A method, system, device and medium for predicting the shape evolution of submarine hydrate mounds

Abstract

The present invention discloses a method, system, device and medium for predicting the shape evolution of submarine hydrate mounds, relating to the field of marine geology. The method includes: obtaining the original image of a target submarine hydrate mound and the corresponding methane gas flow data; using a submarine hydrate mound evolution prediction model to predict the development stage, the shape category in the accumulation stage and the start time of the decomposition stage of the target submarine hydrate mound according to the original image of the target submarine hydrate mound and the corresponding methane gas flow data; wherein, the submarine hydrate mound evolution prediction model includes an image data compression module, an image feature extraction module and an evolution prediction module connected in series, which can perform data compression and feature extraction on the original image, and predict the development stage, the shape category in the accumulation stage and the start time of the decomposition stage of the target submarine hydrate mound based on the feature vector and the corresponding methane gas flow data, so as to achieve an accurate prediction of the shape evolution of the submarine hydrate mound.

Description

Technical Field

[0001] The present invention relates to the field of marine geology, and particularly to a method, system, device and medium for predicting the shape evolution of submarine gas hydrate mounds. Background Art

[0002] The formation of submarine gas hydrate mounds is closely related to the accumulation of shallow gas hydrates. The surface of submarine gas hydrate mounds is usually covered with exposed massive hydrates. The formation of large-scale submarine gas hydrate mounds requires relatively high-throughput hydrocarbon fluid migration conditions. Therefore, the formation of submarine gas hydrate mounds may indicate the convergent fluid migration at the continental margin and the possible occurrence of oil and gas reservoirs or gas hydrate reservoirs below. The prediction of the evolution of submarine gas hydrate mounds is of great significance for the exploration and exploitation of oil and gas reservoirs or gas hydrate reservoirs.

[0003] At present, scholars at home and abroad mainly use methods such as acoustic detection, seismic detection, in-situ observation of the seabed, and geochemical analysis after core sampling to observe submarine gas hydrate mounds, and already have the ability to collect and accumulate in-situ data. However, there are few methods to use these in-situ data to predict the evolution of submarine gas hydrate mounds. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, system, device and medium for predicting the shape evolution of submarine gas hydrate mounds to achieve accurate prediction of the shape evolution of submarine gas hydrate mounds.

[0005] To achieve the above purpose, the present invention provides the following solutions:

[0006] A method for predicting the shape evolution of submarine gas hydrate mounds includes:

[0007] Obtaining the original image of the target submarine gas hydrate mound and the corresponding methane gas flow rate data;

[0008] Using a submarine gas hydrate mound evolution prediction model, based on the original image of the target submarine gas hydrate mound and the corresponding methane gas flow rate data, predicting the development stage, the shape category in the accumulation stage, and the start time of the decomposition stage of the target submarine gas hydrate mound; the development stage includes: a nucleation stage, an accumulation stage, and a decomposition stage; the shape category includes: a circular steep mound, a circular gentle slope mound, an oval steep mound, an oval gentle slope mound, a slender steep mound, and a slender gentle slope mound;

[0009] Among them, the submarine gas hydrate mound evolution prediction model includes an image data compression module, an image feature extraction module, and an evolution prediction module connected in series;

[0010] The image data compression module is used to compress the original image of the target submarine hydrate mound to obtain the compressed data of the target submarine hydrate mound;

[0011] The image feature extraction module is used to extract features from the compressed data of the target submarine hydrate mound to obtain the feature vector of the target submarine hydrate mound;

[0012] The evolution prediction module is used to predict the development stage, the shape category of the accumulation stage, and the start time of the decomposition stage of the target submarine hydrate mound according to the feature vector of the target submarine hydrate mound and the corresponding methane gas flow data.

[0013] Optionally, the method for determining the submarine hydrate mound evolution prediction model specifically includes:

[0014] Obtain a sample data set; the sample data set includes the original images of several labeled sample submarine hydrate mounds and the corresponding methane gas flow data; the labels include: the development stage of the sample submarine hydrate mound, the shape category of the accumulation stage, and the start time of the decomposition stage;

[0015] Perform data augmentation on the original images of the labeled sample submarine hydrate mounds to obtain the enhanced images of the labeled sample submarine hydrate mounds;

[0016] Determine the original images of the labeled sample submarine hydrate mounds and the corresponding methane gas flow data, as well as the enhanced images of the labeled sample submarine hydrate mounds and the corresponding methane gas flow data as the training data set;

[0017] Construct an initial neural network model; the initial neural network model includes an image data compression network based on the MobileNet architecture, an image feature extraction network based on the Inception module, and an evolution prediction network; the image data compression network, the image feature extraction network, and the evolution prediction network are connected in series;

[0018] Use the training data set to train the initial neural network model to obtain the submarine hydrate mound evolution prediction model.

[0019] Optionally, the image data compression network includes a depthwise separable convolutional layer, a first residual network module, and a second residual network module connected in series; the depthwise separable convolutional layer includes a depthwise convolutional layer and a pointwise convolutional layer connected in series;

[0020] Among them, the convolutional kernel size of the depthwise convolutional layer is 3×3, and the convolutional stride is 5.

[0021] Optionally, the image feature extraction network includes a plurality of Inception modules and a fully connected layer connected in series; each of the Inception modules includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a pooling layer, and a channel connection layer; the channel connection layer is respectively connected to the first convolutional layer, the second convolutional layer, the third convolutional layer, and the pooling layer; the channel connection layer is used to splice the outputs of the first convolutional layer, the second convolutional layer, the third convolutional layer, and the pooling layer;

[0022] Among them, the convolutional kernel size of the first convolutional layer is 1×1, the convolutional stride is 1, and the padding method is same; the convolutional kernel size of the second convolutional layer is 3×3, the convolutional stride is 1, and the padding method is same; the convolutional kernel size of the third convolutional layer is 5×5, the convolutional stride is 1, and the padding method is same; the pooling layer is a max pooling layer.

[0023] Optionally, the evolution prediction network includes a normalization processing layer, a first classifier, a second classifier, and a multiple linear regression layer; the normalization processing layer is connected to the first classifier; the first classifier is respectively connected to the second classifier and the multiple linear regression layer;

[0024] The normalization processing layer is used to perform normalization processing on the input feature vector and methane gas flow data to obtain normalized data;

[0025] The first classifier is used to determine the development stage according to the normalized data;

[0026] The second classifier is used to determine the shape category of the accumulation stage according to the normalized data when the development stage is the accumulation stage;

[0027] The multiple linear regression layer is used to predict the start time of the decomposition stage by using the multiple linear regression method according to the normalized data when the development stage is the accumulation stage.

[0028] Optionally, data augmentation is performed on the original image of the labeled sample submarine hydrate mound to obtain an enhanced image of the labeled sample submarine hydrate mound, specifically including:

[0029] Determine the mirror image of the sample submarine hydrate mound according to the original image of the sample submarine hydrate mound;

[0030] Crop the original image and the mirror image of the sample submarine hydrate mound respectively to obtain the cropped image of the sample submarine hydrate mound;

[0031] Use the principal component analysis method to perform color enhancement on the cropped image of the sample submarine hydrate mound to obtain the enhanced image of the sample submarine hydrate mound;

[0032] Determine the label of the original image of the sample submarine gas hydrate mound as the label of the enhanced image of the sample submarine gas hydrate mound, and obtain the enhanced image of the labeled sample submarine gas hydrate mound.

[0033] Optionally, use the training data set to train the initial neural network model to obtain a submarine gas hydrate mound evolution prediction model, specifically including:

[0034] Use the training data set to train the initial neural network model, determine the trained image data compression network as the image data compression module, determine the trained image feature extraction network as the image feature extraction module, and determine the trained evolution prediction network as the evolution prediction module, to obtain a submarine gas hydrate mound evolution prediction model.

[0035] A submarine gas hydrate mound shape evolution prediction system, including:

[0036] A data acquisition module, configured to acquire the original image of the target submarine gas hydrate mound and the corresponding methane gas flow rate data;

[0037] An evolution prediction module, configured to use the submarine gas hydrate mound evolution prediction model to predict the development stage, the shape category in the accumulation stage, and the start time of the decomposition stage of the target submarine gas hydrate mound according to the original image of the target submarine gas hydrate mound and the corresponding methane gas flow rate data; the development stage includes: a nucleation stage, an accumulation stage, and a decomposition stage; the shape categories include: a circular steep mound, a circular gentle slope mound, an oval steep mound, an oval gentle slope mound, a slender steep mound, and a slender gentle slope mound;

[0038] Wherein, the submarine gas hydrate mound evolution prediction model includes an image data compression module, an image feature extraction module, and an evolution prediction module connected in series;

[0039] The image data compression module is configured to perform data compression on the original image of the target submarine gas hydrate mound to obtain the compressed data of the target submarine gas hydrate mound;

[0040] The image feature extraction module is configured to perform feature extraction on the compressed data of the target submarine gas hydrate mound to obtain the feature vector of the target submarine gas hydrate mound;

[0041] The evolution prediction module is configured to predict the development stage, the shape category in the accumulation stage, and the start time of the decomposition stage of the target submarine gas hydrate mound according to the feature vector of the target submarine gas hydrate mound and the corresponding methane gas flow rate data.

[0042] An electronic device includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned method for predicting the shape evolution of submarine hydrate mounds.

[0043] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned method for predicting the shape evolution of submarine hydrate mounds.

[0044] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0045] The method for predicting the shape evolution of submarine hydrate mounds provided by the present invention uses the image data compression module in the submarine hydrate mound evolution prediction model to compress the original image of the target submarine hydrate mound to obtain the compressed data of the target submarine hydrate mound. The image feature extraction module in the submarine hydrate mound evolution prediction model is used to extract features from the compressed data of the target submarine hydrate mound to obtain the feature vector of the target submarine hydrate mound. The evolution prediction module in the submarine hydrate mound evolution prediction model predicts its development stage according to the feature vector of the target submarine hydrate mound and the corresponding methane gas flow data, and when the development stage is in the accumulation stage, predicts its shape category and the start time of the decomposition stage, and can achieve accurate prediction of the shape evolution of submarine hydrate mounds. Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0047] Figure 1 It is a flowchart of the method for predicting the shape evolution of submarine hydrate mounds provided by the present invention;

[0048] Figure 2 It is the structure and working flowchart of the image data compression network based on the MobileNet architecture provided by the present invention;

[0049] Figure 3 It is the structure and working flowchart of the Inception module provided by the present invention;

[0050] Figure 4 It is a module diagram of the system for predicting the shape evolution of submarine hydrate mounds provided by the present invention. Detailed Embodiments

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] The object of the present invention is to provide a method, system, device and medium for predicting the shape evolution of submarine gas hydrate mounds, so as to achieve accurate prediction of the shape evolution of submarine gas hydrate mounds.

[0053] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0054] Embodiment 1

[0055] As Figure 1 shown, the present invention provides a method for predicting the shape evolution of submarine gas hydrate mounds, and the method includes:

[0056] Step 101: Obtain the original image of the target submarine gas hydrate mound and the corresponding methane gas flow rate data.

[0057] Step 102: Use the submarine gas hydrate mound evolution prediction model to predict the development stage, the shape category in the accumulation stage, and the start time of the decomposition stage of the target submarine gas hydrate mound according to the original image of the target submarine gas hydrate mound and the corresponding methane gas flow rate data; the development stage includes: the nucleation stage, the accumulation stage and the decomposition stage; the shape categories include: circular steep mound, circular gentle slope mound, elliptical steep mound, elliptical gentle slope mound, slender steep mound and slender gentle slope mound. The development stage, the shape category in the accumulation stage and the start time of the decomposition stage are used to characterize the evolution of the submarine gas hydrate mound.

[0058] Among them, the submarine gas hydrate mound evolution prediction model includes an image data compression module, an image feature extraction module and an evolution prediction module connected in series; the image data compression module is used to compress the data of the original image of the target submarine gas hydrate mound to obtain the compressed data of the target submarine gas hydrate mound; the image feature extraction module is used to extract features from the compressed data of the target submarine gas hydrate mound to obtain the feature vector of the target submarine gas hydrate mound; the evolution prediction module is used to predict the development stage, the shape category in the accumulation stage, and the start time of the decomposition stage of the target submarine gas hydrate mound according to the feature vector of the target submarine gas hydrate mound and the corresponding methane gas flow rate data.

[0059] Furthermore, the present invention also provides a method for determining an evolution prediction model of submarine gas hydrate mounds, specifically including:

[0060] Step S1: Obtain a sample data set; the sample data set includes a number of original images of sample submarine gas hydrate mounds with labels and corresponding methane gas flow data; the labels include: the development stage of the sample submarine gas hydrate mound, the shape category in the accumulation stage, and the start time of the decomposition stage.

[0061] As a specific implementation manner, step S1 includes: collecting images and data, and manually annotating the images and data.

[0062] 1.1 Continuously collect the original images of sample submarine gas hydrate mounds and methane gas flow data using an underwater camera and a methane sensor.

[0063] 1.2 According to the following rules, manually annotate the development stage and shape category of the submarine gas hydrate mound images that can be visually identified, and the methane gas flow data in the same time period is automatically classified into the corresponding development stage and shape category. Generally, the number of manually annotated images is not less than 75% of the collected data.

[0064] The development stage of the submarine gas hydrate mound can be divided according to Table 1:

[0065] Table 1 Development stage table of submarine gas hydrate mounds

[0066]

[0067] Taking the start time of stage 2 as the 0 moment, after dividing the development stage of the submarine gas hydrate mound, the start time of stage 3 should also be marked.

[0068] Among them, the submarine gas hydrate mounds in stage 2 can be divided into 6 categories in Table 2 according to the combination of the planar shape and the terrain:

[0069] Table 2 Shape category table of submarine gas hydrate mounds

[0070] Planar shape A Planar shape B Planar shape C Terrain 1 A1 B1 C1 Terrain 2 A2 B2 C2

[0071] Among them, the planar shape A is circular, the planar shape B is elliptical, and the planar shape C is elongated; the terrain 1 is a smooth, round, and steep mound, and the terrain 2 is a rough, uneven, and gentle slope mound. Then the shape category A1 is a circular steep mound, the shape category A2 is a circular gentle slope mound, the shape category B1 is an elliptical steep mound, the shape category B2 is an elliptical gentle slope mound, the shape category C1 is an elongated steep mound, and the shape category C2 is an elongated gentle slope mound.

[0072] After the manually labeled images of submarine hydrate mounds and the methane gas flow rate data for the corresponding time periods are processed in step S2, they will be used to train the initial neural network model constructed in step S4. During training, the 3-channel RGB image data of the submarine hydrate mounds and the methane gas flow rate data are used as the input of the neural network, and the development stage manually labeled, the shape categories in the accumulation stage, and the start time in the decomposition stage are used as the output of the neural network.

[0073] Step S2: Perform data augmentation on the original images of the labeled sample submarine hydrate mounds to obtain enhanced images of the labeled sample submarine hydrate mounds.

[0074] Furthermore, step S2 specifically includes: determining the mirror image of the sample submarine hydrate mound according to the original image of the sample submarine hydrate mound; respectively cropping the original image and the mirror image of the sample submarine hydrate mound to obtain the cropped images of the sample submarine hydrate mound; using the Principal Component Analysis (PCA) method to perform color enhancement on the cropped images of the sample submarine hydrate mound to obtain the enhanced images of the sample submarine hydrate mound; determining the label of the original image of the sample submarine hydrate mound as the label of the enhanced image of the sample submarine hydrate mound to obtain the enhanced images of the labeled sample submarine hydrate mounds.

[0075] Specifically, due to the absorption and scattering of incident light by water bodies, the video images collected underwater generally appear blue-green and have an obvious foggy effect. In addition, blurring, low contrast, color distortion, more noise, unclear details, and limited visible range are typical problems that reduce the quality of underwater video images. In order to improve the anti-interference ability of the system, the present invention needs to perform data augmentation on the original images in the sample dataset.

[0076] Improving the quality of the collected submarine hydrate mound images one by one will increase the computational cost. Therefore, the present invention performs data augmentation on the 3-channel RGB image data of the submarine hydrate mounds manually labeled in step S1, including the following two aspects:

[0077] 2.1 Crop the original image and the mirror image of the original image respectively.

[0078] 2.2 PCA color enhancement.

[0079] Step S3: Determine the original images of the labeled sample submarine hydrate mounds and the corresponding methane gas flow rate data, as well as the enhanced images of the labeled sample submarine hydrate mounds and the corresponding methane gas flow rate data as the training dataset.

[0080] Specifically, the labeled original images collected in step S1 and the enhanced images after data augmentation in step S2 are used as the image data in the training dataset, and are jointly used to train the initial neural network model constructed in step S4 in combination with the corresponding methane gas flow data.

[0081] Step S4: Construct an initial neural network model; the initial neural network model includes an image data compression network based on the MobileNet architecture, an image feature extraction network based on the Inception module, and an evolution prediction network; the image data compression network, the image feature extraction network, and the evolution prediction network are connected in series.

[0082] The structures of the image data compression network, the image feature extraction network, and the evolution prediction network are described in detail below.

[0083] 3.1 Image data compression network

[0084] The image data compression network includes a depthwise separable convolutional layer, a first residual network module, and a second residual network module connected in series; the depthwise separable convolutional layer includes a depth convolutional layer and a pointwise convolutional layer connected in series; wherein, the convolutional kernel size of the depth convolutional layer is 3×3, and the convolutional stride is 5.

[0085] Specifically, an image data compression network based on the MobileNet architecture is constructed to compress the original image or the enhanced image processed in step S2.

[0086] Taking the data compression of the original image as an example, the structure and working process of the image data compression network are as Figure 2 shown, including 1 depthwise separable convolutional layer and 2 residual network modules. Compared with ordinary convolution operations, the depthwise separable convolutional layer performs two-step operations of depth convolution and pointwise convolution on RGB image data, which can reduce the computational amount. The depth convolution uses a convolutional kernel of size 3×3 and a convolutional stride of 5.

[0087] 3.2 Image feature extraction network

[0088] The image feature extraction network includes a number of Inception modules connected in series and a fully connected layer; each of the Inception modules includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a pooling layer, and a channel connection layer; the channel connection layer is respectively connected to the first convolutional layer, the second convolutional layer, the third convolutional layer, and the pooling layer; the channel connection layer is used to splice the outputs of the first convolutional layer, the second convolutional layer, the third convolutional layer, and the pooling layer; wherein, the convolutional kernel size of the first convolutional layer is 1×1, the convolutional stride is 1, and the padding method is same; the convolutional kernel size of the second convolutional layer is 3×3, the convolutional stride is 1, and the padding method is same; the convolutional kernel size of the third convolutional layer is 5×5, the convolutional stride is 1, and the padding method is same; the pooling layer is a max pooling layer.

[0089] Specifically, an image feature extraction network based on Inception modules is constructed, specifically an Inception deep neural network, to extract features from the image data processed by the image data compression network. The structure and working process of the Inception module are as Figure 3 shown.

[0090] Among them, each Inception module contains 3 convolutional layers, 1 pooling layer, and 1 channel connection layer. The 3 convolutional layers respectively use convolutional kernels with sizes of 1×1, 3×3, and 5×5, the convolutional strides are all 1, and the padding method is same. The pooling layer is a max pooling layer. The channel connection layer splices the outputs of the 3 convolutional layers and 1 pooling layer as the input to the next Inception module or the final fully connected layer.

[0091] Preferably, the Inception deep neural network is composed of 10 Inception modules and a fully connected layer, and finally outputs a 1-dimensional feature vector.

[0092] 3.3 Evolution Prediction Network

[0093] The evolution prediction network includes a normalization processing layer, a first classifier, a second classifier, and a multiple linear regression layer; the normalization processing layer is connected to the first classifier; the first classifier is respectively connected to the second classifier and the multiple linear regression layer; the normalization processing layer is used to perform normalization processing on the input feature vector and methane gas flow data to obtain normalized data; the first classifier is used to determine the development stage according to the normalized data; the second classifier is used to determine the shape category in the accumulation stage according to the normalized data when the development stage is the accumulation stage; the multiple linear regression layer is used to adopt the multiple linear regression method to predict the start time of the decomposition stage according to the normalized data when the development stage is the accumulation stage.

[0094] Preferably, both the first classifier and the second classifier are Softmax classifiers.

[0095] Specifically, after normalizing the 1D feature vector output by the image feature extraction network and the methane gas flow data, input them into Softmax classifier 1 of the evolution prediction network. Use Softmax classifier 1 to output the development stage of the hydrate mound. For the hydrate mound with output stage 2, use Softmax classifier 2 to output the shape category of the hydrate mound (i.e., the shape category in the accumulation stage), and use the multiple linear regression layer to predict the start time of stage 3 (i.e., the start time of the decomposition stage).

[0096] Step S5: Use the training data set to train the initial neural network model to obtain a subsea hydrate mound evolution prediction model.

[0097] Specifically, use the training data set to train the initial neural network model. Determine the trained image data compression network as the image data compression module, determine the trained image feature extraction network as the image feature extraction module, and determine the trained evolution prediction network as the evolution prediction module to obtain a subsea hydrate mound evolution prediction model.

[0098] Embodiment 2

[0099] To execute the method corresponding to the above Embodiment 1 to achieve the corresponding functions and technical effects, a subsea hydrate mound shape evolution prediction system is provided below. As Figure 4 shown, the system includes:

[0100] A data acquisition module 401, configured to acquire the original image of the target subsea hydrate mound and the corresponding methane gas flow data.

[0101] An evolution prediction module 402, configured to use the subsea hydrate mound evolution prediction model to predict the development stage, the shape category in the accumulation stage, and the start time of the decomposition stage of the target subsea hydrate mound according to the original image of the target subsea hydrate mound and the corresponding methane gas flow data; the development stage includes: a nucleation stage, an accumulation stage, and a decomposition stage; the shape category includes: a circular steep mound, a circular gentle slope mound, an elliptical steep mound, an elliptical gentle slope mound, a slender steep mound, and a slender gentle slope mound.

[0102] Among them, the prediction model for the evolution of submarine hydrate mounds includes an image data compression module, an image feature extraction module, and an evolution prediction module connected in series; the image data compression module is used to compress the original image of the target submarine hydrate mound to obtain the compressed data of the target submarine hydrate mound; the image feature extraction module is used to extract features from the compressed data of the target submarine hydrate mound to obtain the feature vector of the target submarine hydrate mound; the evolution prediction module is used to predict the development stage, the shape category of the accumulation stage, and the start time of the decomposition stage of the target submarine hydrate mound according to the feature vector of the target submarine hydrate mound and the corresponding methane gas flow data.

[0103] Embodiment III

[0104] The embodiment of the present invention also provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor is used to run the computer program so that the electronic device executes the method for predicting the shape evolution of submarine hydrate mounds in Embodiment I. Preferably, the electronic device may be a server.

[0105] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method for predicting the shape evolution of submarine hydrate mounds in Embodiment I.

[0106] In summary, the method, system, device, and medium for predicting the shape evolution of submarine hydrate mounds provided by the present invention adopt a method of fusing multi-sensors and image recognition, and have the following advantages:

[0107] (1) When determining the prediction model for the evolution of submarine hydrate mounds, by performing data enhancement on the original image, the problem of unstable underwater image quality is addressed, and relatively accurate prediction of the evolution of submarine hydrate mounds can also be performed after the collected images are interfered.

[0108] (2) Using a neural network with the MobileNet architecture to process underwater images can reduce the amount of calculation, enabling the system for predicting the shape evolution of submarine hydrate mounds provided by the present invention to be integrated into other underwater detection systems, reducing the occupation of software and hardware resources.

[0109] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0110] In this article, specific examples are used to elaborate on the principles and implementation modes of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation modes and application scopes. To sum up, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A method for predicting the shape evolution of submarine hydrate mounds, characterized in that, Including: Obtaining the original image of the target submarine hydrate mound and the corresponding methane gas flow rate data; Using the submarine hydrate mound evolution prediction model, based on the original image of the target submarine hydrate mound and the corresponding methane gas flow rate data, predicting the development stage of the target submarine hydrate mound, the shape category of the accumulation stage, and the start time of the decomposition stage; The development stage includes: nucleation stage, accumulation stage, and decomposition stage; the shape category includes: circular steep mound, circular gentle slope mound, elliptical steep mound, elliptical gentle slope mound, slender steep mound, and slender gentle slope mound; Among them, the submarine hydrate mound evolution prediction model includes an image data compression module, an image feature extraction module, and an evolution prediction module connected in series; the image data compression module is obtained by training an image data compression network based on the MobileNet architecture; the image feature extraction module is obtained by training an image feature extraction network based on the Inception module; the evolution prediction module is obtained by training an evolution prediction network; The image data compression module is used to perform data compression on the original image of the target submarine hydrate mound to obtain the compressed data of the target submarine hydrate mound; The image feature extraction module is used to perform feature extraction on the compressed data of the target submarine hydrate mound to obtain the feature vector of the target submarine hydrate mound; The evolution prediction module is used to predict the development stage of the target submarine hydrate mound, the shape category of the accumulation stage, and the start time of the decomposition stage according to the feature vector of the target submarine hydrate mound and the corresponding methane gas flow rate data; The evolution prediction network includes a normalization processing layer, a first classifier, a second classifier, and a multiple linear regression layer; the normalization processing layer is connected to the first classifier; the first classifier is respectively connected to the second classifier and the multiple linear regression layer; The normalization processing layer is used to perform normalization processing on the input feature vector and methane gas flow rate data to obtain normalized data; The first classifier is used to determine the development stage according to the normalized data; The second classifier is used to determine the shape category of the accumulation stage according to the normalized data when the development stage is the accumulation stage; The multiple linear regression layer is used to adopt the multiple linear regression method to predict the start time of the decomposition stage according to the normalized data when the development stage is the accumulation stage.

2. The method for predicting the shape evolution of submarine hydrate mounds according to claim 1, wherein The determination method of the submarine hydrate mound evolution prediction model specifically includes: Obtaining a sample data set; the sample data set includes the original images of several labeled sample submarine hydrate mounds and the corresponding methane gas flow rate data; the labels include: the development stage of the sample submarine hydrate mound, the shape category of the accumulation stage, and the start time of the decomposition stage; Performing data enhancement on the original images of the labeled sample submarine hydrate mounds to obtain enhanced images of the labeled sample submarine hydrate mounds; Determine the original image of the labeled sample submarine hydrate mound and the corresponding methane gas flow data, as well as the enhanced image of the labeled sample submarine hydrate mound and the corresponding methane gas flow data as the training dataset; Construct an initial neural network model; the initial neural network model includes an image data compression network based on the MobileNet architecture, an image feature extraction network based on the Inception module, and an evolution prediction network; the image data compression network, the image feature extraction network, and the evolution prediction network are connected in series; Use the training dataset to train the initial neural network model to obtain a submarine hydrate mound evolution prediction model.

3. The method for predicting the shape evolution of submarine hydrate mounds according to claim 2, wherein The image data compression network includes a depthwise separable convolutional layer, a first residual network module, and a second residual network module connected in series; the depthwise separable convolutional layer includes a depth convolutional layer and a pointwise convolutional layer connected in series; Among them, the convolutional kernel size of the depth convolutional layer is 3×3, and the convolutional stride is 5.

4. The method for predicting the shape evolution of submarine hydrate mounds according to claim 2, characterized in that, The image feature extraction network includes a number of Inception modules and a fully connected layer connected in series; each Inception module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a pooling layer, and a channel connection layer; the channel connection layer is respectively connected to the first convolutional layer, the second convolutional layer, the third convolutional layer, and the pooling layer; the channel connection layer is used to splice the outputs of the first convolutional layer, the second convolutional layer, the third convolutional layer, and the pooling layer; Among them, the convolutional kernel size of the first convolutional layer is 1×1, the convolutional stride is 1, and the padding method is same; the convolutional kernel size of the second convolutional layer is 3×3, the convolutional stride is 1, and the padding method is same; the convolutional kernel size of the third convolutional layer is 5×5, the convolutional stride is 1, and the padding method is same; the pooling layer is max pooling.

5. The method for predicting the shape evolution of submarine hydrate mounds according to claim 2, characterized in that Perform data augmentation on the original image of the labeled sample submarine hydrate mound to obtain the enhanced image of the labeled sample submarine hydrate mound, specifically including: Determine the mirror image of the sample submarine hydrate mound according to the original image of the sample submarine hydrate mound; Crop the original image and the mirror image of the sample submarine hydrate mound respectively to obtain the cropped image of the sample submarine hydrate mound; Adopt the principal component analysis method to perform color enhancement on the cropped image of the sample submarine hydrate mound to obtain the enhanced image of the sample submarine hydrate mound; Determine the label of the original image of the sample submarine hydrate mound as the label of the enhanced image of the sample submarine hydrate mound to obtain the enhanced image of the labeled sample submarine hydrate mound.

6. The method for predicting the shape evolution of submarine hydrate mounds according to claim 2, wherein, Use the training dataset to train the initial neural network model to obtain a submarine hydrate mound evolution prediction model, specifically including: Train the initial neural network model using the training data set, determine the trained image data compression network as the image data compression module, determine the trained image feature extraction network as the image feature extraction module, and determine the trained evolution prediction network as the evolution prediction module to obtain a submarine hydrate mound evolution prediction model.

7. A prediction system for the shape evolution of submarine hydrate mounds, characterized in that, It includes: A data acquisition module for acquiring the original image of a target submarine hydrate mound and the corresponding methane gas flow data; An evolution prediction module for using the submarine hydrate mound evolution prediction model to predict the development stage, the shape category of the accumulation stage, and the start time of the decomposition stage of the target submarine hydrate mound according to the original image of the target submarine hydrate mound and the corresponding methane gas flow data; The development stage includes: a nucleation stage, an accumulation stage, and a decomposition stage; the shape categories include: circular steep mound, circular gentle slope mound, elliptical steep mound, elliptical gentle slope mound, elongated steep mound, and elongated gentle slope mound; Among them, the submarine hydrate mound evolution prediction model includes an image data compression module, an image feature extraction module, and an evolution prediction module connected in series; The image data compression module is used to compress the data of the original image of the target submarine hydrate mound to obtain the compressed data of the target submarine hydrate mound; The image feature extraction module is used to extract features from the compressed data of the target submarine hydrate mound to obtain the feature vector of the target submarine hydrate mound; The evolution prediction module is used to predict the development stage, the shape category of the accumulation stage, and the start time of the decomposition stage of the target submarine hydrate mound according to the feature vector of the target submarine hydrate mound and the corresponding methane gas flow data.

8. An electronic device, characterized in that, It includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the submarine hydrate mound shape evolution prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by the processor, it implements the submarine hydrate mound shape evolution prediction method according to any one of claims 1 to 6.

Images