A deep learning-based ctd section data interpolation reconstruction method

Abstract

This invention proposes a deep learning-based CTD cross-section data interpolation and reconstruction method, comprising: a data preprocessing module for extracting effective cross-section data from raw CTD measurement data and performing quality control and standardization; a model building module that utilizes a deep learning model to automatically learn the spatial structural features of CTD data and the nonlinear relationships between multiple variables, enabling more accurate recovery of complex hydrological feature changes and improving data reconstruction accuracy, especially in highly variable areas such as ocean fronts and strata; it possesses adaptive learning capabilities and is suitable for various environmental backgrounds: the deep model can be trained to adapt to CTD data from different sea areas, seasons, or measurement platforms, exhibiting good generalization ability, without relying on manually set parameters or prior models, significantly reducing human intervention and improving the method's versatility and adaptability; and it effectively handles data gaps and anomalies.

Description

Technical Field

[0001] This invention relates to the field of CTD detectors, and more particularly to a deep learning-based method for interpolating and reconstructing CTD cross-sectional data. Background Technology

[0002] CTD (Cyclic Transmission Device) is an important tool for acquiring vertical profile data of the marine environment and is widely used in marine scientific research, climate monitoring, marine engineering and military fields. By measuring the conductivity, temperature and depth of seawater, key parameters such as seawater density and sound speed can be calculated, which is of great significance for understanding ocean dynamic processes, water mass structure and environmental changes.

[0003] In actual marine surveys and observations, CTD equipment is limited by factors such as cost, deployment frequency, and operating environment. The data acquired is often sparsely and irregularly distributed in space, with a large number of missing measurement points or discontinuous profiles, making it difficult to meet the needs of high-precision numerical simulation, refined environmental assessment, and prediction. To compensate for the spatial discontinuity of the data, interpolation methods are usually used to reconstruct and fill in the CTD profile data. Traditional interpolation methods mainly include linear interpolation, spline interpolation, inverse distance weighted interpolation, and Kriging interpolation. Although these methods are simple to implement and computationally efficient, they often fail to accurately capture complex features such as local jumps, cover structures, and small-scale disturbances when processing highly nonlinear marine profile data with significant multi-scale variations. The physical rationality and spatial continuity of the reconstruction results are difficult to guarantee.

[0004] With the rapid development of artificial intelligence technology, deep learning has demonstrated powerful modeling capabilities in tasks such as image reconstruction, spatiotemporal prediction, and real data imputation. In particular, deep neural networks have shown fitting and generalization capabilities far exceeding those of traditional methods when dealing with large-scale, high-dimensional, and nonlinearly correlated datasets. In recent years, some studies have attempted to introduce deep learning into the field of marine data processing, such as marine temperature field prediction, eddy detection, and flow field reconstruction. However, dedicated deep learning methods for CTD cross-section data interpolation and reconstruction are still immature. Therefore, a deep learning-based CTD cross-section data interpolation and reconstruction method is proposed to solve the above problems. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the defects of the existing technology. The present invention proposes a CTD cross-sectional data interpolation and reconstruction method based on deep learning.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a deep learning-based CTD cross-sectional data interpolation and reconstruction method, comprising: The data preprocessing module is used to extract effective cross-sectional data from the raw CTD measurement data and perform quality control and standardization processing. The model building module uses convolutional neural networks or deep learning models that integrate Transformer structures to perform CTD profile interpolation modeling. The interpolation and reconstruction module applies the model to profiles with missing measurement points in real ocean transect observation data. The model evaluation and visualization module is used to quantify the consistency between the interpolation effect and the actual profile, and to provide visualization output through profile plots and error distribution plots. By introducing deep neural networks and combining the physical structure and spatial correlation characteristics of ocean water bodies, high-precision reconstruction of missing parts of CTD data profiles can be achieved.

[0007] Preferably, the data preprocessing module includes outlier removal, depth layer alignment, physical quantity normalization, and vertical interpolation unified sampling depth operation to spatially match and complete data between different flight segments or stations, forming the input and output tensors required for training and prediction.

[0008] Preferably, the deep learning model includes an input layer, a feature extraction layer, an encoder-decoder module, and an output reconstruction layer.

[0009] Preferably, the input layer consists of partial point values ​​of the observed profile data and their corresponding depth coordinates.

[0010] Preferably, the output reconstruction layer is the interpolated reconstruction value of the corresponding depth profile at the missing measurement point.

[0011] Preferably, the model training in the model building module adopts a supervised learning method, using partial sampled values ​​from the complete profile data as input and the complete values ​​as labels for regression learning.

[0012] In a second aspect, the present invention provides an electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of a deep learning-based CTD cross-sectional data interpolation and reconstruction method.

[0013] Thirdly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of a deep learning-based CTD cross-sectional data interpolation and reconstruction method.

[0014] Compared with existing technologies, the beneficial effects of this invention include: by utilizing a deep learning model to automatically learn the spatial structural features and nonlinear relationships between multiple variables in CTD data, this invention can more accurately recover complex hydrological feature changes and improve data reconstruction accuracy, especially in areas with strong variability such as ocean fronts and strata; it has adaptive learning capabilities and is suitable for various environmental backgrounds: the deep model can be trained to adapt to CTD data from different sea areas, different seasons, or different measurement platforms, has good generalization ability, does not rely on manually set parameters or prior models, significantly reduces human intervention, and improves the versatility and adaptability of the method; and it effectively handles data gaps and anomalies. Attached Figure Description

[0015] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Figure 1 The flowchart illustrates a deep learning-based CTD cross-sectional data interpolation and reconstruction method according to an embodiment of the present invention. Detailed Implementation

[0016] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0017] Example 1, referring to Figure 1 As an embodiment of the present invention, a deep learning-based CTD cross-sectional data interpolation and reconstruction method is provided, comprising: Data preprocessing module: This module is used to extract valid cross-sectional data from raw CTD measurement data, perform quality control and standardization processing, including outlier removal, depth layer alignment, physical quantity normalization, and vertical interpolation to unify sampling depth operations; and perform spatial pairing and completion of data between different flight segments or stations to form the input and output tensors required for training and prediction.

[0018] Model building module: This invention employs an encoder-decoder deep learning model that integrates convolutional neural networks (CNN) and Transformers for interpolation and reconstruction of CTD cross-sectional data.

[0019] Preferably, the encoder can be implemented using a CNN, ResNet, or UNet architecture; more preferably, a Transformer module is introduced between the encoder and decoder to enhance global feature modeling capabilities. The overall model structure adopts a UNet-type encoder-decoder framework, which includes the following layers from the input layer to the output layer: Input layer Convolutional Feature Encoding Layer (CNN Encoder) Transformer Global Feature Modeling Layer Decoder (Reconstruction Layer) Output layer Among them, CNN is used to extract local vertical structural features, and Transformer is used to model the global correlation between different depth layers.

[0020] The input data format and tensor dimensions of the model are designed as follows.

[0021] The input data consists of observations of the CTD profile on a certain cross-section, constructed into a two-dimensional tensor form: in, This represents the vertical depth layer number, with a value range of, for example, 64-512; This represents the number of sampling points in the horizontal direction of the cross-section, with a value range of, for example, 16-128. The number of channels corresponds to the number of physical quantities, preferably temperature (T), salinity (S), density, or sound velocity (optional). In a typical embodiment, =2~4.

[0022] For missing test points, use a mask matrix. For missing measurements, fill in zero or the mean value.

[0023] The model convolutional encoder design consists of multiple layers of two-dimensional convolutional modules. Each layer includes a two-dimensional convolutional layer (Conv2D), batch normalization, and a non-linear activation function (ReLU or LeakyReLU). The preferred kernel size is 3. 3 or 5 5. The encoder layer count is preferably 3 to 5 layers, with the number of channels increasing layer by layer, for example: Local features are extracted by gradually compressing the spatial size through downsampling operations (convolution with a stride of 2 or max pooling).

[0024] The Transformer fusion layer is designed at the end of the encoder to combine the feature tensors output by the CNN: Rearranged into sequence form: in, , The feature dimension is (e.g., 128 or 256). This sequence is input into the Transformer encoder module to capture global dependencies between different depth layers and cross-sectional locations.

[0025] The Transformer encoder includes a multi-head self-attention mechanism, a feedforward fully connected network (FFN), and residual connections with layer normalization (LayerNorm). The key parameter settings ranges are as follows: Number of attention heads: 4~8; Number of Transformer layers: 1~4; Hidden dimensions of feedforward network: 2D~4D.

[0026] The Transformer output is then restored to a two-dimensional feature map for use by the decoder.

[0027] The model decoder and skip connections are designed with a decoder structure symmetrical to the encoder, including upsampling layers (deconvolution or bilinear interpolation) and convolutional reconstruction layers. Skip connections are used to concatenate features from corresponding layers in the encoder with features from the decoder to preserve detailed information about the profile structure.

[0028] The model output layer design adopts 1 1. Convolution maps the features to the target physical quantity dimension, outputting the complete CTD profile interpolation result: Interpolation errors are only calculated at locations where measurements are missing; the remaining locations are used to maintain observation consistency.

[0029] The model is trained using supervised learning, constructing training samples by randomly generating missing test masks on the complete CTD profile. The loss function is: Where MSE is the mean squared error, used to ensure numerical longitude; Gradient Loss is the gradient loss, used to constrain the smoothness of the vertical profile. and This is the weighting coefficient, with a value ranging from 0.1 to 1.0.

[0030] Interpolation and reconstruction module: The trained depth model is applied to profiles with missing measurement points in real ocean transect observation data; users can input any partially missing CTD profile data, and the model will automatically generate its complete reconstruction result, keeping the profile trend of temperature, salinity and physical parameters consistent with the original observation characteristics.

[0031] Model evaluation and visualization module: This module is used to quantify the consistency between interpolation results and actual profiles. It supports comparative analysis with traditional methods such as linear interpolation, spline method, and Kriging. Evaluation metrics include RMSE, MAE, R², and profile difference curves. It also supports the visualization output of profile plots and error distribution maps.

[0032] The method of this invention can be packaged as embedded algorithm software and deployed in AUV, ROV, buoy or edge computing platform, supporting on-the-fly interpolation and rapid cross-section reconstruction, and can also perform batch data processing through server-side interface.

[0033] Deep neural networks are used to model the vertical structure and nonlinear characteristics of ocean profiles to improve interpolation accuracy. It supports complex missing patterns, irregular sampling, and data reconstruction with skip-level measurements, resulting in stronger robustness. The model has good generalization ability and can adapt to data from different regions, different flight segments, and different observation platforms; It is beneficial for constructing high-resolution marine environmental fields and improving the quality of input data for subsequent physical, ecological and engineering models.

[0034] This invention proposes a deep learning-based CTD cross-sectional data interpolation and reconstruction method. Addressing the issue of insufficient reconstruction accuracy in traditional interpolation methods when data is sparse, has many outliers, or the cross-sectional structure is complex, this invention provides a more robust and efficient technical solution with the following advantages: Significantly improves interpolation accuracy and robustness: Compared to traditional spatial interpolation algorithms, such as linear interpolation, inverse distance weighting, and minimum curvature, this invention utilizes a deep learning model to automatically learn the spatial structural features of CTD data and the nonlinear relationships between multiple variables. This enables more accurate recovery of complex hydrological feature changes and improves data reconstruction accuracy, especially in areas with strong variability such as ocean fronts and strata.

[0035] It has adaptive learning capabilities and is suitable for various environments: Depth models can be trained to adapt to CTD data from different sea areas, seasons, or measurement platforms. They have good generalization ability, do not rely on manually set parameters or prior models, significantly reduce human intervention, and improve the versatility and adaptability of the method.

[0036] Effective handling of missing and anomaly data: By introducing a data interpolation loss function and an outlier suppression mechanism, missing data and abnormal observations in cross-sections can be automatically identified and repaired, improving the overall quality of CTD data and providing more reliable data support for subsequent environmental analysis, model assimilation, and numerical simulation.

[0037] It can be integrated into various marine observation platforms for real-time reconstruction. The proposed method can be embedded into the data processing systems of autonomous platforms such as AUVs and Argo buoys, and has the ability to reconstruct data while collecting it. This helps to achieve rapid acquisition and online optimization of marine environmental data, and provides strong support for intelligent observation and the development of marine unmanned systems.

[0038] Promote the improvement of intelligent processing capabilities for marine data: The deep learning-guided interpolation reconstruction approach provides a new paradigm for marine data processing, promotes the integration of traditional physical methods with artificial intelligence, and has strong technological advancement and development potential. It can meet the urgent needs for high-quality data and efficient analysis methods under the strategic background of building a maritime power.

[0039] In summary, this invention not only improves the accuracy and adaptability of CTD cross-sectional data interpolation reconstruction, but also has good engineering feasibility and application prospects. It can be widely used in multiple fields such as marine surveys, air-sea coupling research, marine environmental monitoring and marine numerical model data assimilation, and has significant scientific research and industrial value.

[0040] This embodiment also provides an electronic device, including: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of a deep learning-based CTD cross-sectional data interpolation and reconstruction method.

[0041] This embodiment also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of a deep learning-based CTD cross-sectional data interpolation and reconstruction method.

[0042] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for interpolating and reconstructing CTD cross-sectional data based on deep learning proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0043] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0044] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A deep learning-based CTD cross-sectional data interpolation and reconstruction method, characterized in that, include: The data preprocessing module is used to extract effective cross-sectional data from the raw CTD measurement data and perform quality control and standardization processing. The model building module uses convolutional neural networks or deep learning models that integrate Transformer structures to perform CTD profile interpolation modeling. The interpolation and reconstruction module applies the model to profiles with missing measurement points in real ocean transect observation data. The model evaluation and visualization module is used to quantify the consistency between the interpolation effect and the actual profile, and to provide visualization output through profile plots and error distribution plots. By introducing deep neural networks and combining the physical structure and spatial correlation characteristics of ocean water bodies, high-precision reconstruction of missing parts of CTD data profiles can be achieved.

2. The deep learning-based CTD cross-sectional data interpolation and reconstruction method according to claim 1, characterized in that, The data preprocessing module includes outlier removal, depth layer alignment, physical quantity normalization, and vertical interpolation for unified sampling depth operation. It performs spatial pairing and completion of data from different flight segments or stations to form the input and output tensors required for training and prediction.

3. The deep learning-based CTD cross-sectional data interpolation and reconstruction method according to claim 1, characterized in that, The deep learning model includes an input layer, a feature extraction layer, an encoder-decoder module, and an output reconstruction layer.

4. The deep learning-based CTD cross-sectional data interpolation and reconstruction method according to claim 3, characterized in that, The input layer consists of partial point values ​​of the observed profile data and their corresponding depth coordinates.

5. The CTD cross-sectional data interpolation and reconstruction method based on deep learning according to claim 4, characterized in that, The output reconstruction layer is the interpolated reconstruction value of the corresponding depth profile at the missing measurement points.

6. The CTD cross-sectional data interpolation and reconstruction method based on deep learning according to claim 1, characterized in that, The model building module employs supervised learning for model training, using partial sampled values ​​from the complete profile data as input and the complete values ​​as labels for regression learning.

7. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the deep learning-based CTD cross-sectional data interpolation and reconstruction method according to any one of claims 1 to 6.

8. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the deep learning-based CTD cross-sectional data interpolation and reconstruction method according to any one of claims 1 to 6.

Images