Mine dust concentration video intelligent detection device and dust removal method

Abstract

The invention discloses a mine dust concentration video intelligent detection device, which comprises a mine light camera used for shooting a dust video in a mine; the central processing unit is connected with the mining light camera, and the central processing unit is used for processing and analyzing videos shot by the mining light camera so as to detect the dust concentration in the mine; the dust alarm is connected with the central processing unit and is used for acousto-optic alarm; and the dust removal equipment is connected with the central processing unit, and the dust removal equipment receives the instruction of the central processing unit and performs dust removal work. And dust concentration detection on areas with different concentrations under different mines is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of dust concentration detection, and in particular to a mine dust concentration video intelligent detection device and a dust removal method. Background Art

[0002] For a long period of time, coal will remain the main source of energy in my country. After nearly two decades of continuous exploitation of coal resources, most parts of my country have gradually entered the stage of deep mining in mines, with increasingly complex conditions and increasing risks of mine disasters. In GB / T51272-2018 "Design Standard for Intelligent Mines in Coal Industry", it is proposed that dust monitoring, that is, monitoring the dust concentration at each dust-producing location, is required; as we all know, due to the very small pore size of dust particles, it is easy to enter precision instruments, causing equipment paralysis. During the diffusion process, dust will absorb various gases, including harmful gases, which will seriously endanger human health when inhaled into the human body. At the same time, when the dust concentration reaches a certain level, dust explosions will occur under certain conditions, and even continuous explosions will occur.

[0003] Due to the complex underground environment, various factors will affect the dust concentration detection, such as light intensity, temperature, humidity, water mist particle size, etc. Now, by analyzing the error values ​​of various factors on dust concentration detection in different environments, the influence rules of dust concentration detection in different environments are obtained, and various error calibration methods are proposed according to the error generation rules, including: least squares method, error compensation algorithm, BP neural network method, etc. The errors of underground dust concentration detection are calibrated according to various factors respectively, and the accurate values ​​of dust concentration in various underground environments are obtained. Summary of the invention

[0004] The present application provides a mine dust concentration video intelligent detection device, comprising:

[0005] Lightweight mining camera, used to capture dust video inside the mine;

[0006] A central processing unit is connected to the light-weight camera for mining, and the central processing unit is used to process and analyze the video captured by the light-weight camera for mining to detect the dust concentration inside the mine;

[0007] A dust alarm, connected to the central processor, for sound and light alarm;

[0008] A dust removal device is connected to the central processing unit, and the dust removal device receives instructions from the central processing unit and performs dust removal work.

[0009] Optionally, the lightweight camera includes an AI processing chip, the AI ​​processing chip is respectively connected to the ambient light sensor, the humidity sensor and the power supply, and the AI ​​processing chip is connected to the central processing unit;

[0010] The power source is the light type used in mining.

[0011] Optionally, the dust removal equipment is an electric energy spray sprinkler device, and the electric energy spray sprinkler device includes a power supply, a main control circuit board, a water inlet pipe, a water outlet pipe and a spray sprinkler device;

[0012] The power supply is connected to the main control circuit board to provide stabilized direct current to the main control circuit board. A passive relay is installed on the main control circuit board. The passive relay is connected to an electric ball valve. The electric ball valve is installed at the water outlet of the spray sprinkler device. The electric ball valve is used to control the opening and closing of the spray sprinkler device.

[0013] The main control circuit board is connected to the central processing unit.

[0014] Optionally, the dust alarm includes a housing and a control unit, the control unit is connected to the power supply, the control unit includes a lithium manganate ion battery, the power supply charges the lithium manganate ion battery, the control unit is also provided with a control board multivibrator, the control board multivibrator is provided with two triode collector circuits, the triode collector circuit is connected to an LED light board, the LED light board is installed with an LED light tube, and the LED light tube provides a light alarm for the dust alarm;

[0015] The control board multivibrator is also connected to a digital power amplifier, the digital power amplifier is connected to a speaker, and the speaker provides an audible alarm for the dust alarm.

[0016] Optionally, the light-weight camera for mining is connected to the dust alarm via Ethernet, and when the light-weight camera for mining detects that the transmittance of the video image reaches a set threshold, the dust alarm is activated.

[0017] The present application also provides a method for intelligent detection of dust concentration in a mine by video, comprising the following steps:

[0018] Step A: Take an image with a lightweight camera for mining, and locate the dust area by using a light source extraction module in the central processor, distinguish the light source area and the non-light source area in the image, record the light source area as a bright channel image of the dust image, and record the non-light source area as a dark channel image of the dust image;

[0019] Step B: The transmittance is calculated by using the bright channel area prior method for the light source area, and the transmittance is calculated by using the dark channel area prior method for the non-light source area;

[0020] Step C: The brightness of the dust concentration area in the dark channel image of the dust image is greater than that of the dust-free area. The difference between the dark channel value of the clear image after the image defogging algorithm and the dark channel value of the dust image is obtained. The point with a high pixel value in the obtained image is the dust area in the original image, so as to obtain the transmittance corresponding to different dust concentrations and the corresponding dust concentration value;

[0021] Step D: calibrating the dust concentration value error in step C through a dual-parallel hidden layer BP neural network to reduce the dust concentration value detection error caused by changes in light intensity and humidity;

[0022] Step E: The dust concentration value collected by the mining camera is transmitted via the Ethernet, and compared with the dust concentration values ​​of different degrees pre-set in the central server to obtain the dust hazard level of the area to prevent dust from harming the working area.

[0023] Optionally, the method for locating the dust area in step A is specifically: performing defogging processing on the dust image taken by the mining light camera 1 to obtain a clear image, wherein the defogging formula is:

[0024]

[0025] Where J is the pixel of the clear image, A is the atmospheric light value, t(x) is the dark channel transmittance, and t 0 Set the threshold for the image pixels, I(x) is the atmospheric scattering model,

[0026] Optionally, the dark channel region transmittance is:

[0027]

[0028] Among them, I c (x) is the atmospheric scattering model on the RGB image, A is the atmospheric light value, Ω(x) represents the filter window with size x, and c∈{r, g, b} represents each channel on the RGB image.

[0029] Optionally, the transmittance of the bright channel region is:

[0030]

[0031] Among them, I c (x) is the atmospheric scattering model on the RGB image, A is the atmospheric light value, and c∈{r, g, b} represents each channel on the RGB image.

[0032] The beneficial effects of this application are:

[0033] The present invention detects dust images through a lightweight camera for mining, realizes dust detection in the detection area, and the camera detects the underground dust image in real time, so as to prevent the underground dust situation more efficiently. And by analyzing the detected data, the detected data is correspondingly warned and dust removed. The method can reduce the damage of dust to the equipment, can effectively collect the area with a wide detection space, and protect the dust removal equipment. In the process of dust removal, the passive relay in the dust removal device is controlled by the central processor connected to the lightweight camera for mining, and the electric ball valve in the dust removal nozzle is switched, and the corresponding dust removal treatment is performed for different dust concentrations. Secondly, in order to facilitate the staff to handle the dust, remote centralized control spray is adopted. The staff can observe the underground dust concentration through the camera on the ground, and the remote control equipment can more effectively handle the dust area. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of the structure of the invention;

[0035] Figure 2 This is a circuit diagram of the electric energy spray sprinkler device of the present invention;

[0036] Figure 3 The circuit principle block diagram of the intrinsically safe sound and light alarm for mining of the present invention is as follows;

[0037] Figure 4 This is a block diagram of the internal structure of the high frame rate camera for mining of the present invention;

[0038] Figure 5 It is a calibration flow chart of the present invention;

[0039] Figure 6 It is a dual-parallel hidden layer BP neural network of the present invention;

[0040] Figure 7 This is a flow chart of the dust area positioning method of the present invention;

[0041] Figure 8 This is a flow chart of the bright area extraction module of the present invention;

[0042] Fig. 9 This is a flow chart of calculating dust concentration in bright and dark channels of the present invention;

[0043] In the figure: 1. Light camera for mining; 2. Passive relay; 3. Power supply; 4. AI processing chip; 5. Ethernet electrical interface; 6. Dust alarm; 8. Water inlet pipe; 9. Water outlet pipe; 10. Spray sprinkler; 11. Electric ball valve; 12. Control board multivibrator circuit; 13. LED light-emitting tube; 14. Collector circuit of transistor; 15. Speaker; 16. Digital power amplifier; 17. Housing; 18. Lithium manganate ion battery; 19. LED light board; 20. Main control circuit board; 21. Automatic zoom lens; 22. Central processing unit; 23. Spray sprinkler nozzle; 24. Ambient light sensor; 25. Humidity sensor. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The present invention adopts a remote video detection method to detect dust in real time. The dust image is detected by a lightweight camera 1 for mining, and the dust concentration of the detection area is detected. The ambient light sensor 24 is installed at the lens of the lightweight camera 1 for mining to detect the light intensity of the area in the mine in real time. The lightweight camera 1 for mining is connected to a humidity sensor 25 installed underground, and the error correction is performed by comparing the detected dust concentration with the ambient humidity. At the same time, the light source area and the non-light source area in the image can be distinguished by the light source extraction module, and the transmittance values ​​of the light source area and other areas are calculated by using the bright channel prior and the dark channel prior respectively. The brightness of the dust concentration area in the dark channel image of the dust image is greater than that of the dust concentration area, and the difference is calculated by finding the dark channel value of the clear image after the image defogging algorithm and the dark channel value of the dust image. The point with a high pixel value in the obtained image is the dust area in the original image, thereby obtaining the transmittance corresponding to different dust concentrations, and the corresponding dust concentration can be judged according to the obtained image transmittance.

[0046] Then, the error caused by the underground light intensity and humidity on the dust concentration detection is eliminated through the dual-parallel hidden layer BP neural network, so as to prevent the underground dust situation more efficiently. And by analyzing the detected data, the detected data is correspondingly warned and dust-removed. This method can reduce the damage of dust to the equipment, effectively collect the detection area with a wide space, classify the results, and protect the dust removal equipment. In the process of dust removal, the passive relay 2 in the dust removal device is controlled by the central processor 22 connected to the mining lightweight camera 1, and the electric ball valve 11 in the dust removal nozzle is switched, and the corresponding dust removal treatment is carried out for different dust levels. Secondly, in order to facilitate the staff to handle the dust removal, remote centralized control spray is adopted. The staff can observe the underground dust situation through the camera on the ground, and the remote control equipment can handle the dust area more effectively.

[0047] The present invention provides a mine dust concentration video intelligent detection device, comprising:

[0048] Lightweight mining camera 1, used to shoot dust video inside the mine;

[0049] The light-weight camera 1 for mining includes an AI processing chip 4 , which is respectively connected to an ambient light sensor 24 , a humidity sensor 25 and a power supply 3 , and is also connected to the central processor 22 .

[0050] The light-duty mining camera 1 can use a camera of model KBA18 (C).

[0051] The power supply 3 is a mining-use cast-in-place and intrinsically safe DC regulated power supply, with a rated input voltage of AC127V (AC85V-264V), an output voltage of DC12V, an output current of 1.9A, and an aviation plug as the DC output interface with a working voltage indicator light.

[0052] A central processor 22, connected to the light-weight camera 1 for mining, and used for processing and analyzing the video captured by the light-weight camera 1 for mining to detect the dust concentration inside the mine;

[0053] A dust alarm 6, connected to the central processor 22, for sound and light alarm;

[0054] The light-weight camera 1 for mining is connected to the dust alarm 6 via Ethernet 5. When the light-weight camera 1 for mining detects that the transmittance of the video image reaches a set threshold, the dust alarm 6 is activated.

[0055] The electrical interface type of the Ethernet 5 is 4 10 / 100 / 1000M adaptive RJ45 ports, 2 Gigabit SFP ports, industrial-grade operating temperature: -40°C ~ 75°C, wide voltage input: 9.6V ~ 60VDC, IP30 protection: reduce the impact of dust.

[0056] The light-weight camera 1 for mining is connected to the central server 22 for information transmission via Ethernet 5 electrical signals.

[0057] The ambient light sensor 24 adopts LXD / GB3-A1DPS, which has an operating voltage range of 0 to 10 volts, an operating temperature of -25 to +90 degrees Celsius, a storage temperature of -30 to +100 degrees Celsius, and an ambient light intensity ranging from 1 lux to 100,000 lux.

[0058] The humidity sensor 25 is RS-YS-GPRS-Y with a power supply voltage of 10-30 VDC, a humidity detection range of 0% RH-100% RH, 4 probes, a humidity display resolution of 0.1% RH, and a humidity refresh time of 1 s.

[0059] The dust alarm 6 includes a housing 17 and a control unit, the control unit is connected to the power supply 3, the control unit includes a lithium manganate ion battery 18, the power supply 3 charges the lithium manganate ion battery 18, the control unit is also provided with a control board multivibrator 12, the control board multivibrator 12 is provided with two triode collector circuits 14, the triode collector circuit 14 is connected to an LED light board 19, the LED light board 19 is installed with an LED light tube 13, the LED light tube 13 provides a light alarm for the dust alarm 6;

[0060] The control board multivibrator 12 is also connected to a digital power amplifier 16 , and the digital power amplifier 16 is connected to a speaker 15 , and the speaker 15 provides an audible alarm for the dust alarm 6 .

[0061] The dust removal device is connected to the central processor 22, and the dust removal device receives instructions from the central processor 22 and performs dust removal.

[0062] The dust removal equipment is an electric energy spray sprinkler, which includes a power supply 3, a main control circuit board 20, a water inlet pipe 8, a water outlet pipe 9 and a spray sprinkler 10;

[0063] The power supply 3 is connected to the main control circuit board 20 to provide the main control circuit board 20 with stabilized direct current. The main control circuit board 20 is equipped with a passive relay 2, and the passive relay 2 is connected to the electric ball valve 11. The electric ball valve 11 is installed at the water outlet of the spray sprinkler 10. The electric ball valve 11 is used to control the opening and closing of the spray sprinkler 10.

[0064] The main control circuit board 20 is connected to the central processing unit 22 .

[0065] The light-weight camera 1 for mining is connected to the power supply of the electric energy spray sprinkler through Ethernet 5, and the dust concentration values ​​of the three levels are respectively: 0mg·m -3 -300mg·m -3 , 300mg·m -3 -700mg·m -3 , 700mg·m -3 -1000mg·m -3 The size of the electric ball valve 11 can be set. When it is detected that the dust concentration reaches the corresponding threshold, the corresponding electric energy spray sprinkler device is automatically started to remove dust.

[0066] In this embodiment, the AI ​​processing chip 4 may be NervanaNNP-I;

[0067] In this embodiment, the Ethernet 5 uses the ring network layer 2 management industrial switch TL-SG2206 industrial grade, and is installed by rail installation and wall mounting;

[0068] In this embodiment, the dust alarm 6 is a mining intrinsically safe sound and light alarm and is selected as KXB12.

[0069] In this embodiment, the spray sprinkler 10 uses the ZP-127Z mine-use isolation and intrinsically safe automatic sprinkler dust suppression device main control box, and the electric ball valve 11 uses the DFH-20 / 7 mine-use intrinsically safe electric ball valve and water circuits, air circuits, spray assembly systems and other parts.

[0070] The present invention provides a method for intelligent video detection of mine dust concentration based on a mine dust concentration video intelligent detection device, comprising the following steps:

[0071] Step A: take an image through a lightweight camera 1 for mining, and locate the dust area through a light source extraction module in the central processor 1, distinguish the light source area and the non-light source area in the image, record the light source area as a bright channel image of the dust image, and record the non-light source area as a dark channel image of the dust image;

[0072] Step B: The transmittance is calculated by using the bright channel area prior method for the light source area, and the transmittance is calculated by using the dark channel area prior method for the non-light source area;

[0073] Step C: The brightness of the dust concentration area in the dark channel image of the dust image is greater than that of the dust-free area. The difference between the dark channel value of the clear image after the image defogging algorithm and the dark channel value of the dust image is obtained. The point with a high pixel value in the obtained image is the dust area in the original image, so as to obtain the transmittance corresponding to different dust concentrations and the corresponding dust concentration value;

[0074] Step D: calibrating the dust concentration value error in step C through a dual-parallel hidden layer BP neural network to reduce the dust concentration value detection error caused by changes in light intensity and humidity;

[0075] Step E: The dust concentration value collected by the mining camera 1 is transmitted via the Ethernet 5 and compared with the dust concentration values ​​of different degrees pre-set in the central server 22 to obtain the dust hazard level of the area and prevent dust from harming the working area.

[0076] The dust area positioning method:

[0077] The first step is to dehaze the dust image to obtain a clear image;

[0078] Specifically, most of the pixels in the dust image I have at least one channel whose grayscale value has very low intensity and is close to 0, which is the dark channel prior theory. For a clear dust image J, its dark channel value can be expressed by the following formula:

[0079]

[0080] In the formula, J dark represents the dark channel value; J c Represents each channel of the RGB image; Ω(x) represents the filter window with size x.

[0081] The atmospheric scattering model is simplified as follows:

[0082] I(x)=J(x)t(x)+A(x)(1-t(x)) (2)

[0083] Divide both sides of equation (2) by the atmospheric light value A to obtain:

[0084]

[0085] Performing dark channel operation on both sides of equation (3) simultaneously, we obtain:

[0086]

[0087] Substituting formula (1) into formula (4), we can get the formula for calculating the dark channel transmittance:

[0088]

[0089] After obtaining the transmittance, the haze-free image can be restored.

[0090] However, under certain circumstances, the pixels of the defogging image may be too large due to the low transmittance. This problem can be solved by setting the threshold T0. When the transmittance value is less than T0, let t = T0. All the effect pictures in this article are calculated based on T0 = 0.1. Therefore, the final defogging formula is as follows:

[0091]

[0092] The second step is to calculate the dark channel map A1 of the clear image and the dark channel map B1 of the dust image.

[0093] The third step is to calculate the difference between Figure A1 and Figure B1 to obtain Figure C. Because Figure A1 is a dehazed image, the brightness area in Figure C is the dust area in the original image.

[0094] The fourth step is to take the pixel points with pixel values ​​greater than 150 in Figure C in the third step, mark them in the point set A, and find the corresponding coordinates.

[0095] The fifth step is to design a 3*3 window module D to detect the edge of the dust area.

[0096] Step 6: Take the minimum point of the horizontal coordinate among the marked points (if the same horizontal coordinate points appear, take the minimum point of the vertical coordinate) as the center point of the window module D.

[0097] The seventh step is to compare the pixel values ​​of the points around the center point of window module D, mark the pixel points in window module D that belong to point set A to point set B, find the two pixel blocks with the largest difference in prime values ​​between adjacent pixel blocks, move the center point of window module D to the larger pixel block, and mark the pixel coordinates at the center point of window module D to point set C.

[0098] Step 8. Repeat step 4 until the pixel point in point C appears again at the center point of window module D.

[0099] In the ninth step, the maximum and minimum values ​​of the horizontal and vertical coordinates in the point set B are calculated, and the value is determined to be the dust area. The points in this area are removed from the point set A, and the third to sixth steps are repeated until all the points in the point set A are processed.

[0100] The method for extracting the bright area is:

[0101] Since the dark channel prior theory is not effective for detecting light source areas, the bright channel prior theory is used to calculate the image transmittance in the light source area. In addition, due to the uneven lighting in the well and the reflections from various equipment and personnel in the well, the calculation of the video transmittance in the well is seriously affected, thereby reducing the accuracy of the detected dust concentration value. Therefore, it is necessary to distinguish different brightness areas in the image to reduce the impact of uneven lighting on the calculation of image transmittance. In this application, a bright area extraction module is designed to distinguish different bright areas. The bright channel prior algorithm is used to calculate the image transmittance in high-brightness areas, and the dark channel prior algorithm is used to calculate the image transmittance in low-brightness areas, thereby achieving accurate detection of dust concentrations distinguished by different brightness.

[0102] Specifically:

[0103] Step 1: Convert the original image into a grayscale image, set a 15*15 window, start from the upper left corner of the grayscale image, and move horizontally to the right. The 15*15 window can calculate the value of each pixel contained in the window and sum them up;

[0104] Step 2: Lower the pixel value in the 15*15 window in step 1 to position A, return the 15*15 window to the position where the pixel value is the largest, and start to translate downward. When the pixel value reaches the maximum, it reaches the light source;

[0105] Step 3: Change the 15*15 window in step 2 to a 16*16 window, and compare the brightness change between the edge of the 16*16 window and the 15*15 window in step 2:

[0106] If the brightness decreases, it is considered that the previous 15*15 window is the entire light source;

[0107] If the brightness increases, it is considered that the light source area is larger than the 15*15 window. Continue to increase the window area until the pixel value at the edge of the new window is smaller than the pixel value in the window. Calibrate the window size and position at this time, and this position is the final light source position and area size.

[0108] At this time, the window is restored to a window size of 15*15, and the window returns to the previous brightness drop point A, continues to translate to the right, repeats the previous movement, reaches the rightmost point and starts to move horizontally to the right from the 16th row on the left end of the grayscale image. When it reaches the previously calibrated light source area, it jumps directly to the right and continues to translate to the right until it reaches the lower right corner of the grayscale image and ends the movement, indicating that all light sources in the image have been found.

[0109] The dust concentration monitoring method is:

[0110] By using the dust area positioning and light source area extraction methods, it is determined whether the dust is in the light source area. If the dust exists in the light source area, a bright area dust concentration monitoring method is used;

[0111] If dust is not present in the light source area, use the dark area dust concentration monitoring method;

[0112] The dark area dust concentration monitoring method is:

[0113] From formula (5), the dark channel transmittance calculation formula can be obtained:

[0114]

[0115] The bright area dust concentration monitoring method is:

[0116] According to the bright channel prior theory, for most pixels in the dust image I, there is at least one channel whose grayscale value has a very high intensity and is close to the atmospheric light value of the corresponding fog-free image, and for the fog-free image J, its bright channel value is close to 1. For an image, the bright channel prior theory can be expressed by the following formula:

[0117]

[0118]

[0119] In the formula, I bright Indicates the bright channel value of the haze image; J bright (x) represents the bright channel value of a clear and fog-free image; I c and J c Represents each channel of the foggy and fog-free RGB images respectively; Ω(x) represents the filter window with size x; A bright (x) represents the atmospheric light value obtained by the bright channel prior theory.

[0120] By performing bright channel operation on both sides of equation (3), we obtain:

[0121]

[0122] Substituting formula (8) into formula (9), we can get the formula for calculating the transmittance of the bright channel:

[0123]

[0124] The dust concentration value is determined by the image transmittance in the dust area. In this embodiment, the lower the image transmittance, the higher the corresponding dust concentration value.

[0125] In this embodiment, a total of 540 dust images of various concentration values ​​under different light intensities and ambient humidity are collected, the image transmittance is calculated, and the transmittance and dust concentration formula is fitted as follows:

[0126] y=-3268.15×t+3331.2 (12)

[0127] Where y is the dust concentration value, and t is the image transmittance of each pixel i in the image.

[0128] Please refer to Figure 6 As shown, there are many factors that affect dust concentration detection, such as ambient humidity and light intensity. In this embodiment, ambient humidity and light intensity are selected as factors that affect dust concentration detection. The ambient humidity and light intensity are uncertain, discrete and random. In this application, according to the video dust concentration detection, a three-layer BP neural network model with dual parallel hidden layers is constructed, with the three factors of video detection dust concentration, ambient humidity and light intensity as the input layer, and the dust concentration detected by the video as the output layer, to obtain a 3-N-1 model, where N is the number of neurons in the hidden layer;

[0129] The specific method of building a dual-parallel hidden layer BP neural network is:

[0130] Step 1: Data normalization: In the dual-parallel hidden layer BP neural network, the neuron activation functions of the hidden layer and the output layer use logsig and purelin functions respectively. When the input threshold is too large or too small, the neuron output will enter a saturated state. Therefore, the original data needs to be normalized. The normalization method is

[0131] y=(xx min ) / (x max -x min ) (13)

[0132] In the formula: y is the normalized value; x is the original data value; x min is the minimum value of the original data; x max is the maximum value of the original data.

[0133] In this application, a total of 1000 sets of data were collected. According to the data division, the training set data is 600, the test set data is 200, and the validation set data is 200;

[0134] Step 2: Build a dual-parallel hidden layer BP neural network to calibrate the dust concentration error caused by the simultaneous effects of light intensity and humidity. The calibration steps are as follows:

[0135] a) The input layer is the values ​​corresponding to the dust concentration, light intensity and humidity detected by the video, and is input into hidden layers 1 and 2 at the same time. The output layer is the dust concentration detection value, and the output result error is compared.

[0136] b) Replace the network parameters with large errors with those with small errors.

[0137] c) Perform backpropagation on the hidden layer with smaller error, update the hidden layer parameters, and perform forward propagation again.

[0138] d) Compare the outputs of the two hidden layers. Use the hidden layer parameters with smaller errors to replace the network parameters with larger errors again until the output error meets the requirements.

[0139] The above are only a few embodiments of the present application and do not constitute any form of limitation to the present application. Although the present application is disclosed as above with preferred embodiments, it is not intended to limit the present application. Any technician familiar with the profession, without departing from the scope of the technical solution of the present application, using the technical content disclosed above to make slight changes or modifications are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A mine dust concentration video intelligent detection device, characterized in that: include: A light-weight mining camera (1) for capturing dust video inside a mine; A central processing unit (22) is connected to the light-weight mining camera (1), and the central processing unit (22) is used to process and analyze the video captured by the light-weight mining camera (1) to detect the dust concentration inside the mine; A dust alarm (6), connected to the central processor (22) and used for sound and light alarm; A dust removal device is connected to the central processor (22), and the dust removal device receives instructions from the central processor (22) and performs dust removal work.

2. The intelligent video detection device for mine dust concentration according to claim 1 is characterized in that: The lightweight camera (1) comprises an AI processing chip (4), wherein the AI ​​processing chip (4) is respectively connected to an ambient light sensor (24), a humidity sensor (25) and a power source (3), and the AI ​​processing chip (4) is connected to the central processing unit (22).

3. The intelligent video detection device for mine dust concentration according to claim 1 is characterized in that: The dust removal equipment is an electric energy spray sprinkler device, which comprises a power supply (3), a main control circuit board (20), a water inlet pipe (8), a water outlet pipe (9) and a spray sprinkler device (10); The power supply (3) is connected to the main control circuit board (20) to provide the main control circuit board 20 with stabilized direct current. A passive relay (2) is installed on the main control circuit board (20). The passive relay (2) is connected to an electric ball valve (11). The electric ball valve (11) is installed at the water outlet of the spray sprinkler device (10). The electric ball valve (11) is used to control the opening and closing of the spray sprinkler device (10). The main control circuit board (20) is connected to the central processing unit (22).

4. The intelligent video detection device for mine dust concentration according to claim 1 is characterized in that: The dust alarm (6) comprises a housing (17) and a control unit, the control unit being connected to the power source (3), the control unit comprising a manganese oxide lithium ion battery (18), the power source (3) charging the manganese oxide lithium ion battery (18), the control unit also being provided with a control board multivibrator (12), the control board multivibrator (12) being provided with two triode collector circuits (14), the triode collector circuit (14) being connected with an LED light board (19), the LED light board (19) being provided with an LED light tube (13), the LED light tube (13) providing a light alarm for the dust alarm (6); The control board multivibrator (12) is also connected to a digital power amplifier (16), and the digital power amplifier (16) is connected to a loudspeaker (15), and the loudspeaker (15) provides an audible alarm for the dust alarm (6).

5. The intelligent video detection device for mine dust concentration according to claim 4 is characterized in that: The light-weight camera (1) for mining is connected to the dust alarm (6) via Ethernet (5). When the transmittance of the video image detected by the light-weight camera (1) for mining reaches a set threshold, the dust alarm (6) is activated.

6. A method for intelligent detection of dust concentration in mines by video, characterized in that: The steps include: Step A: taking an image with a lightweight mining camera (1), and locating the dust area by distinguishing the light source area and the non-light source area in the image through a light source extraction module in the central processor (1), recording the light source area as a bright channel image of the dust image, and recording the non-light source area as a dark channel image of the dust image; Step B: The transmittance is calculated by using the bright channel area prior method for the light source area, and the transmittance is calculated by using the dark channel area prior method for the non-light source area; Step C: The brightness of the dust concentration area in the dark channel image of the dust image is greater than that of the dust-free area. The difference between the dark channel value of the clear image after the image defogging algorithm and the dark channel value of the dust image is obtained. The point with a high pixel value in the obtained image is the dust area in the original image, so as to obtain the transmittance corresponding to different dust concentrations and the corresponding dust concentration value; Step D: calibrating the dust concentration value error in step C through a dual-parallel hidden layer BP neural network to reduce the dust concentration value detection error caused by changes in light intensity and humidity; Step E: The dust concentration value collected by the mining camera (1) is transmitted via the Ethernet (5) and compared with the dust concentration values ​​of different degrees pre-set in the central server (22) to obtain the dust hazard level in the area and prevent dust from harming the working area.

7. The method for intelligent detection of mine dust concentration by video according to claim 6, characterized in that: The method for locating the dust area in step A is specifically as follows: defogging the dust image taken by the light-weight mining camera 1 to obtain a clear image, wherein the defogging formula is: Where J is the pixel of the clear image, A is the atmospheric light value, t(x) is the dark channel transmittance, t0 is the image pixel setting threshold, and I(x) is the atmospheric scattering model.

8. The method for intelligent detection of mine dust concentration by video according to claim 7, characterized in that: The dark channel area transmittance is: Among them, I c (x) is the atmospheric scattering model on the RGB image, A is the atmospheric light value, Ω(x) represents the filter window with size x, and c∈{r, g, b} represents each channel on the RGB image.

9. The method for intelligent video detection of mine dust concentration according to claim 7 is characterized in that: The transmittance of the bright channel region is: Among them, I c (x) is the atmospheric scattering model on the RGB image, A is the atmospheric light value, and c∈{r, g, b} represents each channel on the RGB image.

Images