Hydropower station equipment state ai video image recognition and intelligent monitoring analysis method

Abstract

The application discloses a hydropower station equipment state AI video image recognition and intelligent monitoring analysis method and relates to the technical field of hydropower station equipment monitoring.The application solves the problem that the surface state of components is analyzed in detail by adopting a video monitoring mode to evaluate whether the corresponding components have abnormal problems such as aging and scratches.The application confirms the accuracy of the determined gradient area, selects feature points on the same horizontal line, confirms the feature center line based on the selected feature points, determines the selected points by the left-right sequence feature difference minimization principle, and determines the corresponding gradient area center point based on the feature center line, so that the accuracy of the determined area center point is higher, the accuracy of the subsequent confirmed feature vector is higher, the accuracy of the subsequent confirmed related abnormal features is higher, and the specific accuracy of the monitored components can be ensured.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of water power station equipment monitoring, in particular to an AI video image recognition and intelligent monitoring analysis method for water power station equipment states. BACKGROUND

[0002] Water power station equipment is a key basis for realizing water energy conversion into electric energy and guaranteeing stable operation of electric power production, and is equipment for converting kinetic energy and potential energy of water into mechanical energy, common types of which include mixed-flow type, axial-flow type and impulse type; the mixed-flow type water turbine is suitable for power stations with medium water head and flow; the axial-flow type water turbine is suitable for power stations with low water head and large flow; and the impulse type water turbine is used for power stations with high water head and small flow.

[0003] An application with the publication number CN115640698A discloses a water power station operation equipment fault early warning system, and particularly relates to the technical field of water power station early warning, which comprises an intelligent monitoring module, an equipment fault analysis module, a cloud service module, a fault evaluation module, a daily maintenance module and an emergency treatment plan module, the intelligent monitoring module is connected with various sensors of the water power station, and real-time acquisition and monitoring of the running state data of various operation equipment are realized, the equipment fault analysis module models and analyzes the operation data of the operation equipment through a fault similarity model, and transmits the analysis result to the fault evaluation module, the cloud service module acquires expert knowledge and big data features in the cloud database, provides algorithm support for the process of the equipment fault analysis module, and simultaneously stores and self-updates data and algorithms, the fault data is classified through the fault evaluation module, data support is provided for solving various faults, and the efficiency and accuracy of solving faults are improved.

[0004] For the related monitoring process of the water power station equipment state, whether the corresponding component has abnormal behavior is generally evaluated based on specific operation parameters of the related components in the corresponding equipment, but the original such mode can only exhibit signals when the component is abnormal, and a large accident may occur in serious cases, and in the actual processing process, the surface state of the component is not analyzed in detail in a video monitoring manner to evaluate whether the corresponding component has abnormal problems such as aging and scratches. SUMMARY

[0005] In view of the deficiencies of the prior art, the AI video image recognition and intelligent monitoring analysis method for water power station equipment states is provided, which solves the problem that the surface state of the component is not analyzed in detail in a video monitoring manner to evaluate whether the corresponding component has abnormal problems such as aging and scratches.

[0006] To achieve the above purpose, the following technical scheme is adopted: the AI video image recognition and intelligent monitoring analysis method for water power station equipment states comprises the following steps:

[0007] Step one, video monitoring of designated relevant components in hydropower station equipment, and real-time generation of monitoring video belonging to corresponding relevant components, processing of specific frames in the monitoring video, confirming the gradient area associated in the corresponding frame, the specific way is:

[0008] S11, for the real-time generated monitoring video, based on the set interception period in intelligent AI, the interception period is the preset period, at the end of each interception period, the specific frame is intercepted, and the specific frame associated with the interception period is locked;

[0009] S12, the pixel value associated with different pixel points in the specific frame is marked as X i-k , where i represents different pixel points, and k represents different relevant components monitored, and the gradient characteristics associated with each group of pixel points are confirmed, and the specific way is:

[0010] S121, the horizontal gradient and vertical gradient associated with the corresponding pixel point are confirmed by using Sobe l algorithm, and the confirmed horizontal gradient is marked as H i-k , and the confirmed vertical gradient is marked as T i-k ;

[0011] S122, using: confirm the gradient characteristics Tz i-k associated with the corresponding pixel point;

[0012] S13, the gradient characteristics Tz i-k associated with different pixel points in the corresponding frame are checked with the preset value Y1, when Tz i-k > Y1, the corresponding pixel point is marked as gradient pixel point, otherwise, no marking is performed;

[0013] S14, a plurality of gradient pixel points appearing in the specific frame are sequentially marked, and the specific area covered between a plurality of adjacent gradient pixel points is marked as gradient area;

[0014] Step two, reprocessing the gradient area confirmed in the specific frame, based on the gradient characteristics of different gradient pixel points in the gradient area in the same horizontal direction, selecting the gradient pixel points with relatively balanced characteristics in the horizontal position as selected points, based on the determined selected points, selecting the characteristic median line in the gradient area, the specific sub-step is:

[0015] S21, a plurality of gradient pixel points in the gradient area in the same horizontal direction are taken as a pending set, a single gradient pixel point is randomly selected from the pending set as a pending point, a group of characteristic value calculation processes is executed to the left of the pending point, and the gradient characteristics associated with the pending point are marked as DD i-kand from the to-be-determined point, the gradient features associated with other gradient pixel points are sequentially sorted to the left to confirm a left sorting sequence, and the value associated with the first position in the left sorting sequence is DD i-k From the first value, the difference between adjacent values is confirmed sequentially, and the difference = previous group of values - next group of values. After the difference between several groups of values associated with the left sorting sequence is processed, the sum of the several groups of values is summed up to confirm the sum value, and the absolute value of the sum value is taken as the left sequence feature of the left sorting sequence;

[0016] Then, a group of feature value calculation processes are performed to the right of the to-be-determined point. From the to-be-determined point, the gradient features associated with other gradient pixel points are sequentially sorted to the right to confirm a right sorting sequence, and the same processing method as the left sorting sequence processing left sequence feature is used to confirm the right sequence feature of the right sorting sequence;

[0017] The difference between the left sequence feature and the right sequence feature is processed to confirm the feature difference, and the absolute value of the feature difference is taken as the process feature of the to-be-determined point;

[0018] S22, other single gradient pixel points in the to-be-determined set are sequentially selected as to-be-determined points, and the same processing method as step S21 is used to confirm the process feature associated with the corresponding to-be-determined point;

[0019] S23, based on different process features associated with different to-be-determined points in the to-be-determined set, the minimum process feature is selected, and the to-be-determined point associated with the minimum process feature is taken as the selected point associated with the to-be-determined set;

[0020] S24, the to-be-determined sets associated with different horizontal directions in the gradient region are confirmed, and the same processing method as steps S21-S23 is used to confirm the selected points associated with the corresponding to-be-determined set;

[0021] Step three, whether the feature line identified in the gradient region is a closed loop is identified. If it is a closed loop, the center point of the closed loop is locked. If it is not a closed loop, the end point is locked, the end points are connected to generate a closed loop, and the center point of the closed loop is locked. The center point of the closed loop is taken as the region center point of the gradient region. The specific method is:

[0022] The closed loop is combined with the two-dimensional coordinate system, the two-dimensional coordinates associated with different point positions on the closed loop ring in the two-dimensional coordinate system are confirmed, the several two-dimensional coordinates confirmed are processed by averaging to lock a group of average coordinates, and the coordinate point associated with the average coordinates in the two-dimensional coordinate system is confirmed. The coordinate point is taken as the center point of the closed loop, and is simultaneously calibrated to the inside of the gradient region based on the position feature as the region center point;

[0023] Step four, according to the different region midpoint marked in different gradient regions, the adjacent region midpoints are connected, and the multi-feature set line vector of the specific frame is locked, and the specific method is as follows: according to the different region midpoints associated with the different gradient regions in the specific frame, the straight line distance of several region midpoints is identified, the two groups of region midpoints with the shortest straight line distance are recorded as adjacent points, and the adjacent points associated with each different region midpoint are connected in a straight line to generate a feature connection line between several region midpoints, and the generated multi-feature connection line is used as the multi-feature set line vector of the specific frame;

[0024] If there are multiple adjacent points for a region midpoint, each adjacent point is connected to the region midpoint in a straight line;

[0025] Step five, after the multi-feature set line vector of the current specific frame is determined, the multi-feature set line vector generated by the related components in the same group in the monitoring period is confirmed, the multi-feature set line vectors generated by the related components in the adjacent monitoring period are checked, the checking result is evaluated, and the monitoring signal is output based on the checking result, and the specific method is as follows:

[0026] S51, based on the mark k, the same related components are confirmed, the multi-feature set line vector generated by the related components in the current monitoring period is confirmed, and the multi-feature set line vector generated by the related components in the next monitoring period is confirmed, and it is identified whether the multi-feature set line vectors generated in the adjacent monitoring period are consistent;

[0027] S511, the endpoints of the two groups of multi-feature set line vectors are coincided, and it is identified whether the two groups of multi-feature set line vectors are completely coincided, if coincided, the monitoring is continued, if not coincided, the group of multi-feature set line vectors is rotated according to the coincided endpoints, and it is identified whether there is a coincidence process in the rotating process:

[0028] If there is, the monitoring is continued;

[0029] If not, the endpoints are exchanged, and it is identified whether there is a coincidence, and the identification process is the same as that when the endpoints are not exchanged, if not, a component abnormal signal is generated for display.

[0030] The application provides a water power station equipment state AI video image recognition and intelligent monitoring analysis method. Compared with the prior art, the following beneficial effects are achieved:

[0031] The application can effectively guarantee that the determined region center point is more accurate, the subsequent confirmed feature vector is more accurate, the subsequent confirmed related abnormal feature is more accurate, and the specific accuracy of the monitored can be guaranteed.

[0032] The corresponding feature line vector is verified and compared, whether the associated feature vector is consistent is identified, based on the specific identification process, whether the corresponding abnormal part exists aging, cracking and other abnormal conditions is evaluated, based on the specific identification result, the corresponding output monitoring signal is confirmed to be displayed, the component feature can be effectively confirmed based on the video monitoring, and the signal output is performed based on the component feature, so that the hydropower station equipment achieves better intelligent monitoring and analysis effect. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The figure is a schematic diagram of the method of the application.

[0034] Figure 2 The figure is a schematic diagram of the determination of the gradient pixel point. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0036] First embodiment

[0037] Please refer to Figure 1 The application provides a hydropower station equipment state AI video image recognition and intelligent monitoring and analysis method, including the following steps:

[0038] Step one, video monitoring of designated related components in hydropower station equipment, and real-time generation of monitoring video belonging to corresponding related components, processing specific frames in monitoring video, confirming the gradient area associated in the corresponding frame, installing high-definition cameras at each key equipment location in the hydropower station (such as generators, transformers, water turbines, gates, etc.), collecting video image data during equipment operation, selecting appropriate industrial cameras based on the environmental characteristics and monitoring needs of hydropower station equipment, such as cameras with high resolution, low illumination, wide dynamic range, etc.; Install the camera in a position that can clearly capture the specified component to ensure comprehensive coverage of the monitoring field. The camera placement process combines the practical experience of relevant operators. Different types of components have different image characteristics, but the processing method of image characteristics is consistent, so the same method can be used to analyze the change state between the corresponding component frames to assess whether the corresponding component has any abnormal problems. The confirmation method of the gradient area is:

[0039] S11, combined Figure 2 For real-time generated monitoring video, based on the interception period set in the intelligent AI, the interception period is the preset period, which is generally 10 min. Frame capture is performed at the end of each interception period to lock the specific frame associated with this interception period.

[0040] S12, the pixel values associated with different pixel points in the specific frame are labeled as X i-k Where i represents different pixel points, and k represents different related components being monitored. Confirm the gradient characteristics associated with each group of pixel points:

[0041] S121, the pixel points are arranged in a grid ordering method, that is, there are eight pixel points around each pixel point. The Sobel algorithm is used to confirm the horizontal gradient and vertical gradient associated with the corresponding pixel point, and the confirmed horizontal gradient is labeled as H i-k The confirmed vertical gradient is labeled as T i-k In the Sobel algorithm, different weight factors are used when calculating the horizontal gradient and the vertical gradient. This pixel point is used as the middle point, and the other eight surrounding pixel points are used as the adaptive points to generate nine pixel point sets. Based on the different weight factors associated with different set positions, the numerical characteristics associated with the corresponding set position are confirmed, and then the numerical characteristics associated with the corresponding pixel point set are summed to lock the horizontal gradient and vertical gradient associated with the corresponding pixel point. Because the Sobel algorithm is commonly used in the prior art to calculate the numerical gradient of the pixel point, further description is not provided here.

[0042] S122, using: Confirm the gradient characteristics Tz associated with the corresponding pixel pointi-k ;

[0043] S13. Associating the gradient features Tz of different pixels within the corresponding frame. i-k The value of Y1 is compared with the preset value Y1, where the specific value of Y1 is determined by the operator based on experience. When Tz i-k When the value is greater than Y1, the corresponding pixel is marked as a gradient pixel; otherwise, no marking is performed. The preset value Y1 varies depending on the monitored component. For a water turbine, the value is generally between 50 and 80, and for a generator, the value is generally between 60 and 90. If the environment has high humidity and dust, the corresponding Y1 value needs to be reduced, and a value between 30 and 50 is generally more appropriate. Therefore, the specific value is determined by the personnel based on the corresponding component and working environment.

[0044] S14. Mark the gradient pixels that appear in a specific frame in sequence, and mark the specific area covered by the gradient pixels as the gradient region (the gradient pixels do not appear in a single group, but generally appear continuously, because the corresponding component contour features are not a point, but a contour region, under the premise of high-definition camera and strong recognition clarity).

[0045] Step 2: Reprocess the gradient regions identified within a specific frame. Based on the gradient features of different gradient pixels along the same horizontal direction within the gradient region, select gradient pixels with corresponding balanced features at horizontal positions as selection points. Based on these selection points, select the feature centerline within the gradient region. Specifically, at a horizontal position within a gradient region, the gradient points associated with that region exhibit a regular variation. To ensure more accurate subsequent monitoring, it is necessary to re-lock the corresponding centerline within the subtle gradient regions to guarantee monitoring accuracy. The specific sub-steps for selecting the feature centerline are as follows:

[0046] S21. Since several pixels belong to a grid-sorted state, the pixels will not be disordered. Therefore, we can directly identify the same pixels in the horizontal direction. Multiple gradient pixels within the gradient region located in the same horizontal direction are considered as a set to be determined. A single gradient pixel is randomly selected from this set as a point to be determined. A set of feature value calculations is then performed to the left of this point, and the gradient feature associated with the point is labeled as DD. i-k Starting from this undetermined point, sort the gradient features associated with other gradient pixels sequentially to the left, confirming a left-sorted sequence. The value associated with the first position in the left-sorted sequence is DD. i-kFrom the first value, the difference value of the adjacent values is confirmed in sequence, the difference value = the previous group of values - the next group of values (the previous group of values here is the adjacent value of the next group of values), after the associated several groups of difference values in the left sorting sequence are processed, the several groups of difference values are summed up, the sum value is confirmed, and the absolute value of the sum value is taken as the left sequence feature of the left sorting sequence;

[0047] Then a group of feature value calculation processes are performed to the right of the to-be-determined point. From the to-be-determined point, the gradient features associated with other gradient pixel points are sorted in sequence to the right, a right sorting sequence is confirmed, and the same processing mode of the left sorting sequence processing left sequence feature is adopted to confirm the right sequence feature of the right sorting sequence;

[0048] The left sequence feature and the right sequence feature are processed by difference, the feature difference value is confirmed, and the absolute value of the feature difference value is taken as the process feature of the to-be-determined point. For example, it is assumed that there are five gradient pixel points in a horizontal direction, and the to-be-determined set to be confirmed is {A, B, C, D, E}. The gradient features associated with each different gradient pixel point are 10, 20, 25, 40, and 60 respectively. When B is selected as the to-be-determined point, the left sorting sequence is 25, 20, and 10, and the right sorting sequence is 25, 40, and 60. Then there are two groups of difference values in the left sorting sequence, which are 5 and 10 respectively. The left sequence feature generated is 15, and there are also two groups of difference values in the left sorting sequence, which are -15 and -20 respectively. The right sequence feature is 35. Then the process feature associated with B as the to-be-determined point is 20 (the absolute value of 35-15), and the process feature associated with D as the to-be-determined point is 10. Therefore, D is selected as the selected point.

[0049] S22, other single gradient pixel points in the to-be-determined set are selected as to-be-determined points in sequence, and the process feature associated with the corresponding to-be-determined point is confirmed by adopting the same processing mode of step S21;

[0050] S23, based on different process features associated with different to-be-determined points in the to-be-determined set, the minimum process feature is selected, and the to-be-determined point associated with the minimum process feature is selected as the selected point associated with the to-be-determined set;

[0051] S24, the to-be-determined set associated with different horizontal directions (each horizontal direction is parallel, that is, different horizontal directions parallel to each other) in the gradient region is confirmed, and the selected point associated with the corresponding to-be-determined set is confirmed by adopting the same processing mode of steps S21-S23;

[0052] S25. Connect the selected points associated with adjacent horizontal directions in sequence to confirm the feature centerline associated with this gradient region. There is a selected point in each horizontal region. After confirming several selected points in sequence, the selected points on the corresponding adjacent horizontal lines can be confirmed. Connect the corresponding selected points to confirm the feature centerline associated with the corresponding gradient region.

[0053] Step 3: Identify whether the confirmed feature midline within the gradient region is a closed loop. If it is a closed loop, lock the center point of this loop. If it is not a closed loop, lock the endpoints, connect the endpoints to generate a closed loop, and lock the center point of this closed loop. The center point of the closed loop is taken as the region center of this gradient region. The specific method for determining this is as follows:

[0054] S31. Confirm whether the feature midline has endpoints. If endpoints exist, it means that it does not belong to a closed loop. Connect the endpoints at both ends of the feature midline with straight lines to generate a closed loop. If there are no endpoints, it means that it belongs to a closed loop.

[0055] S32. Combine this closed loop with a two-dimensional coordinate system (which can also be understood as constructing a two-dimensional coordinate system within the gradient region). Confirm the two-dimensional coordinates associated with different points on the closed loop within the two-dimensional coordinate system. Then, average the confirmed two-dimensional coordinates to lock a set of average coordinates. Confirm the coordinate points associated with these average coordinates within the two-dimensional coordinate system. Use these coordinate points as the center point of the closed loop and simultaneously calibrate them within the gradient region based on their positional characteristics as the region center. Specifically, the center point confirmation method is used here to confirm the midpoint of the corresponding gradient region. The so-called positional characteristics are that the center point within the closed loop has different positional characteristics from different points. Since the closed loop is located within the gradient region, the region center of the gradient region is directly confirmed.

[0056] Step 4: Based on the different region points marked in different gradient regions, connect the adjacent region points to lock the multi-feature set line vector for this specific frame. The specific locking method is as follows:

[0057] Based on the different regions associated with different gradient regions within this specific frame, the straight-line distance between several regions is identified. The two regions with the shortest straight-line distance are recorded as adjacent points (i.e., point 1 and point 2, where point 2 is an adjacent point of point 1 and point 1 is an adjacent point of point 2). The adjacent points associated with each region are connected by straight lines to generate feature lines between several regions. All feature lines are in a connected state. The generated feature lines are used as the multi-feature set line vector of this specific frame.

[0058] If there are multiple adjacent points (i.e. the straight-line distance is in the minimum state, but there are multiple such points, generally only two) in a certain area, each adjacent point is connected to the point in the area by a straight line.

[0059] Step five, after determining the multi-feature set line vector of the current specific frame, for the same group of monitored related components, confirming the multi-feature set line vector generated by the related components in the interception period, checking the multi-feature set line vector generated by the same group of related components in the adjacent interception period, evaluating the checking result, and outputting a monitoring signal based on the checking result, wherein the specific way of checking is:

[0060] S51, based on the mark k, confirming the same related components, confirming the multi-feature set line vector generated by the related components in the current interception period, and confirming the multi-feature set line vector generated by the related components in the next interception period, and identifying and confirming whether the multi-feature set line vectors generated in the adjacent interception period are consistent;

[0061] S511, the endpoints of the two groups of multi-feature set line vectors are coincided, and it is identified and confirmed whether the two groups of multi-feature set line vectors are completely coincided, if coincided, continue to monitor, if not coincided, control a group of multi-feature set line vectors to rotate according to the coincided endpoints, identify whether there is a coincidence process in the rotating process, if there is, continue to monitor;

[0062] S512, if not, perform endpoint exchange (coincide the other endpoint of a group of multi-feature set line vectors with the endpoint of the other group of multi-feature set line vectors, that is, exchange the direction), identify whether there is a coincidence, the identification process is consistent with step S511, if none of them can be coincided, generate a component abnormal signal for display;

[0063] Specifically, some components belong to a rotating state, so that the multi-feature set line vector generated in different rotating states is different, so that the rotating method can be used to specifically compare the coincidence to evaluate whether the corresponding feature line vector has a specific change, so as to determine whether such components have abnormal conditions such as aging, cracking, etc., and specific signal display.

[0064] Some data in the above formula are dimensionless numerical calculations, and the contents not described in detail in the specification are all prior art known to those skilled in the art.

[0065] The above examples are only used to illustrate the technical method of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present application.

Claims

1. A method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status, characterized in that, Includes the following steps: Step 1: Perform video monitoring on the designated components within the hydropower station equipment and generate real-time monitoring videos for the corresponding components. Process specific frames within the monitoring videos to identify the associated gradient regions within the corresponding frames. Step 2: Reprocess the gradient regions identified within a specific frame. Based on the gradient features of different gradient pixels along the same horizontal direction within the gradient region, select gradient pixels with corresponding balanced features at horizontal positions as selected points. Based on the determined selected points, select the feature midline within the gradient region. The specific sub-steps are as follows: S21. For multiple gradient pixels located in the same horizontal direction within the gradient region, form a set of undetermined points. Randomly select a single gradient pixel from the set of undetermined points as an undetermined point. Perform a set of feature value calculations to the left of the undetermined point, and label the gradient feature associated with the undetermined point as DD. i-k Starting from this undetermined point, sort the gradient features associated with other gradient pixels sequentially to the left, confirming a left-sorted sequence. The value associated with the first position in the left-sorted sequence is DD. i-k Starting from the first value, the difference between adjacent values ​​is confirmed sequentially. The difference is equal to the difference between the previous value and the next value. After processing several sets of differences associated with the left sorting sequence, the differences are summed to confirm the total value. The absolute value of the total value is used as the left sequence feature of the left sorting sequence. Then, starting from the undetermined point, perform a set of feature value calculation processes to the right. Starting from the undetermined point, sort the gradient features associated with other gradient pixels in turn to the right to confirm a set of right sorting sequences. Then, use the same processing method as the left sorting sequence to confirm the right sequence features of the right sorting sequence. The left sequence features and the right sequence features are processed by difference to confirm the feature difference. The absolute value of the feature difference is used as the process feature of this point to be determined. S22. Then, select other individual gradient pixels from the set of undetermined points as undetermined points, and use the same processing method as in step S21 to confirm the process characteristics associated with the corresponding undetermined points. S23. Based on the different process characteristics associated with different undetermined points in the undetermined set, select the smallest process characteristic, and take the undetermined point associated with the smallest process characteristic as the selected point associated with this undetermined set. S24. Next, confirm the sets of undetermined points associated with different horizontal directions within the gradient region, and use the same processing method as steps S21-S23 to confirm the selected points associated with the corresponding sets of undetermined points. Step 3: Identify whether the confirmed feature midline within the gradient region is a closed loop. If it is a closed loop, lock the center point of the closed loop. If it is not a closed loop, lock the endpoints, connect the endpoints to generate a closed loop, and lock the center point of the closed loop. Use the center point of the closed loop as the region center of this gradient region. Step 4: Based on the different region points marked in different gradient regions, connect the adjacent region points to lock the multi-feature set line vector for this specific frame. Step 5: After determining the multi-feature set line vectors of the current specific frame, for the same group of monitored related components, confirm the multi-feature set line vectors generated by this related component in the interception period, check the multi-feature set line vectors generated by the same group of related components in adjacent interception periods, evaluate the check results, and output the monitoring signal based on the check results.

2. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1, characterized in that, In step one, the specific method for confirming the gradient region is as follows: S11. For real-time generated monitoring videos, based on the capture period set in the intelligent AI, the capture period is a preset period. At the end of each capture period, frame capture is performed to lock the specific frame associated with this capture period. S12. Label the pixel values ​​associated with different pixels within a specific frame as X. i-k , where i represents different pixels and k represents different related components being monitored, and the gradient features associated with each group of pixels are confirmed; S13. Associating the gradient features Tz of different pixels within the corresponding frame. i-k Check against the preset value Y1, when Tz i-k When the value is greater than Y1, the corresponding pixel is marked as a gradient pixel; otherwise, no marking is performed. S14. The gradient pixels appearing in a specific frame are labeled sequentially, and the specific area covered between several adjacent gradient pixels is labeled as the gradient region.

3. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 2, characterized in that, In step S12, the specific method for confirming the gradient features associated with each group of pixels is as follows: S121. Use the Sobel algorithm to determine the horizontal and vertical gradients associated with the corresponding pixel, and label the determined horizontal gradient as H. i-k The confirmed vertical gradient is calibrated as T. i-k ; S122, Adopted: Confirm the gradient feature Tz associated with the corresponding pixel. i-k .

4. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1, characterized in that, In step two, the selected points associated with adjacent horizontal directions are connected sequentially to confirm the feature midline associated with this gradient region.

5. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1, characterized in that, In step three, if the feature midline has endpoints, it means that it does not belong to a closed loop. Connect the endpoints at both ends of the feature midline with a straight line to generate a closed loop.

6. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1, characterized in that, In step three, if the feature midline has no endpoints, it indicates that it belongs to a closed loop.

7. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1, characterized in that, In step three, the specific method for confirming the region center of this gradient region is as follows: Combine this closed loop with a two-dimensional coordinate system, identify the two-dimensional coordinates associated with different points on the closed loop within the two-dimensional coordinate system, then average the identified two-dimensional coordinates to lock a set of average coordinates, and identify the coordinate points associated with these average coordinates within the two-dimensional coordinate system. Use these coordinate points as the center point of this closed loop, and based on the positional characteristics, simultaneously mark the center point within the gradient region as the region center.

8. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1, characterized in that, In step four, the specific method for locking the line vectors of multiple feature sets within a specific frame is as follows: Based on the different regions associated with different gradient regions within this specific frame, identify the straight-line distance between several regions, record the two groups of regions with the shortest straight-line distance as adjacent points, and connect the adjacent points associated with each different region with a straight line to generate feature lines between several regions. All feature lines are in a connected state, and the generated multiple feature lines are used as the multi-feature set line vector of this specific frame. If a point in a region has multiple adjacent points, then each of the adjacent points is connected to the point in this region by a straight line.

9. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1, characterized in that, In step five, the specific method for verifying the line vectors of the multi-feature set is as follows: S51. Based on the marker k, identify the same related components, identify the multi-feature set line vectors generated by the related components in the current interception period, and then identify the multi-feature set line vectors generated by the related components in the next interception period. Identify and confirm whether the multi-feature set line vectors generated in adjacent interception periods are consistent. S511. Align the endpoints of the two sets of multi-feature set line vectors, identify and confirm whether the two sets of multi-feature set line vectors are completely aligned. If they are aligned, continue monitoring. If they are not aligned, control one set of multi-feature set line vectors to rotate according to the aligned endpoints, and identify whether there is a process of alignment during the rotation. If it exists, continue monitoring; If it does not exist, the endpoints are swapped to identify if there is overlap. The identification process is the same as when the endpoints are not swapped. If there is no overlap, a component abnormality signal is generated and displayed.

Images