CN111080651B - Automatic monitoring method of oil drilling polluted gas based on water flow segmentation - Google Patents

Abstract

The invention belongs to the technical field of automatic monitoring of petroleum drilling polluted gas, and discloses a water flow segmentation-based automatic monitoring method of petroleum drilling polluted gas. The water flow segmentation algorithm is used for segmenting water flow to obtain form information of the water flow, and the anomaly monitoring algorithm is realized through multiple gaussians and is used for identifying an abnormal data frame. Through the automatic monitoring algorithm of the petroleum drilling polluted gas based on water flow segmentation, the water flow of the petroleum drilling drainage pipeline is segmented, whether the polluted gas meets the emission standard or not is detected according to the segmentation result, real-time early warning can be realized when the polluted gas does not meet the emission standard, meanwhile, the problems that manual observation cannot be realized, and the water flow change form cannot be obtained by a water flow sensor in real time are solved, the labor cost can be effectively saved, and the environment is protected.

Description

Automatic monitoring method of oil drilling polluted gas based on water flow segmentation

technical field

The invention belongs to the technical field of automatic monitoring of polluted gas, and in particular relates to an automatic monitoring method for petroleum drilling polluted gas based on water flow segmentation.

Background technique

At present, the closest existing technology: pollutant discharge and treatment, which is related to the environment and people's livelihood, is an indispensable part of industrial production. In the process of industrial exploration and excavation, in addition to the underground useful resources, some pollutants, such as H ₂ S, will also be carried. H ₂ S gas is a poisonous and harmful acid gas, which is directly discharged into the air and has an impact on the surrounding environment. When the content of the polluting gas is higher than the emission standard, the emission should be treated. In addition to being toxic and harmful, H ₂ S gas is also a colorless, odorless gas that can be tolerated in water. The HSO ₄ formed by dissolving H ₂ S gas in water is also colorless, so it is impossible to detect the polluted gas by observing the color change of the water flow. In the actual production process, using chemical reagents directly for detection will consume a lot of chemical reagents. At the same time, it is also very difficult to achieve real-time monitoring. On the other hand, the water capacity of the polluted gas is limited, so when the amount of polluted gas is large, the pressure in the entire environment will increase, and the phenomenon of water jets will appear at the outlet of the drainage pipe. The larger the water flow rate, the more obvious the jet phenomenon, indicating that the greater the internal pressure of the drainage pipe, the greater the pollution gas content.

At present, the video data is collected by the camera, and then the polluted gas is monitored by manually monitoring the changes of the water flow in the drainage pipeline in real time in the background. On the other hand, in the process of industrial production, multiple work points are often working at the same time, so there are many monitoring points required, which requires a lot of human resources, resulting in a waste of human resources. At the same time, artificially watching surveillance video for a long time will cause visual fatigue and lack of concentration, which will lead to the problem of inadequate supervision. At present, it is judged whether the discharge of polluted gas meets the standard according to the change of water flow at the outlet of the drainage pipe. At present, in the monitoring and statistics of water flow, water flow sensors are mainly used. The water flow sensor can obtain statistics of water flow over a period of time. The main principle is that the sensor converts the induction of water flow into electrical signals, which are transmitted to the background, and the background analyzes the water flow data according to the transmitted signals. Its main function is for water flow display and flow accumulation calculation. But water flow sensors are generally used in electric water heaters, solar thermal water heaters, and household water flow statistics. When monitoring the change of water flow in the drainage pipeline, the water flow is suddenly sprayed, and the data obtained by the water flow sensor may be inaccurate. At the same time, due to the impact of the water flow, the installation of the sensor is also a problem. Therefore, the water flow sensor cannot be used to judge whether the pollutant gas discharge meets the standard.

To sum up, the problems existing in the prior art are: in the pollution gas monitoring, there are problems such as waste of human resources, high cost, visual fatigue, and the existing water flow sensor can not meet the actual needs through manual methods.

Most of the existing abnormal data detection methods based on video are in the abnormal behavior of the human body or the crowd, and most of them use the human body posture, skeleton, optical flow information, etc. to represent the change of motion. In practice, the difference between water flow and human body is that the shape of water flow is not fixed, so it is difficult to capture the change of water flow shape through the change of optical flow and other information, that is, the existing video abnormal event detection method is not suitable for polluted gas. monitor. At the same time, the background of the image is complex, involving different light intensities and changes in the background, so it is difficult to cut the background through traditional image algorithms. The present invention can obtain the image area of interest and cut the background through the deep learning method.

SUMMARY OF THE INVENTION

Aiming at the problems existing in the prior art, the present invention provides an automatic monitoring method for oil drilling polluted gas based on water flow segmentation.

The present invention is achieved in this way, an automatic monitoring method for oil drilling pollution gas based on water flow segmentation, and the automatic monitoring method for oil drilling pollution gas based on water flow segmentation comprises the following steps:

The first step is to do an AND operation between the mask and the original image, map the image to the HSV color representation space, and manually adjust the threshold to obtain batch annotation data;

The second step is to preprocess the marked data, adjust the length and width of the image to multiples of 32, convert the image to tensor format, normalize the data, and normalize the pixel values from 0 to 255 to 0 to 1;

In the third step, through the neural network training data, the neural network adopts the encoder-decoder structure to fully explore the potential information of the image, the input of the neural network is the original image, and the output is the segmented image;

In the fourth step, the contour of the segmented image is extracted, and the perimeter and area of the contour will be calculated;

The fifth step is to establish a multivariate Gaussian model by using the two quantities of perimeter and area extracted from the data frame without abnormality in the labeled data;

The sixth step is to input the obtained perimeter and area of the segmented image into the multivariate Gaussian model to determine whether the data frame is abnormal.

Further, the water flow data labeling of the oil drilling pollution gas automatic monitoring method based on water flow segmentation is modeled by making a mask image and HSV color space analysis, and at the same time labeling 100 to 1000 pieces of data in the same scene; HSV is used to pass the color space. The threshold is used to segment the water flow, and the mask is used to eliminate noise points whose color space threshold is in the same range as the water flow. Through the observation and analysis of the water flow sample data, it is concluded that the camera does not move within a certain time range in a fixed scene. The lighting conditions and background in the image are basically unchanged;

The HSV model maps the RGB color space to a conical space, from 0 to 255, representing the change of color from black to white; in the process of making the label of the data set, the pictures are classified into one category through the combination of mask and HSV through code pairs Labeling and processing; Mask is an image modeled based on a background image, each category and scene has a corresponding mask, and most of the background can be removed by directly ANDing the mask with the original image; HSV uses color The spatial threshold adjustment will further process the masked result to obtain an accurate description of the water flow shape.

Further, the water flow segmentation algorithm of the oil drilling pollution gas automatic monitoring method based on water flow segmentation adopts the Unet network structure, and the Unet network realizes pixel-level segmentation to obtain an accurate description of the water flow change; the U-net network consists of a contraction path and An expansion path consists of;

The shrinking path follows a typical convolutional network structure consisting of two repeated 3x3 convolution kernels, both using a rectified linear unit activation function and a 2x2 max-pooling operation with stride 2 for downsampling, and in In each downsampling step, the number of feature channels is doubled;

In the expansion path, each step includes up-sampling the feature map; convolution operation is performed with a convolution kernel of 22 to reduce the number of feature channels by half; then the corresponding cropped feature map in the cascade contraction path; Then two 33 convolution kernels are used for the convolution operation, and both use the ReLU activation function; in the last layer, the 1*1 convolution kernel is used for the convolution operation, and each 64-dimensional feature vector is mapped to the network. output layer.

Further, the automatic monitoring method for oil drilling pollution gas based on water flow segmentation calculates the mean value and variance of all training data during preprocessing, and subtracts the mean value from each picture and compares it with the variance; the data is normalized to 0 to 1, so that many pixels become zero;

The loss function of the neural network is defined as follows:

Pre=∑ _{i, j} P _r (i, j) (1)

Pte=∑ _i,j P _t (i, j) (2)

In the formula, Pre is the output image of the neural network in the training process, Pte represents the corresponding label, loss represents the loss; i, j represent the coordinate position of the pixel point; i=0 in formula (3), n represents the number of images involved in the calculation , const represents a non-zero constant, in order to ensure that the denominator is not 0 when the label is all black; it can be seen from formula (3) that the loss is smaller when the label is closer to the actual predicted value.

Further, the abnormal monitoring method of the oil drilling pollution gas automatic monitoring method based on water flow segmentation adopts the method of multivariate Gaussian modeling. When the frame is an abnormal data frame, the area and perimeter of the frame will change; The model will deviate from the Gaussian model when testing abnormal data; the Gaussian model is calculated according to the perimeter and area of the water flow obtained by normal segmentation; a frame of data of the water flow monitoring video is judged to be abnormal, which is directly determined by the shape and change of the water flow ; Perimeter and area are the two most intuitive quantities to express the shape of an object. In practice, a multivariate Gaussian model is established through the two quantities of perimeter and area to explore the connection between the two quantities. The calculation formula of the multivariate Gaussian is:

Among them, ∑: represents the covariance matrix calculated according to the perimeter and area, u: represents the mean, x: represents the input, ∈: represents the threshold value of the judgment; the formula is used to determine whether the input x belongs to the established multivariate Gaussian distribution, when the output When the probability value is lower than ∈, it is determined that it does not belong to the probability distribution, and ∈ is determined according to the statistical value of normal data and abnormal data.

Another object of the present invention is to provide an automatic monitoring system for oil drilling pollution gas based on water flow segmentation that implements the automatic monitoring method for oil drilling pollution gas based on water flow segmentation, and the automatic monitoring system for oil drilling pollution gas based on water flow segmentation Include the following steps:

Annotation data acquisition module, which is used to perform AND operation between the mask and the original image, map the image to the HSV color representation space, and manually adjust the threshold to obtain batch annotation data;

The data preprocessing module is used to preprocess the marked data, adjust the length and width of the image to multiples of 32, convert the image into tensor format, and normalize the data;

The segmentation image acquisition module is used to train data through the neural network. The neural network adopts the encoder-decoder structure to fully explore the potential information of the image. The input of the neural network is the original image, and the output is the segmented image;

The contour calculation module is used to extract the contour of the segmented image, and will calculate the perimeter and area of the contour;

The model building module is used to establish a multivariate Gaussian model by using the two quantities of perimeter and area extracted from the data frame without abnormality in the labeled data;

The data frame determination module is used to input the obtained perimeter and area of the segmented image into the multivariate Gaussian model to determine whether the data frame is abnormal.

Another object of the present invention is to provide an information data processing terminal for realizing the automatic monitoring method for oil drilling polluted gas based on water flow segmentation.

Another object of the present invention is to provide a computer-readable storage medium, including instructions, which, when run on a computer, cause the computer to execute the automatic monitoring method for oil drilling polluted gas based on water flow segmentation.

Another object of the present invention is to provide an application of the automatic monitoring method for oil drilling polluted gas based on water flow segmentation in automatic monitoring of polluted gas.

Another object of the present invention is to provide an application of the automatic monitoring method for oil drilling polluted gas based on water flow segmentation in pollutant discharge treatment.

To sum up, the advantages and positive effects of the present invention are as follows: the water flow data labeling method of the present invention can label 100 to 1000 pieces of data in the same scene at the same time by making mask images and HSV color space analysis and modeling, saving a lot of manpower resource. The water flow segmentation algorithm mainly adopts the Unet network structure, which can achieve pixel-level segmentation and obtain an accurate description of the water flow change. The abnormality monitoring method adopts the multivariate Gaussian modeling method. Model.

The invention replaces manual real-time monitoring of the water flow of the drainage pipeline by means of machine vision. The invention obtains the jet state information of the water flow by dividing the water flow image, and after the division is completed, whether the current pollution gas discharge meets the standard is obtained through mathematical statistics and analysis combined with the relationship between the actual pollution gas discharge standard and the jet state.

The technical problem to be solved by the present invention is that there are waste of human resources, high cost, visual fatigue in the manual method, and the existing water flow sensor cannot obtain the change of the water flow in real time. The purpose is to obtain the jet of water flow by dividing the water flow image. After the status information is divided, whether the current pollutant gas discharge meets the emission standard is obtained through mathematical statistics and analysis combined with the relationship between the actual pollutant gas emission standard and the injection state.

Through the automatic monitoring algorithm of oil drilling polluted gas based on water flow segmentation, the invention divides the water flow of the oil drilling drainage pipeline, and then detects whether the polluted gas meets the emission standard according to the segmentation result, and can give real-time early warning when the polluted gas does not meet the discharge standard. It overcomes the problems of lack of supervision in manual observation and the inability of water flow sensors to obtain the changing shape of water flow in real time, which can effectively save labor costs (one detection machine can replace the workload of 5-6 people) and protect the environment.

Description of drawings

FIG. 1 is a flowchart of an automatic monitoring method for oil drilling polluted gas based on water flow segmentation according to an embodiment of the present invention.

FIG. 2 is a structural diagram of a segmentation network provided by an embodiment of the present invention.

FIG. 3 is a training accuracy diagram provided by an embodiment of the present invention.

FIG. 4 is a training loss diagram provided by an embodiment of the present invention.

5 is a segmentation result diagram provided by an embodiment of the present invention;

In the figure: (a) original image; (b) segmentation result; (c) original image; (d) segmentation result.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

In view of the existing problems in the prior art, the present invention provides an automatic monitoring method for oil drilling polluted gas based on water flow segmentation. The present invention is described in detail below with reference to the accompanying drawings.

As shown in FIG. 1 , the automatic monitoring method for oil drilling pollution gas based on water flow segmentation provided by the embodiment of the present invention includes the following steps:

S101: Perform an AND operation on the mask and the original image, then map the image to the HSV color representation space, and finally manually adjust the threshold to obtain batch annotation data;

S102: Preprocess the marked data, first adjust the length and width of the image to multiples of 32, then convert the image into tensor format, and finally normalize the data, and normalize the pixel value from 0 to 255 Between 0 and 1, the length and width of the image are adjusted to meet the needs of neural network training;

S103: Through the neural network training data, the neural network adopts the encoder-decoder structure to fully explore the potential information of the image, the input of the neural network is the original image, and the output is the segmented image;

S104: Extract the contour of the segmented image, and calculate the perimeter and area of the contour. In practice, the abnormal situation is judged by manually observing the change of the shape of the water flow, and the most basic representation of the morphological feature is the perimeter and area;

S105: Establish a multivariate Gaussian model by using the two quantities of perimeter and area extracted from the data frame without abnormality in the labeled data. The multivariate Gaussian model can fully explore the relationship between the perimeter and the area of the image, and at the same time according to the output probability The size of the judgment is abnormal;

S106: Input the obtained perimeter and area of the segmented image into a multivariate Gaussian model to determine whether the data frame is abnormal.

The technical solutions of the present invention will be further described below with reference to the accompanying drawings.

The present invention is an automatic monitoring algorithm for oil drilling polluted gas based on water flow segmentation, including three parts: a water flow data labeling method suitable for water flow, a water flow segmentation algorithm, and an abnormality monitoring algorithm. The shape of water flow data is not fixed, and there will be large changes with the blowout phenomenon. Therefore, the method of labeling by Labelme is not applicable. It is proposed to label the data by combining mask and HSV. The water flow segmentation algorithm is used to segment the water flow at the outlet of the drainage pipe to obtain an accurate description of the water flow shape. The abnormality monitoring algorithm detects whether the polluted gas at the current moment meets the emission standard according to the corresponding relationship between the segmentation result and the polluted gas.

The main process of the present invention's automatic monitoring algorithm for oil drilling pollution gas based on water flow segmentation is as follows: (1) first perform an AND operation on the mask and the original image, then map the image to the HSV color representation space, and finally manually adjust the threshold to obtain batches of Label the data; (2) Preprocess the labelled data, first adjust the length and width of the image to be multiples of 32, then convert the image to tensor format, and finally normalize the data to change the pixel value from 0 to 255 is normalized to between 0 and 1, and the length and width of the image are adjusted to meet the needs of neural network training; (3) through the neural network training data, the neural network adopts the encoder-decoder structure to fully explore the image’s characteristics. Potential information, the input of the neural network is the original image, and the output is the segmented image; (4) The contour of the segmented image is extracted, and the perimeter and area of the contour will be calculated. In practice, the abnormal situation is judged by manually observing the change of the water flow shape. , and the most basic representation of morphological features is the perimeter and area of an object; (5) Use the two quantities of perimeter and area extracted from the data frame without anomalies in the labeled data to establish a multivariate Gaussian model, a multivariate Gaussian model. The model can fully explore the relationship between the perimeter and the area of the image, and at the same time determine whether it is abnormal according to the output probability. (6) Input the obtained perimeter and area of the segmented image into the multivariate Gaussian model to determine whether the data frame is abnormal.

In the labeling stage of the water flow data set, the existing tool Labelme is directly used for labeling, and the labeled objects are shown in the segmentation network structure diagram in Figure 2. The left side is the original image and the right side is the Labelme labeling result. It can be seen from Figure 2 that only the rough outline information of the water flow can be obtained by using the tool for labeling, while other parts such as the small water column next to it and the water droplets that are separated from the central area due to jetting will be lost. The workload of labeling data is huge, so multiple people are required to cooperate and each person has a biased understanding of the water flow pattern.

The water flow data labeling method, mainly through the production of mask images and HSV color space analysis and modeling, can label 100 to 1000 pieces of data in the same scene at the same time, saving a lot of human resources. HSV is used to segment the water flow through the threshold of the color space, while the mask is used to eliminate noise points whose color space threshold is in the same range as the water flow. Through the observation and analysis of the water flow sample data, it can be concluded that within a certain time range in a fixed scene, the lighting conditions and background in the image are basically unchanged when the camera does not move. At the same time, referring to related papers on flame segmentation, we know that HSV is a better color representation space, and has achieved good results in the field of flame segmentation. Based on the above conditions, there is a basis for segmenting the water flow as a data label in a fixed scene. The HSV model maps the RGB color space to a conic space, from 0 to 255 representing the change in color from black to white. In the process of making dataset labels, the images that are classified into one category are labeled and processed by the combination of mask and HSV through code. The classification standard is that there is no change in the position of the camera, no change in the background, and little change in light conditions. Mask refers to an image modeled based on a background image. Each category and scene has a corresponding mask. Most of the background can be removed by directly ANDing the mask with the original image. The sample image of the mask is shown in Figure 3 after training accuracy. HSV further processes the result of mask processing by adjusting the threshold value of the color space to obtain an accurate description of the water flow shape.

The water flow segmentation algorithm mainly adopts the Unet network structure, and the neural network structure is shown in Figure 4 after training loss. The Unet network can achieve pixel-level segmentation and obtain accurate descriptions of water flow changes. The U-net network consists of a shrinking path (left) and an expanding path (right). Among them, the shrinking path follows the typical convolutional network structure, which consists of two repeated 3x3 convolution kernels (no padding convolution), and both use a rectified linear unit (ReLU) activation function and a stride for downsampling A 2x2 max-pooling operation of 2, and the number of feature channels is doubled at each downsampling step. In the expansion path, each step consists of up-sampling the feature map; then convolution operation (up-convolution) is performed with a convolution kernel of 22 to reduce the number of feature channels by half; then the corresponding The cropped feature map; then two 33 convolution kernels are used for convolution operation, and both use the ReLU activation function. Due to the missing problem of boundary pixels in each convolution operation, it is necessary to crop the feature map. In the last layer, a 1*1 convolution kernel is used for convolution operation, and each 64-dimensional feature vector is mapped to the output layer of the network.

The main purpose of data preprocessing is to reduce the amount of data for operation and to normalize the data at the same time. In preprocessing, the mean and variance of all training data are calculated first, and then the mean is subtracted from each image, and finally the variance is compared. Through the above processing, the data can be normalized to be between 0 and 1, and at the same time, many pixels will become zero, thereby reducing the amount of data in the operation. The loss function of the neural network is defined as follows:

Pre=∑ _{i, j} P _r (i, j) (1)

Pte=∑ _i,j P _t (i, j) (2)

In the formula, Pre is the image of the output of the neural network during the training process, Pte represents the corresponding label, and loss represents the loss. i,j represent the coordinate position of the pixel point. In formula (3), i=0, n represents the number of images involved in the calculation, and const represents a non-zero constant to ensure that the denominator is not 0 when the label is all black. It can be seen from formula (3) that the loss is smaller when the label is closer to the actual predicted value.

The Adam optimization algorithm is used in neural network optimization. Adam is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process. It can iteratively update the neural network weights based on training data. Adam algorithm is different from traditional stochastic gradient descent. Stochastic gradient descent maintains a single learning rate (i.e. alpha) to update all weights, and the learning rate does not change during training. Adam, on the other hand, designs independent adaptive learning rates for different parameters by calculating the first-order moment estimation and second-order moment estimation of the gradient. The proponents of the Adam algorithm describe it as a set of advantages of two stochastic gradient descent extensions, namely: the adaptive gradient algorithm retains a learning rate for each parameter to improve performance on sparse gradients (i.e. natural language and computer vision problems) . Root mean square propagation adaptively preserves the learning rate for each parameter based on the mean of the nearest magnitudes of the weight gradients. This means that the algorithm has excellent performance on non-stationary and online problems. Adam is a very popular algorithm in the deep learning field because it achieves good results very quickly. The empirical results show that the Adam algorithm has excellent performance in practice and has great advantages over other kinds of stochastic optimization algorithms. Good results have also been achieved in the actual training process. The training accuracy and loss graphs are shown in Figures 3 and 4 below.

As can be seen from Figures 3-4, 95% accuracy can be achieved with only 20 training sessions. The higher the accuracy, the better the water flow segmentation effect. In Figure 3-4, the solid line represents the training accuracy, and the dots represent the verification accuracy. It can be seen that the training accuracy and the verification accuracy almost increase at the same time, that is, the data does not have overfitting. The solid line represents the training loss, and the dashed line represents the validation loss. The simultaneous decrease of the training loss and the validation loss also verifies that the model does not overfit the data. It can be seen from Figure 3-Figure 4 that the neural network model converges quickly, and only 20 times of training can achieve high accuracy.

After training in Figures 3-4, the segmentation results are shown in Figure 5; in Figure 5: (a) the original image; (b) the segmentation result; (c) the original image; (d) the segmentation result.

At present, data in five scenarios have been collected from the field and trained, and some results have been obtained. At present, there are nearly 70,000 data samples. The experimental equipment is titan v, and the framework is keras. It only takes 20 hours to run 150 training runs on a Titan v. Because five upsampling and (2, 2) pooling are performed in the network structure, the input image size needs to be a multiple of 32. The size of the input images during training is (320, 448). In the test phase, the average processing time of a single frame of image is 4ms, which can meet the real-time requirements.

The abnormality monitoring method adopts the multivariate Gaussian modeling method. mold. The Gaussian model is calculated from the perimeter and area of the water flow obtained by normal segmentation. To judge a frame of data of the water flow monitoring video as abnormal, it can be directly determined by the shape and change of the water flow. Perimeter and area are the two most intuitive quantities to express the shape of an object. In practice, a multivariate Gaussian model is established to explore the relationship between the two quantities through the perimeter and the area. The formula for calculating the multivariate Gaussian is as follows:

Among them, ∑: represents the covariance matrix calculated according to the perimeter and area, u: represents the mean, x: represents the input, and ∈: represents the judgment threshold. This formula is used to determine whether the input x belongs to the established multivariate Gaussian distribution, and when the output probability value is lower than ∈, it is determined that it does not belong to the probability distribution. ∈ is determined based on the statistics for normal data and abnormal data.

The Gaussian model is established by perimeter and area, and the methods of using perimeter and area alone or combining the two to establish a multivariate Gaussian model to discriminate abnormal data frames are tested and compared. The comparison results are shown in Table 1. The data used contains a total of 3000 frames of images, of which the number of abnormal data frames is 200. When the perimeter and area are combined to build a multivariate Gaussian model, the detection accuracy reaches 96%.

Table 1 Detection results of abnormal water flow data frame

Qualification precision false detection rate Missing detection rate area 90% 15% 10% perimeter 93% 12% 7% perimeter + area 96% 12% 4%

The automatic monitoring method for oil drilling polluted gas based on water flow segmentation provided by the present invention includes three parts: a water flow data labeling method suitable for water flow, a water flow segmentation algorithm, and an abnormality monitoring algorithm. The water flow data labeling method is used to label water flow data whose shape changes at any time. It can label hundreds to thousands of frames of data at the same time, which can save a lot of time. The water flow segmentation algorithm is used to segment the water flow to obtain the morphological information of the water flow, and the abnormal monitoring algorithm is realized by multivariate Gaussian, which is used to identify abnormal data frames. Through the automatic monitoring algorithm of oil drilling pollution gas based on water flow segmentation, the water flow of oil drilling drainage pipeline is divided and then according to the segmentation results, whether the polluted gas meets the emission standard can be detected in real time. The problems of lack of supervision in manual observation and the inability of water flow sensors to obtain the changing shape of water flow in real time can effectively save labor costs and protect the environment.

The above descriptions are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (5)

1. an automatic monitoring method for oil drilling polluted gas based on water flow division, is characterized in that, the described oil drilling pollution gas automatic monitoring method based on water flow division comprises the following steps: The first step is to do an AND operation between the mask and the original image, map the image to the HSV color representation space, and manually adjust the threshold to obtain batch annotation data; The second step is to preprocess the marked data, adjust the length and width of the image to multiples of 32, convert the image to tensor format, normalize the data, and normalize the pixel values from 0 to 255 to 0 to 1; The third step is to train the data through the neural network. The neural network adopts an encoder-decoder structure. The input of the neural network is the original image, and the output is the segmented image; In the fourth step, the contour of the segmented image is extracted, and the perimeter and area of the contour will be calculated; The fifth step is to establish a multivariate Gaussian model by using the two quantities of perimeter and area extracted from the data frame without abnormality in the labeled data; The sixth step is to input the obtained perimeter and area of the segmented image into the multivariate Gaussian model to determine whether the data frame is abnormal; the water flow data of the method for automatic monitoring of oil drilling pollution gas based on water flow segmentation is marked by making a mask image and HSV color. Spatial analysis and modeling, marking 100 to 1000 pieces of data in the same scene at the same time; HSV is used to segment the water flow through the threshold of the color space, and mask is used to eliminate the noise points whose color space threshold is in the same range as the water flow. The observation and analysis of the water flow sample data shows that within a certain time range in a fixed scene, the lighting conditions and background in the image remain unchanged when the camera does not move; The HSV model maps the RGB color space to a conical space, from 0 to 255, representing the change of color from black to white; in the process of making the label of the data set, the pictures are classified into one category through the combination of mask and HSV through code pairs Labeling and processing; mask is an image modeled according to the background image, there is a corresponding mask under each category and scene, and the mask is directly ANDed with the original image; HSV processes the mask through the threshold adjustment of the color space The post-results are processed to obtain an accurate depiction of the water flow pattern; The abnormal monitoring method of the oil drilling pollution gas automatic monitoring method based on water flow segmentation adopts the method of multivariate Gaussian modeling. When the frame is an abnormal data frame, the area and perimeter of the frame will change; the Gaussian model obtained through normal data is tested in Abnormal data will deviate from the Gaussian model; the Gaussian model is calculated according to the perimeter and area of the water flow obtained by normal segmentation; a frame of data of the water flow monitoring video is judged to be abnormal and directly determined by the shape and change of the water flow; the perimeter and area are the two most intuitive quantities to express the shape of an object. In practice, a multivariate Gaussian model is established through the two quantities of perimeter and area to explore the connection between the two quantities. The calculation formula of the multivariate Gaussian is:

Among them, ∑: represents the covariance matrix calculated according to the perimeter and area, u: represents the mean, x: represents the input, ∈: represents the threshold value of the judgment; the formula is used to determine whether the input x belongs to the established multivariate Gaussian distribution, when the output When the probability value is lower than ∈, it is determined that it does not belong to the multivariate Gaussian distribution, and ∈ is determined according to the statistical values of normal data and abnormal data; The water flow segmentation algorithm of the oil drilling pollution gas automatic monitoring method based on water flow segmentation adopts the U-net network structure, and the U-net network realizes pixel-level segmentation to obtain an accurate description of the water flow change; The path is composed of an expansion path; The shrinking path follows a typical convolutional network structure consisting of two repeated 3x3 convolution kernels, both using a rectified linear unit activation function and a 2x2 max-pooling operation with stride 2 for downsampling, and in In each downsampling step, the number of feature channels is doubled; In the expansion path, each step includes upsampling the feature map; convolution operation is performed with a 2x2 convolution kernel to reduce the number of feature channels by half; then the corresponding cropped feature map in the cascade contraction path; Then two 3x3 convolution kernels are used for convolution operation, and both use ReLU activation function; in the last layer, 1*1 convolution kernel is used for convolution operation, and each 64-dimensional feature vector is mapped to the network. output layer. 2. the oil drilling pollution gas automatic monitoring method based on water flow segmentation as claimed in claim 1 is characterized in that, the oil drilling pollution gas automatic monitoring method based on water flow segmentation calculates the mean value of all training data during preprocessing and variance, each image is subtracting the mean, compared with the variance; normalizing the data to 0 to 1, so that the pixels become zero; The loss function of the neural network is defined as follows: Pre=∑ _{i, j} P _r (i, j) (1) Pte=∑ _i,j P _t (i, j) (2)

In the formula, Pre is the output image of the neural network in the training process, Pte represents the corresponding label, loss represents the loss; i, j represent the coordinate position of the pixel point; i=0 in formula (3), n represents the number of images involved in the calculation , const represents a non-zero constant, in order to ensure that the denominator is not 0 when the label is all black; it can be seen from formula (3) that the loss is smaller when the label is closer to the actual predicted value. 3. An automatic monitoring system for oil drilling pollution gas based on water flow segmentation for implementing the automatic monitoring method for oil drilling pollution gas based on water flow segmentation according to any one of claims 1 to 2, wherein the oil drilling based on water flow segmentation The drilling pollution gas automatic monitoring system includes the following steps: Annotation data acquisition module, which is used to perform AND operation between the mask and the original image, map the image to the HSV color representation space, and manually adjust the threshold to obtain batch annotation data; The data preprocessing module is used to preprocess the marked data, adjust the length and width of the image to multiples of 32, convert the image into tensor format, and normalize the data; The segmented image acquisition module is used to train data through a neural network. The neural network adopts an encoder-decoder structure. The input of the neural network is the original image, and the output is the segmented image; The contour calculation module is used to extract the contour of the segmented image, and will calculate the perimeter and area of the contour; The model building module is used to establish a multivariate Gaussian model by using the two quantities of perimeter and area extracted from the data frame without abnormality in the labeled data; The data frame determination module is used to input the obtained perimeter and area of the segmented image into the multivariate Gaussian model to determine whether the data frame is abnormal; The water flow data labeling of the oil drilling pollution gas automatic monitoring method based on water flow segmentation is modeled by making a mask image and HSV color space analysis, and labeling 100 to 1000 pieces of data in the same scene at the same time; HSV is used for threshold segmentation through the color space The water flow is out, and the mask is used to eliminate the noise points in the same range as the color space threshold and the water flow. Through the observation and analysis of the water flow sample data, it is concluded that in a certain time range in a fixed scene, the image when the camera does not move The lighting conditions and the background are unchanged; The HSV model maps the RGB color space to a conical space, from 0 to 255, representing the change of color from black to white; in the process of making the label of the data set, the pictures are classified into one category through the combination of mask and HSV through code pairs Labeling and processing; mask is an image modeled according to the background image, there is a corresponding mask under each category and scene, and the mask is directly ANDed with the original image; HSV processes the mask through the threshold adjustment of the color space The post-results are processed to obtain an accurate depiction of the water flow pattern; The abnormal monitoring method of the oil drilling pollution gas automatic monitoring method based on water flow segmentation adopts the method of multivariate Gaussian modeling. When the frame is an abnormal data frame, the area and perimeter of the frame will change; the Gaussian model obtained through normal data is tested in Abnormal data will deviate from the Gaussian model; the Gaussian model is calculated according to the perimeter and area of the water flow obtained by normal segmentation; a frame of data of the water flow monitoring video is judged to be abnormal and directly determined by the shape and change of the water flow; the perimeter and area are the two most intuitive quantities to express the shape of an object. In practice, a multivariate Gaussian model is established through the two quantities of perimeter and area to explore the connection between the two quantities. The calculation formula of the multivariate Gaussian is:

Among them, ∑: represents the covariance matrix calculated according to the perimeter and area, u: represents the mean, x: represents the input, ∈: represents the threshold value of the judgment; the formula is used to determine whether the input x belongs to the established multivariate Gaussian distribution, when the output When the probability value is lower than ∈, it is determined that it does not belong to the multivariate Gaussian distribution, and ∈ is determined according to the statistical values of normal data and abnormal data; The water flow segmentation algorithm of the oil drilling pollution gas automatic monitoring method based on water flow segmentation adopts the U-net network structure, and the U-net network realizes pixel-level segmentation to obtain an accurate description of the water flow change; The path is composed of an expansion path; The shrinking path follows a typical convolutional network structure consisting of two repeated 3x3 convolution kernels, both using a rectified linear unit activation function and a 2x2 max-pooling operation with stride 2 for downsampling, and in In each downsampling step, the number of feature channels is doubled; In the expansion path, each step includes upsampling the feature map; convolution operation is performed with a 2x2 convolution kernel to reduce the number of feature channels by half; then the corresponding cropped feature map in the cascade contraction path; Then two 3x3 convolution kernels are used for convolution operation, and both use ReLU activation function; in the last layer, 1*1 convolution kernel is used for convolution operation, and each 64-dimensional feature vector is mapped to the network. output layer. 4 . An information data processing terminal for realizing the automatic monitoring method for oil drilling polluted gas based on water flow segmentation according to any one of claims 1 to 2 . 5. A computer-readable storage medium, comprising instructions that, when executed on a computer, cause the computer to execute the automatic monitoring method for oil drilling polluted gas based on water flow segmentation according to any one of claims 1-2.

Images