Coke oven battery exchange coke rapid inspection method based on short-term dense cascade network

Abstract

The application discloses a coke oven exchange coke rapid inspection method based on a short-term dense cascade network, which comprises non-stop inspection and exchange coke working state detection, and the non-stop inspection step comprises the following steps: S1.1, a robot inspection map is constructed, and a robot movement soft limit starting point and an end point are configured; S1.2, the robot is controlled to move to a suitable position; S1.3, algorithm identification rules are configured, configured inspection data point data packets are imported into the robot through a web, S1.4, the robot performs path planning and inspection through an inspection task and a map; and S1.5, in the inspection process, the robot platform displays inspection results and alarm information in real time according to exchange coke state identification results and in combination with linkage signal rules. The exchange coke working state detection step comprises the following steps: S2.1, a short-term dense cascade network (STDC) is used to segment key components of the exchange coke in an image; and S2.2, angle judgment is performed on the segmented exchange coke.

Description

Technical Field

[0001] This invention relates to a method for detecting the working status of a coke oven exchange cock, and more specifically to a rapid inspection method for a coke oven exchange cock based on a short-term dense cascaded network, belonging to the field of coking oven technology. Background Technology

[0002] The gas exchange cock in a coke oven is used to regulate the gas flow and thus control the heating process, such as temperature control, to achieve optimal coking results. However, the corrosive nature of the gas, coupled with the high temperature and humidity of the working environment, causes the exchange cock to rust and become stuck. Furthermore, impurities such as tar in the coke oven gas increasingly adhere to the cock core, easily leading to breakage of the exchange lever or actuator. Additionally, if the lever falls off or is misaligned, it can affect the exchange of gas and exhaust gas, resulting in unstable temperature control in the coke oven, ultimately causing the coke oven to fail to heat or remain at a persistently high temperature.

[0003] Regarding the issue of detecting the operating status of coke oven exchange cocks, existing products mainly rely on installing multiple sensors within the exchange cock for detection. For example, temperature sensors are installed to detect abnormalities in temperature and thus identify malfunctions in the entire heating equipment. Other products determine malfunctions by monitoring the pressure within the exchange cock.

[0004] The inventors found CN213579834U: an intelligent monitoring system for coke oven exchange anomalies. This system uses multiple temperature detection modules to collect the temperature of the coke oven's opener and closer. The data acquisition module obtains the temperature detection data and transmits it to an industrial computer via a controller, displaying the data on a monitor for the workers. While this device can detect anomalies in the entire heating equipment, it cannot pinpoint the specific component causing the anomaly. Workers are still required to conduct a secondary inspection of all equipment on-site, resulting in a slow response time and significant labor costs. CN217032835U: an alarm device suitable for coke oven exchange cocks. This device detects anomalies in the cocking equipment by collecting the pressure of each cock. It uses an electronic force gauge connected to an exchange cock lever. The lever rotates back and forth at regular intervals, and a fixed reference pressure value is set. By comparing this reference value with the actual pressure, an anomaly is determined. An alarm is triggered if the anomaly exceeds 3 minutes. Although this device can detect anomalies in a particular exchange cock, the detection method simply compares pressure values ​​without considering the influence of other factors, such as environmental changes, measurement errors, or increased tension due to equipment rust. As a result, the detection accuracy is relatively low, which may lead to false alarms or missed alarms, and it does not effectively reduce or save labor costs.

[0005] Furthermore, on-site cock swapping typically occurs in pairs, with a certain distance between the cocks in each pair. By adjusting camera magnification and other parameters, it can be ensured that only one pair of cocks is present in the camera's field of view. However, the state of the swapped cocks changes significantly, and traditional feature point matching methods like SIFT and SURT cannot extract stable feature pairs. Using this as an offset for compensation can lead to misidentification. Our company utilizes a depth-based quality-aware template matching algorithm (QATM) for feature matching to obtain accurate results. While this algorithm offers good matching performance, it requires certain hardware resources. Since there is only one pair of swapped cocks in the detection image, template matching is not practical, and adding template matching further increases the time consumption.

[0006] Therefore, how to achieve non-stop on-site inspection of coke ovens, eliminate offset compensation operations to improve inspection response efficiency, and reduce labor is a technical challenge that needs to be solved. Summary of the Invention

[0007] This application provides a rapid inspection method for coke oven switching cocks based on a short-term dense cascaded network, which solves the problems of low inspection accuracy, time-consuming and labor-intensive, and slow response in the prior art, and achieves the technical effect of high-precision inspection without stopping the machine on site.

[0008] The technical solution of this invention is: a rapid inspection method for coke oven switching cocks based on short-term dense cascaded networks, characterized in that it includes a non-stop inspection step and a switching cock working status detection step, wherein the non-stop inspection step includes:

[0009] S1.1, Construct the robot inspection map and configure the robot's soft limit start and end points;

[0010] S1.2, control the robot to move to a suitable position, adjust the gimbal angle, find the best detection and photography position of the exchange cock, save the position of the device to be inspected in the center of the visible light camera's field of view, collect the inspection target template image, record the gimbal angle and visible light magnification focal length information, and record the distance between the gimbal and the device under test at this position.

[0011] S1.3 Configure the algorithm recognition rules and import the configured inspection data point data package into the robot via the web;

[0012] S1.4, the robot performs path planning and inspection based on the inspection task and map;

[0013] S1.5 During the inspection process, the robot platform displays the inspection results and alarm information in real time based on the exchange cock status recognition results and the linkage signal rules. After the inspection task is completed, the robot platform will automatically generate a report based on the above information for the coke oven staff to view.

[0014] The switching cock working status detection step includes:

[0015] S2.1, Short-Term Dense Concatenation Network (STDC) is used to segment the key components of the exchanged cock in the image;

[0016] S2.2, determine the angle of the split exchange cock.

[0017] Preferably, the S1.4 robot performs path planning and inspection steps based on the inspection task and map, including:

[0018] S1.4.1) Based on the path planning results, the robot starts moving to the end or start point of the soft limit;

[0019] S1.4.2) The robot moves to inspection point A and adjusts the camera magnification and focus;

[0020] S1.4.3) Control the angle of the pan-tilt unit to allow the visible light camera to constantly track the equipment that needs to be inspected at the inspection point;

[0021] S1.4.4) After the robot enters the positioning range of inspection point A, the robot reaches the current inspection point. The positioning range is a distance value s = v * t calculated based on the robot's running speed v and the maximum time t between the robot's judgment of positioning and preparation for taking a picture. For example, if the target position is a, then [as, a + s] is considered the robot's positioning range. In actual testing, if this range is found to be unsuitable, an offset can be configured using the robot's configuration tool, then the robot's positioning range will be [a - s + offset, a + s - offset]. Efforts should be made to ensure that the robot's position is at the target position when it is taking a picture. Even a slight deviation will not have an impact, because the gimbal is constantly being controlled to track the equipment to be inspected at the inspection point during this process.

[0022] S1.4.5) Action group separation: Schedule parallel threads to continue executing the remaining actions of the action group, such as taking pictures and image recognition;

[0023] S1.4.6) Simultaneously, the main thread is scheduled to execute the next action group; the main thread refers to the thread responsible for controlling the robot's movement, executing core functions such as task planning and scheduling. When performing inspection tasks, the main thread plans and schedules the robot's motion trajectory, action groups, and corresponding action parameters according to the pre-set task plan;

[0024] S1.4.7) Repeat steps S1.4.2) to S1.4.6) until the inspection is completed.

[0025] Preferably, step S2.1 includes:

[0026] 2.1.1) The input image is segmented into multiple sub-regions using a Short-Term Dense Concatenated Network (STDC), and a convolution operation is performed on each sub-region to generate multiple feature maps;

[0027] 2.1.2) Using the Short-Term Dense Concatenation Network (STDC), the dimension of the feature map is gradually reduced by pooling operations to obtain a more abstract feature representation;

[0028] 2.1.3) Use the Short-Term Dense Concatenation Network (STDC) to concatenate all feature maps to obtain the final feature representation.

[0029] Preferably, step 2.1.1 includes:

[0030] First, the Short-Term Dense Cascaded Network (STDC) utilizes a detail aggregation module to generate detailed ground reality.

[0031] Then, the binary cross-entropy loss and DICE loss are used to optimize the learning of detailed information. The detailed aggregation module upsamples the feature map and concatenates it with the original feature map to generate detailed ground reality.

[0032] Finally, the Short-Term Dense Concatenation Network (STDC) uses convolutional operations to fuse detailed ground reality with feature maps to generate the final segmentation result.

[0033] Preferably, in the step of generating detailed ground reality using the detail aggregation module in the Short-Term Dense Cascaded Network (STDC),

[0034] First, Feature Space Modulation (FSM) is used to enhance the low-level feature map to enrich the spatial information and ensure that the final alignment result is as close as possible to the true label GT.

[0035] Next, FAM uses deformable convolutional DCN to learn the offset between high-level and low-level features, and then combines the offset to guide the feature alignment process.

[0036] Preferably, the method for controlling the gimbal angle in step S1.4.3 is as follows:

[0037] The robot is currently located at point A. Point A1 is the center of the visible light field of view of the exchange cock to be tested. Let the coordinates of point A1 be x1 and the distance between the robot's coordinates x2 be d1 = x2 - x1. Let the distance between point A1 and the device under test be d2.

[0038] According to the trigonometric function formula, we get: tan(θ)=d2 / d1, where θ represents the angle that the gimbal needs to rotate;

[0039] Move the gimbal to an angle of 90-θ so that the detection exchange cock at point A1 is located in the center of the visible light field of view.

[0040] Preferably, the method for determining the angle of the segmented exchange cock in S2.2 is as follows: extract the masks of the two segmented exchange cocks, and calculate the principal direction of the masks respectively by PCA principal component analysis; the PCA principal component analysis is used to calculate the principal direction of a specific target and to calculate the angles θ1 and θ2 between the principal direction and the horizontal axis; the left angle of the target is θ3, the right angle of the target is θ4, and β is the set positioning error;

[0041] If the included angle |θ1-θ3|<β and |θ2-θ3|<β, then the lever is rotated to the left to complete the rotation.

[0042] If the included angle |θ1-θ4|<β and |θ2-θ4|<β, then the lever is rotated to the right into position;

[0043] If the included angle does not meet the above conditions, or if only one lever is divided, the lever will not be rotated into place, resulting in an abnormality.

[0044] Preferably, step S2.2 further includes determining whether the position of the lever is consistent with the desired target position through a linkage signal; the linkage signal is a signal that controls the movement direction of the exchange cock. After the robot obtains the linkage signal, it waits for a period of time longer than the movement time of the cock before starting to perform the inspection task. During the inspection process, the algorithm program determines the target position of the cock based on the linkage signal in order to make a positioning judgment.

[0045] The technical solution provided in this application has at least the following technical effects or advantages:

[0046] 1. This invention enables the identification of exchanged cocks through non-stop inspection, filling the gap in non-stop inspection in coke oven scenarios.

[0047] 2. This invention's non-stop inspection solution improves inspection efficiency and effectiveness. Simultaneously, it solves the problem of blurry images caused by taking photos while in motion; by controlling the gimbal's movement, this effect is mitigated.

[0048] 3. By using the FAM feature alignment module, edge detection loss, and boundary guidance loss, the semantic segmentation effect of swapped cock state recognition is optimized. Then, the PCA principal component analysis method is used to improve the detection and judgment ability of swapped cock state.

[0049] 4. This invention eliminates the offset compensation matching process, greatly shortening the inspection time.

[0050] 5. The present invention uses an image detection device to detect the working status of the exchange cock, which better meets the inspection needs in actual scenarios and improves inspection efficiency and effectiveness. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of non-stop inspection according to an embodiment of the present invention.

[0052] Figure 2 This is a flowchart of the non-stop inspection process according to an embodiment of the present invention.

[0053] Figure 3 This is a schematic diagram of the lever structure in an embodiment of the present invention. Detailed Implementation

[0054] Because the coke oven gas heating system requires automatic commands from the switchboard's electrical control system every 30 minutes to start the hydraulic oil pump motor, the switching procedure opens and closes the hydraulic solenoid valves, causing hydraulic oil to push the gas through the hydraulic directional valve. When the gas switching cylinder is working, it drives the gas cock to rotate via a chain. This entire process takes less than one minute. To determine if the cock has rotated to the correct position, all equipment status checks must be completed within the time the switchboard motor stops operating. To meet this requirement, this invention's rapid inspection solution uses a non-stop inspection method, completing the inspection within the time the coke oven cock stops moving, preventing errors in determining whether the cock has rotated to the correct position due to its movement.

[0055] The method for detecting the working status of coke oven exchange cock provided by this invention mainly includes the construction of a robot inspection map, the configuration of the exchange cock inspection point algorithm, robot inspection identification, result output and early warning.

[0056] Example

[0057] To better understand the above technical solution, the following will refer to the appendix to the instruction manual. Figures 1-2 The document provides a detailed explanation of the non-stop inspection scheme and the method for detecting the working status of the switching cock, along with specific implementation examples.

[0058] First, non-stop inspection

[0059] Traditional inspection robots typically stop to take photos and perform recognition, ensuring accuracy and image quality. However, this method is inefficient and cannot meet the requirement of completing an inspection within 30 minutes in a coke oven environment. The overall process of non-stop inspection in this embodiment is as follows:

[0060] S1.1 Construct a robot inspection map and configure the start and end points of the robot's soft limit. The soft limit start and end points ensure that the robot can only move within these directions, preventing the robot from running off the track and causing an accident.

[0061] S1.2 Using the inspection robot configuration tool, the robot is moved to a suitable position. By controlling the angle of the gimbal, the optimal position for exchanging cock detection and taking pictures is found. The position of the device under test in the center of the visible light camera's field of view is saved. A template image of the inspection target is acquired, and information such as the gimbal angle, visible light magnification, and focal length is recorded. The distance between the gimbal and the device under test at this position is also recorded. During track installation, the track and the device under test are installed parallel to each other, and the installation distance is also fixed. Using these parameters, plus the fixed height of the gimbal from the device under test, this information is input into the configuration tool. The configuration tool can calculate the distance between the gimbal and the device under test using trigonometric formulas.

[0062] S1.3 Use the configuration tool to configure the algorithm recognition rules, import the configured inspection data point data package into the robot via the web, and create an inspection task on the web to carry out the inspection.

[0063] The S1.4 robot performs path planning and inspection tasks based on inspection missions and maps. The overall process is as follows:

[0064] S1.4.1) Based on the path planning results, the robot starts moving to the end or start point of the soft limit;

[0065] S1.4.2) The robot moves to inspect point A and adjusts the camera magnification and focus;

[0066] S1.4.3) Control the angle of the gimbal so that the visible light camera tracks the equipment to be inspected at the inspection point during its movement. By calculating the distance between the robot and point A, as well as the angle of the gimbal stored at point A and the distance between the gimbal and the equipment being tested, the angle of the gimbal needs to be rotated is calculated using trigonometric formulas if the detection of point A1 needs to be swapped to the center of the visible light field of view.

[0067] (S1.4.4) Once the robot enters the inspection point A's positioning range, i.e., when the robot's current position is between the start and end points of the positioning range A, it is considered to have reached the current inspection point. The positioning range is a distance value s = v * t calculated based on the robot's running speed v and the maximum time t between the robot's assessment of its positioning and its readiness to take a picture. For example, if the target position is a, then [as, a + s] is considered the robot's positioning range. In actual testing, if this range is found to be unsuitable, an offset can be configured using the robot's configuration tool, making the robot's positioning range [a - s + offset, a + s - offset]. It is important to ensure that the robot's position is at the target position when it is taking a picture; even a slight deviation will not have an impact because the gimbal is constantly tracking the equipment to be inspected at the inspection point.

[0068] S1.4.5) Action group separation: Schedule parallel threads to continue executing the remaining actions of the action group, such as taking pictures and image recognition;

[0069] S1.4.6) Simultaneously, schedule the main thread to execute the next action group;

[0070] S1.4.7) Repeat steps S1.4.2 to S1.4.6 until the inspection is completed.

[0071] During the inspection process, the robot platform displays the inspection results and alarm information in real time based on the exchange cock status recognition results and the linkage signal rules. After the inspection task is completed, the robot platform will automatically generate a report based on the above information for the coke oven staff to view. The staff can check and view the report according to the time, task name, etc.

[0072] Second, the exchange cock state detection method

[0073] S2.1 Exchange Cock Key Part Segmentation

[0074] This embodiment uses a Short-Term Dense Concatenated Network (STDC) to segment the key components of the exchanged cock. During the encoding phase, STDC gradually reduces the dimensionality of the feature maps and uses their aggregation for image representation. Specifically, STDC segments the input image into multiple sub-regions, then performs convolution operations on each sub-region to generate multiple feature maps.

[0075] Then, STDC uses pooling operations to gradually reduce the dimensionality of the feature maps, resulting in a more abstract feature representation. Finally, STDC concatenates all feature maps to obtain the final feature representation.

[0076] During the decoding phase, STDC employs detail guidance to instruct lower-level layers to learn spatial details. Specifically, STDC first uses a detail aggregation module to generate detailed ground truths, and then uses binary cross-entropy loss and DICE loss to optimize the learning of detailed information. The detail aggregation module upsamples the feature map and then concatenates it with the original feature map to generate detailed ground truths. Finally, STDC uses convolutional operations to fuse the detailed ground truths with the feature map, generating the final segmentation result.

[0077] In the STDC network, the Attention Refine Module (ARM) does not consider feature alignment during feature aggregation, leading to inconsistencies in the context due to the direct relationship between pixels in the local feature map and the upsampled feature map. This embodiment replaces the original ARM module with a Feature Alignment Module (FAM) and a Feature Selected Module (FSM). This not only solves the pixel offset problem related to high-level feature upsampling but also achieves effective fusion of high-level and low-level features. Therefore, the segmentation results become more accurate for small objects. Specifically, FSM is first used to enhance the low-level feature map to enrich spatial information, ensuring that the final alignment result is as close as possible to the ground truth (GT). Then, FAM uses deformable convolution (DCN) to learn the offset between high-level and low-level features, and the model uses the offset to guide the feature alignment process. This paper combines the offset with the enhanced low-level features to solve the pixel offset problem related to upsampling in the high-level feature map, achieving effective fusion between high-level and low-level features.

[0078] S2.2 performs angle determination on the split exchange cock.

[0079] Two masks are extracted from the key positions of the swapped cock, and the principal directions of the masks are calculated using Principal Component Analysis (PCA). PCA can be used to calculate the principal direction of a specific target and the angle between that direction and the horizontal axis. This angle can be used to determine whether the lever has rotated to the correct position, and through the linkage signal, to determine whether the lever position matches the desired target position. Figure 3 As shown: 1. Crankshaft 2. Lever 3.

[0080] Principal component analysis (PCA) is used to calculate the principal direction of a specific target and the angles θ1 and θ2 between the principal direction and the horizontal axis; the left angle of the target is θ3, the right angle of the target is θ4, and β is the set positioning error;

[0081] If the included angle |θ1-θ3|<β and |θ2-θ3|<β, then the lever is rotated to the left to complete the rotation.

[0082] If the included angle |θ1-θ4|<β and |θ2-θ4|<β, then the lever is rotated to the right into position;

[0083] If the included angle does not meet the above conditions, or if only one lever is divided, the lever will not be rotated into place, resulting in an abnormality.

[0084] The linkage signal is a signal that controls the direction of the exchanging cock's movement. After the robot receives the linkage signal, it waits for a period of time longer than the cock's movement time before starting to perform the inspection task. During the inspection, the algorithm program determines the target position of the cock based on the linkage signal in order to make a positioning judgment.

[0085] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A rapid inspection method for coke oven switching cocks based on short-term dense cascaded networks, characterized in that, The process includes a non-stop inspection step and a cock operation status detection step. The non-stop inspection step includes: S1.1, Construct the robot inspection map and configure the robot's soft limit start and end points; S1.2, control the robot to move to a suitable position, adjust the gimbal angle, find the best detection and photography position of the exchange cock, save the position of the device to be inspected in the center of the visible light camera's field of view, collect the inspection target template image, record the gimbal angle and visible light magnification focal length information, and record the distance between the gimbal and the device under test at this position. S1.3 Configure the algorithm recognition rules and import the configured inspection data point data package into the robot via the web; S1.4, the robot performs path planning and inspection based on the inspection task and map; S1.5 During the inspection process, the robot platform displays the inspection results and alarm information in real time based on the exchange cock status recognition results and the linkage signal rules. After the inspection task is completed, the robot platform will automatically generate a report based on the above information for the coke oven staff to view. The switching cock working status detection step includes: S2.1, Short-Term Dense Concatenation Network (STDC) is used to segment the key components of the exchanged cock in the image; S2.2, Determine the angle of the split exchange cock; The S1.4 robot performs path planning and inspection steps based on inspection tasks and maps, including: S1.4.1) Based on the path planning results, the robot starts moving to the end or start point of the soft limit; S1.4.2) The robot moves to inspection point A and adjusts the camera magnification and focus; S1.4.3) Control the angle of the pan-tilt unit to allow the visible light camera to constantly track the equipment that needs to be inspected at the inspection point; S1.4.4) Once the robot enters the designated area of ​​inspection point A, the robot will arrive at the current inspection point. S1.4.5) Separate the action group and schedule parallel threads to continue executing the remaining actions of the action group; S1.4.6) Simultaneously, schedule the main thread to execute the next action group; S1.4.7) Repeat steps S1.4.2) to S1.4.6) until the inspection is completed; The method for controlling the gimbal angle in step S1.4.3 is as follows: The robot is currently located at point A. Point A1 is the center of the visible light field of view of the exchange cock to be tested. Let the coordinates of point A1 be x1 and the distance between the robot coordinates x2 be d1 = x2 - x1. Let the distance between point A1 and the device under test be d2. According to the trigonometric function formula, we get: tan(θ) = d2 / d1, where θ represents the angle that the gimbal needs to rotate; Move the gimbal to an angle of 90-θ so that the detection exchange cock at point A1 is located in the center of the visible light field of view; Step S2.1 includes: 2.1.1) The input image is segmented into multiple sub-regions using a Short-Term Dense Concatenated Network (STDC), and a convolution operation is performed on each sub-region to generate multiple feature maps; 2.1.2) Using the Short-Term Dense Concatenation Network (STDC), pooling operations are used to gradually reduce the dimension of the feature map, resulting in a more abstract feature representation; 2.1.3) Use the Short-Term Dense Concatenation Network (STDC) to concatenate all feature maps to obtain the final feature representation; Step 2.1.1 includes: First, the Short-Term Dense Cascaded Network (STDC) utilizes a detail aggregation module to generate detailed ground reality. Then, the binary cross-entropy loss and DICE loss are used to optimize the learning of detailed information. The detailed aggregation module upsamples the feature map and concatenates it with the original feature map to generate detailed ground reality. Finally, the Short-Term Dense Concatenation Network (STDC) uses convolutional operations to fuse detailed ground reality with feature maps to generate the final segmentation result. In the step of generating detailed ground reality using the detail aggregation module in the short-term dense cascaded network (STDC), First, Feature Space Modulation (FSM) is used to enhance the low-level feature map to enrich the spatial information and ensure that the final alignment result is as close as possible to the true label GT. Next, FAM uses deformable convolutional DCN to learn the offset between high-level and low-level features, and then combines the offset to guide the feature alignment process.

2. The rapid inspection method for coke oven switching cock based on a short-term dense cascaded network according to claim 1, characterized in that, The method for determining the angle of the segmented exchange cock in S2.2 is as follows: The two exchanged cock masks are extracted from the segmentation, and the principal directions of the masks are calculated by PCA principal component analysis. The PCA principal component analysis is used to calculate the principal direction of a specific target and the angles θ1 and θ2 between the principal direction and the horizontal axis. The left angle of the target is θ3, the right angle of the target is θ4, and β is the set positioning error. If the included angle |θ1-θ3|<β and |θ2-θ3|<β, then the lever is rotated to the left to complete the rotation. If the included angle |θ1-θ4|<β and |θ2-θ4|<β, then the lever is rotated to the right into position; If the included angle does not meet the above conditions, or if only one lever is divided, and the lever has not been rotated into place, then it is determined that there is an abnormality.

3. The rapid inspection method for coke oven switching cock based on a short-term dense cascaded network according to claim 1, characterized in that, The S2.2 step also includes determining whether the position of the lever is consistent with the expected target position through the linkage signal; the linkage signal is the signal that controls the exchanging cock to move in the direction of movement. After the robot obtains the linkage signal, it waits for more than the cock's movement time before starting to perform the inspection task. During the inspection, the algorithm program determines the target position of the cock based on the linkage signal in order to make a positioning judgment.

Images