Remote sensing sea fog detection method and system based on deep learning

Abstract

The invention discloses a remote sensing sea fog detection method and system based on deep learning, and the method comprises the following steps: obtaining an original multiband remote sensing image for sea fog detection, and carrying out the standardized data preprocessing; the preprocessed multiband remote sensing image data are input into a preset deep learning network model, a fused feature map is output, and the deep learning network model comprises an encoder and a decoder which are combined in a jump connection mode; and processing the fused feature map through a SoftMax activation function to generate a pixel-level sea fog probability map, and completing image segmentation of the sea fog boundary. Compared with the prior art, the sea fog detection method can overcome the problems of fuzzy sea fog boundary, insufficient multi-scale feature extraction, shallow space information loss and the like of the sea fog detection method of the existing standard deep learning model.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and more specifically, to a remote sensing sea fog detection method and system based on deep learning. Background Technology

[0002] Sea fog is a common hazardous weather phenomenon in the marine environment, significantly impacting maritime navigation safety, port operations, military activities, and marine economic development. Large-scale, near-real-time monitoring of sea fog using satellite remote sensing technology is currently the mainstream approach.

[0003] Traditional remote sensing methods for detecting sea fog often rely on multispectral thresholding or statistical models. However, these methods often suffer from low detection accuracy, weak model generalization ability, and poor spatiotemporal continuity when dealing with complex and variable marine environments.

[0004] With the development of artificial intelligence technology, deep learning, especially convolutional neural networks (CNNs), has demonstrated powerful capabilities in image segmentation. However, when standard deep learning models are directly applied to sea fog detection, challenges remain, such as blurred sea fog boundaries, insufficient multi-scale feature extraction, and loss of shallow spatial information, leading to room for improvement in segmentation accuracy. Therefore, there is an urgent need to develop a deep learning algorithm that can accurately and robustly achieve end-to-end sea fog detection.

[0005] Therefore, in light of the above requirements and the shortcomings of existing technologies, this application proposes a remote sensing method and system for detecting sea fog based on deep learning. Summary of the Invention

[0006] This invention provides a remote sensing sea fog detection method and system based on deep learning, realizing end-to-end detection from raw remote sensing images to sea fog probability maps.

[0007] The primary objective of this invention is to solve the aforementioned technical problems. The technical solution of this invention is as follows: The first aspect of this invention provides a remote sensing method for detecting sea fog based on deep learning, the method comprising the following steps: S1. Acquire raw multi-band remote sensing images for sea fog detection and perform standardized data preprocessing.

[0008] S2. Input the preprocessed multi-band remote sensing image data into a preset deep learning network model and output the fused feature map. The deep learning network model includes an encoder and a decoder combined through skip connections.

[0009] S3. The fused feature map is processed by the SoftMax activation function to generate a pixel-level sea fog probability map, thus completing the image segmentation of the sea fog boundary.

[0010] A second aspect of the present invention provides a deep learning-based remote sensing sea fog detection system, which is used in the aforementioned deep learning-based remote sensing sea fog detection method and includes: a data preprocessing module, a feature extraction module, a feature recovery module, and a classification output module.

[0011] The data preprocessing module acquires the original multi-band remote sensing images for sea fog detection and performs standardized data preprocessing. The feature extraction module is equipped with an encoder with a preset deep learning network model, receives the preprocessed multi-band remote sensing image data, and extracts feature maps which are input to the feature recovery module.

[0012] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention provides a deep learning-based remote sensing method and system for detecting sea fog. The deep learning network model used is the FogNet architecture, which incorporates skip connections in the encoder and decoder to effectively fuse deep semantic information and shallow spatial details, significantly improving the segmentation accuracy of sea fog boundaries. The residual network blocks used as the basic units of the encoder can effectively avoid gradient vanishing. Through the preprocessing of the original multi-band remote sensing images, the deep learning network model has good adaptability to remote sensing data from different sensors and multiple bands. Attached Figure Description

[0013] Figure 1 This is a flowchart of a deep learning-based remote sensing method for detecting sea fog according to the present invention.

[0014] Figure 2 This is a schematic diagram of the overall architecture of a deep learning network model in one embodiment of the present invention.

[0015] Figure 3 This is a schematic diagram of a deep learning-based remote sensing sea fog detection system according to the present invention. Detailed Implementation

[0016] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0017] 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 therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0018] Example 1 like Figure 1As shown, this invention provides a remote sensing method for detecting sea fog based on deep learning. This method includes the following steps: S1. Acquire raw multi-band remote sensing images for sea fog detection and perform standardized data preprocessing; S2. Input the preprocessed multi-band remote sensing image data into a preset deep learning network model and output the fused feature map. The deep learning network model includes an encoder and a decoder combined through skip connections. S3. The fused feature map is processed by the SoftMax activation function to generate a pixel-level sea fog probability map, thus completing the image segmentation of the sea fog boundary.

[0019] In step S1, the standardized data preprocessing process includes: performing atmospheric correction on the remote sensing image, unifying the original multi-band remote sensing image to a preset geographic coordinate system and projection system, unifying different bands of the multi-band remote sensing image to the same spatial resolution through resampling, and normalizing the data of the multi-band remote sensing image.

[0020] In one specific embodiment, the preprocessing process includes: Atmospheric correction: eliminating the influence of atmospheric scattering and absorption on satellite sensor signals to obtain the true surface / sea surface reflectance. Spatial projection transformation: unifying remote sensing data from different sources and time phases to the same geographic coordinate system and projection system. Resolution alignment: for multispectral data, unifying images of different bands to the same spatial resolution through resampling (such as bilinear interpolation or cubic convolution interpolation). Normalization: scaling multiband data to a specific range (such as 0 to 1 or -1 to 1) to accelerate model convergence and improve training stability. Data augmentation: (optionally) expanding the training dataset through operations such as rotation, flipping, and cropping to improve the model's generalization ability. The deep learning network model used in this invention adopts an encoder-decoder architecture as the whole, with the U-Net structure as the backbone, combined with the ResNet module, to achieve effective extraction of multi-scale features and high-precision spatial localization.

[0021] Example 2 Based on the above embodiment 1, combined with Figure 2 This embodiment elaborates on the specific content of the present invention.

[0022] like Figure 2As shown, the architecture of the preset deep learning network model is the FogNet network architecture, which includes an encoder and a decoder. The encoder includes multiple downsampling parts connected in sequence. Each downsampling part includes at least one residual network block. The path of the decoder is symmetrically set with the path of the encoder. The encoder includes multiple upsampling parts connected in sequence. Each upsampling part is upsampled through a deconvolution operation.

[0023] In each upsampling stage, the decoder obtains the feature map output by the residual network block from the corresponding level of the encoder through skip connections, concatenates it with the upsampled feature map of the decoder at this level, and then inputs it into at least one residual network block for processing.

[0024] In one specific embodiment, the feature extraction and restoration process is as follows: Preprocessed data is input into the FogNet network. The encoder extracts multi-scale deep semantic features of the input image layer by layer through cascaded residual network blocks and downsampling operations. The decoder gradually restores the resolution of the feature maps through upsampling operations such as deconvolution (transposed convolution). Simultaneously, skip connections are used to fuse shallow, high-resolution feature maps from corresponding layers in the encoder path with deep, low-resolution feature maps from the decoder path to preserve spatial details of the sea fog and achieve accurate localization.

[0025] The residual network block structure includes a first convolutional layer, a first normalization layer, a first activation function, a second convolutional layer, a second normalization layer, and a second activation function connected in sequence. The original input of the residual network block is input to the first convolutional layer and the second convolutional layer, respectively. The output of the second convolutional layer is added to the output of the second normalization layer by skipping, and then processed by the second activation function to obtain the output of the residual network block.

[0026] In one specific embodiment, both the first convolutional layer and the second convolutional layer are 3x3 convolutional layers, and both the first activation function and the second activation function are ReLU activation functions.

[0027] The specific process of step S3 is as follows: At the end of the decoder path, the deep learning network model outputs a feature map with the same resolution as the input image, but with the number of channels equal to the number of categories. The feature map is then passed through a SoftMax activation function to convert the score of each pixel in different categories into a probability value between [0, 1], thus obtaining the sea fog detection image segmentation result.

[0028] In one specific embodiment, the SoftMax activation function is used in the last layer of the decoder to convert the multi-channel feature map into a pixel-level probability map, where the value of each pixel represents the probability that it belongs to "sea fog" or "background", thereby achieving end-to-end sea fog image segmentation.

[0029] Example 3 like Figure 3 As shown, the present invention also provides a deep learning-based remote sensing sea fog detection system, which is used in the aforementioned deep learning-based remote sensing sea fog detection method and includes: a data preprocessing module, a feature extraction module, a feature recovery module, and a classification output module.

[0030] The data preprocessing module acquires the original multi-band remote sensing images for sea fog detection and performs standardized data preprocessing. The feature extraction module is equipped with an encoder with a preset deep learning network model, receives the preprocessed multi-band remote sensing image data, and extracts feature maps which are input to the feature recovery module.

[0031] like Figure 2 As shown, in one specific implementation, the preprocessed data is used as the input to the FogNet network.

[0032] After data input, the data first passes through a multi-band positional encoding layer. This layer enables the model to perceive the correlation between spectral information of different bands and pixel positions. The encoder of the model consists of four downsampling stages, each mainly consisting of a residual network block (ResNet-18), a normalization layer (batch normalization), and a ReLU activation function. The residual network block of the data input first passes through a 3x3 convolutional layer, a batch normalization layer, and a ReLU activation function; then it passes through a second 3x3 convolutional layer and a batch normalization layer. Finally, the output of the second batch normalization layer is skip-connected to the original input of the block, and then output after passing through a ReLU activation function. This residual structure effectively solves the gradient vanishing problem in deep network training. The encoder path continuously increases the receptive field and extracts abstract, deep semantic features through layer-by-layer downsampling and stacking of residual blocks.

[0033] The decoder path of the model is symmetrical to the encoder path. Upsampling is performed through deconvolutional layers to gradually restore the resolution of the feature maps to the original input size. In each upsampling stage of the decoder, the output feature map is obtained from the residual network block at the same level as the encoder, i.e., the same resolution. This output feature map is then concatenated with the current upsampled feature map of the decoder through skip connections. The concatenated feature map is then fed into the corresponding module of the decoder, which also consists of a residual network block (ResNet-18), a normalization layer, and a ReLU activation function.

[0034] The skip connection combines the shallow, high-resolution texture and boundary information captured by the encoder with the deep, abstract semantic information recovered by the decoder. The boundary information helps in the localization of the boundary, and the semantic information helps in the classification of the boundary, thereby achieving accurate segmentation of the sea fog boundary.

[0035] At the end of the decoder path, the network outputs a feature map with the same resolution as the input image, but with the same number of channels as the number of categories. In a specific embodiment, the number of categories can be two: sea fog and background. The feature map is then passed through a SoftMax activation function, which converts the score of each pixel in different categories into a probability value between [0, 1]. Ultimately, the output is the sea fog detection image segmentation result, achieving end-to-end detection from the original remote sensing image to the sea fog probability map.

[0036] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0037] Alternatively, if the above embodiments of the present invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0038] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. The icons depicting structural positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A remote sensing method for detecting sea fog based on deep learning, characterized in that, Includes the following steps: S1. Acquire raw multi-band remote sensing images for sea fog detection and perform standardized data preprocessing; S2. Input the preprocessed multi-band remote sensing image data into a preset deep learning network model and output the fused feature map. The deep learning network model includes an encoder and a decoder combined through skip connections. S3. The fused feature map is processed by the SoftMax activation function to generate a pixel-level sea fog probability map, thus completing the image segmentation of the sea fog boundary.

2. The remote sensing sea fog detection method based on deep learning according to claim 1, characterized in that, In step S1, the standardized data preprocessing process includes: performing atmospheric correction on the remote sensing image, unifying the original multi-band remote sensing image to a preset geographic coordinate system and projection system, unifying different bands of the multi-band remote sensing image to the same spatial resolution through resampling, and normalizing the data of the multi-band remote sensing image.

3. The remote sensing sea fog detection method based on deep learning according to claim 2, characterized in that, In step S2, the architecture of the preset deep learning network model is the FogNet network architecture, which includes an encoder and a decoder. The encoder includes multiple downsampling parts connected in sequence. Each downsampling part includes at least one residual network block. The path of the decoder is symmetrically set with the path of the encoder. The encoder includes multiple upsampling parts connected in sequence. Each upsampling part is upsampled through a deconvolution operation.

4. The remote sensing sea fog detection method based on deep learning according to claim 3, characterized in that, In each upsampling stage, the decoder obtains the feature map output by the residual network block from the corresponding level of the encoder through skip connections, and concatenates it with the upsampled feature map of the decoder at this level before inputting it into at least one residual network block for processing.

5. The remote sensing sea fog detection method based on deep learning according to claim 4, characterized in that, The structure of the residual network block includes a first convolutional layer, a first normalization layer, a first activation function, a second convolutional layer, a second normalization layer, and a second activation function connected in sequence.

6. The remote sensing sea fog detection method based on deep learning according to claim 5, characterized in that, The original inputs of the residual network block are respectively input to the first convolutional layer and the second convolutional layer. The output of the second convolutional layer is added to the output of the second normalization layer by skipping, and then processed by the second activation function to obtain the output of the residual network block.

7. The remote sensing sea fog detection method based on deep learning according to claim 5, characterized in that, Both the first and second convolutional layers are 3x3 convolutional layers.

8. The remote sensing sea fog detection method based on deep learning according to claim 5, characterized in that, Both the first activation function and the second activation function are ReLU activation functions.

9. A remote sensing method for detecting sea fog based on deep learning according to claim 5, characterized in that, The specific process of step S3 is as follows: At the end of the decoder path, the deep learning network model outputs a feature map with the same resolution as the input image, but with the number of channels equal to the number of categories. The feature map is then passed through a SoftMax activation function to convert the score of each pixel in different categories into a probability value between [0, 1], thus obtaining the sea fog detection image segmentation result.

10. A deep learning-based remote sensing sea fog detection system, wherein the system is used in the deep learning-based remote sensing sea fog detection method according to any one of claims 1-9, characterized in that, It includes: a data preprocessing module, a feature extraction module, a feature recovery module, and a classification output module; The data preprocessing module acquires the original multi-band remote sensing images for sea fog detection and performs standardized data preprocessing. The feature extraction module is equipped with an encoder with a preset deep learning network model, receives the preprocessed multi-band remote sensing image data, and extracts feature maps which are input to the feature recovery module.

Images