A method, system and device for detecting oil slicks on water

Abstract

The application belongs to the technical field of oil pollution detection, and particularly relates to a water surface oil pollution detection method, system and equipment, which comprises a flexible printed circuit (FPC), an InGaAs infrared sensor, an infrared narrow-band filter and an optical lens. Light enters a visible light camera and the optical lens through an external lens, the visible light camera collects visible light imaging, the optical lens focuses a light beam, and then a specific wavelength of light is screened out through the near-infrared narrow-band filter, the screened light is received by the InGaAs infrared sensor, and the InGaAs infrared sensor is connected with the FPC. The application has the advantages that visible light imaging and short-wave infrared spectrum analysis technology are ingeniously combined, so that the accuracy and reliability of oil pollution detection are significantly improved. Under the background of increasingly severe environmental protection and water resource management, the application undoubtedly provides strong technical support for water surface pollution monitoring.

Description

Technical Field

[0001] This application belongs to the field of oil pollution detection technology, specifically relating to a method, system and equipment for detecting oil pollution on the water surface. Background Technology

[0002] With rapid societal development, the aging of industrial facilities has become increasingly prominent, especially metal pipelines such as oil pipelines, sewage pipelines, and natural gas pipelines. These pipelines are exposed to moisture and corrosive substances in transported raw materials for extended periods, inevitably leading to corrosion and damage, resulting in leaks and ultimately, severe ecological and environmental crises.

[0003] To address this issue, several solutions exist: First, using new materials and technologies to improve pipeline durability. However, given the vast number of existing pipelines, replacing the entire pipeline is not only costly but also potentially wasteful of resources. Second, real-time monitoring via embedded sensors. However, the effectiveness of this method is highly dependent on sensor accuracy; typically, sensors only trigger alarms when severe pipeline damage leads to pollution or environmental changes, exhibiting significant hysteresis. This is particularly problematic in underwater pipeline applications, where failure to detect leaks promptly poses a substantial environmental risk.

[0004] Therefore, there is an urgent need to develop more efficient and accurate monitoring technologies, and combine them with preventative maintenance strategies to reduce the environmental and economic risks caused by pipeline aging and leaks. Summary of the Invention

[0005] To address the aforementioned problems, this application utilizes multispectral imaging technology, combining visible light with short-wave infrared spectroscopy, to achieve efficient detection of oil spills on water surfaces. The technical solution is as follows:

[0006] A method for detecting oil pollution on the water surface includes the following steps:

[0007] S1. Data acquisition and preprocessing;

[0008] S2. Construct the graph structure of the oil film region;

[0009] S3. Construct a dual-link detection model, including a main detection network and a graph-guided network. The main detection network is used to detect visible light images in real time, and the graph-guided network is used to model the spectral-spatial correlation characteristics of the oil film to achieve oil film contamination monitoring.

[0010] Preferably, in step S1, data acquisition involves ensuring the spatiotemporal alignment of the dual-mode data because the oil film exhibits significantly different physical properties in the visible and infrared bands. The method is as follows:

[0011] S101. Arrange heatable metal dots on the calibration plate;

[0012] S102. Simultaneously acquire visible light images and infrared thermal images;

[0013] S103. Achieve pixel-level alignment by solving the perspective transformation matrix:

[0014] ;

[0015] in, These are the pixel coordinates of the visible light image. Let H be the pixel coordinates of the infrared image, and H be a 3×3 homography matrix (solved using the least squares method).

[0016] Preferably, the data acquisition preprocessing is as follows:

[0017] Differentiated preprocessing is employed to address the characteristics of different modalities:

[0018] Visible light channel:

[0019] Enhance the iridescent effect of the oil film using an improved MSRCR algorithm:

[0020] ;

[0021] in, This represents the pixel value at coordinates (x, y) in the enhanced visible light image. This is the weighting coefficient (default is 0.33). This represents the pixel value of the original visible light image at coordinates (x, y). For Gaussian kernel, To enhance the strength parameter (typically 0.5), Color restoration factor;

[0022] Infrared band channel:

[0023] Oil film characteristic peaks:

[0024] ;

[0025] For difference window, It is the wavelength of infrared waves.

[0026] Preferably, step S2, which involves constructing the graph structure of the oil film region, is as follows:

[0027] S21. Node Generation Strategy:

[0028] S210. Initial Segmentation: Input the visible light image into the lightweight DeepLabv3+ network to obtain a pixel-level segmentation mask. Simultaneously, use an infrared spectral angle mapping algorithm to calculate the similarity between each pixel and the standard oil film spectrum.

[0029] ;

[0030] in, Let p be the infrared spectral vector. This is the standard oil film spectral template vector; This indicates finding the norm of a vector;

[0031] Regions with SAM values ​​< 0.2 rad are selected as candidate oil film areas;

[0032] S211. Superpixel clustering:

[0033] An improved SLIC superpixel algorithm is applied to the candidate regions, introducing spectral distance constraints:

[0034] ;

[0035] in, For color space Euclidean distance. Distance in pixel coordinates denoted as Euclidean distance of the infrared band vector, where m, s, and n are normalization coefficients.

[0036] S22. Edge connection mechanism: S220: The establishment of edges must consider both spatial adjacency and spectral similarity.

[0037] Spatial constraints: Delaunay triangulation is used to establish the initial topology, and the maximum side length is set;

[0038] Spectral verification: for each candidate edge Calculate its infrared feature correlation coefficient:

[0039] ;

[0040] in, Let be the infrared feature vector of node i. Let be the infrared feature vector of node j. Infrared feature vectors representing nodes i and j and covariance; and Let represent the standard deviations of the infrared feature vectors of node i and node j, respectively.

[0041] reserve Edges are used to suppress spurious connections;

[0042] S221: Dynamic Update:

[0043] The graph structure is re-evaluated every M frames, and a node merging strategy is used to reduce computational complexity.

[0044] = ;

[0045] in, and They are nodes and nodes spatial coordinates, Represents a node and The Euclidean distance between the position coordinates.

[0046] Preferably, in step S3, the main detection network adopts the YOLOv8 architecture, and a spectral attention sublayer is added to its RGB branch. Spectral pyramid pooling is designed on the IR branch to convert the 128-dimensional spectral vector into multi-scale features, as shown in the following expression:

[0047] ;

[0048] in, This indicates a join operation, which concatenates features resulting from different operations. To perform max pooling on the spectral vector v, the pooling kernel size is 1; To perform average pooling on the spectral vector v, the pooling kernel size is 3; To perform a convolution operation on the spectral vector v, the kernel size is 5.

[0049] Introduce a cross-modal attention mechanism in the FPN layer:

[0050] ;

[0051] in, This represents the activation function, which maps input values ​​to the interval (0,1). This is a visible light feature map. Infrared feature map, This indicates a convolution operation, used to process the concatenated features.

[0052] The detection head outputs key points to predict the oil film contour points.

[0053] Preferably, the graph-guided network in step S3 adopts the GMM model, and the specific method is as follows:

[0054] Node coding layer:

[0055] Using a 3-layer GATv2 network, the number of attention heads in each layer... Node update formula:

[0056] ;

[0057] in, Indicates the updated node eigenvectors; The activation function performs a nonlinear transformation on the summation result; Represents a node The set of neighboring nodes; The attention coefficient is calculated using a gating mechanism. This is a trainable weight matrix used to transform the feature vectors of neighboring nodes; For nodes eigenvectors.

[0058] Image pooling layer:

[0059] Using the TopKPooling method, the top 30% of nodes with the highest importance are retained:

[0060] ;

[0061] It is a node The importance score is used to select and retain nodes with high importance during graph pooling; These are trainable vectors used to calculate the importance of nodes; For nodes eigenvectors; Representing vectors The norm;

[0062] Image readout layer:

[0063] Generate graph-level representations using dynamic routing algorithms:

[0064] ;

[0065] Graph sparsification: Use the K-NN strategy to limit the maximum number of edges in each node;

[0066] in, This represents the total number of nodes in the graph; Represents a node The importance coefficient, calculated from the data, is used to measure the importance of nodes. The degree of contribution to graph-level representation; Represents a node eigenvectors; These are trainable vectors used to calculate the importance coefficients of nodes; For vectors Transpose of; This indicates exponentiation.

[0067] Preferably, it also includes S4 early warning, establishing a three-level early warning mechanism:

[0068] Monitoring Level 1: Events are recorded when ThreatScore > 0.5;

[0069] Warning Level 2: A sound and light alarm is triggered when the score is greater than 0.7 for more than 3 consecutive frames.

[0070] Emergency Level 3: If the score is greater than 0.9, the pollution source tracing procedure will be automatically initiated.

[0071] A water surface oil pollution detection and monitoring system includes a data acquisition module, a data processing module, and a data output module;

[0072] Data acquisition module: Acquires visible light and infrared spectral data, and calibrates the data;

[0073] Data processing module: preprocesses the acquired 3D data, constructs the graph structure of the oil film area, and uses a dual-link detection model to analyze oil film contamination;

[0074] Data output module: Provides early warnings based on oil pollution monitoring results.

[0075] A water surface oil pollution detection and monitoring device includes an InGaAs infrared sensor, a near-infrared narrowband filter, and an optical lens. Light enters a visible light camera and an optical lens through an external lens. The visible light camera acquires visible light images, and the optical lens focuses the light beam. Then, the light of a specific wavelength is filtered out by the near-infrared narrowband filter. The filtered light is received by the InGaAs infrared sensor, which is connected to a flexible circuit board (FPC).

[0076] Preferably, the visible light imaging captured by the camera can provide graphic and boundary information of the oil spill, while the short-wave infrared spectrum can further identify the type and concentration of the oil spill through the difference in spectral reflection and absorption intensity; the visible light camera and optical lens need to zoom synchronously to ensure the synchronization of imaging.

[0077] Compared with the prior art, the beneficial effects of this application are as follows:

[0078] 1. Water and oil films exhibit different reflection characteristics to light of different wavelengths. By analyzing the variation of reflection intensity with wavelength, it can be determined that the difference in reflection intensity between water and oil films is most significant within a specific wavelength range. Selecting this wavelength band for filtering can effectively distinguish the reflected light from water and oil films, thereby achieving accurate detection of oil films.

[0079] 2. The synchronous zoom mechanism achieves synchronized zooming of the two lens groups through gear connection, ensuring the synchronicity and consistency of imaging. This design not only improves detection accuracy but also reduces misjudgments caused by inconsistent focal lengths.

[0080] 3. High-precision data analysis technology, through multispectral data fusion and algorithmic identification, can automatically process visible and infrared spectral data and combine the visible and infrared characteristics of oil stains for comprehensive judgment. This high-precision data analysis technology significantly improves the accuracy and reliability of detection. Attached Figure Description

[0081] Figure 1 This is a schematic diagram of the structure of this application.

[0082] 1-Optical lens, 2-Near-infrared narrowband filter, 3-InGaAs infrared sensor, 4-Flexible circuit board (FPC), 5-Lens, 6-Lens mount. Detailed Implementation

[0083] The technical solution of this application will be described in detail below with reference to specific embodiments and accompanying drawings. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of this application, rather than limitations thereof. Specific technical features can be combined with each other.

[0084] A water surface oil pollution detection and monitoring device includes a visible light camera and a near-infrared camera. The near-infrared camera includes an InGaAs infrared sensor 3, a near-infrared narrowband filter 2, and an optical lens 1. Light enters the visible light camera and the optical lens 1 through an external lens. The visible light camera captures visible light images, and the optical lens 1 focuses the light beam. Then, the light of a specific wavelength is filtered out by the near-infrared narrowband filter 2. The filtered light enters the lens 5 and is received by the InGaAs infrared sensor 3. The InGaAs infrared sensor 2 is connected to a flexible circuit board (FPC).

[0085] Lens 5 is mounted on lens mount 6, and near-infrared narrowband filter 2 can be attached to lens 5.

[0086] Visible light cameras can capture visible light images that provide graphic and boundary information of oil spills, while near-infrared cameras can further identify the type and concentration of oil spills by the differences in spectral reflectance and absorption intensity. Visible light cameras and optical lenses need to zoom synchronously to ensure the synchronicity and consistency of imaging.

[0087] A water surface oil pollution detection and monitoring system includes a data acquisition module, a data processing module, and a data output module;

[0088] Data acquisition module: Acquires visible light and infrared spectral data, and calibrates the data;

[0089] Data processing module: preprocesses the acquired 3D data, constructs the graph structure of the oil film area, and uses a dual-link detection model to analyze oil film contamination;

[0090] Data output module: Provides early warnings based on oil pollution monitoring results.

[0091] Near-infrared cameras can more accurately identify the type and composition of oil spills by capturing the spectral differences between oil films and water bodies in the infrared band. Different types of oil spills exhibit different absorption and reflection characteristics in the infrared band, which can be captured and analyzed by near-infrared cameras, thus enabling precise identification of oil spills.

[0092] When the visible light camera detects an abnormal area on the water surface, the near-infrared camera zooms in sync with the visible light camera to perform multispectral analysis on the suspected leak point. Because oil films and water bodies have significantly different spectral reflectance and absorption characteristics in the short-wave infrared band, the near-infrared camera can capture these differences to accurately identify the spectral characteristics of oil spills. For example, crude oil exhibits a distinct absorption peak in a specific infrared band, while water may show a higher reflectivity. By comparing and analyzing data from the visible light and short-wave infrared bands, the system can more comprehensively identify the presence and characteristics of oil spills.

[0093] As attached Figure 1As shown in the diagram, the core structure and working principle of a near-infrared camera are illustrated in detail. When light enters the camera from the external environment, it first passes through an external lens and then is focused by an optical lens system to ensure that the light accurately converges onto the sensor's photosensitive area. The focused light then passes through a near-infrared narrowband filter, which filters out near-infrared light of a specific wavelength and removes stray light from other bands, ensuring that only light of the target wavelength can enter the subsequent sensor section. This filtering process is crucial for improving imaging accuracy and signal-to-noise ratio. The filtered near-infrared light is then received by an InGaAs infrared sensor. InGaAs sensors are high-performance infrared detectors with advantages such as high sensitivity, wide spectral response range, and low noise, making them particularly suitable for near-infrared detection. After receiving the light signal, the sensor converts it into an electrical signal and connects it to an FPC (flexible printed circuit board) via gold wire. Gold wire, as a highly conductive and low-resistance connecting material, ensures the stability and reliability of the signal during transmission. The other end of the FPC is connected to a computer or other data processing equipment, responsible for transmitting the data collected by the sensor to a network or local system for further processing. This design not only simplifies the camera's structure but also enhances the system's flexibility and scalability. In this way, the near-infrared camera can efficiently capture, transmit, and process near-infrared light signals, making it widely applicable in night vision, security monitoring, medical imaging, and industrial inspection. The near-infrared camera's structural design fully considers the synergistic operation of optics, electronics, and mechanics, ensuring high performance in complex environments. Through the organic combination of optical lenses, narrowband filters, InGaAs sensors, and FPCs, the camera can efficiently complete the acquisition, transmission, and processing of light signals, providing a reliable solution for various application scenarios.

[0094] In practical applications, the collaborative working mechanism of this system demonstrates powerful detection capabilities. The system can monitor the water surface in real time using a visible light camera. Once an abnormal area is detected, the near-infrared camera immediately performs multispectral analysis to determine the type and extent of the oil spill. This collaborative mechanism not only improves detection efficiency but also significantly reduces the false alarm rate, especially under complex environmental conditions such as changes in lighting and water surface fluctuations. The system can also monitor the water surface in real time through the collaborative operation of a visible light camera and a short-wave infrared camera (near-infrared camera), enabling timely detection and handling of oil spill incidents. This real-time monitoring and rapid response mechanism effectively prevents the spread of oil spills and minimizes environmental pollution and economic losses. The collaborative working mechanism of the visible light camera and the short-wave infrared camera greatly improves the accuracy and reliability of detection.

[0095] All data collected by visible light and shortwave infrared cameras is aggregated and fed into efficient data analysis algorithms. These algorithms combine the characteristics of oil slicks on the water surface in the visible and infrared bands, using machine learning or deep learning models to perform multi-dimensional analysis of the data. For example, the algorithms can extract the color and texture features of the oil film in visible light images, while combining them with features such as spectral reflectance and absorptivity in shortwave infrared images to construct a comprehensive discrimination model. This multimodal data analysis method not only improves the accuracy of oil spill detection but also effectively reduces the false alarm rate, especially under complex environmental conditions (such as changes in lighting and water surface fluctuations).

[0096] Furthermore, the system's collaborative working mechanism is also reflected in real-time data processing and feedback. When suspected oil pollution is detected, the system immediately triggers an alarm and uploads relevant data to the monitoring center for further analysis and decision-making. Simultaneously, the system can automatically adjust camera parameters (such as focal length and exposure time) based on the detection results to optimize the quality of subsequent data acquisition. This intelligent detection process not only improves detection efficiency but also provides reliable data support for subsequent pollution control. Through the collaborative work of a visible light camera and a short-wave infrared camera, combined with advanced data analysis algorithms, the system can achieve high-precision, real-time detection of oil pollution on the water surface. This multispectral fusion detection method not only overcomes the limitations of single-band detection but also provides an efficient and reliable solution for monitoring oil pollution on the water surface, with broad application prospects in environmental protection, industrial safety, and other fields.

[0097] A method for detecting oil pollution on the water surface includes the following steps:

[0098] S1. Data acquisition and preprocessing;

[0099] Data acquisition in step S1: Since the oil film exhibits significantly different physical properties in the visible and infrared bands, it is necessary to ensure the spatiotemporal alignment of the dual-mode data. The method is as follows:

[0100] S101. Arrange heatable metal dots on the calibration plate;

[0101] S102. Simultaneously acquire visible light images and infrared thermal images;

[0102] S103. Achieve pixel-level alignment by solving the perspective transformation matrix:

[0103] ;

[0104] in, The original pixel intensity value. express, This is the weighting coefficient (default is 0.33). This represents the pixel value of the original visible light image at coordinates (x, y). For Gaussian kernel, To enhance the strength parameter (typically 0.5), Color restoration factor;

[0105] Data acquisition and preprocessing are as follows:

[0106] Differentiated preprocessing is employed to address the characteristics of different modalities:

[0107] Visible light channel:

[0108] Enhance the iridescent effect of the oil film using an improved MSRCR algorithm:

[0109] ;

[0110] in, This represents the pixel value at coordinates (x, y) in the enhanced visible light image. This is the weighting coefficient (default is 0.33). For Gaussian kernel, To enhance the strength parameter (typically 0.5), Color restoration factor;

[0111] Infrared band channel:

[0112] Oil film characteristic peaks:

[0113] ;

[0114] For difference window, It is the wavelength of infrared waves.

[0115] S2. Construct the graph structure of the oil film region;

[0116] Step S2, which involves constructing the graph structure of the oil film region, is as follows:

[0117] S21. Node Generation Strategy:

[0118] S210. Initial Segmentation: Input the visible light image into the lightweight DeepLabv3+ network to obtain a pixel-level segmentation mask. Simultaneously, use an infrared spectral angle mapping algorithm to calculate the similarity between each pixel and the standard oil film spectrum.

[0119] ;

[0120] in, Let p be the infrared spectral vector. This is the standard oil film spectral template vector; This indicates finding the norm (modulus) of a vector.

[0121] Regions with SAM values ​​< 0.2 rad are selected as candidate oil film areas;

[0122] S211. Superpixel clustering:

[0123] An improved SLIC superpixel algorithm is applied to the candidate regions, introducing spectral distance constraints:

[0124] ;

[0125] in, For color space Euclidean distance. Distance in pixel coordinates denoted as Euclidean distance of the infrared band vector, where m, s, and n are normalization coefficients.

[0126] S22. Edge connection mechanism:

[0127] S220: The establishment of edges must consider both spatial adjacency and spectral similarity:

[0128] Spatial constraints: Delaunay triangulation is used to establish the initial topology, and the maximum side length is set;

[0129] Spectral verification: for each candidate edge Calculate its infrared feature correlation coefficient:

[0130] ;

[0131] in, Let be the infrared feature vector of node i. Let be the infrared feature vector of node j. Infrared feature vectors representing nodes i and j and covariance; and Let represent the standard deviations of the infrared feature vectors of node i and node j, respectively.

[0132] reserve Edges are used to suppress spurious connections;

[0133] S221: Dynamic Update:

[0134] The graph structure is re-evaluated every M frames, and a node merging strategy is used to reduce computational complexity.

[0135] = .

[0136] in, and They are nodes and nodes spatial coordinates, Represents a node and The Euclidean distance between the position coordinates.

[0137] S3. Construct a dual-link detection model, including a main detection network and a graph-guided network. The main detection network is used to detect visible light images in real time, and the graph-guided network is used to model the spectral-spatial correlation characteristics of the oil film to achieve oil film contamination monitoring.

[0138] In step S3, the main detection network adopts the YOLOv8 architecture, and a spectral attention sublayer is added to its RGB branch. Spectral pyramid pooling is designed on the IR branch to convert the 128-dimensional spectral vector into multi-scale features, as shown in the following expression:

[0139] ;

[0140] in, This indicates a join operation, which concatenates features resulting from different operations. To perform max pooling on the spectral vector v, the pooling kernel size is 1; To perform average pooling on the spectral vector v, the pooling kernel size is 3; To perform a convolution operation on the spectral vector v, the kernel size is 5.

[0141] Introduce a cross-modal attention mechanism in the FPN layer:

[0142] ;

[0143] in, This represents the activation function, which maps input values ​​to the interval (0,1). This is a visible light feature map. Infrared feature map, This indicates a convolution operation, used to process the concatenated features.

[0144] The detection head outputs key points to predict the oil film contour points.

[0145] In step S3, the graph-guided network uses the GMM model, and the specific method is as follows:

[0146] Node coding layer:

[0147] Using a 3-layer GATv2 network, the number of attention heads in each layer... Node update formula:

[0148] ;

[0149] in, Indicates the updated node eigenvectors; The activation function performs a nonlinear transformation on the summation result; Represents a node The set of neighboring nodes; The attention coefficient is calculated using a gating mechanism. This is a trainable weight matrix used to transform the feature vectors of neighboring nodes; For nodes eigenvectors.

[0150] Image pooling layer:

[0151] Using the TopKPooling method, the top 30% of nodes with the highest importance are retained:

[0152] ;

[0153] It is a node The importance score is used to select and retain nodes with high importance during graph pooling; These are trainable vectors used to calculate the importance of nodes; For nodes eigenvectors; Representing vectors The norm (modulus).

[0154] Image readout layer:

[0155] Generate graph-level representations using dynamic routing algorithms:

[0156] ;

[0157] Graph sparsification: The k-NN strategy is used to limit the maximum number of edges in each node;

[0158] in, This represents the total number of nodes in the graph; Represents a node The importance coefficient, calculated from the data, is used to measure the importance of nodes. The degree of contribution to graph-level representation; Represents a node eigenvectors; These are trainable vectors used to calculate the importance coefficients of nodes; For vectors Transpose of; This represents the natural constant, which is used here in exponential operations.

[0159] in, This represents the total number of nodes in the graph; The importance coefficient of node i is calculated and used to measure the degree of contribution of node i to the graph-level representation. Represents the feature vector of node i; These are trainable vectors used to calculate the importance coefficients of nodes; This is the transpose of vector u; This represents the natural constant, which is used here in exponential operations.

[0160] S4 early warning system, establishing a three-tiered early warning mechanism:

[0161] Monitoring Level 1: Events are recorded when ThreatScore > 0.5;

[0162] Warning Level 2: A sound and light alarm is triggered when the score is greater than 0.7 for more than 3 consecutive frames.

[0163] Emergency Level 3: If the score is greater than 0.9, the pollution source tracing procedure will be automatically initiated.

[0164] In summary, the surface oil pollution detection system described in this patent application achieves high-precision, real-time detection of surface oil pollution through the collaborative operation of a visible light camera and a short-wave infrared camera, combined with advanced data analysis algorithms. This system not only overcomes the limitations of single-band detection but also provides an efficient and reliable solution for surface oil pollution monitoring. It has broad application prospects and enormous market potential in fields such as environmental protection and industrial safety. With continuous technological advancements and deeper applications, it is believed that this system will play an increasingly important role in future surface oil pollution monitoring, contributing significantly to the protection of our water resources and environmental safety.

[0165] It is worth noting that the design and implementation of this system involves cutting-edge knowledge in multiple technical fields, including optical imaging technology, image processing technology, and machine learning algorithms. During the system's development, we fully drew upon the latest research results and practical experience in related fields both domestically and internationally, and made innovative improvements and optimizations based on actual needs. Therefore, this system not only possesses high technological advancement and practicality, but also excellent scalability and maintainability, enabling it to meet application requirements in various scenarios.

[0166] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for detecting oil pollution on the water surface, characterized in that, Includes the following steps: S1. Data acquisition and preprocessing; S2. Construct the graph structure of the oil film region; Step S2, which involves constructing the graph structure of the oil film region, is as follows: S21. Node Generation Strategy: S210. Initial Segmentation: Input the visible light image into the lightweight DeepLabv3+ network to obtain a pixel-level segmentation mask. Simultaneously, use an infrared spectral angle mapping algorithm to calculate the similarity between each pixel and the standard oil film spectrum. ； in, For pixels p infrared spectral vector, This is the standard oil film spectral template vector; This indicates finding the norm of a vector; Regions with SAM values ​​< 0.2 rad are selected as candidate oil film areas; S211. Superpixel clustering: An improved SLIC superpixel algorithm is applied to the candidate regions, introducing spectral distance constraints: ； in, For color space Euclidean distance. Distance in pixel coordinates is the Euclidean distance of the infrared band vector, and m, s, and n are normalization coefficients; S22. Edge connection mechanism: S220: The establishment of edges must simultaneously consider spatial adjacency and spectral similarity: Spatial constraints: Delaunay triangulation is used to establish the initial topology, and the maximum edge length is set; Spectral verification: for each candidate edge Calculate its infrared feature correlation coefficient: ； in, For nodes i Infrared feature vectors, For nodes j Infrared feature vectors, Represents a node i and j infrared feature vectors and covariance; and Representing nodes respectively i and nodes j Standard deviation of infrared eigenvectors; reserve Edges are used to suppress spurious connections; S221: Dynamic Update: The graph structure is re-evaluated every M frames, and a node merging strategy is used to reduce computational complexity. = ； in, and They are nodes and nodes spatial coordinates, Represents a node and The Euclidean distance between the position coordinates; S3. Construct a dual-link detection model, including a main detection network and a graph-guided network. The main detection network is used to detect visible light images in real time, and the graph-guided network is used to model the spectral-spatial correlation characteristics of the oil film to achieve oil film contamination monitoring. In step S3, the main detection network adopts the YOLOv8 architecture, and a spectral attention sublayer is added to its RGB branch. Spectral pyramid pooling is designed on the IR branch to convert the 128-dimensional spectral vector into multi-scale features, as shown in the following expression: ； in, This indicates a join operation, which concatenates features resulting from different operations. For the spectral vector v Perform max pooling operation; For the spectral vector v Perform average pooling operation; For the spectral vector v Perform convolution operations; Introduce a cross-modal attention mechanism in the FPN layer: ； in, This represents the activation function, which maps input values ​​to the interval (0,1). This is a visible light feature map. Infrared feature map, This indicates a convolution operation, used to process the concatenated features; The detection head outputs key points to predict the oil film contour points.

2. The method for detecting oil pollution on the water surface according to claim 1, characterized in that, Data acquisition in step S1: Since the oil film exhibits significantly different physical properties in the visible and infrared bands, it is necessary to ensure the spatiotemporal alignment of the dual-mode data. The method is as follows: S101. Arrange heatable metal dots on the calibration plate; S102. Simultaneously acquire visible light images and infrared thermal images; S103. Solving the perspective transformation matrix Achieve pixel-level alignment: ； in, These are the pixel coordinates of the visible light image. These are the pixel coordinates corresponding to the infrared image. H It is a 3×3 homography matrix.

3. The method for detecting oil pollution on the water surface according to claim 1, characterized in that, Data acquisition and preprocessing are as follows: Differentiated preprocessing is employed to address the characteristics of different modalities: Visible light channel: Employing the MSRCR algorithm to enhance the iridescent effect of the oil film ; Infrared band channel: Oil film characteristic peaks: ； For difference window, It is the wavelength of infrared waves.

4. The method for detecting oil pollution on the water surface according to claim 1, characterized in that, In step S3, the graph-guided network uses the GMM model, and the specific method is as follows: Node coding layer: Using a 3-layer GATv2 network, the number of attention heads in each layer... Node update formula: ； in, Indicates the updated node eigenvectors; The activation function performs a nonlinear transformation on the summation result; Represents a node The set of neighboring nodes; The attention coefficient is calculated using a gating mechanism. This is a trainable weight matrix used to transform the feature vectors of neighboring nodes; For nodes eigenvectors; Image pooling layer: Using the TopKPooling method, the top 30% of nodes with the highest importance are retained: ； It is a node The importance score is used to select and retain nodes with high importance during graph pooling; These are trainable vectors used to calculate the importance of nodes; For nodes eigenvectors; Representing vectors The norm; Image readout layer: Generate graph-level representations using dynamic routing algorithms: ； Graph sparsification: Use the K-NN strategy to limit the maximum number of edges in each node; in, This represents the total number of nodes in the graph; Represents a node Importance coefficient, Represents a node eigenvectors; These are trainable vectors used to calculate the importance coefficients of nodes; For vectors Transpose of; This indicates exponentiation.

5. The method for detecting oil pollution on the water surface according to claim 4, characterized in that, This also includes S4 early warning, establishing a three-tiered early warning mechanism: Monitoring Level 1: Events are recorded when ThreatScore > 0.5; Warning Level 2: A sound and light alarm is triggered when the score is greater than 0.7 for more than 3 consecutive frames. Emergency Level 3: If the score is greater than 0.9, the pollution source tracing procedure will be automatically initiated.

6. A water surface oil pollution detection and monitoring system, characterized in that, The method for detecting and monitoring oil pollution on the water surface according to any one of claims 1-5 includes a data acquisition module, a data processing module, and a data output module; Data acquisition module: Acquires visible light and infrared spectral data, and calibrates the data; Data processing module: preprocesses the acquired 3D data, constructs the graph structure of the oil film area, and uses a dual-link detection model to analyze oil film contamination; Data output module: Provides early warnings based on oil pollution monitoring results.

7. A device for detecting and monitoring oil pollution on the water surface, characterized in that, The method for detecting and monitoring oil pollution on the water surface according to any one of claims 1-5 includes an InGaAs infrared sensor, a near-infrared narrowband filter, and an optical lens. Light enters the visible light camera and the optical lens through an external lens. The visible light camera acquires visible light images, and the optical lens focuses the light beam. Then, the light of a specific wavelength is filtered out by the near-infrared narrowband filter. The filtered light is received by the InGaAs infrared sensor, which is connected to a flexible circuit board (FPC).

8. The water surface oil pollution detection and monitoring device according to claim 7, characterized in that, Visible light imaging captured by a visible light camera can provide graphic and boundary information of oil spills, while short-wave infrared spectroscopy can further identify the type and concentration of oil spills through differences in spectral reflection and absorption intensity; the visible light camera and optical lens need to zoom synchronously to ensure the synchronization of imaging.

Images