Intelligent identification method and system for falling point of port dust fall fog gun and medium

Abstract

The invention provides an intelligent identification method and system for a falling point of a port dust fall fog gun and a medium, and relates to the technical field of fog gun falling point identification, and the method comprises the steps: constructing an improved MoilNetV2 network of a mixed cavity convolution layer based on odd-even mixing; determining a basic pixel concentration coefficient; determining a contrast coefficient according to the average pixel values of the whole image, the fog gun jet region and the non-fog gun jet region; according to the basic pixel concentration coefficient and the contrast coefficient, constructing a weighting coefficient of a pixel-concentration relationship, and weighting the labels of the training set; on the basis of the weighted training set and the improved MobilNetV2 network, training is carried out to obtain a fog gun jet flow segmentation model; acquiring to-be-identified image data, and extracting a fog gun jet region through the fog gun jet segmentation model; judging a drop point form in the fog gun jet flow area according to whether the target object is hit or not, and performing fitting to obtain a drop point coordinate. The method can effectively improve the accuracy of fog gun drop point identification.

Description

Technical Field

[0001] The invention relates to the technical field of fog cannon landing point recognition, and in particular to an intelligent recognition method, system and medium for the landing point of a fog cannon for port dust reduction. Background Art

[0002] The effectiveness and efficiency of fog cannons in port environment operations are easily significantly affected by a variety of environmental factors. Among them, wind force and wind direction are two key factors that have a direct impact on the stability and accuracy of the fog cannon jet. Since port areas are usually close to open waters, the wind force is often strong and the wind direction may change frequently. This dynamic environmental condition makes the operation of fog cannons complicated and challenging. The constant change in wind direction requires continuous monitoring and adjustment of the position and angle of the fog cannon to ensure that the jet can accurately cover the area that needs to be treated. However, due to the unpredictability of wind direction, it may be difficult to completely avoid the problem of jet drift even if real-time adjustments are made. Therefore, accurately obtaining the actual landing point of the fog cannon water mist can timely adjust the spray direction of the fog cannon, thereby effectively improving the dust reduction and dust suppression effects.

[0003] The existing fog cannon landing point identification methods are as follows: Traditional monitoring methods: In the early days, the fog cannon landing point was mainly determined by manual observation and experience. This method is inefficient, inaccurate, and greatly affected by human factors, making it difficult to achieve real-time and accurate landing point monitoring. Simple sensor application: Simple sensors such as position sensors are used to monitor some motion parameters of fog cannons and calculate the approximate landing point of fog cannons, but they can only obtain limited position information and cannot fully and accurately reflect the actual situation of the fog cannon landing point. They are not adaptable enough to complex environments and changing operating conditions.

[0004] In summary, the existing fog cannon landing point recognition accuracy is low and cannot adapt to complex environments and changing operating conditions. Summary of the invention

[0005] In order to solve the above problems, the present invention provides an intelligent identification method for the landing point of a fog cannon for dust reduction in ports, which can effectively improve the accuracy of fog cannon landing point identification.

[0006] To achieve the above object, the present invention provides the following technical solutions.

[0007] The present invention provides an intelligent identification method for the landing point of a port dust suppression fog cannon, comprising the following steps:

[0008] Constructing an improved MobilNetV2 network; the improved MobilNetV2 network changes the convolution layer of the original MobilNetV2 network into an odd-even mixed hole convolution layer;

[0009] The normalization method is used to construct the relationship between the pixel value and concentration distribution of the fog cannon jet water mist in the image, and the basic pixel concentration coefficient is determined; the contrast coefficient is determined according to the average pixel value of the entire image, the fog cannon jet area, and the non-fog cannon jet area;

[0010] According to the basic pixel concentration coefficient and contrast coefficient, the weighted coefficient of the pixel-concentration relationship is constructed, and the labels of the training set are weighted by the weighted coefficient; based on the weighted training set and the weighted loss function, the improved MobilNetV2 network is trained to obtain the fog cannon jet segmentation model;

[0011] Acquire the image data to be identified, and extract the fog cannon jet area through the fog cannon jet segmentation model;

[0012] According to the mist cannon jet area and the depth information of multiple jet trajectory points obtained by binocular vision, the jet trajectory of the mist cannon jet is linearly fitted to determine the mist cannon jet landing point and the spatial position information of the jet during the jet process.

[0013] Preferably, the determining of the spatial position information of the jet landing point and the jet during the jetting process of the fog cannon comprises the following steps:

[0014] The landing point pattern in the fog cannon jet area is determined by whether the target is hit:

[0015] When the fog cannon jet hits the target, its water mist produces a divergent phenomenon, and the water mist spreads from the central high-pressure area of ​​the hit area to the surrounding areas, so that the fog cannon jet is cut off and the jet end area is enlarged. The jet end area is the divergent jet landing point area;

[0016] When the fog cannon jet fails to hit the target, a direct jet landing point is generated. The depth information of multiple jet trajectory points in the fog cannon jet area is extracted by the binocular camera, and linear fitting is performed on the multiple jet trajectory points to determine the fog cannon jet landing point and the spatial position information of the jet during the direct jet process.

[0017] Preferably, the calculation of the convolution kernel size N of the mixed hole convolution layer is as shown in the following formula:

[0018] N = k + (k-1) (r-1)

[0019] In the formula, k is the width and height of the convolution kernel, and r is the expansion rate;

[0020] The calculation of the receptive field of each dilated convolutional layer is as follows:

[0021]

[0022] In the formula, l f is the receptive field size corresponding to the fth layer, S i is the step length of the i-th layer;

[0023] The expansion rate of the mixed hole convolution layer satisfies the following conditions:

[0024] M i =max[M i+1 -2r i ,M i+1 -2(M i+1 -2r i ),r i ]

[0025] In the formula, r i is the expansion rate of the i-th layer, M i Refers to the maximum expansion rate of the layer; assuming there are n layers of dilated convolutional layers, the default is M n =n, and M2≤k.

[0026] Preferably, the basic pixel concentration coefficient is solved as shown in the following formula:

[0027]

[0028] In the formula, R gc is the basic pixel concentration coefficient; G(x,y) is the pixel value of the fog cannon jet water mist area; min(G(x,y)) is the minimum pixel value of the fog cannon jet water mist area; G r (x, y) is the relative pixel value of the water mist area of ​​the fog cannon jet; max(G r (x, y)) is the maximum relative pixel value of the water mist area of ​​the fog cannon jet.

[0029] Preferably, the contrast coefficient is solved as shown in the following formula:

[0030]

[0031] In the formula, is the average pixel value of the entire image, is the average pixel value of the fog cannon jet area, is the average pixel value of the non-fog cannon jet area, and ε is the contrast coefficient; where the value range of ε is [0,1], reflecting the relative distance between the pixel mean of the fog cannon jet area and the pixel mean of the entire image.

[0032] Preferably, the weighted coefficient of the pixel-concentration relationship is solved as shown in the following formula:

[0033] λ=ε×R gc +1-ε

[0034] In the formula, λ represents the weight of concentration, and its value range is [1-ε, 1].

[0035] Preferably, the construction of the weighted loss function comprises the following steps:

[0036] Constructing the cross entropy loss function L CE :

[0037]

[0038] In the formula, y i ={0,1} is the input image pixel category label, where 1 represents the fog cannon jet water mist category and 0 represents the background category; is the predicted probability when the corresponding category label is 1;

[0039] Construct a weighted loss function L according to the weight coefficient W :

[0040]

[0041] Where N is the number of samples; is the predicted probability when the corresponding category label is 1; is the weight when the corresponding category label is 1;

[0042] According to the cross entropy loss function L CE And the weighted loss function L W Construct the weighted total loss function L:

[0043] L=αL CE +(1-α)L W

[0044] Where α is the control coefficient, which is used to determine the proportion of the weighted part in the total loss function.

[0045] Preferably, when the mist cannon jet does not hit the target object, a direct jet landing point is generated, depth information of multiple jet trajectory points in the mist cannon jet area is extracted by a binocular camera, and linear fitting is performed on the multiple jet trajectory points, including the following steps:

[0046] The depth information of multiple jet trajectory points in the spray area of ​​the fog cannon is extracted based on the binocular camera, and the coordinates of multiple jet trajectory points {(x1, y1), (x2, y2), (x3, y3), …, (x n ,y n )}, where x is the independent variable and y is the dependent variable;

[0047] Construct a linear curve:

[0048] f(x)=a1r1(x)+a2r2(x)+...+a n r n (x)

[0049] In the formula, rk(x) is a set of linearly independent functions, a k is the unknown coefficient, where k=1,2,···,m,m<n;

[0050] Determine the coordinate y i With f(x i ) is the sum of the squares of the distances J(a1,···,a m ) is the minimum fitting target:

[0051]

[0052] In order to minimize J, we can get m The equation is as follows:

[0053]

[0054] Other:

[0055]

[0056] A=[a1,···,a m ] T ,Y=(y1,···,y n ) T

[0057] Then, a1,···,a m The equation is transformed into:

[0058] R T RA=R T Y

[0059] The fitting curve is determined to be a polynomial curve based on the fact that the jet trajectory is close to a parabola:

[0060] f(x)=a1x m +…+a m x+a m+1

[0061] Then, R T RA=R T Y solves the unique solution A=(R T R) -1 R T Y, determine A=[a1,···,a m ] T , and obtain the trajectory curve fitting result of the fog cannon jet in the direct state.

[0062] The present invention also provides an intelligent identification system for the landing point of a port dust suppression fog cannon, the system comprising:

[0063] processor;

[0064] a memory having stored thereon a computer program executable on the processor;

[0065] Wherein, when the computer program is executed by the processor, the steps of the intelligent identification method for the landing point of the port dust reduction fog cannon are implemented.

[0066] The present invention also provides a computer-readable storage medium, on which a data processing program is stored. When the data processing program is executed by a processor, the steps of the intelligent identification method for the landing point of the port dust reduction fog cannon are implemented.

[0067] Beneficial effects of the present invention:

[0068] The present invention proposes an intelligent identification method, system and medium for the landing point of port dust suppression fog cannon. The method uses a hybrid void convolutional network to improve the traditional water jet trajectory segmentation algorithm, improves the MobileNetV2 network, makes it more suitable for fog cannon jet trajectory segmentation extraction, and effectively improves the accuracy of the MobileNet2 network in fog cannon jet recognition. When performing loss calculation, the method introduces the weight of the water mist concentration in the fog cannon jet in the input image to construct a loss function for network training, so that the network understands that the corresponding pixel points of the fog cannon jet it wants to identify are different, thereby reducing the influence of the uncertainty of the image data set labeling method on the network model's recognition of fog cannon jet water mist, and further improving the accuracy of the entire recognition method. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a flow chart of an intelligent identification method for the landing point of a port dust suppression fog cannon according to an embodiment of the present invention;

[0070] Figure 2 This is a schematic diagram of dust reduction by fog cannon radiation;

[0071] Figure 3 This is a schematic diagram of dust reduction by direct spraying of fog cannon;

[0072] Figure 4 This is the schematic diagram of the depth positioning of the binocular camera. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0074] Example 1

[0075] The existing control methods of port dust suppression fog cannon include the following means:

[0076] (1) Use manual and sensor monitoring and control

[0077] Dust suppression by fog cannon is the main method of dust suppression at present. The fog cannon has the functions of on-site manual, handle remote control and PLC remote automatic control. If dust is found manually during use, the fog cannon can be adjusted manually on-site or by handle remote control to spray at the dust location; dust concentration detection sensors can also be set in multiple fixed areas. If the dust concentration in a certain area is detected to be excessive, the fog cannon can be automatically adjusted by PLC to spray and suppress dust at the corresponding location. The common defect of the above two methods is that the action is delayed. The fog cannon is only started to spray and suppress dust after people observe or detect dust. In addition, the degree of automation of manual operation is low and depends on people. There are many uncertain factors. Dust concentration detection can only detect dust at a fixed location. When the area is large, many dust concentration detection sensors are installed, which is costly. In open areas such as ports, the on-site coal dust concentration detected by coal dust concentration sensors is highly random and unreliable.

[0078] (2) Monitoring and control by installing high-definition cameras

[0079] High-definition cameras are installed to monitor the port operations within the field of view, and the dust generation in the port can also be monitored. The data collected by the camera will be uploaded to the server for analysis, and the built-in algorithm will be used for identification and analysis, so as to achieve precise dust reduction for the operating vehicles and dust generation points. There are currently two main solutions for fog cannon spray control: fixed-point / swinging spray start and stop: when coal dust is found, the server sends start and stop and working mode instructions to the fog cannon PLC through the OPC protocol to realize the automatic start and stop of the fog cannon, and achieve water and energy saving while accurately reducing dust. Follow-up start and stop: when coal dust is found, the server calculates the spatial coordinate value of the vehicle, and assigns the fog cannon PLC to control the fog cannon to rotate through the OPC protocol, so as to achieve precise dust reduction with the vehicle. When the fog cannon intelligent identification and analysis system identifies an operating vehicle or a dust generation point, it will send a control signal to the fog cannon PLC controller, so as to achieve targeted dust reduction at the point. However, this method is mainly aimed at large-scale dust reduction, and does not have the function of positioning and identifying the landing point of the fog cannon, and cannot effectively reduce dust in areas with high concentration and small volume of coal dust.

[0080] (3) Through numerical calculation method

[0081] The empirical formula is fitted by simulating and calculating the basic physical information of the jet obtained by the sensor and the actual landing point information of the fog cannon. During use, the initial information is imported and the accurate landing point is calculated through the empirical formula. This method is more commonly used in specific engineering applications, but its scope of application is limited, and its accuracy is not high for complex jet conditions or new working conditions. The open-air port environment is greatly affected by factors such as wind, and the working position is not fixed, resulting in more variables. Therefore, this method is not suitable for open-air port environments.

[0082] However, the above-mentioned technical solutions cannot meet the port environment requirements and have the following technical problems:

[0083] Inaccurate location of the landing point: Traditional methods and simple sensors are difficult to accurately determine the actual landing point of the fog cannon's water mist, which makes it difficult to accurately evaluate and optimize the use of the fog cannon, affecting the actual effectiveness of dust reduction and dust suppression operations. At the same time, due to the special nature of the edge of the fog cannon jet, the traditional semantic segmentation algorithm is prone to large errors in the annotation process, resulting in reduced accuracy.

[0084] The definition of the landing point is vague: the best dust reduction landing point for port fog cannons is vague, and the coal dust generated in port operations is concentrated and highly random. However, traditional methods often rely on fixed monitoring points or regular manual inspections, and cannot determine the effective dust reduction position of the fog cannon in a timely manner, which can easily lead to miscalculation of the dust reduction effect and affect the overall dust reduction effect.

[0085] Lack of real-time performance: The inability to obtain fog cannon landing point information in real time makes it difficult for operators to adjust the working parameters and status of the fog cannon according to actual conditions in a timely manner, resulting in limited flexibility and response speed of operations. For example, in the event of sudden dust pollution, it is impossible to adjust the fog cannon landing point in time to effectively respond.

[0086] Poor environmental adaptability: Under different environmental conditions, such as light changes, bad weather, complex terrain, etc., existing technologies often find it difficult to accurately identify the landing point of fog cannons, and are easily interfered with, resulting in misjudgments or failure to work properly, reducing the reliability and stability of the system.

[0087] Low degree of intelligence: It cannot be effectively integrated and linked with other intelligent devices or systems, making it difficult to achieve automated control and management, which increases the complexity and workload of manual operations and does not conform to the development trend of modern intelligent operations.

[0088] Due to the particularity of the fog cannon jet and the port environment, the fog cannon jet has a wider coverage range than the traditional water cannon jet due to its atomization ability. However, this also makes its range shorter, the jet volume larger, the density lower, and it is easily affected by environmental factors such as the port's side wind. How to accurately find the exact landing point of the fog cannon jet and ensure the dust reduction efficiency of the fog cannon is bound to become an urgent problem to be solved when fog cannons are used in large areas of ports to reduce dust, and an important breakthrough for the application of fog cannons in outdoor open environments.

[0089] To this end, the present invention provides an intelligent identification method for the landing point of a port dust suppression fog cannon, and the specific process is as follows: Figure 1 As shown, the following steps are included:

[0090] S1: Construct an improved MobilNetV2 network; the improved MobilNetV2 network changes the convolution layer of the original MobilNetV2 network into a mixed odd-even mixed void convolution layer.

[0091] S2: Use the normalization method to construct the relationship between the pixel value and concentration distribution of the fog cannon jet water mist in the image, and determine the basic pixel concentration coefficient.

[0092] S3: Determine the contrast coefficient according to the average pixel values ​​of the entire image, the fog cannon jet area, and the non-fog cannon jet area.

[0093] S4: According to the basic pixel concentration coefficient and contrast coefficient, the weighted coefficient of the pixel-concentration relationship is constructed, and the labels of the training set are weighted by the weighted coefficient; based on the weighted training set and the weighted loss function, the improved MobilNetV2 network is trained to obtain the fog cannon jet segmentation model.

[0094] S5: Acquire the image data to be identified, and extract the fog cannon jet area through the fog cannon jet segmentation model.

[0095] S6: Determine the landing point form in the fog cannon jet area based on whether the target object is hit: when it is a divergent jet landing point, the end of the fog cannon jet area is the fog cannon jet landing point; when it is a direct jet landing point, extract the depth information of multiple jet trajectory points in the fog cannon jet area through the binocular camera, perform linear fitting on the multiple jet trajectory points, and determine the fog cannon jet landing point and the spatial position information of the jet during the direct jet process.

[0096] Specifically, in S1, the present invention adopts an improved MobilNetV2 network based on hybrid dilated convolution. The present invention uses the MobilNetV2 network structure to perform fog cannon jet segmentation.

[0097] The advantage of using the MobilNetV2 network structure for fog cannon jet segmentation is that the project can be started quickly and high performance can be achieved with limited resources, which is suitable for industrial environments where ports need to start projects quickly. Through fine-tuning and optimization, the pre-trained model is adjusted to suit various specific application scenarios.

[0098] When performing fog cannon jet trajectory segmentation at the port, it is necessary to obtain as much visual field information as possible in the environment while ensuring resolution. Therefore, the present invention uses dilated convolution, and downsampling is often used in deep networks to increase the receptive field and reduce the amount of calculation, but it will also reduce the spatial resolution. In order not to lose resolution and expand the receptive field at the same time, on the one hand, a large receptive field can segment large targets, and on the other hand, a high resolution can accurately locate targets, and dilated convolution can capture multi-scale contextual information by setting different expansion rates.

[0099] The calculation of the dilated convolution kernel size N is as follows:

[0100] N = k + (k-1) (r-1)

[0101] In the formula, k is the width and height of the convolution kernel, and r is the expansion rate;

[0102] The receptive field affects the level of detail of the information extracted by the network. The receptive field calculation formula for each dilated convolutional layer in the network is as follows:

[0103]

[0104] In the formula, l f is the receptive field size corresponding to the fth layer, S i is the step length of the i-th layer;

[0105] In order to reduce the checkerboard effect, the design of the mixed dilated convolution layer of the present invention follows the odd-even mixing so that it does not have the same common divisor. In order to avoid the dilation rate being too large during the first dilated convolution, which causes the entire network to lose detail information, the dilation rate of the dilated convolution meets the following conditions:

[0106] M i =max[M i+1 -2r i ,M i+1 -2(M i+1 -2r i ),r i ]

[0107] In the formula, r i is the expansion rate of the i-th layer, M i Refers to the maximum expansion rate of the layer; assuming there are n layers of dilated convolutional layers, the default is M n =n, and M2≤k, where k is the convolution kernel size.

[0108] In S2-S4, the present invention proposes a weighted method for the water mist concentration of a fog cannon jet.

[0109] Usually, for an object in semantic segmentation, all its pixels are equivalent. However, due to the diffusion characteristics of fog cannon jet water mist, when marking, the marking is more certain in places with high water mist concentration, and the lower the concentration, the less certain the marking is. Therefore, for water mist marking, its pixels are not equivalent. Therefore, the water mist image semantic segmentation network adds the weight of the water mist concentration in the fog cannon jet in the input image when calculating the loss, so that the network understands that the corresponding pixel points of the fog cannon jet it wants to identify are different, thereby reducing the impact of the uncertainty of the image data set marking method on the network model's recognition of fog cannon jet water mist. The present invention introduces a weighting factor to measure the degree of uncertainty of the marked pixels, avoiding the same treatment of all marked pixels, thereby more accurately identifying fog cannon jet water mist. The water mist concentration is directly related to the water mist pixel value in the camera image. The difference in water mist concentration in the same image can be expressed by the difference in water mist pixel value. Specifically:

[0110] Correctly establishing the relationship between the pixel value of the fog cannon jet water mist in the image and the concentration distribution of the fog cannon jet water mist is a necessary condition for introducing weights. The normalization method is used to establish the pixel-concentration relationship, and the basic pixel concentration coefficient is solved as shown in the following formula:

[0111]

[0112] G r (x,y)=G(x,y)-min(G(x,y))

[0113] In the formula, R gc is the basic pixel concentration coefficient; G(x,y) is the pixel value of the fog cannon jet water mist area; min(G(x,y)) is the minimum pixel value of the fog cannon jet water mist area; G r (x, y) is the relative pixel value of the water mist area of ​​the fog cannon jet; max(G r (x, y)) is the maximum relative pixel value of the water mist area of ​​the fog cannon jet.

[0114] In order to distinguish between fog cannon jets and water mist, the scene background information should also be included in the pixel concentration. Here, the contrast factor is introduced:

[0115]

[0116] In the formula, is the average pixel value of the entire image, is the average pixel value of the fog cannon jet area, is the average pixel value of the non-fog cannon jet area, and ε is the contrast coefficient; where the value range of ε is [0,1], reflecting the relative distance between the pixel mean of the fog cannon jet area and the pixel mean of the entire image. The background area of ​​the scene should be mainly coal piles. The greater the difference between the average pixel value of the fog cannon jet area and the average pixel value of the entire image, the greater the contrast coefficient, indicating that the fog cannon jet is easier to identify.

[0117] The solution of the weighting coefficient of the pixel-concentration relationship in the image is shown in the following formula:

[0118] λ=ε×R gc +1-ε

[0119] In the formula, λ represents the weight of concentration, and its value range is [1-ε, 1]. The lower limit of the pixel-concentration relationship is increased, and the confidence of the network model in the water mist of the fog cannon jet is improved.

[0120] In the training of the semantic segmentation network, it is necessary to evaluate the contribution of each pixel in the water mist area of ​​the fog cannon jet according to the weighted image and construct the weighted loss function through the weighting coefficient, which specifically includes the following steps:

[0121] Constructing the cross entropy loss function L CE :

[0122]

[0123] In the formula, y i ={0,1} is the input image pixel category label, where 1 represents the fog cannon jet water mist category and 0 represents the background category; is the predicted probability when the corresponding category label is 1.

[0124] Construct a weighted loss function L according to the weight coefficient W :

[0125]

[0126] Where N is the number of samples; is the predicted probability when the corresponding category label is 1; is the weight when the corresponding category label is 1.

[0127] According to the cross entropy loss function L CE And the weighted loss function L W Construct the weighted total loss function L:

[0128] L=αL CE +(1-α)L W

[0129] Where α is the control coefficient, which is used to determine the proportion of the weighted part in the total loss function.

[0130] Due to the particularity of the fog cannon jet and the port environment, the end of the jet area is not necessarily the fog cannon jet landing point, and further calculation is required through jet trajectory fitting. In the dust reduction process of the fog cannon, there is a jet water column regardless of whether the fog cannon jet hits the target object, and the fog cannon jet water column has a width and cannot be simply represented by a line; the jet is directly shot or hits the target, and the hit part is an area, not a simple point. Since the fog cannon jet and the target are both displayed two-dimensionally in the image information, the present invention divides the fog cannon landing point into two types: direct landing point and divergent landing point according to the fog cannon jet hitting information. Among them, the direct landing point refers to the effective dust reduction range formed when the fog cannon jet does not hit the target. When the fog cannon jet hits the target, the water mist produces a divergent phenomenon, that is, the water mist diffuses from the central high-pressure area of ​​the hitting area to the surrounding areas, so that the fog cannon jet is cut off, but the end area is enlarged. The divergent landing point is the end area of ​​the jet. Specifically, when it is a divergent jet landing point, such as Figure 2 As shown in the figure, the end of the fog cannon jet area is the fog cannon jet landing point. When it is a direct jet landing point, such as Figure 3 shown.

[0131] Due to the remarkable particularity of the port operation environment, the location where coal dust is generated often changes at any time. This is because the port operations are frequent and complex, and various loading, unloading, transportation and other operations may cause coal dust to be raised and spread. Therefore, the specific location where coal dust is generated is difficult to determine and has great uncertainty.

[0132] If the semantic segmentation algorithm analyzes that the fog cannon jet is a straight jet and the coal dust group happens to be on the trajectory of the fog cannon jet, then in this case, it is necessary to closely combine a detection method specifically for port coal dust. Through this detection method, the spatial position information of the coal dust group can be effectively obtained. With this accurate spatial position information, the angle of the fog cannon can be adjusted accordingly. In this way, the fog cannon can hit the target dust group in the air more accurately, and then carry out effective dust reduction operations. This method can greatly improve the efficiency and effect of dust reduction and ensure the cleanliness and safety of the port environment.

[0133] If after the semantic segmentation process, when the fog cannon jet accurately hits the coal pile, the fog cannon jet will diverge. Specifically, the fine water mist will gradually spread from the middle core area of ​​the fog cannon jet to the surrounding areas in all directions. Generally speaking, in such a situation, the distance between the fog cannon and the coal pile that needs to be treated for dust reduction is relatively close. At this time, the effective dust reduction area played by the fog cannon mainly covers the divergence area and the trajectory area passed by the initial jet of the fog cannon. In the present invention, in order to effectively reduce the calculation amount of the central processing unit (CPU), in this case, the dust reduction area is identified as the jet divergence area. This identification method not only takes into account the actual dust reduction effect, but also optimizes the allocation of computing resources to a certain extent, so that the entire system can function more efficiently during operation, and provides strong technical support for achieving good dust reduction effects.

[0134] In S6, the depth information of multiple jet trajectory points in the jet area of ​​the fog cannon is extracted by a binocular camera, and linear fitting is performed on the multiple jet trajectory points, including the following steps:

[0135] When the fog cannon jet does not hit the target object, the fog cannon jet does not encounter any obstacles, and the fog cannon jet will be in a parabolic form. The depth information of the fog cannon jet can be obtained by performing parallax calculation with the smart cameras on both sides of the fog cannon. The present invention uses a binocular camera to extract the depth information of multiple jet trajectory points in the fog cannon jet area. The principle diagram of binocular camera depth positioning is shown in FIG. Figure 4 As shown,

[0136] The distance between the projection centers of the two cameras is b, and the imaging point of any point P in the three-dimensional space on the left camera is P L , the imaging point of the right camera is P R The three-dimensional space point P is the intersection of the line connecting the projection center point of the two cameras and the imaging point. L and X R are the distances from the imaging points of the left and right cameras to the left imaging plane, respectively. The parallax of point P in the left and right cameras can be defined as follows:

[0137] d=|X L -X R

[0138] Two imaging points P L and P R The distance between them is:

[0139]

[0140] According to the theory of similar triangles, we can conclude that:

[0141]

[0142] Then we get the distance Z from point P to the center plane of the projection:

[0143]

[0144] The depth information of multiple jet trajectory points in the spray area of ​​the fog cannon is extracted based on the binocular camera, and the coordinates of multiple jet trajectory points {(x1, y1), (x2, y2), (x3, y3), …, (x n ,y n )}, where x is the independent variable and y is the dependent variable;

[0145] Construct a linear curve:

[0146] f(x)=a1r1(x)+a2r2(x)+...+a n r n (x)

[0147] In the formula, rk(x) is a set of linearly independent functions, a k is the unknown coefficient, where k=1,2,···,m,m<n;

[0148] Determine the coordinate y i With f(x i ) is the sum of the squares of the distances J(a1,···,a m ) is the minimum fitting target:

[0149]

[0150] In order to minimize J, we can get m The equation is as follows:

[0151]

[0152] Other:

[0153]

[0154] A=[a1,···,a m ] T ,Y=(y1,···,y n ) T

[0155] Then, a1,···,a m The equation is transformed into:

[0156] R T RA=R T Y

[0157] The fitting curve is determined to be a polynomial curve based on the fact that the jet trajectory is close to a parabola:

[0158] f(x)=a1x m +…+a m x+a m+1

[0159] The least squares method was used to fit the curve, R T RA=R T Y solves the unique solution A=(R T R)- 1 R T Y, determine A=[a1,···,a m ] T , and obtain the trajectory curve fitting result of the fog cannon jet in the direct state.

[0160] In summary, through the above jet trajectory fitting algorithm, the trajectory curve of the fog cannon jet in the direct injection state is obtained, so as to obtain the spatial position information of the jet during the direct injection process. Combined with a port coal dust identification method, the fixed-point dust reduction of centralized port coal dust can be achieved.

[0161] By improving the MobileNetV2 network trajectory feature extraction algorithm, the present invention effectively improves the accuracy of dust group identification. In combination with a fog cannon jet curve fitting method, the dust group location information is effectively identified and obtained. This creates the prerequisite for subsequent work. The above method of the present invention has the following technical effects:

[0162] Improve dust reduction efficiency and accuracy: By introducing the MobilNetV2 network structure and dilated convolution technology, the fog cannon jet can be semantically segmented efficiently. This technology not only improves the speed of image processing, but also ensures high-precision dust reduction area identification under limited computing resources.

[0163] Optimizing fog cannon operation strategy: By using the water mist concentration weighted method, the present invention can more accurately identify the water mist distribution in the fog cannon jet, thereby optimizing the fog cannon operation strategy. This method takes into account the effect of water mist concentration on the marking accuracy, so that the model can more accurately distinguish between the fog cannon jet and the background environment.

[0164] Enhance the robustness and adaptability of the system: Through the design of the mixed atrous convolution layer, the present invention reduces the checkerboard effect and enhances the system's adaptability to different expansion rates. This design enables the model to maintain good performance in different port environments.

[0165] Achieve accurate fog cannon jet trajectory fitting: The present invention proposes a fog cannon jet trajectory fitting method based on a polynomial curve, which can accurately calculate the landing point of the fog cannon jet. This method is applicable to both direct and divergent jet conditions, and improves the accuracy of fog cannon operation.

[0166] Reduce the difficulty of manual operation: Through automated camera data collection and intelligent algorithm processing, the present invention significantly reduces the difficulty and labor intensity of manual operation. Workers can easily monitor and manage fog cannon operations through the intuitive interface provided by the system, improving work efficiency.

[0167] The above is an intelligent identification method for the landing point of a port dust reduction fog cannon provided by one embodiment of the present embodiment. Based on the same idea, the present embodiment also provides a corresponding intelligent identification system for the landing point of a port dust reduction fog cannon. The specific definition of the intelligent identification system for the landing point of a port dust reduction fog cannon can be found in the above definition of the intelligent identification method for the landing point of a port dust reduction fog cannon, which will not be repeated here. Each module in the above-mentioned intelligent identification system for the landing point of a port dust reduction fog cannon can be fully or partially implemented by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0168] This embodiment also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 An intelligent identification method for the landing point of port dust suppression fog cannon is provided.

[0169] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0170] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent identification method for the landing point of a port dust suppression fog cannon, characterized in that: The following steps are involved: Constructing an improved MobilNetV2 network; the improved MobilNetV2 network changes the convolution layer of the original MobilNetV2 network into an odd-even mixed hole convolution layer; The normalization method is used to construct the relationship between the pixel value and concentration distribution of the fog cannon jet water mist in the image, and the basic pixel concentration coefficient is determined; the contrast coefficient is determined according to the average pixel value of the entire image, the fog cannon jet area, and the non-fog cannon jet area; According to the basic pixel concentration coefficient and contrast coefficient, the weighted coefficient of the pixel-concentration relationship is constructed, and the labels of the training set are weighted by the weighted coefficient; based on the weighted training set and the weighted loss function, the improved MobilNetV2 network is trained to obtain the fog cannon jet segmentation model; Acquire the image data to be identified, and extract the fog cannon jet area through the fog cannon jet segmentation model; According to the mist cannon jet area and the depth information of multiple jet trajectory points obtained by binocular vision, the jet trajectory of the mist cannon jet is linearly fitted to determine the mist cannon jet landing point and the spatial position information of the jet during the jet process.

2. The intelligent identification method for the landing point of the port dust suppression fog cannon according to claim 1 is characterized in that: The method of determining the landing point of the fog cannon jet and the spatial position information of the jet during the jetting process comprises the following steps: The landing point pattern in the fog cannon jet area is determined by whether the target is hit: When the fog cannon jet hits the target, its water mist produces a divergent phenomenon, and the water mist spreads from the central high-pressure area of ​​the hit area to the surrounding areas, so that the fog cannon jet is cut off and the jet end area is enlarged. The jet end area is the divergent jet landing point area; When the fog cannon jet fails to hit the target, a direct jet landing point is generated. The depth information of multiple jet trajectory points in the fog cannon jet area is extracted by the binocular camera, and linear fitting is performed on the multiple jet trajectory points to determine the fog cannon jet landing point and the spatial position information of the jet during the direct jet process.

3. The intelligent identification method for the landing point of the port dust suppression fog cannon according to claim 1 is characterized in that: The calculation of the convolution kernel size N of the hybrid hole convolution layer is shown in the following formula: N = k + (k-1) (r-1) In the formula, k is the width and height of the convolution kernel, and r is the expansion rate; The calculation of the receptive field of each dilated convolutional layer is as follows: In the formula, l f is the receptive field size corresponding to the fth layer, S i is the step length of the i-th layer; The expansion rate of the mixed hole convolution layer satisfies the following conditions: M i =max[M i+1 -2r i ,M i+1 -2(M i+1 -2r i ),r i ] In the formula, r i is the expansion rate of the i-th layer, M i Refers to the maximum expansion rate of the layer; assuming there are n layers of dilated convolutional layers, the default is M n =n, and M2≤k.

4. The intelligent identification method for the landing point of the port dust suppression fog cannon according to claim 1 is characterized in that: The solution of the basic pixel concentration coefficient is shown in the following formula: G r (x,y)=G(x,y)-min(G(x,y)) In the formula, R gc is the basic pixel concentration coefficient; G(x,y) is the pixel value of the fog cannon jet water mist area; min(G(x,y)) is the minimum pixel value of the fog cannon jet water mist area; G r (x, y) is the relative pixel value of the water mist area of ​​the fog cannon jet; max(G r (x, y)) is the maximum relative pixel value of the water mist area of ​​the fog cannon jet.

5. The intelligent identification method for the landing point of the port dust suppression fog cannon according to claim 4 is characterized in that: The solution of the contrast coefficient is shown in the following formula: In the formula, is the average pixel value of the entire image, is the average pixel value of the fog cannon jet area, is the average pixel value of the non-fog cannon jet area, and ε is the contrast coefficient; where the value range of ε is [0,1], reflecting the relative distance between the pixel mean of the fog cannon jet area and the pixel mean of the entire image.

6. The intelligent identification method for the landing point of the port dust suppression fog cannon according to claim 5 is characterized in that: The solution of the weighted coefficient of the pixel-concentration relationship is shown in the following formula: λ=ε×R gc +1-e In the formula, λ represents the weight of concentration, and its value range is [1-ε, 1].

7. The intelligent identification method for the landing point of the port dust suppression fog cannon according to claim 1 is characterized in that: The construction of the weighted loss function includes the following steps: Constructing the cross entropy loss function L CE : In the formula, y i ={0,1} is the input image pixel category label, where 1 represents the fog cannon jet water mist category and 0 represents the background category; is the predicted probability when the corresponding category label is 1; Construct a weighted loss function L according to the weighting coefficient W : Where N is the number of samples; is the predicted probability when the corresponding category label is 1; is the weight when the corresponding category label is 1; According to the cross entropy loss function L CE And the weighted loss function L W Construct the weighted total loss function L: L=αL CE +(1-α)L W Where α is the control coefficient, which is used to determine the proportion of the weighted part in the total loss function.

8. The intelligent identification method for the landing point of the port dust suppression fog cannon according to claim 2 is characterized in that: When the fog cannon jet does not hit the target object, a direct jet landing point is generated, and the depth information of multiple jet trajectory points in the fog cannon jet area is extracted by a binocular camera, and linear fitting is performed on the multiple jet trajectory points, including the following steps: The depth information of multiple jet trajectory points in the spray area of ​​the fog cannon is extracted based on the binocular camera, and the coordinates of multiple jet trajectory points {(x1, y1), (x2, y2), (x3, y3), …, (x n ,y n )}, where x is the independent variable and y is the dependent variable; Construct a linear curve: f(x)=a1r1(x)+a2r2(x)+...+a n r n (x) In the formula, rk(x) is a set of linearly independent functions, a k is the unknown coefficient, where k=1,2,···,m,m<n; Determine the coordinate y i With f(x i ) is the sum of the squares of the distances J(a1,···,a m ) is the minimum fitting target: In order to minimize J, we can get m The equation is as follows: Other: A=[a1,···,a m ] T ,Y=(y1,···,y n ) T Then, a1,···,a m The equation is transformed into: R T RA=R T Y The fitting curve is determined to be a polynomial curve based on the fact that the jet trajectory is close to a parabola: f(x)=a1x m +…+a m x+a m+1 Then, R T RA=R T Y solves the unique solution A=(R T R) -1 R T Y, determine A=[a1,···,a m ] T , and obtain the trajectory curve fitting result of the fog cannon jet in the direct state.

9. An intelligent identification system for the landing point of port dust suppression fog cannon, characterized in that: The system comprises: processor; a memory having stored thereon a computer program executable on the processor; Wherein, when the computer program is executed by the processor, the steps of the intelligent identification method for the landing point of the port dust reduction fog cannon as described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a data processing program, and when the data processing program is executed by a processor, the steps of the intelligent identification method for the landing point of a port dust reduction fog cannon as described in any one of claims 1 to 8 are implemented.

Images