A method for intelligently intercepting trendy videos at the backend and highly frequently identifying tidal level changes

Abstract

The present invention discloses a method for intelligently intercepting tide arrival videos and identifying tide level changes at high frequencies at the back end. The specific implementation scheme is as follows: real-time low-frequency image acquisition is performed on tide monitoring points, tide level identification is performed, and a low-frequency tide level curve is obtained; data conversion is performed on the low-frequency tide level curve to obtain low-frequency tide level difference data, and low-frequency tide level change type analysis is performed on the low-frequency tide level difference data in combination with tide surge forecast data and historical observation data experience, low-frequency abnormal data is cleaned, and the low-frequency tide surge arrival time is obtained; based on the low-frequency tide surge arrival time, a tide arrival video is obtained by replaying historical videos, a high-frequency tide level change curve is generated based on the tide arrival video, and the high-frequency tide arrival time and high-frequency tide rise time are calculated to generate a tide arrival identification video. The present invention adopts a post-feedback method to efficiently and accurately extract tide level change information from massive tide surge videos and identify high-frequency changes in tide level, thereby saving bandwidth and computing resources.

Description

Technical Field

[0001] The present invention relates to the technical field of tidal bore level monitoring technology, and in particular to a method for intelligently intercepting tide arrival videos and identifying tide level changes at a high frequency through a back-end. Background Art

[0002] Tides are one of Earth's most common natural phenomena, caused by the interaction of gravity and inertia. Tidal patterns can be inferred by measuring changes in ocean water levels. Tidal bores, a special form of tide, have attracted considerable attention for their immense energy and spectacular spectacle.

[0003] Traditional tidal bore monitoring methods rely primarily on contact-based measurement devices, such as float-type water level gauges and radar level gauges. While these devices offer the advantage of high measurement accuracy, they suffer from high installation costs and time constraints. Furthermore, they are limited in their suitability for tidal bore monitoring in complex environments. Furthermore, traditional equipment typically collects data at a low frequency, typically every two seconds. Due to the rapid speed of tidal bores, detailed tidal level information cannot be displayed. In recent years, advancements in science and technology, including video surveillance and computer vision, have made video-based tidal bore monitoring a promising new method. Furthermore, cameras are non-contact measurement methods characterized by ease of installation, low cost, and high frequency. Video recordings can typically reach over 25 frames per second, making them suitable for high-frequency tidal bore monitoring.

[0004] However, since tidal bore level change monitoring only requires identifying the short period of time when the tidal bore arrives, how to efficiently and accurately extract tidal level change information from massive tidal bore videos and identify high-frequency changes in tidal levels, thereby saving bandwidth and computing power resources, has become an urgent problem to be solved. Summary of the Invention

[0005] Based on this, the present invention provides a back-end intelligent method for intercepting tide videos and identifying high-frequency tide changes in order to solve the problem of how to efficiently and accurately extract tide level change information from massive tide surge videos and identify high-frequency tide level changes, thereby saving bandwidth and computing power resources.

[0006] The present invention provides a method for intelligently capturing tide videos and identifying tide level changes at high frequencies, comprising:

[0007] Real-time low-frequency image acquisition is performed on tide monitoring points to identify tide levels and obtain low-frequency tide curves;

[0008] The low-frequency tidal level curve is subjected to data conversion to obtain low-frequency tidal level difference data, low-frequency tidal level change type analysis is performed on the low-frequency tidal level difference data in combination with tidal bore forecast data and historical observation data experience, low-frequency abnormal data is cleaned, and the arrival time of the low-frequency tidal bore is obtained;

[0009] Based on the arrival time of the low-frequency tidal bore, a historical video playback method is used to obtain a tide arrival video, a high-frequency tide level change curve is generated based on the tide arrival video, and the high-frequency tide arrival time and high-frequency tide rise time are calculated to generate a tide arrival recognition video.

[0010] The operation of tide level recognition includes:

[0011] Through collection, labeling and training, a water body semantic segmentation recognition model is obtained. The water body in the image collection is identified and segmented by the water body semantic segmentation recognition model. Then, the total station is calibrated and measured on site using a virtual water gauge to realize the identification of tide level.

[0012] The acquisition of the water body semantic segmentation and recognition model includes:

[0013] Collect a large number of water body images, pre-label the water bodies in the water body images using a large model, manually review and modify the labeling results, and obtain labeling training data;

[0014] The Lovasz-Softmax loss function is selected to directly optimize the IoU index of the water body in the segmented image set by introducing the continuous approximation of the Jaccard index.

[0015] ;

[0016] ;

[0017] Among them, C represents a collection of categories, c represents a single category, represents the error vector for category C, represents the probability of category C predicted by the model, Represents the Lovasz extension.

[0018] The drawing of virtual water ruler includes:

[0019] Collect a low-tide image, draw a straight line along the plumb line on the low-tide image, set the highest tide point A and the lowest tide point B on the straight line, and automatically set a points on the straight line, a+2 points in total. Divide the line segment between points A and B into a+1 equal parts according to the pixel distance to obtain a virtual water gauge.

[0020] The operations of calibrating the total station on site by means of a virtual water gauge include:

[0021] Use a total station to measure the a+2 pixel point of the virtual water gauge to obtain the altitude and tide level value of the a+2 pixel point of the virtual water gauge. At the same time, set the watch position WP of the current camera and record a historical template low tide image TI;

[0022] The specific operations of tide level identification include:

[0023] By regularly collecting low-tide images, the current low-tide image and the historical template low-tide image TI are matched with the template image feature points to determine whether the camera has moved. If the camera has moved, the watch position WP is called to re-collect the low-tide image to obtain the final low-tide image.

[0024] A water body semantic segmentation model is used to identify the water body area in the low tide image, and the intersection point of the water body segmentation area and the virtual water ruler is obtained. The tide level of the intersection point is calculated proportionally through the known tide level values of the two calibrated pixel points above and below the intersection point to realize tide level recognition and finally obtain the low-frequency tide level curve.

[0025] The operations for analyzing low-frequency tidal level change types include:

[0026] Calculate tidal bore forecast data for tide monitoring points and download the current tidal bore forecast data through the web interface;

[0027] Acquiring low-frequency tidal level difference data based on the current tidal bore forecast data and the previous tidal bore forecast data;

[0028] When the low-frequency tidal level difference data is less than -0.5 meters, it is low-frequency abnormal data and needs to be cleaned. When the low-frequency tidal level difference data is greater than 0.5 meters, it is the suspected low-frequency tidal bore arrival time. When the following three conditions are met, the suspected low-frequency tidal bore arrival time is determined to be the low-frequency tidal bore arrival time. If not, it is low-frequency abnormal data and the low-frequency abnormal data is cleaned. The cleaning method of the low-frequency abnormal data is to take the trend mean TM of the 10 groups of low-frequency tidal level difference data before the low-frequency abnormal time, take the low-frequency tidal level data before the low-frequency abnormal time and add the trend mean TM to correct the low-frequency abnormal data.

[0029] The three conditions are specifically:

[0030] Take the four sets of low-frequency tidal level difference data after the suspected low-frequency tidal bore arrival time for calculation and judgment, and sum the four sets of low-frequency tidal level difference data. If the sum result is a positive value, the tide level trend is rising;

[0031] If the low-frequency tidal level difference data at the suspected low-frequency tidal bore arrival time plus the last set of low-frequency tidal level difference data is greater than the sum of the last two sets of low-frequency tidal level difference data among the four sets after the suspected low-frequency tidal bore arrival time, it means that the rising rate of the tidal level trend is gradually weakening;

[0032] Judging from the tidal bore forecast data, the suspected low-frequency tidal bore arrival time is within 30 minutes before and after the tidal bore forecast data.

[0033] The acquisition of the high-frequency tide level change curve includes:

[0034] Based on the arrival time of the low-frequency tidal bore, the historical video playback method is adopted to intercept the 2 minutes before the arrival time of the low-frequency tidal bore as the start time and the 1 minute after the arrival time of the low-frequency tidal bore as the end time, a total of 3 minutes of tidal bore arrival video, and the tidal bore arrival video is standardized and framed to obtain the image set PD, and the tide level is identified on the image set PD to obtain the high-frequency tide level curve.

[0035] The analysis of the high frequency tide curve includes:

[0036] Sequentially obtain the tide data in the high-frequency tide curve, obtain 10 groups of data before and after the tide data time, calculate whether the tide data is the maximum or minimum value, and obtain the peak array BL and the trough array BG;

[0037] Select the values in the crest array BL in turn, subtract the trough data closest to the current time, obtain the tidal range of the crest, and finally obtain the tidal range array BLX of all crests;

[0038] Take the maximum value of the tidal range in the tidal range array BLX of all crests. If there are multiple maximum values, take the time of the first maximum crest tidal range, that is, the high-frequency tide arrival time;

[0039] According to the time series, the first tidal range in BLX is greater than 0.5 meters, which is the high-frequency tide rise time. When the tidal range at the high-frequency tide rise time is the maximum tidal range in the tidal range array BLX of all wave crests, then the high-frequency tide rise time and the high-frequency tide arrival time are the same time.

[0040] The generation of the tide recognition video includes:

[0041] Take the 15 seconds of video before and after the high-frequency tide arrival time, superimpose the recognition results of this 30-second video, and generate the tide arrival recognition video.

[0042] Beneficial effects: The present invention adopts visual recognition technology to realize high-frequency recognition of tide level changes by cameras, which solves the problem of low acquisition frequency of traditional equipment and the inability to collect high-frequency change data of tidal head water level. At the same time, the invention introduces the Lovasz-Softmax loss function and the abnormal data judgment and cleaning method based on the tidal bore change law, thus realizing a tidal bore level monitoring method with simple and convenient installation, low cost, high frequency and high accuracy.

[0043] The arrival time of low-frequency tidal bores is obtained by analyzing the low-frequency tidal curve. A 3-minute video of the arrival of the tidal bore is captured and played back using a post-feedback method to identify high-frequency tidal changes. This significantly shortens the size of the collected tidal bore videos, reduces bandwidth, and reduces the computational effort required for identification.

[0044] By identifying high-frequency tidal changes, performing data analysis on high-frequency tidal curves, extracting peaks and troughs for analysis, and automatically extracting high-frequency tide arrival times and high-frequency tide rise times, the workload of manual extraction is reduced. At the same time, the timeliness of obtaining tidal arrival time data is greatly improved, solving the problem of manual judgment of tidal arrival locations not being timely, and providing data support for tide prevention safety.

[0045] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are provided for a better understanding of the present invention and do not constitute a limitation of the present invention.

[0047] Figure 1 is a flow chart provided according to the present invention;

[0048] Figure 2 is a tide level recognition result diagram provided by the present invention;

[0049] Figure 3 A shadowed tide level anomaly identification diagram provided by the present invention;

[0050] Figure 4 The tidal level anomaly identification diagram for the occurrence of water mist provided by the present invention;

[0051] Figure 5 This is a low-frequency tide level change type analysis diagram provided by the present invention;

[0052] Figure 6 is a low-frequency tide level curve graph provided according to the present invention;

[0053] Figure 7 is a high-frequency tide level curve graph provided according to the present invention;

[0054] Figure 8 This is a tidal bore diagram provided according to the present invention. DETAILED DESCRIPTION

[0055] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0056] like Figure 1As shown, the present invention provides a method for intelligently intercepting tide videos and identifying tide level changes at high frequencies, including:

[0057] S1: Real-time low-frequency image acquisition of tide monitoring points, tide level recognition, and low-frequency tide level curves.

[0058] The operation of tide level recognition includes:

[0059] Through collection, labeling and training, a water body semantic segmentation recognition model is obtained. The water body in the image collection is identified and segmented by the water body semantic segmentation recognition model. Then, the total station is calibrated and measured on site using a virtual water gauge to realize the identification of tide level.

[0060] Low-frequency tide level recognition: low-frequency images are collected at tide monitoring points to perform tide level recognition and obtain low-frequency tide level curves.

[0061] The low-frequency tide level curve is a tide level curve collected and identified once every 1 minute.

[0062] The acquisition of the water body semantic segmentation and recognition model includes:

[0063] Collect a large number of water body images, pre-label the water bodies in the water body images using a large model, manually review and modify the labeling results, and obtain labeling training data;

[0064] The Lovasz-Softmax loss function is selected to directly optimize the IoU index of the water body in the segmented image set by introducing the continuous approximation of the Jaccard index.

[0065] ;

[0066] ;

[0067] Among them, C represents a collection of categories, c represents a single category, represents the error vector for category C, represents the probability of category C predicted by the model, Represents the Lovasz extension.

[0068] Preferably, since the traditional cross entropy loss function is not completely consistent with the IoU evaluation index, the model may fail to significantly improve the actual segmentation effect when optimizing the loss function. The present invention uses the Lovasz-Softmax loss function and directly optimizes the IoU index in the segmentation task by introducing a continuous approximation of the Jaccard index. This can better reflect the goal of the segmentation task, especially when dealing with scenarios with unbalanced categories or blurred boundaries, and can greatly improve the segmentation accuracy and robustness of the model.

[0069] The drawing of virtual water ruler includes:

[0070] Collect a low-tide image, draw a straight line along the plumb line on the low-tide image, set the highest tide point A and the lowest tide point B on the straight line, and automatically set a points on the straight line, a+2 points in total. Divide the line segment between points A and B into a+1 equal parts according to the pixel distance to obtain a virtual water gauge.

[0071] The operations of calibrating the total station on site by means of a virtual water gauge include:

[0072] Use a total station to measure the a+2 pixel point of the virtual water gauge to obtain the altitude and tide level value of the a+2 pixel point of the virtual water gauge. At the same time, set the watch position WP of the current camera and record a historical template low tide image TI;

[0073] The specific operations of tide level identification include:

[0074] By regularly collecting low-tide images, the current low-tide image and the historical template low-tide image TI are matched with the template image feature points to determine whether the camera has moved. If the camera has moved, the watch position WP is called to re-collect the low-tide image to obtain the final low-tide image.

[0075] The water body semantic segmentation model is used to identify the water body area in the low tide image, and the intersection point of the water body segmentation area and the virtual water ruler is obtained. The tide level of the intersection point is calculated proportionally through the known tide level values of the two calibrated pixels above and below the intersection point, realizing tide level recognition (such as Figure 2 Finally, the low-frequency tide level curve is obtained.

[0076] S2: Convert the low-frequency tidal curve to obtain low-frequency tidal difference data, analyze the low-frequency tidal change type of the low-frequency tidal difference data based on the experience of tidal bore forecast data and historical observation data, clean the low-frequency abnormal data, and obtain the arrival time of the low-frequency tidal bore.

[0077] The operations for analyzing low-frequency tidal level change types include:

[0078] Calculate tidal bore forecast data for tide monitoring points and download the current tidal bore forecast data through the web interface;

[0079] Acquiring low-frequency tidal level difference data based on the current tidal bore forecast data and the previous tidal bore forecast data;

[0080] When the low-frequency tidal level difference data is less than -0.5 meters, it is low-frequency abnormal data and needs to be cleaned. When the low-frequency tidal level difference data is greater than 0.5 meters, it is the suspected low-frequency tidal bore arrival time. When the following three conditions are met, the suspected low-frequency tidal bore arrival time is determined to be the low-frequency tidal bore arrival time. If not, it is low-frequency abnormal data and the low-frequency abnormal data is cleaned. The cleaning method of the low-frequency abnormal data is to take the trend mean TM of the 10 groups of low-frequency tidal level difference data before the low-frequency abnormal time, take the low-frequency tidal level data before the low-frequency abnormal time and add the trend mean TM to correct the low-frequency abnormal data.

[0081] The three conditions are specifically:

[0082] Take the four sets of low-frequency tidal level difference data after the suspected low-frequency tidal bore arrival time for calculation and judgment, and sum the four sets of low-frequency tidal level difference data. If the sum result is a positive value, the tide level trend is rising;

[0083] If the low-frequency tidal level difference data at the suspected low-frequency tidal bore arrival time plus the last set of low-frequency tidal level difference data is greater than the sum of the last two sets of low-frequency tidal level difference data among the four sets after the suspected low-frequency tidal bore arrival time, it means that the rising rate of the tidal level trend is gradually weakening;

[0084] Judging from the tidal bore forecast data, the suspected low-frequency tidal bore arrival time is within 30 minutes before and after the tidal bore forecast data.

[0085] like Figure 5 As shown, in order to determine the type of low-frequency tidal level change in the shortest possible time, the present invention takes 4 groups of low-frequency tidal level difference data after the abnormal time for calculation and judgment. First, the last 4 groups of low-frequency tidal level difference data are summed. If the result is a positive value, the tide trend is rising. Secondly, the low-frequency tidal level difference data at the abnormal time plus the last group of low-frequency tidal level difference data is greater than the sum of the last two groups of low-frequency tidal level difference data in the last 4 groups, indicating that the rising rate of the tide trend is gradually weakening. Combined with the tidal bore forecast data, it is judged that the arrival time of the suspected low-frequency tidal bore is within 30 minutes before and after the tidal bore forecast data.

[0086] The abnormal data is due to the fact that shadows cannot be avoided in image recognition (such as Figure 3 As shown), water mist (as Figure 4 Problems such as those shown in the figure above may cause recognition anomalies. In actual tests, the average abnormal data rate is only about 1%, which is a very small amount.

[0087] Furthermore, 0.5 meters is based on historical data from tidal bore monitoring. Under normal circumstances, the tide level will only rise by 0.5 meters within 1 minute when the tidal bore arrives. The ebb tide recedes slowly and will not suddenly drop by 0.5 meters. Figure 6 shown.

[0088] S3: Based on the arrival time of the low-frequency tidal bore, a historical video playback method is used to obtain a tidal arrival video, a high-frequency tidal level change curve is generated based on the tidal arrival video, and the high-frequency tidal arrival time and high-frequency tidal rise time are calculated to generate a tidal arrival recognition video. It should be noted that:

[0089] The acquisition of the high-frequency tide level change curve includes:

[0090] Based on the arrival time of the low-frequency tidal bore, the historical video playback method is adopted to intercept the 2 minutes before the arrival time of the low-frequency tidal bore as the start time and the 1 minute after the arrival time of the low-frequency tidal bore as the end time, a total of 3 minutes of tidal bore arrival video, and the tidal bore arrival video is standardized and framed to obtain the image set PD, and the tide level is identified on the image set PD to obtain the high-frequency tide level curve.

[0091] The present invention adopts a post-feedback method to intercept 3 minutes of tidal arrival video playback recording to perform high-frequency tidal level change recognition, which greatly shortens the size of the collected tidal arrival video, reduces bandwidth, and reduces the calculation amount of recognition.

[0092] Furthermore, the high-frequency tide level curve is because traditional tide level monitoring is usually once every 2 seconds, which cannot accurately reflect the tide level changes when the tidal bore arrives, resulting in the inability to accurately capture the wave crest of the tidal bore. The high-frequency tide level curve refers to a high-frequency tide level curve that is identified 20 times in 1 second.

[0093] The standardized frame extraction is to extract 20 frames of images in the video per 1 second.

[0094] include,

[0095] Take the high-frequency tide level curve of the tidal bore arrival video and perform high-frequency tide level curve data analysis to analyze the high-frequency tide arrival time and high-frequency tide rise time;

[0096] The characteristics of the tide level at the high-frequency tide arrival time are that the high-frequency tide arrival time is the peak, and the tidal difference between the peak of the high-frequency tide arrival time and the first trough in the previous period is the largest.

[0097] There are two characteristics of high-frequency tide rise time. First, the tide level at the high-frequency tide arrival time must be the crest. Second, the tidal difference between the crest of the high-frequency tide arrival time and the first trough of the previous period is greater than 0.5 meters and it is the first to appear in the high-frequency tide level curve. Most of the time, the high-frequency tide arrival time and the high-frequency tide rise time are consistent. Only when the front water appears, the high-frequency tide arrival time and the high-frequency tide rise time are inconsistent. For example, Figure 7 As shown in the figure, due to the presence of forewater, it is necessary to determine the time when the tidal range is the largest, which is the time when the high-frequency tide arrives.

[0098] At the same time, in the high-frequency tide level curve, due to the fast speed of the tide and the fluctuation of the water surface, there is no shadow on the water surface, no reflection on the water surface, and the feature points are obvious, so the tide level recognition is accurate. At the same time, the probability of abnormal data appearing is only 1%, and the probability of appearing in these 3 minutes of a day is even lower. In actual testing, there is no abnormal data and no abnormal data processing is required. Figure 8 shown.

[0099] The analysis of the high frequency tide curve includes:

[0100] Sequentially obtain the tide data in the high-frequency tide curve, obtain 10 groups of data before and after the tide data time, calculate whether the tide data is the maximum or minimum value, and obtain the peak array BL and the trough array BG;

[0101] Select the values in the crest array BL in turn, subtract the trough data closest to the current time, obtain the tidal range of the crest, and finally obtain the tidal range array BLX of all crests;

[0102] Take the maximum value of the tidal range in the tidal range array BLX of all crests. If there are multiple maximum values, take the time of the first maximum value, that is, the high-frequency tide arrival time.

[0103] According to the time series, the first tidal range in BLX is greater than 0.5 meters, which is the high-frequency tide rise time. When the tidal range at the high-frequency tide rise time is the maximum tidal range in the tidal range array BLX of all wave crests, then the high-frequency tide rise time and the high-frequency tide arrival time are the same time.

[0104] The generation of the tide recognition video includes:

[0105] Take the 15 seconds of video before and after the high-frequency tide arrival time, superimpose the recognition results of this 30-second video, and generate the tide arrival recognition video.

[0106] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention shall be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for intelligently capturing tide videos and identifying tide level changes at high frequencies, characterized in that: include: Real-time low-frequency image acquisition is performed on tide monitoring points to identify tide levels and obtain low-frequency tide curves; The low-frequency tidal level curve is subjected to data conversion to obtain low-frequency tidal level difference data, low-frequency tidal level change type analysis is performed on the low-frequency tidal level difference data in combination with tidal bore forecast data and historical observation data experience, low-frequency abnormal data is cleaned, and the arrival time of the low-frequency tidal bore is obtained; The operations for analyzing low-frequency tidal level change types include: Calculate tidal bore forecast data for tide monitoring points and download the current tidal bore forecast data through the web interface; Acquiring low-frequency tidal level difference data based on the current tidal bore forecast data and the previous tidal bore forecast data; When the low-frequency tidal level difference data is less than -0.5 meters, it is low-frequency abnormal data and needs to be cleaned. When the low-frequency tidal level difference data is greater than 0.5 meters, it is the suspected low-frequency tidal bore arrival time. When the following three conditions are met, the suspected low-frequency tidal bore arrival time is determined to be the low-frequency tidal bore arrival time. If not, it is low-frequency abnormal data and the low-frequency abnormal data is cleaned. The cleaning method of the low-frequency abnormal data is to take the trend mean TM of the 10 groups of low-frequency tidal level difference data before the low-frequency abnormal time, take the low-frequency tidal level data before the low-frequency abnormal time and add the trend mean TM to correct the low-frequency abnormal data. The three conditions are specifically: Take the four sets of low-frequency tidal level difference data after the suspected low-frequency tidal bore arrival time for calculation and judgment, and sum the four sets of low-frequency tidal level difference data. If the sum result is a positive value, the tide level trend is rising; If the low-frequency tidal level difference data at the suspected low-frequency tidal bore arrival time plus the last set of low-frequency tidal level difference data is greater than the sum of the last two sets of low-frequency tidal level difference data among the four sets after the suspected low-frequency tidal bore arrival time, it means that the rising rate of the tidal level trend is gradually weakening; Judging from the tidal bore forecast data, the suspected low-frequency tidal bore arrival time is within 30 minutes before and after the tidal bore forecast data; Based on the arrival time of the low-frequency tidal bore, a tide arrival video is obtained by replaying historical videos, a high-frequency tidal level change curve is generated based on the tide arrival video, and the high-frequency tidal arrival time and high-frequency tidal rise time are calculated to generate a tide arrival recognition video; The analysis of the high-frequency tide level change curve includes: Sequentially obtain the tide data in the high-frequency tide change curve, obtain 10 groups of data before and after the tide data time, calculate whether the tide data is the maximum or minimum value, and obtain the peak array BL and the trough array BG; Select the values in the crest array BL in turn, subtract the trough data closest to the current time, obtain the tidal range of the crest, and finally obtain the tidal range array BLX of all crests; Take the maximum value of the tidal range in the tidal range array BLX of all crests. If there are multiple maximum values, take the time of the first maximum crest tidal range, that is, the high-frequency tide arrival time; According to the time series, the first tidal range in BLX is greater than 0.5 meters, which is the high-frequency tide rise time. When the tidal range at the high-frequency tide rise time is the maximum tidal range in the tidal range array BLX of all wave crests, then the high-frequency tide rise time and the high-frequency tide arrival time are the same time.

2. The method of intelligently capturing tide videos and identifying tide level changes at high frequencies according to claim 1, characterized in that: The operation of tide level recognition includes: Through collection, labeling and training, a water body semantic segmentation recognition model is obtained. The water body in the image collection is identified and segmented by the water body semantic segmentation recognition model. Then, the total station is calibrated and measured on site using a virtual water gauge to realize the identification of tide level.

3. The method of intelligently capturing tide videos and identifying tide level changes at high frequencies according to claim 2, characterized in that: The acquisition of the water body semantic segmentation and recognition model includes: Collect a large number of water body images, pre-label the water bodies in the water body images using a large model, manually review and modify the labeling results, and obtain labeling training data; The Lovasz-Softmax loss function is selected to directly optimize the IoU index of the water body in the segmented image set by introducing the continuous approximation of the Jaccard index. ； ； Among them, C represents a collection of categories, c represents a single category, represents the error vector for category C, represents the probability of category C predicted by the model, Represents the Lovasz extension.

4. The method of intelligently capturing tide videos and identifying tide level changes at high frequencies according to claim 3, characterized in that: The drawing of virtual water ruler includes: Collect a low-tide image, draw a straight line along the plumb line on the low-tide image, set the highest tide point A and the lowest tide point B on the straight line, and automatically set a points on the straight line, a+2 points in total. Divide the line segment between points A and B into a+1 equal parts according to the pixel distance to obtain a virtual water gauge. The operations of calibrating the total station on site by means of a virtual water gauge include: Use a total station to measure the a+2 pixel point of the virtual water gauge to obtain the altitude and tide level value of the a+2 pixel point of the virtual water gauge. At the same time, set the watch position WP of the current camera and record a historical template low tide image TI; The specific operations of tide level identification include: By regularly collecting low-tide images, the current low-tide image and the historical template low-tide image TI are matched with the template image feature points to determine whether the camera has moved. If the camera has moved, the watch position WP is called to re-collect the low-tide image to obtain the final low-tide image. A water body semantic segmentation model is used to identify the water body area in the low tide image, and the intersection point of the water body segmentation area and the virtual water ruler is obtained. The tide level of the intersection point is calculated by geometric calculation based on the known tide level values of the two upper and lower calibrated pixel points closest to the intersection point to realize tide level recognition and finally obtain the low-frequency tide level curve.

5. The method of intelligently capturing tide videos and identifying tide level changes at high frequencies according to claim 1, characterized in that: The acquisition of the high-frequency tide level change curve includes: Based on the arrival time of the low-frequency tidal bore, the historical video playback method is adopted to intercept the 2 minutes before the arrival time of the low-frequency tidal bore as the start time and the 1 minute after the arrival time of the low-frequency tidal bore as the end time, a total of 3 minutes of tidal bore arrival video, and the tidal bore arrival video is standardized and framed to obtain the image set PD, and the tide level is recognized on the image set PD to obtain the high-frequency tide level change curve.

6. The method of intelligently capturing tide videos and identifying tide level changes at high frequencies according to claim 1, characterized in that: The generation of the tide recognition video includes: Take the 15 seconds of video before and after the high-frequency tide arrival time, superimpose the recognition results of this 30-second video, and generate the tide arrival recognition video.

Images