SAR image ocean internal wave automatic extraction method

Abstract

The invention belongs to the technical field of ocean remote sensing and deep learning, and particularly relates to an SAR image ocean internal wave automatic extraction method which comprises the following steps: step 1, collecting SAR image data of a target sea area, and preprocessing the image, including radiation correction and geometric correction; 2, marking internal wave signals in the preprocessed SAR image in a manual marking mode, generating a truth value image used for deep learning model training, matching a truth value with an original image, randomly selecting four fifths of the truth value image as a training set, and randomly selecting the rest one fifths of the truth value image as a test set; by adopting the alpha-balanced cross-entropy loss function and the online data enhancement technology, the precision and efficiency of extracting the internal wave signal from the SAR image are effectively improved, especially under the complex imaging condition, the internal wave signal can be accurately extracted, high-precision geographic positioning of the internal wave signal is achieved by combining geographic information, and the method has the advantages of being high in accuracy and high in accuracy. The method is suitable for ocean research and high-precision environment monitoring, and has a wide application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of ocean remote sensing and deep learning, and specifically is a method for automatically extracting ocean internal waves from SAR images. Background Art

[0002] With the global climate change and the continuous changes in the marine environment, the role of ocean observation in studying and understanding the ocean dynamics process is becoming increasingly important. Internal waves, as a wave phenomenon that is widely present in the ocean, not only have a profound impact on the energy distribution and mixing process of the ocean, but also have important significance in the fields of marine environmental assessment and marine resource development. Although traditional ocean observation methods such as buoys and shipborne instruments can provide local high-precision data, they are limited by the observation range and timeliness and cannot meet the needs of large-scale and long-term observations. For this reason, remote sensing technology, especially synthetic aperture radar (SAR), as an observation tool that can work all day and all night, has become an ideal means of monitoring and studying ocean internal waves because of its unique advantage of not being affected by clouds and weather conditions. The high resolution of SAR images enables it to capture detailed ocean surface fluctuations, especially in complex marine environments, providing rich internal wave information.

[0003] Although SAR technology has many advantages, the existing internal wave extraction methods still have some problems when processing SAR images:

[0004] 1. Traditional methods usually rely on manual or semi-automatic processing procedures, but when faced with a large number of high-resolution images, it is time-consuming and labor-intensive, and it is difficult to meet the needs of large-scale data processing, which has become a major obstacle to the research and application of ocean internal waves;

[0005] 2. Although the high-resolution characteristics of SAR images provide rich detailed information, the internal wave signals in the images are complex and diverse, and are easily interfered by speckle noise, which further increases the difficulty of automated extraction. Traditional methods often find it difficult to ensure the accuracy and consistency of extraction when processing these complex signals.

[0006] Therefore, to address the above problems, a method for automatic extraction of ocean internal waves from SAR images was proposed. Summary of the invention

[0007] 1. Technical issues to be resolved

[0008] In view of the shortcomings of the existing technology, the present invention provides a method for automatically extracting ocean internal waves from SAR images, which solves the problem that the existing technology is time-consuming and labor-intensive when faced with a large number of high-resolution images and is difficult to meet the needs of large-scale data processing, and the problem that traditional methods are difficult to ensure the accuracy and consistency of extraction when processing complex and diverse internal wave signals.

[0009] (II) Technical solution

[0010] To achieve the above object, the present invention provides the following technical solution: a method for automatically extracting ocean internal waves from SAR images, comprising the following steps:

[0011] Step 1: Collect SAR image data of the target sea area and preprocess the image, including radiation correction and geometric correction;

[0012] Step 2: manually label the internal wave signals in the preprocessed SAR images to generate true value images for deep learning model training. At the same time, match the true value with the original image, randomly select four fifths of the pairs as training sets, and the remaining one fifth of the pairs as test sets;

[0013] Step 3: Build and train a deep convolutional neural network model based on the improved U-Net architecture. The deep convolutional neural network model uses an alpha-balanced cross-entropy loss function:

[0014] L αBCE To solve the sample imbalance problem, the expression is as follows:

[0015]

[0016] Where y represents the true value, Represents the model output result, αBCE is alpha-balanced cross-entropy, and α∈(0,1) represents the hyperparameter of balancing the influence of samples in the loss function;

[0017] Step 4: Input the preprocessed SAR images and annotated true value data into the deep convolutional neural network model for training, evaluate the extraction effect of the model through the test set, and select the model with the highest index as the final model;

[0018] Step 5: Using the trained deep convolutional neural network model, the ocean internal wave signal is automatically extracted from the SAR image and a binary image of the internal wave signal is generated;

[0019] Step 6: Convert the extracted internal wave signal into the corresponding longitude and latitude position according to the geographic information in the SAR image.

[0020] Preferably, the preprocessing in step 1 is based on SNAP (Sentinel Application Platform) software, including multi-view processing, and the multi-view processing of SAR images is used to reduce speckle noise.

[0021] Preferably, in step 2, marking the internal wave signal in the pre-processed SAR image by manual annotation means: manually extracting and marking the internal wave crest line features accurately by using a stylus pen in conjunction with Sketchbook software.

[0022] Preferably, in step 3, the online data enhancement technology expands the training data including random rotation, scaling, brightness adjustment, etc. of the image.

[0023] Preferably, in step 4, the precision rate, recall rate and F1 are used to evaluate the extraction effect of the model, and only the internal wave position pixels are considered during the calculation process.

[0024] (III) Beneficial effects

[0025] Compared with the prior art, the present invention provides a method for automatically extracting ocean internal waves from SAR images, which has the following beneficial effects:

[0026] 1. This method for automatically extracting ocean internal waves from SAR images significantly optimizes the extraction process of ocean internal wave signals in SAR images by introducing deep convolutional neural network technology, solving the problems of cumbersome steps, time-consuming and labor-intensive traditional methods.

[0027] 2. The method for automatically extracting ocean internal waves from SAR images effectively improves the accuracy and efficiency of extracting internal wave signals from SAR images by adopting the alpha-balanced cross-entropy loss function and online data enhancement technology. Especially under complex imaging conditions, the present invention can accurately extract internal wave signals and realize high-precision geographic positioning of internal wave signals by combining geographic information. It is suitable for marine research and high-precision environmental monitoring and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0029] Figure 1 It is a schematic diagram of the process of the present invention;

[0030] Figure 2 It is a schematic diagram of the deep convolutional neural network structure in the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Specific examples are given below.

[0033] See also Figure 1 The present invention provides a method for automatically extracting ocean internal waves from SAR images, comprising the following steps:

[0034] Step 1: Collect SAR image data of the target sea area and preprocess the image, including radiation correction and geometric correction;

[0035] Step 2: manually label the internal wave signals in the preprocessed SAR images to generate true value images for deep learning model training. At the same time, match the true value with the original image, randomly select four fifths of the pairs as training sets, and the remaining one fifth of the pairs as test sets;

[0036] Step 3: Build and train a deep convolutional neural network model based on the improved U-Net architecture. The deep convolutional neural network model uses an alpha-balanced cross-entropy loss function:

[0037] L αBCE To solve the sample imbalance problem, the expression is as follows:

[0038]

[0039] Where y represents the true value, Represents the model output result, αBCE is alpha-balanced cross-entropy, and α∈(0,1) represents the hyperparameter of balancing the influence of samples in the loss function;

[0040] Step 4: Input the preprocessed SAR images and annotated true value data into the deep convolutional neural network model for training, evaluate the extraction effect of the model through the test set, and select the model with the highest index as the final model;

[0041] Step 5: Using the trained deep convolutional neural network model, the ocean internal wave signal is automatically extracted from the SAR image and a binary image of the internal wave signal is generated;

[0042] Step 6: Convert the extracted internal wave signal into the corresponding longitude and latitude position according to the geographic information in the SAR image.

[0043] Example 1

[0044] Further Figure 1 As shown, the preprocessing in step 1 is based on SNAP (Sentinel Application Platform) software, including multi-view processing, and the multi-view processing of SAR images is used to reduce speckle noise.

[0045] The above scheme: In this embodiment, the northern part of the South China Sea was selected as the research area, 116 ENVISAT ASAR image data of the area were collected, and the images were batch preprocessed based on SNAP software; the preprocessing steps included radiation correction, geometric correction and multi-look processing, radiation correction and geometric correction are used to eliminate geometric distortion and radiation inconsistency in the image, and multi-look processing is used to reduce speckle noise, thereby improving the visibility of internal wave features and the consistency of the image.

[0046] Example 2

[0047] Further Figure 1 As shown, in step 2, marking the internal wave signal in the pre-processed SAR image by manual annotation means: manually using a stylus pen in conjunction with Sketchbook software to accurately extract and mark the internal wave crest line features.

[0048] The above scheme: After the image preprocessing is completed, the internal wave crest line features in the preprocessed SAR image are manually annotated using a stylus pen combined with Sketchbook software to generate high-precision true value data, and the true value data is paired with the original image, and one tenth of them is selected as a test set, and the remaining pairs are used as training sets. Compared with other methods of marking with a mouse, the stylus pen can better fit the internal wave direction and the true value is more accurate.

[0049] Example 3

[0050] Further Figure 1 As shown, in step 3, the online data enhancement technology expands the training data including random rotation, scaling, brightness adjustment, etc. of the image.

[0051] The above scheme: In order to solve the sample imbalance problem caused by the sparse internal wave signal, the alpha-balanced cross-entropy loss function is used in the model training process. In order to improve the robustness and diversity of the model, the online data enhancement technology is used, including random rotation, scaling, brightness adjustment of the image, etc., to simulate the manifestation of internal waves under different imaging conditions. Multiple trainings are performed, and the precision, recall and F1-score of the model in the test set are calculated. The model with the highest index is selected as the final model. This method accelerates the model convergence by giving more weight to the internal wave recognition results, which is beneficial to improving the accuracy of the model in extracting internal waves. In addition, this method also expands the training data through online data enhancement technology.

[0052] Example 4

[0053] Further Figure 1 As shown in Figure 2, in step 4, precision, recall and F1 are used to evaluate the extraction effect of the model, and only the internal wave position pixels are considered during the calculation process.

[0054] The above scheme: using the trained deep convolutional neural network model, this embodiment automatically extracts the ocean internal wave signal in the SAR image and generates a binary image of the internal wave signal. On this basis, combined with the geographic information of the original SAR image, the extracted internal wave signal is converted into the corresponding longitude and latitude position to ensure the geographic accuracy of the extraction result to meet the needs of high-precision ocean research and application.

[0055] Working principle: First, determine the geographical scope of the target sea area, collect SAR image data of the area, and pre-process the image; then, use the stylus to accurately mark the internal wave crest line features in the pre-processed SAR image to generate a high-precision true value image for model training; then, build and train a deep convolutional neural network model based on the improved U-Net architecture; finally, use the trained model to automatically extract the ocean internal wave signal in the SAR image and generate a binary image of the internal wave signal. According to the geographic information in the original SAR image, the extracted internal wave signal is converted into the corresponding longitude and latitude position to ensure the geographic accuracy of the extraction result.

[0056] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for automatically extracting ocean internal waves from SAR images, characterized in that: The following steps are involved: Step 1: Collect SAR image data of the target sea area and preprocess the image, including radiation correction and geometric correction; Step 2: Label the internal wave signals in the preprocessed SAR images by manual annotation to generate true value images for deep learning model training. At the same time, match the true value with the original image, randomly select four fifths of the pairs as training sets, and the remaining one fifth of the pairs as test sets; Step 3: Build and train a deep convolutional neural network model based on the improved U-Net architecture. The deep convolutional neural network model uses an alpha-balanced cross-entropy loss function as L αBCE To solve the sample imbalance problem, the expression is as follows: Where y represents the true value, Represents the model output result, αBCE is alpha-balanced cross-entropy, and α∈(0,1) represents the hyperparameter of balancing the influence of samples in the loss function; Step 4: Input the preprocessed SAR images and annotated true value data into the deep convolutional neural network model for training, evaluate the extraction effect of the model through the test set, and select the model with the highest index as the final model; Step 5: Using the trained deep convolutional neural network model, the ocean internal wave signal is automatically extracted from the SAR image and a binary image of the internal wave signal is generated; Step 6: Convert the extracted internal wave signal into the corresponding longitude and latitude position according to the geographic information in the SAR image.

2. The method for automatically extracting ocean internal waves from SAR images according to claim 1, characterized in that: The preprocessing in step 1 is based on SNAP (Sentinel Application Platform) software, including multi-view processing. The multi-view processing of SAR images is used to reduce speckle noise.

3. The method for automatically extracting ocean internal waves from SAR images according to claim 1, characterized in that: In the step 2, marking the internal wave signal in the pre-processed SAR image by manual labeling means: manually extracting and marking the internal wave crest line features by using a stylus pen in conjunction with Sketchbook software.

4. The method for automatically extracting ocean internal waves from SAR images according to claim 1, characterized in that: In step 3, the online data enhancement technology expands the training data including random rotation, scaling, brightness adjustment, etc. of the image.

5. The method for automatically extracting ocean internal waves from SAR images according to claim 1, characterized in that: In step 4, the precision, recall and F1 are used to evaluate the extraction effect of the model, and only the internal wave position pixels are considered during the calculation process.

Images