Marine oil spill detection and tracking method
By establishing a marine oil spill target detection model and combining it with high-precision turntable motion, real-time acquisition and processing of sea surface video information has solved the problems of small coverage area and low efficiency in marine oil spill detection, realizing real-time oil spill detection and tracking within a large field of view and reducing costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2024-11-29
- Publication Date
- 2026-05-29
AI Technical Summary
Existing methods for detecting and tracking marine oil spills suffer from problems such as limited coverage, low detection efficiency, and high costs.
A marine oil spill target detection model was established and trained. Video information of the sea surface was collected in real time. The inter-frame spatiotemporal correlation was used to identify and track targets of interest. The model was combined with high-precision turntable motion locking and target tracking. The ResNet-18 neural network model and Fourier transform technology were used. Noise samples were generated from simulated images for training. Hardware acceleration was achieved using FPGA.
It enables real-time detection and tracking of marine oil spills within a wide field of view, improving detection efficiency and coverage while reducing costs.
Smart Images
Figure CN122116258A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of target tracking technology, and in particular to a method for detecting and tracking marine oil spills. Background Technology
[0002] Marine oil spills pose a significant environmental risk, with substantial negative impacts on marine ecosystems, human health, and cultural resources. Marine oil spills primarily originate from accidents on offshore oil platforms, leaks in near-shore or offshore oil pipelines, leaks during maritime transport, and collisions between ships. With the further development of the marine oil and gas industry, the monitoring and early warning of marine oil spills have become increasingly urgent. When an oil spill occurs, it is crucial to promptly identify its location, volume, and spread trend and control it. Failure to address it in a timely manner can lead to widespread oil pollution of seawater, causing mass deaths of marine fish, birds, algae, and marine mammals, and having long-term impacts on aquaculture, coastal tourism, and the marine ecological environment. Furthermore, the toxic benzene compounds in crude oil entering the food chain can ultimately pose a serious threat to human health.
[0003] Currently, there are two main methods used for monitoring marine oil spills: direct detection and remote sensing. Direct detection includes on-site detection and buoy detection. On-site detection involves personnel traveling by aircraft or ship to the oil spill area to visually inspect various information about the spill, or taking samples on-site for analysis and identification in a laboratory. This method requires a significant amount of time and manpower to locate and classify the oil spill area, and has a serious time lag. Buoy detection, on the other hand, uses sensors mounted on fixed buoys at sea to analyze samples and determine parameters such as the type and volume of oil spills. Buoy detection is highly sensitive and provides reliable data, but its detection coverage area is small and its detection efficiency is low. In addition, direct detection cannot monitor the large-scale oil slick spread caused by major oil spills in real time. Remote sensing monitoring, categorized by altitude, includes ground-based, airborne, and spaceborne platforms. It's a technology that senses electromagnetic waves, visible light, and infrared radiation reflected or emitted by distant targets, measuring, analyzing, and identifying target materials without direct contact. This technology can monitor oil spills over large areas, acquiring information such as the area, type, and thickness of the spill. Furthermore, continuous satellite tracking and observation of the spread direction can determine efficient methods for oil spill cleanup. However, remote sensing technology has limitations due to the long re-observation time and short spatial resolution of satellite platforms, as well as its susceptibility to weather and wind conditions. Airborne remote sensing monitoring of oil spills is expensive, lacks continuity, and has a small coverage area, making it unsuitable for offshore oil platforms located far from shore. Summary of the Invention
[0004] The purpose of this application is to provide a method for detecting and tracking marine oil spills, in order to solve the problems of small coverage area, low detection efficiency and high cost of existing marine oil spill detection and tracking methods.
[0005] To achieve the above objectives, this application provides a method for detecting and tracking marine oil spills, including:
[0006] Step S100: Establish and train a marine oil spill target detection model;
[0007] Step S200: Collect video information of the sea surface in real time, and split the video information into frames to obtain multiple frames of images;
[0008] Step S300: Input the i-th to i+n-th frame images into the trained marine oil spill target detection model. The marine oil spill target detection model automatically identifies the target of interest in each frame image and obtains n+1 identification results, where i≥1 and n≥2.
[0009] Step S400: Use inter-frame spatiotemporal correlation to determine the degree of association of the n+1 recognition results to obtain the pixel location information of the target of interest;
[0010] Step S500: Control the high-precision turntable movement using the pixel position information of the target of interest to lock onto and track the target of interest; and,
[0011] Step S600: Let i = i + 1, and return to step S300.
[0012] Optionally, the steps for training the marine oil spill target detection model include:
[0013] Acquire multiple high-quality images of the actual sea surface;
[0014] Noise is added to the actual sea surface image to obtain multiple low-quality sea surface images; and,
[0015] The marine oil spill target detection model is trained using the actual high-quality sea surface image and the low-quality sea surface image as samples.
[0016] Optionally, the number of low-quality sea surface images is greater than the number of high-quality actual sea surface images.
[0017] Optionally, noise can be added to the actual sea surface image using a simulation image generation method.
[0018] Optionally, the images from frame i to frame (i+n) can be subjected to Fourier transform before being input into the trained marine oil spill target detection model.
[0019] Optionally, the step of controlling the high-precision turntable movement using the position of the target of interest includes:
[0020] Convert the pixel location information of the target of interest into physical location information; and,
[0021] The high-precision turntable is controlled in real time using the physical location information of the target of interest.
[0022] Optionally, the marine oil spill target detection model is a ResNet-18 neural network model.
[0023] Optionally, a camera can be used to collect video information of the sea surface in real time, and the camera is mounted on the high-precision turntable.
[0024] The marine oil spill detection and tracking method provided in this application includes the following steps: S100: establishing and training a marine oil spill target detection model; S200: acquiring real-time video information of the sea surface and splitting the video information into frames to obtain multiple frames of images; S300: inputting the i-th to i+n-th frames of images into the trained marine oil spill target detection model, wherein the marine oil spill target detection model automatically identifies targets of interest in each frame of image to obtain n+1 identification results, where i≥1, n≥2; S400: using inter-frame spatiotemporal correlation to determine the degree of association of the n+1 identification results to obtain the pixel position information of the target of interest; S500: using the pixel position information of the target of interest to control the movement of a high-precision turntable to lock and track the target of interest; and S600: setting i = i+1 and returning to step S300. This application trains a marine oil spill target detection model and uses real-time video information collected from the sea surface to identify targets of interest. It combines inter-frame spatiotemporal tracking methods to obtain the pixel position information of the targets of interest, performs real-time feedback control on the high-precision turntable motion, locks and tracks the targets of interest, and simultaneously acquires video information for iterative processing, thereby achieving real-time detection and tracking of small targets with a large field of view in marine oil spills. Attached Figure Description
[0025] Figure 1 This is a flowchart of a marine oil spill detection and tracking method provided in an embodiment of this application. Detailed Implementation
[0026] The specific embodiments of this application will now be described in more detail with reference to the accompanying drawings. The advantages and features of this application will become clearer from the following description. It should be noted that the drawings are all in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of this application.
[0027] Figure 1This is a flowchart illustrating a method for detecting and tracking marine oil spills provided in this embodiment. Figure 1 As shown, the marine oil spill detection and tracking method includes:
[0028] Step S100: Establish and train a marine oil spill target detection model;
[0029] Step S200: Collect video information of the sea surface in real time, and split the video information into frames to obtain multiple frames of images;
[0030] Step S300: Input the i-th to i+n-th frame images into the trained marine oil spill target detection model. The marine oil spill target detection model automatically identifies the target of interest in each frame image and obtains n+1 identification results, where i≥1 and n≥2.
[0031] Step S400: Use inter-frame spatiotemporal correlation to determine the degree of association of the n+1 recognition results to obtain the pixel location information of the target of interest;
[0032] Step S500: Control the high-precision turntable movement using the pixel position information of the target of interest to lock onto and track the target of interest; and,
[0033] Step S600: Let i = i + 1, and return to step S300.
[0034] Specifically, step S100 is executed first to establish the marine oil spill target detection model. The marine oil spill target detection model can be a ResNet-18 neural network model. The ResNet-18 neural network model uses residual units, which enables the deep neural network to learn "residuals". This avoids the gradient vanishing problem during backpropagation when updating weights, reduces the training difficulty of the deep neural network, and makes the network performance better than that of a general convolutional neural network.
[0035] Next, the marine oil spill target detection model is trained. This includes acquiring multiple high-quality images of the actual sea surface; adding noise to these images to obtain multiple low-quality images; and using both the high-quality and low-quality images as samples to train the marine oil spill target detection model. A training database is constructed by combining the actual and low-quality images. The low-quality images are generated using a simulation image generation method, cleverly solving the problem of obtaining difficult-to-obtain actual sea surface images. By using the actual sea surface images as simulation input and adding noise to them through simulation, low-quality images with significant noise are obtained. This provides the clear-blurred image pairs needed for training the marine oil spill target detection model. The simulation method generates a large training set of samples, effectively solving the data problem required for training the marine oil spill target detection model.
[0036] Optionally, since high-quality images of the actual sea surface are difficult to obtain, the number of low-quality sea surface images can be greater than the number of high-quality images of the actual sea surface.
[0037] Next, step S200 is executed, which involves continuously acquiring video information of the target of interest (the oil spill area on the sea surface) using a camera or other equipment, and then splitting the video information into frames to obtain multiple frames of images.
[0038] In step S300, the images from frame i to frame (i+n) are input into the trained marine oil spill target detection model. The marine oil spill target detection model automatically identifies the target of interest in each frame, obtaining n+1 identification results, where i ≥ 1 and n ≥ 2. Since the images from frame i to frame (i+n) are continuous images, the n+1 identification results should have spatiotemporal correlation.
[0039] For example, in the initial state, i=1, n=9, the images from frame 1 to frame 10 are input into the trained marine oil spill target detection model, and the marine oil spill target detection model automatically identifies the target of interest in the images from frame 1 to frame 10, obtaining 10 identification results.
[0040] In this embodiment, the images from frame i to frame (i+n) are subjected to Fourier transform before being input into the trained marine oil spill target detection model. Transforming the images from frame i to frame (i+n) to the frequency domain before inputting them into the marine oil spill target detection model, and using frequency domain dot multiplication instead of time domain convolution calculations, results in less computation and lower complexity.
[0041] Step S400 involves using inter-frame spatiotemporal correlation to determine the correlation degree of the n+1 recognition results, thereby obtaining the pixel location information of the target of interest, i.e., the pixel location of the target of interest in the image. It should be understood that the pixel location information of the target of interest obtained at this time actually corresponds to the pixel location of the target of interest in the (i+n)th frame of the image.
[0042] In step S500, the pixel position information of the target of interest is converted into physical position information, that is, the physical position of the target of interest on the sea surface. The physical position information of the target of interest is used to control the movement of the high-precision turntable in real time to lock onto and track the target of interest. The high-precision turntable is equipped with a camera (for video acquisition) and a 3D high-precision image-stabilized gyroscope.
[0043] The calculations in step S500 are performed by the core control and processing module, which is the hardware core of the entire tracking and stabilization algorithm. This module is primarily responsible for the synchronization control of various modules, image processing, stabilization control, storage, and communication. In this application, an FPGA is used for hardware acceleration. After the camera acquires image data and caches it on the information processing and control board, the imaging algorithm module inside the FPGA reads the image data serially, pixel by pixel. The imaging algorithm module operates at a clock frequency of 125MHz, and the target detection and tracking algorithm module requires approximately 10ms to process one image, which meets the requirements for high-speed real-time tracking. Stabilization control is achieved using the aforementioned three-dimensional high-precision stabilization gyroscope and the high-precision turntable. The high-precision turntable is positioned and controlled using real-time detected physical positions to achieve high-precision locking and tracking of the target of interest. Using an integrated algorithm-hardware-control approach, target information is obtained through a real-time target detection algorithm based on video data. Combined with an inter-frame spatiotemporal tracking method, the core control and processing module performs real-time feedback control on the high-precision turntable, further acquiring video for iterative processing to achieve real-time detection and tracking of small targets with a large field of view.
[0044] Next, step S600 is executed, setting i = i + 1, and then returning to step S300. For example, the images from frame 2 to frame 11 are input into the trained marine oil spill target detection model. The marine oil spill target detection model automatically identifies the target of interest in the images from frame 2 to frame 11, obtaining 10 identification results. Then, step S400 obtains the pixel position of the target of interest in the image from frame 11.
[0045] In summary, the marine oil spill detection and tracking method provided in this application includes the following steps: S100: establishing and training a marine oil spill target detection model; S200: acquiring real-time video information of the sea surface and splitting the video information into frames to obtain multiple frames of images; S300: inputting the i-th to i+n-th frames of images into the trained marine oil spill target detection model, wherein the marine oil spill target detection model automatically identifies targets of interest in each frame of image to obtain n+1 identification results, where i≥1, n≥2; S400: using inter-frame spatiotemporal correlation to determine the degree of association of the n+1 identification results to obtain the pixel position information of the target of interest; S500: using the pixel position information of the target of interest to control the movement of a high-precision turntable to lock and track the target of interest; and S600: setting i = i+1 and returning to step S300. This application trains a marine oil spill target detection model and uses real-time video information collected from the sea surface to identify targets of interest. It combines inter-frame spatiotemporal tracking methods to obtain the pixel position information of the targets of interest, performs real-time feedback control on the high-precision turntable motion, locks and tracks the targets of interest, and simultaneously acquires video information for iterative processing, thereby achieving real-time detection and tracking of small targets with a large field of view in marine oil spills.
[0046] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.
[0047] It should also be noted that although preferred embodiments have been disclosed above, these embodiments are not intended to limit this application. Any person skilled in the art can make many possible variations and modifications to the technical solutions of this application, or modify them into equivalent embodiments, without departing from the scope of the technical solutions of this application. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of this application, without departing from the content of the technical solutions of this application, shall still fall within the scope of protection of the technical solutions of this application.
[0048] It should also be understood that, unless otherwise specified or indicated, the terms “first,” “second,” “third,” etc., in the specification are used only to distinguish the various components, elements, and steps in the specification, and not to indicate the logical or sequential relationships between the various components, elements, and steps.
[0049] Furthermore, it should be recognized that the terminology described herein is used only to describe particular embodiments and is not intended to limit the scope of this application. It must be noted that the singular forms “a” and “an” used herein and in the appended claims include plural bases unless the context clearly indicates otherwise. For example, a reference to “a step” or “an apparatus” means a reference to one or more steps or apparatuses, and may include secondary steps and secondary apparatuses. All conjunctions used should be understood in the broadest sense. Also, the word “or” should be understood to have the definition of logical “or” rather than logical “exclusive OR”, unless the context clearly indicates otherwise. Furthermore, implementations of the methods and / or devices in the embodiments of this application may include performing selected tasks manually, automatically, or in combination.
Claims
1. A method for detecting and tracking marine oil spills, characterized in that, include: Step S100: Establish and train a marine oil spill target detection model; Step S2 00: Real-time acquisition of video information from the sea surface, and splitting the video information into frames to obtain multiple frames of images; Step S300: Input the i-th to i+n-th frame images into the trained marine oil spill target detection model. The marine oil spill target detection model automatically identifies the target of interest in each frame image and obtains n+1 identification results, where i≥1 and n≥2. Step S400: Use inter-frame spatiotemporal correlation to determine the degree of association of the n+1 recognition results to obtain the pixel location information of the target of interest; Step S500: Control the high-precision turntable movement using the pixel position information of the target of interest to lock onto and track the target of interest; and, Step S600: Let i = i + 1, and return to step S300.
2. The marine oil spill detection and tracking method as described in claim 1, characterized in that, The steps for training the marine oil spill target detection model include: Acquire multiple high-quality images of the actual sea surface; Noise is added to the actual sea surface image to obtain multiple low-quality sea surface images; and, The marine oil spill target detection model is trained using the actual high-quality sea surface image and the low-quality sea surface image as samples.
3. The marine oil spill detection and tracking method as described in claim 2, characterized in that, The number of low-quality sea surface images is greater than the number of high-quality actual sea surface images.
4. The marine oil spill detection and tracking method as described in claim 2, characterized in that, Noise is added to the actual sea surface image using a simulation image generation method.
5. The marine oil spill detection and tracking method as described in claim 1, characterized in that, The images from frame i to frame (i+n) are subjected to Fourier transform and then input into the trained marine oil spill target detection model.
6. The marine oil spill detection and tracking method as described in claim 1, characterized in that, The steps of controlling the high-precision turntable movement using the position of the target of interest include: Convert the pixel location information of the target of interest into physical location information; and, The high-precision turntable is controlled in real time using the physical location information of the target of interest.
7. The marine oil spill detection and tracking method as described in claim 1, characterized in that, The marine oil spill target detection model is a ResNet-18 neural network model.
8. The marine oil spill detection and tracking method as described in claim 1, characterized in that, The camera is used to collect real-time video information of the sea surface, and the camera is set on the high-precision turntable.