A pressure gauge identification method and device in an industrial valve pressure test process

Abstract

The present application belongs to the field of instrument identification, and particularly relates to a pressure gauge identification method and device in the process of industrial valve pressure testing. The method extracts instrument images from the collected video stream, performs distortion correction and illumination normalization processing; then performs feature extraction on the processed images, identifies the instrument type based on the training model of multiple instrument types, and corrects the angle-deformed instrument images through a spatial transformation network; then performs spatial sorting on the identified instrument images based on the center point coordinates of the detection frame, generates a data array containing instrument information; finally, performs enhancement processing on the single instrument image in the data array, locates the dial area and performs perspective correction, determines the geometric parameters of the dial, calculates the original reading based on the parameters, and outputs the final reading combined with the tilt compensation. Through the above image processing and identification steps, the accurate identification and reading of the pressure gauge in the process of industrial valve pressure testing are realized, and the identification efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the field of instrument identification, and specifically relates to a method and device for identifying pressure gauges during industrial valve pressure testing. Background Technology

[0002] As a core energy source for modern industry, the safety of oil and natural gas transportation systems directly impacts national economic development and public safety. With the development of oil pipelines towards larger diameters, higher pressures, and thicker walls on the seabed, the accuracy and reliability of pressure performance testing for oil and gas valves, as key control components of pipeline systems, have become crucial for ensuring pipeline safety. In industrial valve pressure testing, pressure gauges, as core monitoring instruments, are critical for determining valve sealing performance and pressure resistance through real-time acquisition and accurate identification of their readings. Traditional manual reading methods are not only inefficient but also susceptible to human error and environmental interference, failing to meet the real-time and accuracy requirements of modern, intelligent manufacturing supervision. Therefore, automated pressure gauge identification technology is urgently needed to support the intelligent upgrade of valve pressure testing.

[0003] For automated identification technology of industrial instruments, relevant research and applications already exist in the existing technology. For example, patent application CN119296110A discloses an automatic identification method for industrial pointer instrument readings based on deep learning. This method uses an industrial camera to acquire images of pointer instruments, uses the YOLOX-DC model to detect the instrument panel, and uses the PM-SwinUnet model to segment the pointer and scale. Finally, it calculates the reading based on the angle method, providing a deep learning solution for automated instrument identification. In addition, patent application CN120260022A discloses an artificial intelligence identification method for instrument panels based on edge computing. It uses the YOLO object detection model to train and identify the instrument panel area, combines a ResNet18 neural network to segment the pointer and scale, and uses a spatial transformation network to obtain the instrument panel image in the front view to optimize the recognition effect, further promoting the intelligent development of instrument identification in industrial scenarios.

[0004] However, existing technologies still have significant limitations in pressure gauge identification in the specific scenario of industrial valve pressure testing: First, although patent application CN119296110A achieves automatic reading of pointer instruments, it does not involve simultaneous acquisition by multiple cameras and spatial positioning and sorting of images, making it difficult to adapt to the scenario of simultaneous monitoring of multiple pressure gauges in valve testing and failing to meet the requirement of correlation between test data and instrument position; Second, although patent application CN120260022A optimizes the dial image through spatial transformation, it does not incorporate targeted processing such as super-resolution noise reduction and tilt compensation in the dial reading calculation, making the identification accuracy easily affected by complex lighting conditions and instrument angle deviations at the valve testing site; Third, existing technologies do not design image preprocessing schemes for special environments such as high-frequency vibration and oil pollution interference in valve pressure testing, resulting in insufficient image quality optimization in actual testing scenarios, which in turn affects the accuracy of subsequent feature extraction and reading calculation. Therefore, existing technologies are difficult to directly apply to industrial valve pressure testing scenarios, and there is an urgent need for a pressure gauge identification method and equipment that can adapt to simultaneous monitoring of multiple gauges, has strong anti-interference capabilities, and high identification accuracy. Summary of the Invention

[0005] To address the aforementioned problems, in a first aspect, this invention proposes a method for identifying pressure gauges during industrial valve pressure testing, comprising the following steps:

[0006] The instrument images are extracted from the acquired video stream, and the extracted instrument images are subjected to distortion correction and illumination normalization.

[0007] Feature extraction is performed on the instrument images after illumination normalization, the instrument type is identified based on the training model of multiple types of instruments, and the angular deformation of the instrument images is corrected through a spatial transformation network.

[0008] Based on the coordinates of the center point of the detection frame, the identified instrument images are spatially sorted to generate a data array containing instrument information;

[0009] The individual instrument images in the data array are enhanced, the dial area in the instrument image is located and perspective correction is performed to determine the geometric parameters of the dial, the original reading of the dial is determined based on the geometric parameters of the dial, and the final reading is output in combination with tilt compensation.

[0010] Furthermore, the step of extracting instrument images from the acquired video stream and performing distortion correction and illumination normalization on the extracted images includes the following steps:

[0011] Hardware-level time calibration is performed on the camera's acquisition time;

[0012] Instrument images are extracted from the acquired video stream at preset time intervals;

[0013] Distortion correction is performed on the extracted instrument images based on pre-calibrated camera intrinsic parameters;

[0014] Local contrast enhancement and overall light compensation are performed on the instrument image after distortion correction.

[0015] Furthermore, the step of extracting features from the instrument image after illumination normalization, identifying the instrument type based on a training model for multiple instrument types, and correcting the angularly distorted instrument image through a spatial transformation network includes the following steps:

[0016] Feature extraction is performed on the instrument image after illumination normalization.

[0017] Acquire scene image data for different types of industrial instruments;

[0018] The recognition model is trained based on the scene image data.

[0019] The trained recognition model is used to identify instrument images and automatically corrects angularly distorted instrument images.

[0020] Furthermore, the step of spatially sorting the identified instrument images based on the coordinates of the center point of the detection frame to generate a data array containing instrument information includes the following steps:

[0021] Based on the coordinates of the center point of the instrument detection frame, an ordered list is generated in ascending order of x-axis coordinates and y-axis coordinates.

[0022] For instruments in the same column in an ordered list, if the difference in y-coordinate between instruments in the same column is less than a preset coordinate threshold, they are determined to be installed in parallel, and a suffix is ​​added to the number to distinguish them.

[0023] The sorted instrument information is encapsulated into a data array.

[0024] Furthermore, the instrument information includes the instrument's location coordinates, type, name, and ID.

[0025] Further, the enhancement processing of a single instrument image in the data array, locating the dial area in the instrument image and performing perspective correction to determine the dial's geometric parameters, determining the original dial reading based on the dial's geometric parameters, and outputting the final reading in conjunction with tilt compensation, includes the following steps:

[0026] Super-resolution reconstruction and noise reduction are performed on the dial area of ​​the instrument image;

[0027] Perform instance segmentation on the denoised dial area and output the mask of the dial elliptical region;

[0028] The ellipse parameters of the mask region are fitted, and the perspective transformation matrix is ​​calculated based on the ellipse parameters to map the elliptical dial to a perfect circle. If there is non-uniform distortion in the edge region of the dial, the four corners of the dial are identified by combining FAST corner detection, and the edge region of the dial is subdivided and corrected by mesh subdivision.

[0029] Detect the arc boundary of the dial, and combine gradient change analysis to extract the geometric parameters of the dial based on the pointer outline and the digital edge.

[0030] The original reading of the dial is determined based on the geometric parameters, and the final reading is output after tilt compensation.

[0031] Furthermore, the ellipse parameters include the center, major axis, and minor axis;

[0032] The geometric parameters of the dial include the center, the tip of the pointer, and the starting scale.

[0033] Furthermore, determining the original reading of the dial based on its geometric parameters and outputting the final reading in conjunction with tilt compensation includes the following steps:

[0034] The angle of the pointer tip relative to the center of the circle in the current dial is determined based on the coordinates of the center and the pointer tip.

[0035] Based on the angular relationship between the pointer tip and the starting scale, the original reading of the dial is calculated using the range mapping formula;

[0036] If the dial tilt angle is greater than the set tilt angle, the original dial reading will be compensated for tilt, and the final reading will be output after compensation; if it is less than or equal to the set tilt angle, the calculated original dial reading will be used as the final reading.

[0037] Furthermore, the range mapping formula is as follows:

[0038]

[0039] Where angle is the angle of the pointer tip relative to the center of the circle in the current dial, angle_E is the angle corresponding to the pointer tip of the end scale; angle_S is the angle corresponding to the pointer tip of the starting scale; and Max and Min are the maximum and minimum scales of the dial, respectively.

[0040] The formula for tilt compensation is:

[0041]

[0042] in, This is the original reading on the dial. The final reading is after tilt compensation, where θ is the dial tilt angle.

[0043] Secondly, this invention proposes a pressure gauge identification system for industrial valve pressure testing, comprising:

[0044] The preprocessing module is used to extract instrument images from the acquired video stream and perform distortion correction and illumination normalization on the extracted instrument images.

[0045] The instrument recognition and extraction module is used to extract features from the instrument images after illumination normalization, identify the instrument type based on the training model of multiple types of instruments, and correct the angular deformation of the instrument images through the spatial transformation network.

[0046] The image localization and sorting module is used to spatially sort the identified instrument images based on the coordinates of the center point of the detection box, and generate a data array containing instrument information.

[0047] The dial reading calculation module is used to enhance a single instrument image in the data array, locate the dial area in the instrument image and determine the geometric parameters of the dial after perspective correction, determine the original reading of the dial based on the geometric parameters of the dial, and output the final reading in combination with tilt compensation.

[0048] Thirdly, the present invention proposes a pressure gauge identification device in the process of industrial valve pressure testing, including a server, a camera, an edge computing box, a network transmission device, and a mobile device;

[0049] The server is used to receive identification tasks and, according to the identification tasks, schedule the camera of the instrument to be identified to collect the video stream information of the instrument panel.

[0050] A camera is used to capture video streams from the dashboard.

[0051] An edge computing box, used to perform the aforementioned recognition method or configured with the aforementioned recognition system;

[0052] Mobile devices are used to view the recognition results and related test data of the edge computing boxes;

[0053] Network transmission equipment is used to provide network connectivity for cameras, edge computing boxes, and mobile devices.

[0054] The beneficial effects of this invention are:

[0055] This invention utilizes extensive machine learning training to cover instrument samples under various conditions, including multiple types, perspectives, and lighting. Combined with built-in feature extraction and angle correction optimization algorithms, it significantly improves the accuracy of image recognition. It can stably identify instruments with different angle deformations, complex lighting (too bright, too dark, backlight), and various types of industrial instruments, solving the problems of poor adaptability and high false detection rate of traditional recognition methods in complex scenarios.

[0056] This invention optimizes the dial area in the identified instrument image through multiple steps, including instrument positioning, over-resolution noise reduction, dial ellipse positioning, perspective correction, precise positioning of scale and pointer, and angle compensation. This effectively eliminates the influence of factors such as image distortion, lighting interference, and viewing angle deviation on the readings, resulting in extremely small errors in the final output instrument values, thus providing reliable data support for valve pressure testing.

[0057] This invention acquires instrument video streams in real time through multiple cameras, extracts key frames at fixed intervals and continuously monitors instrument readings. Combined with edge computing boxes for real-time data processing and storage, it can summarize and analyze data throughout the entire test cycle and generate fluctuation charts. This replaces the inefficient traditional manual reading mode, realizes automated and continuous monitoring of the stress test process, and significantly improves test efficiency and data integrity.

[0058] This invention constructs an independent system based on an edge computing box, and establishes a local area network through an industrial-grade wireless router. Each instrument panel, camera, and edge computing box can be flexibly combined into an independent unit, and it supports multi-device expansion and protocol adaptation such as GB / T28181. It can quickly adapt to valve testing scenarios of different scales (such as single-unit testing and multi-unit parallel testing). At the same time, the edge computing box's wide temperature design (-20℃ to +60℃) and IP40 protection level ensure the stable operation of the system in complex industrial environments and reduce deployment and maintenance costs.

[0059] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart of the pressure gauge identification method during industrial valve pressure testing in this embodiment of the invention;

[0062] Figure 2 It shows Figure 1 Detailed flowchart of step S1;

[0063] Figure 3 It shows Figure 1 Detailed flowchart of step S2;

[0064] Figure 4 It shows Figure 1 Detailed flowchart of step S3;

[0065] Figure 5 A detailed flowchart of step S4 of the present invention is shown;

[0066] Figure 6 A schematic diagram of a pressure gauge identification device during industrial valve pressure testing in an embodiment of the present invention is shown.

[0067] Figure 7 A flowchart illustrating the execution process of the pressure gauge identification device during industrial valve pressure testing in an embodiment of the present invention is shown. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] This invention proposes a pressure gauge identification method during industrial valve pressure testing. Through collaborative processing of video preprocessing, instrument identification and extraction, image positioning and sorting, and dial reading calculation, it achieves automated and high-precision identification of pressure gauges during industrial valve pressure testing. Figure 1 As shown, the following provides a detailed explanation of each step:

[0070] S1. Extract instrument images from the acquired video stream, and perform distortion correction and illumination normalization processing on the extracted instrument images.

[0071] S2. Extract features from the instrument image after illumination normalization, identify the instrument type based on the training model of multiple types of instruments, and correct the angularly deformed instrument image through a spatial transformation network.

[0072] S3. Based on the coordinates of the center point of the detection frame, the identified instrument images are spatially sorted to generate a data array containing instrument information;

[0073] S4. Enhance the individual instrument image in the data array, locate the dial area in the instrument image and perform perspective correction to determine the geometric parameters of the dial, determine the original reading of the dial based on the geometric parameters of the dial, and output the final reading in combination with tilt compensation.

[0074] In one embodiment of the present invention, the video preprocessing in step S1 is fundamental to ensuring the accuracy of subsequent recognition. Through synchronous correction and image optimization, high-quality image input is provided for instrument recognition, such as... Figure 2 As shown, the specific steps include:

[0075] S11. Hardware-level time calibration is performed using PTP (Precision Time Protocol) to ensure that the acquisition time difference of multiple industrial-grade cameras is controlled within milliseconds (≤1ms). This synchronization mechanism can avoid data association errors caused by the acquisition time offset of different cameras, and is especially suitable for scenarios where multiple pressure gauges are monitored synchronously in valve pressure testing.

[0076] S12. Extract keyframes from the raw video stream transmitted from the camera at fixed time intervals. In this embodiment, the frame extraction interval is set to 250ms / frame (i.e., 4 frames are extracted per second). The extracted keyframes are temporarily stored in the GPU shared memory queue of the edge computing box, which reduces the amount of data transmission while preserving the dynamic changes during the stress test and ensuring data integrity.

[0077] S13. Based on the pre-calibrated camera intrinsics, the extracted image is distorted using the fisheye.undistort function (fisheye lens distortion correction) of a computer vision library (such as OpenCV, Open Source Computer Vision Library).

[0078] Specifically, the camera intrinsic parameters include the camera matrix K and the distortion coefficient D. For example, for industrial cameras with a lens focal length of 6-12mm, the distortion of the dial edges caused by the fisheye effect can be eliminated by this algorithm, restoring the dial outline to its true geometric shape.

[0079] S14. The CLAHE (Contrast Limitation Adaptive Histogram Equalization) algorithm is used to enhance local contrast, and the Retinex-MSR (Multi-Scale Retina Enhancement) algorithm is combined for overall light compensation.

[0080] In one embodiment of the present invention, the CLAHE algorithm enhancement process parameters are set to clipLimit=2.0 and tileGrid Size=8×8, which can highlight the detailed differences between the dial markings and the hands, and avoid feature loss due to local over-darkness. The Retinex-MSR algorithm performs overall light compensation, which can balance the brightness of images under backlight, shadow, or strong light, so that the grayscale distribution of the dial tends to be consistent under different lighting conditions, thereby improving the stability of subsequent recognition.

[0081] In one embodiment of the present invention, the instrument identification and extraction process in step S2 can be based on a deep learning model to achieve accurate classification and angle correction of the instruments, such as... Figure 3 As shown, the specific steps are as follows:

[0082] S21. The YOLOv8+ResNet (You Only Look Once version 8 + residual network) joint architecture is used to extract features from the image after illumination normalization.

[0083] S22. Acquire scene image data of different industrial instruments;

[0084] S23: The recognition model is trained based on the scene image data using a cosine learning rate (cosine lr) combined with an exponential moving average (EMA) strategy;

[0085] S24. The trained recognition model is used to recognize instrument images, and the instrument images with angular deformation are automatically corrected by STN (Spatial Transformer Network); angular deformation includes dial deformation caused by yaw angle and pitch angle.

[0086] In one embodiment of the present invention, step S21 is specifically performed as follows:

[0087] The input image is uniformly adjusted to 640×640 pixels, and after normalization (pixel values ​​are scaled to the range of 0~1) and mean-variance standardization (based on the mean μ=0.485 and variance σ=0.229 of the training set statistics), it enters the feature extraction network.

[0088] The initial features are extracted through a Conv+BN+SiLU (convolution + normalization + Swish function) combined layer, outputting a 256-channel feature map. This feature map is divided into two paths: one path is directly passed through the ResNet residual structure, and the other path is processed by a 3×3 small convolution kernel and then fused with the residual path, finally outputting feature maps of three scales: 80×80, 40×40, and 20×20, to adapt to instrument targets of different sizes.

[0089] In one embodiment of the present invention, the types of industrial instruments in step S22 include, but are not limited to, 12 types of industrial instruments such as pressure gauges, temperature gauges, differential pressure gauges, level gauges, and flow meters, with a sample size of ≥10,000 images for each type of instrument (including scenes with different brands, ranges, stains, lighting, and angles).

[0090] In step S23, the training endpoint of the recognition model can be mAP@0.5≥0.95 on the validation set, with a false detection rate ≤1%, which can stably recognize various instruments and output category labels.

[0091] In step S24, STN learns a 3×3 affine transformation matrix to rotate, scale, and translate the detected instrument area, turning the tilted dial into a front view and eliminating the influence of angular deviation on subsequent reading calculations.

[0092] In one embodiment of the present invention, step S3, to achieve the correspondence between multi-table data and physical locations, requires spatial positioning and sorting of the identified instruments, such as... Figure 4 As shown, the specific steps include:

[0093] S31. Using the center point coordinates of the instrument detection frame as the sorting basis, first sort by the x-axis coordinate in ascending order (from left to right), then sort by the y-axis coordinate in ascending order (from top to bottom), to generate an ordered list. For example, for three instruments with center point coordinates of (320,240), (640,240), and (320,480), the sorting result is (320,240)→(320,480)→(640,240).

[0094] S32. For instruments in the same column in an ordered list, if the difference in y-coordinate between instruments in the same column is less than a preset coordinate threshold, it is determined to be installed in parallel, and a suffix is ​​added after the number to distinguish them (e.g., "Table 1", "Table 1-1", "Table 1-2"). For example, the preset coordinate threshold is 20px.

[0095] S33. Encapsulate the sorted instrument information into a JSON data array. The instrument information includes, but is not limited to, the instrument's location coordinates, type, name, and ID, as shown in the example below:

[0096] json[{"ID":"M001","type":"Pressure gauge","x":320,"y":240,"number":"1"},{"ID":"M002","type":"Pressure gauge","x":325,"y":255,"number":"1-1"}, {"ID":"M003","type":"Thermometer","x":640,"y":240,"number":"2"}).

[0097] In one embodiment of the present invention, in step S4, the final reading is output through image enhancement, region localization, parameter extraction, and compensation calculation, as shown below. Figure 5 As shown, the specific steps include:

[0098] S41. The Real-ESRGAN-x2 (Real Scene Enhanced Super-Resolution Generative Adversarial Network (2x Scaled Version)) model is used to perform super-resolution reconstruction of the dial area of ​​the instrument image and combined with the BM3D (Block Matching 3D Filtering) algorithm for noise reduction.

[0099] The super-resolution reconstruction process can enlarge the image size by 2 times, improving the detail clarity of the scale and pointer; after noise reduction, the image PSNR (peak signal-to-noise ratio) can be improved by ≈3dB, effectively reducing misidentification caused by oil stains and scratches.

[0100] S42. Mask R-CNN (mask-based region convolutional neural network) is used to segment the denoised dial region and output the mask of the dial elliptical region.

[0101] S43. Fit the ellipse parameters of the ellipse mask region using the fitEllipse function (a function for fitting ellipses in OpenCV), calculate the perspective transformation matrix based on the ellipse parameters, and map the elliptical dial to a perfect circle; the ellipse parameters are the center, major axis, and minor axis.

[0102] S44: If there is non-uniform distortion in the edge area of ​​the dial, the four corners of the dial are identified by combining FAST (accelerated segmentation test feature, a fast corner detection method), and the edge area is subdivided and corrected by mesh subdivision; this step can eliminate non-uniform perspective distortion (such as edge scale stretching).

[0103] S45: Detect the arc boundary of the dial using Hough transform, and extract the geometric parameters of the dial by combining gradient change analysis to locate the pointer profile and digital edge.

[0104] The geometric parameters include the center of the circle, the tip of the pointer, and the starting scale.

[0105] Center point O(x0,y0): The geometric center of the dial;

[0106] Pointer tip P(xp,yp): The endpoint of the pointer furthest from the center of the circle;

[0107] Starting scale S(xs,ys): The scale point corresponding to the minimum range of the dial.

[0108] S46: Determine the original reading of the dial based on the geometric parameters and output the final reading in combination with tilt compensation; specifically including the following steps:

[0109] S461: Determine the angle between the pointer tip and the center of the circle in the current dial based on the coordinates of the center and the pointer tip.

[0110] S462: Based on the angular relationship between the pointer tip and the starting scale, the original reading of the dial is calculated using the range mapping formula;

[0111] S463: If the dial tilt angle is greater than the set tilt angle, the original dial reading will be compensated for tilt and the final reading will be output. If it is less than or equal to the set tilt angle, the calculated original dial reading will be used as the final reading.

[0112] In one embodiment of the present invention, the formula for calculating the angle between the pointer tip and the center of the circle is as follows:

[0113] angle=atan2(yp-y0,xp-x0)-atan2(ys-y0,xs-x0);

[0114] The range mapping formula is as follows:

[0115]

[0116] Where angle is the angle of the current pointer tip relative to the center of the circle, angle_E is the angle corresponding to the end pointer tip, that is, the angle between the ray from the center of the circle to the end pointer point (corresponding to the maximum range Max) and the reference axis (usually the positive x-axis direction); angle_S is the angle corresponding to the starting pointer tip, that is, the angle between the ray from the center of the circle to the starting point (xs, ys) and the reference axis (usually the positive x-axis direction);

[0117] The formulas for calculating angle_E and angle_S are:

[0118] angle_E = atan2(ye-y0,xe-x0) (where xe and ye are the x and y coordinates of the tip of the pointer at the end of the scale);

[0119] angle_S=atan2(ys-y0,xs-x0) (where xs and ys are the horizontal and vertical coordinates of the starting point pointer).

[0120] In one embodiment of the present invention, the tilt angle is set to 5°. That is, if the camera tilt causes the angle θ between the dial plane normal vector and the optical axis to be greater than 5°, the reading is corrected by the following formula to compensate for the tilt and eliminate the measurement error caused by the viewing angle deviation. The tilt compensation formula is as follows:

[0121]

[0122] in, This is the original reading on the dial. The final reading after tilt compensation is θ, which is the dial tilt angle, i.e., the angle between the dial plane normal vector and the camera optical axis.

[0123] Based on the same inventive concept, another embodiment of the present invention proposes a pressure gauge identification system for industrial valve pressure testing, comprising:

[0124] The preprocessing module is used to extract instrument images from the acquired video stream and perform distortion correction and illumination normalization on the extracted instrument images.

[0125] The instrument recognition and extraction module is used to extract features from the instrument images after illumination normalization, identify the instrument type based on the training model of multiple types of instruments, and correct the angular deformation of the instrument images through the spatial transformation network.

[0126] The image localization and sorting module is used to spatially sort the identified instrument images based on the coordinates of the center point of the detection box, and generate a data array containing instrument information.

[0127] The dial reading calculation module is used to enhance a single instrument image in the data array, locate the dial area in the instrument image and determine the geometric parameters of the dial after perspective correction, determine the original reading of the dial based on the geometric parameters of the dial, and output the final reading in combination with tilt compensation.

[0128] Another embodiment of the present invention proposes a pressure gauge identification device in the process of industrial valve pressure testing, including hardware combinations and a data transfer mechanism between the hardware combinations. Figure 6 As shown, the hardware assembly includes a camera, an edge computing box, a network transmission device, and a mobile device, wherein...

[0129] A camera is used to capture video streams from the dashboard.

[0130] The server is used to receive identification tasks and, according to the identification tasks, schedule the camera of the instrument to be identified to collect the video stream information of the instrument panel.

[0131] The edge computing box is equipped with a pressure gauge recognition system for the industrial valve pressure test process to identify the instrument panel video stream information scheduled by the server; specifically, it includes a preprocessing module, an instrument recognition and extraction module, an image positioning and sorting module, and a dial reading calculation module.

[0132] Mobile devices are used to view the recognition results and related test data of the edge computing boxes;

[0133] Network transmission devices are used to provide network connectivity for servers, cameras, edge computing boxes, and mobile devices. Examples of network transmission devices include routers and corporate intranets.

[0134] In one embodiment of the present invention, the camera is an industrial camera with a 5-megapixel global shutter, which can clearly capture dial details; it supports HDR function to adapt to complex lighting environments; it uses Ethernet power supply to achieve integrated power supply and data transmission; the lens focal length is manually adjustable from 6 to 12 mm, which is suitable for shooting distances of 0.8 to 1.5 m.

[0135] Installation method: Secure it to the tripod using magnetic attraction or clamps. Place the tripod 0.8-1.5m in front of the instrument panel, ensuring that the pitch angle is ≤15° and the yaw angle is ≤30° to reduce dial deformation.

[0136] In one embodiment of the present invention, the specific information of the edge computing box is as follows:

[0137] Core configuration: It adopts a TPU graphics processing chip, supports 16-channel 1080P video monitoring, 38-channel 1080P video decoding and 2-channel encoding; it is compatible with GB / T28181 national standard protocol and 300W / 500W / 800W / 1200W high-definition cameras.

[0138] Environmental adaptability: Operating temperature -20℃ to +60℃, supports 12V DC power supply, power consumption ≤25W; equipped with IP40 protection rating and fanless heat dissipation design, it can operate stably in dusty and vibrating industrial environments.

[0139] Functions: Performs algorithms such as video preprocessing, instrument recognition and extraction, and dial reading calculation, while also realizing video stream storage and feature learning.

[0140] In one embodiment of the present invention, the network transmission device is a wireless router, and the enterprise intranet uses the router to construct a specific network environment, providing dedicated and stable network connections for internal enterprise devices (such as office computers). Specifically:

[0141] Features: Supports dual gigabit WAN / LAN interfaces for high-speed wired connections; supports 4G / 5G redundant links to ensure automatic switching in case of network interruption; integrates VPN (Virtual Private Network) and firewall functions to ensure data transmission security.

[0142] Function: To build a local area network between the camera and the edge computing box, and to enable wireless data backhaul between the edge computing box and the handheld terminal.

[0143] In one embodiment of the present invention, the mobile device includes a smartphone or tablet. By installing dedicated software, the pressure gauge recognition results, pressure holding time, and pressure value trend graph (such as a line graph) can be viewed in real time, and historical data query and export are supported.

[0144] In one embodiment of the present invention, the actual application process of the identification device is as follows: Figure 7 As shown, the specific process is as follows:

[0145] Task initiation: The tablet acts as the terminal. Enter the valve number, click "Start Task," and send the identification task to the server, specifying which valve to be detected.

[0146] Execution and processing: The server schedules the output of camera images according to the recognition task. The camera collects video streams from valves and associated instrument panels (multi-pointer meters) and transmits them back. The edge computing box, based on the video stream, performs recognition of the workbench ID and instrument readings (in real time and continuously to ensure data timeliness). At the same time, it uses algorithms to maintain the real-time correlation between the recognition results and the actual pointers, and accurately interprets the dial values.

[0147] Output Results: After the identification is completed, the instrument identification results (including the final reading of the instrument, fluctuation spectrum, etc.) are generated and output. The detection process video can also be attached and displayed on terminals such as tablets, which connects the task distribution-data acquisition and processing-result presentation link, and reflects the equipment collaboration and functional value in the business process.

[0148] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying pressure gauges during industrial valve pressure testing, characterized in that, Includes the following steps: The instrument images are extracted from the acquired video stream, and the extracted instrument images are subjected to distortion correction and illumination normalization. Feature extraction is performed on the instrument images after illumination normalization, the instrument type is identified based on the training model of multiple types of instruments, and the angular deformation of the instrument images is corrected through a spatial transformation network. Based on the coordinates of the center point of the detection frame, the identified instrument images are spatially sorted to generate a data array containing instrument information; The process involves enhancing individual instrument images in the data array, locating the dial area within the instrument image, performing perspective correction to determine the dial's geometric parameters, determining the original dial reading based on these geometric parameters, and then combining this with tilt compensation to output the final reading. This includes the following steps: Super-resolution reconstruction and noise reduction are performed on the dial area of ​​the instrument image; Perform instance segmentation on the denoised dial area and output the mask of the dial elliptical region; The ellipse parameters of the mask region are fitted, and the perspective transformation matrix is ​​calculated based on the ellipse parameters to map the elliptical dial to a perfect circle. If there is non-uniform distortion in the edge region of the dial, the four corners of the dial are identified by combining FAST corner detection, and the edge region of the dial is subdivided and corrected by mesh subdivision. Detect the arc boundary of the dial, and combine gradient change analysis to extract the geometric parameters of the dial based on the pointer outline and the digital edge. The original reading of the dial is determined based on the geometric parameters, and the final reading is output after tilt compensation.

2. The pressure gauge identification method during industrial valve pressure testing according to claim 1, characterized in that, The process of extracting instrument images from the acquired video stream and performing distortion correction and illumination normalization on the extracted images includes the following steps: Hardware-level time calibration is performed on the camera's acquisition time; Instrument images are extracted from the acquired video stream at preset time intervals; Distortion correction is performed on the extracted instrument images based on pre-calibrated camera intrinsic parameters; Local contrast enhancement and overall light compensation are performed on the instrument image after distortion correction.

3. The pressure gauge identification method during industrial valve pressure testing according to claim 1, characterized in that, The process of extracting features from the instrument image after illumination normalization, identifying the instrument type based on a training model for multiple instrument types, and correcting angularly distorted instrument images through a spatial transformation network includes the following steps: Feature extraction is performed on the instrument image after illumination normalization. Acquire scene image data for different types of industrial instruments; The recognition model is trained based on the scene image data. The trained recognition model is used to identify instrument images and automatically corrects angularly distorted instrument images.

4. The pressure gauge identification method during industrial valve pressure testing according to claim 1, characterized in that, The step of spatially sorting the identified instrument images based on the coordinates of the center point of the detection frame to generate a data array containing instrument information includes the following steps: Based on the coordinates of the center point of the instrument detection frame, an ordered list is generated in ascending order of x-axis coordinates and y-axis coordinates. For instruments in the same column in an ordered list, if the difference in y-coordinate between instruments in the same column is less than a preset coordinate threshold, they are determined to be installed in parallel, and a suffix is ​​added to the number to distinguish them. The sorted instrument information is encapsulated into a data array.

5. The pressure gauge identification method during industrial valve pressure testing according to claim 4, characterized in that, The instrument information includes the instrument's location coordinates, type, name, and ID.

6. The pressure gauge identification method during industrial valve pressure testing according to claim 1, characterized in that, The ellipse parameters include the center, major axis, and minor axis; The geometric parameters of the dial include the center, the tip of the pointer, and the starting scale.

7. The pressure gauge identification method during industrial valve pressure testing according to claim 6, characterized in that, The process of determining the original reading of the dial based on its geometric parameters and outputting the final reading in conjunction with tilt compensation includes the following steps: The angle of the pointer tip relative to the center of the circle in the current dial is determined based on the coordinates of the center and the pointer tip. Based on the angular relationship between the pointer tip and the starting scale, the original reading of the dial is calculated using the range mapping formula; If the dial tilt angle is greater than the set tilt angle, the original dial reading will be compensated for tilt, and the final reading will be output after compensation; if it is less than or equal to the set tilt angle, the calculated original dial reading will be used as the final reading.

8. The pressure gauge identification method during industrial valve pressure testing according to claim 7, characterized in that, The range mapping formula is as follows: Where angle is the angle of the pointer tip relative to the center of the circle in the current dial, angle_E is the angle corresponding to the pointer tip of the end scale; angle_S is the angle corresponding to the pointer tip of the starting scale; and Max and Min are the maximum and minimum scales of the dial, respectively. The formula for tilt compensation is: in, This is the original reading on the dial. The final reading is after tilt compensation, where θ is the dial tilt angle.

9. A pressure gauge identification system for industrial valve pressure testing, characterized in that, include: The preprocessing module is used to extract instrument images from the acquired video stream and perform distortion correction and illumination normalization on the extracted instrument images. The instrument recognition and extraction module is used to extract features from the instrument images after illumination normalization, identify the instrument type based on the training model of multiple types of instruments, and correct the angular deformation of the instrument images through the spatial transformation network. The image localization and sorting module is used to spatially sort the identified instrument images based on the coordinates of the center point of the detection box, and generate a data array containing instrument information. The dial reading calculation module is used to enhance a single instrument image in the data array, locate the dial area in the instrument image, perform perspective correction to determine the dial's geometric parameters, determine the original dial reading based on the dial's geometric parameters, and output the final reading after combining tilt compensation. The module includes the following steps: Super-resolution reconstruction and noise reduction are performed on the dial area of ​​the instrument image; Perform instance segmentation on the denoised dial area and output the mask of the dial elliptical region; The ellipse parameters of the mask region are fitted, and the perspective transformation matrix is ​​calculated based on the ellipse parameters to map the elliptical dial to a perfect circle. If there is non-uniform distortion in the edge region of the dial, the four corners of the dial are identified by combining FAST corner detection, and the edge region of the dial is subdivided and corrected by mesh subdivision. Detect the arc boundary of the dial, and combine gradient change analysis to extract the geometric parameters of the dial based on the pointer outline and the digital edge. The original reading of the dial is determined based on the geometric parameters, and the final reading is output after tilt compensation.

10. A pressure gauge identification device for industrial valve pressure testing, characterized in that, This includes servers, cameras, edge computing boxes, network transmission equipment, and mobile devices; The server is used to receive identification tasks and, according to the identification tasks, schedule the camera of the instrument to be identified to collect the video stream information of the instrument panel. A camera is used to capture video streams from the dashboard. An edge computing box, used to perform the identification method according to any one of claims 1-8 or configured with the identification system according to claim 9; Mobile devices are used to view the recognition results and related test data of the edge computing boxes; Network transmission equipment is used to provide network connectivity for cameras, edge computing boxes, and mobile devices.

Images