Image recognition-based intelligent substation pressure plate state detection method and system

Abstract

The application discloses an image recognition-based intelligent substation pressure plate state detection method and system, which comprises the following steps: acquiring an image of a substation control cabinet, dividing a detection area according to the arrangement rule of the pressure plate in the image, and constructing an area index containing the position and on-off state of the pressure plate; comparing the current position of the pressure plate with the initial position, calculating the position offset, screening the detection area with the position offset exceeding a preset range to mark the abnormal area, and forming abnormal characteristic data; using the abnormal characteristic data to correct the judgment rule, combining the on-off state of the abnormal area to first correct the preset range of the on-off state judgment in the judgment rule, generating an optimized dynamic judgment rule, applying the dynamic judgment rule to all detection areas, judging whether the current position and on-off state of the pressure plate conform to the judgment rule, marking the pressure plate not conforming to the judgment rule, combining the marked abnormal characteristics to second correct the judgment rule, forming an updated dynamic judgment rule, and outputting a detection report.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for power equipment, and more specifically, relates to a method and system for detecting the status of pressure plates in intelligent substations based on image recognition. Background Technology

[0002] Accurate detection of switchboard status is crucial for the safe operation of the power system during substation operation and maintenance. The switchboard's engagement / disengagement status indicates the operating status of related equipment or circuits; abnormal states can lead to control command failures, equipment malfunctions, or even system failures. However, existing technologies for switchboard status detection have the following problems:

[0003] (1) Manual inspection is inefficient and prone to errors.

[0004] Currently, many substations still rely on manual inspections to record the status of pressure plates. This method not only consumes a lot of manpower and time, but is also prone to errors in status recording due to human negligence, especially in scenarios where the pressure plates are arranged in a complex or numerous manner, making it difficult to guarantee accuracy.

[0005] (2) Image recognition accuracy is limited

[0006] Existing image recognition technologies, when detecting the status of pressure plates, typically rely solely on simple color recognition or position matching, making it difficult to handle complex situations such as irregular pressure plate arrangement or blurred insertion / removal markings. For example, pressure plates may experience positional shifts, changes in arrangement characteristics, or degradation of color markings due to long-term use, rendering traditional methods ineffective in identifying anomalies.

[0007] (3) Lack of dynamic adaptive rules

[0008] Traditional detection methods often rely on static rules to set criteria for judging the state of the pressure plate, making it difficult to dynamically optimize these rules based on real-time changes in the detection environment. For example, a fixed threshold cannot handle changes in ambient light or the diverse characteristics of abnormal pressure plates when judging the deployment or retraction status of the pressure plate. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a method and system for detecting the status of pressure plates in intelligent substations based on image recognition.

[0010] The present invention adopts the following technical solution.

[0011] The first aspect of the present invention provides a method for detecting the status of pressure plates in intelligent substations based on image recognition, comprising the following steps:

[0012] Step 1: Obtain an image of the substation control cabinet. Divide the image into multiple detection areas according to the arrangement rules of the pressure plates. Each detection area contains pressure plates that are physically adjacent or functionally related. Construct an area index based on the initial position and engagement / disengagement status of the pressure plates in each detection area.

[0013] Step 2: Based on the region index, determine the position offset between the current position and the initial position of the pressure plate in each detection region. Detection regions with position offsets exceeding the preset range are marked as abnormal regions. Extract the boundary contour, arrangement features, and deployment / retraction status of the pressure plates in the abnormal regions to form abnormal feature data.

[0014] Step 3: Set judgment rules by analyzing abnormal feature data. Specifically, set the position offset range in the judgment rules by using the position offset of the pressure plate in the abnormal area, set the range of arrangement features in the judgment rules by using the data distribution of arrangement features, and set the range of the judgment of the engagement and disengagement status by combining the data of engagement and disengagement status, thereby generating dynamic judgment rules.

[0015] Step 4: Apply the dynamic judgment rule to all detection areas, determine whether the current position and deployment / retraction status of the pressure plate meet the judgment rule, mark the pressure plate that does not meet the judgment rule, extract the area index and abnormal features corresponding to the pressure plate that does not meet the judgment rule, combine the marked abnormal features to correct the judgment rule, form the updated dynamic judgment rule, and output a detection report containing the pressure plate status, area index and abnormal features.

[0016] Preferably, the step of dividing the image into multiple detection regions according to the arrangement rules of the pressure plates, each detection region containing pressure plates that are physically adjacent or functionally related, includes:

[0017] The pressure plates that are physically adjacent are divided as follows:

[0018] For pressure plates arranged in a one-dimensional linear pattern, the physical proximity is determined by whether the distance between the center points of adjacent pressure plates is uniform, and pressure plates with continuous spacing within the set range are divided into the same detection area.

[0019] For pressure plates arranged in a two-dimensional matrix, the detection area is divided into rows or columns based on the alignment of the boundary lines in the horizontal or vertical directions and the uniformity of the center point spacing.

[0020] The functionally related pressure plates are classified as follows:

[0021] Based on the correlation of the pressure plates in the circuit control logic, the pressure plates involved in the same functional module are classified into the same detection area;

[0022] In accordance with the design specifications for substation control cabinets, the pressure plates are logically divided according to functional zones to ensure that the pressure plates in the detection area are functionally consistent.

[0023] The principles of division include:

[0024] If the number of pressure plates in the detection area exceeds the maximum processing unit, it will be subdivided according to the physical proximity rule.

[0025] If there is a conflict between physical location and functional zoning, the detection area shall be divided according to the pressure plate with the relevant function.

[0026] Preferably, the construction of the area index, which includes the position of the pressure plate and the deployment / retraction status, includes:

[0027] The region index is stored as a two-dimensional array of key-value pairs:

[0028] The two-dimensional array includes the positions of the pressure plates corresponding to the rows and columns, and each cell records the pixel coordinates, deployment and retraction status and boundary information of the pressure plates;

[0029] Key-value pairs include those with the region number as the key, storing the number, coordinates, deployment / retraction status, and boundary information of the pressure plates within the region.

[0030] Preferably, the step of determining the positional offset between the current position and the initial position of the pressure plate within each detection area based on the region index, and marking detection areas with positional offsets exceeding a preset range as abnormal areas, includes:

[0031] The boundary contour of the pressure plate is obtained through edge detection. The geometric center of the pressure plate in the image is calculated based on the contour pixels as the current position. The center coordinates of the current position are compared with the initial position, and the quantization offset is calculated using the following formula:

[0032]

[0033]

[0034] In the formula, , , and These are the pixel coordinates of the boundary points; Total number of pixels; initial coordinates of the pressure plate ;

[0035] The preset range is adjusted based on the offset statistics of the abnormal pressure plate:

[0036] Statistically analyze the distribution data of the pressure plate offset within the detection area; dynamically set the offset range based on the 95th percentile of the offset distribution.

[0037] Preferably, the step of extracting the boundary contour, arrangement features, and deployment / retraction status of the pressure plates within the abnormal region to form abnormal feature data includes:

[0038] Edge detection is used to obtain the boundary contour of the pressure plate, and the features of the boundary contour are quantified, including boundary length, number of corner points, and closure. The boundary length is the total number of pixels of the contour line segment, which is used to determine whether there is a defect in the pressure plate. The number of corner points is obtained by the Harris corner detection algorithm to obtain the number of feature points on the boundary, which is used to verify the integrity of the boundary. The closure is used to determine whether the contour forms a closed curve, and the distance between the first and last boundary points is calculated for detection.

[0039] Extracting arrangement features includes determining whether the arrangement conforms to the rules by calculating the spacing between adjacent pressure plates and the alignment deviation; the pressure plate spacing is calculated using the following formula:

[0040]

[0041] In the formula, and These are the pixel coordinates of the boundary points;

[0042] Fit a straight line to the center points of all pressure plates, calculate the average deviation from the line to the point, and determine misalignment if the deviation exceeds the threshold.

[0043] The extraction of the engagement / disengagement status includes using the color of the markings on the pressure plate surface, with red indicating engagement and green indicating disengagement; the engagement / disengagement marking area is extracted using color space segmentation, and the engagement / disengagement status is determined by statistically analyzing the proportions of the extracted marking colors.

[0044] Preferably, the step of setting judgment rules by analyzing abnormal feature data includes:

[0045] The basis for revising the judgment rules is as follows:

[0046] Based on the extracted abnormal feature data, including abnormal position offset, abnormal arrangement feature, and abnormal deployment / retreat status, the rule parameters are dynamically adjusted by analyzing the distribution statistics of the abnormal feature data. This includes setting the offset range based on the 95th percentile of the abnormal area, adjusting the arrangement feature weight according to the abnormal frequency, and dynamically correcting the deployment / retreat status threshold based on the mean of the color components.

[0047] By statistically analyzing the frequency of abnormal pressure plate arrangement characteristics, the weight of the arrangement characteristics is dynamically adjusted when the abnormality rate exceeds 30%, and the range of arrangement parameters is modified at the same time.

[0048] Preferably, the step of setting the range for judging the surrender / retreat status based on the surrender / retreat status data and generating dynamic judgment rules includes:

[0049] The criteria for judging the activation / deactivation status are revised based on the color components and status switching frequency of the abnormal area, including dynamically adjusting the color threshold and setting an upper limit for the status switching frequency, including 10 times / minute.

[0050] The color threshold is dynamically adjusted as follows:

[0051]

[0052] In the formula, and The thresholds for red and green are respectively; and These are the average values ​​for the throwing and retreating states, respectively. and These are the standard deviations for the throwing and withdrawing states, respectively.

[0053] The state transition frequency is used to detect whether the number of state changes of the pressure plate within a certain period of time is abnormal.

[0054] Preferably, the step of applying the dynamic judgment rule to all detection areas, determining whether the current position and deployment / retraction status of the pressure plate conform to the judgment rule, and marking pressure plates that do not conform to the judgment rule includes:

[0055] Position offset determination includes:

[0056] For each pressure plate, calculate the offset between the current position and the initial position, and compare it with the optimized offset range; offsets that exceed the range are marked as abnormal.

[0057] Permutation feature determination includes:

[0058] Based on the arrangement characteristics of the pressure plates, determine whether they conform to the optimized arrangement rules; those with alignment deviations exceeding the threshold are marked as abnormal.

[0059] The determination of the surrender status includes:

[0060] The activation / deactivation status of the pressure plate is analyzed by color and switching frequency to determine whether it conforms to the activation / deactivation status rules; abnormal colors and switching frequencies are marked as abnormal.

[0061] Preferably, the step of extracting the region index and abnormal features corresponding to the pressure plate that does not conform to the judgment rule, and combining the marked abnormal features to correct the judgment rule to form an updated dynamic judgment rule includes:

[0062] During the rule determination process, all pressure plate features marked as abnormal are extracted, including offsets and directions that exceed the offset range and are recorded as position offset abnormalities; arrangement center points that exceed the spacing range and deviate from the arrangement center point and are recorded as arrangement feature abnormalities; color components and switching frequencies that do not conform to the deployment and deactivation rules and are recorded as deployment and deactivation status abnormalities.

[0063] The extracted abnormal features are bound to the corresponding detection area's region index data, including location binding and region binding. The location binding records the abnormal features by the index position of the pressure plate in a two-dimensional array. The region binding generates an abnormal feature list for each detection area, recording the area number and abnormal pressure plate information.

[0064] A second aspect of the present invention provides an intelligent substation pressure plate status detection system based on image recognition, comprising: an image acquisition module, a feature extraction module, a rule optimization module, and an anomaly detection module;

[0065] The image acquisition module is used to acquire images of the substation control cabinet, divide the detection area according to the arrangement rules of the pressure plates in the image, and record the initial position and engagement / disengagement status of the pressure plates in each detection area to construct an area index containing the position and engagement / disengagement status of the pressure plates.

[0066] The feature extraction module is used to compare the current position of the pressure plate with the initial position based on the region index, calculate the position offset, filter out the detection areas whose position offset exceeds the preset range and mark them as abnormal areas, and extract the boundary contour, arrangement features and deployment / retraction status of the pressure plate in the abnormal area to form abnormal feature data.

[0067] The rule optimization module is used to correct the preset range of position offset and the priority of arrangement features in the judgment rules by using abnormal feature data, and to correct the preset range of the judgment of the surrender and retreat status in the first correction of the surrender and retreat status in the abnormal area, thereby generating optimized dynamic judgment rules.

[0068] The anomaly detection module applies the optimized dynamic judgment rules to all detection areas, determines whether the current position and deployment / retraction status of the pressure plate conform to the rules, marks the pressure plates that do not conform to the rules, extracts the area index and anomaly features corresponding to the pressure plates that do not conform to the rules, and combines the marked anomaly features to revise the judgment rules a second time, forming updated dynamic judgment rules, and outputs a detection report containing the pressure plate status, area index and anomaly features.

[0069] Compared with existing technologies, the beneficial effects of the present invention include at least the following: The intelligent substation pressure plate status detection method and system based on image recognition provided by the present invention can effectively solve the detection problems in complex scenarios such as pressure plate position offset, irregular arrangement, and abnormal operation / deactivation status by introducing a dynamic judgment rule optimization mechanism, thereby achieving automated, efficient, and high-precision detection of pressure plate status; through two rule corrections and closed-loop optimizations, the system has good dynamic adaptability and can adjust the judgment criteria in real time for different detection environments; finally, it outputs a detailed detection report containing pressure plate status, area index, and abnormal features, providing accurate and reliable data support for the operation, maintenance, and system optimization of substation pressure plates, significantly reducing manual intervention and improving detection efficiency and accuracy. Attached Figure Description

[0070] Figure 1 This is a network architecture diagram of the ECA attention mechanism provided in accordance with an embodiment of the present invention;

[0071] Figure 2This is a diagram illustrating the angle cost calculation process according to an embodiment of the present invention;

[0072] Figure 3 This is a distance cost calculation principle diagram provided according to an embodiment of the present invention;

[0073] Figure 4 This is a schematic diagram of IoU cost calculation provided in accordance with an embodiment of the present invention;

[0074] Figure 5 These are detection effect diagrams of YOLOE, YOLOE +ECA, YOLOE +SIoU, and the algorithm of this invention provided according to embodiments of the present invention;

[0075] Figure 6 This is an architecture diagram of an intelligent substation pressure plate status monitoring system provided according to an embodiment of the present invention;

[0076] Figure 7 This is a front-end interface diagram of a WeChat mini program provided in accordance with an embodiment of the present invention;

[0077] Figure 8 This is a diagram of the substation control cabinet management interface provided according to an embodiment of the present invention;

[0078] Figure 9 This is a diagram illustrating the pressure plate detection function according to an embodiment of the present invention;

[0079] Figure 10 This is a visualization of the detection results provided in accordance with the embodiments of the present invention;

[0080] Figure 11 This is an example image of a substation control cabinet pressure plate being acquired according to an embodiment of the present invention;

[0081] Figure 12 This is a diagram illustrating the noise reduction effect according to an embodiment of the present invention;

[0082] Figure 13 This is a diagram showing the engagement and disengagement of the pressure plate according to an embodiment of the present invention;

[0083] Figure 14 This is an example diagram of the labeled pressure plate engagement / disengagement status detection dataset provided in accordance with an embodiment of the present invention. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0085] Example 1 of the present invention provides a method for detecting the status of pressure plates in intelligent substations based on image recognition, comprising the following steps:

[0086] Step 1: Obtain an image of the substation control cabinet. Divide the image into multiple detection areas according to the arrangement rules of the pressure plates. Each detection area contains pressure plates that are physically adjacent or functionally related. Construct an area index based on the initial position and engagement / disengagement status of the pressure plates in each detection area.

[0087] Preferably, the step of dividing the detection area according to the arrangement rules of the pressure plates in the image includes:

[0088] The pressure plates are distributed in the substation control cabinet according to physical arrangement rules, including one-dimensional linear arrangement and two-dimensional matrix arrangement;

[0089] In the image of a substation control cabinet, the matrix-arranged pressure plates are characterized by aligned horizontal and vertical boundary lines and uniform spacing between center points; the linearly arranged pressure plates are characterized by aligned vertical boundary lines and uniform spacing between adjacent center points.

[0090] Based on the arrangement characteristics of the pressure plates, including alignment and spacing uniformity, the image is divided into multiple detection regions.

[0091] Preferably, the construction of the area index, which includes the position of the pressure plate and the deployment / retraction status, includes:

[0092] The region index is stored as a two-dimensional array of key-value pairs:

[0093] The two-dimensional array includes the positions of the pressure plates corresponding to the rows and columns, and each cell records the pixel coordinates, deployment and retraction status and boundary information of the pressure plates;

[0094] Key-value pairs include those with the region number as the key, storing the number, coordinates, deployment / retraction status, and boundary information of the pressure plates within the region.

[0095] Step 2: Based on the region index, determine the position offset between the current position and the initial position of the pressure plate in each detection region. Detection regions with position offsets exceeding the preset range are marked as abnormal regions. Extract the boundary contour, arrangement features, and deployment / retraction status of the pressure plates in the abnormal regions to form abnormal feature data.

[0096] Preferably, the step of comparing the current position of the pressure plate with its initial position, calculating the position offset, and filtering out detection areas whose position offset exceeds a preset range and marking them as abnormal areas includes:

[0097] The boundary contour of the pressure plate is obtained through edge detection. The geometric center of the pressure plate in the image is calculated based on the contour pixels as the current position. The center coordinates of the current position are compared with the initial position, and the quantization offset is calculated using the following formula:

[0098]

[0099]

[0100] In the formula, , , and These are the pixel coordinates of the boundary points; Total number of pixels; initial coordinates of the pressure plate ;

[0101] The preset range is adjusted based on the offset statistics of the abnormal pressure plate:

[0102] Statistically analyze the distribution data of the pressure plate offset within the detection area; dynamically set the offset range based on the 95th percentile of the offset distribution.

[0103] Preferably, the step of extracting the boundary contour, arrangement features, and deployment / retraction status of the pressure plates within the abnormal region to form abnormal feature data includes:

[0104] Edge detection is used to obtain the boundary contour of the pressure plate, and the features of the boundary contour are quantified, including boundary length, number of corner points, and closure. The boundary length is the total number of pixels of the contour line segment, which is used to determine whether there is a defect in the pressure plate. The number of corner points is obtained by the Harris corner detection algorithm to obtain the number of feature points on the boundary, which is used to verify the integrity of the boundary. The closure is used to determine whether the contour forms a closed curve, and the distance between the first and last boundary points is calculated for detection.

[0105] Extracting arrangement features includes determining whether the arrangement conforms to the rules by calculating the spacing between adjacent pressure plates and the alignment deviation; the pressure plate spacing is calculated using the following formula:

[0106]

[0107] In the formula, and These are the pixel coordinates of the boundary points;

[0108] Fit a straight line to the center points of all pressure plates, calculate the average deviation from the line to the point, and determine misalignment if the deviation exceeds the threshold.

[0109] The extraction of the engagement / disengagement status includes using the color of the markings on the pressure plate surface, with red indicating engagement and green indicating disengagement; the engagement / disengagement marking area is extracted using color space segmentation, and the engagement / disengagement status is determined by statistically analyzing the proportions of the extracted marking colors.

[0110] Step 3: Set judgment rules by analyzing abnormal feature data. Specifically, set the position offset range in the judgment rules by using the position offset of the pressure plate in the abnormal area, set the range of arrangement features in the judgment rules by using the data distribution of arrangement features, and set the range of the judgment of the engagement and disengagement status by combining the data of engagement and disengagement status, thereby generating dynamic judgment rules.

[0111] Specifically, the judgment rules are dynamically set, and the specific process is as follows:

[0112] Setting the position offset range

[0113] The abnormal feature data includes the positional offset of the pressure plate within the abnormal region. This offset is calculated from the coordinate difference between the current and initial positions and quantized as Euclidean distance. By statistically analyzing the distribution characteristics of these offsets, the mean and 95th percentile are used to set the upper and lower limits of the offset range, ensuring that this range covers most normal offsets. For example, when the offset distribution is between 1 and 10 pixels, the offset range can be set to [1, 10] pixels.

[0114] Setting the range of arrangement features

[0115] The abnormal feature data also includes the arrangement characteristics of the pressure plates, including the spacing and alignment deviation. The spacing is obtained by calculating the distance distribution between the center points of adjacent pressure plates, and the alignment deviation is determined by fitting a straight line to the center points and calculating the deviation from the line.

[0116] Arrangement Spacing: Statistically analyze the distribution range of the spacing between adjacent pressure plates, and set the mean plus or minus the standard deviation as the allowable spacing range. For example, if the spacing is distributed between 15 and 25 pixels, then set the arrangement spacing range to [15, 25] pixels.

[0117] Alignment Deviation: Calculate the maximum deviation of the center point from the fitted line and set the allowable deviation threshold using the 95th percentile. For example, if the alignment deviation ranges from 0 to 3 pixels, then set the alignment deviation threshold to 3 pixels.

[0118] Setting the range for judging the throw / retreat status

[0119] The engagement / disengagement status information is represented by color components (such as the ratio of red to green components), and the abnormal feature data includes the color component ratio distribution of each pressure plate. By statistically analyzing the distribution range of the red and green components, the judgment threshold for engagement / disengagement status is dynamically set.

[0120] Red component threshold (cast state): For example, if the proportion of red component is concentrated between 65% and 85%, then set the red component threshold to [65%, 85%].

[0121] Green component threshold (retreat state): For example, if the proportion of green component is concentrated between 70% and 90%, then set the green component threshold to [70%, 90%].

[0122] Generate dynamic decision rules

[0123] By integrating the position offset range, arrangement feature range, and deployment / retraction status judgment range, a dynamic judgment rule is formed. This rule can comprehensively consider position, arrangement, and status characteristics to determine the status of the pressure plates in all detection areas, providing a unified judgment standard for subsequent detection and anomaly marking.

[0124] Preferably, the preset range of position offset and the priority of arrangement features in the correction determination rule include:

[0125] The basis for revising the judgment rules is as follows:

[0126] Based on the extracted abnormal feature data, including abnormal position offset, abnormal arrangement feature, and abnormal deployment / retreat status, the rule parameters are dynamically adjusted by analyzing the distribution statistics of the abnormal feature data. This includes setting the offset range based on the 95th percentile of the abnormal area, adjusting the arrangement feature weight according to the abnormal frequency, and dynamically correcting the deployment / retreat status threshold based on the mean of the color components.

[0127] The priority of the arrangement features includes the spacing and alignment of the pressure plates; by statistically analyzing the frequency of the arrangement features of abnormal pressure plates, the weight of the arrangement features is dynamically adjusted when the abnormality rate exceeds 30%, and the range of arrangement parameters is modified at the same time.

[0128] Preferably, the optimized dynamic determination rule is generated by combining the preset range of the deployment / retreat status judgment in the first correction judgment rule of the deployment / retreat status of the abnormal area, including:

[0129] The criteria for judging the activation / deactivation status are revised based on the color components and status switching frequency of the abnormal area, including dynamically adjusting the color threshold and setting an upper limit for the status switching frequency, including 10 times / minute.

[0130] The color threshold is dynamically adjusted as follows:

[0131]

[0132] In the formula, and The thresholds for red and green are respectively; and These are the average values ​​for the throwing and retreating states, respectively. and These are the standard deviations for the throwing and withdrawing states, respectively.

[0133] The state transition frequency is used to detect whether the number of state changes of the pressure plate within a certain period of time is abnormal.

[0134] Step 4: Apply the dynamic judgment rule to all detection areas, determine whether the current position and deployment / retraction status of the pressure plate meet the judgment rule, mark the pressure plate that does not meet the judgment rule, extract the area index and abnormal features corresponding to the pressure plate that does not meet the judgment rule, combine the marked abnormal features to correct the judgment rule, form the updated dynamic judgment rule, and output a detection report containing the pressure plate status, area index and abnormal features.

[0135] Preferably, the step of applying the optimized dynamic judgment rule to all detection areas, determining whether the current position and deployment / retraction status of the pressure plate conform to the judgment rule, and marking pressure plates that do not conform to the judgment rule includes:

[0136] Position offset determination:

[0137] For each pressure plate, calculate the offset between the current position and the initial position, and compare it with the optimized offset range; offsets that exceed the range are marked as abnormal.

[0138] Permutation feature determination:

[0139] Based on the arrangement characteristics of the pressure plates (such as spacing and alignment), determine whether they conform to the optimized arrangement rules; those with alignment deviations exceeding the threshold are marked as abnormal.

[0140] Pitching / retirement status determination:

[0141] The activation / deactivation status of the pressure plate is analyzed by color and switching frequency to determine whether it conforms to the activation / deactivation status rules; abnormal colors and switching frequencies are marked as abnormal.

[0142] Preferably, the step of extracting the region index and abnormal features corresponding to the pressure plate that does not conform to the judgment rule, and combining the marked abnormal features to revise the judgment rule a second time, forms an updated dynamic judgment rule, including:

[0143] During the rule determination process, all pressure plate features marked as abnormal are extracted, including offsets and directions that exceed the offset range and are recorded as position offset abnormalities; arrangement center points that exceed the spacing range and deviate from the arrangement center point and are recorded as arrangement feature abnormalities; color components and switching frequencies that do not conform to the deployment and deactivation rules and are recorded as deployment and deactivation status abnormalities.

[0144] The extracted abnormal features are bound to the corresponding detection area's region index data, including location binding and region binding. The location binding records the abnormal features by the index position of the pressure plate in a two-dimensional array. The region binding generates an abnormal feature list for each detection area, recording the area number and abnormal pressure plate information.

[0145] Example 2 of the present invention provides an intelligent substation pressure plate status detection system based on image recognition, including: an image acquisition module, a feature extraction module, a rule optimization module, and an anomaly detection module;

[0146] The image acquisition module is used to acquire images of the substation control cabinet, divide the detection area according to the arrangement rules of the pressure plates in the image, and record the initial position and engagement / disengagement status of the pressure plates in each detection area to construct an area index containing the position and engagement / disengagement status of the pressure plates.

[0147] The feature extraction module is used to compare the current position of the pressure plate with the initial position based on the region index, calculate the position offset, filter out the detection areas whose position offset exceeds the preset range and mark them as abnormal areas, and extract the boundary contour, arrangement features and deployment / retraction status of the pressure plate in the abnormal area to form abnormal feature data.

[0148] The rule optimization module is used to correct the preset range of position offset and the priority of arrangement features in the judgment rules by using abnormal feature data, and to correct the preset range of the judgment of the surrender and retreat status in the first correction of the surrender and retreat status in the abnormal area, thereby generating optimized dynamic judgment rules.

[0149] The anomaly detection module applies the optimized dynamic judgment rules to all detection areas, determines whether the current position and deployment / retraction status of the pressure plate conform to the rules, marks the pressure plates that do not conform to the rules, extracts the area index and anomaly features corresponding to the pressure plates that do not conform to the rules, and combines the marked anomaly features to revise the judgment rules a second time, forming updated dynamic judgment rules, and outputs a detection report containing the pressure plate status, area index and anomaly features.

[0150] like Figure 1 As shown, the ECA attention mechanism affects the input feature map. Global average pooling (GAP) was performed to reduce the feature map dimension from H×W×C to 1×1×C. Then, the output feature vector was processed with a sigmoid activation function. Finally, a one-dimensional convolution was used to weight the input based on its importance, capturing more important information from the input map to obtain a feature map with channel attention. Important features were assigned higher weights, while less important features were assigned lower weights.

[0151] The ECA attention mechanism has relatively low computational overhead because it only involves global average pooling and one-dimensional convolution operations, without requiring expensive matrix multiplication, thus exhibiting high efficiency. This study introduces the ECA attention mechanism into a convolutional neural network to improve the accuracy and precision of the pressure plate throwing / retreating state detection model.

[0152] In object detection, the loss function typically consists of three parts: classification loss, localization loss, and confidence loss. These three parts measure the model's accuracy in predicting the object's class, location, and presence, respectively. In YOLOE, both classification and confidence losses are calculated using the cross-entropy loss function. The localization loss measures the model's accuracy in predicting the bounding box's location. In YOLOE, the Smooth L1 loss function is commonly used to calculate the localization loss, and its formula is shown below:

[0153] (0-1)

[0154] in, It is the number of targets in the image. It is the first The true location information of each target includes the center coordinates of the target bounding box. and width and height ), It is the model's prediction of the first The location information of each target. The Smooth L1 loss function is designed to balance squared loss and absolute loss, using squared loss when the error is small and absolute loss when the error is large. However, this introduction of smoothness can cause the loss function to become unstable when approaching a threshold. , Transitional changes can sometimes make it difficult for the model to converge. Meanwhile, the performance of the Smooth L1 loss function can be affected by hyperparameters (such as thresholds), which typically need to be manually tuned. Choosing inappropriate hyperparameter values ​​can negatively impact the model's training performance.

[0155] To address the shortcomings of the Smooth L1 loss function, this invention introduces SIoU Loss to replace the original Smooth L1 loss function in the localization loss calculation. SIoU redefines the relevant loss function by incorporating directionality into the cost of the loss function. The SIoU loss function consists of four cost functions: angle, distance, shape, and IoU.

[0156] (1) Angle cost

[0157] The SIoU loss function aims to minimize the number of variables in distance-related "wondering" by adding an angle-aware component. Essentially, the model will attempt to make a prediction first on either the X or Y axis (whichever is closest), and then continue approaching along the relevant axis. To achieve this, the convergence process will first attempt to minimize... ,if Otherwise minimize The cost calculation process is as follows: Figure 2 As shown.

[0158] The formula for calculating the angle is as follows:

[0159] (0-2)

[0160] In the formula, certain specific values

[0161] (0-3)

[0162] (0-4)

[0163] (0-5)

[0164] (2) Distance cost

[0165] The SIoU loss function redefines distance cost by considering the angle cost defined above. Its calculation formula is as follows:

[0166] (0-6)

[0167] in , and They are represented as follows:

[0168] (0-7)

[0169] It can be seen that when When the distance reaches 0, the contribution of distance cost decreases significantly. Conversely, when... near When the angle increases, Δ increases. Therefore, the problem becomes increasingly difficult as the angle increases. Furthermore, as the angle increases... Time priority is assigned based on distance value. It's important to note that when... → When the value is 0, distance cost becomes normal. Distance cost calculation is as follows: Figure 3 As shown.

[0170] (3) Shape cost

[0171] The shape cost of the SIoU loss function is defined as follows:

[0172]

[0173] (0-9)

[0174] and The value defines the shape cost, and its value is unique for each dataset. The value is a very important term in this equation; it controls how much attention is required to determine the shape cost. If Setting the value to 1 will immediately optimize the shape, thereby affecting the shape's free movement.

[0175] (4) IoU cost

[0176] In the SIoU loss function, the IoU cost is typically included as part of the loss function and is used to measure the model's accuracy in localizing the target. The principle of IoU cost calculation is as follows: Figure 4 As shown.

[0177] Based on the above calculations of cost losses, the final formula for calculating the loss function is defined as follows:

[0178]

[0179]

[0180] The SIoU loss function can be used to train object detection models. By minimizing the IoU cost, the model can better learn the positioning accuracy of the pressure plate target, thereby improving the accuracy of object detection.

[0181] The IoU cost of the SIoU loss function is a measure of the degree of overlap between two bounding boxes. It is used to evaluate the accuracy of the model in localizing the target and has the characteristics of robustness, interpretability, and training optimization.

[0182] To highlight the effectiveness of this invention's improvement method for the YOLOE model and to quantify the improvement results, comparative experiments were designed to demonstrate the performance enhancement of the YOLOE network model by the two research improvements of this invention. This invention proposes two improvement methods based on the YOLOvE model: the first method introduces the ECA attention mechanism, and the second method replaces the Smooth L1 loss function with the SIoU LOSS loss function. Firstly, to emphasize the improved detection performance of the improved YOLOE network, the model's detection performance is compared with that of various mainstream object detection models in the previous chapter in a table to demonstrate the model's improvement. Simultaneously, group ablation experiments were conducted; ablation experiments are an important method for evaluating models or algorithms.

[0183] To progressively study the optimization of model detection performance by various improvement methods and demonstrate the role and contribution of each improvement, we will delve into the model's internal mechanisms. The ECA attention mechanism and SIoU loss function were deployed into the initial YOLOE network model, and performance testing was performed using the same self-built substation control cabinet pressure plate dataset.

[0184] Table 1. Comparison of detection results between the improved algorithm and the mainstream algorithm.

[0185]

[0186] In the comparative experiment of improved performance, the improved YOLOE model was trained and tested on a substation control cabinet pressure plate dataset. The detection performance was compared with the aforementioned mainstream algorithms. As shown in Table 1, the improved deep learning YOLOE network model achieved enhanced detection performance, with an accuracy of 97.8%, a recall of 96.9%, and a mean accuracy per second (mAP) of 90.4%, representing improvements of 0.9%, 1%, and 2.1% respectively compared to the initial YOLOE object detection model. Besides the improvements in accuracy and mean accuracy per second, the most significant improvement was in the detection speed, reaching 29.86 frames per second. This represents an increase of 5.87 frames per second compared to the original YOLOE object detection model, and even compared to the faster YOLOv7, it still showed an improvement of 2.09 frames per second. It is evident that the improved YOLOE model significantly enhances both detection accuracy and speed, validating the effectiveness of the improved YOLOE method studied in this design. The improved YOLOE method studied in this design scheme has achieved the detection accuracy required by the engineering project. Since the accuracy of manual inspection cannot be expressed in numerical form, this design scheme does not compare data with manual inspection.

[0187] To progressively study the optimization of model detection performance by various improvement methods and demonstrate the role and contribution of each improvement, this design scheme deploys the ECA attention mechanism and SIoU loss function into the initial YOLOE network model, respectively, and uses the same self-built substation control cabinet pressure plate dataset for performance testing. The ablation experiment data is shown in Table 2.

[0188] Table 2 Ablation Experiment Results

[0189]

[0190] A comparative ablation study shows that both introducing the ECA attention mechanism and replacing the SIoU loss function improve the detection accuracy and speed of the YOLOE deep learning model. Comparing the detection accuracy of the two methods, the introduction of the ECA attention mechanism significantly improves both the model's accuracy and mean precision, increasing them by 0.6 and 1.3 percentage points respectively. This demonstrates the significant effect of the ECA attention mechanism in enhancing model detection performance. In contrast, while replacing the SIoU loss function did not significantly improve detection performance, it introduced directionality, leading to faster and more accurate convergence of the deep learning model, resulting in a qualitative leap in detection speed. The speed increased from 24.02 frames per second to 28.18 frames per second. Therefore, the two improvement methods for the YOLOE detection model presented in this invention increase the accuracy and speed of model detection in multiple aspects of the results, validating the effectiveness of the proposed improvement method.

[0191] like Figure 5 The image shows the model's detection performance on substation pressure plate samples. Upon closer inspection, it was found that the YOLOE detection model had the lowest average accuracy, with some pressure plates achieving only 0.85 accuracy. While the influence of individual photographic input cannot be ruled out, the detection performance was significantly inferior. However, the two YOLOE-based improvements studied in this design scheme improved the model's detection performance, achieving an average accuracy of 0.91.

[0192] like Figure 6 As shown, this system uses a front-end and back-end combined approach to achieve intelligent monitoring of substation control cabinet status. The front-end uses WeChat as the platform and has designed a substation control cabinet management mini-program with QR code scanning function to obtain relevant cabinet information, and simultaneously takes and uploads photos of the control cabinet to the back-end monitoring.

[0193] Since the mini-program and the backend are not on the same local area network, an Alibaba Cloud server is rented as the external network server, and frp is used for intranet penetration. frp can expose the intranet host to the Internet through a server with a public IP address, thus enabling direct access to the intranet host from the external network. frp has a server and a client; the server needs to be installed on the server with a public IP address, and the client is installed on the intranet host. Through frp intranet penetration, the image and data requests uploaded by the mini-program are passed to the local server, and then Nignx is used to provide reverse proxy settings for the server to communicate with the mini-program.

[0194] Communication between the mini-program and the local server is via the HTTP protocol. The HTTP protocol uses a request / response model. The client sends a request message to the server, which includes the request method, URL, protocol version, request headers, and request data. The server responds with a status line, which includes the protocol version, success or error code, server information, response headers, and response data. Therefore, this system completes communication between the front-end and back-end by pre-setting the response interface and data.

[0195] The software front-end uses the JavaScript framework and the WeChat Developer Tools platform to design and implement a smart substation management mini-program. WeChat mini-programs offer advantages such as low development cost, fast operation speed, and high user engagement. Therefore, this system chose WeChat mini-programs for its front-end development.

[0196] The backend server framework uses the Python Django application framework. The Django framework has a powerful database access component, ORM, which facilitates database access and invocation. At the same time, it has a full range of functionalities, making it easy for developers to use.

[0197] The WeChat mini program's front-end user interface mainly includes a login interface, a substation control cabinet management interface, and a single inspection and patrol photo upload interface.

[0198] like Figure 7 As shown, on the user login interface, operators enter the username and password pre-registered on the backend server. If the information is correct, the system verifies the current employee's permission level and directs them to different area management and inspection order pages. All interfaces below use the highest administrator privileges as an example.

[0199] After entering the correct username and password, you will be taken to the mini-program homepage, which is the function selection page. The "Station Cabinet Management" button on the page corresponds to the substation control cabinet management interface, and the "Pressure Plate Detection" button corresponds to the single-inspection scanning, photo taking, and uploading interface.

[0200] Click the "Substation Control Cabinet Management" button on the homepage to enter the substation control cabinet management interface. On this page, you can perform CRUD operations on substation control cabinet information. Taking the substation site information interface and the "Add Control Cabinet Information" interface as examples, the interfaces are as follows: Figure 8 As shown.

[0201] As mentioned above Figures 1 to 8As shown, the complete substation control cabinet information management system includes functions for adding, deleting, modifying, and querying information for each substation site, area, and electrical control cabinet. The "Code" button on the site information interface generates a QR code containing the site's information, used to provide the single-inspection function with information about the detected control cabinets, facilitating the retrieval of the initial control panel's on / off status. On the control cabinet information editing page, by entering the row and column numbers of the control cabinet's control panels, the system checks whether the number of control panel on / off statuses in the task table exceeds the limit, reducing errors in the initial on / off status input by staff.

[0202] like Figure 9 As shown, the operator clicks the pressure plate detection button to enter the barcode scanning and pressure plate image capture and upload interface. Barcode scanning utilizes QR codes for different areas and control cabinets to achieve the single-inspection and patrol inspection functions described in the requirements analysis. The patrol inspection function was developed to detect the pressure plate status of all control cabinets within the entire area. Multiple images can be uploaded simultaneously for processing. Taking a single control cabinet inspection as an example, the visualization effect of the inspection results is demonstrated.

[0203] Since this system uses manual mobile phone photography to acquire images of the electrical control cabinet, to reduce the influence of external factors, this design adds two auxiliary lines to the photo upload interface. This is to help operators minimize the angle at which the control cabinet is tilted when taking photos, which could affect model recognition and result comparison, and cause errors in the pressure plate position sorting. Therefore, two horizontal and vertical lines are added to facilitate the correction of image angles.

[0204] Inspectors scan a QR code to obtain information about the electrical control cabinets for individual or routine inspections, retrieve the cabinet number, and search for preset pressure plate activation / deactivation status data. Taking the control cabinet in the image below as an example, correct or incorrect initial pressure plate activation / deactivation statuses are set, and the current control cabinet is monitored. The inspection results are visualized, as shown in the image. Figure 10 As shown.

[0205] The visualization of the above detection results shows that if there is an error between the target detection model result and the preset pressure plate deployment / retraction status when the control cabinet is created, the backend server will use the location information to select the misaligned pressure plate in the original uploaded detection path and return the image to the WeChat mini program frontend, so that staff can make timely corrections.

[0206] First, the front end pre-stores QR code information, substation area information, and the set activation / deactivation status information of the control cabinet pressure plates under normal working conditions within that area. The QR code information includes the area number of each substation area and the cabinet number of each control cabinet within that substation area. When storing the activation / deactivation status information of the control cabinet pressure plates, the activation / deactivation status is extracted based on the color of the markings on the pressure plate surface: red indicates activation, and green indicates deactivation. The activation / deactivation marking area is extracted using color space segmentation, and the activation / deactivation status is determined by statistically analyzing the proportions of the extracted marking colors.

[0207] Then determine the need for individual inspection or patrol inspection. When selecting the individual inspection mode, scan the QR code of the current control cabinet and take a photo of the current control cabinet and upload it. When selecting the patrol inspection mode, scan the QR code of the current substation area and take photos of all control cabinets in order of cabinet number and upload them.

[0208] After the control cabinet photos are uploaded, the location information and engagement / disengagement status of each pressure plate in the current photo are obtained. The pressure plates are then sorted based on their location information and by calculating their Euclidean distance. Finally, the ordered pressure plate engagement / disengagement statuses are retrieved and marked with data to form a result array. The substation area where the control cabinet is located and its corresponding cabinet number are obtained using the QR code information in the current photo. The preset engagement / disengagement status of the corresponding control cabinet is retrieved from pre-stored information, and the engagement / disengagement statuses of the two pressure plates are compared. If a difference is found, the pressure plate statuses at different locations are marked in the photo, generating an incorrect pressure plate annotation image, which is sent to the staff. If they match, a correct comparison is indicated. The collected control cabinet pressure plate images are shown below. Figure 11 As shown.

[0209] Before using the substation pressure plate dataset to test the performance of the target detection model, in order to reduce the training difficulty and training time of the model, it is necessary to preprocess the images in the pressure plate dataset to ensure the quality of the pressure plate images.

[0210] During the acquisition and network transmission of substation pressure plate images, interference and changes caused by unpredictable factors may occur. These interferences and changes are called image noise. This noise affects the quality and visual effect of the acquired images. In other words, noisy images will affect the training and detection of subsequent models, increasing the difficulty and time of training. Therefore, eliminating noise in images is an important task in image preprocessing.

[0211] Gaussian noise, salt-and-pepper noise, and mixed noise are three common types of noise in image processing. Gaussian noise is caused by electromagnetic interference and other factors during image acquisition and transmission. It causes random fluctuations in pixel values, resulting in blurring and distortion. It's called Gaussian noise because the intensity of these fluctuations always follows a Gaussian distribution. Salt-and-pepper noise produces numerous random dots on the image, resembling salt and pepper, which can be white or black. These dots are distributed throughout the image, visually similar to the static noise seen on television; it's sometimes called impulse noise. Mixed noise, as the name suggests, is not a single type of noise but a mixture of multiple noises. When this type of noise appears in an image, the combined effect of these noises makes the image much blurrier than when only one type of noise is present. Because these noises affect image quality and thus target detection, image noise removal is necessary. There are generally three main categories of methods for removing noise from images: filter-based, model-based, and learning-based methods. Each has its own characteristics and drawbacks. For different image scenarios, the most suitable method should be chosen based on the specific circumstances. Considering the application environment of this design, which requires denoising images of control cabinets, model-based methods build a model to describe the noise and signal in the image. However, their computational complexity is too high and does not suit practical applications. Learning-based methods, on the other hand, learn the statistical characteristics of noise by training on a large number of noisy and noise-free images, thereby suppressing noise. Learning-based methods can achieve better denoising results in some cases, but require a large amount of training data. Therefore, filter-based methods are more effective. Median filtering, bilateral filtering, and Gaussian filtering are commonly used filtering algorithms that can be used to remove image noise. This design uses Non-Local Means Denoising (NLMeans) for image denoising in the dataset.

[0212] NLMeans is an image denoising method based on local similarity. It reduces noise by finding the mean of similar regions in an image. It applies nonlocal mean filtering to each pixel in the image, calculates the similarity of the pixel's surrounding neighborhood, and uses the mean of the similar regions as the new value for the center pixel. The denoising effect is as follows: Figure 12 As shown:

[0213] By filtering, organizing, and denoising the collected pressure plate images, a self-built control cabinet pressure plate dataset was obtained, consisting of 6,000 images, all with the metal screen of a substation electrical control cabinet as the background.

[0214] Based on past substation inspection practices, the engagement / disengagement status of the pressure plate is categorized into three types: engaged, disengaged, and removed. The engagement / disengagement status of the pressure plate is as follows: Figure 13As shown:

[0215] Based on the three states of the substation pressure plate being engaged or disengaged, the substation pressure plate dataset was organized and labeled. This study selected 6000 images as the dataset. To use it for training, the dataset needed to be manually labeled using the labelme tool to obtain the corresponding label for each image. During labeling, a bounding box should be drawn around the object being detected, ensuring accurate size and position. After labeling, the location for storing the dataset images and labels should be selected for easy retrieval and program access. This design uses the YOLO format for labeling, such as... Figure 14 The diagram shows how to annotate a dataset. First, draw the annotation boxes, then select the label format. Next, LabelMe will automatically generate a txt file (label file) containing the target information, which stores the target category and coordinate information.

[0216] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0217] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for detecting the status of pressure plates in an intelligent substation based on image recognition, characterized in that, Includes the following steps: Step 1: Obtain an image of the substation control cabinet. Divide the image into multiple detection areas according to the arrangement rules of the pressure plates. Each detection area contains pressure plates that are physically adjacent or functionally related. Construct an area index based on the initial position and engagement / disengagement status of the pressure plates in each detection area. Step 2: Based on the region index, determine the position offset between the current position and the initial position of the pressure plate in each detection region. Detection regions with position offsets exceeding the preset range are marked as abnormal regions. Extract the boundary contour, arrangement features, and deployment / retraction status of the pressure plates in the abnormal regions to form abnormal feature data. Step 3: Set judgment rules by analyzing abnormal feature data. Specifically, set the position offset range in the judgment rules by using the position offset of the pressure plate in the abnormal area, set the range of arrangement features in the judgment rules by using the data distribution of arrangement features, and set the range of the judgment of the engagement and disengagement status by combining the data of engagement and disengagement status, thereby generating dynamic judgment rules. The basis for revising the judgment rules is as follows: Based on the extracted abnormal feature data, including abnormal position offset, abnormal arrangement feature, and abnormal deployment / retreat status, the rule parameters are dynamically adjusted by analyzing the distribution statistics of the abnormal feature data. This includes setting the offset range by the 95th percentile of the abnormal area, adjusting the arrangement feature weight according to the abnormal frequency, and dynamically correcting the deployment / retreat status threshold by the mean of the color components. By statistically analyzing the frequency of abnormal pressure plate arrangement characteristics, the weight of the arrangement characteristics is dynamically adjusted when the abnormality rate exceeds 30%, and the range of arrangement parameters is modified at the same time. The process of combining data on the deployment and withdrawal status to define the range for determining the deployment and withdrawal status and generating dynamic determination rules includes: The criteria for judging the activation / deactivation status are revised based on the color components and status switching frequency of the abnormal area, including dynamically adjusting the color threshold and setting an upper limit for the status switching frequency, including 10 times / minute. The color threshold is dynamically adjusted as follows: T 红 =μ 红 -s 红 ,T 绿 =μ 绿 -s 绿 In the formula, T 红 and T 绿 Thresholds for red and green, respectively; μ 红 and μ 绿 σ represents the mean values ​​for the throwing and retreating states, respectively; 红 and σ 绿 These are the standard deviations of the throwing and retreating states, respectively. The state transition frequency is used to detect whether the number of state changes of the pressure plate within a certain period of time is abnormal. Step 4: Apply the dynamic judgment rule to all detection areas, determine whether the current position and deployment / retraction status of the pressure plate meet the judgment rule, mark the pressure plate that does not meet the judgment rule, extract the area index and abnormal features corresponding to the pressure plate that does not meet the judgment rule, combine the marked abnormal features to correct the judgment rule, form the updated dynamic judgment rule, and output a detection report containing the pressure plate status, area index and abnormal features.

2. The method for detecting the status of pressure plates in an intelligent substation based on image recognition according to claim 1, characterized in that: The image is divided into multiple detection regions according to the arrangement rules of the pressure plates. Each detection region contains pressure plates that are physically adjacent or functionally related, including: The pressure plates that are physically adjacent are divided as follows: For pressure plates arranged in a one-dimensional linear pattern, the physical proximity is determined by whether the distance between the center points of adjacent pressure plates is uniform, and pressure plates with continuous spacing within the set range are divided into the same detection area. For pressure plates arranged in a two-dimensional matrix, the detection area is divided into rows or columns based on the alignment of the boundary lines in the horizontal or vertical directions and the uniformity of the center point spacing. The functionally related pressure plates are classified as follows: Based on the correlation of the pressure plates in the circuit control logic, the pressure plates involved in the same functional module are classified into the same detection area; In accordance with the design specifications for substation control cabinets, the pressure plates are logically divided according to functional zones to ensure that the pressure plates in the detection area are functionally consistent. The principles of division include: If the number of pressure plates in the detection area exceeds the maximum processing unit, it will be subdivided according to the physical proximity rule. If there is a conflict between physical location and functional zoning, the detection area shall be divided according to the pressure plate with the relevant function.

3. The method for detecting the status of pressure plates in an intelligent substation based on image recognition according to claim 1, characterized in that: The construction of the area index, which includes the position of the pressure plate and its deployment / retraction status, includes: The region index is stored as a two-dimensional array of key-value pairs: The two-dimensional array includes the positions of the pressure plates corresponding to the rows and columns, and each cell records the pixel coordinates, deployment and retraction status and boundary information of the pressure plates; Key-value pairs include those with the region number as the key, storing the number, coordinates, deployment / retraction status, and boundary information of the pressure plates within the region.

4. The method for detecting the status of pressure plates in an intelligent substation based on image recognition according to claim 1, characterized in that: The method of determining the positional offset between the current and initial positions of the pressure plate within each detection area based on the region index, and filtering detection areas whose positional offset exceeds a preset range as abnormal areas, includes: The boundary contour of the pressure plate is obtained through edge detection. The geometric center of the pressure plate in the image is calculated based on the contour pixels as the current position. The center coordinates of the current position are compared with the initial position, and the quantization offset is calculated using the following formula: In the formula, x, y, x i and y i These are the pixel coordinates of the boundary points; N is the total number of pixels; the initial coordinates of the pressure plate are (x0, y0); The preset range is adjusted based on the offset statistics of the abnormal pressure plate: Statistically analyze the distribution data of the pressure plate offset within the detection area; dynamically set the offset range based on the 95th percentile of the offset distribution.

5. The method for detecting the status of pressure plates in an intelligent substation based on image recognition according to claim 1 or 4, characterized in that: The extraction of the boundary contours, arrangement features, and deployment / retraction status of the pressure plates within the abnormal area forms abnormal feature data, including: Edge detection is used to obtain the boundary contour of the pressure plate, and the features of the boundary contour are quantified, including boundary length, number of corner points, and closure. The boundary length is the total number of pixels of the contour line segment, which is used to determine whether there is a defect in the pressure plate. The number of corner points is obtained by the Harris corner detection algorithm to obtain the number of feature points on the boundary, which is used to verify the integrity of the boundary. The closure is used to determine whether the contour forms a closed curve, and the distance between the first and last boundary points is calculated for detection. Extracting arrangement features includes determining whether the arrangement conforms to the rules by calculating the spacing between adjacent pressure plates and the alignment deviation; the pressure plate spacing is calculated using the following formula: In the formula, x i+1 and y i+1 These are the pixel coordinates of the boundary points; Fit a straight line to the center points of all pressure plates, calculate the average deviation from the line to the point, and determine misalignment if the deviation exceeds the threshold. The extraction of the engagement / disengagement status includes using the color of the markings on the pressure plate surface, with red indicating engagement and green indicating disengagement; the engagement / disengagement marking area is extracted using color space segmentation, and the engagement / disengagement status is determined by statistically analyzing the proportions of the extracted marking colors.

6. The method for detecting the status of pressure plates in an intelligent substation based on image recognition according to claim 1, characterized in that: The process of applying dynamic judgment rules to all detection areas, determining whether the current position and deployment / retraction status of the pressure plate conform to the judgment rules, and marking pressure plates that do not conform to the judgment rules includes: Position offset determination includes: For each pressure plate, calculate the offset between the current position and the initial position, and compare it with the optimized offset range; offsets that exceed the range are marked as abnormal. Permutation feature determination includes: Based on the arrangement characteristics of the pressure plates, determine whether they conform to the optimized arrangement rules; those with alignment deviations exceeding the threshold are marked as abnormal. The determination of the surrender status includes: The activation / deactivation status of the pressure plate is analyzed by color and switching frequency to determine whether it conforms to the activation / deactivation status rules; abnormal colors and switching frequencies are marked as abnormal.

7. The method for detecting the status of pressure plates in an intelligent substation based on image recognition according to claim 1, characterized in that: The process of extracting the region index and abnormal features corresponding to the pressure plates that do not conform to the judgment rules, and combining the marked abnormal features to correct the judgment rules, forms an updated dynamic judgment rule, including: During the rule determination process, all pressure plate features marked as abnormal are extracted, including offsets and directions that exceed the offset range and are recorded as position offset abnormalities; arrangement center points that exceed the spacing range and deviate from the arrangement center point and are recorded as arrangement feature abnormalities; color components and switching frequencies that do not conform to the deployment and deactivation rules and are recorded as deployment and deactivation status abnormalities. The extracted abnormal features are bound to the corresponding detection area's region index data, including location binding and region binding. The location binding records the abnormal features by the index position of the pressure plate in a two-dimensional array. The region binding generates an abnormal feature list for each detection area, recording the area number and abnormal pressure plate information.

8. An intelligent substation pressure plate condition detection system based on image recognition, comprising: The module comprises an image acquisition module, a feature extraction module, a rule optimization module, and an anomaly detection module; its features are: The image acquisition module is used to acquire images of the substation control cabinet, divide the detection area according to the arrangement rules of the pressure plates in the image, and record the initial position and engagement / disengagement status of the pressure plates in each detection area to construct an area index containing the position and engagement / disengagement status of the pressure plates. The feature extraction module is used to compare the current position of the pressure plate with the initial position based on the region index, calculate the position offset, filter out the detection areas whose position offset exceeds the preset range and mark them as abnormal areas, and extract the boundary contour, arrangement features and deployment / retraction status of the pressure plate in the abnormal area to form abnormal feature data. The rule optimization module is used to correct the preset range of position offset and the priority of arrangement features in the judgment rules by using abnormal feature data, and to correct the preset range of the judgment of the surrender and retreat status in the first correction of the surrender and retreat status in the abnormal area, thereby generating optimized dynamic judgment rules. The anomaly detection module applies the optimized dynamic judgment rules to all detection areas, determines whether the current position and deployment / retraction status of the pressure plate conform to the rules, marks the pressure plates that do not conform to the rules, extracts the area index and anomaly features corresponding to the pressure plates that do not conform to the rules, and combines the marked anomaly features to revise the judgment rules a second time, forming updated dynamic judgment rules, and outputs a detection report containing the pressure plate status, area index and anomaly features.

Images