptability of the method to high noise environments, such as dealing with smoke interference in chemical explosions. Compared with neural network methods, rough set calculation is simple and the rules are highly interpretable, suitable for embedded system applications.

[0109] The above-mentioned method provided by the embodiments of the present application can be applied to various application scenarios, including but not limited to: in chemical plant explosion monitoring, the method can quickly identify the dynamic behavior pattern of explosion flame by collecting acoustic vibration signals and multi-spectral signals, using persistent homology algorithm and symbolic dynamics fusion features, and outputting detection results in real time, effectively dealing with smoke or strong light interference, and ensuring safe production. In industrial blasting scenarios such as mine or tunnel blasting operations, the method can accurately distinguish explosion flame from environmental interference by topological analysis of sound waves and micro-vibrations combined with optical features, adapt to complex terrain and dust environment, provide reliable flame detection support, reduce false alarm rate and improve response speed. The robustness and real-time performance of the method make it significantly practical in the fields of chemical industry, mining, etc., providing an efficient solution for safety monitoring and emergency response.

[0110] The above describes specific embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different than the order in the embodiments and still achieve the desired result. In addition, the processes depicted in the drawings do not necessarily require the particular order shown or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing are also possible or advantageous.

[0111] According to another aspect, embodiments provide an explosion source flame detection system. Figure 3 A schematic block diagram of the explosion source flame detection system according to one embodiment is shown. As shown, the system 300 includes: Figure 3

[0112] A signal acquisition unit 301 configured to acquire multi-modal signals of an explosion source flame, the multi-modal signals including acoustic vibration type signals and optical type signals.

[0113] A feature set generation unit 302 configured to preprocess the acoustic vibration type signals and the optical type signals to generate an acoustic vibration feature set and an optical feature set.

[0114] A pattern recognition unit 303 configured to use a persistent homology algorithm to perform topological analysis on the acoustic vibration feature set to identify a dynamic behavior pattern of the explosion source flame, and obtain a dynamic behavior pattern recognition result of the explosion source flame. ​

[0115] The comprehensive feature generation unit 304 is configured to transform the dynamic behavior pattern recognition result and the optical feature set into symbolic sequences by using a symbolic dynamics algorithm, analyze the cross-modal symbolic pattern by using a transition probability matrix, and generate a comprehensive feature representation of the explosion flame.

[0116] The detection result generation unit 305 is configured to determine whether there is an explosion source flame by using a classification algorithm based on the comprehensive feature representation, and output a detection result.

[0117] As an implementable manner, the feature set generation unit 302 can be configured to perform multi-scale decomposition on the acoustic vibration signal and the micro-vibration signal by using a discrete wavelet transform, extract time-frequency features at each scale, and remove environmental noise interference by using an adaptive threshold filtering to generate an acoustic vibration feature set when pre-processing the acoustic vibration signal and the optical signal to generate the acoustic vibration feature set and the optical feature set.

[0118] As an implementable manner, the pattern recognition unit 303 can be configured to map the acoustic vibration feature set to a high-dimensional point cloud representation, construct a persistent homology barcode, and extract topological features of the dynamic behavior pattern of the explosion flame, including the duration of 0-dimensional connected components and 1-dimensional ring structures when performing topological analysis on the acoustic vibration feature set by using a persistent homology algorithm to identify the dynamic behavior pattern of the explosion source flame.

[0119] As an implementable manner, the comprehensive feature generation unit 304 can be configured to discretize the dynamic behavior pattern recognition result and the optical feature set into finite symbolic sequences respectively, construct a joint transition probability matrix, calculate the conditional entropy of the cross-modal symbolic sequences, and generate a comprehensive feature representation of the explosion flame when transforming the dynamic behavior pattern recognition result and the optical feature set into symbolic sequences by using a symbolic dynamics algorithm, analyzing the cross-modal symbolic pattern by using a transition probability matrix, and generating a comprehensive feature representation of the explosion flame, wherein the conditional entropy is used to quantify the correlation between the two types of signals in the instantaneous dynamics of the explosion flame.

[0120] As an implementable manner, the comprehensive feature generation unit 304 can be configured to discretize the dynamic behavior pattern recognition result and the optical feature set into finite symbolic sequences by using an adaptive partition coding method when discretizing the dynamic behavior pattern recognition result and the optical feature set into finite symbolic sequences and constructing a joint transition probability matrix, wherein the adaptive partition coding dynamically adjusts the symbol division threshold according to the instantaneous intensity dynamics of the explosion flame.

[0121] As an implementable manner, the detection result generation unit 305 can be configured to, when judging whether there is an explosion source flame by using a classification algorithm, adopt a sparse representation based classifier, perform sparse coding on the comprehensive feature representation, construct a shared dictionary of explosion flame features, and judge whether there is an explosion source flame by using a dictionary reconstruction error.

[0122] As an implementable manner, the detection result generation unit 305 can be configured to, when judging whether there is an explosion source flame by using a classification algorithm, adopt a rough set theory based classification method, perform attribute reduction on the comprehensive feature representation, extract a core feature subset of the explosion flame, and judge whether there is an explosion source flame by using a rough set decision rule, wherein the decision rule is adapted to the detection requirement in a high noise environment by dynamic updating.

[0123] Each of the embodiments in the specification is described in a progressive manner, and the same and similar parts of each of the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, it is described more simply, and the relevant part can be referred to the part of the method embodiment. The system embodiment described above is only illustrative, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0124] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0125] In addition, the embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the steps of the method in any one of the preceding method embodiments.

[0126] An electronic device, comprising:

[0127] one or more processors; and a memory associated with the one or more processors and storing program instructions that, when executed by the one or more processors, perform the steps of any of the preceding method embodiments.

[0128] The present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of any of the preceding method embodiments.

[0129] wherein, Figure 4 An exemplary architecture of the electronic device is shown, which can specifically include a processor 410, a video display adapter 411, a disk drive 412, an input / output interface 413, a network interface 414, and a memory 420. The processor 410, the video display adapter 411, the disk drive 412, the input / output interface 413, the network interface 414, and the memory 420 can be communicatively connected through a communication bus 430.

[0130] The processor 410 can be implemented in the form of a general-purpose CPU, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute related programs to implement the technical solutions provided by the present application.

[0131] The memory 420 can be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory 420 can store an operating system 421 for controlling the operation of the electronic device 400, a basic input / output system (BIOS) 422 for controlling the low-level operation of the electronic device 400. In addition, a web browser 423, a data storage management system 424, and an explosion source flame detection system 425, etc. can also be stored. The explosion source flame detection system 425 can be an application program that specifically implements the above steps in the embodiments of the present application. In summary, when the technical solutions provided by the present application are implemented by software or firmware, the related program codes are stored in the memory 420 and executed by the processor 410.

[0132] The input / output interface 413 is configured to connect an input / output module to realize information input and output. The input / output module can be configured as a component in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. The input device can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device can include a display, a speaker, a vibrator, an indicator light, etc.

[0133] The network interface 414 is configured to connect a communication module (not shown in the figure) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0134] The bus 430 includes a channel for transmitting information between various components (for example, the processor 410, the video display adapter 411, the disk drive 412, the input / output interface 413, the network interface 414, and the memory 420) of the device.

[0135] It should be noted that although the above device only shows the processor 410, the video display adapter 411, the disk drive 412, the input / output interface 413, the network interface 414, the memory 420, and the bus 430, in the specific implementation process, the device can also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device can also only contain the components necessary to implement the scheme of the present application, and does not have to contain all the components shown in the figure.

[0136] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and a general hardware platform. Based on such understanding, the technical solutions of the present application can be embodied in the form of a computer program product, which can be stored in a storage medium such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0137] The technical solutions provided by the present application are described in detail above, and the principle and implementation manner of the present application are described by applying specific examples. The above description of the embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range can be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method of flame detection of an explosive source, characterized in that The method comprises: Collecting multi-modal signals of explosion source flame, the multi-modal signals comprising acoustic vibration signals and optical signals; Preprocessing the acoustic vibration signals and the optical signals to generate an acoustic vibration feature set and an optical feature set; Performing topological analysis on the acoustic vibration feature set using a persistent homology algorithm to identify dynamic behavior patterns of the explosion source flame, to obtain a dynamic behavior pattern recognition result of the explosion source flame; Converting the dynamic behavior pattern recognition result and the optical feature set into symbolic sequences using a symbolic dynamics algorithm, analyzing cross-modal symbolic patterns through a transition probability matrix, and generating a comprehensive feature representation of the explosion flame; Based on the comprehensive feature representation, determining whether there is an explosion source flame by a classification algorithm, and outputting a detection result.

2. The explosive source flame detection method of claim 1, wherein, The preprocessing of the acoustic vibration signals and the optical signals to generate an acoustic vibration feature set and an optical feature set comprises: Performing multi-scale decomposition on acoustic signals and micro-vibration signals using discrete wavelet transform, extracting time-frequency features at each scale, and removing environmental noise interference through adaptive threshold filtering to generate an acoustic vibration feature set; Using fast Fourier transform to analyze the frequency distribution of multi-spectral signals, extracting spectral line features in the wavelength range of 400 nanometers to 2500 nanometers, and generating an optical feature set.

3. The blast source flame detection method of claim 1, wherein, The topological analysis on the acoustic vibration feature set using a persistent homology algorithm to identify dynamic behavior patterns of the explosion source flame comprises: Mapping the acoustic vibration feature set to a high-dimensional point cloud representation, constructing a persistent homology barcode, and extracting topological features of the dynamic behavior patterns of the explosion flame, the topological features including the duration of 0-dimensional connected components and 1-dimensional ring structures.

4. The blast source flame detection method of claim 1, wherein, The conversion of the dynamic behavior pattern recognition result and the optical feature set into symbolic sequences using a symbolic dynamics algorithm, the analysis of cross-modal symbolic patterns through a transition probability matrix, and the generation of a comprehensive feature representation of the explosion flame comprise: Discretizing the dynamic behavior pattern recognition result and the optical feature set into finite symbolic sequences respectively, and constructing a joint transition probability matrix; Calculating the conditional entropy of the cross-modal symbolic sequences to generate a comprehensive feature representation of the explosion flame, wherein the conditional entropy is used to quantify the relevance of the two types of signals in the instantaneous dynamics of the explosion flame.

5. The explosive source flame detection method of claim 4, wherein, The discretization of the dynamic behavior pattern recognition result and the optical feature set into finite symbolic sequences respectively, and the construction of a joint transition probability matrix comprise: Discretizing the dynamic behavior pattern recognition result and the optical feature set into finite symbolic sequences through an adaptive partition coding method, wherein the adaptive partition coding dynamically adjusts the symbol division threshold according to the instantaneous intensity dynamics of the explosion flame.

6. The explosive source flame detection method of claim 1, wherein, The determination of whether there is an explosion source flame by a classification algorithm comprises: Using a sparse representation-based classifier to perform sparse coding on the comprehensive feature representation, constructing a shared dictionary of explosion flame features, and determining whether there is an explosion source flame through dictionary reconstruction error.

7. The blast source flame detection method of claim 1, wherein, The determination of whether there is an explosion source flame by a classification algorithm comprises: The classification method based on rough set theory is used to perform attribute reduction on the comprehensive feature representation, extract a core feature subset of the explosion flame, and determine whether there is an explosion source flame through a rough set decision rule, wherein the decision rule is updated dynamically to adapt to the detection requirements in a high-noise environment.

8. An explosive source flame detection system characterized by, The system comprises: a signal acquisition unit configured to acquire multi-modal signals of the explosion source flame, the multi-modal signals comprising acoustic vibration signals and optical signals; a feature set generation unit configured to preprocess the acoustic vibration signals and the optical signals to generate an acoustic vibration feature set and an optical feature set; a pattern recognition unit configured to perform topological analysis on the acoustic vibration feature set using a persistent homology algorithm to identify a dynamic behavior pattern of the explosion source flame, and obtain a dynamic behavior pattern recognition result of the explosion source flame; a comprehensive feature generation unit configured to convert the dynamic behavior pattern recognition result and the optical feature set into a symbolic sequence using a symbolic dynamics algorithm, analyze cross-modal symbolic patterns through a transition probability matrix, and generate a comprehensive feature representation of the explosion flame; a detection result generation unit configured to determine whether there is an explosion source flame based on the comprehensive feature representation through a classification algorithm, and output a detection result.

9. An electronic device, comprising: comprise: one or more processors; and a memory associated with the one or more processors, the memory being configured to store program instructions that, when executed by the one or more processors, perform the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

Images