A method and system for identifying the opening and closing state of a valve in a thermal power plant based on image matching

Abstract

The application relates to the technical field of image recognition, and particularly discloses a valve opening and closing state recognition method and system for a thermal power plant based on image matching, which comprises the following steps: collecting environment parameters of a valve area; adaptively adjusting optical parameters based on the environment parameters, triggering pre-arranged air curtain cleaning when dust concentration exceeds a standard, and obtaining an anti-interference original image; performing mechanical structure guided analysis on the original image based on a valve three-dimensional mechanical structure library, and dynamically segmenting a region of interest of a key mechanical component; extracting morphological features and topological structure features of the region of interest, and obtaining a mechanical feature vector; comparing the mechanical feature vector with a standard state feature template library, and performing weighted fusion, to obtain a state confidence; and outputting a recognition result or triggering an artificial review prompt according to the state confidence; and the application improves the robustness and accuracy of automatic recognition of the valve opening and closing state of the thermal power plant.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for recognizing the valve switching status of a thermal power plant based on image matching. Background Technology

[0002] Valves in thermal power plants are critical actuators controlling the flow of media such as steam, water, and oil. Accurate identification of valve open / closed states directly impacts the safe and stable operation of the unit. Currently, valve status identification in thermal power plants primarily relies on manual inspection and visual confirmation or monitoring methods based on contact sensors. Manual inspection is labor-intensive, inefficient, and poses safety risks in high-temperature and high-pressure areas. Contact sensor methods require modifications to existing valves, resulting in high deployment costs and long deployment cycles. However, existing image-based valve status identification methods lack prior knowledge guidance mechanisms tailored to the inherent mechanical structural features of valves. Image segmentation processes rely on general object detection or color / texture-based segmentation methods, making it difficult to accurately locate critical mechanical components such as valve stems and handwheels in the complex environment of a thermal power plant with intersecting pipelines and cluttered backgrounds. This leads to inaccurate extraction of regions of interest and subsequent status determination failures.

[0003] In existing technologies, image-based valve status recognition methods rely excessively on superficial features such as paint color and valve opening scale markings during feature extraction. They lack multi-dimensional feature extraction methods for the inherent mechanical positional relationships of valves, making it difficult to extract stable and reliable discriminative features even under conditions of valve corrosion and damaged markings. This leads to a significant decrease in status recognition accuracy. Furthermore, existing status discrimination mechanisms typically employ single-feature template matching, lacking multi-feature fusion mechanisms. When some features fail due to local occlusion, uneven lighting, or other reasons, the system's fault tolerance is insufficient, resulting in a high false positive rate and failing to meet the high reliability requirements of thermal power plants. Therefore, there is an urgent need to develop an image-matching-based valve on / off status recognition method for thermal power plants to address the problems of low accuracy and poor robustness in valve status recognition under complex industrial environments, thereby improving the reliability and practicality of automated valve status monitoring in thermal power plants. Summary of the Invention

[0004] This invention provides a method and system for identifying the valve switching status of thermal power plants based on image matching, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for recognizing the valve switching status of a thermal power plant based on image matching, comprising: S1: The environmental parameters of the valve area are collected by the image acquisition device in the valve area of ​​the thermal power plant. The environmental parameters include light intensity parameters, dust concentration parameters and vibration amplitude parameters. S2: Based on the environmental parameters, the optical parameters of the image acquisition device are adaptively adjusted, and when the dust concentration parameter exceeds the preset dust threshold, the front air curtain cleaning of the valve area is triggered to acquire the anti-interference original image of the valve area. S3: Based on a preset 3D mechanical structure library of valves, perform mechanical structure-guided image analysis on the anti-interference original image, and dynamically segment the region of interest of the mechanical structure from the anti-interference original image; S4: Perform multi-dimensional image processing on the region of interest, extract and quantify the morphological and topological features of the current mechanical position in the valve region, and obtain the mechanical feature vector of the valve region; S5: Compare the mechanical feature vector with the pre-stored standard state feature template library to obtain the feature similarity value of the mechanical feature vector. Then, perform weighted fusion on the feature similarity value to obtain the state confidence of the valve area. S6: When the confidence level of the status exceeds the preset confidence threshold, output the switch status identification result of the valve area; when the confidence level of the status does not exceed the confidence threshold, mark the valve area as uncertain and trigger a manual review prompt.

[0006] In a preferred embodiment, the image acquisition device in the valve area of ​​the thermal power plant acquires environmental parameters of the valve area, including light intensity parameters, dust concentration parameters, and vibration amplitude parameters, including: An image acquisition device integrating environmental sensors is deployed in the valve area. The image acquisition device includes a visible light camera, an infrared fill light, and a vibration sensor. Based on the visible light camera, the light intensity parameters of the valve area are collected, and the infrared fill light provides an auxiliary light source when the light is insufficient. Based on the vibration sensor, the vibration amplitude parameters of the valve area are collected; based on the environmental sensor, the dust concentration parameters of the valve area are collected. The light intensity parameter, the dust concentration parameter, and the vibration amplitude parameter are aggregated into the environmental parameters of the valve area.

[0007] In a preferred embodiment, the process of adaptively adjusting the optical parameters of the image acquisition device based on the environmental parameters, and triggering a pre-air curtain cleaning of the valve area when the dust concentration parameter exceeds a preset dust threshold, to acquire an anti-interference original image of the valve area, includes: Extract the light intensity parameter from the environmental parameters, and compare the light intensity parameter with the light reference range of the valve area; When the light intensity parameter deviates from the light reference range, an exposure compensation parameter is generated based on the direction and degree of deviation of the light intensity parameter, wherein the deviation direction includes the direction of deviation to the higher range and the direction of deviation to the lower range. Based on the exposure compensation parameters, the exposure time and gain coefficient of the image acquisition device are adjusted in a coordinated manner to bring the imaging brightness of the image acquisition device back to the illumination reference range. When the vibration amplitude parameter in the environmental parameters exceeds the vibration threshold of the valve area, the optical image stabilization mechanism of the image acquisition device is activated to compensate for image jitter during the acquisition process and obtain the adjusted optical parameters. Based on the adjusted optical parameters, when the dust concentration parameter exceeds the preset dust threshold, the pre-air curtain cleaning of the valve area is triggered, and the original anti-interference image of the valve area is acquired.

[0008] In a preferred embodiment, the step of triggering a pre-air curtain cleaning of the valve area based on the adjusted optical parameters when the dust concentration parameter exceeds a preset dust threshold, and acquiring an anti-interference original image of the valve area, includes: The dust concentration parameter in the environmental parameters is compared with the dust threshold of the valve area in real time to generate a dust exceedance judgment result; When the dust exceedance determination result is that the dust exceeds the limit, an air curtain start command is sent to the front air curtain cleaning mechanism of the image acquisition device. After receiving the air curtain activation command, the valve area is cleaned by the front air curtain based on the continuous airflow curtain of the front air curtain cleaning mechanism, and the cleaned image acquisition device is obtained. Based on the cleaned image acquisition device, an anti-interference original image of the valve area is acquired.

[0009] In a preferred embodiment, the preset three-dimensional mechanical structure library of valves includes: Obtain three-dimensional geometric models of various valves in thermal power plants; The three-dimensional geometric model is structurally analyzed to extract the spatial positional relationship data of key mechanical components of various valves in the fully open and fully closed states; Based on the spatial positional relationship data, open-state mechanical structure templates and closed-state mechanical structure templates for various valves are established; The open-state mechanical structure template and the closed-state mechanical structure template are indexed and arranged to form the valve three-dimensional mechanical structure library.

[0010] In a preferred embodiment, the image analysis guided by the mechanical structure of the original image against interference, based on a preset three-dimensional mechanical structure library of valves, dynamically segments the region of interest of the mechanical structure from the original image against interference, including: Edge detection and contour extraction are performed on the original anti-interference image to obtain the contour boundary of the valve body in the original anti-interference image; The valve body type is identified by coarsely matching the outline boundary with the three-dimensional geometric model in the valve three-dimensional mechanical structure library. Based on the open-state mechanical structure template and the closed-state mechanical structure template, determine the estimated location region of the key mechanical component in the anti-interference original image; Using the boundary coordinates of the estimated location region as the cropping reference, coordinate constraint cropping is performed on the anti-interference original image to obtain the region of interest of the key mechanical component.

[0011] In a preferred embodiment, the step of performing multi-dimensional image processing on the region of interest to extract and quantify the morphological and topological features of the current mechanical position in the valve region, thereby obtaining a mechanical feature vector of the valve region, includes: The region of interest is subjected to grayscale processing and binarization segmentation to separate the connected regions of the key mechanical components; Based on the connected regions, morphological processing is performed on the region of interest to extract its morphological features; Based on the spatial relationship between the connected regions, the topological features of the region of interest are extracted; The morphological features and the topological features are combined to form the mechanical feature vector of the valve region.

[0012] In a preferred embodiment, the step of comparing the mechanical feature vector with a pre-stored standard state feature template library to obtain a feature similarity value of the mechanical feature vector, and then performing weighted fusion on the feature similarity values ​​to obtain the state confidence of the valve region, includes: Each feature dimension in the mechanical feature vector is compared with the corresponding standard feature dimension in the standard state feature template library one by one to obtain the feature similarity value of each feature dimension. Based on the image quality parameters of the region of interest, calculate the feature confidence parameter for each feature dimension; Based on the feature similarity values ​​and the feature confidence parameters, a covariance matrix between feature dimensions is established; Based on the feature similarity value, the feature confidence parameter, and the covariance matrix, the state fusion confidence of the valve region is calculated using a weighted fusion calculation formula. The weighted fusion calculation formula is as follows: ; in, The state fusion confidence level of the valve region; The total number of feature dimensions in the mechanical feature vector. For the first Feature similarity values ​​for each feature dimension For the first Feature credibility parameters for each feature dimension The covariance matrix between the feature dimensions is... The inverse of the covariance matrix is ​​the i-th row and j-th column. Column elements, The inverse covariance matrix is ​​the first... The sum of the absolute values ​​of all elements in the row. It is a regularization stabilization factor.

[0013] In a preferred embodiment, when the state confidence level exceeds a preset confidence threshold, the valve region's on / off state identification result is output; when the state confidence level does not exceed the confidence threshold, the valve region is marked as having an uncertain state and a manual review prompt is triggered, including: When the state confidence does not exceed the preset confidence threshold, a state uncertainty marker for the valve area is generated, and the anti-interference original image, the mechanical feature vector, and the state confidence are packaged into a manual review data package. The manual review data packet is pushed to the maintenance personnel's maintenance terminal; After receiving the manual review data packet, the operation and maintenance terminal displays the anti-interference original image and the status confidence level for operation and maintenance personnel to perform manual status judgment and generate manual judgment results. The system receives the manual judgment result from the operation and maintenance terminal, outputs the manual judgment result as the final state recognition result, and feeds the manual judgment result back to the standard state feature template library for template update and iteration.

[0014] To address the aforementioned problems, the present invention also provides an image matching-based valve on / off status recognition system for thermal power plants, the system comprising: An environmental parameter acquisition module is used to acquire environmental parameters of the valve area in a thermal power plant through an image acquisition device in the valve area. The environmental parameters include light intensity parameters, dust concentration parameters, and vibration amplitude parameters. An adaptive image acquisition module is used to adaptively adjust the optical parameters of the image acquisition device based on the environmental parameters, and to trigger the pre-air curtain cleaning of the valve area when the dust concentration parameter exceeds a preset dust threshold, thereby acquiring an anti-interference original image of the valve area. The mechanical structure guided segmentation module is used to perform mechanical structure guided image analysis on the anti-interference original image based on a preset valve 3D mechanical structure library, and dynamically segment the region of interest of the mechanical structure from the anti-interference original image; The multidimensional feature extraction module is used to perform multidimensional image processing on the region of interest, extract and quantify the morphological and topological features of the current mechanical position in the valve region, and obtain the mechanical feature vector of the valve region. The fusion and discrimination module is used to compare the mechanical feature vector with the pre-stored standard state feature template library to obtain the feature similarity value of the mechanical feature vector. The feature similarity value is then weighted and fused to obtain the state confidence of the valve area. The status output and verification module is used to output the valve area's on / off status identification result when the status confidence exceeds a preset confidence threshold; when the status confidence does not exceed the confidence threshold, the valve area is marked as having an uncertain status and a manual verification prompt is triggered.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention provides prior knowledge through a pre-set 3D mechanical structure library of valves, and performs mechanical structure-guided image analysis on the acquired images. It dynamically segments the regions of interest (ROIs) covering key mechanical components such as valve stems and handwheels, effectively eliminating the influence of intersecting pipelines and background clutter in thermal power plants. This solves the problems of inaccurate ROI localization and low segmentation accuracy in complex industrial scenes by existing image recognition methods. This invention abandons the extraction method that relies on the appearance features of valves, such as paint color and opening scale, which are easily corroded by the environment. Instead, it extracts the morphological and topological features of inherent mechanical structures, such as the valve stem length-to-diameter ratio, handwheel deflection angle, and the relative position of the valve stem and packing gland. This fundamentally solves the technical problem of valve appearance corrosion and label damage causing status recognition failure, and significantly improves the feature stability and discrimination reliability in the image matching process.

[0016] 2. This invention introduces a multi-feature weighted fusion mechanism in the state discrimination stage. Based on the confidence parameters of each feature dimension and the covariance matrix between feature dimensions, multiple similarity values ​​are weighted and fused to generate a state confidence score as the discrimination criterion. This effectively solves the problem of high misjudgment rate in existing single-feature template matching methods when some features fail due to local occlusion, uneven lighting, etc. Furthermore, when the state confidence score does not exceed a preset threshold, the system automatically triggers a manual review mechanism and feeds back the manual discrimination result to the standard state feature template library for iterative updates, achieving continuous optimization of the recognition model. Therefore, this invention provides a non-contact state recognition solution that does not require modification of existing valve equipment, solving the problems of low accuracy and poor robustness in state recognition caused by changes in lighting, dust, water vapor, cluttered backgrounds, and valve appearance deterioration in complex industrial environments. This improves the reliability and practicality of automated monitoring of valve opening and closing states in thermal power plants. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for recognizing the valve switch status in a thermal power plant based on image matching, according to an embodiment of the present invention. Figure 2 A functional block diagram of a valve switch status recognition system for thermal power plants based on image matching, provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a method for recognizing the valve switch status of a thermal power plant based on image matching. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for recognizing the valve switch status of a thermal power plant based on image matching can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0020] Reference Figure 1The diagram shown is a flowchart illustrating a method for recognizing the valve switch status of a thermal power plant based on image matching, according to an embodiment of the present invention. In this embodiment, the method for recognizing the valve switch status of a thermal power plant based on image matching includes: S1: The environmental parameters of the valve area are collected by the image acquisition device in the valve area of ​​the thermal power plant. The environmental parameters include light intensity parameters, dust concentration parameters and vibration amplitude parameters. In this embodiment of the invention, the image acquisition device for the valve area in the thermal power plant acquires environmental parameters of the valve area. These environmental parameters include light intensity parameters, dust concentration parameters, and vibration amplitude parameters, including: An image acquisition device integrating environmental sensors is deployed in the valve area. The image acquisition device includes a visible light camera, an infrared fill light, and a vibration sensor. Based on the visible light camera, the light intensity parameters of the valve area are collected, and the infrared fill light provides an auxiliary light source when the light is insufficient. Based on the vibration sensor, the vibration amplitude parameters of the valve area are collected; based on the environmental sensor, the dust concentration parameters of the valve area are collected. The light intensity parameter, the dust concentration parameter, and the vibration amplitude parameter are aggregated into the environmental parameters of the valve area.

[0021] An image acquisition device integrating environmental sensors is deployed in the valve area. The image acquisition device includes a visible light camera, an infrared fill light, and a vibration sensor. The visible light camera is encapsulated in an industrial-grade protective shell, and a high-temperature resistant dustproof glass cover is installed in front of the lens. The image acquisition device is fixedly installed at a preset observation position directly in front of the valve by a bracket. The preset observation position maintains a fixed observation distance and observation angle with the valve to ensure that the field of view of the visible light camera completely covers the mechanical parts area of ​​the valve.

[0022] The visible light camera continuously senses the ambient lighting conditions in the valve area while in operation. The ambient lighting conditions include constant light from the power plant's lighting system, interference from reflected light from the pipe insulation layer, and changes in natural light from the maintenance passage. The visible light camera converts the sensed ambient lighting conditions into light intensity parameters for the valve area. The light intensity parameters are output as electrical signals and transmitted to the data processing unit of the image acquisition device.

[0023] Infrared supplementary lights provide an auxiliary light source when the light intensity parameters indicate insufficient light. The auxiliary light source illuminates the valve area with invisible infrared wavelengths, ensuring that the visible light camera can still capture clear valve images under low-light conditions.

[0024] The vibration sensor is mechanically fastened to the bottom of the image acquisition device housing. The vibration sensor continuously senses the mechanical vibrations transmitted to the image acquisition device from the operation of the thermal power plant equipment. The mechanical vibrations include low-frequency vibrations from the operation of adjacent pumps, pulse vibrations from the flow of media in the steam pipes, and environmental vibrations transmitted from the plant floor. The vibration sensor converts the sensed mechanical vibrations into vibration amplitude parameters for the valve area.

[0025] An environmental sensor is installed on the side of the image acquisition device housing. The environmental sensor continuously senses the dust particle concentration around the valve area. The dust particles come from fly ash produced by coal combustion in thermal power plants, metal shavings generated during equipment maintenance, and suspended particulate matter carried by the plant ventilation system. The environmental sensor converts the sensed dust particle concentration into a dust concentration parameter for the valve area.

[0026] The data processing unit aligns and synchronizes the received light intensity parameters, dust concentration parameters, and vibration amplitude parameters according to a unified timestamp, and aggregates them into environmental parameters for the valve area in the form of data packets.

[0027] The beneficial effects are as follows: This embodiment, by deploying an image acquisition device integrating a visible light camera, an infrared supplementary light, a vibration sensor, and an environmental sensor in the valve area, realizes real-time perception and parameterized characterization of lighting conditions, dust concentration, and mechanical vibration in the complex industrial environment of a thermal power plant, providing an accurate and reliable data foundation for subsequent adaptive adjustment of optical parameters and triggering of air curtain cleaning.

[0028] S2: Based on the environmental parameters, the optical parameters of the image acquisition device are adaptively adjusted, and when the dust concentration parameter exceeds the preset dust threshold, the front air curtain cleaning of the valve area is triggered to acquire the anti-interference original image of the valve area. In this embodiment of the invention, the step of adaptively adjusting the optical parameters of the image acquisition device based on the environmental parameters, and triggering the pre-air curtain cleaning of the valve area when the dust concentration parameter exceeds a preset dust threshold, to acquire an anti-interference original image of the valve area, includes: Extract the light intensity parameter from the environmental parameters, and compare the light intensity parameter with the light reference range of the valve area; When the light intensity parameter deviates from the light reference range, an exposure compensation parameter is generated based on the direction and degree of deviation of the light intensity parameter, wherein the deviation direction includes the direction of deviation to the higher range and the direction of deviation to the lower range. Based on the exposure compensation parameters, the exposure time and gain coefficient of the image acquisition device are adjusted in a coordinated manner to bring the imaging brightness of the image acquisition device back to the illumination reference range. When the vibration amplitude parameter in the environmental parameters exceeds the vibration threshold of the valve area, the optical image stabilization mechanism of the image acquisition device is activated to compensate for image jitter during the acquisition process and obtain the adjusted optical parameters. Based on the adjusted optical parameters, when the dust concentration parameter exceeds the preset dust threshold, the pre-air curtain cleaning of the valve area is triggered, and the original anti-interference image of the valve area is acquired.

[0029] Based on the adjusted optical parameters, when the dust concentration exceeds a preset dust threshold, a pre-air curtain cleaning of the valve area is triggered, and an anti-interference original image of the valve area is acquired, including: The dust concentration parameter in the environmental parameters is compared with the dust threshold of the valve area in real time to generate a dust exceedance judgment result; When the dust exceedance determination result is that the dust exceeds the limit, an air curtain start command is sent to the front air curtain cleaning mechanism of the image acquisition device. After receiving the air curtain activation command, the valve area is cleaned by the front air curtain based on the continuous airflow curtain of the front air curtain cleaning mechanism, and the cleaned image acquisition device is obtained. Based on the cleaned image acquisition device, an anti-interference original image of the valve area is acquired.

[0030] The data processing unit extracts the light intensity parameter from the environmental parameters and compares the light intensity parameter with the light reference range of the valve area. The light reference range is set in advance according to the light range when the visible light camera can output the best image quality under standard imaging conditions. The upper limit of the light reference range corresponds to the light threshold at which the image is not overexposed, and the lower limit of the light reference range corresponds to the lowest light threshold at which the image is not distorted.

[0031] When the illumination intensity parameter deviates from the illumination reference range, the data processing unit determines the direction of deviation, which can be either too high or too low. A high deviation indicates excessive ambient light, potentially causing highlight clipping in the image; a low deviation indicates insufficient ambient light, potentially leading to loss of image detail. The data processing unit generates exposure compensation parameters based on the direction and degree of deviation. When the deviation is too high, the exposure compensation parameter is negative to reduce the amount of light entering the image; when the deviation is too low, the exposure compensation parameter is positive to increase the amount of light entering the image. Based on the exposure compensation parameters, the data processing unit adjusts the exposure time and gain coefficient of the image acquisition device accordingly. When the exposure compensation parameter is positive, it prioritizes extending the exposure time and appropriately increasing the gain coefficient to enhance image brightness; when the exposure compensation parameter is negative, it prioritizes shortening the exposure time and appropriately decreasing the gain coefficient to avoid noise amplification. This coordinated adjustment brings the image brightness output by the visible light camera back to the illumination reference range.

[0032] The data processing unit simultaneously compares the vibration amplitude parameter with the preset vibration threshold of the valve area. The preset vibration threshold is set in advance based on the upper limit of vibration amplitude that the visible light camera can withstand under the condition of the maximum allowable image offset. When the vibration amplitude parameter exceeds the preset vibration threshold, the data processing unit sends a stabilization activation signal to the visible light camera. The visible light camera then activates its built-in optical image stabilization mechanism. The optical image stabilization mechanism uses an internal gyroscope sensor to sense the vibration displacement direction and amplitude of the lens in real time, and drives the optical compensation lens to perform a constant amplitude offset movement in the opposite direction of the vibration displacement direction. This cancels out the interference of mechanical vibration on the imaging optical path at the optical level, and the adjusted optical parameters are obtained through the action of the optical image stabilization mechanism.

[0033] The data processing unit compares the dust concentration parameter with the preset dust threshold of the valve area in real time based on the adjusted optical parameters. The preset dust threshold is set in advance according to the maximum dust concentration that can maintain image clarity on the lens surface of the visible light camera under acceptable dust adhesion conditions. When the dust concentration parameter is greater than or equal to the preset dust threshold, the dust exceedance judgment result is exceeded; when the dust concentration parameter is less than the preset dust threshold, the dust exceedance judgment result is not exceeded.

[0034] When the dust exceedance determination result is "exceeding the limit," the data processing unit sends an air curtain activation command to the front air curtain cleaning mechanism of the image acquisition device. The front air curtain cleaning mechanism includes a compressed air storage tank, a solenoid valve, and an annular airflow nozzle. The annular airflow nozzle is arranged in a ring around the outer edge of the visible light camera lens. Upon receiving the air curtain activation command, the solenoid valve opens, and high-pressure clean air is delivered to the annular airflow nozzle through the solenoid valve, forming a continuous annular airflow curtain in front of the lens. This airflow curtain prevents dust particles from adhering to the lens surface and blows away any already attached dust particles, keeping the optical path in front of the visible light camera lens clean. Under the continuous action of the airflow curtain, the visible light camera acquires images of the valve area based on adjusted optical parameters, ultimately outputting an anti-interference original image of the valve area. This anti-interference original image features moderate exposure, stable image, and clear imaging.

[0035] The beneficial effects are as follows: This embodiment achieves the linkage and adaptive adjustment of exposure time and gain coefficient by comparing the light intensity parameter with the light reference range, which solves the problem of image overexposure or underexposure caused by sudden changes in light conditions at thermal power plants. It automatically activates the optical image stabilization mechanism by comparing the vibration amplitude parameter with the preset vibration threshold, which suppresses the interference of equipment operation vibration on image acquisition stability. It automatically triggers the front air curtain cleaning mechanism by comparing the dust concentration parameter with the preset dust threshold, which solves the technical problem of image blurring caused by dust adhesion. The triple protection mechanism works together to ensure high-quality output of anti-interference original images.

[0036] S3: Based on a preset 3D mechanical structure library of valves, perform mechanical structure-guided image analysis on the anti-interference original image, and dynamically segment the region of interest of the mechanical structure from the anti-interference original image; In this embodiment of the invention, the preset three-dimensional mechanical structure library of valves includes: Obtain three-dimensional geometric models of various valves in thermal power plants; The three-dimensional geometric model is structurally analyzed to extract the spatial positional relationship data of key mechanical components of various valves in the fully open and fully closed states; Based on the spatial positional relationship data, open-state mechanical structure templates and closed-state mechanical structure templates for various valves are established; The open-state mechanical structure template and the closed-state mechanical structure template are indexed and arranged to form the valve three-dimensional mechanical structure library.

[0037] The system, based on a pre-defined 3D mechanical structure library for valves, performs mechanical structure-guided image analysis on the original, interference-resistant image, dynamically segmenting the region of interest (ROI) of the mechanical structure from the original, interference-resistant image, including: Edge detection and contour extraction are performed on the original anti-interference image to obtain the contour boundary of the valve body in the original anti-interference image; The valve body type is identified by coarsely matching the outline boundary with the three-dimensional geometric model in the valve three-dimensional mechanical structure library. Based on the open-state mechanical structure template and the closed-state mechanical structure template, determine the estimated location region of the key mechanical component in the anti-interference original image; Using the boundary coordinates of the estimated location region as the cropping reference, coordinate constraint cropping is performed on the anti-interference original image to obtain the region of interest of the key mechanical component.

[0038] In thermal power plants, various valves have different mechanical structural characteristics. Gate valves open and close by moving the valve stem up and down. Globe valves achieve sealing by moving the valve disc along the valve seat axis via the valve stem. Ball valves control the flow path by rotating the ball via the valve stem. The spatial positions of the key mechanical components of various valves differ significantly between the fully open and fully closed states. These differences form the physical basis for using image recognition to determine the valve's open / closed state.

[0039] 3D geometric models of various valves in thermal power plants were obtained using 3D modeling tools. These models accurately described the external contours and internal mechanical structures of each valve using point cloud data and triangular mesh patches. The valve types acquired included at least gate valves, globe valves, and ball valves. Structural analysis was performed on the 3D geometric models, dividing them into several key mechanical components according to their mechanical functions. These key mechanical components included at least the valve stem, handwheel, packing gland, and actuator linkage. The structural analysis process involved feature surface extraction and feature edge identification of the 3D geometric models, separating each key mechanical component into independent geometric units. For each separated key mechanical component, spatial positional relationship data was extracted in both fully open and fully closed states. This spatial positional relationship data included the center coordinates, principal axis direction vectors, outer contour dimensions, and relative distances and angles between each key mechanical component. Based on spatial positional relationship data, open-state mechanical structure templates and closed-state mechanical structure templates for various types of valves are established. The open-state mechanical structure template fully describes the spatial layout relationship of each key mechanical component of the valve in the fully open state, and the closed-state mechanical structure template fully describes the spatial layout relationship of each key mechanical component of the valve in the fully closed state. The two types of templates store the position coordinates, direction vectors and size parameters of each key mechanical component in a structured data format.

[0040] The open-state and closed-state mechanical structure templates are classified and archived according to valve type, and a unique index identifier is assigned to each type of template. The index identifier enables quick retrieval and retrieval of mechanical structure templates for specific valve types in specific states, ultimately forming a three-dimensional mechanical structure library for valves.

[0041] When performing mechanical structure-guided image analysis based on the valve 3D mechanical structure library to combat interference in the original image, the first step is to perform edge detection processing on the original image. Edge detection processing calculates the gray-level gradient amplitude of each pixel in the image and marks pixels with gray-level gradient amplitudes exceeding a preset edge threshold as edge pixels, thereby extracting the boundary lines of objects in the image. Based on this, contour extraction is performed, and adjacent edge pixels are connected into closed contour curves according to connectivity rules to obtain the contour boundary of the valve body in the original image.

[0042] The contour boundary is coarsely matched with the 3D geometric models in the valve 3D mechanical structure library. This coarse matching process compares the shape similarity between the 2D shape described by the contour boundary and the 2D projected contour of the 3D geometric model at the same viewing angle, selecting the 3D geometric model with the highest shape similarity as the matching result. This identifies the valve type of the valve body in the anti-interference original image. Based on the identified valve type, the corresponding open and closed mechanical structure templates are retrieved from the valve 3D mechanical structure library. The spatial positional relationship data of each key mechanical component in the template is mapped from the 3D coordinate system to the 2D pixel coordinate system of the anti-interference original image. The mapping process is based on the known viewing angle and distance of the image acquisition device, using perspective projection transformation to convert the 3D spatial coordinates into 2D image coordinates, thus determining the estimated position regions of each key mechanical component in the anti-interference original image.

[0043] Using the boundary coordinates of the estimated location area as the cropping reference, coordinate constraint cropping is performed on the original image to resist interference, removing the background area outside the cropping range and retaining only the image content within the cropping range. The region of interest of the key mechanical components is obtained through coordinate constraint cropping. The region of interest is a local image containing only key mechanical components such as valve stem, handwheel, and packing gland.

[0044] The beneficial effects are as follows: This embodiment provides an accurate source of prior knowledge for image analysis by establishing a three-dimensional mechanical structure library of valves. It achieves automatic identification of valve types by coarsely matching the contour boundaries of the anti-interference original image with the three-dimensional geometric model. It achieves automatic generation of the estimated position area of ​​key mechanical components by mapping the spatial position relationship data in the three-dimensional mechanical structure template to the two-dimensional image coordinate system. It performs coordinate constraint clipping based on the estimated position area, effectively eliminating the interference of intersecting pipelines and background clutter in the thermal power plant. It solves the technical problem of inaccurate positioning of the region of interest in complex industrial environments by existing image recognition methods.

[0045] S4: Perform multi-dimensional image processing on the region of interest, extract and quantify the morphological and topological features of the current mechanical position in the valve region, and obtain the mechanical feature vector of the valve region; In this embodiment of the invention, the step of performing multi-dimensional image processing on the region of interest to extract and quantify the morphological and topological features of the current mechanical position in the valve region to obtain the mechanical feature vector of the valve region includes: The region of interest is subjected to grayscale processing and binarization segmentation to separate the connected regions of the key mechanical components; Based on the connected regions, morphological processing is performed on the region of interest to extract its morphological features; Based on the spatial relationship between the connected regions, the topological features of the region of interest are extracted; The morphological features and the topological features are combined to form the mechanical feature vector of the valve region.

[0046] The region of interest (ROI) is converted to grayscale. Grayscale conversion transforms the red, green, and blue color values ​​of each pixel in the ROI into a single grayscale value using a weighted average. The grayscale value ranges from 0 to 255. Grayscale conversion eliminates the interference of color information on subsequent image segmentation. The grayscale image is then binarized. Binarization segmentation uses a grayscale threshold to mark pixels with grayscale values ​​greater than or equal to the threshold as foreground pixels and pixels with grayscale values ​​less than the threshold as background pixels. Foreground pixels are uniformly set to white, and background pixels are uniformly set to black. After binarization, the grayscale image is converted into a binary image containing only black and white pixel values. In the binary image, white areas correspond to the physical parts of the key mechanical components, and black areas correspond to the background. Connected regions of the key mechanical components are separated from the ROI through binarization segmentation. A connected region is a continuous area composed of all adjacent white pixels in the binary image, and each connected region physically corresponds to an independent key mechanical component.

[0047] Morphological processing of the region of interest (ROI) is performed based on the connected regions. Morphological features include at least the aspect ratio of the visible portion of the valve stem, the deflection angle of the handwheel plane relative to the valve body reference plane, and the visible area percentage of the packing gland. The aspect ratio of the visible portion of the valve stem is obtained by measuring the ratio of the principal axis length to the secondary axis length of the valve stem's connected region. The principal axis length is the maximum span of the connected region along the valve stem's extension direction, and the secondary axis length is the maximum span of the connected region perpendicular to the principal axis direction. A larger aspect ratio indicates that the valve stem extends further, corresponding to the valve being open; a smaller aspect ratio indicates that the valve stem retracts further, corresponding to the valve being closed. The deflection angle of the handwheel plane relative to the valve body reference plane is obtained by calculating the angle between the principal axis direction of the handwheel's connected region and a preset reference direction. The deflection angle value corresponds to the angle rotated by the handwheel from the fully closed position to the current position. The visible area percentage of the packing gland is obtained by calculating the ratio of the pixel area of ​​the packing gland's connected region to the total area of ​​the ROI. Changes in the visible area percentage are related to the valve stem's lifting and lowering motion.

[0048] Topological features are extracted based on the spatial relationships between connected regions. These features include at least the relative positions of the valve stem and packing gland, the distance between the handwheel spokes and the valve stem tip, and the contact state between the actuator feedback rod and the limit switch. The relative positions of the valve stem and packing gland are calculated by the longitudinal offset between the centroid coordinates of the valve stem's connected region and the packing gland's connected region. A positive offset indicates the valve stem is raised, while a negative offset indicates it is lowered. The distance between the handwheel spokes and the valve stem tip is calculated by the pixel distance between the nearest edge point of the handwheel spokes' connected region and the nearest edge point of the valve stem tip's connected region. The contact state between the actuator feedback rod and the limit switch is determined by whether there is pixel overlap between the actuator feedback rod's connected region and the limit switch's connected region. The morphological and topological features are then arranged in a fixed order to form an ordered numerical sequence, which constitutes the mechanical feature vector of the valve region.

[0049] The beneficial effects are as follows: This embodiment separates key mechanical components from the background into independent connected regions through grayscale processing and binarization segmentation. It extracts inherent mechanical structural features such as valve stem length-to-diameter ratio, handwheel deflection angle, and packing gland visible area ratio through morphological processing. It abandons appearance features that are easily corroded by the environment, such as valve surface paint color and opening scale scale. It extracts the spatial positional relationship between key mechanical components through topological structural features. The combination of morphological features and topological structural features enables the mechanical feature vector to completely describe the mechanical positional state of the valve from multiple dimensions, which significantly improves the accuracy and robustness of subsequent state discrimination.

[0050] S5: Compare the mechanical feature vector with the pre-stored standard state feature template library to obtain the feature similarity value of the mechanical feature vector. Then, perform weighted fusion on the feature similarity value to obtain the state confidence of the valve area. In this embodiment of the invention, the step of comparing the mechanical feature vector with a pre-stored standard state feature template library to obtain a feature similarity value of the mechanical feature vector, and then performing weighted fusion on the feature similarity values ​​to obtain the state confidence of the valve region, includes: Each feature dimension in the mechanical feature vector is compared with the corresponding standard feature dimension in the standard state feature template library one by one to obtain the feature similarity value of each feature dimension. Based on the image quality parameters of the region of interest, calculate the feature confidence parameter for each feature dimension; Based on the feature similarity values ​​and the feature confidence parameters, a covariance matrix between feature dimensions is established; Based on the feature similarity value, the feature confidence parameter, and the covariance matrix, the state fusion confidence of the valve region is calculated using a weighted fusion calculation formula. The weighted fusion calculation formula is as follows: ; in, The state fusion confidence level of the valve region; The total number of feature dimensions in the mechanical feature vector. For the first Feature similarity values ​​for each feature dimension For the first Feature credibility parameters for each feature dimension The covariance matrix between the feature dimensions is... The inverse matrix of the covariance matrix is ​​the nth... Line 1 Column elements, The inverse covariance matrix is ​​the first... The sum of the absolute values ​​of all elements in the row. It is a regularization stabilization factor.

[0051] The standard state feature template library pre-stores standard feature dimension values ​​of various valves in the fully open and fully closed states obtained by the same feature extraction method. For continuous feature dimensions, the difference ratio is calculated by dividing the absolute value of the difference between the current value and the standard value by the larger of the two values, and then subtracting the difference ratio from one to obtain the similarity value. For discrete feature dimensions, such as the contact state between the actuator feedback rod and the limit switch, the similarity value is one when the current state is completely consistent with the standard state, and zero when the two are inconsistent. The feature similarity value of each feature dimension is obtained by comparing them one by one.

[0052] Based on the image quality parameters of the region of interest (ROI), the feature confidence parameter for each feature dimension is calculated. The image quality parameters are obtained by calculating the mean gradient magnitude and grayscale variance for each connected region corresponding to a key mechanical component within the ROI. The mean gradient magnitude is obtained by calculating the square root of the sum of the squares of the grayscale differences of each pixel within the connected region in the horizontal and vertical directions, and then averaging this value over all pixels in the region. A larger mean gradient magnitude indicates clearer edges and richer image details in the region. The grayscale variance is obtained by calculating the sum of the squares of the differences between the grayscale values ​​of each pixel within the connected region and the region's mean grayscale value, then dividing by the total number of pixels in the region. A larger grayscale variance indicates richer grayscale levels and higher contrast in the region. The feature confidence parameter is obtained as the ratio of the mean gradient magnitude to the grayscale variance. When the image of a region containing a key mechanical component is blurry, both the mean gradient magnitude and grayscale variance are low, resulting in a low corresponding feature confidence parameter.

[0053] A covariance matrix is ​​established based on feature similarity values ​​and feature confidence parameters. The covariance matrix is ​​a square matrix, where the number of rows and columns equals the total number of feature dimensions in the mechanical feature vector. The covariance matrix is ​​then used to construct the covariance matrix for each feature dimension. Line 1 The elements of the column represent the first... The first feature dimension and the first feature dimension The covariance value between the feature dimensions is calculated by taking the feature dimension as the covariance value. The feature similarity value of the i-th feature dimension and the i-th feature dimension The covariance matrix is ​​obtained by multiplying the feature similarity values ​​of each feature dimension, subtracting the product of the two means, and then averaging the values ​​over multiple historical data sets. The covariance matrix fully describes the linear correlation between all feature dimensions.

[0054] Based on feature similarity values, feature confidence parameters, and the covariance matrix, the state fusion confidence of the valve region is calculated using a weighted fusion formula. The calculation process is as follows: First, the inverse of the covariance matrix is ​​calculated. The inverse matrix is ​​transformed into its mathematical inverse through matrix inversion. Then, the sum of the absolute values ​​of all elements in each row of the inverse matrix is ​​calculated to obtain the independence weight corresponding to each feature dimension. The larger the sum of the absolute values ​​of the rows, the more significant the difference between the first and second rows. The weaker the correlation between a feature dimension and other dimensions, the greater its weight in the fusion process. The smaller the sum of the absolute values ​​of the rows, the more likely it is that the first row is the first row. The stronger the correlation between a feature dimension and other dimensions, the higher the information redundancy. The weight of this feature dimension is weakened during the fusion process to avoid redundant information being calculated repeatedly. The numerator of the formula is calculated by multiplying the feature similarity value of each feature dimension by its corresponding feature confidence parameter and the sum of the absolute values ​​of the rows of the inverse matrix corresponding to that feature dimension, and then summing these products over all feature dimensions. The denominator is calculated by multiplying the feature confidence parameter of each feature dimension by the sum of the absolute values ​​of the rows of the inverse matrix corresponding to that feature dimension, summing these products over all feature dimensions, and then adding a regularization stabilization factor. The regularization stabilization factor is a very small positive number used to prevent the denominator from being zero, which would lead to computational anomalies. Finally, the state fusion confidence is obtained by dividing the numerator by the denominator.

[0055] The feature similarity value is derived from the dimension-by-dimensional similarity comparison between the mechanical feature vector and the standard state feature template library. The feature confidence parameter is derived from the ratio of the mean gradient magnitude of each connected region in the region of interest to the gray variance. The covariance matrix is ​​derived from the covariance calculation between the feature similarity values ​​of each feature dimension. The sum of the absolute values ​​of each row of the inverse matrix is ​​derived from the sum of the absolute values ​​of each row after the matrix inversion operation of the covariance matrix. The regularization stability factor is a preset minimum positive number.

[0056] When the feature similarity values ​​and feature confidence parameters of all feature dimensions are high, the state fusion confidence is close to one; when the feature confidence parameters of some feature dimensions are low, the contribution of that feature dimension to the state fusion confidence is weakened, and when the th feature dimension has a low confidence parameter, the state fusion confidence is close to one. When the covariance of the first feature dimension is large with other dimensions, the inverse matrix is... The smaller the sum of the absolute values ​​of the rows, the weaker the weight of that feature dimension in the formula, thus suppressing the repeated calculation of redundant information.

[0057] The beneficial effects are as follows: This embodiment ensures the refinement and traceability of the comparison process through dimension-by-dimensional similarity comparison. It realizes the quantitative evaluation of image quality of each feature dimension by calculating the feature credibility parameter based on the ratio of the gradient magnitude mean to the gray-level variance. By establishing the covariance matrix between feature dimensions and introducing the sum of the absolute values ​​of each row of the inverse matrix as the dimension-by-dimensional weight in the weighted fusion formula, the interference of information redundancy between feature dimensions on the fusion result is effectively eliminated. It comprehensively considers the three dimensions of feature similarity, feature credibility and feature independence, and significantly improves the robustness of valve switching status recognition in thermal power plants in complex industrial environments.

[0058] S6: When the confidence level of the status exceeds the preset confidence threshold, output the switch status identification result of the valve area; when the confidence level of the status does not exceed the confidence threshold, mark the valve area as uncertain and trigger a manual review prompt.

[0059] In this embodiment of the invention, when the state confidence level exceeds a preset confidence threshold, the valve region's on / off state identification result is output; when the state confidence level does not exceed the confidence threshold, the valve region is marked as having an uncertain state and a manual review prompt is triggered, including: When the state confidence does not exceed the preset confidence threshold, a state uncertainty marker for the valve area is generated, and the anti-interference original image, the mechanical feature vector, and the state confidence are packaged into a manual review data package. The manual review data packet is pushed to the maintenance personnel's maintenance terminal; After receiving the manual review data packet, the operation and maintenance terminal displays the anti-interference original image and the status confidence level for operation and maintenance personnel to perform manual status judgment and generate manual judgment results. The system receives the manual judgment result from the operation and maintenance terminal, outputs the manual judgment result as the final state recognition result, and feeds the manual judgment result back to the standard state feature template library for template update and iteration.

[0060] The data processing unit compares the state fusion confidence level with a preset confidence threshold, which is a judgment benchmark value set in advance according to the reliability requirements of valve state identification in thermal power plants. When the state fusion confidence level exceeds the preset confidence threshold, the data processing unit determines that the mechanical feature vector of the current valve matches a certain standard feature vector in the standard state feature template library sufficiently well, and can deterministically output the valve on / off state identification result. The on / off state identification result includes fully open state, fully closed state, and intermediate transition state. The data processing unit writes the on / off state identification result, along with the corresponding valve identification information and identification timestamp, into the state identification result database, and simultaneously pushes the on / off state identification result to the distributed control system of the thermal power plant for operators to view and access.

[0061] When the state fusion confidence level does not exceed the preset confidence threshold, the data processing unit generates a state uncertainty marker for the valve area. This marker, indicated by a specific identification code, indicates that the identification result is unreliable and requires manual intervention. The data processing unit packages the anti-interference original image, mechanical feature vector, and state fusion confidence level into a manual review data package. The anti-interference original image provides complete visual information of the valve area for maintenance personnel to visually assess. The mechanical feature vector provides detailed values ​​for each feature dimension for maintenance personnel to analyze the specific reasons for the identification failure. The state fusion confidence level provides the confidence level of this automatic identification for maintenance personnel to evaluate the reliability of the automatic identification system. The data processing unit pushes the manual review data packet to the maintenance personnel's terminal via a wired or wireless communication link. After receiving the manual review data packet, the maintenance terminal displays the anti-interference original image and the status fusion confidence level on the display interface. The maintenance personnel visually observe the valve stem position, handwheel angle, and actuator feedback rod status in the anti-interference original image, and perform a comprehensive analysis based on the values ​​of each feature dimension in the mechanical feature vector. The maintenance personnel make a manual judgment on the valve's opening and closing status based on their professional knowledge and field experience. The maintenance personnel submit the manual judgment result to the data processing unit through the input interface of the maintenance terminal.

[0062] After receiving the manual judgment results from the operation and maintenance terminal, the data processing unit outputs these results as the final state recognition results to the state recognition result database and pushes them to the distributed control system. Simultaneously, the data processing unit feeds back the manual judgment results to the standard state feature template library for template updates and iterations. The template update and iteration process is as follows: the data processing unit associates the mechanical feature vector extracted from the original anti-interference image with the standard state type label corresponding to the manual judgment result, adds this mechanical feature vector as a new training sample to the standard state feature template library, and updates the standard feature vectors corresponding to this type label in the standard state feature template library. The update method involves fusing the original standard feature vectors and the newly added mechanical feature vectors using a weighted average. The original standard feature vectors have higher weights to maintain the stability of historical data, while the newly added mechanical feature vectors have lower weights to introduce new feature information. Through template updates and iterations, the standard state feature template library can continuously adapt to changes in the mechanical structural characteristics of the valve caused by wear, corrosion, and maintenance during long-term operation.

[0063] The beneficial effects are as follows: This embodiment realizes a hierarchical decision-making mechanism of automatic judgment and manual review by comparing the state fusion confidence score with a preset confidence threshold. When the automatic recognition result is reliable, the switch state recognition result is directly output to avoid unnecessary manual intervention. When the automatic recognition result is unreliable, the manual review mechanism is automatically triggered to package the anti-interference original image, mechanical feature vector and state fusion confidence score and push them to the operation and maintenance terminal. By feeding back the manual judgment result to the standard state feature template library for template update and iteration, the continuous learning and adaptive optimization of the recognition system is realized, which solves the technical problem that the mechanical structure characteristics of the valve gradually change during long-term operation, resulting in a decrease in the accuracy of automatic recognition.

[0064] like Figure 2 The diagram shown is a functional block diagram of a valve switch status recognition system for thermal power plants based on image matching, provided by an embodiment of the present invention.

[0065] The image matching-based valve switch status recognition system 100 for thermal power plants described in this invention can be installed in an electronic device. Depending on the functions implemented, the image matching-based valve switch status recognition system 100 may include an environmental parameter acquisition module 101, an adaptive image acquisition module 102, a mechanical structure guided segmentation module 103, a multi-dimensional feature extraction module 104, a fusion discrimination module 105, and a status output and verification module 106. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0066] In this embodiment, the functions of each module / unit are as follows: The environmental parameter acquisition module 101 is used to acquire environmental parameters of the valve area through an image acquisition device in the valve area of ​​a thermal power plant. The environmental parameters include light intensity parameters, dust concentration parameters, and vibration amplitude parameters. The adaptive image acquisition module 102 is used to adaptively adjust the optical parameters of the image acquisition device based on the environmental parameters, and trigger the pre-air curtain cleaning of the valve area when the dust concentration parameter exceeds the preset dust threshold, so as to acquire the anti-interference original image of the valve area. The mechanical structure guided segmentation module 103 is used to perform mechanical structure guided image analysis on the anti-interference original image based on a preset valve three-dimensional mechanical structure library, and dynamically segment the region of interest of the mechanical structure from the anti-interference original image; The multidimensional feature extraction module 104 is used to perform multidimensional image processing on the region of interest, extract and quantify the morphological features and topological features of the current mechanical position in the valve region, and obtain the mechanical feature vector of the valve region. The fusion discrimination module 105 is used to compare the mechanical feature vector with the pre-stored standard state feature template library to obtain the feature similarity value of the mechanical feature vector, and to perform weighted fusion on the feature similarity value to obtain the state confidence of the valve area. The status output and verification module 106 is used to output the valve region's switch status identification result when the status confidence exceeds a preset confidence threshold; and to mark the valve region as having an uncertain status and trigger a manual verification prompt when the status confidence does not exceed the confidence threshold.

[0067] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0068] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0069] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0070] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0071] The embodiments of this application can acquire and process relevant data based on an artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for recognizing the valve on / off status in a thermal power plant based on image matching, characterized in that, The method includes: S1: The environmental parameters of the valve area are collected by the image acquisition device in the valve area of ​​the thermal power plant. The environmental parameters include light intensity parameters, dust concentration parameters and vibration amplitude parameters. S2: Based on the environmental parameters, the optical parameters of the image acquisition device are adaptively adjusted, and when the dust concentration parameter exceeds a preset dust threshold, a pre-air curtain cleaning of the valve area is triggered to acquire an anti-interference original image of the valve area, including: Extract the light intensity parameter from the environmental parameters, and compare the light intensity parameter with the light reference range of the valve area; When the light intensity parameter deviates from the light reference range, an exposure compensation parameter is generated based on the direction and degree of deviation of the light intensity parameter, wherein the deviation direction includes the direction of deviation to the higher range and the direction of deviation to the lower range. Based on the exposure compensation parameters, the exposure time and gain coefficient of the image acquisition device are adjusted in a coordinated manner to bring the imaging brightness of the image acquisition device back to the illumination reference range. When the vibration amplitude parameter in the environmental parameters exceeds the vibration threshold of the valve area, the optical image stabilization mechanism of the image acquisition device is activated to compensate for image jitter during the acquisition process and obtain the adjusted optical parameters. Based on the adjusted optical parameters, when the dust concentration parameter exceeds the preset dust threshold, the front air curtain cleaning of the valve area is triggered, and the anti-interference original image of the valve area is acquired. S3: Based on a pre-defined 3D mechanical structure library for valves, image analysis guided by the mechanical structure is performed on the original image to resist interference. The region of interest for the mechanical structure is dynamically segmented from the original image, including: Edge detection and contour extraction are performed on the original anti-interference image to obtain the contour boundary of the valve body in the original anti-interference image; The valve body type is identified by coarsely matching the outline boundary with the three-dimensional geometric model in the valve three-dimensional mechanical structure library. Based on the open-state mechanical structure template and the closed-state mechanical structure template, determine the estimated location region of the key mechanical component in the anti-interference original image; Using the boundary coordinates of the estimated location region as the cropping reference, coordinate constraint cropping is performed on the anti-interference original image to obtain the region of interest of the key mechanical component; The preset three-dimensional mechanical structure library for valves includes: Obtain three-dimensional geometric models of various valves in thermal power plants; The three-dimensional geometric model is structurally analyzed to extract the spatial positional relationship data of key mechanical components of various valves in the fully open and fully closed states; Based on the spatial positional relationship data, open-state mechanical structure templates and closed-state mechanical structure templates for various valves are established; The open-state mechanical structure template and the closed-state mechanical structure template are indexed and arranged to construct the valve three-dimensional mechanical structure library; S4: Perform multi-dimensional image processing on the region of interest, extract and quantify the morphological and topological features of the current mechanical position in the valve region, and obtain the mechanical feature vector of the valve region; S5: Compare the mechanical feature vector with the pre-stored standard state feature template library to obtain the feature similarity value of the mechanical feature vector. Then, perform weighted fusion of the feature similarity values ​​to obtain the state confidence of the valve region, including: Each feature dimension in the mechanical feature vector is compared with the corresponding standard feature dimension in the standard state feature template library one by one to obtain the feature similarity value of each feature dimension. Based on the image quality parameters of the region of interest, calculate the feature confidence parameter for each feature dimension; Based on the feature similarity values ​​and the feature confidence parameters, a covariance matrix between feature dimensions is established; Based on the feature similarity value, the feature confidence parameter, and the covariance matrix, the state fusion confidence of the valve region is calculated using a weighted fusion calculation formula. The weighted fusion calculation formula is as follows: ； in, The state fusion confidence level of the valve region; The total number of feature dimensions in the mechanical feature vector. For the first Feature similarity values ​​for each feature dimension For the first Feature credibility parameters for each feature dimension The covariance matrix between the feature dimensions is... The inverse matrix of the covariance matrix is ​​the nth... Line 1 Column elements, The inverse matrix of the covariance matrix is ​​the nth... The sum of the absolute values ​​of all elements in the row. As a regularization stabilization factor; S6: When the confidence level of the status exceeds the preset confidence threshold, output the switch status identification result of the valve area; when the confidence level of the status does not exceed the confidence threshold, mark the valve area as uncertain and trigger a manual review prompt.

2. The method for recognizing the valve switching status of a thermal power plant based on image matching as described in claim 1, characterized in that, The image acquisition device in the valve area of ​​the thermal power plant collects environmental parameters of the valve area, including light intensity parameters, dust concentration parameters, and vibration amplitude parameters. An image acquisition device integrating environmental sensors is deployed in the valve area. The image acquisition device includes a visible light camera, an infrared fill light, and a vibration sensor. Based on the visible light camera, the light intensity parameters of the valve area are collected, and the infrared fill light provides an auxiliary light source when the light is insufficient. Based on the vibration sensor, the vibration amplitude parameters of the valve area are collected; based on the environmental sensor, the dust concentration parameters of the valve area are collected. The light intensity parameter, the dust concentration parameter, and the vibration amplitude parameter are aggregated into the environmental parameters of the valve area.

3. The method for recognizing the valve switching status of a thermal power plant based on image matching as described in claim 2, characterized in that, Based on the adjusted optical parameters, when the dust concentration exceeds a preset dust threshold, a pre-air curtain cleaning of the valve area is triggered, and an anti-interference original image of the valve area is acquired, including: The dust concentration parameter in the environmental parameters is compared with the dust threshold of the valve area in real time to generate a dust exceedance judgment result; When the dust exceedance determination result is that the dust exceeds the limit, an air curtain start command is sent to the front air curtain cleaning mechanism of the image acquisition device. After receiving the air curtain activation command, the valve area is cleaned by the front air curtain based on the continuous airflow curtain of the front air curtain cleaning mechanism, and the cleaned image acquisition device is obtained. Based on the cleaned image acquisition device, an anti-interference original image of the valve area is acquired.

4. The method for recognizing the valve switching status of a thermal power plant based on image matching as described in claim 3, characterized in that, The process involves multi-dimensional image processing of the region of interest to extract and quantify the morphological and topological features of the current mechanical position within the valve region, resulting in a mechanical feature vector for the valve region, including: The region of interest is subjected to grayscale processing and binarization segmentation to separate the connected regions of the key mechanical components; Based on the connected regions, morphological processing is performed on the region of interest to extract its morphological features; Based on the spatial relationship between the connected regions, the topological features of the region of interest are extracted; The morphological features and the topological features are combined to form the mechanical feature vector of the valve region.

5. The method for recognizing the valve switching status of a thermal power plant based on image matching as described in claim 1, characterized in that, When the confidence level of the state exceeds a preset confidence threshold, the switch state identification result of the valve area is output. When the status confidence level does not exceed the confidence threshold, the valve area is marked as having an uncertain status and a manual review prompt is triggered, including: When the state confidence does not exceed the preset confidence threshold, a state uncertainty marker for the valve area is generated, and the anti-interference original image, the mechanical feature vector, and the state confidence are packaged into a manual review data package. The manual review data packet is pushed to the maintenance personnel's maintenance terminal; After receiving the manual review data packet, the operation and maintenance terminal displays the anti-interference original image and the status confidence level for operation and maintenance personnel to perform manual status judgment and generate manual judgment results. The system receives the manual judgment result from the operation and maintenance terminal, outputs the manual judgment result as the final state recognition result, and feeds the manual judgment result back to the standard state feature template library for template update and iteration.

6. A valve on / off status recognition system for thermal power plants based on image matching, characterized in that, The system for implementing the image matching-based valve on / off status recognition method for thermal power plants as described in claim 1 includes: An environmental parameter acquisition module is used to acquire environmental parameters of the valve area in a thermal power plant through an image acquisition device in the valve area. The environmental parameters include light intensity parameters, dust concentration parameters, and vibration amplitude parameters. An adaptive image acquisition module is used to adaptively adjust the optical parameters of the image acquisition device based on the environmental parameters, and to trigger pre-air curtain cleaning of the valve area when the dust concentration parameter exceeds a preset dust threshold, thereby acquiring an anti-interference original image of the valve area, including: Extract the light intensity parameter from the environmental parameters, and compare the light intensity parameter with the light reference range of the valve area; When the light intensity parameter deviates from the light reference range, an exposure compensation parameter is generated based on the direction and degree of deviation of the light intensity parameter, wherein the deviation direction includes the direction of deviation to the higher range and the direction of deviation to the lower range. Based on the exposure compensation parameters, the exposure time and gain coefficient of the image acquisition device are adjusted in a coordinated manner to bring the imaging brightness of the image acquisition device back to the illumination reference range. When the vibration amplitude parameter in the environmental parameters exceeds the vibration threshold of the valve area, the optical image stabilization mechanism of the image acquisition device is activated to compensate for image jitter during the acquisition process and obtain the adjusted optical parameters. Based on the adjusted optical parameters, when the dust concentration parameter exceeds the preset dust threshold, the front air curtain cleaning of the valve area is triggered, and the anti-interference original image of the valve area is acquired. The mechanical structure guided segmentation module is used to perform image analysis guided by mechanical structure analysis on the original, anti-interference image based on a preset 3D mechanical structure library of valves. It dynamically segments the region of interest (ROI) of the mechanical structure from the original, anti-interference image, including: Edge detection and contour extraction are performed on the original anti-interference image to obtain the contour boundary of the valve body in the original anti-interference image; The valve body type is identified by coarsely matching the outline boundary with the three-dimensional geometric model in the valve three-dimensional mechanical structure library. Based on the open-state mechanical structure template and the closed-state mechanical structure template, determine the estimated location region of the key mechanical component in the anti-interference original image; Using the boundary coordinates of the estimated location region as the cropping reference, coordinate constraint cropping is performed on the anti-interference original image to obtain the region of interest of the key mechanical component; The preset three-dimensional mechanical structure library for valves includes: Obtain three-dimensional geometric models of various valves in thermal power plants; The three-dimensional geometric model is structurally analyzed to extract the spatial positional relationship data of key mechanical components of various valves in the fully open and fully closed states; Based on the spatial positional relationship data, open-state mechanical structure templates and closed-state mechanical structure templates for various valves are established; The open-state mechanical structure template and the closed-state mechanical structure template are indexed and arranged to construct the valve three-dimensional mechanical structure library; The multidimensional feature extraction module is used to perform multidimensional image processing on the region of interest, extract and quantify the morphological and topological features of the current mechanical position in the valve region, and obtain the mechanical feature vector of the valve region. The fusion and discrimination module compares the mechanical feature vectors with a pre-stored standard state feature template library to obtain feature similarity values. These feature similarity values ​​are then weighted and fused to obtain the state confidence score of the valve region, including: Each feature dimension in the mechanical feature vector is compared with the corresponding standard feature dimension in the standard state feature template library one by one to obtain the feature similarity value of each feature dimension. Based on the image quality parameters of the region of interest, calculate the feature confidence parameter for each feature dimension; Based on the feature similarity values ​​and the feature confidence parameters, a covariance matrix between feature dimensions is established; Based on the feature similarity value, the feature confidence parameter, and the covariance matrix, the state fusion confidence of the valve region is calculated using a weighted fusion calculation formula. The weighted fusion calculation formula is as follows: ； in, The state fusion confidence level of the valve region; The total number of feature dimensions in the mechanical feature vector. For the first Feature similarity values ​​for each feature dimension For the first Feature credibility parameters for each feature dimension The covariance matrix between the feature dimensions is... The inverse matrix of the covariance matrix is ​​the nth... Line 1 Column elements, The inverse matrix of the covariance matrix is ​​the nth... The sum of the absolute values ​​of all elements in the row. As a regularization stabilization factor; The status output and verification module is used to output the valve area's on / off status identification result when the status confidence exceeds a preset confidence threshold; when the status confidence does not exceed the confidence threshold, the valve area is marked as having an uncertain status and a manual verification prompt is triggered.

Images