Intelligent recognition and positioning method of oceanic internal wave based on multi-source remote sensing image

Abstract

This invention discloses an intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery, belonging to the field of marine remote sensing technology. The key technical solutions include the following steps: determining the geographical location and time range of the target sea area; collecting multi-temporal high-resolution remote sensing image data from synthetic aperture radar (SAR) and optical satellite imagery; labeling the internal wave features of the preprocessed multi-temporal high-resolution remote sensing image data to generate ground truth images; using a trained deep convolutional neural network model to perform pixel-level analysis on the input remote sensing images to extract ocean internal wave features and generate binarized images of the internal wave features; and converting the extracted binarized images of the internal wave features into corresponding latitude and longitude locations based on the geographical location information of the original remote sensing images. This achieves high-precision geographic positioning of internal waves, is suitable for marine research and high-precision environmental monitoring, and has broad application prospects.

Description

Technical Field

[0001] This invention relates to the field of marine remote sensing technology, and more specifically, to a method for intelligent identification and positioning of marine internal waves based on multi-source remote sensing images. Background Technology

[0002] Internal ocean waves are an important oceanic phenomenon widely present in marginal seas and continental shelves, significantly influencing large-scale tidal energy transfer and small-scale mixing processes. Traditional methods for observing internal waves primarily rely on oceanographic instruments, which, while providing accurate data, are costly, complex to operate, and unable to cover the needs of large-scale and long-term observations. With the development of remote sensing technology, especially the application of synthetic aperture radar and optical satellites, new possibilities have emerged for achieving large-scale, long-term observations of internal ocean waves.

[0003] However, existing methods for extracting internal waves still face the following pressing problems that need to be addressed: Traditional methods for extracting internal waves rely primarily on manual or semi-automatic processing. While these methods offer some accuracy, they are cumbersome, involving extensive image preprocessing, manual feature annotation, and subsequent analysis. This time-consuming and labor-intensive approach makes it difficult to handle large-scale, long-term data processing demands. This inefficiency has become a major bottleneck restricting the research and application of ocean internal waves, particularly when processing large-scale, multi-temporal remote sensing data.

[0004] The complexity of multi-source remote sensing imagery increases the difficulty of automated processing. Different sensors (such as SAR and optical sensors) capture significantly different internal wave characteristics due to differences in their imaging principles. These differences in spatial resolution, illumination conditions, and imaging angles result in diverse and inconsistent manifestations of internal wave characteristics across different images, further increasing the complexity and difficulty of automatically extracting internal wave information. Therefore, how to uniformly and accurately extract consistent internal wave characteristics from multi-source remote sensing imagery in this complex and variable environment has become a crucial technical problem that urgently needs to be solved in the field of marine remote sensing. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method for intelligent identification and positioning of ocean internal waves based on multi-source remote sensing images.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent identification and localization of internal ocean waves based on multi-source remote sensing imagery, comprising the following steps: Determine the geographical location and time range of the target sea area, collect multi-temporal high-resolution remote sensing image data from synthetic aperture radar imagery and optical satellite imagery, and preprocess the multi-temporal high-resolution remote sensing image data; The internal wave characteristics of preprocessed multi-temporal high-resolution remote sensing image data are labeled to generate ground truth images. A deep convolutional neural network model based on an improved U-Net architecture is constructed. The deep convolutional neural network model is trained using ground truth images and combined with online data augmentation techniques. The deep convolutional neural network model is embedded with a feature attention mechanism to enhance the ability to recognize internal wave features. Using a trained deep convolutional neural network model, pixel-level analysis is performed on the input remote sensing image to extract ocean internal wave features, and a binarized image of the internal wave features is generated. Based on the geographical location information of the original remote sensing image, the extracted internal wave features are binarized into the corresponding latitude and longitude locations.

[0007] Preferably, the preprocessing includes radiometric correction, geometric correction, and multi-view processing of the synthetic aperture radar image; Among them, multi-view processing is used to reduce speckle noise and perform radiometric and geometric corrections on optical satellite imagery to improve image consistency and comparability.

[0008] Preferably, Sketchbook software is used to extract and label the internal wave features of the preprocessed multi-temporal high-resolution remote sensing image data.

[0009] Preferably, the deep convolutional neural network model uses the Mathews correlation coefficient as the loss function.

[0010] Preferably, the feature attention mechanism is a compression and activation module.

[0011] Preferably, the online data augmentation technique includes random rotation, scaling, brightness adjustment, and noise injection operations on the training image to simulate the manifestation of internal wave features under different imaging conditions.

[0012] Preferably, the synthetic aperture radar imagery includes ENVISAT ASAR imagery; The optical satellite imagery includes MODIS imagery and Himawari-8AHI imagery.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This invention clearly defines the specific geographical location and time range of the target sea area, and based on this, collects multi-temporal high-resolution remote sensing image data covering the region. This data includes synthetic aperture radar (SAR) images and optical satellite images, taking into account internal wave characteristic information under different imaging conditions. After collection, these multi-temporal image data are preprocessed to eliminate data errors, unify data standards, and construct a deep convolutional neural network model based on an improved U-Net architecture. This architecture can well preserve the spatial details of the images and adapt to the pixel-level recognition requirements of internal waves. Simultaneously, the model incorporates a feature attention mechanism to enhance its ability to capture key internal wave features in the images and avoid interference from background information. During the training phase, the generated ground truth images are used as supervision signals, and online data augmentation techniques are combined to enrich the diversity of training data, allowing the model to adapt to internal wave characteristics under different imaging conditions, thereby improving the model's generalization ability and recognition accuracy. After the model training is complete, new remote sensing images are input into the model. The model performs pixel-by-pixel analysis of the images, distinguishing between internal wave regions and non-internal wave regions, and finally generating a binarized image of internal wave features, clearly presenting the distribution range of internal waves in the image.

[0014] By combining the geographic location information inherent in the original remote sensing images, the marked internal wave regions in the binarized images are mapped to the actual geographic space and converted into corresponding latitude and longitude positions, thereby achieving precise positioning of ocean internal waves and providing specific spatial information support for ocean monitoring, scientific research and other scenarios.

[0015] This invention significantly simplifies the internal wave extraction process by introducing deep learning technology, solving the problems of cumbersome steps and time-consuming processes in traditional methods, and greatly improving the efficiency of processing large-scale, multi-temporal remote sensing data. Addressing the issue of inconsistency in internal wave features in multi-source remote sensing images, this invention employs multi-scale feature fusion and an adaptive weighting mechanism, enabling the model to accurately extract internal wave features in complex and variable environments. Furthermore, by combining the extracted internal wave features with the geographic information of the images, this invention achieves high-precision geographic positioning of internal waves, making it suitable for marine research and high-precision environmental monitoring, and possessing broad application prospects. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing images is provided for embodiments of the present invention. Figure 2 This invention provides a schematic diagram of the structure of a deep convolutional neural network in an intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing images, which is an embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0020] Reference Figures 1-2 As shown.

[0021] The embodiments further illustrate the intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing images proposed in this invention.

[0022] A method for intelligent identification and localization of internal ocean waves based on multi-source remote sensing imagery, comprising the following steps: Determine the geographical location and time range of the target sea area, collect multi-temporal high-resolution remote sensing image data from synthetic aperture radar imagery and optical satellite imagery, and preprocess the multi-temporal high-resolution remote sensing image data; Preprocessing includes radiometric correction, geometric correction, and multi-look processing of synthetic aperture radar images; Among them, multi-view processing is used to reduce speckle noise and perform radiometric and geometric corrections on optical satellite imagery to improve image consistency and comparability.

[0023] Synthetic aperture radar imagery includes ENVISAT ASAR imagery; Optical satellite imagery includes MODIS imagery and Himawari-8AHI imagery.

[0024] First, clarify the geographical location and time range of the target sea area. For example, select a specific latitude and longitude range as the target area and determine a continuous monitoring period of 3 months. Collect multi-temporal high-resolution remote sensing image data in this area, which includes synthetic aperture radar imagery and optical satellite imagery.

[0025] Preprocessing these multi-temporal high-resolution remote sensing images aims to improve their quality and consistency. Synthetic Aperture Radar (SAR) image preprocessing includes radiometric correction, geometric correction, and multi-view processing. Radiometric correction eliminates the sensor's own radiometric errors, ensuring the image's grayscale values ​​accurately reflect the actual scattering characteristics of ground features. Geometric correction corrects geometric distortions caused by terrain undulations and sensor attitude changes during imaging, ensuring the image's pixel positions correspond to actual geographic coordinates. Multi-view processing involves fusing multiple observations of the same area. Specifically, the image is divided into multiple sub-views in the range and azimuth directions, and the data from each sub-view are averaged to reduce the inherent speckle noise of SAR images. For example, after multi-view processing, the speckle noise intensity of ENVISAT ASAR images can be significantly reduced, making subsequent internal wave characteristics easier to identify.

[0026] Preprocessing of optical satellite imagery mainly includes radiometric correction and geometric correction. These two operations serve the same purpose as the corresponding processing of synthetic aperture radar (SAR) imagery: to unify the radiometric standards and geographic coordinate systems of different imagery, thereby improving consistency and comparability. For example, for Himawari-8 AHI and MODIS imagery, radiometric correction ensures that the brightness values ​​of the same ground features are within the same order of magnitude, while geometric correction ensures that the same latitude and longitude point is located at the same pixel position in different imagery.

[0027] Synthetic Aperture Radar (SAR) imagery utilizes ENVISAT ASAR images, which possess all-weather, day-and-night imaging capabilities, making them suitable for acquiring marine data under complex weather conditions. Optical satellite imagery encompasses ENVISAT ASAR, MODIS, and Himawari-8AHI images. MODIS images possess multispectral characteristics, providing rich optical features, while Himawari-8AHI images offer high temporal resolution, enabling high-frequency marine monitoring. The combination of multiple image sources allows for complementary advantages, providing more comprehensive information support for subsequent internal wave analysis.

[0028] The internal wave characteristics of preprocessed multi-temporal high-resolution remote sensing image data are labeled to generate ground truth images. The internal wave characteristics of preprocessed multi-temporal high-resolution remote sensing image data were extracted and labeled using Sketchbook software.

[0029] For the preprocessed multi-temporal high-resolution remote sensing images, the internal wave characteristics of these images are manually labeled to generate ground truth images corresponding to the original images. Internal wave characteristics contain key information about the morphology and distribution of internal waves in the image. For example, in ENVISAT ASAR images, internal waves appear as striped, alternating bright and dark textures. During labeling, all pixels corresponding to these striped areas need to be marked as internal wave regions, while background areas without internal waves are marked as non-target areas, thus clarifying the actual location and extent of internal waves in the image.

[0030] The Sketchbook software was used to extract and mark internal wave features. The crest line is one of the most crucial features of internal waves, representing the location of maximum amplitude during propagation. In remote sensing images, it is typically the boundary between light and dark areas of a striped texture. Using Sketchbook's pixel-level editing capabilities, operators can manually trace the outline of the crest line pixel-by-pixel along the light-dark boundary of the internal wave stripes in the image. For example, in the Himawari-8AHI image, the internal wave crest line appears as a line with a sudden change in brightness in the infrared band. Using the software's fine-touch tools, operators can accurately trace and mark this brightness change line, ensuring that the ground truth image not only includes the area of ​​the internal wave but also records the detailed features of the crest line.

[0031] A deep convolutional neural network model based on the improved U-Net architecture is constructed. The deep convolutional neural network model is trained using ground truth images and combined with online data augmentation techniques. The deep convolutional neural network model is embedded with a feature attention mechanism to enhance the ability to recognize internal wave features. The deep convolutional neural network model uses the Mathews correlation coefficient as the loss function.

[0032] The feature attention mechanism is a compression and activation module.

[0033] Online data augmentation techniques include random rotation, scaling, brightness adjustment, and noise injection of training images to simulate the behavior of internal wave features under different imaging conditions.

[0034] First, a deep convolutional neural network model based on the improved U-Net architecture is built. The U-Net architecture itself has a symmetrical structure of encoder and decoder. The encoder is responsible for extracting multi-scale feature information from remote sensing images, and the decoder maps these features back to the original image size through upsampling operations, thereby achieving pixel-level feature segmentation. The improved architecture can better adapt to the detailed extraction requirements of internal wave features in remote sensing images. For example, when processing internal wave strip textures in ENVISAT ASAR images, the encoder captures features from local textures to overall shape layer by layer, while the decoder can accurately restore the spatial location of these features.

[0035] To enhance the model's ability to identify internal wave features, a feature attention mechanism is embedded in the deep convolutional neural network model, specifically a compression and activation module. The compression and activation module works by first compressing the feature maps output from the convolutional layers. Global average pooling is used to condense the feature information of each channel into a single value, thereby capturing the global dependencies between channels. Then, activation is performed. Fully connected layers learn the importance weights of different channels, and these weights are assigned to the feature maps of the corresponding channels. This allows the compression and activation module to automatically focus on key feature channels related to internal waves. For example, when processing images containing both internal waves and sea clutter, the compression and activation module increases the weight of the channel containing the internal wave crest, suppressing interference from clutter channels, thus more accurately identifying internal wave features.

[0036] During the model training phase, the generated ground truth images are used as supervisory signals, combined with online data augmentation techniques to optimize the training effect. Online data augmentation dynamically transforms the input training images during training, specifically including random rotation, scaling, brightness adjustment, and noise injection. Random rotation can simulate the internal wave morphology under different imaging angles; for example, rotating the image by 30 degrees to correspond to different sensor observation orientations. Scaling can adapt to the feature scale of images with different resolutions; for example, reducing the image to 80% of its original size to match the internal wave performance of low-resolution optical images. Brightness adjustment can simulate the characteristics of optical images under different lighting conditions; for example, increasing image brightness to correspond to the imaging effect on sunny days. Noise injection simulates sensor imaging noise; for example, adding slight Gaussian noise to the image to approximate the speckle characteristics of actual SAR images. These operations significantly enrich the diversity of training data, allowing the model to adapt to the internal wave characteristics under different imaging conditions.

[0037] The model uses the Mathews correlation coefficient as the loss function. The Mathews correlation coefficient can comprehensively consider four types of prediction results: true positive, true negative, false positive, and false negative. It effectively balances the problem of uneven distribution of positive and negative samples. For example, when the proportion of internal wave areas in marine images is small, this loss function will not mask the prediction error of internal wave areas due to a large number of correct predictions of background areas. It can more accurately reflect the model's recognition performance of small-area internal wave features, thereby guiding the model to converge to the ideal recognition effect more efficiently.

[0038] Using a trained deep convolutional neural network model, pixel-level analysis is performed on the input remote sensing image to extract ocean internal wave features, and a binarized image of the internal wave features is generated. Based on the geographical location information of the original remote sensing image, the extracted internal wave features are binarized into the corresponding latitude and longitude locations.

[0039] The preprocessed remote sensing imagery is input into a trained deep convolutional neural network model. The model performs pixel-level analysis to extract internal wave features from the ocean. Pixel-level analysis refers to the model judging each pixel in the image one by one, distinguishing whether the pixel belongs to an internal wave region or a non-internal wave region. The final result is a binary image of the internal wave features. In this image, pixels representing internal waves are marked with a specific value (e.g., 1), while background pixels (non-internal waves) are marked with another value (e.g., 0). When an ENVISAT ASAR image is input, the model identifies striped internal wave texture regions in the image, marking the pixels in these regions as 1, while pixels in other areas of the sea surface are marked as 0, thus clearly distinguishing the extent and shape of internal waves.

[0040] After obtaining the binarized image of the internal wave features, the geographic location information of the original remote sensing image is combined to complete the localization of the internal waves. The original remote sensing image itself carries corresponding geographic coordinate parameters, such as the latitude and longitude range and projection method information corresponding to each pixel in the image. Using this information, the pixels marked as internal waves in the binarized image are mapped one by one to the actual geographic space, thereby converting them into corresponding latitude and longitude locations. If the geographic parameters corresponding to a certain internal wave pixel in the original MODIS image are longitude 115°E and latitude 20°N, then the actual geographic location of the internal wave region corresponding to that pixel in the binarized image is the sea area point corresponding to 115°E and 20°N. In this way, the distribution location of internal waves in the actual ocean can be determined.

[0041] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0042] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence 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 ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent identification and localization of ocean internal waves based on multi-source remote sensing imagery, characterized in that, The method includes the following steps: Determine the geographical location and time range of the target sea area, collect multi-temporal high-resolution remote sensing image data from synthetic aperture radar imagery and optical satellite imagery, and preprocess the multi-temporal high-resolution remote sensing image data; The internal wave characteristics of preprocessed multi-temporal high-resolution remote sensing image data are labeled to generate ground truth images. A deep convolutional neural network model based on an improved U-Net architecture is constructed. The deep convolutional neural network model is trained using ground truth images and combined with online data augmentation techniques. The deep convolutional neural network model is embedded with a feature attention mechanism to enhance the ability to recognize internal wave features. Using a trained deep convolutional neural network model, pixel-level analysis is performed on the input remote sensing image to extract ocean internal wave features, and a binarized image of the internal wave features is generated. Based on the geographical location information of the original remote sensing image, the extracted internal wave features are binarized into the corresponding latitude and longitude locations.

2. The intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery according to claim 1, characterized in that, The preprocessing includes radiometric correction, geometric correction, and multi-view processing of synthetic aperture radar images; Among them, multi-view processing is used to reduce speckle noise and perform radiometric and geometric corrections on optical satellite imagery to improve image consistency and comparability.

3. The intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery according to claim 1, characterized in that, The internal wave characteristics of preprocessed multi-temporal high-resolution remote sensing image data were extracted and labeled using Sketchbook software.

4. The intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery according to claim 1, characterized in that, The deep convolutional neural network model uses the Mathews correlation coefficient as the loss function.

5. The intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery according to claim 1, characterized in that, The feature attention mechanism is a compression and activation module.

6. The intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery according to claim 1, characterized in that, The online data augmentation technology includes random rotation, scaling, brightness adjustment, and noise injection operations on the training images to simulate the manifestation of internal wave features under different imaging conditions.

7. The intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery according to claim 1, characterized in that, The synthetic aperture radar imagery includes ENVISAT ASAR imagery. The optical satellite imagery includes MODIS imagery and Himawari-8AHI imagery.

Images