A method for detecting internal waves in the ocean based on buoy observation data

Abstract

The application relates to the technical field of artificial intelligence and discloses a marine internal wave detection method based on buoy observation data, characterized in that: profile sea current observation data of a buoy is acquired, a flow velocity profile time sequence diagram is drawn, the starting and ending ranges of internal waves shown in the flow velocity profile time sequence diagram are marked, a marine internal wave flow velocity profile image sample library is constructed, a deep learning neural network model is trained by using the marine internal wave flow velocity profile image sample library, then, the trained deep learning neural network model is used to detect a to-be-detected flow velocity profile time sequence diagram, and the occurrence and ending time of detected internal waves are calculated. The application solves the problems existing in the traditional internal wave detection method based on remote sensing data by using the profile sea current observation data of the buoy, is not affected by atmospheric environmental factors, can timely and continuously detect internal waves in real time, and can detect unobvious internal wave phenomena.

Description

Technical Field

[0001] This invention belongs to the technical field of artificial intelligence, specifically relating to a method for detecting internal ocean waves based on buoy observation data. Background Technology

[0002] Currently, the detection technology for internal waves in the ocean mainly relies on remote sensing image data. However, this technology is affected by atmospheric environmental factors such as cloud cover and precipitation, and can only be performed under cloudless or lightly cloudy conditions. Furthermore, satellite orbital periods are typically long, requiring waiting for tens of minutes or even hours to acquire the latest remote sensing images, resulting in significant delays in internal wave detection. Additionally, the resolution of remote sensing image data is usually low, meaning that remote sensing satellites can only observe relatively obvious internal wave phenomena. Since internal wave activity occurs underwater, surface phenomena are not readily apparent, leading to omissions in internal wave detection and further threatening the safety of maritime operations. Summary of the Invention

[0003] This invention provides a method for detecting internal ocean waves based on buoy observation data. Using buoy profile current observation data as the data foundation, a deep learning neural network model is used to predict internal ocean waves. The buoy is unaffected by atmospheric environmental factors and can acquire the latest data every few seconds or minutes. Moreover, the buoy can acquire hydrological information such as temperature, salinity, depth, and current velocity in local sea areas, enabling accurate observation of various obvious or subtle internal wave phenomena. This provides an accurate data foundation for the prediction of the deep learning neural network model and improves the subsequent prediction accuracy.

[0004] This invention can be achieved through the following technical solutions:

[0005] A method for detecting internal ocean waves based on buoy observation data involves acquiring ocean current observation data from buoy profiles, plotting a current velocity profile time series diagram, marking the start and end ranges of internal waves shown in the current velocity profile time series diagram, constructing an ocean internal wave current velocity profile image sample library, training a deep learning neural network model using the ocean internal wave current velocity profile image sample library, and then using the trained deep learning neural network model to detect the current velocity profile time series diagram to be inspected, and calculating the occurrence and end times of the detected internal waves.

[0006] Furthermore, based on the inherent characteristics of the velocity profile time series, the start and end times of the velocity are marked as (M, N), the image size is W × H, and the detected internal wave positions are marked as (x, N). i ,y i ,w i ,h i ), where x i ,y i This indicates the coordinates of the position where the internal wave was first detected, w i,h i The width and height of the rectangle representing the start and end ranges of the detected internal wave are used to calculate the start time m of the internal wave using the following equation. i and end time n i .

[0007]

[0008]

[0009] Furthermore, the labelImg tool is used to label the start and end ranges of internal waves displayed in the flow velocity time-series profile, including obvious internal waves and indistinct internal waves. The obvious internal waves are set to have two wave cores, with the core being red, the upper wave core being dark blue, and the lower wave core being dark yellow. The indistinct internal waves are set to have two wave cores, with the upper wave core being light blue and the lower wave core being light yellow.

[0010] Furthermore, the deep learning neural network models include the Faster RCNN model and the YOLOv5 model, and recall and mean precision (mAP) are used as evaluation metrics.

[0011] The beneficial technical effects of this invention are as follows:

[0012] Compared with traditional methods for detecting and predicting internal waves, this invention uses deep learning target detection technology to construct an ocean internal wave detection model based on buoy profile ocean current observation data, thereby achieving accurate detection of internal waves and improving detection efficiency. Furthermore, by using buoy profile ocean current observation data, it solves the problems existing in traditional internal wave detection methods based on remote sensing data, is not affected by atmospheric environmental factors, and can continuously detect internal waves in a timely manner, and can detect inconspicuous internal wave phenomena. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0014] Figure 2 This is a schematic diagram illustrating the obvious internal wave of the present invention;

[0015] Figure 3 This is a schematic diagram of the indistinct internal wave of the present invention;

[0016] Figure 4 This is a schematic diagram of the network structure of the ocean internal wave detection model of the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings and preferred embodiments.

[0018] like Figure 1As shown, this invention provides a method for detecting internal ocean waves based on buoy observation data. The method involves acquiring ocean current observation data from buoy profiles, plotting a current velocity profile time series diagram, marking the start and end ranges of internal waves displayed in the current velocity profile time series diagram, constructing an ocean internal wave current velocity profile image sample library, training a deep learning neural network model using the ocean internal wave current velocity profile image sample library, and then using the trained deep learning neural network model to detect the current velocity profile time series diagram to be inspected, and calculating the occurrence and end times of the detected internal waves. Specifically:

[0019] S1. Acquire ocean current observation data from buoy profiles, plot time-series velocity profiles, label the start and end ranges of internal waves in the images, and construct a sample library of ocean internal wave velocity profile images. For example, acquire ocean current observation data from buoy profiles in the South China Sea (May 10, 2020 - May 29, 2020), and use MATLAB to plot a total of 8040 time-series velocity profiles. Among them, 7083 images from May 10 to May 26 are used as training data, and the remaining 957 images from May 27 to May 29 are used as test data for analysis of detection results.

[0020] S2. Based on the Pascal VOC data annotation format for target detection, use the LabelImg annotation tool to annotate all internal wave velocity time series plots. During annotation, both obvious and indistinct internal waves need to be annotated simultaneously. Obvious internal waves have two cores, with the core being red, the upper core being dark blue, and the lower core being dark yellow, showing clear layering and very dark colors, as shown in the attached figure. Figure 2 Even the less obvious internal waves have two wave nuclei: the upper wave nucleus is light blue, and the lower wave nucleus is light yellow, both being lighter in color, as shown in the attached image. Figure 3 .

[0021] The occurrence intervals of internal waves are marked in the images to generate a labeling file. Finally, internal wave time matching is performed based on the start and end times of each time series image and the position of the marked internal wave in the image to obtain a sample library of ocean internal wave velocity profile images.

[0022] S3. Based on a database of ocean internal wave velocity profile images, an ocean internal wave detection model is established using the Faster R-CNN deep learning object detection algorithm. The network structure is as follows: Figure 4 The Backbone network uses ResNet-50. Parameters such as enhancement method, learning rate, and number of iterations are adjusted to optimize model accuracy. Recall and mean precision (mAP) are used as evaluation metrics. The parameters of the ocean internal wave detection model are saved after the training termination condition is met.

[0023] After obtaining the trained ocean internal wave detection model, the remaining images in the ocean internal wave velocity profile image sample library are used to test the ocean internal wave detection model. The width and height of the image of the region where the internal wave is located and the start position of the internal wave are obtained. After the detection is completed, the start and end time of each detected internal wave are calculated. The operator checks whether the results are accurate. If they are not accurate, the test images are labeled and put into the training set to retrain the model.

[0024] S4. When an internal wave is detected, output the position of the internal wave. The position is represented by a rectangle, and the representation method is B(x). i ,y i ,w i ,h i ), where x i ,y i Let w be the coordinates of the top-left corner of the rectangle containing the internal wave. i ,h i Given the width and height of the rectangle, calculate the time of occurrence and termination of the internal wave, as follows:

[0025] The time series plot of the internal wave velocity to be detected has dimensions of W×H, the start and end times for recording the velocity are (M,N), and the location of an internal wave predicted by the ocean internal wave detection model is (x). i ,y i ,w i ,h i The starting coordinate of the internal wave is x. i The ending coordinate is x i +w i ,

[0026] The start time m of the internal wave i for:

[0027]

[0028] The end time n of the inner wave i for:

[0029]

[0030] Compared with traditional methods for detecting internal waves based on remote sensing images and manual judgment, the deep learning detection model for ocean internal waves constructed using buoy profile ocean current observation data in this invention can effectively overcome the defects of existing detection methods based on remote sensing images and replace manual judgment of the occurrence of internal waves. This invention can accurately and continuously detect internal waves in real time, and can overcome the disadvantages of traditional methods such as long delay, poor resistance to environmental interference, frequent missed alarms, and high labor costs.

[0031] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples. Various changes or modifications can be made to these embodiments without departing from the principles and essence of the present invention. Therefore, the scope of protection of the present invention is defined by the appended claims.

Claims

1. A method for detecting internal ocean waves based on buoy observation data, characterized in that: Acquire ocean current observation data from buoy profiles, plot a current velocity profile time series diagram, mark the start and end ranges of internal waves shown in the current velocity profile time series diagram, construct an ocean internal wave current velocity profile image sample library, train a deep learning neural network model using the ocean internal wave current velocity profile image sample library, then use the trained deep learning neural network model to detect the current velocity profile time series diagram to be examined, and calculate the occurrence and end times of the detected internal waves; Based on the inherent characteristics of the velocity profile time series diagram, the start and end times of the velocity on it are marked as follows: Its image size is The detected internal wave position is marked as ,in, This indicates the coordinates of the position where the internal wave was first detected. The width and height of the rectangle representing the start and end ranges of the detected internal wave are used to calculate the start time of the internal wave using the following equation. and end time ; Using the labelImg annotation tool, the start and end ranges of internal waves displayed in the flow velocity time-series profile are labeled, including obvious internal waves and indistinct internal waves. The obvious internal waves are set to have two wave cores, with the core being red, the upper wave core being dark blue, and the lower wave core being dark yellow. The indistinct internal waves are set to have two wave cores, with the upper wave core being light blue and the lower wave core being light yellow.

2. The ocean internal wave detection method based on buoy observation data according to claim 1, characterized in that: The deep learning neural network models include Faster RCNN and YOLOv5, and recall and mean precision (mAP) are used as evaluation metrics.

Images