Human-changed gas station monitoring method based on visual algorithm

Abstract

The invention relates to an artificially changed gas station monitoring method based on a visual algorithm, and belongs to the technical field of meteorological monitoring, and the method comprises the steps: S1, building a visible light camera nearby a gas station, and obtaining a video picture; s2, identifying the video picture through a target detection network, and judging whether a target object changing the monitoring index of the gas station exists or not; and S3, when the target object is detected, judging the residence time of the area in which the target object is detected, and if a judgment condition is met, reporting to carry out manual processing. The surroundings of the atmosphere detection station are monitored through the neural network vision algorithm, atmosphere detection errors caused by human factors are avoided, and the detection accuracy of the atmosphere detection station is improved.

Description

Technical Field

[0001] The invention relates to the technical field of meteorological monitoring, in particular to a method for monitoring a gas station with artificial changes based on a visual algorithm. Background Art

[0002] Urban atmospheric monitoring stations are usually deployed in key locations in the city, such as industrial areas, major traffic arteries and residential areas to monitor pollutants in the air, including particulate matter (PM2.5 and PM10), sulfur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO) and ozone (O3). These data are essential for formulating environmental protection policies, evaluating the effectiveness of pollution control measures, and protecting public health. Urban atmospheric monitoring stations (air quality monitoring stations) are an infrastructure for urban air environmental protection, and are facilities dedicated to collecting and analyzing the concentration of pollutants in the air. They are essential for assessing and controlling air quality. However, there may be some problems with artificial changes in the accuracy of detection at air quality monitoring stations, such as fog cannons around the detection stations, and people spraying water mist with water pipes. These behaviors will cause the detection data to be localized, which is different from the entire city atmosphere. In the end, the detection data is inaccurate, which will cause more serious problems, such as: Data distortion: If the background of the monitoring station is artificially changed, such as by planting a large number of trees near the monitoring station or using air purification equipment, the monitoring data may be distorted. Such data cannot accurately reflect the actual air quality conditions.

[0003] Misleading decision-making: Environmental policies and measures based on distorted data may lead to waste of resources and policy failure because they may not target the true sources of pollution and pollution levels.

[0004] Loss of public trust: If the public discovers that air quality data is being manipulated, it could negatively impact the credibility of governments and environmental protection agencies, reducing their support for environmental protection measures.

[0005] Legal liability: Artificially changing the background of a testing station to conceal the true level of pollution may violate environmental protection regulations, and the individuals or institutions involved may face legal liability.

[0006] Therefore, a set of methods and solutions are needed to address the problem of local deviations caused by human intervention. Summary of the invention

[0007] In order to solve the above-mentioned problems in the prior art and to improve the accuracy of meteorological monitoring, the present invention provides a method for artificially changing a gas station monitoring method based on a visual algorithm.

[0008] The artificial change gas station monitoring method based on visual algorithm includes the following steps: Step S1, set up a visible light camera near the gas station to obtain video images; Step S2: Identify the video image through the target detection network to determine whether there is a target object that changes the gas station monitoring index; Step S3: When a target object is detected, the residence time of the area where the target object is detected is judged. If the judgment condition is met, it is reported for manual processing.

[0009] Furthermore, the target object includes mist and spray.

[0010] Furthermore, the target detection network in step S2 uses resnet as the basic backbone network, uses akconv as the basic neuron to extract image features, and finally uses two convolutional layers to fuse the regression feature map as the Head part.

[0011] Furthermore, before the step S2 recognizes the video image through the target detection network, it also includes human body recognition on the image. When a person is recognized, the image area with the person is enlarged, and then the enlarged area of ​​the video image is recognized through the target detection network.

[0012] Furthermore, the human body recognition process is specifically as follows: resnet plus ca attention mechanism is used as the backbone network to extract image features, and two convolutional layers are used as the head part to fuse the regression feature map.

[0013] Furthermore, the use of resnet plus ca attention mechanism as the backbone network to extract human features of the image is specifically as follows: First, the input feature map is globally averaged pooled in both the width and height directions to obtain feature maps in both the width and height directions respectively; Secondly, the two parallel stages of width and height are merged, the width and height features are transposed to the same dimension, and then the width and height features are merged together by stacking, and the features are obtained by convolution + normalization + activation function; Then, separate the width and height parallel stages, separate the width and height features, and then transpose them to obtain two feature layers of width and height respectively; Finally, after adjusting the number of channels using 1x1 convolution, the sigmoid function is used to expand the width and height to obtain the attention in the width and height dimensions, and then multiply it with the original features to obtain the CA attention mechanism; Furthermore, before the step S2 recognizes the video image through the target detection network, it also includes identifying the drone in the image. When the drone is detected, the image area with the drone is enlarged, and then the enlarged area of ​​the video image is recognized through the target detection network.

[0014] Furthermore, the identification of drones in images is specifically as follows: resnet plus ca attention mechanism is used as the backbone network to extract image features, and two convolutional layers are used as the head part to fuse the regression feature map.

[0015] Furthermore, the determination of the dwell time in step S3 is specifically as follows: setting the time interval for camera recognition and the maximum threshold for recognizing the target object within a period of time, and reporting when the number of times the target object is recognized within a period of time exceeds the maximum threshold and the target position is not much different.

[0016] The beneficial effects of the present invention are as follows: the present invention monitors the surroundings of the atmosphere detection station through a neural network visual algorithm, avoids errors in atmosphere detection caused by human factors, and improves the accuracy of detection at the atmosphere detection station. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the monitoring method in Example 1; Figure 2 This is the target detection network structure diagram; Figure 3 This is a flow chart of the monitoring method in Example 4. DETAILED DESCRIPTION

[0018] 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.

[0019] Example 1, reference Figure 1 ,The method of monitoring gas station by artificial change based on visual ,algorithm includes the following steps: Step S1, set up a visible light camera near the gas station to obtain video images; Step S2: Identify the video image through the target detection network to determine whether there is a target object that changes the gas station monitoring index; Step S3: When a target object is detected, the residence time of the area where the target object is detected is judged. If the judgment condition is met, it is reported for manual processing.

[0020] In Example 1, the target object includes mist and spray.

[0021] In Example 1, Figure 2As shown, the target detection network in step S2 uses resnet as the basic backbone network, uses akconv as the basic neuron to extract image features, and finally uses two convolution layers to fuse the regression feature map as the head part. Because the water mist will always change its shape, a special convolution is used to extract features. Therefore, a changing shape will be given when initializing the convolution. A variable convolution kernel is used to give the convolution kernel any number of parameters and any sampling shape to solve the problem that the convolution kernel with a fixed sample shape and a square cannot adapt well to the changing target. In Example 1, the determination of the dwell time in step S3 is specifically as follows: setting the time interval for camera recognition and the maximum threshold for identifying the target object within a period of time, and reporting when the number of times the target object is recognized within a period of time exceeds the maximum threshold and the target position is not much different. For example, recognition is performed every 5 seconds, and when the number of times the mist and water spray recognition is recognized reaches more than 4 times within 5 minutes and the target position is not much different, reporting is performed.

[0022] Example 2, based on Example 1, in order to improve detection efficiency, before the video image is recognized through the target detection network, it also includes human body recognition on the image. When a person is recognized, the image area with the person is enlarged, and then the enlarged area of ​​the video image is recognized through the target detection network.

[0023] Among them, the human body recognition process is specifically as follows: using resnet plus ca attention mechanism as the backbone network to extract image features, and using two convolutional layers to fuse the regression feature map as the head part.

[0024] Among them, using resnet plus ca attention mechanism as the backbone network to extract the human features of the picture is as follows: First, the input feature map is globally averaged pooled in both the width and height directions to obtain feature maps in both the width and height directions respectively; Secondly, the two parallel stages of width and height are merged, the width and height features are transposed to the same dimension, and then the width and height features are merged together by stacking, and the features are obtained by convolution + normalization + activation function; Then, separate the width and height parallel stages, separate the width and height features, and then transpose them to obtain two feature layers of width and height respectively; Finally, after adjusting the number of channels using 1x1 convolution, the sigmoid function is used to expand the width and height respectively to obtain the attention in the width and height dimensions, and then multiply it with the original features to obtain the CA attention mechanism.

[0025] For example: Assume that the shape of the input feature layer is [H, W, C]. After average pooling in the width direction, the shape of the feature layer is [H, 1, C], thus mapping the feature to the high dimension; after average pooling in the height direction, the shape of the feature layer is [1, W, C], thus mapping the feature to the wide dimension. Then merge the two parallel stages, transpose the width and height to the same dimension, and then stack them to merge the width and height features together. At this time, we get the feature layer: [1, H+W, C], and use convolution + normalization + activation function to obtain the features. It should be noted that the number of convolution channels here is generally smaller, and scaling can reduce the number of parameters. The shape of the feature layer after convolution is [1, H+W, C / r], where r is the scaling factor. Then it is divided into two parallel stages again, and the width and height are separated into: [1, H,C / r] and [1, W, C / r], and then transposed. We obtain two feature layers [H, 1, C / r] and [1, W, C / r]. We then use 1x1 convolution to adjust the number of channels and take sigmoid to obtain the attention in the width and height dimensions. The former is expanded in width, and the latter is expanded in height. Then multiplying them together with the original features is the CA attention mechanism.

[0026] Among them, if global average pooling is used, then

[0027] in, Represents the output value of the cth channel after global average pooling; Represents the value of the i-th row and j-th column of the c-th channel in the original feature map F; H represents the length; W represents the width.

[0028] Example 3, based on Example 1, before the video image is recognized through the target detection network, it also includes identifying the drone in the image. When the drone is detected, the image area with the drone is enlarged, and then the enlarged area of ​​the video image is recognized through the target detection network.

[0029] The specific method of identifying drones in images is as follows: resnet plus ca attention mechanism is used as the backbone network to extract image features, and two convolutional layers are used as the head part to fuse the regression feature map.

[0030] Example 4, in order to solve the problem of local deviation caused by human changes and further improve efficiency, Figure 3 As shown, the artificial change gas station monitoring method based on visual algorithm includes: ①. Set up a visible light camera near the gas station to obtain pictures; ② Deploy fog and water spray recognition algorithms based on the camera ③. Check whether there is any person intruding in the video area ④. Detect whether there are drones in the air ⑤. If ③④ detects a target, zoom in on the target and pass it to ② to check if there is fog, spray or other objects that change the gas station monitoring indicators on the zoomed in interface ⑥Submit it to manual processing and proceed to the second step of confirmation.

[0031] In the description of the embodiments of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "center", "top", "bottom", "top", "bottom", "inside", "outside", "inner side", "outer side" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. Among them, "inside" refers to an internal or enclosed area or space. "Periphery" refers to the area surrounding a specific component or a specific area.

[0032] In the description of the embodiments of the present invention, the terms "first", "second", "third", and "fourth" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", and "fourth" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0033] In the description of the embodiments of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "install", "connect", "connect", and "assemble" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0034] In the description of the embodiments of the present invention, specific features, structures, materials or characteristics may be combined in a suitable manner in any one or more embodiments or examples.

[0035] In the description of the embodiments of the present invention, it should be understood that "-" and "~" represent a range between two values, and the range includes the endpoints. For example: "AB" represents a range greater than or equal to A and less than or equal to B. "A~B" represents a range greater than or equal to A and less than or equal to B.

[0036] In the description of the embodiments of the present invention, the term "and / or" herein is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " herein generally indicates that the associated objects before and after are in an "or" relationship.

[0037] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The method for monitoring a gas station by artificially changing the visual algorithm is characterized in that: The following steps are involved: Step S1, set up a visible light camera near the gas station to obtain video images; Step S2: Identify the video image through the target detection network to determine whether there is a target object that changes the gas station monitoring index; Step S3: When a target object is detected, the residence time of the area where the target object is detected is judged. If the judgment condition is met, it is reported for manual processing.

2. The method for monitoring a gas station by artificial changes based on a visual algorithm according to claim 1 is characterized in that: The target objects include mist and spray.

3. The method for monitoring a gas station by artificial changes based on a visual algorithm according to claim 1 is characterized in that: The target detection network in step S2 uses resnet as the basic backbone network, uses akconv as the basic neuron to extract image features, and finally uses two convolutional layers to fuse the regression feature map as the head part.

4. The method for monitoring a gas station by artificial changes based on a visual algorithm according to claim 1 is characterized in that: Before the step S2 recognizes the video image through the target detection network, it also includes human body recognition on the image. When a person is recognized, the image area with the person is enlarged, and then the enlarged area of ​​the video image is recognized through the target detection network.

5. The method for monitoring a gas station by artificially changing the gas station based on a visual algorithm according to claim 4 is characterized in that: The human body recognition process is specifically as follows: resnet plus ca attention mechanism is used as the backbone network to extract image features, and two convolutional layers are used as the head part to fuse the regression feature map.

6. The method for monitoring a gas station by artificial changes based on a visual algorithm according to claim 5 is characterized in that: The use of resnet plus ca attention mechanism as the backbone network to extract human features of images is specifically as follows: First, the input feature map is globally averaged pooled in both the width and height directions to obtain feature maps in both the width and height directions respectively; Secondly, the two parallel stages of width and height are merged, the width and height features are transposed to the same dimension, and then the width and height features are merged together by stacking, and the features are obtained by convolution + normalization + activation function; Then, separate the width and height parallel stages, separate the width and height features, and then transpose them to obtain two feature layers of width and height respectively; Finally, after adjusting the number of channels using 1x1 convolution, the sigmoid function is used to expand the width and height respectively to obtain the attention in the width and height dimensions, and then multiply it with the original features to obtain the CA attention mechanism.

7. The method for monitoring a gas station by artificial changes based on a visual algorithm according to claim 1, characterized in that: Before the step S2 recognizes the video image through the target detection network, it also includes identifying the drone in the image. When the drone is detected, the image area with the drone is enlarged, and then the enlarged area of ​​the video image is recognized through the target detection network.

8. The method for monitoring a gas station by artificial changes based on a visual algorithm according to claim 7 is characterized in that: The specific method of identifying drones in images is as follows: resnet plus ca attention mechanism is used as the backbone network to extract image features, and two convolutional layers are used as the head part to fuse the regression feature map.

9. The method for monitoring a gas station by artificial changes based on a visual algorithm according to claim 1, characterized in that: The determination of the dwell time in step S3 is specifically as follows: setting the time interval for camera recognition and the maximum threshold for recognizing the target object within a period of time, and reporting when the number of times the target object is recognized within a period of time exceeds the maximum threshold and the target position is not much different.

Images