Coastal wind power estimation method and system based on water wave form optical characteristics

Abstract

The invention discloses a coastal wind power estimation method and system based on water wave form optical characteristics. The method comprises the following steps: acquiring sea surface video data of a pre-selected coastal area; intercepting a continuous video with a stable water wave area; analyzing the intercepted continuous video into a plurality of frames of continuous images, and extracting water wave features; the water wave characteristics comprise sea wave vibration amplitude, frequency and sunlight irradiation angle; the process of extracting the water wave features comprises the following steps: identifying reflection point pixels in all images, determining the proportion of reflection pixel points, calculating the vibration frequency of sea waves and extracting the vibration amplitude; and inputting the extracted water wave features into the constructed wind power grade measurement model based on the water wave form to obtain a coastal site wind power grade evaluation result. According to the method, the wind power of the site is evaluated through an image analysis method, hardware facilities with high cost do not need to be arranged, long-term maintenance of equipment is not depended on, the small-range wind field measurement requirement is met, and the wind field change accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent wind assessment, and in particular to a coastal wind estimation method and system based on water ripple morphological optical characteristics. Background Art

[0002] Accurate acquisition of coastal wind is an important support for marine operations. Traditional coastal wind assessment requires the use of punctuation data from meteorological stations combined with on-site manual measurement. It is not possible to accurately measure the small-scale wind force of the entire coastal area. The existing coastal wind measurement tools are relatively common, such as the WRS3000 offshore measurement system, which is an integrated multi-layer wind speed and direction sensor, temperature and humidity sensor, and ocean physical parameter (wave, current) measurement module. It supports long-term continuous monitoring, but the hardware cost and deployment cost are very high, and this system requires long-term maintenance, regular calibration of sensors, and long-term reliance on professional technicians for inspection and maintenance. The maintenance cost is high. When plankton attaches, it will also affect the accuracy of the test. Therefore, it does not meet the current requirements for coastal wind testing. In the existing technology, satellite remote sensing images are also used for wind measurement, but it depends on the problem of spatial resolution. The resolution of microwave scatterometers is usually above 10 kilometers, which is not accurate enough for small-scale wind field changes. Although SAR has high resolution, it has a long revisit cycle and cannot be monitored in real time. Summary of the invention

[0003] Therefore, the purpose of the present invention is to provide a coastal wind force estimation method and system based on the optical characteristics of water ripple morphology, eliminating the influencing factors such as tides and ship driving, and realizing fully automated assessment of the wind force level in a designated sea area.

[0004] In order to achieve the above object, the present invention provides a method for estimating coastal wind force based on optical characteristics of water ripple morphology, comprising the following steps: S1. Acquire sea surface video data of a pre-selected coastal area; S2, intercepting a continuous video with a stable water ripple area in the sea surface video data; S3, using the intercepted continuous video to analyze multiple frames of continuous images, and extracting water ripple features from the multiple frames of continuous images; the water ripple features include wave vibration amplitude, vibration frequency and sunlight irradiation angle; extracting water ripple features from the multiple frames of continuous images includes: identifying reflective point pixels in all images, determining the proportion of reflective pixel points, calculating the vibration frequency and vibration amplitude of the waves; S4. Input the extracted water ripple features into the constructed wind force level calculation model based on water ripple morphology to obtain the coastal site wind force level assessment results.

[0005] Further preferably, in S1, the pre-selected coastal area is selected according to the following rules: Select sea surface areas with relatively little human disturbance; Set up a high-definition camera along the selected sea surface area and adjust the camera position to look down and observe the sea surface; Mark the highest high tide and lowest low tide along the coastal section of the selected area to ensure that the selected sea surface area is a seawater area all day long during observation; The sea surface observation area is selected at the lowest low tide, between 30 meters and 200 meters from the edge of the sea surface.

[0006] Further preferably, in S2, intercepting a continuous video with a stable water ripple area in the sea surface video data comprises the following steps: The video stream captured by the camera is clipped to delete the aerial area and only keep the sea surface area; Parse the retained video stream to obtain frame images, and extract from the frame images at specific time intervals; Perform ship recognition on the extracted frame image to determine whether the ship is moving; Video clips without ships or with stationary ships are captured, and the captured video clips are spliced ​​to obtain a continuous video with a stable water ripple area.

[0007] Further preferably, in S2, the step of performing ship recognition on the extracted frame image to determine whether the ship is moving includes the following process: Convert the extracted frame image to grayscale space; Use the Canny operator to extract the grayscale images of 10 frames and obtain the edge information image; The edge information image is processed by straight line pattern matching: the pixel value is calculated pixel by pixel, and the pixel value of 0 represents black and 1 represents white. If there is a continuous pixel area with a value of 1 and the continuous length is greater than the threshold, this area is marked as straight line edge information; In the entire image, if high-density straight line edge information appears within an actual distance of 50 meters, the entity of the straight line edge information is identified as a ship.

[0008] Further preferably, when capturing a video clip without a ship or a stationary ship, the method further includes: Parsing the captured video clip into a second frame of image; All the second frame images are processed in layers to obtain multiple image segments, which are numbered from top to bottom as: P1, P2, ..., P n； A second Canny operator is used to extract edge information from the second frame image, and an image with the least edge information is selected as a candidate stable water ripple area image; The edge information density of the candidate stable water ripple area image is extracted again according to the layered image segments, and the abnormal wave area is eliminated as the final selected stable water ripple area.

[0009] Further preferably, in S3, the extraction of the vibration frequency of the ocean waves comprises the following steps: Identify the reflective pixels in all images, determine the proportion of reflective pixels and perform curve fitting, and use the average value of the peak interval and trough interval of the fitting curve as the vibration frequency of the waves; The vibration amplitude of the waves is extracted by calculating the proportion of reflective pixels in each image among all the images, identifying some images with the highest proportion of reflective pixels, and calculating the average proportion of reflective pixels as the vibration amplitude of the waves.

[0010] Further preferably, the method further includes constructing a mapping relationship between the extracted water pattern features and the wind force level, including: Divide the daily sunshine period into multiple sub-periods of 10 minutes; According to the sampling date and sub-period, the angle of sunlight in three-dimensional space relative to the coastal area is determined; Construct a database to store the wave vibration amplitude, vibration frequency and three-dimensional sunlight angle as a data item; Through historical weather data, the corresponding wind force level is filled in for each data item in the database.

[0011] Further preferably, in S4, the wind force level calculation model uses Tensorflow as a machine learning framework; The daily wave vibration amplitude, vibration frequency and sunlight angle in the constructed database are used as input samples, and the corresponding wind force level is used as the sample label; Using the sample library formed by samples and labels, the neural network is trained in a weighted cycle until the model converges.

[0012] Further preferably, the sample library is prepared in the following form: WaveRange n = aWaveRange n_0 + bWaveRange n_1 + cWaveRange n_2 +dWaveRange n_3； WaveFreq n =aWaveFreq n_0 +bWaveFreq n_1 + cWaveFreq n_2+dWaveFreq n_3； WindLevel n = aWindLevel n_0+ bWindLevel n_1 +cWindLevel n_2 +dWindLevel n_3； Among them, WaveRange n Represents the wave vibration amplitude sample; WaveFreq n Represents the wave vibration frequency sample; WindLevel n Represents a wind force level sample; the current wave vibration amplitude data is recorded as WaveRange n_0 , the wave vibration frequency is recorded as WaveFreq n_0 , wind force level is recorded as WindLevel n_0 ; The data of the previous three moments are recorded as WaveRange n_1 、WaveRange n_2 、Waverange n_3 , WaveFreq n_1 、WaveFreq n_2 、WaveFreq n_3 ,WindLevel n_1 、WindLevel n_2 、WindLevel n_3 ; a, b, c, d are weighted empirical coefficients respectively.

[0013] The present invention also provides a coastal wind force estimation system based on water ripple morphology optical characteristics, which is used to implement the coastal wind force estimation method based on water ripple morphology optical characteristics, comprising: Coastal sensor module: used to obtain sea surface video data of pre-selected coastal areas; A video segment capture module for a sea surface stable water ripple area, used to capture continuous video with a stable water ripple area in the sea surface video data; Water ripple morphology optical feature extraction module: Analyze the intercepted continuous video into multiple frames of continuous images, and extract water ripple features from the multiple frames of continuous images; A water ripple morphology and optical characteristics corresponding to wind force level database construction module is used to form a sample library of extracted water ripple features and corresponding wind force levels, and use a deep learning algorithm to train the constructed wind force level measurement model based on water ripple morphology; The coastal real-time wind force estimation module inputs the extracted water ripple features into the constructed wind force level estimation model based on water ripple morphology to obtain the coastal on-site wind force level assessment results.

[0014] The coastal wind force estimation method and system based on the optical characteristics of water ripple morphology disclosed in this application evaluates the oscillation amplitude and oscillation frequency of the waves through image analysis to evaluate the wind force on site. There is no need to deploy expensive hardware facilities, and the detection results do not rely on long-term maintenance of the equipment. It meets the needs of close-range and small-scale wind field measurement and improves the accuracy of wind field changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flow chart of a coastal wind force estimation method based on the optical characteristics of water ripple morphology of the present invention; Figure 2 It is a module block diagram of a coastal wind force estimation method based on the optical characteristics of water ripple morphology of the present invention; Figure 3 It is a schematic diagram of sea surface area extraction of the coastal wind force estimation method based on the optical characteristics of water ripple morphology of the present invention; Figure 4 It is a schematic diagram of extracting the "stable water ripple area" of the coastal wind estimation method based on the optical characteristics of water ripple morphology of the present invention. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below through the accompanying drawings and specific embodiments.

[0017] like Figure 1 As shown, an embodiment of the present invention provides a method for estimating coastal wind force based on optical characteristics of water ripple morphology, comprising the following steps: S1. Acquire sea surface video data of a pre-selected coastal area. It should be noted that the camera is arranged at a suitable position in the coastal area so that the nearshore sea surface can be observed at both high tide and low tide.

[0018] The pre-selected coastal areas are selected according to the following rules: Select sea surface areas with relatively little human disturbance; Set up a high-definition camera along the selected sea surface area and adjust the camera position to look down and observe the sea surface; Mark the highest high tide and lowest low tide along the coastal section of the selected area to ensure that the selected sea surface area is a seawater area all day long during observation; The sea surface observation area is selected at the lowest low tide, between 30 meters and 200 meters from the edge of the sea surface.

[0019] The specific implementation steps are: Step 1.1: In coastal areas where wind force measurement is required, choose coastal sections with fewer ships and fewer people (water sports) on the sea surface as much as possible to ensure that the subsequent collection of optical characteristics of wave ripples is not interfered with by humans; Step 1.2: Mark the highest high tide state and the lowest low tide state in the selected coastal section; Step 1.3: Set up a high-definition camera on the coast, 4-6 meters above the ground; Step 1.4, adjust the camera observation angle. The camera should look down at the sea surface. The sea surface observation follows two principles: 1) It is a seawater area all day (at high tide and low tide); 2) Choose an appropriate sea surface for observation (the farther the observation distance, the greater the error; if the observation distance is too close, the water surface is prone to turbidity, which affects the subsequent water pattern optical feature extraction); Step 1.5: According to the above principles, the sea surface observation area should be selected at the lowest low tide, between 30 meters and 200 meters from the shore (sea edge).

[0020] S2, intercepting a continuous video with a stable water ripple area in the sea surface video data; The specific steps include: a) Edit the video stream captured by the camera, delete the aerial area, and keep only the sea surface area. This is mainly to avoid misjudgment caused by non-sea surface targets (such as clouds, airplanes, etc.); Figure 3 shown.

[0021] b) The sea surface line vector concentration recognition method is used for the reserved area to identify the ship. The sea surface line vector concentration recognition method includes analyzing the characteristics of the sea surface image. Only in the area where the ship is located in the image, there will be a large number of straight line vectors and regular surface vectors (unchanged rectangles, arcs, etc.). Therefore, the ship in the sea surface can be quickly identified based on this. Note: Other sea surface images, such as water ripples (including dark surfaces and reflective light surfaces), will not have highly concentrated straight line and regular surface elements. The specific steps include the following: b1) Parse the retained video stream to obtain frame images, and extract them from the frame images at specific time intervals; for example, if the video segment length is 10 minutes and the camera sampling frequency is 15 frames, then there are One frame image is extracted every 1 minute, for a total of 10 frames, which are used for subsequent ship identification.

[0022] b2) Perform ship recognition on the extracted frame image to determine whether the ship is moving, including: Convert 10 frames of images to grayscale space (assuming they were originally in RGB color space); Use the Canny operator to extract the grayscale images of 10 frames and obtain the edge information image; The edge information image is processed by "straight line mode" matching: the pixel value is calculated pixel by pixel (0 is black, 1 is white). If there is a continuous pixel area with a value of 1, and the continuous length is greater than the threshold (to avoid interference from water marks), it indicates that the area is "straight line edge information"; In the entire image, if a high density of "straight edge information" appears within the actual distance of 50 meters, the entity of the straight edge information is identified as a ship; it can be explained that since the camera is fixed and the position of the sea surface of the image taken is fixed, the actual length can be converted from the length of the target object in the image; Determine whether the ship is moving: If the high-density area of ​​"straight line edge information" is the same in 10 frames of images, it is determined that the ship has not moved in the corresponding time period and is in a stationary state; otherwise, it is determined that the ship is moving.

[0023] b3) intercepting video clips without ships or with stationary ships, and splicing the intercepted video clips to obtain a continuous video with a stable water ripple area.

[0024] Further preferably, when capturing a video clip without a ship or a stationary ship, the method further includes: Parsing the captured video clip into a second frame of image; All the second frame images are processed in layers to obtain multiple image segments, which are numbered from top to bottom as: P1, P2, ..., P n； A second Canny operator is used to extract edge information from the second frame image, and an image with the least edge information is selected as a candidate stable water ripple area image; The edge information density of the candidate stable water ripple area image is extracted again according to the layered image segments, and the abnormal wave area is eliminated as the final selected stable water ripple area.

[0025] S3, using the intercepted continuous video to analyze multiple frames of continuous images, and extracting water ripple features from the multiple frames of continuous images; the water ripple features include wave vibration amplitude, vibration frequency and sunlight irradiation angle; extracting water ripple features from the multiple frames of continuous images includes: identifying reflective point pixels in all images, determining the proportion of reflective pixel points, calculating the vibration frequency and vibration amplitude of the waves; Further preferably, in S3, the extraction of the vibration frequency of the ocean waves comprises the following steps: Identify the reflective pixels in all images, determine the proportion of reflective pixels and perform curve fitting, and use the average value of the peak interval and trough interval of the fitting curve as the vibration frequency of the waves; The vibration amplitude of the waves is extracted by calculating the proportion of reflective pixels in each image among all the images, identifying some images with the highest proportion of reflective pixels, and calculating the average proportion of reflective pixels as the vibration amplitude of the waves.

[0026] Further preferably, the method further includes constructing a mapping relationship between the extracted hydrological characteristics and the wind force level, including: Divide the daily sunshine period into multiple sub-periods of 10 minutes; According to the sampling date and sub-period, the angle of sunlight in three-dimensional space relative to the coastal area is determined; Construct a database to store the wave vibration amplitude, vibration frequency and three-dimensional sunlight angle as a data item; Through historical weather data, the corresponding wind force level is filled in for each data item in the database.

[0027] Judging the frequency of sea wave vibration, since the azimuth of sunlight (or other natural light) is relatively fixed in a short period of time (such as 10 minutes), the vibration frequency of the sea wave can be evaluated by the periodic reflection characteristics of the sea wave. The implementation method is as follows: 301) extracting the video stream of the “stable water pattern area” within a time period (e.g., 10 minutes); 302) Parse the video stream into 9000 frame images; 303) converting all frame images into grayscale space; 304) For each frame image, calculate the pixel value (0~255) pixel by pixel; 305) Setting a threshold to identify the light-reflecting property of the pixel, such as 200, that is, a pixel point with a pixel value greater than 200 is identified as reflecting light at this moment (presenting an optical property close to white); 306) Calculate the proportion of reflective pixels in each of the 9000 images; 307) Draw a quadratic curve with 9000 values, and calculate the interval between the peaks (maximum values) and the interval between the troughs (minimum values). Considering that the local waves on the sea surface may produce irregular vibrations, the 500 values ​​with the largest and smallest intervals are eliminated. 308) Take the average value of all the peak intervals and trough intervals to finally get the vibration frequency of the wave: convert it through the sampling rate of the frame image, such as 15 frames / second, then the time difference between the two images is 1 / 15 second; The amplitude of the sea wave vibration is judged by the reflection of the sea surface, which is related to the direct angle of sunlight. Therefore, the time of the video segment needs to be considered. The implementation method is as follows: 311) Same as step 301, parsing the video segment into 9000 frame images, and calculating the proportion of reflective pixels in each image; 312) To reduce the impact of irregular changes in waves, the 1,000 frames with the highest reflection ratio were removed; 313) Among the remaining 8000 images, take the 500 images with the highest reflection ratio and take their average value; 314) together with the moment (date and time) and the average of the highest reflective ratio.

[0028] S4. Input the extracted water ripple features into the constructed wind force level calculation model based on water ripple morphology to obtain the coastal site wind force level assessment result. It should be noted that it is also necessary to establish a wind force level database corresponding to the optical features of water ripple morphology, and use the database to train the wind force level calculation model; The steps for constructing a database of wind force levels corresponding to water ripple morphological optical characteristics include: Step 4.1, divide the daily sunshine period into multiple "sub-periods" of 10 minutes; Step 4.2: Based on the sampling date and “sub-period”, the angle of sunlight in three-dimensional space relative to the coastal area can be determined; Step 4.3, construct a database to store the wave vibration amplitude, vibration frequency and three-dimensional sunlight irradiation angle as a data item; Step 4.4: Through manual on-site judgment, complete the corresponding wind force level for each data item in the database.

[0029] In S4, the wind force level calculation model uses Tensorflow as a machine learning framework; the daily wave vibration amplitude, vibration frequency and sunlight angle in the constructed database are used as input samples, and the corresponding wind force level is used as a sample label; Using the sample library formed by samples and labels, the neural network is trained in a weighted cycle until the model converges.

[0030] Further preferably, the sample library is prepared in the following form: WaveRange n = aWaveRange n_0 + bWaveRange n_1 + cWaveRange n_2 +dWaveRange n_3； WaveFreq n =aWaveFreq n_0 +bWaveFreq n_1+ cWaveFreq n_2 +dWaveFreq n_3； WindLevel n = aWindLevel n_0+ bWindLevel n_1 +cWindLevel n_2 +dWindLevel n_3； Among them, WaveRange n Represents the wave vibration amplitude sample; WaveFreq n Represents the wave vibration frequency sample; WindLevel n Represents a wind force level sample; the current wave vibration amplitude data is recorded as WaveRange n_0 , the wave vibration frequency is recorded as WaveFreq n_0 , wind force level is recorded as WindLevel n_0 ; The data of the previous three moments are recorded as WaveRange n_1 、WaveRange n_2 、Waverange n_3 , WaveFreq n_1 、WaveFreq n_2 、WaveFreq n_3 ,WindLevel n_1 、WindLevel n_2 、WindLevel n_3 ; a, b, c, d are weighted empirical coefficients respectively.

[0031] The present invention also provides a coastal wind force estimation system based on water ripple morphology optical characteristics, which is used to implement the coastal wind force estimation method based on water ripple morphology optical characteristics, comprising: Coastal sensor module 10: used to obtain sea surface video data of pre-selected coastal areas; sensors are deployed at appropriate locations in coastal areas so that the nearshore sea surface can be observed at both high tide and low tide. First, in coastal areas where wind force needs to be measured, coastal sections with fewer ships and fewer people (water sports) on the sea surface are selected as much as possible to ensure that the subsequent collection of optical characteristics of water ripples of waves is not interfered with by humans; then, in the selected coastal sections, mark the highest high tide state and the lowest low tide state of the tide; then, set up a high-definition camera on the coast, and the camera is set up 4-6 meters from the ground; then, adjust the camera observation angle, the camera should look down at the sea surface, and observe the sea surface according to two principles: 1) It is a seawater area all day long (at high tide and low tide); 2) Select an appropriate sea surface for observation (the farther the observation distance, the greater the error; if the observation distance is too close, the water surface is prone to turbidity, which affects the subsequent extraction of water ripple optical characteristics); finally, according to the above principles, the sea surface observation area should be selected at the lowest low tide state, between 30 meters and 200 meters from the shore (sea surface edge).

[0032] The video segment capture module 20 of the sea surface stable water ripple area is used to capture the continuous video with the stable water ripple area in the sea surface video data. Through the intelligent processing of the video camera, the influence of tide and ship movement on the waves is eliminated, and the "stable water ripple area" is selected. First, it is necessary to identify whether there is a ship in the video segment captured by the camera in real time: first, the video stream captured by the camera is cropped to avoid misjudgment caused by non-sea surface targets (such as clouds in the sky, airplanes, etc.); Figure 3 As shown in the figure, analyzing the characteristics of the sea surface image, only the area where the ship is located in the image will have a large number of straight line vectors and regular surface vectors (unchanged rectangles, arcs, etc.), so the ship in the sea surface can be quickly identified based on this; Explanation: Other images of the sea surface, such as water ripples (including dark surfaces and reflective surfaces), will not have highly concentrated straight lines and regular surface elements; Then, the video stream is parsed into frame images; if the video segment length is 10 minutes and the camera sampling frequency is 15 frames, there are a total of Frame images are extracted every 1 minute, for a total of 10 frames for subsequent ship identification; then, the 10 frames are converted to grayscale space (assuming that the original color space is RGB), and the Canny operator is used to extract the grayscale images of the 10 frames to obtain edge information images; then the edge information image is processed by "straight line mode" matching: the pixel value is calculated pixel by pixel (0 is black, 1 is white), if there is a continuous pixel area with a value of 1, and the continuous length is greater than the threshold (avoiding the interference of the water pattern part), it indicates that the area is "straight line edge information"; finally, in the entire image, if a high-density "straight line edge information" appears within the actual distance of 50 meters, the entity of the straight line edge information is identified as a ship; it is explained that since the camera is fixed and the position of the sea surface of the image taken is fixed, the actual length can be converted through the length of the target object in the image. Then, it is necessary to determine whether the ship is moving: if the high-density area of ​​"straight line edge information" is the same in the 10 frames, it is determined that the ship has not moved in the corresponding time period and is in a stationary state; otherwise, it is determined that the ship is moving. Next, in the video segment, a sub-video segment without a ship or a stationary ship is intercepted; further, the influence of regular waves caused by natural tides needs to be eliminated: the sub-video segment intercepted above is also analyzed into continuous frame images; the wave characteristics of natural tides are analyzed: usually the whole wave of the coastline rushes to the coastline in stages; 10 frame images are equally selected from the sub-video segment. If the sub-video segment is not continuous, only the sub-video segment with the longest duration is considered for equal division; the 10 frames of images are layered, such as Figure 4 As shown, it is divided into multiple image segments, and the image segments are numbered from top to bottom: P1, P2, ..., P n , where the value of n can be set manually by the operator; for the 10 frame images, convert them into grayscale images (avoid the influence of color and light, and judge only from the morphological features), and then use the Canny operator to extract edge information; it is necessary to adjust the Canny operator parameters so that the "small edges" (normal water ripples) are avoided as much as possible during the edge information extraction process; among the 10 frame images, select the image with the least edge information as the "alternative stable water ripple area" image; edge information ratio: the ratio of edge information pixels in the edge information image to all image range pixels. Finally, for the "alternative stable water ripple area" (such as Figure 4 The uppermost video segment in the video segment should also be removed from the abnormal wave area in the video segment, such as Figure 4 In the circle part of the top video segment, the Canny operator is used to extract edge information, and the density of the edge information is judged; it is noted that the parameters of the Canny operator in this step are different from those in step 2.4; Figure 4 The "final selected area" shown is the "stable water ripple area" adopted by the present invention.

[0033] The water ripple morphology optical feature extraction module 30 uses the intercepted continuous video to analyze multiple frames of continuous images, and extracts water ripple features from the multiple frames of continuous images; The extracted water ripple features and the corresponding wind force levels form a sample library, and the deep learning algorithm is used to train the constructed wind force level measurement model based on water ripple morphology. Through continuous video frame images, the regular characteristics of water ripples are calculated to assess the vibration frequency and wave size of waves affected by wind, thereby providing data support for subsequent wind force assessment. The first is the judgment of wave vibration frequency. Since the azimuth of sunlight (or other natural light) is relatively fixed in a short period of time (such as 10 minutes), the vibration frequency of waves can be evaluated by the periodic reflection characteristics of waves: within a time period (such as 10 minutes), extract the video stream of the "stable water ripple area"; parse the video stream into 9,000 frame images; convert all frame images to grayscale space; for each frame image, calculate its pixel value (0~255) pixel by pixel; set a threshold to mark the reflective characteristics of the pixel, such as 200, that is, the pixel point with a pixel value greater than 200 is marked as the pixel point in the grayscale space. The image is always reflective (presenting an optical property close to white); calculate the proportion of reflective pixels in each of the 9000 images; draw a quadratic curve with the 9000 values, and calculate the interval between the curve crests (maximum values) and crests, and the interval between the troughs (minimum values) and troughs; considering that local waves on the sea surface may produce irregular vibrations, the 500 values ​​with the largest and smallest intervals are eliminated; take the average of all crest intervals and trough intervals, and finally get the vibration frequency of the waves: convert it through the sampling rate of the frame image, such as 15 frames / second, then the time difference between the two images is 1 / 15 second. Next is the determination of the amplitude of wave vibration. Since the size of waves is determined by the reflection of the sea surface, which is related to the direct angle of sunlight, the time of the video segment must be considered: the same method for determining the frequency of wave vibration is used to parse the video segment into 9,000 frames, and calculate the proportion of reflective pixels in each image; to reduce the impact of irregular changes in waves, remove the 1,000 frames with the highest reflective ratio; among the remaining 8,000 images, take the 500 frames with the highest reflective ratio and take their average; and record them together with the moment (date and time) and the average of the highest reflective ratio.

[0034] The morphological characteristics of water ripples are mainly the vibration frequency and amplitude of waves. The corresponding database of water ripple morphological characteristics and wind force levels is used, and the date and time (direct angle of sunlight) must also be considered. First, the daily sunshine period is divided into multiple "sub-periods" in units of 10 minutes; based on the sampling date and "sub-period", the angle of sunlight in three-dimensional space relative to the coastal area can be determined; then, a database is constructed to store the wave vibration amplitude, vibration frequency and three-dimensional angle of sunlight as a data item; finally, each data item in the database is filled with the corresponding wind force level through manual on-site judgment.

[0035] The coastal real-time wind force estimation module 40 inputs the extracted water ripple features into a constructed wind force level estimation model based on water ripple morphology to obtain a coastal on-site wind force level assessment result.

[0036] The wind force level calculation model based on water ripple morphology uses a deep learning model to evaluate the wind force level on the coast. First, a recurrent neural network model is built based on the Tensorflow2.0 machine learning framework; then, based on the aforementioned database, a sample library is made: the samples are the wave vibration amplitude, vibration frequency and sunlight angle, and the sample labels are the manually calibrated on-site wind force levels. The recurrent neural network is trained until the model converges; finally, the trained model is used to fully automatically evaluate the wind force level at the real-time coastal site through the optical characteristics of the sea surface water ripple morphology collected by the camera.

[0037] Obviously, the above embodiments are only examples for clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from them are still within the protection scope of the invention.

Claims

1. A coastal wind force estimation method based on the optical characteristics of water ripple morphology, characterized in that: The following steps are involved: S1. Acquire sea surface video data of a pre-selected coastal area; S2, intercepting a continuous video with a stable water ripple area in the sea surface video data; S3, using the intercepted continuous video to analyze multiple frames of continuous images, and extracting water ripple features from the multiple frames of continuous images; The water pattern characteristics include the amplitude of wave vibration, the vibration frequency and the angle of sunlight exposure; The process of extracting water ripple features from multiple continuous frames of images includes: identifying reflective pixels in all images, determining the proportion of reflective pixels, calculating the vibration frequency and vibration amplitude of the waves; S4. Input the extracted water ripple features into the constructed wind force level calculation model based on water ripple morphology to obtain the coastal site wind force level assessment results.

2. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 1 is characterized in that: In S1, the pre-selected coastal area is selected according to the following rules: Select sea surface areas with relatively little human disturbance; Set up a high-definition camera along the selected sea surface area and adjust the camera position to look down and observe the sea surface; Mark the highest high tide and lowest low tide along the coastal section of the selected area to ensure that the selected sea surface area is a seawater area all day long during observation; The sea surface observation area is selected at the lowest low tide, between 30 meters and 200 meters from the edge of the sea surface.

3. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 1 is characterized in that: In S2, a continuous video with a stable water ripple area is intercepted from the sea surface video data, including the following steps: The video stream captured by the camera is clipped to delete the aerial area and only keep the sea surface area; Parse the retained video stream to obtain frame images, and extract from the frame images at specific time intervals; Perform ship recognition on the extracted frame image to determine whether the ship is moving; Video clips without ships or in which the ships are stationary are intercepted, and the intercepted video clips are spliced ​​to obtain a continuous video with a stable water ripple area.

4. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 3 is characterized in that: In S2, the process of performing ship recognition on the extracted frame image to determine whether the ship is moving includes the following process: Convert the extracted frame image to grayscale space; Use the Canny operator to extract the grayscale images of 10 frames and obtain the edge information image; The edge information image is processed by straight line pattern matching: the pixel value is calculated pixel by pixel, and the pixel value of 0 represents black and 1 represents white. If there is a continuous pixel area with a value of 1 and the continuous length is greater than the threshold, this area is marked as straight line edge information; In the entire image, if high-density straight line edge information appears within an actual distance of 50 meters, the entity of the straight line edge information is identified as a ship.

5. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 3 is characterized in that: When capturing footage without a vessel or with a stationary vessel, this also includes: Parsing the captured video clip into a second frame of image; All the second frame images are processed in layers to obtain multiple image segments, which are numbered from top to bottom as: P1, P2, ..., P n ; A second Canny operator is used to extract edge information from the second frame image, and an image with the least edge information is selected as a candidate stable water ripple area image; The edge information density of the candidate stable water ripple area image is extracted again according to the layered image segments, and the abnormal wave area is eliminated as the final selected stable water ripple area.

6. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 1, characterized in that: In S3, the extraction of the vibration frequency of the ocean wave comprises the following steps: Identify the reflective pixels in all images, determine the proportion of reflective pixels and perform curve fitting, and use the average value of the peak interval and trough interval of the fitting curve as the vibration frequency of the waves; The vibration amplitude of the waves is extracted by calculating the proportion of reflective pixels in each image among all the images, identifying some images with the highest proportion of reflective pixels, and calculating the average proportion of reflective pixels as the vibration amplitude of the waves.

7. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 1, characterized in that: The method also includes establishing a mapping relationship between the extracted water pattern features and the wind force level, including: Divide the daily sunshine period into multiple sub-periods of 10 minutes; According to the sampling date and sub-period, the angle of sunlight in three-dimensional space relative to the coastal area is determined; Construct a database to store the wave vibration amplitude, vibration frequency and three-dimensional sunlight angle as a data item; Through historical weather data, the corresponding wind force level is filled in for each data item in the database.

8. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 7, characterized in that: In S4, the wind force level calculation model uses Tensorflow as a machine learning framework; The daily wave vibration amplitude, vibration frequency and sunlight angle in the constructed database are used as input samples, and the corresponding wind force level is used as the sample label; Using the sample library formed by samples and labels, the neural network is trained in a weighted cycle until the model converges.

9. The coastal wind force estimation method based on water ripple morphological optical characteristics according to claim 8, characterized in that: The sample library is prepared in the following form: WaveRange n = aWaveRange n_0 + bWaveRange n_1 + cWaveRange n_2 + dWaveRange n_3； WaveFreq n = aWaveFreq n_0 + bWaveFreq n_1 + cWaveFreq n_2 + dWaveFreq n_3； WindLevel n = aWindLevel n_0+ bWindLevel n_1 +cWindLevel n_2 +dWindLevel n_3； Among them, WaveRange n Represents the wave vibration amplitude sample; WaveFreq n Represents the wave vibration frequency sample; WindLevel n Represents a wind force level sample; the current wave vibration amplitude data is recorded as WaveRange n_0 , the wave vibration frequency is recorded as WaveFreq n_0 , wind force level is recorded as WindLevel n_0 ; The data of the previous three moments are recorded as WaveRange n_1 、WaveRange n_2 、Waverange n_3 , WaveFreq n_1 、WaveFreq n_2 、WaveFreq n_3 , WindLevel n_1 、WindLevel n_2 、WindLevel n_3 ; a, b, c, d are weighted empirical coefficients respectively.

10. A coastal wind force estimation system based on the optical characteristics of water ripple morphology, characterized in that: The method for estimating coastal wind force based on water ripple morphological optical characteristics according to any one of claims 1 to 9 above comprises: Coastal sensor module: used to obtain sea surface video data of pre-selected coastal areas; A video segment capture module for the sea surface stable water ripple area, used to capture continuous video with a stable water ripple area in the sea surface video data; Water ripple morphology optical feature extraction module: Analyze the intercepted continuous video into multiple frames of continuous images, and extract water ripple features from the multiple frames of continuous images; A water ripple morphology and optical characteristics corresponding to wind force level database construction module is used to form a sample library of extracted water ripple features and corresponding wind force levels, and use a deep learning algorithm to train the constructed wind force level measurement model based on water ripple morphology; The coastal real-time wind force estimation module inputs the extracted water ripple features into the constructed wind force level estimation model based on water ripple morphology to obtain the coastal on-site wind force level assessment results.

Images