A monitoring method for the operating state of a pantograph based on key point detection

Abstract

The present invention discloses a monitoring method for the running state of a pantograph based on key point detection, including collecting pantograph images and training to obtain a pantograph detection model; annotating key points of the pantograph on the pantograph images and training an HRNet model to obtain a key point detection model; detecting the pantograph from the pantograph image to be detected through the pantograph detection model, and detecting the key points of the pantograph from the intercepted pantograph area through the key point detection model; obtaining a quantitative index value of the tilt posture of the pantograph bow according to the key point coordinates in the detected key points of the pantograph bow; assigning an abnormal alarm weight set for the quantitative index of the tilt posture of the pantograph bow according to the bow posture amplitude; assigning an abnormal alarm weight set for the quantitative index of the jitter posture of the pantograph bow according to the inter-frame change of the quantitative index of the tilt posture of the pantograph bow between the current frame and the previous frame; taking an image sequence as an alarm unit and outputting an alarm result when the threshold condition is met.

Description

Technical Field

[0001] The present invention relates to the field of rail transit, and specifically to a method for monitoring the operating state of a pantograph based on key point detection. Background Art

[0002] A pantograph is an electrical device installed on the top of a train that obtains electrical energy by friction contact with an overhead catenary. With the increase in the operating speed and mileage of high-speed trains, the probability of the pantograph malfunctioning during operation also increases. To ensure train operation safety and timely detect pantograph faults and potential safety hazards, it is extremely important to monitor the operating state of the pantograph in real time.

[0003] In the early stage, pantograph fault detection mainly relied on manual inspection. After the train stopped in the depot and the power was cut off, workers climbed onto the roof to check for faults in each component. The disadvantages were long detection time and high labor consumption. Currently, online automatic pantograph fault detection mainly falls into two categories: contact type and non-contact type. The contact type uses various sensors installed on the train to collect information such as the mechanical characteristics and electrical parameters of key components of the pantograph and catenary, and further analyzes this information to judge the operating state of the current pantograph and catenary equipment. The disadvantages are interference with train operation, poor versatility, and the influence of signal acquisition noise on the detection results. Non-contact image detection mainly uses high-resolution cameras to extract pantograph features for abnormal state detection. Due to its advantages in flexibility, robustness, and detection performance, it has become the mainstream technical direction for pantograph detection.

[0004] In the existing image-based pantograph state detection and early warning, common methods combine edge detection (such as the Sobel operator) and binarization methods (such as Otsu's method) with empirical thresholds for judgment. However, in actual operation, affected by different lighting, environments, weather, etc., such methods have many false alarms. In recent years, deep learning has shone brightly in the field of images, and there are also some pantograph state monitoring methods based on deep learning. However, most of them only use conventional detectors for supervised learning training to detect common abnormal situations, such as spark arcing, foreign objects, and structural defects. Such supervised learning detectors are limited by the lack of pantograph defect samples, and the trained detection models are only effective under specific conditions such as specific lines, lighting, and weather, with poor versatility. In addition, almost all existing deep learning technical methods analyze single-frame images and cannot monitor dynamic abnormal situations, such as pantograph jitter. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method for monitoring the operating state of a pantograph based on key point detection, including the following steps:

[0006] Step 1: Collect pantograph images, label the pantograph detection frame, perform data cleaning and data augmentation on the pantograph images to obtain a pantograph image set, and use the pantograph image set to train the YOLOv5 model. After training is completed, a pantograph detection model is obtained.

[0007] Step 2: Design and label 10 pantograph key points on the pantograph images after data cleaning and data augmentation to obtain a pantograph image set containing key points, and use the pantograph image set containing key points to train the HRNet model to obtain a key point detection model.

[0008] Step 3: Detect the pantograph from the pantograph image to be detected through the pantograph detection model, intercept the pantograph area through the pantograph detection frame, and detect the pantograph key points from the intercepted pantograph area through the key point detection model.

[0009] Step 4: Obtain the pantograph bow tilt attitude quantization index value according to the coordinates of the 10 detected pantograph key points; when the pantograph bow tilt attitude quantization index is within the set threshold range, assign the abnormal alarm weight set for the pantograph bow tilt attitude quantization index according to the bow attitude amplitude.

[0010] Obtain the pantograph bow jitter attitude quantization index value according to the inter-frame change of the pantograph bow tilt attitude between the pantograph image with the detected pantograph key points and the previous frame image of this image; when the pantograph bow jitter attitude quantization index is within the set threshold range, assign the abnormal alarm weight set for the pantograph attitude quantization index according to the bow attitude amplitude.

[0011] Step 5: Take an image sequence as an alarm unit, add a sliding window of size k to the video input, input k frames of images at a time, obtain the abnormal alarm weight for each frame of image, take the average of the k-frame alarm weights, and output the alarm result when the threshold condition is met.

[0012] Further, obtaining the pantograph bow tilt attitude quantization index value according to the coordinates of the 10 detected pantograph key points includes: obtaining the bow bar plane angle according to the average value of the horizontal angles of the left and right connection points of the front carbon slider and the horizontal angles of the left and right endpoints of the front carbon slider; obtaining the bow support plane angle according to the average value of the horizontal angles of the left and right support points and the horizontal angles of the left and right endpoints of the rear carbon slider; obtaining the horn plane angle according to the horizontal angle of the left and right horn points; obtaining the vertical distance between the key points of the bow bar according to the average value of the vertical differences between the left and right connection points of the front carbon slider and the vertical differences between the left and right endpoints of the front carbon slider; obtaining the vertical distance between the key points of the bow support according to the average value of the vertical differences between the left and right support points and the vertical differences between the left and right endpoints of the rear carbon slider; obtaining the vertical distance between the key points of the horn according to the vertical difference between the left and right horn points.

[0013] Further, when the quantified index of the pantograph bowtilt attitude is within the set threshold range, an abnormal alarm weight set for the pantograph bowtilt attitude is assigned according to the bow attitude amplitude, including: setting a threshold range for each value of the quantified index of the pantograph bowtilt attitude, and assigning the alarm weight within the set alarm weight threshold range according to the size of the quantified index value of the pantograph bowtilt attitude; each quantified index of the pantograph bowtilt attitude will obtain an alarm weight; the comprehensive alarm weight of all is: taking the mean value between the quantified indexes calculated by the same key points, and taking the maximum value between the quantified indexes calculated by different key points.

[0014] Further, the quantified index value of the pantograph bow jitter attitude is obtained according to the frame - to - frame change of the pantograph bowtilt attitude between the pantograph image detected at the key point of the pantograph and the previous frame image of this image, including:

[0015] Obtaining deta_bow bar plane angle according to the absolute value of the difference between the current frame and the previous frame of the bow bar plane angle;

[0016] Obtaining deta_bow support plane angle according to the absolute value of the difference between the current frame and the previous frame of the bow support plane angle;

[0017] Obtaining deta_ram's horn plane angle according to the absolute value of the difference between the current frame and the previous frame of the vertical distance between the bow support key points;

[0018] Obtaining deta_vertical distance of bow bar key points according to the absolute value of the difference between the current frame and the previous frame of the vertical distance between the bow bar key points;

[0019] Obtaining deta_vertical distance of bow support key points according to the absolute value of the difference between the current frame and the previous frame of the vertical distance between the bow support key points;

[0020] Obtaining deta_vertical distance of ram's horn key points according to the absolute value of the difference between the current frame and the previous frame of the vertical distance between the bow support key points.

[0021] Further, when the quantified index of the pantograph bow jitter attitude is within the set threshold range, an abnormal alarm weight set for the pantograph attitude is assigned according to the bow attitude amplitude, including: setting a threshold range for each value of the quantified index of the pantograph bow jitter attitude, and assigning the alarm weight within the set alarm weight threshold range according to the size of the quantified index value of the pantograph bow jitter attitude; each value of the quantified index of the pantograph bow jitter attitude obtains an alarm weight; the comprehensive alarm weight of all is: taking the mean value between the quantified indexes calculated by the same key points, and taking the maximum value between the quantified indexes calculated by different key points.

[0022] The beneficial effects of the present invention are as follows: continuous expansion and strong versatility. Strong versatility means that the algorithm can adapt to changes in different scenarios such as railway lines, lighting, and weather, avoiding the need for customized algorithm adjustments for specific scenarios and greatly improving the economic benefits of the algorithm.

[0023] Most of the existing technologies focus on single-frame image analysis and do not utilize the temporal information of video input. Therefore, they can only detect morphological abnormalities of the pantograph and cannot detect abnormal changes in the pantograph. Even if there are no morphological or structural abnormalities in the pantograph, the abnormal changes during the operation of the pantograph are potential safety hazards. The present invention realizes the detection of abnormal changes in the pantograph jitter based on key point detection, and can accurately locate potential safety hazards according to information such as the pole number and eliminate them in advance.

[0024] Compared with traditional image processing methods, deep learning methods have stronger robustness to complex scenarios and relatively fewer false alarms. In addition, considering the continuity of abnormal states in real scenarios (not just one-frame image alarm), the present invention uses an image sequence as the alarm unit, effectively suppressing false alarms caused by complex backgrounds in single-frame images and reducing the number of false alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic flow chart of a method for monitoring the running state of a pantograph based on key point detection;

[0026] Figure 2 It is a schematic diagram of an embodiment of a method for monitoring the running state of a pantograph based on key point detection;

[0027] Figure 3 It is a schematic diagram of a pantograph detection frame and key point annotation. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] The technical solution of the present invention will be further described in detail below with reference to the drawings, but the protection scope of the present invention is not limited to the following description.

[0029] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Usually, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0030] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention. It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0031] Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0032] The features and performance of the present invention will be further described in detail below in conjunction with embodiments.

[0033] As Figure 1 shown, a pantograph operating state monitoring method based on key point detection includes the following steps:

[0034] Step 1, collect pantograph images, label the pantograph detection frame, and perform data cleaning and data augmentation on the pantograph images to obtain a pantograph image set. Use the pantograph image set to train the yolov5 model, and obtain the pantograph detection model after training is completed;

[0035] Step 2, design and label 10 pantograph key points on the pantograph images after data cleaning and data augmentation to obtain a pantograph image set containing key points. Use the pantograph image set containing key points to train the HRNet model to obtain the key point detection model;

[0036] Step 3, detect the pantograph from the pantograph image to be detected through the pantograph detection model, intercept the pantograph area through the pantograph detection frame, and detect the pantograph key points of the intercepted pantograph area through the key point detection model;

[0037] Step 4, obtain the pantograph bow tilt attitude quantization index value according to the coordinates of the 10 detected pantograph key points; when the pantograph bow tilt attitude quantization index is within the set threshold range, assign the set abnormal alarm weight to the pantograph bow tilt attitude quantization index according to the bow attitude amplitude;

[0038] According to the inter-frame change of the pantograph bow tilt posture between the pantograph image detecting the key points of the pantograph and the previous frame image of this image, obtain the quantization index value of the pantograph bow jitter posture; when the quantization index of the pantograph bow jitter posture is within the set threshold range, assign the set abnormal alarm weight to the pantograph posture quantization index according to the bow posture amplitude;

[0039] Step 5, take an image sequence as an alarm unit, add a sliding window of size k to the video input, input k frames of images at a time, obtain the abnormal alarm weight for each frame of image, take the average of the alarm weights of k frames, and output the alarm result when the threshold condition is met.

[0040] The method of obtaining the quantization index value of the pantograph bow tilt posture according to the coordinates of the detected 10 key points of the pantograph includes: obtaining the bow bar plane angle according to the average value of the horizontal angles of the connections of the left and right connection points of the front carbon slider and the horizontal angles of the connections of the left and right end points of the front carbon slider; obtaining the bow support plane angle according to the average value of the horizontal angles of the connections of the left and right support points and the horizontal angles of the connections of the left and right end points of the rear carbon slider; obtaining the horn plane angle according to the horizontal angle of the connection of the left and right horn points; obtaining the vertical distance between the key points of the bow bar according to the average value of the difference in the vertical coordinates of the left and right connection points of the front carbon slider and the difference in the vertical coordinates of the left and right end points of the front carbon slider; obtaining the vertical distance between the key points of the bow support according to the average value of the difference in the vertical coordinates of the left and right support points and the difference in the vertical coordinates of the left and right end points of the rear carbon slider; obtaining the vertical distance between the key points of the horn according to the difference in the vertical coordinates of the left and right horn points.

[0041] When the quantization index of the pantograph bow tilt posture is within the set threshold range, assign the set abnormal alarm weight to the quantization index of the pantograph bow tilt posture according to the bow posture amplitude, including: setting a threshold range for each quantization index value of the pantograph bow tilt posture, and assigning the alarm weight within the set alarm weight threshold range according to the size of the quantization index value of the pantograph bow tilt posture; each quantization index of the pantograph bow tilt posture will obtain an alarm weight; the comprehensive alarm weight of all is: taking the average value between the quantization indexes calculated by the same key points, and taking the maximum value between the quantization indexes calculated by different key points.

[0042] The method of obtaining the quantization index value of the pantograph bow jitter posture according to the inter-frame change of the pantograph bow tilt posture between the pantograph image detecting the key points of the pantograph and the previous frame image of this image includes:

[0043] Obtain deta_bow bar plane angle according to the absolute value of the difference in the bow bar plane angle between the current frame and the previous frame;

[0044] Obtain deta_bow support plane angle according to the absolute value of the difference in the bow support plane angle between the current frame and the previous frame;

[0045] The deta_horn plane angle is obtained according to the absolute value of the vertical distance difference between the current frame and the previous frame.

[0046] Deta_vertical spacing of the key points of the arch is obtained according to the absolute value of the difference between the vertical spacing of the key points of the arch in the current frame and the previous frame;

[0047] Deta_bow support key point vertical spacing is obtained according to the absolute value of the difference between the current frame and the previous frame;

[0048] The deta_horn key point vertical spacing is obtained according to the absolute value of the difference between the vertical spacing of the bow support key points in the current frame and the previous frame.

[0049] When the pantograph bow shake posture quantitative index is within the set threshold range, the abnormal alarm weight set for the pantograph posture quantitative index is assigned according to the bow posture amplitude, including: setting a threshold range for each pantograph bow shake posture quantitative index value, and allocating the alarm weight within the set alarm weight threshold range according to the size of the pantograph bow shake posture quantitative index value; each pantograph bow shake posture quantitative index value is assigned an alarm weight; the comprehensive alarm weights are: taking the average between the quantitative indicators calculated at the same key points, and taking the maximum value between the quantitative indicators calculated at different key points.

[0050] Specifically, Figure 2 As shown, the following steps are included:

[0051] Step 1: Collect the pantograph image data in operation, mark the pantograph detection frame, and after routine data cleaning and data enhancement, train the yolov5 model for pantograph detection.

[0052] Step 2: According to the appearance characteristics of the pantograph, 10 key points were designed, such as Figure 3 As shown, the pantograph key points are marked on the image collected in the previous step, and the HRNet model is trained to detect the key points.

[0053] Key design principles of pantograph:

[0054] The pantograph is a rigid body. When it is deformed, its structural edges and joints generally undergo certain changes in shape or position, which means that these local areas are more sensitive to abnormal changes. Based on this characteristic, 10 key points are designed for the double-slide pantograph at the carbon slider, the end of the horn and each joint.

[0055] Step 3: After the first two steps of model training are completed, the model inference step is performed. Yolov5 detects the pantograph, and the pantograph area is cut out by the predicted detection frame. The HRNet+DarkPose model is used to detect the pantograph key points in this area.

[0056] Step 4: According to the bow attitude characteristics of two common pantograph abnormal conditions, namely bow tilt and bow jitter, taking the coordinates of 10 key points as input, calculate the pantograph attitude quantization index. For the abnormal situation of bow jitter, not only the key points of the current image frame are used, but also the bow attitude is calculated based on the inter-frame changes of the key points. When the calculated pantograph attitude quantization index is within the artificially set threshold range, a certain alarm weight is assigned according to the amplitude of the bow attitude.

[0057] 1) Bow tilt

[0058] Pantograph attitude quantization index:

[0059] a. Bow bar plane angle: The average value of the horizontal angle between the connection line of the left and right connection points of the front carbon sliding bar and the horizontal angle between the connection line of the left and right end points of the front carbon sliding bar.

[0060] b. Bow support plane angle: The average value of the horizontal angle between the connection line of the left and right support points and the horizontal angle between the connection line of the left and right end points of the rear carbon sliding bar.

[0061] c. Horn plane angle: The horizontal angle between the connection line of the left and right horn points.

[0062] d. Vertical distance between key points of the bow bar: The average value of the difference in the vertical coordinates of the left and right connection points of the front carbon sliding bar and the difference in the vertical coordinates of the left and right end points of the front carbon sliding bar.

[0063] e. Vertical distance between key points of the bow support: The average value of the difference in the vertical coordinates of the left and right support points and the difference in the vertical coordinates of the left and right end points of the rear carbon sliding bar.

[0064] f. Vertical distance between key points of the horn: The difference in the vertical coordinates of the left and right horn points.

[0065] Manually set a threshold range for each quantization index. For example, the threshold range of the bow bar plane angle is set to [4, 6]. When the bow bar plane angle is within the threshold range, a certain alarm weight is evenly distributed according to the quantization value. The alarm weight also has an assignment range, for example, [0.5, 1]. Suppose the bow bar plane angle is equal to 5, and the alarm weight is 0.5 + (1 - 0.5) * (5 - 4) / (6 - 4) = 0.75.

[0066] Each pantograph attitude quantization index will obtain an alarm weight. The principle of integrating all alarm weights is: take the average value between the quantization indexes calculated by the same key points, and take the maximum value between the quantization indexes calculated by different key points.

[0067] 2) Bow jitter

[0068] Pantograph attitude quantization index:

[0069] a. Deta_bow bar plane angle: The absolute value of the difference between the bow bar plane angle of the current frame and the previous frame.

[0070] b. deta_ Bow support plane angle: The absolute value of the difference in the bow support plane angle between the current frame and the previous frame.

[0071] c. deta_ Ram's horn plane angle: The absolute value of the difference in the ram's horn plane angle between the current frame and the previous frame.

[0072] d. deta_ Vertical distance between key points of the bow strip: The absolute value of the difference in the vertical distance between key points of the bow strip between the current frame and the previous frame.

[0073] e. deta_ Vertical distance between key points of the bow support: The absolute value of the difference in the vertical distance between key points of the bow support between the current frame and the previous frame.

[0074] f. deta_ Vertical distance between key points of the ram's horn: The absolute value of the difference in the vertical distance between key points of the ram's horn between the current frame and the previous frame.

[0075] Step 5: Considering the continuity of abnormal states in the real scene (not just one-frame image alarm), an image sequence is used as the alarm unit to effectively suppress false alarms caused by complex backgrounds in single-frame images and greatly reduce the number of false alarms. A sliding window of size k is added to the video input, and k frames of images are taken as input at a time. Each frame of image obtains the abnormal alarm weight through Steps 1 - 4, and then the average value of the multi-frame alarm weights is taken. When the threshold condition is met, the alarm result is output.

[0076] The above is only the preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications, and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in related fields. And the changes and alterations made by those skilled in the art that do not depart from the spirit and scope of the present invention shall all be within the protection scope of the appended claims of the present invention.

Claims

1. A monitoring method for the running state of a pantograph based on key point detection, characterized in that The steps are as follows: Step 1: Collect pantograph images, label the pantograph detection frames, perform data cleaning and data augmentation on the pantograph images to obtain a pantograph image set, and use the pantograph image set to train the YOLOv5 model. After training is completed, a pantograph detection model is obtained; Step 2: Design and label 10 pantograph key points on the pantograph images after data cleaning and data augmentation to obtain a pantograph image set containing key points, and use the pantograph image set containing key points to train the HRNet model to obtain a key point detection model; Step 3: Detect the pantograph from the pantograph image to be detected through the pantograph detection model, intercept the pantograph area through the pantograph detection frame, and detect the pantograph key points from the intercepted pantograph area through the key point detection model; Step 4: Obtain the quantization index value of the pantograph bow tilt posture according to the coordinates of the 10 detected pantograph key points; when the quantization index value of the pantograph bow tilt posture is within the set threshold range, assign the abnormal alarm weight set for the quantization index of the pantograph bow tilt posture according to the bow posture amplitude; Obtain the quantization index value of the pantograph bow jitter posture according to the inter-frame change of the pantograph bow posture between the pantograph image with detected key points and the previous frame image of this image; when the quantization index value of the pantograph bow jitter posture is within the set threshold range, assign the abnormal alarm weight set for the quantization index of the pantograph bow jitter posture according to the bow posture amplitude; Step 5: Use an image sequence as an alarm unit, add a sliding window of size k to the video input, take k frames of images for input at a time, obtain the abnormal alarm weight for each frame of image, take the average of the alarm weights of k frames, and output the alarm result when the threshold condition is met; The obtaining of the quantization index value of the pantograph bow jitter posture according to the inter-frame change of the pantograph bow posture between the pantograph image with detected key points and the previous frame image of this image includes: Obtain deta_bow bar plane angle according to the absolute value of the difference in the bow bar plane angle between the current frame and the previous frame; Obtain deta_bow support plane angle according to the absolute value of the difference in the bow support plane angle between the current frame and the previous frame; Obtain deta_horn plane angle according to the absolute value of the difference in the horn plane angle between the current frame and the previous frame; Obtain deta_bow bar key point vertical spacing according to the absolute value of the difference in the vertical spacing of the bow bar key points between the current frame and the previous frame; Obtain deta_bow support key point vertical spacing according to the absolute value of the difference in the vertical spacing of the bow support key points between the current frame and the previous frame; Obtain deta_horn key point vertical spacing according to the absolute value of the difference in the vertical spacing of the horn key points between the current frame and the previous frame.

2. The pantograph running state monitoring method based on key point detection according to claim 1, characterized in that The method for obtaining the quantitative index value of the pantograph tilt attitude based on the coordinates of 10 key pantograph points detected includes: obtaining the bow plane angle based on the average value of the horizontal angle between the left and right connection points of the front carbon slider and the horizontal angle between the left and right end points of the front carbon slider; obtaining the bow support plane angle based on the average value of the horizontal angle between the left and right support points and the horizontal angle between the left and right end points of the rear carbon slider; obtaining the horn plane angle based on the horizontal angle between the left and right horn points; obtaining the vertical distance between key points of the bow strip based on the average value of the difference in the vertical coordinates of the left and right connection points of the front carbon slider and the difference in the vertical coordinates of the left and right end points of the front carbon slider; obtaining the vertical distance between key points of the bow support based on the average value of the difference in the vertical coordinates of the left and right support points and the difference in the vertical coordinates of the left and right end points of the rear carbon slider; obtaining the vertical distance between key points of the horn based on the difference in the vertical coordinates of the left and right horn points.

3. The pantograph operation state monitoring method based on key point detection according to claim 2, characterized in that When the quantitative index value of the pantograph tilt attitude is within the set threshold range, an abnormal alarm weight is assigned to the quantitative index of the pantograph tilt attitude according to the amplitude of the bow attitude, including: setting a threshold range for each quantitative index of the pantograph tilt attitude, and assigning the alarm weight within the set alarm weight threshold range according to the size of the quantitative index value of the pantograph tilt attitude; each quantitative index of the pantograph tilt attitude will obtain an alarm weight; the comprehensive alarm weight of all is: the alarm weights of the quantitative indexes calculated by the same key points are averaged, and the alarm weights of the quantitative indexes calculated by different key points take the maximum value.

4. The monitoring method for the operating state of a pantograph based on key point detection according to claim 3, characterized in that, When the quantitative index value of the pantograph vibration attitude is within the set threshold range, an abnormal alarm weight is assigned to the quantitative index of the pantograph attitude according to the amplitude of the bow attitude, including: setting a threshold range for each quantitative index of the pantograph vibration attitude, and assigning the alarm weight within the set alarm weight threshold range according to the size of the quantitative index value of the pantograph vibration attitude; each quantitative index value of the pantograph vibration attitude obtains an alarm weight; the comprehensive alarm weight of all is: the alarm weights of the quantitative indexes calculated by the same key points are averaged, and the alarm weights of the quantitative indexes calculated by different key points take the maximum value.

Images