A shoreline identification method, device, equipment and medium based on data enhancement

Abstract

The application discloses a kind of shore line identification method, device, equipment and medium based on data enhancement, method includes: obtaining remote sensing image dataset;According to the cloud data enhancement and mixed enhancement processing of isolation trigger algorithm to remote sensing image dataset, obtain training dataset;Training data set is input into the neural network model constructed in advance and is handled to obtain shore line identification model;Obtain the remote sensing image data to be identified, and the remote sensing image data to be identified are input into shore line identification model, and shore line identification result is obtained.The application embodiment is through cloud layer data enhancement and batch mixing to carry out data enhancement to training data, can make model in complex and changeable dataset simultaneously learn the advanced feature of spectral value and water body form, improve the shore line identification accuracy of remote sensing image, can be widely applied in shore line identification technical field.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of shoreline identification, and in particular to a shoreline identification method and device based on data enhancement, equipment and medium. BACKGROUND

[0002] The shoreline identification technology is one of important applications of geographic information systems (GIS), remote sensing technology and computer vision in the field of ocean. In the related technology, the shoreline identification method mainly depends on manual interpretation and simple image processing technology. These methods have large workload, low efficiency, and are difficult to cope with large-scale or dynamic change scenes. In summary, the technical problems existing in the related technology need to be improved. SUMMARY

[0003] The main purpose of the embodiments of the present application is to provide a shoreline identification method and device based on data enhancement, which can improve the automation and accuracy of shoreline identification.

[0004] To achieve the above-mentioned purpose, one aspect of the embodiments of the present application provides a shoreline identification method based on data enhancement, which comprises:

[0005] Obtain a remote sensing image data set;

[0006] According to the isolation trigger algorithm, the remote sensing image data set is processed by cloud data enhancement and mixed enhancement to obtain a training data set;

[0007] The training data set is input into a pre-constructed neural network model for training processing to obtain a shoreline identification model;

[0008] Obtain a to-be-identified remote sensing image data, input the to-be-identified remote sensing image data into the shoreline identification model, and obtain a shoreline identification result.

[0009] In some embodiments, the cloud data enhancement and mixed enhancement processing of the remote sensing image data set according to the isolation trigger algorithm to obtain the training data set comprises the following steps:

[0010] The cloud layer data and batch data are obtained by screening from the remote sensing image data set;

[0011] The remote sensing image data set is processed by random enhancement strategy trigger to obtain first to-be-processed data and second to-be-processed data;

[0012] The first to-be-processed data is processed by cloud data enhancement according to the cloud layer data to obtain first enhanced data;

[0013] The second to-be-processed data is processed by mixed enhancement according to the batch data to obtain second enhanced data;

[0014] determining the first enhancement data and the second enhancement data as the training data set.

[0015] In some embodiments, the cloud data enhancement processing of the first to-be-processed data according to the cloud layer data to obtain the first enhancement data comprises the following steps:

[0016] performing data preprocessing on the cloud layer data to obtain normalized data and a transparent mask layer;

[0017] performing random rotation processing on the normalized data to obtain morphological enhancement data;

[0018] performing random cloud layer covering processing on the first to-be-processed data through the morphological enhancement data in combination with the transparent mask layer to obtain the first enhancement data.

[0019] In some embodiments, the mixed enhancement processing of the second to-be-processed data according to the batch data to obtain the second enhancement data comprises the following steps:

[0020] performing random region segmentation processing on the batch data to obtain a segmented image;

[0021] covering the segmented image to the second to-be-processed data in sequence to obtain the second enhancement data.

[0022] In some embodiments, before the training processing of the training data set into the pre-constructed neural network model to obtain the shoreline recognition model, the method further comprises pre-constructing the neural network model, specifically comprising the following steps:

[0023] constructing an encoder-decoder backbone network according to a reverse residual structure and a linear bottleneck layer;

[0024] connecting the encoder-decoder backbone network through a skip link module to obtain the neural network model.

[0025] In some embodiments, the training processing of the training data set into the pre-constructed neural network model to obtain the shoreline recognition model comprises the following steps:

[0026] performing cross-validation division processing on the training data set to obtain a training data combination;

[0027] training and verifying the training data combination into the neural network model to obtain the shoreline recognition model.

[0028] In some embodiments, the training and verifying of the training data combination into the neural network model to obtain the shoreline recognition model comprises the following steps:

[0029] combining the training data into the neural network model to obtain a prediction result;

[0030] performing loss calculation processing on the prediction result according to a dice loss function to obtain a training loss value;

[0031] updating parameters of the neural network model according to a learning rate decay strategy in combination with the training loss value to obtain a shoreline identification model.

[0032] To achieve the above object, another aspect of the embodiment of the present application proposes a shoreline identification device based on data enhancement, which comprises:

[0033] A first module is configured to acquire a remote sensing image dataset.

[0034] A second module is configured to perform cloud data enhancement and mixed enhancement processing on the remote sensing image dataset according to an isolation trigger algorithm to obtain a training dataset.

[0035] A third module is configured to input the training dataset into a pre-constructed neural network model to perform training processing and obtain a shoreline identification model.

[0036] A fourth module is configured to acquire to-be-identified remote sensing image data, input the to-be-identified remote sensing image data into the shoreline identification model, and obtain a shoreline identification result.

[0037] To achieve the above object, another aspect of the embodiment of the present application proposes an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the method described above when executing the computer program.

[0038] To achieve the above object, another aspect of the embodiment of the present application proposes a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method described above.

[0039] The embodiments of the present application at least have the following beneficial effects: The present application provides a shoreline identification method, device, equipment and medium based on data enhancement, which obtains a remote sensing image dataset; performs cloud data enhancement and mixed enhancement processing on the remote sensing image dataset according to an isolation trigger algorithm to obtain a training dataset; inputs the training dataset into a pre-constructed neural network model for training processing to obtain a shoreline identification model; obtains to-be-identified remote sensing image data, inputs the to-be-identified remote sensing image data into the shoreline identification model, and obtains a shoreline identification result. The embodiments of the present application perform data enhancement on the training data through cloud layer data enhancement and batch mixing, which can enable the model to simultaneously learn the advanced features of spectral values and water body shapes in complex and changeable datasets, and improve the shoreline identification accuracy of the remote sensing image. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of a shoreline identification method based on data enhancement provided by the embodiments of the present application;

[0041] Figure 2 is a schematic diagram of cloud layer data provided by the embodiments of the present application;

[0042] Figure 3 is a schematic diagram of spectral data and visual display provided by the embodiments of the present application;

[0043] Figure 4 is a color image schematic diagram of cloud layer shielding provided by the embodiments of the present application;

[0044] Figure 5 is a mask image schematic diagram of cloud layer shielding provided by the embodiments of the present application;

[0045] Figure 6 is a schematic diagram of random region segmentation provided by the embodiments of the present application;

[0046] Figure 7 is a schematic diagram of mixed enhancement provided by the embodiments of the present application;

[0047] Figure 8 is a structural schematic diagram of a neural network model provided by the embodiments of the present application;

[0048] Figure 9 is a schematic diagram of an identification result not using the present scheme provided by the embodiments of the present application;

[0049] Figure 10 is a schematic diagram of an identification result using the present scheme provided by the embodiments of the present application. DETAILED DESCRIPTION

[0050] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application. When the following description refers to the accompanying drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with embodiments of the present application. They are merely examples of apparatuses and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.

[0051] It can be understood that the terms "first", "second" and the like used in the present application can be used herein to describe various concepts, but unless specifically stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiments of the present application, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information. Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon determination" or "in response to a determination".

[0052] The terms "at least one", "multiple", "each", "any" and the like used in the present application include one, two or more than two, multiple includes two or more than two, each refers to each of the corresponding plurality, and any refers to any one of the plurality.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0054] Coastline recognition technology is one of the important applications of geographic information systems (GIS), remote sensing technology, and computer vision in the field of ocean. Coastline, the boundary between land and water, is of great significance for coastal resource management, environmental protection, coastal erosion monitoring, and island boundary demarcation. With the rapid development of remote sensing technology and satellite imaging technology, large-scale, high-resolution surface images can be obtained, providing a wealth of raw data for coastline recognition. However, the complexity of the natural properties of the coastline and human activity interference makes the automated and intelligent recognition of the coastline a challenging task. In remote sensing water body recognition, cloud cover is an important problem that directly affects the completeness of water body extraction. If the cloud layer is just above the water body, the spectral information reflected by the water body is difficult to be received by the satellite, so the location of the water body on the remote sensing image lacks available water body information, resulting in the inability to extract the true coastline. In small-scale studies, the coastline can be extracted by flexibly selecting areas without cloud cover. However, in large-scale studies, cloud cover is a common problem in remote sensing images, making it difficult to quickly and efficiently extract large-scale water body coastlines from remote sensing images.

[0055] In the study of remote sensing water body recognition and coastline extraction, the presence of clouds is a major challenge. Cloud cover can cause the spectral information reflected by the water body to be incomplete and not fully received by satellite sensors, resulting in the loss of important information at the water body location on the remote sensing image, which affects the accurate recognition of the water body and the precise extraction of the coastline. In particular, in large-scale water body monitoring, the universality and randomness of cloud cover increase the complexity of identification and analysis. Traditional coastline recognition methods mainly rely on manual interpretation and simple image processing techniques, which are labor-intensive, inefficient, and difficult to cope with large-scale or dynamic scenarios. Therefore, researchers have begun to explore the use of automated and intelligent methods to improve the accuracy and efficiency of coastline recognition.

[0056] Coastline recognition using machine learning and deep learning techniques is a new method of automated digital coastline generation that can replace traditional manual annotation. Convolutional neural networks (CNNs) are used for feature extraction and classification to identify the coastline area in images. This method can automatically learn the features related to coastline recognition in images without complex feature engineering or human intervention. By training a neural network model, the location of the coastline can be automatically identified from satellite images. The advantage of this method is that it can handle complex image features, adapt to different terrains and lighting conditions, and improve recognition accuracy and robustness as more data is used for training. The embodiments of the present application improve the automation level and accuracy of coastline recognition by applying deep learning techniques. By optimizing the neural network model and algorithm, the goal is to develop a technology that can quickly and accurately identify and monitor coastline changes from a large number of satellite images to support key applications such as coastal resource management, environmental monitoring and policy making. This will bring new technical breakthroughs to the field of coastline recognition and is expected to be widely used in related fields.

[0057] Therefore, in the embodiments of the present application, a coastline recognition method, device, equipment and medium based on data enhancement are provided. The scheme obtains a remote sensing image dataset, performs cloud data enhancement and mixed enhancement processing on the remote sensing image dataset according to an isolation trigger algorithm to obtain a training dataset, inputs the training dataset into a pre-constructed neural network model for training processing to obtain a coastline recognition model, obtains to-be-recognized remote sensing image data, inputs the to-be-recognized remote sensing image data into the coastline recognition model, and obtains a coastline recognition result. The embodiments of the present application perform data enhancement on the training data through cloud layer data enhancement and batch mixing, which can enable the model to simultaneously learn the advanced features of spectral values and water body shapes in complex and variable datasets, thereby improving the coastline recognition accuracy of remote sensing images.

[0058] The embodiment of the application provides a kind of shore line identification method based on data enhancement, it is related to shore line identification technical field.The shore line identification method based on data enhancement provided by the embodiment of the application can be applied in terminal, also can be applied in server, it can also be software running in terminal or server.In some embodiments, terminal can be smart phone, tablet computer, notebook computer, desktop computer, smart speaker, smart watch and vehicle terminal etc., but not limited to this;Server end can be configured as independent physical server, can also be configured as the server cluster or distributed system of multiple physical servers, can also be configured as the cloud server of providing cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN and basic cloud computing services such as big data and artificial intelligence platform, server can also be a node server in blockchain network;Software can be application for realizing shore line identification method based on data enhancement etc., but not limited to the above form.

[0059] The application can be used in many general or special computer system environments or configurations. For example: personal computer, server computer, handheld device or portable device, tablet device, multi-processor system, microprocessor-based system, set-top box, programmable consumer electronics, network PC, small computer, large computer, distributed computing environment including any of the above systems or devices, etc. The application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The application can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0060] Figure 1 The method in the embodiment of the application provides a kind of shore line identification method based on data enhancement, Figure 1 The method in the embodiment of the application provides a kind of shore line identification method based on data enhancement,

[0061] Step S101, obtain remote sensing image data set;

[0062] Step S102, according to the cloud data enhancement and hybrid enhancement processing of isolation trigger algorithm to the remote sensing image data set, obtains training data set;

[0063] Step S103, the training data set is input into the neural network model constructed in advance and is trained to obtain shore line identification model;

[0064] Step S104, obtaining to-be-identified remote sensing image data, inputting the to-be-identified remote sensing image data into the coastline identification model, and obtaining a coastline identification result.

[0065] The steps S101 to S104 shown in the embodiments of the present application can obtain the remote sensing image dataset through online data platform download, or obtain the remote sensing image dataset through the remote sensing image of the Sentinel-2 satellite as a data source, and take the remote sensing image dataset as the original training dataset. Then, cloud data enhancement and hybrid enhancement processing are performed on the remote sensing image dataset according to the isolated trigger algorithm. The isolated trigger algorithm means that only one data enhancement strategy is triggered at any time point. Such processing can ensure that the effect of each enhancement strategy can be fully utilized, while avoiding excessive complexity of the training data, and ensuring the effectiveness and efficiency of the model training. In the embodiments of the present application, the data enhancement strategies include a cloud data enhancement strategy and a hybrid enhancement strategy. The cloud data enhancement strategy can fuse thick cloud layer data into the original training data, enable the model to learn high-level features when processing the occluded water body area, enhance the model's recognition ability of water body continuity, thereby improving the generalization ability of the remote sensing image coastline identification and the coherence of large-scale river waterways, and improving the accuracy of coastline identification. The hybrid enhancement strategy destroys the integrity of the original data image, can generate a large amount of training data online, not only increases the difficulty of the data, but also enables the model to learn higher-level water body knowledge in a complex data environment, thereby improving the accuracy of coastline identification. The training dataset after the data enhancement processing is input into the pre-constructed neural network model for training processing, to obtain the coastline identification model. In view of the problem of many noise points in the surface water identification, the coherence of the natural coastline is focused on, and a thick cloud layer shielding training strategy is proposed. Through the way of fusing the semantic segmentation technology and the image restoration technology principle, a deep learning model capable of generating more natural coastline is trained. Thus, the to-be-identified remote sensing image data for coastline identification is obtained, the to-be-identified remote sensing image data is input into the coastline identification model for identification, and a coastline identification result is obtained. The model of the embodiments of the present application is sensitive to the natural coastline, can determine whether the current water element can be identified as the water body of the natural coastline according to the cross-pixel information on the remote sensing image, and thus generates a more regular coastline identification result.

[0066] In some embodiments, the cloud data enhancement and hybrid enhancement processing on the remote sensing image dataset according to the isolated trigger algorithm to obtain the training dataset includes the following steps:

[0067] Filtering cloud layer data and batch data from the remote sensing image dataset;

[0068] perform random enhancement strategy triggering processing on the remote sensing image dataset to obtain first to-be-processed data and second to-be-processed data;

[0069] perform cloud data enhancement processing on the first to-be-processed data according to the cloud layer data to obtain first enhanced data;

[0070] perform mixed enhancement processing on the second to-be-processed data according to the batch data to obtain second enhanced data;

[0071] determine the first enhanced data and the second enhanced data as the training dataset.

[0072] In the embodiment of the present application, cloud layer data and batch data are screened from a remote sensing image dataset, wherein the cloud layer data is remote sensing image data of a period when cloud layers cover more, and the batch data is data randomly extracted from the remote sensing image dataset. The mixed enhancement method and the cloud enhancement method are isolated by the isolation triggering algorithm in the embodiment of the present application. When data processing is performed, one enhancement strategy is triggered every time, but both enhancement strategies are not triggered at the same time. That is, only one of the cloud enhancement and the batch mixed data enhancement scheme is randomly executed when data enhancement is performed each time, and the probability of triggering of the two is 50% each. The embodiment of the present application can randomly classify the remote sensing image dataset to obtain first to-be-processed data and second to-be-processed data before data enhancement processing is performed, wherein the first to-be-processed data is data processed by cloud data enhancement processing of the cloud layer data, and the first enhanced data can be obtained after cloud data enhancement; the second to-be-processed data is data processed by mixed enhancement processing of the batch data, and the second enhanced data can be obtained after mixed enhancement processing; and finally the first enhanced data and the second enhanced data are determined as the training dataset. The embodiment of the present application can make the effect of each enhancement strategy be fully utilized, while avoiding overcomplication of the training data, and ensuring the effectiveness and efficiency of model training.

[0073] In some embodiments, the cloud data enhancement processing on the first to-be-processed data according to the cloud layer data to obtain first enhanced data comprises the following steps:

[0074] perform data preprocessing on the cloud layer data to obtain normalized data and a transparent mask layer;

[0075] perform random rotation processing on the normalized data to obtain morphological enhancement data;

[0076] perform random cloud layer coverage processing on the first to-be-processed data by the morphological enhancement data in combination with the transparent mask layer to obtain the first enhanced data.

[0077] In the embodiment of the present application, the cloud layer data is screened from the remote sensing image data set, the spectral data of the thick cloud layer that can completely block the ground information is extracted from the image data of the period with more cloud coverage, and is used as the blocking data used in the self-supervised learning of the deep learning model of the embodiment of the present application. Please refer to Figure 2 , Figure 2 The embodiment of the present application provides a schematic diagram of cloud layer data. Then, the cloud layer data is preprocessed, the cloud layer data is normalized, the z-score and chopping truncation strategy are used, the Z-score is the result of the difference between an observation value or a data point and the average divided by the standard deviation, the chopping truncation strategy is to remove outliers from the normalized data, by taking only the data within the range after the normalized data, and then normalizing the data, so that the processed data will not be affected by outliers. The embodiment of the present application normalizes the spectral type data to 0 to 1 to obtain normalized data, which is convenient for subsequent deep learning. Please refer to Figure 3 , Figure 3 The embodiment of the present application generates a transparent mask layer, which is a black area in the visualization icon in Figure 3 . The transparent mask layer is used for transparent calculation when blocking, only the non-transparent area (cloud) is overlaid on the cloud-free remote sensing image for training. Then, the cloud shape of the normalized data is randomly rotated by a random rotation angle, so as to increase the original cloud data shape, and finally the random coverage site is set, the rectangular cloud data in the normalized data is intercepted, and a pixel point in the first to-be-processed data can be arbitrarily selected as the coverage center site. The rectangular center point of the randomly rotated cloud map is overlapped with the coverage site, and the non-black edge cloud layer remote sensing data is used to replace the original first to-be-processed data through the transparent mask layer, i.e. the black edge mask data in the visualization area, to obtain the first enhanced data. Please refer to Figure 4 and Figure 5 , the embodiment of the present application adds unknown thick cloud layer data to the original first to-be-processed data to block the waterway, forces the model to fit and restore the blocked area of the waterway according to the unblocked area, can obtain the self-supervised training data enhancement part with complete thick cloud layer addition, and through the cloud layer blocking method, the spectral data of the water body pixels under the natural coastline can be completely blocked, and the final water body true value data is not changed. For Figure 4 and Figure 5The red portion, due to the complete lack of spectral information in the obscured area, cannot yield the correct estuary shoreline if the model only establishes a paradigm based on water spectral data. The model must rely on the spectral information of the water surrounding the obscured cloud layer to model the holistic and continuous characteristics of the water channels along the estuary shoreline at a higher level of abstraction to achieve the final training score. Based on the cloud-enhanced training strategy proposed in this application, the model can effectively learn the overall characteristics of the target object and infer the natural shoreline based on the target object's own information combined with surrounding information, thereby effectively suppressing the generation of fragmented land-based water bodies. The cloud enhancement method in this application mainly refers to extracting thick cloud images from different remote sensing images and overlaying them onto the remote sensing images used for model training, causing the thick clouds to partially obscure the water body and interrupt the surface water system. In this case, the deep learning model will have to use spectral data from the upstream and downstream of the water system to refine the shoreline inference and actively learn the continuity features of the water body. This ensures that the final extraction effect is not affected by features such as paddy fields, generating a smoother and more even shoreline and improving the shoreline recognition results.

[0078] In some embodiments, performing hybrid enhancement processing on the second data to be processed based on the batch data to obtain the second enhanced data includes the following steps:

[0079] The batch data is subjected to random region segmentation to obtain segmented images;

[0080] The segmented images are sequentially overlaid onto the second data to be processed to obtain the second enhanced data.

[0081] In this embodiment, random regions are selected from a randomly selected batch of data, and the segmented random region images are sequentially overlaid onto the second data to be processed, thereby destroying the integrity of the original data image and increasing the amount of training data. In this embodiment, when performing random region segmentation, the ground truth values ​​corresponding to the random regions are also included. Figure 1 The segmentation process is then performed, overwriting the images to be trained and compromising the integrity of the original data. (Refer to...) Figure 6 and Figure 7 , Figure 6 Images obtained by randomly segmenting batch data. Figure 7 To overlay the segmented image onto the second batch of data to be processed, this embodiment selects random regions (along with their corresponding ground truth maps) from multiple images in a randomly generated batch of data and sequentially overlays them onto the next image, thereby disrupting the integrity of the original data image. This embodiment can generate a large amount of training data, which not only increases the difficulty of the data but also enables the model to learn more advanced water body knowledge in a complex data environment, thereby improving the accuracy of shoreline identification.

[0082] In some embodiments, before the training data set is input into the pre-constructed neural network model for training processing to obtain the shoreline identification model, the method further comprises pre-constructing the neural network model, specifically comprising the following steps:

[0083] An encoder-decoder backbone network is constructed according to the inverted residual structure and the linear bottleneck layer;

[0084] The encoder-decoder backbone network is connected by a skip link module to obtain the neural network model.

[0085] In the embodiments of the present application, the UNet++ model with the mobilenet_v2 network structure as the backbone network can be used to construct the neural network model, and the generative adversarial network (GANs) and other network models can also be used as the neural network model. Referring to Figure 8 , the encoder-decoder backbone network is constructed according to the inverted residual structure and the linear bottleneck layer in the embodiments of the present application, and the original feature map can be initialized and up-sampled and down-sampled by the encoder-decoder backbone network. The mobilenet_v2 network structure mainly adopts lightweight design and focuses on implementing efficient deep learning models on mobile devices and embedded devices. The inverted residual structure first performs dimensionality increasing processing through 1x1 convolution, then performs feature extraction through depthwise separable convolution (Depthwise Separable Convolution), and finally performs dimensionality reduction processing through 1x1 convolution, which can help capture more feature information, especially in low-dimensional space. The linear bottleneck layer adopts a linear activation function (such as ReLU6) instead of the traditional ReLU activation function. This is because in low-dimensional space, the ReLU activation function may destroy feature information, while the linear activation function can avoid this situation. The UNet++ model is a deep supervision encoder-decoder network, in which the subnetworks of the encoder and the decoder are connected through a series of nested and dense skip links. Therefore, the encoder-decoder backbone network is connected by the skip link module in the embodiments of the present application, so that the network can utilize feature information of both shallow and deep layers at the same time, improving the accuracy of segmentation. The neural network model is constructed by using deep learning technology in the embodiments of the present application to simulate and generate missing or incomplete water body and shoreline information, thereby improving the overall analysis and prediction accuracy.

[0086] In some embodiments, the training data set is input into the pre-constructed neural network model for training processing to obtain the shoreline identification model, comprising the following steps:

[0087] cross-validation division processing is performed on the training data set to obtain a plurality of training data combinations, and then the plurality of training data combinations are respectively input into a neural network model for training and verification, and the model that performs best on the verification set is taken as the shoreline identification model.

[0088] The training data combination is input into the neural network model for training and verification to obtain a shoreline identification model.

[0089] In the embodiments of the present application, a plurality of training data combinations are obtained by performing cross-validation division processing on the training data set, and then the plurality of training data combinations are respectively input into a neural network model for training and verification, and the model that performs best on the verification set is taken as the shoreline identification model. Specifically, the training data set is re-divided by a five-fold cross-validation method to form five different training set and verification set combinations. At the same time, the original verification set (val) is redefined as the test set (test) of the final model. This division method allows five independent models to be trained, each model selects the epoch that performs best on its corresponding verification set as the final model, and the comprehensive performance is evaluated on the unified test set. It should be noted that the training data set is also preprocessed before cross-validation in the embodiments of the present application, and the z-score standardization method is used to preprocess the remote sensing image data. The formula of this method is as follows:

[0090]

[0091] In the formula, X is the observed value, μ is the mean of the data set, and σ is the standard deviation of the data set. This method converts the original data to a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation, thereby achieving data standardization. When processing the Sent i ne l-2 remote sensing image data, the embodiments of the present application also calculate the mean and variance of each channel. The data processed by the z-score standardization method performs more stable and consistent performance in the subsequent model training and inference process. This standardization method not only helps to improve the convergence speed of the algorithm, but also reduces the deviation between different channels, thereby improving the overall generalization ability of the model.

[0092] In some embodiments, the inputting the training data combination into the neural network model for training and verification to obtain a shoreline identification model comprises the following steps:

[0093] The training data combination is input into the neural network model to obtain a prediction result.

[0094] According to the dice loss function, the prediction result is subjected to loss calculation processing to obtain a training loss value.

[0095] According to the learning rate decay strategy, the parameters of the neural network model are updated in combination with the training loss value to obtain a shoreline identification model.

[0096] In the embodiments of the present application, since the data augmentation is adopted to increase the learning difficulty of the original data set, the shoreline recognition model is forced to learn higher-level features, which leads to the fact that the training set score and the test set score are very close during the entire training process, indicating that the model can effectively generalize the learned data features. However, there is a significant difference between the training set score and the test set score when the model is trained without the data augmentation method of the embodiments of the present application, which indicates that the model has a serious overfitting problem, meaning that the model mainly learns the data distribution characteristics suitable for the training set, and these characteristics are not completely applicable to the related learning task. The embodiments of the present application specifically obtain a prediction result by inputting the training data into the neural network model, and then obtain a training loss value by loss calculation and processing on the prediction result according to the dice loss function. The expression of the dice loss function is as follows:

[0097]

[0098] In the formula, |X∩Y| represents the intersection of two sets X and Y, and |X| and |Y| are the sizes of the two sets respectively. The value range of the coefficient is between 0 and 1, where 1 represents complete similarity and 0 represents no similarity. The dice loss (DiceLoss) is a function derived from the dice coefficient and is used as a loss function in machine learning models, especially in segmentation tasks. The embodiments of the present application increase the similarity between the predicted segmentation and the real situation by minimizing the dice loss. Finally, the parameters of the neural network model are updated according to the learning rate decay strategy combined with the training loss value to obtain the shoreline recognition model. Specifically, the model training method of the embodiments of the present application adopts the poly learning rate decay strategy, the total training cycle number is set to 1000, the loss function adopts the dice loss (Dice Loss), and the evaluation index is set as the EMA Dice index average dice coefficient. The expression of the learning rate decay strategy is as follows:

[0099]

[0100] In the formula, Learning Rate represents the learning rate at the current iteration number, Initial Learning Rate is the initial learning rate, Current Epoch is the current iteration generation number, Max Epochs is the total iteration number, and Power is a power parameter that controls how fast the learning rate decreases. Through setting 1000 epochs of training, the embodiments of the present application can maximize the advantages of the online enhancement strategy, so that the model can learn very complex high-level features from difficult and easy interlaced tasks. In addition, the poly learning rate decays very slowly, and the model can gradually transition from a high learning rate to a low learning rate very smoothly, so that the global optimal solution of the model is obtained in the change of the learning rate.

[0101] Next, the scheme of the embodiments of the present application will be described and explained in detail in combination with specific application examples:

[0102] The embodiments of the present application can be applied to the field of shoreline identification technology, and can identify the shoreline of large-scale remote sensing images. Specifically, the remote sensing image data set as the original training data is processed by cloud data enhancement and hybrid enhancement, so as to emphasize the strategy of deliberately obscuring and destroying the original terrain according to the design idea of morphological obscuring mask, input the training data set after data enhancement into the pre-constructed neural network model for training to obtain a shoreline identification model, so as to force the model to generate invisible ground features based on the learned high-level features of the coherence and integrity of waterways and land in the case of not knowing the real spectral value of the ground, and has the advantages of suppressing small water bodies, generating overall land blocks and smooth waterways. The embodiments of the present application have the advantage of fewer water body noise points in batch processing of large-scale remote sensing images and studying macro waterway and land changes, thereby improving the efficiency and accuracy of shoreline identification. The embodiments of the present application realize the learning of high-level features of the model in processing the obscured water body area by fusing thick cloud layer data into the original training data, enhance the recognition ability of the model to water body continuity, thereby improving the generalization ability of remote sensing image water body recognition and the coherence of large-scale river waterways, and through the self-supervised learning method of thick cloud obscuring and batch hybrid isolation, the image consistency of the original training image is destroyed, and the model is forced to learn the high-level features of the spectrum value and water body shape. Referring to Figure 9 and Figure 10 , Figure 9 is the identification result not using the scheme, Figure 10To use the recognition result of the scheme, it can be known by comparison that the model of the embodiment of the application can effectively learn the overall characteristics of the target object, and the natural shoreline is inferred according to the information of the target object itself combined with the surrounding information, thereby effectively suppressing the generation of fragmented land water bodies. Specifically, the cloud enhancement adopted in the scheme is mainly used to improve the generalization ability of the model, so that when the model performs shoreline identification, it not only judges whether the current pixel point is a water body based on the remote sensing spectral data of the pixel point, but also learns more spectral data of the surrounding pixels, thereby achieving the final effect of suppressing land water body noise and highlighting the natural shoreline water body path. The embodiment of the application can optimize the data of the land area and the cloud layer shielding, promote the performance of the model in generating more overall water bodies and land areas, and effectively suppress the phenomenon of a large number of point-shaped water field area noise in large-scale remote sensing coastal images.

[0103] The embodiment of the application also provides a shoreline identification device based on data enhancement, which can implement the shoreline identification method based on data enhancement.

[0104] The first module is configured to obtain a remote sensing image data set.

[0105] The second module is configured to perform cloud data enhancement and hybrid enhancement processing on the remote sensing image data set according to an isolation trigger algorithm to obtain a training data set.

[0106] The third module is configured to input the training data set into a pre-constructed neural network model for training processing to obtain a shoreline identification model.

[0107] The fourth module is configured to obtain to-be-identified remote sensing image data, input the to-be-identified remote sensing image data into the shoreline identification model, and obtain a shoreline identification result.

[0108] It can be understood that the contents in the above method embodiments are all applicable to the device embodiments, the device embodiments specifically implement the functions of the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0109] The embodiment of the application also provides an electronic device including a memory and a processor, the memory stores a computer program, and the processor implements the above shoreline identification method based on data enhancement when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0110] It can be understood that the contents in the above method embodiments are all applicable to the device embodiments, the device embodiments specifically implement the functions of the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0111] The embodiment of the application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the shore line identification method based on data enhancement.

[0112] It can be understood that the contents in the method embodiments are applicable to the storage medium embodiments, the storage medium embodiments specifically realize the functions of the method embodiments, and the beneficial effects achieved by the storage medium embodiments are the same as the beneficial effects achieved by the method embodiments.

[0113] The memory is a non-transitory computer readable storage medium, and can be used to store a non-transitory software program and a non-transitory computer executable program. In addition, the memory can include a high-speed random access memory, and can also include a non-transitory memory, for example, at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor, and the remote memory can be connected to the processor through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0114] The embodiment of the application provides a shore line identification method, device, equipment and medium based on data enhancement. The method comprises the following steps: acquiring remote sensing image data set; performing cloud data enhancement and mixed enhancement processing on the remote sensing image data set according to an isolation trigger algorithm to obtain a training data set; inputting the training data set into a pre-constructed neural network model for training processing to obtain a shore line identification model; acquiring remote sensing image data to be identified; inputting the remote sensing image data to be identified into the shore line identification model to obtain a shore line identification result. The embodiment of the application can perform data enhancement on the training data through cloud layer data enhancement and batch mixing, can enable the model to simultaneously learn high-level features of spectral values and water body shapes in complex and changeable data sets, and improves the shore line identification accuracy of the remote sensing image.

[0115] The embodiments described in the embodiments of the application are used to more clearly illustrate the technical solutions of the embodiments of the application, and do not constitute a limitation on the technical solutions provided by the embodiments of the application. It can be understood by those skilled in the art that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the application are also applicable to similar technical problems.

[0116] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the application, and can include more or fewer steps than the figures, or combine certain steps, or different steps.

[0117] The apparatus embodiments described above are merely exemplary, and the units described as separate units can or can not be physically separate, i.e., can be located in one place, or can be distributed over multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purposes of the embodiments.

[0118] Those skilled in the art can understand that all or some of the steps in the method disclosed above, the functional modules / units in the system and the device can be implemented as software, firmware, hardware and appropriate combinations thereof.

[0119] The terms "first", "second", "third", "fourth" and the like in the description of the application and in the claims of the foregoing drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so construed can be interchanged, such that the embodiments of the application described herein can be carried out in other than the order discussed herein without departing from the scope of the application. Further, the terms "comprise" and "comprising" and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product or apparatus that comprises a list of steps or units does not necessarily comprise only those steps or units but can include other not expressly listed steps or units.

[0120] It should be understood that in the present application, "at least one" means one or more, and "multiple" means two or more. "And / or" is used to describe the relationship between the associated objects, which means that there can be three relationships, for example, "A and / or B" can mean that there are three cases: only A, only B, and A and B at the same time, where A and B can be singular or plural. The character " / " generally represents that the associated objects before and after are in an "or" relationship. "At least one of the following" or the like means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can mean a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0121] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented by other manners. For example, the apparatus embodiments described above are merely illustrative, for example, the division of the above units is merely a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some interfaces, apparatuses or units, and can be electrical, mechanical or other forms.

[0122] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0123] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0124] If the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that makes a contribution or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method of each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0125] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, but this does not limit the scope of the rights of the embodiments of the present application. Any modification, equivalent replacement and improvement made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the rights of the embodiments of the present application.

Claims

1. A shoreline identification method based on data augmentation, characterized in that, The method includes the following steps: Obtain remote sensing image datasets; The remote sensing image dataset is subjected to cloud data augmentation and hybrid augmentation processing based on the isolation triggering algorithm to obtain the training dataset. The training dataset is input into a pre-built neural network model for training to obtain a shoreline recognition model; Acquire remote sensing image data to be identified, input the remote sensing image data to be identified into the shoreline identification model, and obtain the shoreline identification result; The process of performing cloud data augmentation and hybrid augmentation on the remote sensing image dataset according to the isolation triggering algorithm to obtain the training dataset includes the following steps: Cloud data and batch data are obtained by filtering from the remote sensing image dataset; The remote sensing image dataset is subjected to random augmentation strategy triggering processing to obtain the first data to be processed and the second data to be processed. The cloud data is used to perform cloud data enhancement processing on the first data to be processed to obtain the first enhanced data. The second data to be processed is subjected to hybrid enhancement processing based on the batch data to obtain the second enhanced data; The first augmented data and the second augmented data are determined as the training dataset; The step of performing cloud data enhancement processing on the first data to be processed based on the cloud data to obtain the first enhanced data includes the following steps: The cloud data is preprocessed to obtain normalized data and a transparent mask layer; The normalized data is randomly rotated to obtain morphologically enhanced data; By combining the transparent mask layer with the morphological enhancement data, the first data to be processed is subjected to random cloud layer coverage processing to obtain the first enhanced data; The step of performing hybrid enhancement processing on the second data to be processed based on the batch data to obtain the second enhanced data includes the following steps: The batch data is subjected to random region segmentation to obtain segmented images; The segmented images are sequentially overlaid onto the second data to be processed to obtain the second enhanced data; Before inputting the training dataset into the pre-built neural network model for training to obtain the shoreline recognition model, the method further includes pre-building the neural network model, specifically comprising the following steps: The encoder-decoder backbone network is constructed based on the inverted residual structure and the linear bottleneck layer. The encoder-decoder backbone network is connected by a skip link module to obtain the neural network model.

2. The method according to claim 1, characterized in that, The step of inputting the training dataset into a pre-built neural network model for training to obtain a shoreline recognition model includes the following steps: The training dataset is subjected to cross-validation partitioning to obtain training data combinations; The training data is combined and input into the neural network model for training and verification to obtain the shoreline recognition model.

3. The method according to claim 2, characterized in that, The step of combining the training data and inputting it into the neural network model for training and verification to obtain the shoreline recognition model includes the following steps: The training data is combined and input into the neural network model to obtain the prediction result; The prediction results are processed using the dice loss function to obtain the training loss value. The parameters of the neural network model are updated using the learning rate decay strategy combined with the training loss value to obtain the shoreline recognition model.

4. A shoreline identification device based on data augmentation, characterized in that, The device includes: The first module is used to acquire remote sensing image datasets; The second module is used to perform cloud data augmentation and hybrid augmentation processing on the remote sensing image dataset according to the isolation triggering algorithm to obtain a training dataset. The third module is used to input the training dataset into a pre-built neural network model for training to obtain a shoreline recognition model. The fourth module is used to acquire remote sensing image data to be identified, input the remote sensing image data to be identified into the shoreline identification model, and obtain the shoreline identification result; The second module is used to perform cloud data augmentation and hybrid augmentation processing on the remote sensing image dataset according to the isolation triggering algorithm to obtain a training dataset, including: Cloud data and batch data are obtained by filtering from the remote sensing image dataset; The remote sensing image dataset is subjected to random augmentation strategy triggering processing to obtain the first data to be processed and the second data to be processed. The cloud data is used to perform cloud data enhancement processing on the first data to be processed to obtain the first enhanced data. The second data to be processed is subjected to hybrid enhancement processing based on the batch data to obtain the second enhanced data; The first augmented data and the second augmented data are determined as the training dataset; The step of performing cloud data enhancement processing on the first data to be processed based on the cloud data to obtain the first enhanced data includes: The cloud data is preprocessed to obtain normalized data and a transparent mask layer; The normalized data is randomly rotated to obtain morphologically enhanced data; By combining the transparent mask layer with the morphological enhancement data, the first data to be processed is subjected to random cloud layer coverage processing to obtain the first enhanced data; The step of performing hybrid enhancement processing on the second data to be processed based on the batch data to obtain the second enhanced data includes: The batch data is subjected to random region segmentation to obtain segmented images; The segmented images are sequentially overlaid onto the second data to be processed to obtain the second enhanced data; In the third module, before inputting the training dataset into a pre-built neural network model for training to obtain the shoreline recognition model, the device further includes pre-building the neural network model, specifically including: The encoder-decoder backbone network is constructed based on the inverted residual structure and the linear bottleneck layer. The encoder-decoder backbone network is connected by a skip link module to obtain the neural network model.

5. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 3.

Images