Intelligent detection method and system for railway catenary suspension device based on linear array camera

Abstract

The application discloses a kind of railway overhead line system suspension device intelligent detection method and system based on linear array camera, belong to railway operation and maintenance safety detection technical field.The application uses TDI linear array camera, speed sensor and IMU collaborative work.Intelligent control module adjusts TDI integration rank according to vehicle speed synchronous camera line frequency, to inhibit uniform speed blur, adapt to illumination change;At the same time, according to IMU data, real-time image restoration operator is generated, to characterize vibration blur.After data processing module receives image, the operator is used to carry out dynamic deconvolution restoration to correct vibration blur, and then the defect is identified by deep learning model.The application realizes all-weather, high-definition automatic detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of railway operation safety detection technology, and particularly relates to a method and system for automatically and intelligently detecting a catenary suspension device based on a linear array camera applied to a high-speed railway. BACKGROUND

[0002] As the artery of modern transportation, the safe and stable operation of high-speed railways is of great importance. The catenary is a key infrastructure for direct power supply to high-speed trains, and its structural state is directly related to train safety. The catenary is suspended on the support structure through a large number of suspension devices, such as suspension insulators, positioning tubes, wire clamps, and droppers. Under the long-term effects of aerodynamic impact during high-speed train operation (usually above 300 km / h), mechanical impact of the pantograph, and complex natural environments such as wind, rain, snow, and temperature difference, the fasteners (such as bolts and pins) of these suspension devices are prone to looseness, shedding, and other defects such as damage and excessive wear of the components themselves. If these seemingly minor defects are not discovered and addressed in a timely manner, they may cause major failures and pose a serious threat to train safety.

[0003] Currently, there are two main ways to detect catenary suspension devices: traditional manual inspection and detection vehicles equipped with machine vision systems. Manual inspection usually needs to be carried out during the "window point" of railway operation (i.e., the short window period of no train operation at night), and the operating personnel perform close-range visual inspection by ladder car or climbing the support column. This method has many inherent drawbacks: first, the labor intensity is high, and the operating efficiency is extremely low, which cannot meet the needs of long-distance and high-density detection; second, the detection results are heavily dependent on personnel experience, and are highly subjective, easily affected by factors such as fatigue and weather, and there is a risk of missed detection and false detection; third, high-altitude and night-time operations themselves have high safety risks.

[0004] In order to overcome the shortcomings of manual inspection, the industry has developed detection vehicles equipped with machine vision systems, among which systems equipped with ordinary area array cameras are the most common. However, under high-speed running conditions, such existing detection systems face two insurmountable technical bottlenecks, resulting in detection results far from ideal:

[0005] First, the motion blur problem is serious. High-speed motion is the enemy of face array camera imaging. At a speed of 360 km / h (i.e. 100 m / s), even if a high-speed industrial camera with 1 / 2000 s (i.e. 500 microseconds) is used, during the exposure time, the train has already traveled 50 mm. Such a large displacement reflected in the image will cause a long smear on the originally clear object edge, causing serious motion blur. For catenary detection, the defect features to be identified are often very small, such as the small gap between the nut and the screw due to loosening, the shedding of the cotter pin, and the hairline cracks on the surface of the insulator. Under severe motion blur, these key small detail features are completely submerged and cannot be clearly imaged, which fundamentally limits the recognition ability of subsequent image algorithms, resulting in a high rate of missed detection of key defects.

[0006] Second, the imaging quality consistency is poor. The detection vehicle will experience extremely complex and variable lighting environments during a detection task, such as the alternation of day and night, frequent entry and exit of tunnels, passing under bridges or in the shadow area of tall buildings, etc. The illuminance of the environmental light can change dramatically by tens of thousands of lux in an instant. The traditional face array camera system usually adopts an automatic exposure strategy, but its response speed is far from keeping up with the changes in light during high-speed travel, resulting in severe underexposure of the image at the moment of entering the tunnel (commonly known as "black frame"), and severe overexposure at the moment of exiting the tunnel (commonly known as "white frame"). Such images that are bright and dark at random and have extremely inconsistent quality are "fatal" for deep learning recognition models that rely on data-driven. The performance of the algorithm is highly dependent on the consistency of the training data and the test data, and unstable imaging quality can severely damage this consistency, resulting in poor robustness of the algorithm, a significant increase in false positives and false negatives, and making it difficult to achieve stable and reliable performance in actual applications.

[0007] Therefore, how to obtain high-quality images without motion blur, uniform exposure, and clear details under high-speed, all-weather, and all-scenario conditions, and on this basis, to carry out efficient and accurate intelligent defect recognition, is a technical problem that needs to be solved in the current railway operation and maintenance field. SUMMARY

[0008] The purpose of the present application is to provide an innovative railway catenary suspension device intelligent detection method and system based on a linear array camera, which solves the problems of uniform speed blur under high-speed motion, vibration blur, and imaging quality consistency under complex lighting environments through the synergistic compensation of time delay integration (TDI) hardware technology and software restoration algorithms.

[0009] The present application provides a railway catenary suspension device intelligent detection system based on a linear array camera, comprising:

[0010] An image acquisition module (100) is installed on a detection vehicle for acquiring image data of a catenary suspension device, the image acquisition module (100) comprising:

[0011] a TDI line array camera (101) with its lens optical axis pointing to the catenary suspension device to be detected;

[0012] a continuous linear light source (102) providing continuous and uniform illumination for the scanning field of view of the TDI line array camera (101);

[0013] a speed sensor (104) for acquiring real-time running speed of the detection vehicle; and

[0014] an inertial measurement unit (103) for detecting high-frequency vibration data of the detection vehicle in a direction perpendicular to the running direction;

[0015] an intelligent control module (200) electrically connected to the image acquisition module (100) for synchronously controlling the line scanning frequency of the TDI line array camera (101) according to the real-time running speed, and dynamically adjusting the integration order of the TDI line array camera (101) according to the illumination variation of the detection environment; at the same time, the module is also used for generating an image restoration operator in real time according to the high-frequency vibration data;

[0016] a data processing module (300) for receiving image data acquired by the TDI line array camera (101), dynamically restoring the image data by using the image restoration operator to correct the non-linear motion blur caused by vibration, and processing the restored image by using a defect identification model to identify defects of the catenary suspension device.

[0017] Optionally, the intelligent control module (200) is configured to preferentially increase the integration order of the TDI line array camera (101) to increase the exposure amount when the ambient light decreases, and then increase the output power of the continuous linear light source (102) when the integration order reaches an upper limit.

[0018] Optionally, the sampling axis of the inertial measurement unit (103) is parallel or at a preset angle to the pixel arrangement direction of the TDI line array camera (101).

[0019] Optionally, the image restoration operator is a dynamic point spread function, and the intelligent control module (200) calculates the parameters of the function in real time according to the amplitude and frequency of the high-frequency vibration data.

[0020] Optionally, the dynamic restoration processing of the data processing module (300) is a row-by-row deconvolution operation based on Wiener filtering or Lucy-Richardson algorithm, and each row or array of image data collected is corrected for blurring using the dynamic point spread function.

[0021] Optionally, the defect identification model is a two-stage deep learning model, the first stage uses a target detection network to identify and locate the suspension device in the image, and the second stage uses an image classification network to classify the defect types of the located suspension devices.

[0022] The application also provides a line array camera-based intelligent detection method for railway catenary suspension devices using the intelligent detection system.

[0023] Multi-source data synchronous acquisition step: when the detection vehicle is running, the vehicle speed is acquired in real time through a speed sensor (104), and high-frequency vibration data of the vehicle in a direction perpendicular to the running direction are acquired in real time through an inertial measurement unit (103);

[0024] Adaptive exposure and synchronous scanning step: the line scanning frequency of a TDI line array camera (101) is controlled synchronously according to the vehicle speed, and the integration order of the TDI line array camera (101) is dynamically adjusted according to the environmental light intensity, so that image acquisition is performed under the illumination of a continuous linear light source (102);

[0025] Dynamic blur modeling step: an image restoration operator for characterizing nonlinear motion blur is generated in real time according to the high-frequency vibration data;

[0026] Image restoration and defect identification step: the collected image data are subjected to dynamic restoration processing using the image restoration operator, and then the restored image is input into a defect identification model to identify the defects of the catenary suspension device.

[0027] Optionally, in the adaptive exposure and synchronous scanning step, dynamically adjusting the integration order of the TDI line array camera (101) includes: presetting a target image gray scale range, when the average gray scale value of the real-time collected image is lower than the range, gradually increasing the integration order until the gray scale value meets the standard or the integration order reaches the maximum value of the camera.

[0028] Optionally, the dynamic blur modeling step includes: converting the high-frequency vibration data into pixel displacement of the TDI line array camera (101) in an integration period through a preset transfer function model, and constructing a point spread function describing the displacement as an image restoration operator.

[0029] Optionally, after the image restoration and defect identification step, the method further comprises a report generation step: combining the identified defect type with the real-time acquired geographic position information or line mileage information to generate a detection report containing the defect type and the accurate position.

[0030] Compared with the prior art, the present application has the following advantages:

[0031] 1) Innovative blur suppression path: creatively proposes the collaborative scheme of "TDI suppression of uniform blur + software restoration correction of vibration blur". The TDI line array camera perfectly offsets the smear caused by the uniform straight-line motion of the train through the synchronous transfer of charge packets and the measured object; at the same time, IMU is first introduced to accurately quantify random vibration, and image restoration algorithm is used to remove this non-linear blur at a specific point, realizing the suppression of all types of motion blur.

[0032] 2) Stronger light adaptability: using the multi-stage integration characteristics of the TDI line array camera, the exposure amount is dynamically adjusted by 2N times (N is the integration level) without changing the power of the light source or the shutter speed. This makes the system have faster response speed, wider adjustment range and more stable imaging gray scale in scenes such as entering and exiting tunnels where the light changes dramatically.

[0033] 3) Deep integration of software and hardware, high system robustness: the present application deeply integrates the hardware characteristics of IMU sensors and TDI line array cameras with advanced image processing algorithms to form a complete "perception-prediction-compensation-correction" closed loop. This scheme not only has clear imaging, but also has strong robustness for imaging quality under harsh line conditions (such as high-frequency vibration caused by track irregularities). BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0035] Figure 1 : Structure block diagram of the intelligent detection system in the embodiment of the present application.

[0036] Figure 2 : Flow chart of the intelligent detection method in the embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical scheme and advantages of the present application more clear, the following will combine the drawings and specific embodiments to describe the present application in detail. Referring to Figure 1The application provides an intelligent detection system which can be deployed on a comprehensive detection train of a high-speed railway. The system mainly comprises an image acquisition module 100, an intelligent control module 200, a data processing module 300 and an optional positioning module 400.

[0038] Embodiment 1: Adaptive exposure control based on TDI integration step adjustment

[0039] This embodiment aims to illustrate the adaptive adjustment capability of the application in the light mutation scene and its superiority.

[0040] Detection scene: The detection train drives into a tunnel (<1 Lux) with almost no natural light at a speed of 350 km / h (about 97.2 m / s) from a sunny daylight environment with an illumination intensity of about 100,000 Lux.

[0041] System configuration: The image acquisition module 100 adopts a 8K resolution TDI line array camera 101 (such as Piranha XL series) of Teledyne DALSA Company, which supports a maximum of 256 integration levels. A continuous LED linear light source with a rated power of 500 W is equipped, and its output power is constant at 80% in this embodiment. The ideal 8-bit image average gray value target interval preset by the intelligent control module 200 is [150, 220].

[0042] Control process:

[0043] Stable operation outside the tunnel: In the strong light environment outside the tunnel, to avoid overexposure, the intelligent control module 200 sets the integration level of the TDI line array camera 101 to the lowest 4 levels. At this time, the environmental light is the main imaging light source, and the LED light source is used as a fill light. The average gray value of the collected image is stable at about 185, which is within the target interval.

[0044] Instantaneous entry into the tunnel: When the camera enters the tunnel entrance, the environmental illumination drops to nearly 0 within a few meters. If no adjustment is made, only the constant light source and 4-level integration, the image gray value will drop to below 20, causing the image content to be almost black, i.e. "black frame", and all suspended device information is lost.

[0045] Adaptive adjustment starts: The image gray value monitoring unit built-in the intelligent control module 200 detects that the gray value is far below the threshold value 150 at the first acquisition line after the camera enters the tunnel. The control module immediately starts the adaptive exposure program.

[0046] Exponential integral level increase: To achieve the fastest response, the control module does not increase the integral level step by step, but uses an exponential step to jump up. For example, in the first control cycle (about a few milliseconds), the integral level is increased from 4 levels to 32 levels; in the second cycle, according to the gray value feedback, it is continuously increased to 128 levels; in the third cycle, it is fine-tuned to 256 levels. The whole process is completed within a distance of less than 10 meters (about 0.1 second) of train advancement.

[0047] Restore stable imaging: Under the 256-level integration, the photosensitive capability (exposure) of the camera is increased by 256 / 4=64 times compared with the 4-level integration. At this time, relying entirely on the illumination of the continuous linear light source 102, the average gray value of the image quickly rises and stabilizes at about 175, re-entering the ideal interval.

[0048] Technical effect: This embodiment realizes instantaneous and accurate compensation for light mutation scenes through electronic and exponential TDI integral level adjustment. Compared with adjusting high-power LED light sources (which usually involves power response delay and impact on LED life), this method responds faster, controls more finely, and has less hardware wear and tear. It ensures data continuity and high quality in extreme light change areas such as tunnel entrances, avoiding defects and missed detection caused by "black frames" or underexposure.

[0049] Embodiment 2: Non-linear motion blur correction based on IMU and speed sensor

[0050] This embodiment aims to detail how the present application accurately corrects non-linear motion blur caused by random vibration of the vehicle body, which is difficult to handle by traditional methods.

[0051] Detection scenario: A train is detected at a speed of 300km / h (about 83.3m / s), passing through a region with track irregularities (such as welds or settlement areas), causing non-periodic vertical high-frequency vibration of the vehicle body.

[0052] System configuration: On the base rigidly connected to the TDI linear array camera 101, an Xsens MTi series 6-axis inertial measurement unit 103 (IMU) is installed, with its Y-axis (vertical) precisely aligned with the pixel arrangement direction of the camera chip. The sampling frequency of the IMU is set to 10kHz, much higher than the vibration frequency.

[0053] Control and processing process:

[0054] Multi-sensor data synchronous acquisition: When the train passes through the area, the IMU collects real-time vertical vibration acceleration data. At the same time, the speed sensor 104 provides a stable pulse signal for the intelligent control module 200 to calculate the camera's line frequency, ensuring that the TDI charge transfer is synchronized with the train's forward speed to suppress uniform motion blur.

[0055] Dynamic blur modeling: The intelligent control module 200 processes the collected acceleration signals in real time. By integrating twice with respect to time t, the vertical displacement in the integration period of each line scan is calculated, which is converted into pixel unit offset. For example, it is calculated that the whole camera is offset by 0.8 pixels upward in the integration period of the 1000th line image; the 1001th line is offset by 0.3 pixels downward.

[0056] Generating dynamic PSF: Based on the series of line-by-line varying pixel offsets, the module dynamically generates a corresponding point spread function (PSF) for each line (or a small group of lines). This PSF is no longer a single Gaussian blur kernel, but an asymmetric function that describes the direction and length of the trailing of the line image.

[0057] Line-by-line deconvolution restoration: After the data processing module 300 receives the original image containing vibration blur, it starts the accelerated GPU-based Lucy-Richardson deconvolution algorithm. Instead of using the same blur kernel for the entire image, the algorithm uses a "sliding window" approach to call the dynamic PSF corresponding to each line or block generated by the intelligent control module, and performs fine image restoration.

[0058] Technical effect: The original image shows that the edge of the insulator string on the suspension device presents a wavy and irregular ghosting, and the hairline crack on it is completely covered by the blur. After the processing of the method of the embodiment, the profile of the insulator restores to a straight and sharp state, the blur is effectively "reversed", and the crack defect with a width of only a few pixels is clearly visible. This proves that the present application can quantify and correct random vibration, greatly improving the detection ability of micro defects.

[0059] Example 3: High-precision defect identification in a comprehensive scene

[0060] This embodiment aims to demonstrate the overall performance of the present application in complex and comprehensive working conditions, in which the technical modules work cooperatively to achieve high-precision defect identification.

[0061] Detection scene: night (ambient light <1 Lux), detect the train at a speed of 360 km / h (100 m / s) passing through a river-crossing bridge. The bridge itself has a small vibration of its own frequency, and the train may also have a slight lateral sway due to wind load.

[0062] System cooperative workflow:

[0063] Exposure and synchronization: The intelligent control module 200 raises the TDI integration level to the highest 256 levels to rely entirely on the on-board continuous linear light source 102 for imaging, and deals with the night light environment. At the same time, according to the speed sensor 104 signal, the TDI line array camera 101 row frequency is accurately locked at 70kHz to match the vehicle speed of 100m / s, and the uniform motion blur in the forward direction is perfectly suppressed.

[0064] Multi-dimensional vibration perception and modeling: The IMU simultaneously detects vertical vibration and lateral sway data from the bridge. The intelligent control module 200 fuses the data in these two dimensions in real time to generate a two-dimensional, tilted PSF for each row of images, which accurately describes the blur pattern caused by the vertical and lateral combined motion at that moment.

[0065] Image restoration and stitching: The data processing module 300 first uses a dynamic deconvolution algorithm similar to that of Example 2 to restore the image using a two-dimensional PSF to eliminate all vibration and sway blur. Then, the restored row-by-row clear data is seamlessly stitched into a continuous, high-quality catenary two-dimensional image.

[0066] Two-stage intelligent identification: The high-quality image is sent to the defect identification model. In the first stage, the lightweight YOLOv7-tiny network slides on the image at an extremely high speed (such as 200FPS) to quickly frame all the positions of the "suspension device". In the second stage, for each framed region (ROI), a ResNet-50 classification network based on pre-training on a large number of high-quality images is called to perform fine classification to determine whether it is one of the 12 preset states such as "normal", "bolt marker line misalignment", "split pin missing", or "positioning tube wear exceeds limit".

[0067] Technical effects and data support: On a test set containing 10,000 defect samples collected in the night bridge area (of which 500 are "bolt marker line misalignment 2mm" and "split pin missing" and other minor defects), the end-to-end detection performance of the system of the present application is as follows: the overall accuracy reaches 99.6%, and the recall rate reaches 99.1%. In particular, for "split pin missing", a tiny target that is easily submerged in blur, the recognition accuracy is as high as 99.5%, and the false negative rate is less than 0.5%. As a comparison, if the IMU and dynamic restoration functions of the present application are turned off, and only the TDI line array camera 101 is used to suppress uniform blur, the overall accuracy of the model decreases to 92%, and the false negative rate of "split pin missing" increases to more than 30%. This fully proves that the software and hardware cooperative compensation scheme proposed in the present application plays an irreplaceable key role in ensuring the accuracy and reliability of detection in extreme working conditions.

[0068] Example 4: Adaptive secondary fine imaging based on preliminary diagnosis feedback

[0069] The embodiment aims to disclose a more advanced system configuration and method with closed-loop feedback and adaptive detailed inspection capability to solve the problem of confirming and fine evidence collection of suspicious defects in high-speed inspection, and significantly reduce the defect false alarm rate.

[0070] Technical challenge: In conventional high-speed scanning, some defects (such as hidden cracks of insulators, small loose mark line dislocations of bolts, early fatigue texture changes of components, etc.) may appear as “suspected” or “low confidence” targets in one imaging. If it is directly reported as a defect, it may produce a large number of false alarms and increase the burden of manual review; if ignored, it may miss the real early hidden danger.

[0071] System upgrade configuration:

[0072] Dual-camera system: On the basis of the original image acquisition module 100, a secondary imaging unit is added downstream (for example, 1 meter behind) along the train running direction. The unit is composed of a high-resolution area array camera (for example, 25 million pixels) and a high-intensity, programmable point / surface light source (for example, LED array light source). The original TDI line array camera 101 system serves as the “wide-area scanning subsystem”, and the newly added area array camera system serves as the “focus detailed inspection subsystem”.

[0073] Low-latency processing unit: In the data processing module 300, a lightweight neural network is deployed for real-time preliminary diagnosis (Real-time Preliminary Diagnosis, RPD). The network is specially trained, and its primary goal is not to accurately classify defects, but to quickly identify “high suspicion regions” (Region of High Suspicion, ROHS) from the TDI line array camera image stream in extremely low latency (for example, within 5 milliseconds).

[0074] Adaptive secondary fine imaging workflow:

[0075] 1) Wide-area scanning and real-time preliminary screening: The train is detected at a high speed of 360 km / h (100 m / s). The TDI line array camera 101 serves as the “wide-area scanning subsystem” to continuously acquire catenary images. The acquired image data is streamed to the RPD network in real time.

[0076] 2) Triggering and positioning: Assuming that the RPD network identifies a ROHS of a suspected insulator crack at a certain time. The system immediately records the accurate line mileage position of the ROHS.

[0077] 3) Prediction and Preparation: According to the current vehicle speed of 100 m / s and the camera interval of 1 meter, the intelligent control module 200 accurately calculates that the ROHS will be directly below the "focus inspection subsystem" in seconds.

[0078] 4) Adaptive parameter optimization and execution: In this short 10 ms preparation time, the intelligent control module 200 dynamically generates and issues the optimal imaging instructions for the focus inspection subsystem according to the suspected defect type output by the RPD network. For example:

[0079] For suspected "cracks": the control instruction sets the illumination mode of the LED array light source to "low angle, high intensity grazing light", which is the most likely to produce obvious shadows on the cracks, thereby improving the contrast. At the same time, the exposure time, gain and other parameters of the area array camera are also set to the most suitable mode for capturing crack details.

[0080] For suspected "loose bolts": the control instruction sets the light source to "high angle, diffuse reflection illumination" to eliminate metal reflection and most clearly capture the anti-loose marker line on the bolt.

[0081] For suspected "wear": the control instruction may trigger the area array camera to perform "high-speed burst mode" to continuously capture 3-5 frames of images within a few milliseconds, providing data for subsequent three-dimensional estimation of wear degree through multi-view geometry.

[0082] 5) Fine imaging and data fusion: At the moment, the focus inspection subsystem accurately performs shooting according to the optimized parameters and obtains one or more "close-up" images of the ROHS with extremely high resolution and signal-to-noise ratio.

[0083] Final diagnosis: The data processing module 300 performs data fusion on the wide-area scanning image and the fine inspection image. For example, the inspection image is used as a basis to confirm or exclude suspected targets in the wide-area scanning image. Finally, only the defects confirmed by the inspection image are recorded in the final detection report.

[0084] Technical effect: Through this "scanning + inspection" closed-loop adaptive mechanism, the system can improve the overall defect detection confidence by an order of magnitude. In a field test, the RPD network identified 1200 suspected defect points, and if all were reported, the false positive rate would be as high as 70% after manual verification. After the secondary imaging confirmation of this embodiment, the system finally only reported 380 defect points, and the accuracy rate reached 99.8% after verification. This method greatly improves the reliability of detection without sacrificing the speed of inspection, and significantly reduces the workload of manual review in the back end.

[0085] The above merely describes preferred embodiments of the present application, but is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A line array camera-based intelligent detection system for railway catenary suspension devices, characterized in that, The application relates to a railway overhead line system (OHS) suspension device defect detection system, which comprises the following components: an image acquisition module (100) mounted on a detection vehicle for acquiring image data of the OHS, the image acquisition module (100) comprising: a TDI line array camera (101) with a lens optical axis pointing towards the OHS suspension device to be detected; a continuous linear light source (102) providing continuous and uniform illumination for the scanning field of view of the TDI line array camera (101); a speed sensor (104) for acquiring real-time running speed of the detection vehicle; and an inertial measurement unit (103) for detecting high-frequency vibration data of the detection vehicle in a direction perpendicular to the running direction; an intelligent control module (200) electrically connected to the image acquisition module (100) for synchronously controlling the line scanning frequency of the TDI line array camera (101) according to the real-time running speed and dynamically adjusting the integration order of the TDI line array camera (101) according to illumination changes of the detection environment; meanwhile, the module is also used for generating an image restoration operator in real time according to the high-frequency vibration data; the intelligent control module is configured to: perform twice integration of vibration acceleration data collected by the inertial measurement unit with respect to time, calculate the vertical displacement of the TDI line array camera in the integration period of each line scanning, convert the vertical displacement into a pixel unit offset, and dynamically generate a corresponding dynamic point spread function as the image restoration operator for each line image or line group based on the pixel unit offset; a data processing module (300) for receiving image data collected by the TDI line array camera (101), dynamically restoring the image data by using the image restoration operator, and processing the restored image by using a defect identification model to identify defects of the OHS suspension device, wherein the dynamic restoration processing is performed in a sliding window manner, a dynamic point spread function corresponding to each line or block is called for inverse convolution operation to correct nonlinear motion blur caused by vibration, and the restored image is processed by using a defect identification model to identify defects of the OHS suspension device, the image restoration operator is a dynamic point spread function, and the intelligent control module (200) calculates parameters of the function in real time according to the amplitude and frequency of the high-frequency vibration data, the dynamic restoration processing of the data processing module (300) is a line-by-line inverse convolution operation based on the Wiener filtering or Lucy-Richardson algorithm, and the dynamic point spread function is used for blur correction of each line or block of collected image data.

2. The line array camera-based intelligent detection system for railway overhead line suspension devices according to claim 1, characterized in that, The intelligent control module (200) is configured to: when the ambient light decreases, preferentially increase the integration order of the TDI line array camera (101) to increase the exposure amount, and when the integration order reaches an upper limit, then increase the output power of the continuous linear light source (102).

3. The line array camera-based intelligent detection system for railway overhead line suspension devices according to claim 1, characterized in that, The sampling axis of the inertial measurement unit (103) is parallel or at a preset angle to the pixel arrangement direction of the TDI line array camera (101).

4. The line array camera-based intelligent detection system for railway overhead line suspension devices according to claim 1, characterized in that, The defect identification model is a two-stage deep learning model, the first stage adopts a target detection network to identify and locate the suspension device in the image, and the second stage adopts an image classification network to classify the types of defects of the located suspension devices.

5. A method for detecting the suspension device of railway overhead line system based on linear array camera using the intelligent detection system according to any one of claims 1-4, characterized in that, The method comprises the following steps: A multi-source data synchronous acquisition step: when the vehicle is running, the vehicle speed is acquired in real time through a speed sensor (104), and high-frequency vibration data of the vehicle in a direction perpendicular to the running direction are acquired in real time through an inertial measurement unit (103); An adaptive exposure and synchronous scanning step: the line scanning frequency of a TDI linear array camera (101) is controlled synchronously according to the vehicle speed, and the integration level of the TDI linear array camera (101) is dynamically adjusted according to the intensity of environmental light, so that image acquisition is performed under the illumination of a continuous linear light source (102); A dynamic blur modeling step: an image restoration operator for representing nonlinear motion blur is generated in real time according to the high-frequency vibration data; An image restoration and defect identification step: the collected image data is dynamically restored by using the image restoration operator, and then the restored image is input into a defect identification model to identify the defects of the overhead contact line suspension device.

6. The intelligent detection method for railway catenary suspension device according to claim 5, characterized in that, In the adaptive exposure and synchronous scanning step, the dynamic adjustment of the integration level of the TDI linear array camera (101) comprises: presetting a target image gray scale range, when the average gray scale value of the real-time collected image is lower than the range, gradually increasing the integration level until the gray scale value meets the standard or the integration level reaches the maximum value of the camera.

7. The intelligent detection method for railway catenary suspension device according to claim 5, characterized in that, The dynamic blur modeling step comprises: converting the high-frequency vibration data into pixel displacement of the TDI linear array camera (101) in the integration period through a preset transfer function model, and constructing a point spread function describing the displacement as an image restoration operator.

8. The intelligent detection method for railway catenary suspension device according to claim 5, characterized in that, After the image restoration and defect identification step, a report generation step is further included: combining the identified defect type with the real-time acquired geographic location information or line mileage information to generate a detection report containing the defect type and accurate position.

Images