A coke oven exhaust damper state recognition method and device

Abstract

The present application relates to the technical field of neural network, and more particularly to a coke oven exhaust valve state recognition method, which comprises: obtaining a coke oven exhaust valve image to be recognized; performing target detection processing on the image to be recognized to obtain a detection frame set; performing screening and recognition processing on the detection frame set to obtain the component state of a target valve group in the image to be recognized, wherein the screening and recognition processing is a processing procedure for screening and recognizing the components of the target valve group and the components of an additional valve in the detection frame set. The method accurately identifies the state of the coke oven exhaust valve, improves the recognition accuracy of the recognition method, and ensures the work efficiency and reliability of the coke oven.

Description

Technical Field

[0001] This invention relates to the field of neural network technology, and in particular to a method and device for identifying the status of a coke oven exhaust gas switch. Background Technology

[0002] As a crucial component of the energy, chemical, and steel industries, coal coking and chemical production involves harsh working environments and hazardous locations. The coke oven heating system in coal coking operations requires continuous 24-hour operation and air / exhaust gas replacement every 20-30 minutes to maintain continuous production. Traditional manual inspection methods are labor-intensive, lack information technology, and are prone to missed inspections, false inspections, and gas leaks, making them unsuitable for the safety management needs of coal coking operations.

[0003] In recent years, with the enhancement of computing power and the continuous development of artificial intelligence and deep learning technologies, significant progress has been made in the field of target detection. Based on this, developing a method for automatically identifying and monitoring the status of coke oven exhaust gas switch-off devices using artificial intelligence and deep learning technologies is of significant research importance and is an effective means to improve plant operation and maintenance efficiency and safety.

[0004] Currently, existing vision-based detection and recognition methods can detect and identify the status of coke oven exhaust gas switches in images. However, additional coke oven exhaust gas switches or other objects may exist in the image, causing existing detection and recognition methods to misjudge the status of coke oven exhaust gas switches or fail to exclude such additional targets, resulting in low recognition accuracy. Summary of the Invention

[0005] This application provides a method and device for identifying the status of coke oven exhaust gas switch, which solves the technical problem of low identification accuracy of coke oven exhaust gas switch status in the prior art. It achieves accurate identification of the status of coke oven exhaust gas switch, improves the identification accuracy of the identification method, and ensures the working efficiency and reliability of the coke oven.

[0006] In a first aspect, embodiments of the present invention provide a method for identifying the status of a coke oven exhaust gas switch, comprising:

[0007] Acquire the image to be identified from the coke oven exhaust gas switch;

[0008] The image to be identified is subjected to target detection processing to obtain a set of detection boxes;

[0009] The detection box set is subjected to filtering and recognition processing to obtain the component status of the target opener group in the image to be recognized. The filtering and recognition processing is a process of filtering and recognizing the components of the target opener group and the components of additional openers in the detection box set.

[0010] Preferably, the step of filtering and identifying the detection box set to obtain the component status of the target opener / closer group in the image to be identified includes:

[0011] The detection frame set is subjected to a preliminary screening process to obtain a preliminary screening detection frame set, wherein the detection frame set includes a fan wheel detection frame set, a grinding rod detection frame set, and a cover plate detection frame set;

[0012] The initial screening detection frame set is grouped to obtain a fan wheel detection frame set, and the target fan wheel detection frame set is selected from the fan wheel detection frame set.

[0013] Based on the target fan wheel detection frame group, determine the corresponding weight rod detection frame and the corresponding cover plate detection frame for each target fan wheel detection frame in the target fan wheel detection frame group;

[0014] The component status of the target opener / closer group is obtained based on each target fan wheel detection frame, the corresponding counterweight detection frame, and the corresponding cover plate detection frame.

[0015] Preferably, the initial screening of the detection frame set to obtain the pre-screened detection frame set includes:

[0016] Obtain each fan wheel detection frame in the detection frame set;

[0017] If the size of each fan wheel detection frame in the detection frame set meets the preset size, then the detection frame set is determined as the detection frame set after initial screening, and the detection frame set after initial screening is output.

[0018] Preferably, the step of grouping the initial screening detection frame set to obtain a fan wheel detection frame set includes:

[0019] In the fan wheel detection frame set after the initial screening, two fan wheel detection frames of the same fan wheel color are divided into a fan wheel detection frame group to obtain the fan wheel detection frame group set.

[0020] Preferably, the step of selecting the target fan wheel detection frame group from the set of fan wheel detection frame groups includes:

[0021] When the fan wheel detection frame group consists of two fan wheel detection frame groups, the fan wheel spacing of each fan wheel detection frame group is obtained based on the center point positions of the two fan wheel detection frames in each fan wheel detection frame group.

[0022] If the fan wheel spacing of either of the two fan wheel detection frame groups is greater than a preset spacing threshold, and the fan wheel spacing of the other fan wheel detection frame group is less than the preset spacing threshold, then the other fan wheel detection frame group is determined as the target fan wheel detection frame group.

[0023] Preferably, the step of selecting the target fan wheel detection frame group from the fan wheel detection frame group group further includes:

[0024] When the fan wheel detection frame group consists of two fan wheel detection frame groups, if the difference between the fan wheel spacing of the two fan wheel detection frame groups is within the difference threshold range, then the center spacing of each fan wheel detection frame group in the two fan wheel detection frame groups is obtained based on the center point of the fan wheel spacing of each fan wheel detection frame group in the two fan wheel detection frame groups and the center point of the image to be identified.

[0025] The fan wheel detection frame group with the smallest center-to-center distance among the two fan wheel detection frame groups is selected, and the fan wheel detection frame group with the smallest center-to-center distance is selected as the target fan wheel detection frame group.

[0026] Preferably, the step of selecting the target fan wheel detection frame group from the fan wheel detection frame group group further includes:

[0027] When the fan wheel detection frame group is a single fan wheel detection frame group, the fan wheel detection frame group is directly used as the target fan wheel detection frame group.

[0028] Preferably, determining the corresponding counterweight detection frame and the corresponding cover plate detection frame for each target fan wheel detection frame in the target fan wheel detection frame group includes:

[0029] Using the center point of the fan wheel spacing of the target fan wheel detection frame group as the base point, extend a preset distance to the left and right sides from the base point to obtain the left endpoint, the right endpoint, a first line connecting the left endpoint to the base point, and a second line connecting the right endpoint to the base point. Connect the left endpoint to the bottom edge of the image to be identified to obtain the left line, and connect the right endpoint to the bottom edge of the image to be identified to obtain the right line. Based on the first line, the second line, the left line, and the right line, the display area of ​​the target fan wheel detection frame group is obtained.

[0030] Based on the horizontal coordinate of the base point, the grinding wheel detection frame and the cover plate detection frame in the display area are divided to obtain the left grinding wheel detection frame set and the left cover plate detection frame set of the left target fan wheel detection frame in the target fan wheel detection frame group, and the right grinding wheel detection frame set and the right cover plate detection frame set of the right target fan wheel detection frame in the target fan wheel detection frame group.

[0031] The left target fan wheel detection frame and the corresponding left cover plate detection frame are selected from the left target fan wheel detection frame set and the left target fan wheel detection frame set, respectively. The right target fan wheel detection frame and the corresponding left cover plate detection frame are selected from the right target fan wheel detection frame set and the right target fan wheel detection frame set.

[0032] Preferably, obtaining the component state of the target opener / closer group based on each target fan wheel detection frame, the corresponding counterweight detection frame, and the corresponding cover plate detection frame includes:

[0033] The component status of the left opener / closer is obtained based on the left target fan wheel detection frame, the corresponding left weight rod detection frame, and the corresponding left cover plate detection frame.

[0034] The component status of the right-side opener / closer is obtained based on the right-side target fan wheel detection frame, the corresponding right-side weight rod detection frame, and the corresponding right-side cover plate detection frame.

[0035] The component status of the target opener group is obtained based on the component status of the left opener group and the component status of the right opener group.

[0036] Based on the same inventive concept, in a second aspect, the present invention also provides a device for identifying the status of a coke oven exhaust gas switch, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for identifying the status of a coke oven exhaust gas switch.

[0037] One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages:

[0038] In this embodiment of the invention, an image of the coke oven exhaust gas switch to be identified is first acquired. Then, target detection processing is performed on the image to be identified to obtain a set of detection boxes. Based on the structure and state of each component of a single switch, multiple detection boxes are identified, improving the accuracy and precision of the model's detection and identification, and enhancing the model's response speed and operating efficiency. Next, the detection box set is filtered and identified to obtain the component states of the target switch group in the image to be identified. This filtering and identification process involves filtering and identifying the components of the target switch group and components of additional switches within the detection box set. By filtering and identifying components of additional switches outside the target switch group in the image to be identified, the accuracy and precision of the model's detection and identification are further improved, accurately obtaining the component states of the target switch group. This facilitates the identification of the working state of the target switch group and the corresponding coke oven combustion state, ensuring the working efficiency and reliability of the target switch group. Attached Figure Description

[0039] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0040] Figure 1 A flowchart illustrating the steps of the method for identifying the status of a coke oven exhaust gas switch in an embodiment of the present invention is shown.

[0041] Figure 2 A schematic diagram of the coke oven exhaust gas switch in an embodiment of the present invention is shown.

[0042] Figure 3 A structural schematic diagram of the component state of a single coke oven exhaust gas switch in an embodiment of the present invention is shown;

[0043] Figure 4 The diagram illustrates the calculation principle of the Intersection over Union (IoU) of a single predicted bounding box in an embodiment of the present invention.

[0044] Figure 5 This diagram illustrates the structure of the image content in an embodiment of the present invention, where the fan wheel detection frame set consists of two fan wheel detection frame sets.

[0045] Figure 6 This illustration shows another structural diagram of the image content in an embodiment of the present invention, where the fan wheel detection frame group is composed of two fan wheel detection frame groups.

[0046] Figure 7 A schematic diagram of the display area of ​​the target fan wheel detection frame in an embodiment of the present invention is shown. Detailed Implementation

[0047] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0048] Example 1

[0049] The first embodiment of the present invention provides a method for identifying the status of a coke oven exhaust gas switch, such as... Figure 1 As shown, it includes:

[0050] S101, Obtain the image to be identified from the coke oven exhaust gas switch;

[0051] S102, Perform target detection processing on the image to be recognized to obtain a set of detection boxes;

[0052] S103, perform filtering and recognition processing on the detection box set to obtain the component status of the target opener / closer group in the image to be recognized. The filtering and recognition processing is the process of filtering and recognizing the components of the target opener / closer group and the components of the additional opener / closer in the detection box set.

[0053] The identification method in this embodiment is applied to the scenario of monitoring coke oven exhaust gas on / off switch (hereinafter referred to as the switch). For example Figure 2 As shown, this scenario typically involves multiple groups of openers / closers. Each group consists of two openers / closers, meaning each group includes two openers / closers positioned adjacent to each other on the left and right. The fan wheels of each opener / closer within the same group are the same color. Figure 2 The diagonal lines represent red, and the checkered pattern represents yellow. Each opening / closing mechanism includes a fan wheel, a counterweight, and a cover plate. A camera track is positioned above multiple sets of opening / closing mechanisms in this scene. Figure 2 The solid black line shown above represents the camera track. Multiple cameras are mounted on the drive module of the camera track, and these cameras are positioned directly in front of the opening / closing mechanism to take pictures. The drive module of the camera track drives the multiple cameras to move in a cyclical manner, allowing each camera to capture images of a different group of opening / closing mechanisms. Figure 2 The arrows indicate the direction of movement. Since the camera positions are essentially fixed, each camera's image will contain a maximum of two groups of openers / closers, i.e., four openers / closers, as shown below. Figure 2 The boxes shown represent the content captured by the camera. The transmission module can be configured according to actual needs, such as a conveyor belt.

[0054] Below, in conjunction with Figure 1 The following details the specific implementation steps of the method for identifying the status of the coke oven exhaust gas switch provided in this embodiment:

[0055] Before executing step S101, the initial detection and recognition model is trained to obtain the trained detection and recognition model. Then, the trained detection and recognition model is tested to obtain the target detection and recognition model.

[0056] The specific process of training the initial detection and recognition model to obtain the trained detection and recognition model is as follows: First, a training image set of the coke oven exhaust gas switch is obtained. Each training image in the training image set carries a manually labeled label, which is the labeled bounding box and the ground truth bounding box. Then, the initial detection and recognition model is trained using the labeled image set until the trained detection and recognition model meets the training constraints. The detection and recognition model that meets the training constraints is then determined as the trained detection and recognition model.

[0057] Because acquiring a large number of images of abnormal coke oven combustion states is not feasible in the context of coke oven exhaust gas switch-off devices, it is necessary to deduce the combustion state of the coke oven through structural and state analysis of the switch-off devices. Therefore, the state identification of a coke oven exhaust gas switch-off device group is decoupled into the state identification of different components of a single switch-off device. Subsequent filtering and identification processing methods are then used to combine the identified states of the components, outputting the final identification result. This approach solves the problem of difficulty in acquiring images of abnormal coke oven combustion states.

[0058] The actual process of obtaining the labeled image set of coke oven exhaust gas switch devices involves labeling each training image using the labelme labeling tool. Specifically, a single switch device is decomposed into three components to be detected: the fan wheel, the counterweight, and the cover plate. Each of these three components exists in two different states. The combustion state of the coke oven needs to be determined by the combined states of a set of two switch devices (left and right). To facilitate group identification, the fan wheels belonging to the same switch device group are painted with alternating red and yellow colors. Figure 3 As shown, Figure 3 (a) shows the red fan wheel at the top, with red represented by a diagonal line pattern. Figure 3 (b) shows the yellow fan wheel at the bottom, with yellow represented by a checkered pattern. Figure 3 (c) shows the state where the grinding wheel is at the top. Figure 3 (d) indicates the state where the grinding wheel is at the bottom. Figure 3 (e) indicates that the cover is in the closed state. Figure 3 (f) indicates that the cover is in the open state.

[0059] Therefore, the components of each training image are divided into four categories, each corresponding to two states: red fan wheel up (down), yellow fan wheel up (down), weight rod up (down), and cover plate open (closed). There are a total of eight corresponding labels, which are represented in the algorithm engineering as r_sl_up, r_sl_down; y_sl_up, y_sl_down; tg_up, tg_down; gb_open, gb_close.

[0060] Furthermore, to improve the diversity and quality of the image data for the coke oven exhaust gas switch, image data augmentation methods were employed during the acquisition of the training image set, such as Mosaic data augmentation and MixUp data augmentation. The image data augmentation methods can be set according to actual needs, such as stitching, copying, and scaling. During the training phase, these methods can increase the diversity of samples before the images are input into the model, thereby improving the model's robustness and generalization ability.

[0061] After obtaining the training image set, the initial detection and recognition model is trained using the labeled image set until the trained model meets the training constraints. The model that meets these constraints is then designated as the trained detection and recognition model. The initial model can be set according to actual needs; for example, for YOLO series models, the YOLOX-tiny model is preferred. Training constraints can be that the current loss function value of the detection and recognition model is not greater than a loss function threshold, the number of iterations meets an iteration threshold, or other constraints. The loss function, loss function threshold, and iteration threshold can be set according to actual needs. For example, the current detection and recognition model is designated as the trained detection and recognition model when the current loss function value is not greater than the loss function threshold.

[0062] After obtaining the trained detection and recognition model, it needs to be tested and validated. Specifically, a test image set of the coke oven exhaust gas switch is acquired. The trained detection and recognition model is then tested using this image set. If the evaluation metrics of the trained detection and recognition model meet the preset evaluation metrics, it means the model has passed the test and is adopted as the target detection and recognition model. If the evaluation metrics of the trained detection and recognition model do not meet the preset evaluation metrics, it means the model has failed the test, and further model training is required. The test image set can also be augmented to increase sample diversity.

[0063] Specifically, based on the actual industrial situation, recall rate is adopted as the evaluation index for the target detection stage, and the specific formula (1) is as follows:

[0064]

[0065] Specifically, TP (true positive) and FN (false negative) are calculated based on the Intersection over Union (IoU) between the manually labeled data and the predicted data generated by the detection and recognition model (i.e., the true bounding box and the predicted bounding box generated by the detection and recognition model). TP is a correct prediction of a positive sample, while FN is an incorrect prediction of a positive sample. TP represents the total number of true positives, and FN represents the total number of false negatives. The calculation method of IoU is shown in formula (2):

[0066]

[0067] like Figure 4As shown, the denominator of formula (2) represents the total area occupied by the overlap between the predicted bounding box G and the true bounding box P, and the numerator represents the area occupied by the overlapping part of the two boxes. Therefore, the value of IoU is between 0 and 1, and the larger the value, the more accurate the predicted bounding box. The corresponding IoU threshold is set according to actual needs. If, during the evaluation process, the IoU value for each predicted bounding box is not less than the IoU threshold, it means that the trained detection and recognition model has correctly identified the predicted bounding box, and the predicted bounding box is marked as a true example. If the IoU value is less than the IoU threshold, the predicted bounding box is marked as a false negative example.

[0068] In the Recall metric calculation, the Recall result is obtained based on the total number of true positives and the total number of false negatives. If the result value is not less than the result threshold, it means that the trained detection and recognition model has passed the test, and the trained detection and recognition model is then identified as the target detection and recognition model. If the result value is less than the result threshold, it means that the trained detection and recognition model has failed the test, and the model training continues. The result threshold can be set according to actual needs.

[0069] Table 1 shows the test results of the object detection and recognition model under different IoU thresholds. It can be seen that by setting an appropriate threshold, the accuracy of the detection and recognition model can reach up to 96%. Furthermore, time-based tests show that the object detection and recognition model requires approximately 0.1 seconds for detection inference on a single image, meeting the requirements for real-time detection.

[0070] Table 1

[0071] 0.50 0.96 0.55 0.96 0.60 0.96 0.65 0.95 0.70 0.94 0.75 0.92 0.80 0.90 0.85 0.87 0.90 0.83 0.95 0.79

[0072] After obtaining the target detection and recognition model, it is used. Step S101 is executed to acquire the image to be recognized from the coke oven exhaust gas switch. Specifically, in the scenario of the coke oven exhaust gas switch, the current image of the coke oven exhaust gas switch, i.e., the image to be recognized, is acquired in real time by a camera to ensure real-time monitoring of the coke oven's combustion status.

[0073] Next, step S102 is executed to perform target detection processing on the image to be recognized, and a set of detection boxes is obtained.

[0074] Specifically, the object detection and recognition model performs object detection processing on the image to be recognized, obtaining a set of detection boxes. The preferred object detection and recognition model is the YOLOX-tiny model, which enables fast and efficient detection of the image, accurately identifying each component of the opener / closer within the image and improving detection and recognition efficiency. The object detection and recognition model identifies the image to be recognized, outputting a set of detection boxes. This set includes multiple detection boxes, each with its location information, label, and confidence level. The location information of each detection box can be the horizontal and vertical coordinates of its top-left and bottom-right corners, or can be set according to actual needs. The label of each detection box indicates the category of the opener / closer, such as a fan wheel, a weight, or a cover plate. A higher confidence level for each detection box indicates greater confidence in the object detection and recognition model's detection results.

[0075] In the actual testing of the detection and recognition model, among the three components of a single opener / closer, the fan wheel component achieved a detection rate and state recognition accuracy of over 99% due to its advantages such as simple structure, distinctive features, and complete external features. Therefore, the fan wheel was selected as the core for screening and grouping to accurately identify the components of the target opener / closer group in the image to be recognized.

[0076] Then, step S103 is executed to perform filtering and recognition processing on the detection box set to obtain the component status of the target opener / closer group in the image to be recognized. The filtering and recognition processing is a process of filtering and recognizing the components of the target opener / closer group and the components of the additional opener / closer in the detection box set.

[0077] Specifically, the image to be identified captures a maximum of four openers / closers, and the identification result of the image represents only one group of openers / closers. The specific process for obtaining the component state of the target opener / closer group in the image to be identified is as follows: a) Perform initial screening on the detection frame set to obtain a pre-screened detection frame set, which includes a fan wheel detection frame set, a weight rod detection frame set, and a cover plate detection frame set. b) Group the pre-screened detection frame set to obtain a fan wheel detection frame group set, and then select the target fan wheel detection frame group from this group. c) Based on the target fan wheel detection frame group, determine the weight rod detection frame and the corresponding cover plate detection frame for each target fan wheel detection frame in the target fan wheel detection frame group. d) Based on each target fan wheel detection frame, its corresponding weight rod detection frame, and its corresponding cover plate detection frame, obtain the component state of the target opener / closer group.

[0078] Step a requires initial screening of the dimensions of each fan wheel detection frame in each detection frame set to avoid errors in the recognition result caused by factors inherent to the image itself. The initial screening process involves first acquiring each fan wheel detection frame in the detection frame set. Each fan wheel detection frame is then evaluated. If the dimensions of each fan wheel detection frame in the detection frame set conform to a preset size, the detection frame set is determined as the initially screened detection frame set and output. The preset size is set according to actual needs. If the size of any fan wheel detection frame in the detection frame set does not conform to the preset size, the image to be recognized is discarded, and the recognition error information for the image is output.

[0079] To illustrate this, let's take a fan wheel detection frame as an example. If the length (or width) of the fan wheel detection frame is greater than one-third of the length (or width) of the image to be recognized, or less than one-twentieth of the length (or width) of the image to be recognized, it indicates that the size of the fan wheel detection frame does not conform to the preset size. This image to be recognized will lead to incorrect model recognition, often due to factors such as lighting and shadows in the image. Therefore, the initial screening process involves screening the detection frame set. After initial screening, the reasonableness of the image to be recognized is ensured. It also avoids model recognition of unreasonable images to be recognized, which would lead to incorrect recognition results, thus improving the model's recognition accuracy, precision, and operating efficiency.

[0080] Step b involves first grouping two fan wheel detection frames of the same fan wheel color into a single fan wheel detection frame group within the initial screening detection frame set. Then, the target fan wheel detection frame group is selected from this group.

[0081] Specifically, since the image to be identified displays a maximum of four openers / closers, the image content may contain one opener / closer, one group of openers / closers, three openers / closers, or four openers / closers. After initial screening, the fan wheel detection boxes in the detection box set are grouped, with two fan wheels of the same color grouped together to obtain a fan wheel detection box set. This facilitates subsequent identification by fan wheel detection box group, obtaining the target fan wheel detection box set, and improving the model's recognition accuracy, precision, and operational efficiency.

[0082] After obtaining the set of fan wheel detection boxes, it is necessary to determine whether the set consists of a single fan wheel detection box group. This group is directly used as the target fan wheel detection box group, enabling rapid and efficient identification of the target result in the image. If there are two fan wheel detection box groups, the fan wheel spacing of each group is first determined based on the center point positions of the two fan wheel detection boxes within each group. For example, if the center point of one fan wheel detection box is A1(x1, y1) and the center point of the other fan wheel detection box is A2(x2, y2), the distance between A1 and A2 is the fan wheel spacing of that group. The target fan wheel detection box group is then obtained based on the fan wheel spacing of each group. This method can accurately identify the target fan wheel detection frame group based on the fan wheel spacing of each fan wheel detection frame group, improving the model's recognition accuracy and precision. The calculation method is fast and the running efficiency is improved.

[0083] Due to the camera's perspective, two special situations may occur during the shooting process: Situation 1: The image content contains only one complete set of openers / closes, but there is one exhaust gas opener / closer of the same color but different set on each side. Situation 2: The image content contains two complete sets of openers / closes. Following the procedure below, the image to be recognized will only output the result for one set of openers / closes.

[0084] When there are two fan wheel detection frame groups, if the fan wheel spacing in one of the two fan wheel detection frame groups is greater than a preset spacing threshold, and the fan wheel spacing in the other fan wheel detection frame group is less than the preset spacing threshold, it indicates that the difference between the fan wheel spacings of the two fan wheel detection frame groups is not within the difference threshold range, and there is a significant difference in the fan wheel spacings between the two fan wheel detection frame groups. In this case, the other fan wheel detection frame group is determined as the target fan wheel detection frame group. Both the preset spacing threshold and the difference threshold range can be set according to actual needs.

[0085] Specifically, such as Figure 5 As shown, a set of red fan wheel opening and closing devices is located between the opening and closing devices of two yellow fan wheels. For the image to be recognized, the two fan wheel detection frames representing the yellow fan wheels are divided into a first fan wheel detection frame group g. 黄 The two fan wheel detection frames representing the red fan wheels are divided into a second fan wheel detection frame group g. 红 Based on the center point positions of the two fan wheel detection frames representing the yellow fan wheels, the fan wheel spacing D of the first fan wheel detection frame group is obtained. 黄 P 黄左 for Figure 5 The leftmost point indicates the center point of the detection frame for the yellow fan wheel, P. 黄右 for Figure 5The center point of the detection frame for the yellow fan wheel is shown on the right. Since the center points of the two detection frames for the yellow fan wheel are on the same horizontal line, D... 黄 =P 黄右 x-coordinate - P 黄左 The x-coordinate is calculated. Similarly, based on the center point positions of the two fan wheel detection frames representing the red fan wheels, the fan wheel spacing D of the second fan wheel detection frame group is obtained. 红 P 红左 for Figure 5 The leftmost point indicates the center point of the detection frame for the red fan wheel, P. 红右 for Figure 5 The rightmost point represents the center point of the detection frame for the red fan wheel. Since the center points of the two detection frames for the red fan wheel are on the same horizontal line, D... 红 =P 红右 x-coordinate - P 红左 The x-coordinate.

[0086] For D 黄 and D 红 The judgment is made, and the result is shown in formula (3):

[0087]

[0088] Among them, D 标准 =1.2*(D 红 +D 黄 ) / 2. Here, D 标准 This is the preset spacing threshold.

[0089] According to formula (3), D 黄 and D 红 respectively with D 标准 Compare the fan wheel spacings of the two fan wheel detection frame groups to determine if there is a significant difference. If the fan wheel spacing D of the first fan wheel detection frame group... 黄 Greater than D 标准 And the fan wheel spacing D of the second fan wheel detection frame group 红 Less than D 标准 This indicates a significant difference in the spacing between the two fan wheels, so the second fan wheel detection frame group g is then moved to the next group. 红 The target fan wheel detection frame group is identified. If the fan wheel spacing D of the second fan wheel detection frame group... 红 Greater than D 标准 And the fan wheel spacing D of the first fan wheel detection frame group 黄 Less than D 标准 This indicates a significant difference in the spacing between the two fan wheels, so the first fan wheel detection frame group g is... 黄 The target fan wheel detection frame group has been identified.

[0090] When there are two fan wheel detection box sets, and the fan wheel spacing between the two sets differs significantly, the target fan wheel detection box set can be identified simply by measuring the fan wheel spacing. This method provides a simple, fast, and efficient way to identify the target fan wheel detection box set, improving the model's recognition accuracy and precision, simplifying the identification and filtering steps, and enhancing response speed, operational efficiency, and model computational efficiency.

[0091] Next, when there are two fan wheel detection box sets, if the difference between the fan wheel spacings of the two fan wheel detection box sets is within the difference threshold range, it indicates that there is no significant difference between the fan wheel spacings of the two fan wheel detection box sets. Then, based on the center point of the fan wheel spacing of each of the two fan wheel detection box sets and the center point of the image to be identified, the center spacing of each of the two fan wheel detection box sets is obtained. The fan wheel detection box set with the smallest center spacing is selected from the center spacings of the two fan wheel detection box sets and designated as the target fan wheel detection box set.

[0092] Specifically, such as Figure 6 As shown, the image content consists of two complete opening / closing device groups. The two detection boxes on the left represent the red fan wheels, denoted as g'. 红 The two fan wheel detection boxes on the right, representing the yellow fan wheels, are denoted as g'. 黄 For the image to be identified, g' 红 The fan wheel spacing and g' 黄 The fan wheel spacing is consistent, indicating that the difference in fan wheel spacing between the two is within the difference threshold range. Therefore, according to g' 红 The center point M of the fan wheel spacing 红 Center point O of the image to be identified 中心 , to obtain g' 红 Center spacing S 红 M 红 O 中心 According to g' 黄 The center point M of the fan wheel spacing 黄 Center point O of the image to be identified 中心 , to obtain g' 黄 Center spacing S 黄 M 黄 O 中心 .

[0093] The center point M(x) of any fan wheel spacing M y M ) to O 中心 The formula for calculating the center distance S between (X, Y) is shown in (4):

[0094]

[0095] For S 黄 and S 红 The judgment is made, and the result is shown in formula (5):

[0096]

[0097] According to formula (5), S 黄 and S 红 Compare. If S 黄 Not less than S 红 Then g' 红 As the target fan wheel detection box group. If S 黄 Less than S 红 Then g' 黄 This serves as the target fan wheel detection box group. Thus, the fan wheel detection box group with the smallest center-to-center distance among the two fan wheel detection box groups is selected as the target fan wheel detection box group. This simple, fast, and efficient method of identifying the target fan wheel detection box group improves the model's recognition accuracy and precision, simplifies the recognition and filtering steps, and enhances response speed, operational efficiency, and model computational efficiency.

[0098] Execute steps c, c1, as follows Figure 7 As shown, the center point of the fan wheel spacing in the target fan wheel detection frame group is taken as the base point. Figure 7 As shown in the diagram, N extends a preset distance to the left and right from the base point, resulting in the left endpoint, the right endpoint, a first line connecting the left endpoint to the base point, and a second line connecting the right endpoint to the base point. The preset distances can be set according to actual needs. A left-side line is obtained by connecting the left endpoint to the bottom edge of the image to be recognized, and a right-side line is obtained by connecting the right endpoint to the bottom edge of the image to be recognized. Based on the first line, the second line, the left line, and the right line, the display area of ​​the target fan wheel detection frame group is obtained. The display area is... Figure 7 The area enclosed by the box shown.

[0099] By setting up the display area, redundant components and falsely detected components outside the display area can be filtered out, which facilitates the subsequent identification of the target opener group corresponding to the target fan wheel detection frame group, improves the model's recognition accuracy and precision, and enhances the model's operating efficiency and response speed.

[0100] c2. Based on the horizontal coordinate of the base point, divide the display area into the target fan wheel detection frame and the cover plate detection frame, and obtain the left target fan wheel detection frame set and the left cover plate detection frame set of the left target fan wheel detection frame in the target fan wheel detection frame group, as well as the right target fan wheel detection frame set and the right cover plate detection frame set of the right target fan wheel detection frame in the target fan wheel detection frame group.

[0101] Specifically, with Figure 5Taking the target fan wheel detection frame group as an example Figure 5 The target fan wheel detection frame group consists of two fan wheel detection frames representing the red fan wheels. In the display area of ​​the target fan wheel detection frame group, if the x-coordinate of the weight rod detection frame in the display area is less than the x-coordinate of the base point, then the weight rod detection frame is divided into the left weight rod detection frame of the left target fan wheel detection frame, thus obtaining the left weight rod detection frame set. If the x-coordinate of the weight rod detection frame in the display area is greater than the x-coordinate of the base point, then the weight rod detection frame is divided into the right weight rod detection frame of the right target fan wheel detection frame, thus obtaining the right weight rod detection frame set. Similarly, if the x-coordinate of the cover plate detection frame in the display area is less than the x-coordinate of the base point, then the cover plate detection frame is divided into the left cover plate detection frame of the left target fan wheel detection frame, thus obtaining the left cover plate detection frame set. If the x-coordinate of the cover plate detection frame in the display area is greater than the x-coordinate of the base point, then the cover plate detection frame is divided into the right cover plate detection frame of the right target fan wheel detection frame, thus obtaining the right cover plate detection frame set.

[0102] c3. Filter out the left target fan wheel detection frame and the corresponding left cover plate detection frame from the left target fan wheel detection frame set and the left target fan wheel detection frame set, and filter out the right target fan wheel detection frame and the corresponding left cover plate detection frame from the right target fan wheel detection frame set and the right target fan wheel detection frame set.

[0103] Specifically, each detection frame includes a confidence level. Therefore, the left-side weight detection frame with the highest confidence level from the left-side weight detection frame set is selected as the left-side weight detection frame corresponding to the left-side target fan wheel detection frame. Similarly, the left-side cover plate detection frame with the highest confidence level from the left-side cover plate detection frame set is selected as the left-side cover plate detection frame corresponding to the left-side target fan wheel detection frame. Likewise, the right-side weight detection frame with the highest confidence level from the right-side weight detection frame set is selected as the right-side weight detection frame corresponding to the right-side target fan wheel detection frame, and the right-side cover plate detection frame with the highest confidence level from the right-side cover plate detection frame set is selected as the right-side cover plate detection frame corresponding to the right-side target fan wheel detection frame. This method accurately identifies the left-side weight detection frame and left-side cover plate detection frame corresponding to the left-side target fan wheel detection frame, as well as the right-side weight detection frame and right-side cover plate detection frame corresponding to the right-side target fan wheel detection frame, demonstrating high precision and accuracy in the recognition results.

[0104] Execute step d to obtain the component status of the left opener / closer based on the left target fan wheel detection frame, the corresponding left counterweight detection frame, and the corresponding left cover plate detection frame; obtain the component status of the right opener / closer based on the right target fan wheel detection frame, the corresponding right counterweight detection frame, and the corresponding right cover plate detection frame; and obtain the component status of the target opener / closer group based on the component status of the left and right opener / closer groups.

[0105] Specifically, the left target fan wheel detection frame, the corresponding left counterweight detection frame, and the corresponding left cover plate detection frame are combined to obtain the component state of the left opener / closer. The right target fan wheel detection frame, the corresponding right counterweight detection frame, and the corresponding right cover plate detection frame are combined to obtain the component state of the right opener / closer. Finally, the component states of the left and right opener / closer groups are combined to obtain the component state of the target opener / closer group.

[0106] The component status of a target gate group reflects the combustion status of the coke oven corresponding to that group. For example, in actual industrial situations, within the same gate group, the left and right fan wheels must be in the same state, and the fan wheel and the pressure rod must also be in the same up or down position. The cover can only be opened when the pressure rod (fan wheel) is in the down position. If a gate identifies that the cover is closed when the pressure rod is in the down position, it indicates that the gate group has malfunctioned and requires manual repair, or that the coke oven corresponding to that gate group has malfunctioned and requires manual repair.

[0107] Furthermore, the detection and recognition model in this embodiment is based on a multi-threaded data communication mechanism. This mechanism separates model initialization and the transmission of the address of the image to be recognized, and uses multi-threading technology to enable simultaneous inference of multiple images, achieving efficient communication between the server and the client for real-time detection.

[0108] One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages:

[0109] In this embodiment of the invention, an image of the coke oven exhaust gas switch to be identified is first acquired. Then, target detection processing is performed on the image to be identified to obtain a set of detection boxes. Based on the structure and state of each component of a single switch, multiple detection boxes are identified, improving the accuracy and precision of the model's detection and identification, and enhancing the model's response speed and operating efficiency. Next, the detection box set is filtered and identified to obtain the component states of the target switch group in the image to be identified. This filtering and identification process involves filtering and identifying the components of the target switch group and components of additional switches within the detection box set. By filtering and identifying components of additional switches outside the target switch group in the image to be identified, the accuracy and precision of the model's detection and identification are further improved, accurately obtaining the component states of the target switch group. This facilitates the identification of the working state of the target switch group and the corresponding coke oven combustion state, ensuring the working efficiency and reliability of the target switch group.

[0110] Example 2

[0111] Based on the same inventive concept, a second embodiment of the present invention also provides a device for identifying the status of a coke oven exhaust gas switch, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the methods described above for identifying the status of a coke oven exhaust gas switch. This device includes, but is not limited to, a personal computer, an industrial computer, a tablet computer, or other smart devices.

[0112] Since the coke oven exhaust gas switch status identification device described in this embodiment is the same device used to implement the coke oven exhaust gas switch status identification method in Embodiment 1 of this application, those skilled in the art can understand the specific implementation method and various variations of the coke oven exhaust gas switch status identification device in this embodiment based on the coke oven exhaust gas switch status identification method described in Embodiment 1 of this application. Therefore, how the coke oven exhaust gas switch status identification device implements the method in Embodiment 1 of this application will not be described in detail here. Any device used by those skilled in the art to implement the coke oven exhaust gas switch status identification method in Embodiment 1 of this application falls within the scope of protection of this application.

[0113] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (systems), or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0117] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0118] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying the status of a coke oven exhaust gas switch, characterized in that, include: Acquire the image to be identified from the coke oven exhaust gas switch; The image to be identified is subjected to target detection processing to obtain a set of detection boxes; The detection box set is filtered and identified to obtain the component states of the target opener / closer group in the image to be identified, including: The detection frame set is subjected to a preliminary screening process to obtain a preliminary screening detection frame set, wherein the detection frame set includes a fan wheel detection frame set, a grinding rod detection frame set, and a cover plate detection frame set; The initial screening detection frame set is grouped to obtain a fan wheel detection frame set, and the target fan wheel detection frame set is selected from the fan wheel detection frame set. Based on the target fan wheel detection frame group, determine the corresponding weight rod detection frame and the corresponding cover plate detection frame for each target fan wheel detection frame in the target fan wheel detection frame group; Based on each target fan wheel detection frame, the corresponding counterweight detection frame, and the corresponding cover plate detection frame, the component status of the target opener / closer group is obtained; wherein, the screening and identification process is a process of screening and identifying the components of the target opener / closer group and the components of additional openers / closers in the detection frame set.

2. The method as described in claim 1, characterized in that, The initial screening of the detection frame set to obtain the initially screened detection frame set includes: Obtain each fan wheel detection frame in the detection frame set; If the size of each fan wheel detection frame in the detection frame set meets the preset size, then the detection frame set is determined as the detection frame set after initial screening, and the detection frame set after initial screening is output.

3. The method as described in claim 2, characterized in that, The process of grouping the initial screening detection frame set to obtain the fan wheel detection frame set includes: In the fan wheel detection frame set after the initial screening, two fan wheel detection frames of the same fan wheel color are divided into a fan wheel detection frame group to obtain the fan wheel detection frame group set.

4. The method as described in claim 3, characterized in that, The step of selecting the target fan wheel detection frame group from the set of fan wheel detection frame groups includes: When the fan wheel detection frame group consists of two fan wheel detection frame groups, the fan wheel spacing of each fan wheel detection frame group is obtained based on the center point positions of the two fan wheel detection frames in each fan wheel detection frame group. If the fan wheel spacing of either of the two fan wheel detection frame groups is greater than a preset spacing threshold, and the fan wheel spacing of the other fan wheel detection frame group is less than the preset spacing threshold, then the other fan wheel detection frame group is determined as the target fan wheel detection frame group.

5. The method as described in claim 4, characterized in that, The step of selecting the target fan wheel detection frame group from the set of fan wheel detection frame groups further includes: When the fan wheel detection frame group consists of two fan wheel detection frame groups, if the difference between the fan wheel spacing of the two fan wheel detection frame groups is within the difference threshold range, then the center spacing of each fan wheel detection frame group in the two fan wheel detection frame groups is obtained based on the center point of the fan wheel spacing of each fan wheel detection frame group in the two fan wheel detection frame groups and the center point of the image to be identified. The fan wheel detection frame group with the smallest center-to-center distance among the two fan wheel detection frame groups is selected, and the fan wheel detection frame group with the smallest center-to-center distance is selected as the target fan wheel detection frame group.

6. The method as described in claim 3, characterized in that, The step of selecting the target fan wheel detection frame group from the set of fan wheel detection frame groups further includes: When the fan wheel detection frame group is a single fan wheel detection frame group, the fan wheel detection frame group is directly used as the target fan wheel detection frame group.

7. The method according to any one of claims 2-6, characterized in that, The step of determining the corresponding weight detection frame and the corresponding cover plate detection frame for each target fan wheel detection frame in the target fan wheel detection frame group includes: Using the center point of the fan wheel spacing of the target fan wheel detection frame group as the base point, extend a preset distance to the left and right sides from the base point to obtain the left endpoint, the right endpoint, a first line connecting the left endpoint to the base point, and a second line connecting the right endpoint to the base point. Connect the left endpoint to the bottom edge of the image to be identified to obtain the left line, and connect the right endpoint to the bottom edge of the image to be identified to obtain the right line. Based on the first line, the second line, the left line, and the right line, the display area of ​​the target fan wheel detection frame group is obtained. Based on the horizontal coordinate of the base point, the grinding wheel detection frame and the cover plate detection frame in the display area are divided to obtain the left grinding wheel detection frame set and the left cover plate detection frame set of the left target fan wheel detection frame in the target fan wheel detection frame group, and the right grinding wheel detection frame set and the right cover plate detection frame set of the right target fan wheel detection frame in the target fan wheel detection frame group. The left target fan wheel detection frame and the corresponding left cover plate detection frame are selected from the left target fan wheel detection frame set and the left target fan wheel detection frame set, respectively. The right target fan wheel detection frame and the corresponding left cover plate detection frame are selected from the right target fan wheel detection frame set and the right target fan wheel detection frame set.

8. The method as described in claim 7, characterized in that, The step of obtaining the component status of the target opener / closer group based on each target fan wheel detection frame, the corresponding counterweight detection frame, and the corresponding cover plate detection frame includes: The component status of the left opener / closer is obtained based on the left target fan wheel detection frame, the corresponding left weight rod detection frame, and the corresponding left cover plate detection frame. The component status of the right-side opener / closer is obtained based on the right-side target fan wheel detection frame, the corresponding right-side weight rod detection frame, and the corresponding right-side cover plate detection frame. The component status of the target opener group is obtained based on the component status of the left opener group and the component status of the right opener group.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1-8.

Images