A marine oil spill monitoring method and system based on a multi-band fluorescence spectrum neural network

Abstract

The present application relates to the technical field of marine oil spill monitoring, in particular to a marine oil spill monitoring method and system based on multi-band fluorescence spectrum neural network. The method comprises the following steps: acquiring fluorescence spectrum and fluorescence image data of different oil types and different oil film thicknesses of oil spills; preprocessing the fluorescence spectrum and fluorescence image data of marine oil spills respectively; constructing three-dimensional fluorescence spectrum data and thickness difference matrix by using the preprocessed fluorescence spectrum; extracting band features and thickness features of the three-dimensional fluorescence spectrum data and thickness difference matrix by using a deep neural network; the present application effectively integrates marine oil spill fluorescence spectrum and fluorescence image data, and provides a reliable basis for accurate identification and early warning of oil spills.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine oil spill monitoring, in particular to a marine oil spill monitoring method and system based on a multi-band fluorescence spectrum neural network. BACKGROUND

[0002] The sources of marine oil spills usually include natural seepage, ship accidents and leaks from offshore engineering (drilling platforms, oil pipelines, etc.). Oil is a complex mixture containing many compounds, and the hydrocarbon organic matter contains many highly toxic and persistent harmful substances. Not only does the change in the physical properties of the ocean water cause irreversible damage, but it also destroys marine biodiversity and even causes biological extinction. Therefore, rapid detection and monitoring of marine oil spills are of great significance to the ecological environment of the deep sea and coastal areas.

[0003] Traditional remote sensing technology is a common marine oil spill monitoring and early warning method, and the related satellite, airborne and shipborne instruments and equipment are relatively mature. The principle is based on the large reflectivity difference between oil film and seawater in different wave bands (microwave, visible light and infrared wave bands). Static images or dynamic video information formed by reflected light spectrum are used for marine oil spill monitoring and early warning. However, the reflected light spectrum information of marine oil spills and seawater only has overall differences, and remote sensing technology still has certain difficulties in identifying oil types and inverting oil film thickness. Moreover, traditional remote sensing technology relies on sunlight as the light source, which is affected by fog, clouds and other weather conditions, and cannot be used at night. Ultraviolet-induced fluorescence technology is an active remote sensing detection technology. According to the laser excitation of some organic matter contained in oil at the ultraviolet wave band, the oil can emit fluorescence with a longer wavelength than the excitation wavelength. The sensing and detection method is based on the analysis of the characteristic peaks and intensity information of the fluorescence spectrum. It has strong detection capability for different oil types and thin oil films. Through the analysis and processing of fluorescence information, real-time monitoring and rapid early warning of marine oil spills can be achieved.

[0004] In the existing marine oil spill monitoring using fluorescence information, the fluorescence information in a limited range is usually obtained at a fixed vertical angle, ignoring the fluorescence directional pattern characteristics of marine oil spills under the action of wind and waves. Secondly, the fluorescence spectrum can only collect point data in a small area, and the fluorescence spectrum characteristics of different oil types are obvious. However, although the fluorescence image can collect data in a larger area, the feature difference in the image is small and cannot effectively distinguish oil types and oil film thickness. By using the characteristic information of the fluorescence spectrum and fusing it into the analysis and processing of the fluorescence image, the problem of breaking through the fluorescence spectrum point detection to the fluorescence image large-range accurate detection is solved, and the problem of heterogeneous data fusion of fluorescence spectrum and image is solved, thereby improving the accuracy of the monitoring and early warning model. SUMMARY

[0005] In order to solve the above-mentioned problems, the application provides a marine oil spill monitoring method and system based on a multi-band fluorescence spectrum neural network. The application is based on fluorescence spectrum characteristics and fluorescence image data, and fuses spectrum and image heterogeneous data characteristics, thereby providing a new method for marine oil spill monitoring and early warning.

[0006] In the first aspect, the application provides a marine oil spill monitoring method based on a multi-band fluorescence spectrum neural network, which adopts the following technical scheme:

[0007] A marine oil spill monitoring method based on a multi-band fluorescence spectrum neural network comprises the following steps:

[0008] Fluorescence spectrum and fluorescence image data of different oil types and different oil film thicknesses of oil spills are acquired.

[0009] The fluorescence spectrum and fluorescence image data of the marine oil spill are preprocessed respectively.

[0010] Three-dimensional fluorescence spectrum data and thickness difference matrices are constructed by using the preprocessed fluorescence spectrum.

[0011] Band features and thickness features are extracted from the three-dimensional fluorescence spectrum data and the thickness difference matrices by using a deep neural network.

[0012] Feature extraction is performed on the fluorescence image data based on a joint convolutional neural network.

[0013] The extracted features are analyzed and predicted after feature fusion based on a Lasso regression model.

[0014] An early warning result is output.

[0015] Further, the preprocessing of the fluorescence spectrum and fluorescence image data of the marine oil spill respectively comprises moving average smoothing and Logistic normalization of the fluorescence spectrum data of the marine oil spills of different oil types and different oil film thicknesses, so as to ensure the consistency and integrity of the data. The moving average smoothing is represented as follows:

[0016] ,

[0017] wherein, represents the smoothed result of the fluorescence spectrum intensity at x i , f(x i-1 ), f(x i ) and f(x i+1 ) respectively represent the measured fluorescence spectrum intensity at x i-1 , x i and x i+1 , and the Logistic function is represented as follows: The Logistic normalization process is represented as follows:

[0018] ,

[0019] wherein e represents the base of natural logarithm, g(x i ) represents the fluorescence spectrum intensity after the Logistic function operation at x i wavelength, f(x i ) represents the fluorescence spectrum intensity measured at x i wavelength, represents the normalized result of the fluorescence spectrum intensity at x i wavelength, g(x) max and g(x) min respectively represent the maximum and minimum values of the fluorescence spectrum intensity after the Logistic function operation.

[0020] Further, the preprocessing of the fluorescence spectrum and the fluorescence image data of the marine oil spill respectively further comprises preprocessing of the fluorescence image, edge detection is performed by using Sobel operators in horizontal and vertical directions respectively, so as to detect the edge of the glassware containing the marine oil spill sample, and the fluorescence image is cropped according to the edge detection result, and only the image of the simulated marine oil spill area is reserved.

[0021] Further, the construction of the three-dimensional fluorescence spectrum data and the thickness difference matrix by using the preprocessed fluorescence spectrum comprises the following steps: a mathematical model between the fluorescence spectrum intensity and the oil film thickness is constructed by using the preprocessed fluorescence spectrum, the fluorescence spectrum intensity and the oil film thickness follow an exponential decay model, the three-dimensional fluorescence spectrum data is constructed based on the model and combined with the acquired fluorescence spectrum data of different oil types and oil film thicknesses; the mathematical relationship between the wave band characteristics in the fluorescence spectrum data and the oil film thickness is used to construct the thickness difference matrix, the Euclidean distance is used to calculate the fluorescence spectrum intensity difference of adjacent oil film thicknesses at different wavelengths, wherein the expression of the exponential decay model is as follows:

[0022] ,

[0023] wherein I(d) is the fluorescence intensity when the oil film thickness is d, I0 is the initial fluorescence intensity when the oil film thickness tends to zero, is the intensity decay coefficient, represents the decay speed of the fluorescence signal with the increase of the thickness, and the oil film thickness (mm).

[0024] Further, the construction of the three-dimensional fluorescence spectrum data and the thickness difference matrix by using the preprocessed fluorescence spectrum further comprises the following steps: the three-dimensional fluorescence spectrum data of each oil type with different oil film thicknesses is constructed by taking each oil type as a unit and differentiating the oil film thickness; and the thickness difference matrix is constructed according to the Euclidean distance, each element a oil in the thickness difference matrix A is expressed as:

[0025] ,

[0026] wherein f t (x i ) and f t+1 (x i ) represent the fluorescence spectral intensity of a certain oil film between adjacent thicknesses t and t+1 at x i wavelength, and f j (x i ) represents the fluorescence spectral intensity of a certain oil film with a film thickness of j at x i wavelength; the three-dimensional fluorescence spectral data is subjected to a pooling operation in the frequency domain using Fourier transform to extract the waveband features of the fluorescence spectrum, including transforming the fluorescence spectral curves of different oil film thicknesses of a certain oil using discrete two-dimensional Fourier transform, and the transformed frequency domain curve is:

[0027] ,

[0028] wherein M and N represent the data lengths of the wavelength range and the film thickness range of the fluorescence spectrum, respectively; k and l represent the frequency indexes of the fluorescence spectrum in the wavelength and film thickness directions, respectively.

[0029] Further, the waveband features and thickness features of the three-dimensional fluorescence spectral data and the thickness difference matrix are extracted using a deep neural network, including extracting the waveband features and thickness features of the three-dimensional fluorescence spectral data and the thickness difference matrix using a differentiable pooling deep neural network DP-DNN; wherein the three-dimensional fluorescence spectral data is subjected to a pooling operation in the frequency domain using Fourier transform to extract the waveband features of the fluorescence spectrum; the thickness difference matrix is subjected to a pooling operation using a globally learnable pooling method to extract the film thickness features of the fluorescence spectrum; the globally learnable pooling generates spatial weights through an attention mechanism and performs weighted summation, and is expressed as:

[0030] ,

[0031] wherein x cij is an element of the thickness difference matrix , and the output is ij wherein the attention weight a is generated through a small learning network of a 1x1 convolution and ReLU, and is expressed as: .

[0032] Further, the feature extraction of the fluorescence image data based on the joint convolutional neural network comprises extracting the texture features of the marine oil spill fluorescence image by using a convolutional neural network (CNN), a deep residual network (ResNet) and a densely connected network (DenseNet), and capturing the key features of the marine oil spill by performing convolution operation on the multi-band fluorescence image; when performing the convolution operation on the filtered multi-band fluorescence image, a k×k size convolution kernel is used to slide on the image from the top left corner step by step to cover the entire image, and each pixel point of the image after the convolution operation is weighted and valued according to the band features of the fluorescence spectrum, and the weighted sum result of each pixel point of the image is taken as the pixel value of the point, which is expressed as:

[0033] ,

[0034] wherein w c represents the weight value extracted according to the fluorescence spectrum band feature; I(x, y) is the pixel value of a certain band fluorescence image at (x, y); K(i, j) is the weight of the convolution kernel at position (i, j), represents the pixel value of the feature map at position (x, y) after convolution; P(x, y) represents the pixel value of a certain band fluorescence image at position (x, y) after fusion of the fluorescence spectrum features.

[0035] Further, the extracted features are analyzed and predicted after fusion based on the Lasso regression model, comprising: after the fusion of the band features of the fluorescence spectrum extracted from the differentiable pooling deep neural network, the image features extracted from the oil film thickness features and the fluorescence image data, the fusion result is input into the Lasso regression model for analysis and prediction; the Bayesian deep learning framework is integrated in the Lasso regression model, and the uncertainty of the prediction is further quantified, so as to realize the accurate prediction and early warning of the marine oil spill, wherein first, the three types of features are encoded respectively to obtain the vector representation: Fs: fluorescence spectrum band feature vector; Ft: oil film thickness feature vector; Fi: fluorescence image texture feature vector, and then the three types of features are spliced in the feature dimension to form a unified high-dimensional feature representation: wherein represents the vector splicing operation; the spliced fusion features are input into the Lasso regression model for analysis and prediction, and the Lasso regression realizes feature selection and sparse modeling by introducing L1 regularization in the loss function, and the optimization objective function is:

[0036] ,

[0037] wherein, represents the true oil spill monitoring value of the jth sample, is a bias term, a regression coefficient corresponding to the i-th feature, is a regularization parameter for controlling the sparsity of the model.

[0038] Further, the Lasso regression model based on the extracted features after feature fusion for analysis and prediction, further comprising jointly optimizing the deep neural network structure parameters and the Lasso regression regularization hyperparameter λ by using genetic algorithm GA and particle swarm optimization PSO, after multiple rounds of iterative optimization, the trained model uses the fused feature vector F_concat for monitoring and early warning, outputs the prediction result of marine oil spill and the corresponding uncertainty estimate, and provides a reliable basis for accurate identification and early warning.

[0039] In a second aspect, a marine oil spill monitoring system based on a multi-band fluorescence spectrum neural network comprises:

[0040] The data acquisition module is configured to acquire fluorescence spectrum and fluorescence image data of different oil types and different oil film thickness oil spills.

[0041] The preprocessing module is configured to preprocess the fluorescence spectrum and fluorescence image data of the marine oil spill.

[0042] The difference module is configured to construct a three-dimensional fluorescence spectrum data and a thickness difference matrix using the preprocessed fluorescence spectrum.

[0043] The spectral feature module is configured to extract band features and thickness features from the three-dimensional fluorescence spectrum data and the thickness difference matrix using a deep neural network.

[0044] The image feature module is configured to extract features from the fluorescence image data based on a joint convolutional neural network.

[0045] The prediction module is configured to analyze and predict the extracted features after feature fusion based on a Lasso regression model.

[0046] The early warning module is configured to output an early warning result.

[0047] In a third aspect, the present application provides a computer readable storage medium, wherein a plurality of instructions are stored, the instructions being adapted to be loaded and executed by a processor of a terminal device.

[0048] In a fourth aspect, the present application provides a terminal device comprising a processor and a computer readable storage medium, the processor being configured to implement the instructions; the computer readable storage medium is configured to store a plurality of instructions, the instructions being adapted to be loaded and executed by the processor to implement a marine oil spill monitoring method based on a multi-band fluorescence spectrum neural network.

[0049] In summary, the present application has the following beneficial technical effects:

[0050] (1) The present application proposes a marine oil spill monitoring and early warning system based on multi-band fluorescence spectrum neural network, which combines the feature extraction capabilities of differentiable pooling deep neural network DP-DNN, convolutional neural network CNN and deep residual network (ResNet), dense connection network (DenseNet), realizes the accurate prediction and early warning of marine oil spill through pooling operation and multi-band convolution operation. This method effectively integrates marine oil spill fluorescence spectrum and fluorescence image data, and provides reliable basis for accurate identification and early warning of oil spill.

[0051] (2) The present application innovatively fuses the spectrum neural network with the dual channel of multi-band fluorescence spectrum, so that the system can comprehensively consider the spectral band characteristics and oil film thickness characteristics, and extract the texture features in the fluorescence image. This dual channel design enhances the model's monitoring and early warning ability for complex marine oil spills, especially in the process of processing marine oil fluorescence spectrum data, greatly improving the accuracy and efficiency of monitoring.

[0052] (3) By using the differentiable pooling deep neural network DP-DNN to process three-dimensional fluorescence spectrum data and thickness difference matrix, the system solves the contradiction between spatial resolution and spectral resolution in the direction feature extraction and oil thickness identification of the current fluorescence monitoring method. At the same time, based on the fluorescence spectrum feature fluorescence image data, the spectral and image heterogeneous data features are fused, and the genetic algorithm (GA) and particle swarm optimization (PSO) method are used to optimize the structure and hyperparameters of the neural network, so that the system has higher robustness and accuracy when facing complex marine oil spill monitoring and early warning tasks. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a schematic diagram of a marine oil spill monitoring method based on multi-band fluorescence spectrum neural network of the present application embodiment 1.

[0054] Figure 2 is a multi-band different oil film thickness fluorescence spectrum schematic diagram of the present application embodiment 1.

[0055] Figure 3 is a multi-band different oil film thickness fluorescence image schematic diagram of the present application embodiment 1. DETAILED DESCRIPTION

[0056] The present application will be further described in detail below with reference to the accompanying drawings.

[0057] Embodiment 1

[0058] Reference Figure 1The marine oil spill monitoring method based on the multi-band fluorescence spectrum neural network of the embodiment comprises the following steps:

[0059] Obtain fluorescence spectrum and fluorescence image data of oil spills of different oil types and different oil film thicknesses;

[0060] Preprocess the marine oil spill fluorescence spectrum and fluorescence image data respectively;

[0061] A mathematical model between fluorescence spectrum intensity and oil film thickness is constructed by using the preprocessed oil spill fluorescence spectrum, and the mathematical model between fluorescence spectrum intensity and oil film thickness follows an exponential decay model. Based on the above mathematical model, three-dimensional fluorescence spectrum data are constructed by combining the obtained fluorescence spectrum data of different oil types and oil film thicknesses. A thickness difference matrix is constructed by using the mathematical relationship between the waveband features in the fluorescence spectrum data and the oil film thickness. The fluorescence spectrum intensity difference of adjacent oil film thicknesses at different wavelengths is calculated using the Euclidean distance;

[0062] The waveband features and thickness features of the three-dimensional fluorescence spectrum data and the thickness difference matrix are extracted by using a differentiable pooling deep neural network DP-DNN. The waveband features of the fluorescence spectrum are extracted by using Fourier transform to perform a pooling operation on the three-dimensional fluorescence spectrum data in the frequency domain. The oil film thickness features of the fluorescence spectrum are extracted by using a global learnable pooling method to perform a pooling operation on the thickness difference matrix;

[0063] The texture features of the marine oil spill fluorescence image are extracted by using a convolutional neural network CNN and a deep residual network (ResNet) and a densely connected network (DenseNet), which include capturing key features of the marine oil spill by performing a convolution operation on the multi-band fluorescence image;

[0064] After the waveband features and the oil film thickness features of the fluorescence spectrum extracted from the differentiable pooling deep neural network and the image features extracted from the fluorescence image data are fused, they are input into a Lasso regression model for analysis and prediction. The Lasso regression model integrates a Bayesian deep learning framework, further quantifies the uncertainty of the prediction, and thus realizes accurate prediction and early warning of the marine oil spill;

[0065] Genetic algorithm (GA) and particle swarm optimization (PSO) methods are used to optimize the structure and hyperparameters of the neural network, further improving the performance and accuracy of the model.

[0066] Specifically, the method comprises the following steps:

[0067] Obtain fluorescence spectrum and fluorescence image data of oil spills of different oil types and different oil film thicknesses

[0068] Step 1, obtain marine oil spill fluorescence spectrum data and images;

[0069] The original spectrum of seawater is collected by adding 20 ml of seawater to a culture dish, and after standing, the preset excitation wavelength range (250-550 nm) and emission wavelength range (260-1700 nm) are set. The oil sample is added to the culture dish in the concentration gradient (1-10 μL with 1 μL as the concentration gradient, 10-50 μL with 5 μL as the concentration gradient, 50-100 μL with 10 μL as the concentration gradient, 100-500 μL with 50 μL as the concentration gradient, 500 μL-1 ml with 100 μL as the concentration gradient, 1-3 ml with 1 ml as the concentration gradient). Each time, 5 spectrum data of the corresponding concentration are saved under the set excitation-emission wavelength matrix, which are the center point and the four points around it. The average spectrum of 3 collected spectra is collected each time. The fluorescence spectrum data of different concentrations of the other three oil samples are collected according to the above steps;

[0070] Similarly, under the selected excitation wavelength and corresponding emission band filter combination, the original image of seawater is collected, and different filters are replaced to obtain fluorescence images of different waveband ranges; the oil sample is added according to the same concentration gradient, and the image under the concentration is saved each time, and the filter is replaced to collect the corresponding waveband image; the multi-waveband fluorescence image collection of the four oil samples under each concentration is completed in turn;

[0071] Step 2, pretreatment of marine oil spill fluorescence spectrum data and image;

[0072] In data processing, the moving average smoothing and Logistic normalization of marine oil spill fluorescence spectrum data of different oil types and different oil film thicknesses are carried out to ensure the consistency and integrity of the data, wherein the moving average smoothing formula is:

[0073] ,

[0074] , i wherein f(x i-1 ) represents the smoothed result of the fluorescence spectrum intensity at x i , f(x i+1 ) and f(x i-1 ) represent the measured fluorescence spectrum intensity at x i , x i+1 , respectively, that is, the smoothed value of the fluorescence spectrum intensity at each point is the average value of the point and the points before and after it.

[0075] The Logistic normalization process is represented as:

[0076] ,

[0077] , ​

[0078] Where e represents the base of the natural logarithm, g(x) i ) indicates that in x i The fluorescence spectral intensity after Logistic function calculation at the specified wavelength, f(x) i ) indicates that in x i Fluorescence spectral intensity measured at the specified wavelength. Indicates in x i The result of normalizing the fluorescence spectrum intensity at the specified wavelength, g(x). max and g(x) min These represent the maximum and minimum fluorescence spectral intensities after the Logistic function operation, respectively.

[0079] For the preprocessing of fluorescence images, Sobel operators in the horizontal and vertical directions were used for edge detection to detect the edges of glass containers containing marine oil spill samples. Based on the edge detection results, the fluorescence images were cropped to retain only the images of the simulated marine oil spill areas.

[0080] The Sobel operator sets 3x3 convolution kernels G(x) and G(y) in the horizontal and vertical directions, respectively:

[0081] , ,

[0082] For each fluorescence image I(i, j), convolution operations are performed using G(x) and G(y) respectively to obtain the gradient components of each pixel (i, j) in the x and y directions. and ;

[0083] For each pixel (i, j) in the image, the total gradient magnitude is calculated using Euclidean distance combined with the gradients in the x and y directions. The calculation process is expressed as follows:

[0084] ,

[0085] set up The threshold is 150; after calculating any pixel in the fluorescence image If the value is greater than the threshold, the point is identified as an edge point; if the value is less than the threshold, it is identified as a non-edge point. Based on the edge detection results, the image is cropped, retaining only the central region containing the oil spill sample.

[0086] Step 3: Construct a mathematical model;

[0087] An exponential decay model between fluorescence intensity and oil film thickness was constructed using the pretreated oil spill fluorescence spectrum. Its mathematical expression is as follows:

[0088] ,

[0089] where I(d) is the fluorescence intensity when the oil film thickness is d, I0 is the initial fluorescence when the oil film thickness approaches zero, is the intensity decay coefficient, represents the decay speed of the fluorescence signal with the increase of the thickness oil film thickness (mm).

[0090] Further construct a three-dimensional fluorescence spectrum data, specifically, measure the fluorescence intensity of each oil at M excitation wavelengths (λ_ex1, λ_ex2,..., λ_exM) and N emission wavelengths (λ_em1, λ_em2,..., λ_emN), and set D oil film thickness gradients, thereby forming a three-dimensional data tensor , which comprehensively characterizes the three-dimensional response relationship of the fluorescence intensity with the excitation wavelength, the emission wavelength and the oil film thickness. Further construct a thickness difference matrix A oil Each element in the matrix A is represented as:

[0091] ,

[0092] where f t (x i ) and f t+1 (x i ) respectively represent the fluorescence spectrum intensity of a certain oil film between adjacent thicknesses t and t+1 at x i wavelength, and f j (x i ) represents the fluorescence spectrum intensity of a certain oil film with a thickness of j at x i wavelength.

[0093] Step 4, extract the waveband feature and the thickness feature:

[0094] In the waveband feature and thickness feature extraction stage of the three-dimensional fluorescence spectrum data and the thickness difference matrix using the differentiable pooling deep neural network DP-DNN, the Fourier transform is used to perform the pooling operation on the three-dimensional fluorescence spectrum data in the frequency domain, and the waveband feature of the fluorescence spectrum is extracted; the global learnable pooling method is used to perform the pooling operation on the thickness difference matrix, and the oil film thickness feature of the fluorescence spectrum is extracted; specifically, the input layer receives the three-dimensional fluorescence data F and the thickness difference matrix A oil ; the Fourier pooling layer performs DFT on F and takes the low-frequency component; the learnable pooling layer generates spatial weights through the attention mechanism, and outputs the thickness feature vector after weighted summation. At the same time, the generation network of the attention weight is specifically a combination of a 1×1 convolution layer, a ReLU activation function and a Softmax layer, which enhances the operability.

[0095] The three-dimensional fluorescence spectrum data is pooled in the frequency domain using Fourier transform to extract the waveband features of the fluorescence spectrum, including transforming the fluorescence spectrum curves of different oil film thicknesses of a certain oil using discrete two-dimensional Fourier transform, and the frequency domain curves after transformation are:

[0096] ,

[0097] wherein M and N represent the data lengths of the wavelength range and the oil film thickness range of the fluorescence spectrum respectively; k and l represent the frequency indexes of the fluorescence spectrum in the wavelength and oil film thickness directions respectively;

[0098] The global learnable pooling method is used to pool the thickness difference matrix to extract the oil film thickness features of the fluorescence spectrum, including that the global learnable pooling generates spatial weights through an attention mechanism and performs weighted summation:

[0099] ,

[0100] wherein x cij is an element of the thickness difference matrix , and the output is obtained, wherein the attention weight a ij is generated through a small learning network of 1x1 convolution and ReLU , and can be represented as: .

[0101] Step 5, extracting the texture features of the fluorescence image:

[0102] For the fluorescence image, the convolutional neural network CNN and the deep residual network (ResNet) and the densely connected network (DenseNet) are used to extract the texture features of the marine oil spill fluorescence image; the specific process is as follows:

[0103] The input is the multi-band fluorescence image after Sobel edge detection and cropping, and the size of each image is HxWxC, wherein C is the number of bands. The standardization processing is performed on each band image:

[0104] ,

[0105] wherein and are the mean and standard deviation of the cth band respectively.

[0106] Further, the shallow feature extraction (CNN) is used to extract the texture features of the marine oil spill fluorescence image, including that when performing convolution operation on the filtered multi-band fluorescence image, a convolution kernel of kxk size is used to slide on the image from the top left corner step by step to cover the entire image;

[0107] Further, the deep feature extraction (ResNet + DenseNet) is used to extract middle and high-level texture features, which is specifically represented as follows: the gradient vanishing is alleviated by the skip connection of the ResNet module, and the output of the Lth layer is:

[0108] ,

[0109] Each layer of the DenseNet module receives the features of all previous layers as input, enhancing feature reuse,

[0110] ,

[0111] Each pixel point of the image after the convolution operation is weighted according to the waveband characteristics of the fluorescence spectrum, and the result of the weighted sum of each pixel point of the image is taken as the pixel value of the point, which can be expressed as:

[0112] ,

[0113] where w c represents the weight value extracted according to the waveband characteristics of the fluorescence spectrum; I(x,y) is the pixel value of a certain waveband fluorescence image at (x,y); K(i,j) is the weight of the convolution kernel at position (i,j), represents the pixel value of the feature map at position (x,y) after convolution; P(x,y) represents the pixel value of a certain waveband fluorescence image at position (x,y) after fusion of the fluorescence spectrum characteristics.

[0114] Finally, the feature map is converted into a feature vector F i ∈R Di by global average pooling (GAP), which is a vector representation of the image texture feature and is used for subsequent feature fusion and prediction analysis.

[0115] Step 6, the three features are spliced and fused:

[0116] The waveband features of the fluorescence spectrum extracted from the differentiable pooling deep neural network, the oil film thickness features, and the image features extracted from the fluorescence image data are spliced into fusion features according to the latitude features, and the fusion features are input into the Lasso regression model to output the confidence interval of the marine oil spill monitoring and prediction.

[0117] 1. First, the three types of features are encoded to obtain vector representations:

[0118] F s : fluorescence spectrum waveband feature vector (from DP-DNN);

[0119] F t : oil film thickness feature vector (from thickness difference matrix);

[0120] Fi : Fluorescence image texture feature vector (from joint CNN);

[0121] Map each feature vector to the same dimension D using a fully connected layer:

[0122] ,

[0123] Subsequently, the three types of features are concatenated in the feature dimension to form a unified high-dimensional feature representation:

[0124] where denotes the vector concatenation operation;

[0125] 2. Finally, the concatenated fusion features are input into the Lasso regression (Least Absolute Shrinkage and Selection Operator) model for analysis and prediction. Lasso regression effectively realizes feature selection and sparse modeling by introducing L1 regularization in the loss function, and its optimization objective function is:

[0126] ,

[0127] where, denotes the true oil spill monitoring value of the jth sample, is the bias term, is the regression coefficient corresponding to the ith feature, is the regularization parameter used to control the sparsity of the model.

[0128] 3. To quantify the prediction uncertainty, integrate the Bayesian deep learning framework into the Lasso regression model, assuming that the regression coefficient obeys the prior distribution: , estimate the posterior distribution p by variational inference or MCMC sampling, and the prediction distribution is,

[0129] ,

[0130] By establishing a probability distribution assumption for the regression coefficient , the uncertainty of the prediction result is quantified, and the confidence interval of the marine oil spill monitoring prediction is output.

[0131] Step 7, optimize the parameters:

[0132] To further improve the performance and robustness of the model, the genetic algorithm (GA) and particle swarm optimization (PSO) are used to jointly optimize the deep neural network structure parameters (including the number of convolutional layers, the size of the convolution kernel, and the pooling method) and the regularization hyperparameter λ of the Lasso regression. The optimization process aims to find the optimal parameter combination to minimize the prediction error of the model on the validation set.

[0133] First, the parameters to be optimized are encoded into a vector form that can be processed by the optimization algorithm. The parameter vector is defined as follows:

[0134] Number of convolutional layers: L ∈ {1, 2,..., L max}

[0135] Size of convolution kernel: K ∈ {3, 5, 7}

[0136] Pooling method: P ∈ {MaxPooling, AveragePooling, GlobalAveragePooling}

[0137] Lasso regularization hyperparameter: λ ∈ [0.001, 1.0]

[0138] For the above mixed type parameters (discrete and continuous coexist), as shown in Figure 2 the embodiment designs the coding strategy respectively: for GA, mixed coding of real numbers and integers is adopted; for PSO, continuous real number coding is adopted, and the discrete parameters are rounded or mapped to the nearest value in the pre-defined set. A two-stage hybrid optimization strategy is adopted: on the one hand, in the GA (global exploration) stage, the genetic algorithm is used to perform rough search in the global range, and the potential parameter region is selected through selection, crossover and mutation operations, to avoid the optimization process from falling into local optimum too early; on the other hand, in the PSO (local fine search) stage, the optimal individual obtained in the GA stage is taken as the initial particle swarm of PSO, and the particle swarm algorithm is used to perform fine search in the neighborhood of the high-quality solution, so as to efficiently approach the global optimal solution.

[0139] To verify the effectiveness of the GA-PSO hybrid optimization strategy, as shown in Figure 3As shown, the present embodiment performs a system experiment on a marine oil spill SAR image dataset. The experimental setup includes: dividing the dataset into a training set, a validation set, and a test set in a ratio of 7:2:1; selecting the optimized and unoptimized models for comparison in terms of accuracy, F1 score, and AUC. The results show that the model optimized by GA-PSO has an accuracy of 98.2% and an F1 score of 0.976 on the test set, significantly better than the unoptimized model (accuracy 94.5%, F1 score 0.932) and the results of single optimization algorithm (GA or PSO). In addition, the ablation experiment further shows that the hybrid strategy is superior to the single optimization method in terms of convergence speed and stability.

[0140] After multiple rounds of iterative optimization, the model obtained by training is used to analyze the fused feature vectors monitoring and early warning, outputting the prediction results and corresponding uncertainty estimates of marine oil spills, providing reliable basis for accurate identification and early warning.

[0141] Embodiment 2

[0142] The present embodiment provides a marine oil spill monitoring system based on multi-band fluorescence spectrum neural network, comprising:

[0143] The data acquisition module is configured to acquire fluorescence spectrum and fluorescence image data of oil spills of different oil types and different oil film thicknesses;

[0144] The preprocessing module is configured to preprocess the fluorescence spectrum and fluorescence image data of marine oil spills respectively;

[0145] The difference module is configured to construct a three-dimensional fluorescence spectrum data and a thickness difference matrix using the preprocessed fluorescence spectrum;

[0146] The spectral feature module is configured to extract band features and thickness features from the three-dimensional fluorescence spectrum data and the thickness difference matrix using a deep neural network;

[0147] The image feature module is configured to extract features from the fluorescence image data based on a joint convolutional neural network;

[0148] The prediction module is configured to analyze and predict the extracted features after feature fusion based on a Lasso regression model;

[0149] The early warning module is configured to output early warning results.

[0150] A computer-readable storage medium having a plurality of instructions stored therein, the instructions being adapted to be loaded and executed by a processor of a terminal device, the instructions being a marine oil spill monitoring method based on a multi-band fluorescence spectrum neural network.

[0151] A terminal device comprises a processor and a computer readable storage medium, the processor is used for realizing instructions; the computer readable storage medium is used for storing a plurality of instructions, the instructions are suitable for being loaded by the processor and executing the kind of marine oil spill monitoring method based on multi-band fluorescence spectrum neural network.

[0152] The above are preferred embodiments of the present application, not limited by the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for monitoring marine oil spills based on a multi-band fluorescence spectral neural network, characterized in that, include: Obtain fluorescence spectra and fluorescence image data of oil spills of different oil types and with different oil film thicknesses; The fluorescence spectra and fluorescence image data of marine oil spills were preprocessed respectively; Three-dimensional fluorescence spectral data and a thickness difference matrix are constructed using preprocessed fluorescence spectra. This includes building a mathematical model of the relationship between fluorescence intensity and oil film thickness, where the relationship follows an exponential decay model. Based on this model and combined with fluorescence spectral data of different oil types and oil film thicknesses, three-dimensional fluorescence spectral data are constructed. A thickness difference matrix is ​​constructed using the band characteristics of the fluorescence spectral data and the mathematical relationship between oil film thickness. Euclidean distance is used to calculate the fluorescence intensity difference between adjacent oil film thicknesses at different wavelengths. The expression for the exponential decay model is as follows: ， in, I ( d The oil film thickness is... d fluorescence intensity at that time I 0 represents the initial fluorescence intensity when the oil film thickness approaches zero. The intensity attenuation coefficient, The oil film thickness represents the rate attenuation of the fluorescence signal as the thickness increases. This also includes constructing three-dimensional fluorescence spectral data for each oil type, distinguishing different oil film thicknesses; and building a thickness difference matrix based on Euclidean distance. A oil Each element in Represented as: ， in, f t ( x i )and f t+1 ( x i ) respectively represent a certain type of oil film in x i Adjacent thickness at wavelength t and t Fluorescence spectral intensity between +1, f j ( x i ) indicates a certain type of oil film in x i Oil film thickness at wavelength is j The fluorescence spectral intensity was determined; Fourier transform was used to perform pooling operations on the three-dimensional fluorescence spectral data in the frequency domain to extract the band features of the fluorescence spectrum, including using discrete two-dimensional Fourier transform to transform the fluorescence spectral curves of a certain oil with different oil film thicknesses. The transformed frequency domain curves are as follows: ， in, M and N The data lengths represent the wavelength range of the fluorescence spectrum and the oil film thickness range, respectively. k and l These represent the frequency indices of the fluorescence spectrum in the wavelength and oil film thickness directions, respectively. Deep neural networks were used to extract band features and thickness features from three-dimensional fluorescence spectral data and thickness difference matrices. Feature extraction of fluorescence image data based on joint convolutional neural networks yields texture features of marine oil spill fluorescence images; The extracted band features, thickness features, and texture features are analyzed and predicted based on the Lasso regression model after feature fusion. Output the warning results.

2. The marine oil spill monitoring method based on a multi-band fluorescence spectral neural network according to claim 1, characterized in that, The preprocessing of the fluorescence spectral and fluorescence image data of marine oil spills includes moving average smoothing and logistic normalization of the fluorescence spectral data of marine oil spills with different oil types and oil film thicknesses to ensure data consistency and integrity. The average smoothing is expressed as follows: ， in, Indicates in x i The result after smoothing the fluorescence spectrum intensity at the specified wavelength. f ( x i-1 ), f ( x i )and f ( x i+1 ) respectively represent in x i-1 , x i and x i+1 The fluorescence spectral intensity measured at a given wavelength is expressed by the Logisti function as follows: The Logistic normalization process is represented as: ,in, e The base of the natural logarithm. g ( x i ) indicates in x i Fluorescence spectral intensity after Logistic function calculation at a given wavelength f ( x i ) indicates in x i Fluorescence spectral intensity measured at a specific wavelength Indicates in x i The result after normalizing the fluorescence spectral intensity at the specified wavelength. g ( x ) max and g ( x ) min These represent the maximum and minimum fluorescence spectral intensities after the Logistic function operation, respectively.

3. The marine oil spill monitoring method based on a multi-band fluorescence spectral neural network according to claim 2, characterized in that, The preprocessing of the fluorescence spectrum and fluorescence image data of marine oil spills includes preprocessing of the fluorescence image by using Sobel operators in the horizontal and vertical directions for edge detection to detect the edges of the glass containers containing marine oil spill samples, and cropping the fluorescence image based on the edge detection results to retain only the image of the simulated marine oil spill area.

4. The marine oil spill monitoring method based on a multi-band fluorescence spectral neural network according to claim 3, characterized in that, The method of extracting band and thickness features from three-dimensional fluorescence spectral data and thickness difference matrix using a deep neural network includes using a differentiable pooling deep neural network (DP-DNN) to extract band and thickness features from the three-dimensional fluorescence spectral data and thickness difference matrix. Specifically, Fourier transform is used to perform pooling operations on the three-dimensional fluorescence spectral data in the frequency domain to extract band features of the fluorescence spectrum; a globally learnable pooling method is used to perform pooling operations on the thickness difference matrix to extract oil film thickness features from the fluorescence spectrum; the globally learnable pooling method generates spatial weights through an attention mechanism and performs weighted summation, as shown below: ， in, x cij It is a thickness difference matrix The elements, output Attention weights a ij Through a small learning network of 1x1 convolution and ReLU Generate, represented as: .

5. The marine oil spill monitoring method based on a multi-band fluorescence spectral neural network according to claim 4, characterized in that, The feature extraction of fluorescence image data based on a joint convolutional neural network includes using a convolutional neural network (CNN), a deep residual network (ResNet), and a densely connected network (DenseNet) to extract texture features of marine oil spill fluorescence images. Key features of marine oil spills are captured by performing convolution operations on multi-band fluorescence images. Specifically, when performing convolution operations on filtered multi-band fluorescence images, a k×k convolution kernel is gradually slid across the image from the top left corner, covering the entire image. After the convolution operation, each pixel in the image is weighted according to the band characteristics of the fluorescence spectrum, and the weighted sum of each pixel is used as its pixel value. ， in, w c This represents the weight value extracted based on the characteristics of the fluorescence spectral bands; I ( x , y ) is the input fluorescence image of a certain band in ( x , y The pixel value of ); K ( i , j ) is the position of the convolution kernel ( i , j The weight of ) This indicates the position of the convolutional feature map. x , y The pixel value of ); P ( x , y ) represents a fluorescence image of a certain band with fused fluorescence spectral features at position ( x , y The pixel value of ).

6. The marine oil spill monitoring method based on a multi-band fluorescence spectral neural network according to claim 5, characterized in that, The Lasso regression model is used to analyze and predict the extracted band features, thickness features, and texture features after feature fusion. This includes fusing the band features of fluorescence spectra extracted from differentiable pooling deep neural networks, oil film thickness features, and image features extracted from fluorescence image data, and then inputting these features into the Lasso regression model for analysis and prediction. A Bayesian deep learning framework is integrated into the Lasso regression model to further quantify the uncertainty of the prediction, thereby achieving accurate prediction and early warning of marine oil spills. First, the three types of features are encoded separately to obtain vector representations: Fs: fluorescence spectrum band feature vector; Ft: oil film thickness feature vector; Fi: fluorescence image texture feature vector. Then, the three types of features are concatenated along the feature dimension to form a unified high-dimensional feature representation. ,in This represents a vector concatenation operation; the concatenated and fused features are input into a Lasso regression model for analysis and prediction. Lasso regression achieves feature selection and sparse modeling by introducing L1 regularization into the loss function. Its optimization objective function is: ， in, Indicates the first j The actual oil spill monitoring value of each sample For bias terms, For the corresponding number i The regression coefficients of each feature, This is a regularization parameter used to control the sparsity of the model.

7. The marine oil spill monitoring method based on a multi-band fluorescence spectral neural network according to claim 6, characterized in that, The Lasso regression model is used to analyze and predict the extracted band features, thickness features, and texture features after feature fusion. It also includes using the genetic algorithm (GA) and particle swarm optimization (PSO) to jointly optimize the structural parameters of the deep neural network and the regularization hyperparameter λ of the Lasso regression. After multiple rounds of iterative optimization, the trained model uses the fused feature vector F_concat for monitoring and early warning, outputting the prediction results of marine oil spills and the corresponding uncertainty estimates, providing a reliable basis for accurate identification and early warning.

8. A marine oil spill monitoring system based on a multi-band fluorescence spectral neural network, executing the marine oil spill monitoring method based on a multi-band fluorescence spectral neural network as described in claim 1, characterized in that, include: The data acquisition module is configured to acquire fluorescence spectrum and fluorescence image data of oil spills of different oil types and different oil film thicknesses; The preprocessing module is configured to preprocess the fluorescence spectral and fluorescence image data of marine oil spills, respectively. The difference module is configured to construct three-dimensional fluorescence spectral data and a thickness difference matrix using preprocessed fluorescence spectra; The spectral feature module is configured to extract band features and thickness features from three-dimensional fluorescence spectral data and thickness difference matrix using a deep neural network. The image feature module is configured to extract features from fluorescence image data based on a joint convolutional neural network. The prediction module is configured to analyze and predict the extracted features after feature fusion based on the Lasso regression model. The early warning module is configured to output early warning results.

Images