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

Abstract

The invention discloses a hydropower station equipment state AI video image recognition and intelligent monitoring analysis method, relates to the technical field of hydropower station equipment monitoring, and solves the problem that whether a corresponding part has aging, scars and other abnormal problems or not is evaluated by performing specific analysis on the surface state of the part without adopting a video monitoring mode. The method comprises the following steps: determining the precision of a determined gradient region, selecting feature points on the same transverse horizontal line, determining a feature center line based on the selected feature points, determining a selected point through a left and right sequence feature difference value minimization principle, and determining a region center point of a corresponding gradient region based on the feature center line. Therefore, higher accuracy of the determined central point of the area can be effectively guaranteed, the accuracy of the subsequently confirmed feature vector is higher, the accuracy of the subsequently confirmed related abnormal feature is higher, and the specific accuracy of monitoring can be guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower station equipment monitoring, and specifically to an AI video image recognition and intelligent monitoring and analysis method for the status of hydropower station equipment. Background Art

[0002] Hydropower station equipment is the key foundation for a hydropower station to convert water energy into electrical energy and ensure the stable operation of power production. Equipment that converts the kinetic energy and potential energy of water into mechanical energy, common types include Francis, Kaplan, Pelton, etc.; Francis turbines are suitable for power stations with medium water heads and flows; Kaplan turbines are suitable for power stations with low water heads and large flows; Pelton turbines are used in power stations with high water heads and small flows.

[0003] The application with the publication number CN115640698A discloses a fault warning system for operating equipment in a hydropower station, specifically related to the technical field of hydropower station warning, including an intelligent monitoring module, an equipment fault analysis module, a cloud service module, a fault assessment module, a daily maintenance module, and an emergency treatment plan module. The intelligent monitoring module is connected to various sensors of the hydropower station to obtain and monitor the operation status data of each operating equipment in real time. The equipment fault analysis module models and analyzes the operation data of the operating equipment through a fault similarity model, and transmits the analysis results to the fault assessment module. The cloud service module obtains expert knowledge and big data features in the cloud database to provide algorithm support for the process of the equipment fault analysis module, and at the same time stores, self-updates data and algorithms. The fault assessment module classifies the fault data for fault assessment to provide data support for solving various faults, improving the efficiency and accuracy of solving faults.

[0004] Regarding the relevant monitoring process of the status of hydropower station equipment, it generally evaluates whether there are abnormal behaviors of the corresponding components based on the specific operating parameters of the relevant components in the corresponding equipment. However, the original such method can only display signals when the components are abnormal, and serious accidents may occur in severe cases. In the actual processing process, the method of video monitoring is not used to specifically analyze the surface status of the components to evaluate whether there are abnormal problems such as aging and scars in the corresponding components. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides an AI video image recognition and intelligent monitoring and analysis method for the status of hydropower station equipment, which solves the problem that the method of video monitoring is not used to specifically analyze the surface status of the components to evaluate whether there are abnormal problems such as aging and scars in the corresponding components.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: The AI video image recognition and intelligent monitoring and analysis method for the status of hydropower station equipment includes the following steps:

[0007] Step 1: Conduct video monitoring on specified relevant components within the hydropower station equipment, and generate monitoring videos belonging to the corresponding relevant components in real time. Process specific frame images within the monitoring videos to confirm the associated gradient regions. The specific method is as follows:

[0008] S11. For the real-time generated monitoring video, based on the preset capture period set within the intelligent AI, the capture period is a preset period. At the end of each capture period, frame images are captured to lock the specific frame images associated with this capture period.

[0009] S12. Calibrate the pixel values associated with different pixel points within the specific frame image as X i-k , where i represents different pixel points, and k represents different relevant components being monitored. Confirm the gradient features associated with each group of pixel points. The specific method is as follows:

[0010] S121. Use the Sobel algorithm to confirm the horizontal gradient and vertical gradient associated with the corresponding pixel point, and calibrate the confirmed horizontal gradient as H i-k , and calibrate the confirmed vertical gradient as T i-k ;

[0011] S122. Use: to confirm the gradient feature Tz associated with the corresponding pixel point i-k ;

[0012] S13. Compare the gradient feature Tz i-k associated with different pixel points within the corresponding frame image with the preset value Y1. When Tz i-k > Y1, calibrate the corresponding pixel point as a gradient pixel point; otherwise, do not perform any calibration.

[0013] S14. Calibrate the several gradient pixel points that appear within the specific frame image in sequence, and calibrate the specific area covered between several adjacent gradient pixel points as the gradient region.

[0014] Step 2: Re-process the confirmed gradient region within the specific frame image. Based on the gradient features of different gradient pixel points in the same horizontal direction within the gradient region, select the gradient pixel points with relatively balanced features in the horizontal position as the selected points. Based on the determined several selected points, select the feature midline within the gradient region. The specific sub-steps are as follows:

[0015] S21. Take multiple gradient pixel points in the same horizontal direction within the gradient region as a pending set. Randomly select a single gradient pixel point from the pending set as the pending point. Execute a set of eigenvalue calculation processes to the left of the pending point, and calibrate the gradient feature associated with the pending point as DD i-k, and starting from the undetermined point, sort the gradient features associated with other gradient pixel points to the left, and confirm a set of left sorting sequences. The value associated with the first position in the left sorting sequence is DD i-k , starting from the first value, the difference of the adjacent values ​​is confirmed in sequence, and the difference = the previous group of values ​​- the next group of values. After processing several groups of differences associated with the left sorting sequence, the several groups of differences are summed up to confirm the total value, and the absolute value of the total value is used as the left sequence feature of the left sorting sequence;

[0016] Then, a set of eigenvalue calculation processes is performed from the undetermined point to the right. Starting from the undetermined point, the gradient features associated with other gradient pixel points are sorted to the right in turn to confirm a set of right sorting sequences, and the same processing method as the left sorting sequence is used to process the left sequence features to confirm the right sequence features of the right sorting sequence;

[0017] Perform difference processing on the left sequence feature and the right sequence feature, confirm the feature difference, and use the absolute value of the feature difference as the process feature of the pending point;

[0018] S22, selecting other single gradient pixel points from the pending set as pending points in turn, and using the same processing method as step S21 to confirm the process features associated with the corresponding pending points;

[0019] S23, based on different process characteristics associated with different pending points in the pending set, selecting the minimum process characteristic, and using the pending point associated with the minimum process characteristic as the selected point associated with the pending set;

[0020] S24, confirming the pending sets associated with different horizontal directions in the gradient region, and using the same processing method as steps S21-S23 to confirm the selected points associated with the corresponding pending sets;

[0021] Step 3: Identify whether the feature midline confirmed in 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. The center point of the closed loop is used as the midpoint of the gradient region. The specific method is as follows:

[0022] Combine this closed loop with the two-dimensional coordinate system, confirm the two-dimensional coordinates associated with different points on the closed loop from the two-dimensional coordinate system, then average the confirmed two-dimensional coordinates, lock a set of mean coordinates, and confirm the coordinate point associated with this mean coordinate in the two-dimensional coordinate system, use this coordinate point as the center point of this closed loop, and synchronously calibrate it to the inside of the gradient area as the midpoint of the area based on the position feature;

[0023] Step 4: Connect the midpoints of adjacent regions according to the midpoints of different regions calibrated in different gradient regions, and lock the multi-feature set line vector for this specific frame drawing. The specific method is as follows: Based on the midpoints of different regions associated with different gradient regions in this specific frame drawing, identify the straight-line distances between several midpoints of regions. Denote the two sets of midpoints of regions with the shortest straight-line distance as adjacent points, and make a straight-line connection between the adjacent points associated with each midpoint of a different region to generate feature connection lines between several midpoints of regions. All the generated feature connection lines are in a connected state, and use the multiple generated feature connection lines as the multi-feature set line vector for this specific frame drawing;

[0024] If a midpoint of a region has multiple adjacent points, then a straight-line connection is made between each existing adjacent point and this midpoint of the region;

[0025] Step 5: After determining the multi-feature set line vector of the current specific frame drawing, for the same group of monitored related components, confirm the multi-feature set line vector generated by this related component during 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 result, and output a monitoring signal based on the check result. The specific method is as follows:

[0026] S51: Based on the label k, confirm the same related components, confirm the multi-feature set line vector generated by the related components during the current interception period, and then confirm the multi-feature set line vector generated by the related components during the next interception period, and identify and confirm whether the multi-feature set line vectors generated in adjacent interception periods are consistent;

[0027] S511: Overlap 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 completely overlap. If they overlap, continue to monitor. If they do not overlap, control one set of multi-feature set line vectors to rotate based on the overlapping endpoints, and identify whether there is an overlapping process during the rotation:

[0028] If there is, continue to monitor;

[0029] If there is no such situation, perform endpoint swapping, identify whether there is an overlapping situation, and the identification process is the same as when endpoint swapping is not performed. If no overlap can be achieved in both cases, generate a component anomaly signal for display.

[0030] The present invention provides a method for AI video image recognition and intelligent monitoring and analysis of the state of hydropower station equipment. Compared with the prior art, it has the following beneficial effects:

[0031] The present invention effectively ensures higher accuracy of the determined regional center point by performing accuracy confirmation on the determined gradient region, selecting feature points on the same horizontal line, confirming the feature center line based on the selected feature points, determining the selected points based on the principle of minimizing the difference between the left and right sequence features, and determining the corresponding gradient region center point based on the feature center line. This enables higher accuracy of the subsequently confirmed feature vectors, higher accuracy of the subsequently confirmed relevant abnormal features, and ensures the specific accuracy of the monitored object.

[0032] Verify and compare the corresponding feature line vectors, identify whether the associated feature vectors are consistent, evaluate whether there are abnormal conditions such as aging and cracking of the corresponding abnormal parts based on the specific identification process, and confirm the display of the corresponding output monitoring signal based on the specific identification result. This can effectively confirm the component features based on video monitoring and output signals based on the component features, enabling better intelligent monitoring and analysis of the corresponding hydropower station equipment. Description of the Drawings

[0033] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0034] Figure 2 It is a schematic diagram of the determination of the gradient pixel points of the present invention. Detailed Embodiments

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0036] First Embodiment

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

[0038] Step 1: Conduct video monitoring on specified relevant components within the hydropower station equipment, and generate monitoring videos belonging to the corresponding relevant components in real time. Process specific frame images in the monitoring videos to confirm the associated gradient regions. Install high-definition cameras at various key equipment positions in the hydropower station (such as generators, transformers, turbines, gates, etc.) to collect video image data during equipment operation. Select appropriate industrial-grade cameras according to the environmental characteristics and monitoring requirements of the hydropower station equipment, such as cameras with high resolution, low illumination, wide dynamic range, etc.; install the cameras at positions where the specified components can be clearly photographed to ensure comprehensive monitoring coverage. The installation process of the cameras combines the practical experience of relevant operators. Different types of components have different image features, but the processing methods of the image features are the same, so the same method can be used to analyze the change status between the frame images of the corresponding components to evaluate whether there are abnormal problems with the corresponding components. The method for confirming the gradient region is as follows:

[0039] S11. Combine Figure 2 , for the real-time generated monitoring video, based on the interception period set within the intelligent AI, the interception period is a preset period, and the preset period generally takes a value of 10 minutes. Intercept frame images at the end of each interception period to lock the specific frame images associated with this interception period;

[0040] S12. Calibrate the pixel values associated with different pixel points in the specific frame image as X i-k , where i represents different pixel points, and k represents different relevant components being monitored. Confirm the gradient features associated with each group of pixel points:

[0041] S121. When the pixel points are arranged, they belong to the grid sorting method, that is, there are eight pixel points around each pixel point. Use the Sobel algorithm to confirm the horizontal gradient and vertical gradient associated with the corresponding pixel point, and calibrate the confirmed horizontal gradient as H i-k , and calibrate the confirmed vertical gradient as T i-k . In the Sobel algorithm, when calculating the horizontal gradient and vertical gradient, there are different weight factors. Take this pixel point as the middle point and extract the other eight surrounding pixel points as matching points to generate nine groups of pixel point sets. Based on the different weight factors associated with different set positions, confirm the numerical features associated with the corresponding set positions, and then sum the numerical features associated with the corresponding pixel point sets to lock the horizontal gradient and vertical gradient associated with the corresponding pixel point. Since the method of using the Sobel algorithm to obtain the numerical gradient of pixel points is relatively common in the prior art, it will not be elaborated here;

[0042] S122. Use: to confirm the gradient feature Tz associated with the corresponding pixel pointi-k ;

[0043] S13. Compare the gradient feature Tz associated with different pixel points within the corresponding intra-frame with a preset value Y1. The specific value of Y1 is determined by the operator based on experience. When Tz i-k > Y1, mark the corresponding pixel points as gradient pixel points; otherwise, do not perform any marking. The preset value Y1 has different values for different monitored components. For a water turbine, it is generally appropriate to take values between 50 and 80. For a generator, it is generally more appropriate to take values between 60 and 90. If the corresponding environment has high humidity and dust, then the value of Y1 needs to be reduced, and it is generally more appropriate to take values between 30 and 50. Therefore, its specific value is specifically determined by the personnel according to the corresponding component and the working environment; i-k > Y1, mark the corresponding pixel points as gradient pixel points; otherwise, do not perform any marking. The preset value Y1 has different values for different monitored components. For a water turbine, it is generally appropriate to take values between 50 and 80. For a generator, it is generally more appropriate to take values between 60 and 90. If the corresponding environment has high humidity and dust, then the value of Y1 needs to be reduced, and it is generally more appropriate to take values between 30 and 50. Therefore, its specific value is specifically determined by the personnel according to the corresponding component and the working environment;

[0044] S14. Mark the several gradient pixel points that appear within the specific intra-frame in sequence, and mark the specific area covered between several adjacent gradient pixel points as the gradient area (the gradient pixel points do not appear in single groups and generally appear continuously because the contour feature of the corresponding component is not a single point but a contour area, assuming a high-definition camera with strong recognition clarity);

[0045] Step 2. Re-process the gradient area confirmed within the specific intra-frame. Based on the gradient features of different gradient pixel points in the same horizontal direction within the gradient area, select the gradient pixel points with relatively balanced features in the horizontal position as the selected points. Based on the determined several selected points, select the feature midline within the gradient area. Specifically, assume that in a horizontal position of a gradient area, the gradient points associated within the gradient area change regularly. In such a regular change situation, in order to ensure more accurate subsequent monitoring accuracy, it is necessary to lock the specific midline again within the fine gradient area to ensure the subsequent monitoring accuracy. The specific sub-steps for selecting the feature midline are as follows:

[0046] S21. Since several pixel points are in a grid sorting state, the pixel points will not be disordered. Therefore, the same pixel points in the horizontal direction can be directly confirmed. Consider multiple gradient pixel points in the same horizontal direction within the gradient area as a pending set, and randomly select a single gradient pixel point from the pending set as the pending point. Execute a set of eigenvalue calculation processes to the left from the pending point, and mark the gradient feature associated with the pending point as DD i-k , and starting from this pending point, sort the gradient features associated with other gradient pixel points to the left in sequence to confirm a set of left sorting sequences. The value associated with the first position in the left sorting sequence is DD i-k, starting from the first value, the difference of the adjacent values ​​is confirmed in sequence, and the difference = the previous group of values ​​- the next group of values ​​(the previous group of values ​​here is the adjacent values ​​of the next group of values). After processing several groups of differences associated with the left sorting sequence, the several groups of differences are summed up to confirm the total value, and the absolute value of the total value is used as the left sequence feature of the left sorting sequence;

[0047] Then, a set of eigenvalue calculation processes is performed from the undetermined point to the right. Starting from the undetermined point, the gradient features associated with other gradient pixel points are sorted to the right in turn to confirm a set of right sorting sequences, and the same processing method as the left sorting sequence is used to process the left sequence features to confirm the right sequence features of the right sorting sequence;

[0048] The left sequence features and the right sequence features are subjected to difference processing, the feature difference is confirmed, and the absolute value of the feature difference is used as the process feature of this pending point. For example: it is proposed that there are five gradient pixel points in a horizontal direction, then the confirmed pending set is {A, B, C, D, E}, and the gradient features associated with each different gradient pixel point are: 10, 20, 25, 40, 60, respectively. When B is selected as the pending point, the left sorting sequence is: 25, 20, 10, and the right sorting sequence is 25, 40, 60. Then there are two sets of differences in the left sorting sequence, which are 5 and 10 respectively, and the resulting left sequence feature is 15. There are also two sets of differences in the left sorting sequence, which are -15 and -20 respectively, and the right sequence feature is 35. Then when B is the pending point, the process feature associated with it is 20 (the absolute value of 35-15). When D is selected as the pending point, the associated process feature is 10, so it is best to select D as the selected point;

[0049] S22, selecting other single gradient pixel points from the pending set as pending points in turn, and using the same processing method as step S21 to confirm the process features associated with the corresponding pending points;

[0050] S23, based on different process characteristics associated with different pending points in the pending set, selecting the minimum process characteristic, and using the pending point associated with the minimum process characteristic as the selected point associated with the pending set;

[0051] S24, confirming the pending sets associated with different horizontal directions (each horizontal direction is parallel, that is, different horizontal directions parallel to each other) in the gradient region, and using the same processing method as steps S21-S23 to confirm the selected points associated with the corresponding pending sets;

[0052] S25. Connect the selected points associated with adjacent horizontal directions in sequence, and confirm the characteristic midline associated with this gradient region. There is a selected point within the points associated with each horizontal region. After sequentially confirming a number of selected points, the selected points on the corresponding adjacent horizontal lines can be confirmed therefrom, and the corresponding selected points are connected to confirm the characteristic midline associated with the corresponding gradient region;

[0053] Step 3. Identify whether the characteristic midline confirmed within the gradient region is a closed loop. If it is a closed loop, lock the center point of this closed loop. If it is not a closed loop, lock the endpoints, connect the endpoints to form a closed loop, and lock the center point of this closed loop. Take the center point of the closed loop as the region midpoint of this gradient region. The specific determination method is as follows:

[0054] S31. Confirm whether there are endpoints on the characteristic midline. If there are endpoints, it means it does not belong to a closed loop, then directly connect the endpoints at both ends of this characteristic midline to form a closed loop. If there are no endpoints, it means it belongs to a closed loop;

[0055] S32. Combine this closed loop with the two-dimensional coordinate system (it 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 ring from the two-dimensional coordinate system, then perform mean processing on the confirmed several two-dimensional coordinates, lock a set of mean coordinates, and confirm the coordinate point associated with this mean coordinate within the two-dimensional coordinate system. Take this coordinate point as the center point of this closed loop, and synchronously calibrate it to the inside of the gradient region based on the position characteristics as the region midpoint. Specifically, here, the method of confirming the center point is adopted to confirm the midpoint of the corresponding gradient region. The so-called position characteristics mean that the center point within the closed loop has different position characteristics from different points. Since the closed loop is located within the gradient region, the region midpoint of the gradient region is directly confirmed;

[0056] Step 4. Connect the adjacent region midpoints according to the different region midpoints calibrated within different gradient regions, and lock the multi-feature set line vector regarding this specific frame drawing. The specific locking method is as follows:

[0057] Based on the different region midpoints associated with different gradient regions within this specific frame drawing, identify the straight-line distances between several region midpoints. Denote the two sets of region midpoints with the shortest straight-line distance as adjacent points (that is, point 1 and point 2, point 2 is the adjacent point of point 1, and point 1 is the adjacent point of point 2), and directly connect the adjacent points associated with each different region midpoint to generate the characteristic connection lines between several region midpoints. All the generated characteristic connection lines are in a connected state, and take the generated multiple characteristic connection lines as the multi-feature set line vector of this specific frame drawing;

[0058] If there are multiple adjacent points in a certain area (that is, the straight-line distances are all in the minimum state, but there are multiple such points, generally only two), then each existing adjacent point is connected to the midpoint of this area by a straight line;

[0059] Step 5: After determining the multi-feature set line vectors of the current specific frame drawing, for the same group of monitored related components, confirm the multi-feature set line vectors generated by this related component during the interception period, compare the multi-feature set line vectors generated by the same group of related components in adjacent interception periods, evaluate the comparison results, and output a monitoring signal based on the comparison results. The specific method of comparison is as follows:

[0060] S51: Based on the label k, confirm the same related components, confirm the multi-feature set line vectors generated by the related components during the current interception period, and then confirm the multi-feature set line vectors generated by the related components in the next interception period, and identify and confirm whether the multi-feature set line vectors generated in adjacent interception periods are consistent;

[0061] S511: Overlap the endpoints of the two groups of multi-feature set line vectors, identify and confirm whether the two groups of multi-feature set line vectors completely overlap. If they overlap, continue to monitor. If they do not overlap, control one group of multi-feature set line vectors to rotate according to the overlapping endpoints, and identify whether there is an overlapping process during the rotation. If there is, continue to monitor;

[0062] S512: If not, perform endpoint swapping (overlap the other endpoint of a certain group of multi-feature set line vectors with the endpoint of the current other group of multi-feature set line vectors, that is, swap the directions), and identify whether there is an overlapping situation. The identification process is the same as in step S511. If they still do not overlap, generate a component abnormality signal for display;

[0063] Specifically, some components are in a rotating state. Then, in different rotating states, the generated multi-feature set line vectors are different. Then, the rotation method can be used to specifically compare the overlapping situations to evaluate whether there are specific changes in the corresponding feature line vectors, so as to determine whether such components are aging, cracked, etc., and perform specific signal display.

[0064] Some of the data in the above formula are numerically calculated after removing the dimensions, and the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

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

Claims

1. AI video image recognition and intelligent monitoring and analysis method for hydropower station equipment status, characterized in that: The following steps are involved: Step 1: Perform video monitoring on the designated relevant components in the hydropower station equipment, and generate monitoring videos belonging to the corresponding relevant components in real time, process specific frames in the monitoring video, and confirm the gradient area associated with the corresponding frames; Step 2: reprocess the gradient region confirmed in the specific frame, select the gradient pixel points with corresponding balanced features in the horizontal position as selected points based on the gradient features of different gradient pixel points in the same horizontal direction in the gradient region, and select the feature center line in the gradient region based on the determined several selected points; Step 3: Identify whether the feature midline confirmed in 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. The center point of the closed loop is used as the regional midpoint of the gradient region. Step 4: Connect the midpoints of adjacent regions according to the midpoints of different regions marked in different gradient regions, and lock the multi-feature set line vectors about this specific frame; Step 5. After the multi-feature set line vector of the current specific frame is determined, for the same group of monitored related components, the multi-feature set line vector generated by this related component in the interception period is confirmed, and the multi-feature set line vectors generated by the same group of related components in adjacent interception periods are checked, the check results are evaluated, and the monitoring signal is output 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 is characterized in that: In the step 1, the specific method of confirming the gradient region is: S11. For the surveillance video generated in real time, based on the capture period set in the intelligent AI, the capture period is a preset period, and frame capture is performed at the end of each group of capture periods, and the specific frame associated with the current capture period is locked; S12, the pixel values ​​associated with different pixel points in a specific frame are calibrated as X i-k , where i represents different pixels, and k represents different related components monitored, and the gradient features associated with each group of pixels are confirmed; S13, the gradient features Tz associated with different pixels in the corresponding frame i-k Check with the preset value Y1, when Tz i-k When >Y1, the corresponding pixel point is calibrated as a gradient pixel point, otherwise, no calibration is performed; S14, calibrating a number of gradient pixel points appearing in a specific frame in sequence, and calibrating a specific area covered by a number of adjacent gradient pixel points as a gradient area.

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

4. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1 is characterized in that: In step 2, the specific sub-steps for determining the selected point are: S21, multiple gradient pixel points in the same horizontal direction in the gradient region are regarded as a pending set, a single gradient pixel point is randomly selected from the pending set as a pending point, a set of eigenvalue calculation processes are performed to the left of the pending point, and the gradient feature associated with the pending point is calibrated as DD i-k , and starting from the undetermined point, sort the gradient features associated with other gradient pixel points to the left, and confirm a set of left sorting sequences. The value associated with the first position in the left sorting sequence is DD i-k , starting from the first value, the difference of the adjacent values ​​is confirmed in sequence, and the difference = the previous group of values ​​- the next group of values. After processing several groups of differences associated with the left sorting sequence, the several groups of differences are summed up to confirm the total value, and the absolute value of the total value is used as the left sequence feature of the left sorting sequence; Then, a set of eigenvalue calculation processes is performed from the undetermined point to the right. Starting from the undetermined point, the gradient features associated with other gradient pixel points are sorted to the right in turn to confirm a set of right sorting sequences, and the same processing method as the left sorting sequence is used to process the left sequence features to confirm the right sequence features of the right sorting sequence; Perform difference processing on the left sequence feature and the right sequence feature, confirm the feature difference, and use the absolute value of the feature difference as the process feature of the pending point; S22, selecting other single gradient pixel points from the pending set as pending points in turn, and using the same processing method as step S21 to confirm the process features associated with the corresponding pending points; S23, based on different process characteristics associated with different pending points in the pending set, selecting the minimum process characteristic, and using the pending point associated with the minimum process characteristic as the selected point associated with the pending set; S24, confirming the pending sets associated with different horizontal directions in the gradient region, and using the same processing method as steps S21-S23 to confirm the selected points associated with the corresponding pending sets.

5. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 4 is characterized in that: In the step 2, the selected points associated with adjacent horizontal directions are connected in sequence to confirm the characteristic center line associated with the gradient region.

6. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1 is characterized in that: In the step 3, if the feature midline has endpoints, it means that it does not belong to a closed loop. The endpoints at both ends of the feature midline are connected by straight lines to generate 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 is characterized in that: In the step 3, if the feature midline has no endpoints, it means that it belongs to a closed loop.

8. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1 is characterized in that: In step 3, the specific method of confirming the midpoint of the gradient region is: Combine this closed loop with the two-dimensional coordinate system, confirm the two-dimensional coordinates associated with different points on the closed loop from the two-dimensional coordinate system, then average the confirmed two-dimensional coordinates, lock a set of mean coordinates, and confirm the coordinate point associated with this mean coordinate in the two-dimensional coordinate system, use this coordinate point as the center point of this closed loop, and synchronously calibrate it to the inside of the gradient area as the midpoint of the area based on the position feature.

9. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1 is characterized in that: In step 4, the specific method of locking the multi-feature set line vector in the feature frame is: According to the different regional midpoints associated with different gradient regions in this specific frame, the straight-line distances between the midpoints of several regions are identified, the two groups of regional midpoints with the shortest straight-line distances are recorded as adjacent points, and the adjacent points associated with the midpoints of each different region are connected by straight lines to generate feature lines between the midpoints of several regions, and the several feature lines are all in a connected state, and the generated multiple feature lines are used as the multi-feature set line vector of this specific frame; If there are multiple adjacent points to a midpoint in a region, each adjacent point is connected to the midpoint in the region by a straight line.

10. The method for AI video image recognition and intelligent monitoring and analysis of hydropower station equipment status according to claim 1 is characterized in that: In step 5, the specific method of checking the multi-feature set line vector is: S51, based on the mark k, confirm the same related components, confirm the multi-feature set line vectors generated by the related components in the current interception cycle, and then confirm the multi-feature set line vectors generated by the related components in the next interception cycle, and identify and confirm whether the multi-feature set line vectors generated in adjacent interception cycles are consistent; S511, overlap the endpoints of two groups of multi-feature set line vectors, identify and confirm whether the two groups of multi-feature set line vectors completely overlap, if overlap, continue to monitor, if not overlap, control one group of multi-feature set line vectors to rotate according to the overlapped endpoints, and identify whether there is overlap during the rotation process: If present, continue monitoring; If not, the endpoints are swapped to identify whether there is overlap. The identification process is the same as when no endpoint swap is performed. If there is no overlap, a component abnormality signal is generated for display.

Images