CN116013024B - A mine fire monitoring method based on visible light visual feature fusion - Google Patents

Abstract

The invention discloses a mine external factor fire monitoring method with visible light visual characteristics fused, which mainly comprises the following steps: a visible light camera is installed at the optimal position of the area to be monitored of the mine; dividing a fire disaster area by adopting an improved seed area growth algorithm; calculating static characteristics and dynamic characteristics of a fire area, and establishing a fire sample data set and an identification model; respectively training and testing an identification model; the mine fire monitoring system monitors whether a mine external fire exists or not in real time through the identification model; when the presence is identified, a fire alarm is activated. The mine external fire monitoring method can overcome the problems of the existing monitoring method, and can improve the identification speed, accuracy and reliability of early fire on the basis of realizing underground large-area monitoring, thereby meeting the actual requirements of coal mine safety production.

Description

A mine fire monitoring method based on visible light visual feature fusion

Technical Field

The present invention relates to a mine external cause fire monitoring method integrating visible light visual features, in particular to a vision-based target detection technology, an image segmentation technology, a visual feature extraction technology, and a fire monitoring technology.

Background Art

In recent years, statistics on major accidents in coal mines in my country show that mine fire is one of the five major disasters in coal mine accidents. At present, relevant researchers have insufficient understanding of the disaster-causing mechanism and disaster relief risks of mine fires, resulting in the lack of effective technical methods in the perception and identification of mine fires, causing major accidents of mine fires to occur from time to time, causing huge economic losses, and in serious cases, causing casualties in the mines. When a fire occurs, flames, smoke, toxic and harmful gases are usually produced. At the same time, accident investigations show that in mine fire accidents, deaths caused by trauma and burns account for less than 20%, and deaths caused by carbon monoxide poisoning and suffocation account for more than 80%. The higher the concentration of carbon monoxide and the longer the duration, the more serious the damage to the human body, until death. Therefore, in coal mines, it is crucial to timely discover high-temperature heat sources or early fires, quickly locate the location of disaster sources, accurately identify disaster types, and launch emergency plans and emergency rescue in a targeted manner for coal production safety.

At present, researchers have conducted a lot of theoretical research, experimental analysis and field tests to realize the monitoring and early warning of external causes of mine fires, proposed a variety of perception and early warning methods, and achieved a lot of research results. However, the existing technologies all involve the influence of many factors such as coal geological conditions, mining methods and processes, sensor technology, etc., resulting in high omission and false alarm rates of existing coal mine fire early warning systems. At the same time, the existing monitoring technology cannot achieve early fire warning, nor can it locate and track the source of disasters, which makes it difficult to meet the needs of coal mine safety production. In addition, in the harsh environment underground, the sensors relied on by the disaster perception device are easily interfered by environmental factors, and these sensors can only monitor the installation location or a small area around it, making it difficult to achieve large-scale monitoring of disasters in mines.

In the visual monitoring of external fires in mines, the existing methods for monitoring external fires in mines based on near-infrared and far-infrared bands have defects to varying degrees. For example, near-infrared monitoring can obtain clearer video images than visible light bands in harsh environments, but it is easily affected by interfering light sources such as tunnel lights, mining lamps, and car lights; far-infrared monitoring has the advantages of large monitoring space, strong anti-interference ability, and good real-time performance, but it is easily affected by absorbent gases, dust, and ambient temperature, resulting in limited monitoring distance. Visible light visual monitoring not only has the advantages of large monitoring space, rich geometric information, and high spatiotemporal resolution, but also has color characteristics that cannot be expressed by infrared monitoring and long-distance monitoring performance. For this reason, fire monitoring based on visible light visual features has become one of the important research directions for monitoring external fires in mines. However, the current visible light fire visual monitoring method is only applicable to specific scenarios of out-of-control fire sources or non-interfering light sources, and cannot meet the visual monitoring needs of external fires in mines in the early stages and complex environments.

Aiming at the problems existing in the visual monitoring of visible light fires, combined with the special environment of coal mines, the present invention uses the visual features of the spread of external fires in mines to realize a method for monitoring external fires in mines with the fusion of visible light visual features. By using external equipment such as explosion-proof cameras as video image acquisition devices, video images of the monitoring area can be acquired in real time. Only a small number of cameras are deployed in key areas underground to achieve large-scale monitoring of coal mines. At the same time, the video images collected by the monitoring equipment are analyzed by computer vision and image processing technology, and the visual features of the external fires in mines are combined to identify the suspected fire targets in the video images, so as to achieve rapid, accurate and reliable identification of high-temperature heat sources and early fires in the monitoring area. Compared with the existing monitoring and early warning methods, the mine external fire identification method based on the present invention will be faster, more accurate and more reliable. In addition, in the process of emergency rescue in mines, fire monitoring based on visible light vision can also assist rescue personnel in making predictions on the disaster situation on the scene and formulating corresponding emergency plans.

Summary of the invention

The technical problem to be solved by the present invention is to overcome the problems of false alarm and missed alarm when the above-mentioned existing visible light visual monitoring technology is used to identify external causes of mine fires, and to improve the recognition speed, accuracy and reliability of early fires on the basis of realizing large-area monitoring underground, thereby meeting the actual needs of safe production in coal mines.

The present invention specifically adopts the following technical solutions to solve the above technical problems:

A mine fire monitoring method based on visible light visual feature fusion, characterized in that the mine fire monitoring method includes the following steps:

Step 1: Based on the spatial characteristics, geological conditions, coal quality characteristics and mining impact of the mine area to be monitored, a visible light camera is installed at the optimal location of the area to be monitored;

Step 2: Sample the collected visible light video at equal intervals to obtain visible light images of the t-th frame and the t-1-th frame; use the improved seed region growing algorithm to extract the fire area in the t-th frame and the t-1-th frame;

Step 3: Calculate the static features and dynamic features of the fire area; the static features are circularity, rectangularity, eccentricity, color, texture and sharp corners; the dynamic features are area change rate, centroid movement, perimeter growth rate, similarity coefficient and flashing frequency; fuse the static features and dynamic features to construct a feature vector of the fire area in the t-th frame image;

Step 4: By looping steps 2 to 3, a sample data set of external causes of mine fires is constructed; a portion of samples in the sample data set are randomly selected to form a training data set for the neural network model, and the neural network model is trained;

Step 5: Use the remaining samples as a test data set, input them into the trained neural network model, use the predicted value to determine whether there is an external cause of mine fire, and use an objective evaluation method to analyze the comprehensive performance of the neural network model;

Step 6: Use the improved seed region growing algorithm to extract the suspected fire area in the current frame and the previous frame in the video stream of the monitored area; construct the feature vector of the suspected fire area in the current frame through step 3, and input the feature vector of the suspected fire area in the previous frame into the trained neural network model to identify whether there is an external cause of fire in the mine.

Furthermore, the optimal position is that the closest distance to the monitored target area shall not be less than 5 meters, and the farthest distance shall not be greater than 50 meters. The installation position must avoid fixed interfering light sources, or be installed in the direction of light, installed on a solid roof or tunnel side wall, and must not affect normal construction operations underground.

Furthermore, the improved seed region growing algorithm includes: smoothing and filtering the t-th frame and t-1 frame images, selecting pixels with a grayscale of 1 in the filtered image as seed points; using the 3×3 neighborhood of the seed point as the starting growth area, and performing 8-neighborhood expansion; selecting the grayscale mean and variance of the neighborhood pixels as region growing conditions, and when the grayscale mean and variance of the neighborhood pixels and the seed point are less than a set growth threshold, merging the pixels corresponding to the seed point into the grown area; the grayscale mean is calculated by Calculated; the variance is obtained by Calculated; the regional growth condition is In the formula, _fm (x,y) and _fv (x,y) are the mean and variance of the neighborhood corresponding to the seed point f(x,y); seedpoint( _x0 , _y0 ) is the grayscale of the seed point f(x,y); _T1 and _T2 are the growth thresholds, _which are related to the variance _fv (x,y); r is the neighborhood radius, r= ₁ .

Furthermore, the texture feature is obtained by calculating the image entropy of the fire area or the suspected fire area. Where R _IE is the image entropy; _{Pi is} the probability of gray level i appearing in the suspected fire area; L is the gray level of the visible light image.

Furthermore, the sharp corner feature calculation process is as follows:

Step 1: Use the traversal method to find the left base point and the right base point of the suspected fire area, and use the first-order differential operator to obtain the derivative of the "left base point-highest point-right base point" connection curve at each edge point;

Step 2: Determine whether there is a zero-crossing point based on the derivative of the edge point. If there is a zero-crossing point and the left derivative is positive and the right derivative is negative, the edge point is determined to be a local sharp corner.

Step 3: Find the coordinates of the left and right edge points in the Nth row below each sharp corner, and calculate the width corresponding to the sharp corner based on the coordinates of the left and right edge points. If the aspect ratio of the sharp corner is less than 0.5, the sharp corner is judged to be a flame sharp corner, otherwise it is judged to be noise or burrs.

Furthermore, the flash frequency acquisition process is as follows: using the similarity coefficient to find the same suspected fire area in the t-th frame and the t-1 frame, and calculating the cumulative grayscale difference change rate corresponding to the same suspected fire area. Where _Fj is the cumulative grayscale difference change rate of the j-th suspected fire area; M and N are the total number of pixels in the j-th suspected fire area; Vt _,i is the i-th pixel value in the suspected fire area in the t-th frame image; Vt _-1,i is the i-th pixel value in the suspected fire area in the t-1-th frame image; Δt is the time interval between the t-1-th frame and the t-th frame.

Furthermore, the similarity coefficient In the formula, μ _t and μ _t-1 are the average grayscale of the suspected fire source area in the t-th frame and t-1 frame respectively; σ _t and σ _t-1 are the standard deviations of the suspected fire source area in the t-th frame and t-1 frame respectively; C ₁ and C ₂ are constants to avoid the denominator being 0; usually C ₁ = (K ₁ ·L) ² and C ₂ = (K ₂ ·L) ² , generally K ₁ = 0.01 and K ₂ = 0.03, and L is the dynamic range of the grayscale image, L = 255.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a mine external fire monitoring system according to the present invention;

FIG2 is a flow chart of a method for monitoring external causes of fire in a mine according to the present invention;

FIG3 is a schematic diagram of extracting the flame corner of a mine fire caused externally according to the present invention.

DETAILED DESCRIPTION

In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail and completely below in conjunction with the accompanying drawings and specific implementation methods. The embodiments should not be construed as limiting the scope of use of the present invention.

As shown in Figure 1, the mine external fire monitoring system is divided into an above-ground part and an underground part, which is used to monitor the mine external fire. Its main components include:

1. Information processing server (101): responsible for storing video images collected by the visible light camera (105), and analyzing whether the collected video images contain suspected areas of external fire in the mine. If there is a suspected fire area, the visual features of the suspected fire area are extracted and input into the trained BP neural network. The predicted value is used to determine whether it is a fire caused by external factors in the mine. If it is a fire caused by external factors in the mine, the alarm module (107) sends out sound, light, vibration and other alarm signals through the communication network, and the monitoring server (102) sends out alarm prompts and human-computer interaction on the monitoring screen when the alarm module sends out the alarm information.

2. Monitoring server (102): responsible for displaying monitoring data of the coal mine monitoring area. At the same time, the monitoring server (102) and the information processing server (101) are connected to each other in communication, displaying real-time images of the monitoring area, and providing alarm prompts and human-computer interaction after the alarm module (107) issues an alarm message. Production management personnel can retrieve and query historical data stored in the information processing server (101) through the monitoring server (102), and the monitoring server (102) is connected to the core switch (103) through a communication line to access the mine communication network.

3. Core switch (103): The core management and switching device of the mine communication network, responsible for the management and data exchange of all devices connected to the mine communication network, with routing function, connected to the external Internet through a firewall.

4. Ring network switch (104): underground switching equipment of the mine communication network, installed underground, with multiple ring network switches connected in a ring network manner.

5. Visible light camera (105): an image acquisition device installed in key monitoring areas in the mine, responsible for collecting video images of monitoring areas prone to fire, such as transportation tunnels, ventilation tunnels, communication tunnels, heading tunnels, cutting eyes, coal mining faces, heading faces, electromechanical chambers, underground substations, etc. in the mine. The video images can be color images, grayscale images or pseudo-color images; the visible light camera (105) adopts a mining explosion-proof camera.

6. Communication substation (106): One end is connected to the binocular vision camera (105) for communication, and the other end is connected to the ring network switch (104) for communication. It can be connected to the devices at both ends via a wireless communication network or a wired communication network. In this example, wired communication is used for illustration.

7. Alarm module (107): uses sound, light, vibration and other alarm methods, directly communicates with the monitoring server (102) through the existing communication network, and communicates with the information processing server (101) through the core switch (103); when the alarm module (107) receives the alarm signal sent by the information processing server (101), it uses one or more alarm methods such as sound, light, and vibration to alarm, prompting the staff to deal with the scene and activate the emergency plan.

The process of the mine external fire monitoring method as shown in FIG2 includes:

1. Initialization (201): In combination with the spatial characteristics, geological conditions, coal quality characteristics and mining impacts of the mine, a visible light camera is installed at the optimal position of the key monitoring area in the mine; the key monitoring area includes the transportation tunnel, ventilation tunnel, communication tunnel, heading tunnel, opening eye, coal mining face, heading face, electromechanical chamber, underground substation and other areas prone to disaster hazards in the mine; the optimal position is that the closest distance to the monitoring target area shall not be less than 5 meters, and the farthest distance shall not be more than 50 meters. The installation position must avoid fixed interfering light sources, or be installed in the direction of light, installed on a solid roof or tunnel side wall, and must not affect normal construction operations underground.

2. Segmenting the fire area in the image (202): sampling the visible light video at equal intervals, obtaining the visible light images of the t-th frame and the t-1-th frame, and extracting the fire area in the t-th frame and the t-1-th frame using an improved seed region growing algorithm; the improved seed region growing algorithm comprises: performing smoothing filtering on the t-th frame and the t-1-th frame images, selecting pixels with a grayscale of 1 in the filtered images as seed points; using the 3×3 neighborhood of the seed point as the starting growth area, and performing 8-neighborhood expansion; selecting the grayscale mean and variance of the neighborhood pixels as region growth conditions, and when the grayscale mean and variance of the neighborhood pixels and the seed point are less than a set growth threshold, merging the pixels relative to the seed point into the grown area;

The grayscale mean is obtained by Calculated; the variance is obtained by Calculated; the regional growth condition is In the formula, _fm (x,y) and _fv (x,y) are the mean and variance of the neighborhood corresponding to the seed point f(x,y); seedpoint( _x0 , _y0 ) is the grayscale of the seed point f(x,y); _T1 and _T2 are the growth thresholds, _which are related to the variance _fv (x,y); r is the neighborhood radius, r= ₁ .

As the neighboring pixels are continuously incorporated into the grown region, T ₁ and T ₂ will also change with the variance. Through a large number of experiments, it is known that if the threshold of region growth is too small, the suspected fire region is prone to partial holes, which is not conducive to the extraction of fire features; if the threshold of region growth is too large, the suspected fire region is prone to over-segmentation. For this reason, the thresholds T ₁ and T ₂ are selected as T ₁ = 2f _v (x, y) and T ₂ = f _v (x, y) respectively.

3. Calculate the visual features of the fire area (203): the visual features include static features and dynamic features; the static features are circularity, rectangularity, eccentricity, color, texture and sharp corners; the dynamic features are area change rate, centroid movement, perimeter growth rate, similarity coefficient and flashing frequency; the static features and dynamic features are fused to construct a feature vector of the suspected fire area in the t-th frame image;

The texture feature is obtained by calculating the image entropy of the fire area. The image entropy calculation formula is: Where R _IE is the image entropy; _{Pi is} the probability of gray level i appearing in the suspected fire area; L is the gray level of the visible light image.

The flash frequency acquisition process mainly includes: using the similarity coefficient to find the same suspected fire area in the t-th frame and the t-1 frame, and calculating the cumulative grayscale difference change rate corresponding to the same suspected fire area. Where _Fj is the cumulative grayscale difference change rate of the j-th suspected fire area; M and N are the total number of pixels in the j-th suspected fire area; Vt _,i is the i-th pixel value in the suspected fire area in the t-th frame image; Vt _-1,i is the i-th pixel value in the suspected fire area in the t-1-th frame image; Δt is the time interval between the t-1-th frame and the t-th frame.

Similarity coefficient In the formula, μ _t and μ _t-1 are the average grayscale of the suspected fire source area in the t-th frame and t-1 frame respectively; σ _t and σ _t-1 are the standard deviations of the suspected fire source area in the t-th frame and t-1 frame respectively; C ₁ and C ₂ are constants to avoid the denominator being 0; usually C ₁ = (K ₁ ·L) ² , C ₂ = (K ₂ ·L) ² , generally K ₁ = 0.01, K ₂ = 0.03, L is the dynamic range of the grayscale image, L = 255.

4. Establishing a sample data set and training a recognition model (204): by looping steps 202 to 203, a sample data set for constructing a mine external cause fire is obtained in the video image; based on the visual features, a model structure and model parameters are obtained to establish a recognition model for the mine external cause fire; the recognition model adjusts the model structure and model parameters according to the visual features of the mine external cause fire in different environments and the evolution dynamics of the mine external cause fire, and continuously improves the recognition model; the recognition model uses at least one of a neural network model and a feature matching model; 60% of the samples in the sample data set are randomly selected to form a training data set for the recognition model, and the recognition model is trained;

5. Test and analyze the recognition model (205): The remaining 40% of the samples are used as a test data set and input into the trained neural network model or feature matching model. The predicted value is used to determine whether there is an external cause of fire in the mine, and an objective evaluation method is used to analyze the comprehensive performance of the neural network model or feature matching model. The objective evaluation method uses three indicators, namely, accuracy (ACC), detection rate (TPR), and false positive rate (FPR), for quantitative analysis. The larger the ACC and TPR, and the smaller the FPR, the better the recognition performance of the recognition model. In the formula, TP is the positive sample predicted to be positive; TN is the negative sample predicted to be negative; FP is the negative sample predicted to be positive; FN is the positive sample predicted to be negative.

6. Whether there is an external cause fire in the mine (206): The external cause fire monitoring system of the mine collects video images in the monitoring area in real time, and uses the improved seed region growing algorithm to extract the suspected fire area in the current frame and the previous frame in the video stream; through step 203, the feature vector of the suspected fire area in the current frame is constructed, and the feature vector of the suspected fire area in the previous frame is input into the trained neural network model to identify whether there is an external cause fire in the mine. When the neural network model recognizes that there is an external cause fire in the video image, the external cause fire monitoring system of the mine executes (207) in sequence, otherwise it returns to execute (205).

7. Mine external fire warning (207): When a mine external fire is identified, the information processing server sends an alarm signal to the alarm module. After receiving the alarm signal, the alarm module uses one or more of the following alarm methods: sound, light, and vibration to alert the staff to handle the scene and activate the emergency plan.

As shown in Figure 3, the flame corner extraction diagram of the external cause fire in the mine, the flame corner feature (in the fire image, the flame corner is the maximum point (highest point) of the edge contour curve, and the curvature corresponding to the highest point is large.) The calculation process is:

Step 1: Use the traversal method to find the left base point and the right base point of the suspected fire area, and use the first-order differential operator to obtain the derivative of the "left base point-highest point-right base point" connection curve at each edge point;

Step 2: Determine whether there is a zero-crossing point based on the derivative of the edge point. If there is a zero-crossing point and the left derivative is positive and the right derivative is negative, the edge point is determined to be a local highest point (cusp);

Step 3: Find the coordinates of the left and right edge points in the Nth row below each sharp corner, and calculate the width corresponding to the sharp corner based on the coordinates of the left and right edge points. If the aspect ratio of the sharp corner is less than 0.5, the sharp corner is judged to be a flame sharp corner, otherwise it is judged to be noise or burrs.

Claims (7)

1. The mine external factor fire monitoring method with the visible light visual characteristics fused is characterized by comprising the following steps of:

Step 1: the method comprises the steps that by combining the spatial characteristics, geological conditions, coal quality characteristics and mining influence of a mine area to be monitored, a visible light camera is installed at the optimal position of the area to be monitored;

Step 2: sampling the acquired visible light video at equal intervals to acquire visible light images of a t frame and a t-1 frame; extracting fire areas in the t frame and the t-1 frame by adopting an improved seed area growth algorithm;

Step 3: calculating static characteristics and dynamic characteristics of a fire area; the static characteristics are circularity, rectangularity, eccentricity, color, texture and sharp angle; the dynamic characteristics are area change rate, mass center movement, perimeter growth rate, similarity coefficient and flickering frequency; the obtained static features and dynamic features are fused, and feature vectors of fire areas in the t-th frame image are constructed;

Step 4: through the circulation steps 2 to 3, a sample data set of the mine external fire disaster is constructed; randomly extracting a part of samples in the sample data set to form a training data set of the neural network model, and training the neural network model;

Step 5: taking the rest of samples as a test data set, inputting a trained neural network model, judging whether a mine external fire disaster exists or not by adopting a predicted value, and analyzing the comprehensive performance of the neural network model by adopting an objective evaluation method;

Step 6: extracting fire suspected areas of a current frame and a previous frame in a video stream of a monitoring area by adopting an improved seed area growth algorithm; and 3, constructing a feature vector of a fire suspected area in the current frame, inputting the feature vector of the fire suspected area in the previous frame into a trained neural network model, and identifying whether a mine external fire exists.

2. The method for monitoring mine external fire disaster by fusion of visual characteristics of visible light according to claim 1, wherein the optimal position is that the shortest distance from the target area to be monitored is not less than 5m, the longest distance is not more than 50m, and the installation position is required to avoid fixed interference light sources or be installed in the forward direction and is installed on a firm top plate or a tunnel side wall, so that the underground normal construction operation cannot be influenced.

3. The mine external fire monitoring method of claim 1, wherein said modified seed region growing algorithm comprises: smoothing filtering is carried out on the t-th frame image and the t-1 frame image, and pixel points with gray level of 1 in the filtered images are selected as seed points; taking a 3×3 neighborhood of the seed point as an initial growth area, and performing 8 neighborhood expansion; selecting the gray average value and variance of the neighborhood pixels as region growing conditions, and merging the pixel points opposite to the seed points into a grown region when the gray average value and variance of the neighborhood pixels and the seed points are smaller than a set growing threshold;

the gray level average value passes Calculating to obtain; the variance is passed throughCalculating to obtain; the region growth conditions are as followsWherein f _m (x, y) and f _v (x, y) are respectively the mean and variance of the corresponding neighborhood of the seed point f (x, y); seedpoint (x ₀,y₀) is the gray scale of the seed point f (x, y); t ₁ and T ₂ are growth thresholds, T ₁ and T ₂ are related to variance f _v (x, y), respectively; r is the neighborhood radius, r=1.

4. The mine external factor fire monitoring method of claim 1, wherein the texture features are obtained by calculating image entropy of a fire area, and the image entropy calculation formula is as followsWherein R _IE is image entropy; p _i is the probability of gray level i in the suspected fire area; l is the gray scale of the visible light image.

5. The mine external factor fire monitoring method based on visible light visual feature fusion according to claim 1, wherein the sharp angle feature calculation process is as follows:

Step 1: searching a left base point and a right base point of a fire suspected region by adopting a traversal method, and solving the derivative of a 'left base point-highest point-right base point' connecting curve at each edge point by a first-order differential operator;

step 2: judging whether zero crossing points exist according to the derivative of the edge points, and judging that the edge points are local sharp angles if the zero crossing points exist, the left derivative is positive, and the right derivative is negative;

step 3: searching the coordinates of a left edge point and a right edge point of the N line below each sharp angle, calculating the width corresponding to the sharp angle according to the coordinates of the left edge point and the right edge point, judging the sharp angle as a flame sharp angle if the aspect ratio of the sharp angle is smaller than 0.5, and judging the sharp angle as noise or burr if the aspect ratio of the sharp angle is smaller than 0.5.

6. The mine external factor fire monitoring method with visible light visual characteristic fusion according to claim 1, wherein the flicker frequency obtaining process is as follows: searching fire suspected areas which tend to be the same in the t frame and the t-1 frame by adopting a similarity coefficient, and calculating the corresponding cumulative gray difference change rate of the fire suspected areas which tend to be the same, wherein the cumulative gray difference change rate is calculatedWherein F _j is the cumulative gray level difference change rate of the j fire suspected region; m, N is the total number of pixels in the j-th fire suspected region; v _t,i is the ith pixel value in the fire suspected area in the t frame image; v _t-1,i is the ith pixel value in the fire suspected area in the t-1 th frame image; Δt is the time interval between the t-1 th frame and the t-th frame.

7. The method for monitoring mine external factor fire disaster by fusing visual characteristics of visible light according to claim 1, wherein the similarity coefficient is as followsMu _t、μ_t-1 in the formula is the average gray scale of the suspected area of the fire source in the t frame and the t-1 frame respectively; sigma _t、σ_t-1 is the standard deviation of the fire source suspected region of the t frame and the t-1 frame respectively; c ₁,C₂ is a constant, avoiding denominator being 0; taking C ₁＝(K₁·L)²,C₂＝(K₂·L)² generally K ₁＝0.01,K₂ =0.03, L is the dynamic range of the gray scale image, and l=255.