Fire detection methods and systems based on binocular vision and ultraviolet light

Abstract

This invention provides a fire detection method and system based on binocular vision and ultraviolet light, relating to the field of fire source localization technology. The invention continuously acquires ultraviolet and visible light images; analyzes the intrinsic properties of each pixel in both images, constructs the diagonal terms of a QUBO matrix based on the analysis results, analyzes the collaborative relationships between each pixel and other pixels, constructs the off-diagonal terms of the QUBO matrix based on the analysis results, and generates a QUBO matrix based on the diagonal and off-diagonal terms; defines the flame state of each pixel, and constructs a global energy calculation function based on the QUBO matrix and the flame state of each pixel; uses the minimization of the global energy calculation function value as the optimization objective, and identifies flame regions in the image using a quantum annealing algorithm; and performs binocular vision analysis on the identified visible light image using a binocular recognition algorithm to locate the position of the flame in space.

Description

Technical Field

[0001] This invention relates to the field of fire source location technology, specifically to a fire detection method and system based on binocular vision and ultraviolet light. Background Technology

[0002] Traditional flame recognition technologies primarily rely on visible light or infrared images, employing methods based on threshold segmentation, motion detection, or deep learning. However, these methods exhibit significant drawbacks in complex scenarios such as charging stations: visible light-based methods are susceptible to high-frequency transient interference such as electrical sparks, resulting in a high false alarm rate; while infrared solutions, although sensitive to heat sources, struggle to distinguish between the normal operating heat of charging equipment and a real flame. More critically, these traditional methods operate on a sequential "detect first, then associate" processing model, failing to achieve global collaborative optimization of pixel-level features, leading to insufficient capabilities in addressing key issues such as blurred flame boundaries and propagation path prediction. Especially when dealing with ultraviolet images, existing technologies neither effectively utilize the specific response of the ultraviolet band to flame radiation nor possess mechanisms to suppress ultraviolet-specific noise such as corona discharge, significantly limiting the practicality and reliability of traditional solutions in charging stations.

[0003] In the prior art, CN119107437B discloses a method for identifying fire sources in charging stations based on automatic tracking and positioning using machine vision. This method involves simultaneously acquiring thermal and visible light images of the target area and preprocessing them. Suspicious areas are located through threshold segmentation. Feature extraction is performed on the suspicious areas in both the thermal and visible light images to obtain flame thermal and color features. A flame visual recognition coefficient is obtained by analyzing the flame color features. Edge and internal temperature coefficients are obtained through mathematical analysis of the flame thermal features. A fire source identification coefficient is obtained through mathematical analysis of the flame visual recognition coefficient, edge temperature coefficient, and internal temperature coefficient. The fire source is identified by comparing the fire source identification coefficient with a preset threshold. However, this prior art still has shortcomings. It uses a sequential processing mode of "detect first, then associate," which cannot achieve global collaborative optimization of pixel-level features, resulting in insufficient ability to handle key issues such as blurred flame boundaries and prediction of diffusion paths.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a fire detection method and system based on binocular vision and ultraviolet light to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A fire detection method based on binocular vision and ultraviolet light, comprising the following steps:

[0008] Step 1: Deploy a binocular imaging system at the charging location to continuously acquire ultraviolet and visible light images;

[0009] Step 2: Analyze the intrinsic properties of each pixel in the ultraviolet and visible light images, and construct the diagonal terms of the QUBO matrix based on the analysis results. The intrinsic properties include radiation properties, motion properties, and scintillation properties.

[0010] Step 3: Analyze the collaborative relationship between each pixel and other pixels, construct the off-diagonal terms of the QUBO matrix based on the analysis results, and generate the QUBO matrix based on the diagonal and off-diagonal terms of the QUBO matrix.

[0011] Step 4: Define the flame state of each pixel, and construct a global energy calculation function based on the QUBO matrix and the flame state of each pixel; with the optimization objective of minimizing the value of the global energy calculation function, continuously adjust the flame state of each pixel through the quantum annealing algorithm to identify the flame region in ultraviolet and visible light images;

[0012] Step 5: Simultaneously identify the flame region in the binocular visible light image through Step 4, and perform binocular visual analysis on the identified visible light image using a binocular recognition algorithm to locate the position of the flame in space.

[0013] Furthermore, the binocular imaging system includes a binocular ultraviolet-visible dual-spectrum camera, with each eye simultaneously acquiring ultraviolet and visible light images, and mapping each pixel of the ultraviolet and visible light images acquired by the same eye one-to-one.

[0014] Furthermore, the radiation and motion attributes are obtained by analyzing ultraviolet images, and the scintillation attribute is obtained by analyzing visible light images; a self-evaluation scoring system is constructed by analyzing the scintillation, radiation, and motion attributes.

[0015] The system presets the ultraviolet radiation intensity range and ultraviolet deviation amplitude of the flame. If the ultraviolet radiation intensity of a pixel in the current ultraviolet image is within the ultraviolet radiation intensity range of the flame, the radiation attribute score of this pixel is set to 1. Otherwise, the deviation between the ultraviolet radiation intensity and the ultraviolet radiation intensity range is calculated and recorded as the intensity deviation. If the intensity deviation is not less than the ultraviolet deviation amplitude, the radiation attribute score of this pixel is set to 0. If the intensity deviation is less than the ultraviolet deviation amplitude, the radiation attribute score of this pixel is set according to the intensity deviation and the ultraviolet deviation amplitude, using the following formula:

[0016] ;

[0017] ;

[0018] in, The first in the current ultraviolet image Radiation attribute score for each pixel The first in the current ultraviolet image Intensity deviation of each pixel This represents the ultraviolet deviation amplitude. The first in the current ultraviolet image The intensity of ultraviolet radiation at each pixel The range of ultraviolet radiation intensity. This refers to the index of a pixel in the ultraviolet image;

[0019] Motion attributes are evaluated using optical flux. A preset optical flux threshold is used. If the optical flux of a pixel in the current ultraviolet image is greater than the optical flux threshold, the motion attribute score of that pixel is 1. Conversely, if the optical flux of a pixel is not greater than the optical flux threshold, the motion attribute score of that pixel is 0.

[0020] The brightness differences between adjacent frames in the current frame and the previous nine visible light images are analyzed to evaluate flicker attributes. A first brightness difference threshold and a second brightness difference threshold are preset. If the brightness difference of a pixel in the current image is not less than the second brightness difference threshold, the pixel's motion attribute score is 1; if the brightness difference is not greater than the first brightness difference threshold, the pixel's motion attribute score is 0; if the brightness difference is greater than the first brightness difference threshold but less than the second brightness difference threshold, the pixel's flicker attribute score is 0. The formula is:

[0021] ;

[0022] ;

[0023] in, For the current visible light image, the first The flicker attribute score for each pixel The first brightness difference threshold, The second brightness difference threshold, For the first Frame and the The first frame of the visible light image The brightness difference of each pixel For the first The first frame of the visible light image The brightness value of each pixel. For the first In a frame of visible light image The brightness value of each pixel. This is the index of the frame, and the indices are arranged in chronological order. The index of the current frame is 10.

[0024] The diagonal terms of the QUBO matrix are represented as follows:

[0025] in, For the QUBO matrix, the first... OK Column elements, For the ultraviolet image Motion attribute score for each pixel , and These are the weighting coefficients, and , .

[0026] Furthermore, the specific logic for calculating horizontal and vertical optical flux is as follows: The horizontal and vertical gradients of each pixel in the ultraviolet image are obtained; a spatial gradient matrix is ​​constructed from the horizontal and vertical gradients; a temporal gradient matrix is ​​calculated using the pixel values ​​of the current ultraviolet image and the pixel values ​​of the corresponding positions in the previous ultraviolet image; the horizontal and vertical optical fluxes of each pixel in the ultraviolet image are calculated based on the spatial and temporal gradient matrices; and the optical flux of each pixel is calculated based on the horizontal and vertical optical fluxes.

[0027] Furthermore, the radiation attribute score, motion attribute score, and scintillation attribute score are combined to form an attribute representation vector, denoted as: ;

[0028] Based on the representation vectors of adjacent pixels, the off-diagonal terms of the QUBO matrix are constructed using the following formula:

[0029] ;

[0030] in, For the QUBO matrix, the first... OK Column elements, for and Mahalanobis distance, For pixels and pixels distance, , and Let be the scale parameter, where , ， , This is the index of the pixel in the ultraviolet image, and .

[0031] Furthermore, the state of the flame is set to... , ,in, For the first The state of each pixel, if Then the first If each pixel is a non-flame pixel, then Then the first Each pixel represents a flame pixel;

[0032] The global energy calculation function is:

[0033] ;

[0034] in, The global energy function, For the first The state of each pixel.

[0035] The present invention further provides a fire detection system based on binocular vision and ultraviolet light, the system being used to implement the above-mentioned fire detection method based on binocular vision and ultraviolet light, specifically including:

[0036] The image acquisition module is used to deploy a binocular imaging system at the charging site to continuously acquire ultraviolet and visible light images.

[0037] The attribute analysis module is used to analyze the intrinsic attributes of each pixel in ultraviolet and visible light images, and construct the diagonal terms of the QUBO matrix based on the analysis results. The intrinsic attributes include radiation attributes, motion attributes, and scintillation attributes.

[0038] The correlation analysis module is used to analyze the collaborative relationship between each pixel and other pixels, construct the off-diagonal terms of the QUBO matrix based on the analysis results, and generate the QUBO matrix based on the diagonal and off-diagonal terms of the QUBO matrix.

[0039] The quantum annealing module is used to define the flame state of each pixel. It constructs a global energy calculation function based on the QUBO matrix and the flame state of each pixel. With the minimization of the global energy calculation function value as the optimization objective, the flame state of each pixel is continuously adjusted through the quantum annealing algorithm to identify the flame region in ultraviolet and visible light images.

[0040] The fire source location module is used to simultaneously identify the flame area in the binocular visible light image through step 4, and to perform binocular visual analysis on the identified visible light image through a binocular recognition algorithm to locate the position of the flame in space.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This solution fundamentally changes the traditional technical approach to flame recognition through a quantum annealing-driven QUBO optimization framework. Firstly, it innovatively constructs a joint optimization model that integrates multi-dimensional features from both ultraviolet and visible light, along with spatial correlation constraints, enabling the system to simultaneously handle pixel-level judgment and regional consistency.

[0043] Through the physical optimization process of the quantum annealer, the global optimal solution for 2 million-level pixel variables was achieved, solving the local optimality problem caused by the traditional serial algorithm and the insufficient ability of the traditional serial algorithm to handle key issues such as flame boundary ambiguity, and greatly improving the positioning accuracy of the flame boundary. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0045] Figure 2 The fitted curve of the flashing attribute score to the diagonal item when the motion attribute score is 1;

[0046] Figure 3 The fitted curve of the radiation attribute score to the diagonal item when the motion attribute score is 1;

[0047] Figure 4 The fitted curve of the flashing attribute score to the diagonal item when the motion attribute score is 1;

[0048] Figure 5 The fitted curve of the radiation attribute score to the diagonal item when the motion attribute score is 1;

[0049] Figure 6 This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0051] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0052] Example:

[0053] Please see Figure 1 The present invention provides a technical solution:

[0054] A fire detection method based on binocular vision and ultraviolet light, comprising the following steps:

[0055] Step 1: Deploy a binocular imaging system at the charging location to continuously acquire ultraviolet and visible light images;

[0056] The binocular imaging system includes a binocular ultraviolet-visible dual-spectrum camera, with each eye simultaneously acquiring ultraviolet and visible light images, and mapping each pixel of the ultraviolet and visible light images acquired by the same eye one-to-one.

[0057] The ultraviolet-visible dual-spectrum camera splits incident light into ultraviolet and visible light bands using a beam splitter or dichroic mirror. These bands are received by ultraviolet and visible light sensors, respectively, forming corresponding ultraviolet and visible light images. Since the incident light is the same, the positions of the acquired ultraviolet and visible light images are in one-to-one correspondence. By simply following the positional relationship, each pixel of the ultraviolet and visible light images can be mapped one-to-one. This is existing technology and will not be elaborated further here.

[0058] Step 2: Analyze the intrinsic properties of each pixel in the ultraviolet and visible light images, and construct the diagonal terms of the QUBO matrix based on the analysis results. The intrinsic properties include radiation properties, motion properties, and scintillation properties.

[0059] The radiation and motion properties are obtained by analyzing ultraviolet images, and the scintillation property is obtained by analyzing visible light images.

[0060] The system presets the ultraviolet radiation intensity range and ultraviolet deviation amplitude of the flame. If the ultraviolet radiation intensity of a pixel in the current ultraviolet image is within the ultraviolet radiation intensity range of the flame, the radiation attribute score of this pixel is set to 1. Otherwise, the deviation between the ultraviolet radiation intensity and the ultraviolet radiation intensity range is calculated and recorded as the intensity deviation. If the intensity deviation is not less than the ultraviolet deviation amplitude, the radiation attribute score of this pixel is set to 0. If the intensity deviation is less than the ultraviolet deviation amplitude, the radiation attribute score of this pixel is set according to the intensity deviation and the ultraviolet deviation amplitude, using the following formula:

[0061] ;

[0062] ;

[0063] in, The first in the current ultraviolet image Radiation attribute score for each pixel The first in the current ultraviolet image Intensity deviation of each pixel This represents the ultraviolet deviation amplitude. The first in the current ultraviolet image The intensity of ultraviolet radiation at each pixel The range of ultraviolet radiation intensity. This refers to the index of a pixel in the ultraviolet image;

[0064] The range of ultraviolet radiation intensity and the ultraviolet deviation of the flame are determined by experts in the field based on the specific circumstances. Experts in the field may be invited to demonstrate the ultraviolet radiation intensity of the flame at the charging site in order to determine the range of ultraviolet radiation intensity and the ultraviolet deviation of the flame. This is prior art and will not be elaborated here.

[0065] If the radiation attribute score is 1, then the corresponding pixel has the same ultraviolet radiation attribute as the flame, and the pixel can be identified as a flame pixel only from the perspective of radiation.

[0066] If the radiation attribute score is 0, the corresponding pixel is completely inconsistent with the ultraviolet radiation attribute of the flame. From the perspective of radiation alone, the pixel can be identified as a non-flame pixel.

[0067] If the radiation property score is The corresponding pixel has similar ultraviolet radiation properties to those of a flame, with a similarity to [missing information]. It depends on the size. The larger the size, the higher the similarity.

[0068] Motion attributes are evaluated using optical flux. A preset optical flux threshold is used. If the optical flux of a pixel in the current ultraviolet image is greater than the optical flux threshold, the motion attribute score of that pixel is 1. Conversely, if the optical flux of a pixel is not greater than the optical flux threshold, the motion attribute score of that pixel is 0.

[0069] The optical flux threshold is determined by experts in the field based on specific circumstances. The optical flux is calculated using the following method. Experts in the field are invited to demonstrate the optical flux of the flame at the charging site in order to determine the optical flux threshold. This is prior art and will not be elaborated here.

[0070] Because the motion properties (light flux) of flames and the environment differ greatly, the motion property scores are only 0 and 1, directly reflecting the difference between flame and non-flame areas from the motion perspective;

[0071] The specific logic for calculating horizontal and vertical optical flux is as follows: Obtain the horizontal and vertical gradients of each pixel in the ultraviolet image; construct a spatial gradient matrix from the horizontal and vertical gradients; calculate the temporal gradient matrix using the pixel values ​​of the current ultraviolet image and the corresponding pixel values ​​in the previous frame of the ultraviolet image; calculate the horizontal and vertical optical flux of each pixel in the ultraviolet image based on the spatial and temporal gradient matrices; and calculate the optical flux of each pixel based on the horizontal and vertical optical fluxes.

[0072] The specific logic underlying the calculation of the horizontal and vertical gradients is as follows:

[0073] ;

[0074] ;

[0075] in, For the ultraviolet image The horizontal gradient of each pixel For the current ultraviolet image, the first Vertical gradient of each pixel; , For coordinate parameters;

[0076] The spatial gradient matrix is ​​represented as follows:

[0077] ;

[0078] in, For the ultraviolet image of the first The spatial gradient matrix of each pixel;

[0079] The time gradient matrix is ​​represented as:

[0080] ;

[0081] ;

[0082] in, For the current ultraviolet image, the first The temporal gradient matrix of each pixel For the ultraviolet image Temporal gradient of each pixel The current ultraviolet image corresponds to the previous frame of the ultraviolet image. The pixel value of each pixel;

[0083] Since the object of analysis is a single pixel, the temporal gradient matrix is ​​one-dimensional. Since the time interval of the current ultraviolet image being compared is only one frame, the temporal gradient is directly represented by the difference between pixels.

[0084] The specific formulas used to calculate horizontal and vertical optical flux are as follows:

[0085] ;

[0086] in, For the current ultraviolet image, the first Horizontal light flux per pixel For the current ultraviolet image, the first Vertical optical flux per pixel;

[0087] The optical flux per pixel is:

[0088] ;

[0089] in, For the current ultraviolet image, the first Optical flux per pixel;

[0090] The brightness differences between adjacent frames in the current frame and the previous nine visible light images are analyzed to evaluate flicker attributes. A first brightness difference threshold and a second brightness difference threshold are preset. If the brightness difference of a pixel in the current image is not less than the second brightness difference threshold, the pixel's motion attribute score is 1; if the brightness difference is not greater than the first brightness difference threshold, the pixel's motion attribute score is 0; if the brightness difference is greater than the first brightness difference threshold but less than the second brightness difference threshold, the pixel's flicker attribute score is 0. The formula is:

[0091] ;

[0092] ;

[0093] in, For the current visible light image, the first The flicker attribute score for each pixel The first brightness difference threshold, The second brightness difference threshold, For the first Frame and the The first frame of the visible light image The brightness difference of each pixel For the first The first frame of the visible light image The brightness value of each pixel. For the first In a frame of visible light image The brightness value of each pixel. This is the index of the frame, and the indices are arranged in chronological order. The index of the current frame is 10.

[0094] In charging environments, there is a problem of electric spark flickering. The ultraviolet radiation intensity and motion intensity of electric sparks differ significantly from the surrounding environment, as do the ultraviolet radiation intensity and motion intensity of flames. During identification, if only these two points are considered, electric sparks may be misidentified as flames. However, the flickering characteristics of flames and electric sparks differ. Generally, the flickering degree of flames is lower than that of electric sparks. This embodiment quantifies this difference in flickering degree through a flickering attribute score. The first average brightness difference threshold is used to represent the greater flickering degree of flames. Considering that there are a few cases where the flickering of flames can be comparable to that of electric sparks, the value between the first average brightness difference threshold and the second average brightness difference threshold is used to represent the probability that the current flickering condition corresponds to a flame. The larger the value, the lower the probability. The flickering attribute score can be used as a penalty to prevent electric sparks from being mistaken for flames.

[0095] Please see Figure 2-3 , Figure 2-3 for The table below shows the fitted curves of the flicker attribute score and radiation attribute score to the diagonal item when the motion attribute score is 1.

[0096] Table 1: Data Fitting Table for the First Diagonal Term

[0097]

[0098] As the radiation attribute score gradually decreases and the scintillation attribute score slowly increases, the absolute value of the negative values ​​of the diagonal terms of the QUBO matrix shows a decreasing trend. Specifically, when... It decreased from 0.95 to 0.75 (a decrease of 21%), while It rose from 0.02 to 0.25 (an increase of 11.5 times). The value increased from -0.505 to -0.425. This change clearly reflects the effect of the weighting design: the decrease in radiation properties is the cause... The decrease in negative values ​​is the main reason, while the increase in scintillation properties plays a secondary role in suppression. This trend indicates that the system's determination of flame intensity decreases as radiation characteristics weaken and scintillation interference increases, which aligns with the design requirements of prioritizing flame characteristics and suppressing interference characteristics in fire detection.

[0099] Please see Figure 4-5 , Figure 4-5 for The following table shows the fitting curves of the flicker attribute score and radiation attribute score to the diagonal item when the motion attribute score is 0.

[0100] Table 2: Data Fitting Table for the Second Diagonal Term

[0101]

[0102] when It decreased from 0.98 to 0.75 (a decrease of 23.5%) and It rose from 0.02 to 0.25 (an increase of 11.5 times). The value increased from -0.49 to -0.375. This change indicates that the system primarily relies on radiation properties for judgment, with scintillation attribute scoring playing a supporting role.

[0103] The diagonal terms of the QUBO matrix are represented as follows:

[0104] ;

[0105] in, For the QUBO matrix, the first... OK Column elements, For visible light images Motion attribute score for each pixel , and These are the weighting coefficients, and , .

[0106] The energy cost of classifying a single pixel as a flame is directly determined. The smaller the value, the lower the energy cost of classifying a single pixel as a flame. This quantifies the "cost" or "benefit" of a pixel independently becoming a flame. Negative values ​​encourage the pixel to be classified as a flame (reducing the total system energy), while positive values ​​inhibit the pixel from being classified as a flame (increasing the total system energy). The radiation attribute score reflects the similarity of the pixel's ultraviolet radiation properties to those of a flame. The larger the value, the higher the similarity; this reduces the system's overall energy and encourages the pixel to be identified as a flame. The similarity between the motion characteristics of a pixel and a flame is considered. A higher value indicates a greater similarity, reducing the system's overall energy and encouraging the pixel to be identified as a flame. The flicker attribute score reflects the probability of an electric spark pixel being misidentified as a flame; a higher value increases this probability, increasing the system's overall energy and serving as a penalty to discourage the pixel from being identified as a flame. Ultraviolet radiation intensity is the most representative indicator of flame characteristics, and because ultraviolet signals are less affected by ambient light and flame ultraviolet radiation is stable, its weight is set to the maximum. While the motion characteristics of flames and the environment typically differ significantly, the weight of flame intensity is less than that of ultraviolet radiation intensity due to potential light flow fluctuations caused by swaying obstructions in the environment. Flicker attribute is easily affected by changes in ambient light, and the flicker characteristics of flames and interference may overlap; therefore, its weight is set to the minimum.

[0107] Step 3: Analyze the collaborative relationship between each pixel and other pixels, construct the off-diagonal terms of the QUBO matrix based on the analysis results, and generate the QUBO matrix based on the diagonal and off-diagonal terms of the QUBO matrix.

[0108] The radiation attribute score, motion attribute score, and scintillation attribute score are used to construct an attribute representation vector, denoted as follows: ;

[0109] Based on the representation vectors of adjacent pixels, the off-diagonal terms of the QUBO matrix are constructed using the following formula:

[0110] ;

[0111] in, For the QUBO matrix, the first... OK Column elements, for and Mahalanobis distance, For pixels and pixels distance, , and Let be the scale parameter, where , ， , This is the index of the pixel in the ultraviolet image, and .

[0112] ;

[0113] in, It is the transpose symbol. It is the inverse of the covariance matrix of all attribute representation vectors, and it already contains the values ​​for each pixel. Calculating the inverse of the covariance matrix is ​​a conventional technique used by those skilled in the art.

[0114] The QUBO matrix contains core parameters describing the interactions between pixels, and its design directly determines the continuity and anti-interference capability of the flame region. Real flames exhibit continuous propagation characteristics in space, and the states of adjacent pixels should be highly correlated. Interference such as electric sparks and metallic reflections are usually isolated points or localized areas and need to be decoupled from the real flame. The system is guided to find the globally optimal flame distribution through coupled off-diagonal terms, rather than a simple superposition of isolated pixels. Considering the covariance between features of different pixels (the correlation between color, brightness, and motion) is more robust than Euclidean distance. If there are large differences in brightness but consistent motion, they may still be identified as the same flame. It is a Gaussian kernel with a distance between two pixels, which controls the attenuation of coupling with distance, resulting in stronger correlation between pixels that are close together.

[0115] Step 4: Define the flame state of each pixel, and construct a global energy calculation function based on the QUBO matrix and the flame state of each pixel; with the optimization objective of minimizing the value of the global energy calculation function, continuously adjust the flame state of each pixel through the quantum annealing algorithm to identify the flame region in ultraviolet and visible light images;

[0116] Set the flame state to , ,in, For the first The state of each pixel, if Then the first If each pixel is a non-flame pixel, then Then the first Each pixel represents a flame pixel;

[0117] The global energy calculation function is:

[0118] ;

[0119] in, The global energy function, For the first The state of each pixel;

[0120] The QUBO matrix is ​​converted to a format supported by the quantum annealer and input into the quantum annealer. Quantum annealing is then performed. During the quantum annealing process, each pixel may correspond to... or Select during calculation or Continuously adjust The corresponding flame identification result is output until the global energy function value reaches its minimum. This is a conventional technique for those skilled in the art to perform quantum annealing using a quantum annealer under a given QUBO matrix and global energy, and will not be elaborated here.

[0121] By optimizing the physical process of the quantum annealer, a global optimal solution for 2 million pixel variables is achieved, solving the local optima problem caused by traditional serial algorithms and their insufficient ability to handle key issues such as flame boundary ambiguity. This significantly improves the positioning accuracy of the flame boundary. Traditional methods employ a serial "detection-post-processing" paradigm, independently judging individual pixels and then weakly associating them using algorithms such as CRF, leading to ambiguity in the boundary region due to local decision conflicts. This scheme transforms the flame state decision of 2 million pixels into a unified quadratic optimization problem using the QUBO matrix, and utilizes the parallel tunneling effect of the quantum annealer to simultaneously process the correlation relationships of all pixels.

[0122] Step 5: Simultaneously identify the flame region in the binocular visible light image through Step 4, and perform binocular visual analysis on the identified binocular visible light image using a binocular recognition algorithm to locate the position of the flame in space.

[0123] The existing technology can be used to obtain the spatial coordinates of flames by using a binocular recognition algorithm. Based on the parameters of the binocular camera, the flame pixels are mapped to a three-dimensional coordinate system. The process of obtaining the spatial coordinates of the flame includes: calculating the correspondence of flame pixels based on the SIFT algorithm, obtaining the matching relationship of flame pixels, and after identifying flame pixels in images from different perspectives, finding the correspondence of these flame pixels in different images through a feature point matching algorithm.

[0124] The correspondence of the fire source feature points is calculated based on the SIFT algorithm to obtain the feature point matching relationship. After identifying flame pixels in images from different viewpoints, the correspondence of these feature points in different images is found through a feature point matching algorithm. This can be done using some computer vision algorithms, such as SIFT (Scale Invariant Feature Transform) or ORB (Oriented Fast and Rotated Breeze), etc. This embodiment uses SIFT (Scale Invariant Feature Transform). The spatial coordinates of the flame are obtained by calculating the parameters of the stereo camera and the matching relationship of the SIFT based on the triangulation method. Through the known camera parameters and the matching relationship of the flame pixels, the three-dimensional coordinates of the target point can be calculated using the triangulation method. Common methods include Direct Linear Transform (DLT) or nonlinear optimization-based methods, such as optimizing the coordinates of the three-dimensional point by minimizing the reprojection error. The calculated three-dimensional coordinates are usually relative to the camera coordinate system. If it is necessary to map them to the global world coordinate system, the position and orientation of the camera need to be considered. This can be achieved by coordinate transformation using camera extrinsic parameters.

[0125] The present invention further provides a fire detection system based on binocular vision and ultraviolet light, the system being used to implement the above-mentioned fire detection method based on binocular vision and ultraviolet light, specifically including:

[0126] The image acquisition module is used to deploy a binocular imaging system at the charging site to continuously acquire ultraviolet and visible light images.

[0127] The attribute analysis module is used to analyze the intrinsic attributes of each pixel in ultraviolet and visible light images, and construct the diagonal terms of the QUBO matrix based on the analysis results. The intrinsic attributes include radiation attributes, motion attributes, and scintillation attributes.

[0128] The correlation analysis module is used to analyze the collaborative relationship between each pixel and other pixels, construct the off-diagonal terms of the QUBO matrix based on the analysis results, and generate the QUBO matrix based on the diagonal and off-diagonal terms of the QUBO matrix.

[0129] The quantum annealing module is used to define the flame state of each pixel. It constructs a global energy calculation function based on the QUBO matrix and the flame state of each pixel. With the minimization of the global energy calculation function value as the optimization objective, the flame state of each pixel is continuously adjusted through the quantum annealing algorithm to identify the flame region in ultraviolet and visible light images.

[0130] The fire source location module is used to simultaneously identify the flame area in the binocular visible light image through step 4, and to perform binocular visual analysis on the identified visible light image through a binocular recognition algorithm to locate the position of the flame in space.

[0131] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0132] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0133] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0134] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that cannot be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A fire detection method based on binocular vision and ultraviolet, characterized in that, The specific steps include: Step 1: arranging a binocular imaging system at a charging site, and continuously acquiring binocular ultraviolet images and visible light images; Step 2: analyzing the self attributes of each pixel point of the ultraviolet image and the visible light image collected by the same eye, constructing the diagonal elements of the QUBO matrix based on the analysis results, and the self attributes including radiation attributes, motion attributes and flicker attributes; Step 3: analyzing the cooperative relationship between each pixel point and other pixel points, constructing the non-diagonal elements of the QUBO matrix based on the analysis results, and generating the QUBO matrix based on the diagonal elements of the QUBO matrix and the non-diagonal elements of the QUBO matrix; Step 4: defining the flame state of the pixel point, constructing a global energy calculation function based on the QUBO matrix and the flame state of each pixel point, and taking the minimization of the global energy calculation function value as the optimization objective, and continuously adjusting the flame state of each pixel point through the quantum annealing algorithm to identify the flame region of the ultraviolet image and the visible light image; Step 5: identifying the flame region of the visible light image of the binocular at the same time through step 4, and performing binocular vision analysis on the identified visible light image through a binocular recognition algorithm to locate the position of the flame in space; The radiation attributes and the motion attributes are obtained by analyzing the ultraviolet image, and the flicker attributes are obtained by analyzing the visible light image; the flicker attributes, the radiation attributes and the motion attributes are analyzed to construct a self attribute scoring system; The ultraviolet radiation intensity range and the ultraviolet deviation amplitude of the preset flame are defined, if the ultraviolet radiation intensity of a pixel point of the current ultraviolet image is within the ultraviolet radiation intensity range of the flame, the radiation attribute score of the pixel point is defined as 1, otherwise, the deviation of the ultraviolet radiation intensity from the ultraviolet radiation intensity range is calculated, which is denoted as intensity deviation, if the intensity deviation is not less than the ultraviolet deviation amplitude, the radiation attribute score of the pixel point is defined as 0, if the intensity deviation is less than the ultraviolet deviation amplitude, the radiation attribute score of the pixel point is set according to the intensity deviation and the ultraviolet deviation amplitude, and the formula is: wherein, is a radiation property score of a pixel point in the current ultraviolet image, is an intensity deviation of a pixel point in the current ultraviolet image, is an ultraviolet deviation amplitude, is an ultraviolet radiation intensity of a pixel point in the current ultraviolet image, is an ultraviolet radiation intensity range, is an index of a pixel point in the ultraviolet image;​​​ The motion attribute is evaluated by light flow, a light flow threshold is preset, if the light flow of a pixel point of the current ultraviolet image is greater than the light flow threshold, the motion attribute score of the pixel point is 1, otherwise, if the light flow of the pixel point is not greater than the light flow threshold, the motion attribute score of the pixel point is 0; The luminance difference between adjacent frames in the current frame and the previous nine visible light images is analyzed to evaluate the flicker attribute. A first luminance difference threshold and a second luminance difference threshold are preset. If the luminance difference of a pixel point in the current image is not less than the second luminance difference threshold, the motion attribute score of the pixel point is 1. If the luminance difference is not greater than the first luminance difference threshold, the motion attribute score of the pixel point is 0. If the luminance difference is greater than the first luminance difference threshold and less than the second luminance difference threshold, the flicker attribute score of the pixel point is , and the formula is: wherein, is a flicker attribute score of a current pixel point in the current visible light image, is a first luminance difference threshold, is a second luminance difference threshold, is a luminance difference of a pixel point in the first frame and the second frame visible light image, is a luminance value of a pixel point in the first frame visible light image, is an index of a frame, the indexes are arranged in a time sequence, and the index of the current frame is 10.​​​​​​​​​ The diagonal elements of the QUBO matrix are represented as: wherein is the element of the QUBO matrix in the row column, is the motion attribute score of the pixel point of the ultraviolet image, , and are weight coefficients, and , .

2. The fire detection method based on binocular vision and ultraviolet according to claim 1, characterized in that: The binocular imaging system includes binocular ultraviolet-visible light dual-spectrum cameras, each eye simultaneously collects ultraviolet images and visible light images, and each pixel point of the ultraviolet image and the visible light image collected by the same eye is one-to-one mapped.

3. The fire detection method based on binocular vision and ultraviolet according to claim 2, characterized in that: The horizontal gradient and the vertical gradient of each pixel point of the ultraviolet image are obtained, the horizontal gradient and the vertical gradient are constructed into a spatial gradient matrix, a time gradient matrix is calculated through the pixel value of the current ultraviolet image and the pixel value of the corresponding position of the previous frame ultraviolet image, and the horizontal light flow and the vertical light flow of each pixel point of the ultraviolet image are calculated according to the spatial gradient matrix and the time gradient matrix; The light flow of each pixel point is calculated according to the horizontal light flow and the vertical light flow.

4. The fire detection method based on binocular vision and ultraviolet according to claim 1, characterized in that: The radiation attribute score, the motion attribute score, and the flicker attribute score are combined to form an attribute characterization vector, denoted as The non-diagonal elements of the QUBO matrix are constructed according to the representative vectors of adjacent pixel points, and the formula is: wherein, is the element of the QUBO matrix in the i-th row and j-th column, is the Mahalanobis distance of and is the distance between pixel point and pixel point , , and are scale parameters, wherein, , ， , is the index of the pixel point in the ultraviolet image, and .​​​ 5. The method according to claim 4, wherein: The state of the flame is defined as , wherein, is the state of the pixel, if then the pixel is a non-flame pixel, if then the pixel is a flame pixel; The global energy calculation function is: wherein, is a global energy function, is the state of the th pixel.

6. A dual-vision and ultraviolet-based fire detection system, characterized by: The system is used to realize the fire detection method based on binocular vision and ultraviolet in any one of claims 1-5, specifically comprising: An image acquisition module is used to arrange a binocular imaging system in a charging place, continuously acquire ultraviolet images and visible light images; An attribute analysis module is used to analyze the self attributes of each pixel point of the ultraviolet image and the visible light image, construct the diagonal items of the QUBO matrix based on the analysis results, and the self attributes include radiation attributes, motion attributes and flicker attributes; An association analysis module is used to analyze the cooperative relationship between each pixel point and other pixel points, construct the non-diagonal items of the QUBO matrix based on the analysis results, and generate the QUBO matrix based on the diagonal items of the QUBO matrix and the non-diagonal items of the QUBO matrix; A quantum annealing module is used to define the flame state of the pixel point, construct a global energy calculation function based on the QUBO matrix and the flame state of each pixel point, and minimize the global energy calculation function value as the optimization target, and continuously adjust the flame state of each pixel point through the quantum annealing algorithm to identify the flame area of the ultraviolet image and the visible light image; A fire source positioning module is used to simultaneously identify the flame area of the visible light image of the binocular at step 4, perform binocular vision analysis on the identified visible light image through a binocular recognition algorithm, and locate the position of the flame in the space.

Images