A method for monitoring the status of a gas exchange cock switch

Abstract

The present invention provides a method for monitoring the switch status of a gas exchange cock, including: a robot equipped with a visible light camera and a supplemental lighting device patrols a task point, collects images of the gas exchange cock device and valve device status identification values, and obtains an original device image; enhances a local area of ​​the original device image to obtain preprocessed image data; detects the text on the valve identification plate based on the preprocessed image data, and uses a neural network-based segmentation algorithm to obtain text area position information; performs text content recognition based on the text area position information to obtain a valve device number for distinguishing multiple groups of devices; establishes a one-to-one correspondence between the valve device status identification value and the valve device number; and determines the valve status based on the rotation angles of the two identification plates, and combines the two with the valve device number to obtain a valve monitoring result. This method enables real-time monitoring of the valve's switch status and whether a fault has occurred.

Description

Technical Field

[0001] The present invention relates to the field of gas switches, and in particular to a method for monitoring the state of a gas exchange cock switch. Background Art

[0002] The gas exchange cock is a specialized piece of equipment for coke ovens, and monitoring its on / off status plays a crucial role in production safety. The coking industry faces safety risks in coke oven basements due to gas leaks and explosions, as well as the heavy workload of daily inspections and maintenance. Therefore, the development and application of robot-based imaging technology for equipment monitoring can prevent dangerous accidents at the source.

[0003] This method extracts valve identifiers through connected region segmentation and then calculates the minimum bounding rectangle of the connected region. The coordinate parameters of the rectangle are used to determine the valve device's on / off status in the image's two-dimensional coordinate system. Before determining the connected region, this method first requires binarization of the image. This binarization algorithm requires setting a threshold: image grayscale values ​​above the threshold are set to 255, and image grayscale values ​​below the threshold are set to 0. The selection of the threshold directly affects the effectiveness of the binarization process. Natural scenes are subject to unstable lighting conditions, and binarization with a constant threshold results in excessive or missing neighboring pixels, erroneous information about key targets, and ultimately affects the determination of the connected region. This method struggles to detect valve identifiers in industrial scenes with natural lighting.

[0004] Valve stem target tracking determines stem displacement, focusing only on stem deviation in a specified direction. The valve's open / close status is determined based on the stem displacement and its pre-set threshold. The image acquisition device and valve stem are mounted relatively parallel to each other. This recognition method is not suitable for identifying the rotation mechanism of a gas exchange cock. The operating state of a gas exchange cock is determined by the status of two devices. Each device rotates between 0° and 90°. Normal operation is at either 0° or 90°, while an abnormality indicates an angle between 0° and 90°. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a method for monitoring the status of a gas exchange cock switch that overcomes the above problems or at least partially solves the above problems.

[0006] According to one aspect of the present invention, a method for monitoring the status of a gas exchange cock switch is provided, comprising:

[0007] A robot equipped with a visible light camera and supplementary lighting equipment patrols the task point, collects images of the gas exchange cock equipment and valve equipment status identification values, and obtains the original image of the equipment;

[0008] enhancing a local area of ​​the original image of the device to obtain pre-processed image data;

[0009] Detecting the valve identification plate text based on the preprocessed image data, and obtaining text area position information using a segmentation algorithm based on a neural network;

[0010] Perform text content recognition based on the text area position information to obtain valve device numbers for distinguishing multiple groups of devices;

[0011] Matching the valve device status identification value with the valve device number one by one;

[0012] The state of the valve is determined according to the rotation angles identified by the two identification plates, and combined with the valve equipment number to obtain the valve monitoring result.

[0013] Optionally, enhancing a local area of ​​the original image of the device to obtain pre-processed image data specifically includes:

[0014] Limiting the histogram height limits the local contrast, limits the image noise amplification, and prevents excessive local contrast enhancement:

[0015] Set the adaptive histogram equalization sliding window size to M*M;

[0016] Derivative the local mapping function to obtain a histogram and the slope S of the local mapping function;

[0017]

[0018] Limiting the slope of the local mapping function is equivalent to limiting the histogram height and limiting the contrast intensity;

[0019]

[0020] Set the maximum slope S max , get the maximum height of the histogram, cut off the height above H max Histogram, thus enhancing local contrast within time;

[0021]

[0022] Optionally, detecting the valve identification plate text according to the preprocessed image data and using a neural network-based segmentation algorithm to obtain text area position information specifically includes:

[0023] The image is input into the neural network, and feature maps of different sizes are obtained through bottom-up convolution operations and top-down upsampling;

[0024] The feature map obtained by convolution is fused with the feature map obtained by upsampling, and convolution is used to eliminate the aliasing effect, and the resulting feature map is unified into a 1 / 4 size feature map F;

[0025] After convolution and transposition convolution operations on the 1 / 4 size feature map F, the segmentation probability map P and threshold map T are obtained;

[0026] The segmentation probability map P and the threshold map T are processed to obtain an approximate binary map;

[0027] Approximate the standard binary step function to ensure differentiability and obtain an approximate binary graph;

[0028]

[0029] The segmentation probability map P, the threshold map T and the approximate binary map are trained through neural network supervised learning to update weight parameters;

[0030] Test reasoning to obtain the bounding rectangle coordinate information pt1, pt2, pt3, pt4 of the target area text;

[0031] Calculate the deflection angle relative to the current image coordinate system, where the deflection angle is between [0° and 90°];

[0032] judging the switch state and abnormal state of the valve device according to the deflection angle;

[0033] k=-(pt2[1]-pt1[1]) / (pt2[0]-pt1[0])

[0034] res=arctan(k)*57.29577.

[0035] Optionally, the performing of text content recognition based on the text area position information to obtain valve device numbers for distinguishing multiple groups of devices specifically includes:

[0036] Scaling the image containing the text region position information to a size of 32×W×1 to obtain an image sequence P1;

[0037] The image sequence P1 is processed through the convolutional network to obtain a feature sequence P2 of size 1×(W / 4)×512;

[0038] Set T = (W / 4), D = 512, use a bidirectional LSTM network to learn the feature sequence, and output the predicted true value distribution R;

[0039] The prediction distribution is converted into a final label sequence for each frame through a fully connected network and activation function, which is then input into the CTC module and classified in time series to obtain the prediction result.

[0040] Convert the nameplate content to the equipment number.

[0041] The present invention provides a method for monitoring the switch status of a gas exchange cock, including: a robot equipped with a visible light camera and a supplemental lighting device patrols a task point, collects images of the gas exchange cock device and valve device status identification values, and obtains an original device image; enhances a local area of ​​the original device image to obtain preprocessed image data; detects the text on the valve identification plate based on the preprocessed image data, and uses a neural network-based segmentation algorithm to obtain text area location information; performs text content recognition based on the text area location information to obtain a valve device number for distinguishing multiple groups of devices; establishes a one-to-one correspondence between the valve device status identification value and the valve device number; and determines the valve status based on the rotation angles of the two identification plates, and combines the two with the valve device number to obtain a valve monitoring result. This method enables real-time monitoring of the valve's switch status and whether a fault has occurred.

[0042] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 A flow chart of a method for monitoring the status of a gas exchange cock switch provided by an embodiment of the present invention;

[0045] Figure 2 The embodiment of the present invention provides a design-assisted identification sign and text meaning. DETAILED DESCRIPTION

[0046] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0047] The terms "comprises" and "comprising" and any variations thereof in the description, embodiments, claims and drawings of the present invention are intended to cover non-exclusive inclusions, for example, including a series of steps or units.

[0048] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0049] like Figure 1 As shown, the visible light camera gimbal is equipped with a supplementary light device to obtain image data including the gas exchange cock device;

[0050] Enhance the overall image, suppress uninteresting features, and obtain pre-processed image data;

[0051] Valve identification plate text area detection, using a neural network-based segmentation algorithm to obtain text area location information:

[0052] The image input network passes through the convolutional network to obtain feature maps of different sizes;

[0053] Construct a feature pyramid, unify the scale, and obtain the feature map F;

[0054] The feature map F predicts the segmentation probability map P and threshold map T;

[0055] Graphs P and T estimate the binary graph B;

[0056] The neural network performs supervised training on the P, T, and B special graphs;

[0057] Traditional image segmentation algorithms use a standard binarization method, which sets pixels below a threshold to 0 and pixels above the threshold to 1. Standard binarization is not differentiable, making it impossible to optimize network weight parameters.

[0058] The differentiable binarization algorithm is used to ensure the feasibility of the network training model optimization process;

[0059] The post-processing operation scales the image size to a multiple of 32 and obtains the text location information through the network structure;

[0060] The deflection angle of the current image coordinate system is calculated using the coordinate information of the circumscribed rectangle of the text area in the image. The angle is between [0°-90°]. The valve device's on / off status and abnormal status are determined based on the angle information.

[0061] Valve identification plate text area recognition, traditional text recognition includes image preprocessing, character segmentation and character recognition; it often requires modeling of specific scenes to obtain prior information, and the recognition effect cannot meet the needs of changing and complex scenes.

[0062] A deep learning algorithm is used to recognize text content: a convolutional network extracts text area feature sequences; a bidirectional LSTM network is used to predict the feature sequence, learn each feature of the sequence, and output the distribution of predicted true values; the feature sequence is input into the CTC module to obtain the signboard result; the signboard content is converted to the device number; the image enhancement adopts the histogram averaging method; the text detection method is used to obtain the text area information and four-point coordinate information; when building the text detection model, a lightweight backbone network is selected to make a trade-off between detection accuracy and speed; the standard feature channel parameters are modified to further balance the accuracy and efficiency; when fusing feature maps of different scales, the convolution feature map is reduced to the same number of channels; the weight update strategy is optimized so that the network achieves better convergence accuracy during the training phase; the text polygon area is cropped and perspective transformed to correct the text area into a rectangular frame; the direction is corrected using a direction classifier; the text is recognized based on the rectangular frame containing the text area; the switch status of the valve device is indirectly through the pre-designed signboard structure, and the signboard is attached to the valve device and linked to it.

[0063] Image enhancement processing can improve image quality and reduce the complexity of subsequent processing operations.

[0064] Text detection and positioning uses a deep learning neural network and supports tilt angle detection.

[0065] A new direction classifier network structure is added to the polygonal area of ​​text detection to realize the angle and direction correction of the rectangular area of ​​text. It is more robust and accurate than the traditional connected domain detection algorithm for target position.

[0066] After processing the above steps, the image area containing text is obtained and a lightweight backbone network is used to ensure accuracy and efficiency. The CTC algorithm is used to solve the prediction of indefinite-length text, which has great advantages over traditional image segmentation and character recognition methods.

[0067] A method for automatically identifying the status of a gas exchange cock device includes: acquiring a device image and improving the image quality using image enhancement technology; detecting text box coordinates based on a segmentation algorithm, calculating the valve device rotation angle based on the coordinate information, and judging various valve working states based on the valve deflection angle value; identifying the identification content in the text area, obtaining the current valve device number and valve status, and pushing them to a monitoring system.

[0068] A robot equipped with a visible light camera and supplementary lighting equipment patrols the task point and collects images of the gas exchange valve equipment.

[0069] Enhance the local area of ​​the original image of the device; limit the local contrast by limiting the histogram height, limit the image noise amplification, and prevent excessive local contrast enhancement:

[0070] Set the adaptive histogram equalization sliding window size to M*M;

[0071] Derivative the local mapping function to obtain a histogram, that is, the slope S of the local mapping function;

[0072]

[0073] Limiting the slope of the local mapping function is equivalent to limiting the histogram height, and thus limiting the contrast intensity;

[0074]

[0075] Set the maximum slope S max , that is, get the maximum height of the histogram and cut off the value above H max Histogram, thus enhancing local contrast within time;

[0076]

[0077] Valve identification plate text area detection, using a neural network-based segmentation algorithm to obtain text area location information:

[0078] The image is input into the neural network, and feature maps of different sizes are obtained through bottom-up convolution operations and top-down upsampling;

[0079] The feature map obtained by convolution is fused with the feature map obtained by upsampling, and convolution is used to eliminate the aliasing effect, and the size is unified into a 1 / 4 feature map F;

[0080] After convolution and transposition convolution operations on the 1 / 4 size feature map F, the segmentation probability map P and threshold map T are obtained;

[0081] The segmentation probability map P and the threshold map T are processed to obtain an approximate binary map;

[0082] Approximate the standard binary step function to ensure differentiability;

[0083]

[0084] Through neural network supervised learning, the segmentation probability map P, threshold map T and approximate binary map are trained and the weight parameters are updated;

[0085] Test reasoning to obtain the bounding rectangle coordinate information pt1, pt2, pt3, pt4 of the target area text;

[0086] Calculate the deflection angle equivalent to the current image coordinate system, which is between [0° and 90°]; determine the valve equipment switch status and abnormal status based on the angle information;

[0087] k=-(pt2[1]-pt1[1]) / (pt2[0]-pt1[0])

[0088] res=arctan(k)*57.29577

[0089] Perform text content recognition on the detected text area image, obtain the device number, and distinguish multiple groups of devices;

[0090] Resize the image containing the text area to 32×W×1 to obtain the P1 image sequence;

[0091] After P1 passes through the convolutional network, it obtains the P2 feature sequence of size 1×(W / 4)×512;

[0092] Set T = (W / 4), D = 512, use a bidirectional LSTM network to learn the feature sequence, and output the predicted true value distribution R;

[0093] The prediction distribution is converted into a final label sequence for each frame through a fully connected network and activation function, which is then input into the CTC module and classified in time series to obtain the prediction result.

[0094] The content of the identification plate is converted into the equipment number, and the valve equipment status identification value corresponds to the valve equipment number one by one;

[0095] The status of the valve is determined based on the rotation angles identified by the two identification plates and combined with the numbers, and the valve monitoring results are finally pushed.

[0096] like Figure 2 As shown, the auxiliary identification sign designed.

[0097] Beneficial effects: The auxiliary identification sign solution solves the current situation where it is difficult to directly identify valve equipment and is difficult to achieve; it also solves the problem of numbering multiple groups of valve equipment;

[0098] In response to traditional image recognition algorithms such as connected domain detection of valve stem rotation status and target tracking solutions, deep learning-based text detection and text recognition algorithms, text direction detection algorithms, and pre-processing stage image enhancement are used to solve the real-time monitoring of the operating status of gas exchange valve equipment in the special environment of coking coal coke oven chambers.

[0099] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for monitoring the status of a gas exchange cock switch, characterized in that: The monitoring method comprises: A robot equipped with a visible light camera and supplementary lighting equipment patrols the task point, collects images of the gas exchange cock equipment and valve equipment status identification values, and obtains the original image of the equipment; enhancing a local area of ​​the original image of the device to obtain pre-processed image data; Detecting the valve identification plate text based on the preprocessed image data, and obtaining text area position information using a segmentation algorithm based on a neural network; Perform text content recognition based on the text area position information to obtain valve device numbers for distinguishing multiple groups of devices; Matching the valve device status identification value with the valve device number one by one; Obtaining the rotation angle of the identification plate according to the position information of the text area; the valve device state identification value is used to indicate the state of the valve; The state of the valve is determined according to the rotation angles identified by the two identification plates, and combined with the valve equipment number to obtain the valve monitoring result.

2. A method for monitoring the status of a gas exchange cock switch according to claim 1, characterized in that: The enhancing of the local area of ​​the original image of the device to obtain the pre-processed image data specifically includes: Limiting the histogram height limits the local contrast, limits the image noise amplification, and prevents excessive local contrast enhancement: Set the adaptive histogram equalization sliding window size to ; Derivative the local mapping function to obtain the histogram and the slope of the local mapping function ;H ist (i) is defined as follows: it represents the number of pixels at the i-th gray level in the original histogram, and m(i) represents the cumulative distribution function of the local histogram of the sliding window; Limiting the slope of the local mapping function is equivalent to limiting the histogram height and contrast intensity, S max为 Limit the maximum slope, H max is the corresponding maximum histogram height; ； Set the maximum slope , get the maximum height of the histogram, cut off the height above Histogram, thereby enhancing the local contrast within the time period. T is the set threshold used to determine whether to truncate the number of pixels in the current grayscale level. L is a constant used to adjust the grayscale level, which is set according to actual needs. 。 3. A method for monitoring the status of a gas exchange cock switch according to claim 1, characterized in that: Detecting the valve identification plate text based on the preprocessed image data and obtaining the text area position information using a neural network-based segmentation algorithm specifically includes: The image is input into the neural network, and feature maps of different sizes are obtained through bottom-up convolution operations and top-down upsampling; The feature map obtained by convolution is fused with the feature map obtained by upsampling, and convolution is used to eliminate the aliasing effect, and the size is unified into a 1 / 4 feature map. ; The feature map of 1 / 4 size After convolution and transposed convolution operations, the segmentation probability map is obtained and threshold map ; The segmentation probability map and the threshold map After processing, an approximate binary image is obtained; Approximate the standard binary step function to ensure differentiability and obtain an approximate binary graph; Through neural network supervised learning, the segmentation probability map , the threshold map Training is performed with the approximate binary image to update weight parameters; Test reasoning to obtain the coordinate information of the bounding rectangle of the target area text ; Calculate the deflection angle relative to the current image coordinate system, where the deflection angle is between [0° and 90°]; The switch state and abnormal state of the valve device are determined according to the deflection angle.

4. A method for monitoring the status of a gas exchange cock switch according to claim 1, characterized in that: The text content recognition is performed based on the text area position information to obtain the valve device number for distinguishing multiple groups of devices, specifically including: Scale the image containing the position information of the text area to Size, get image sequence ;W represents the width of the original image; The image sequence After the convolutional network, we get Characteristic sequence of size ; Use a bidirectional LSTM network to learn the feature sequence and output the predicted true value distribution ; The prediction distribution is converted into a final label sequence for each frame through a fully connected network and activation function, which is then input into the CTC module and classified in time series to obtain the prediction result. Convert the nameplate content to the equipment number.

Images