A method for segmenting ocean and non-ocean areas in remote sensing images based on pyramid mechanism

Abstract

The invention provides a remote sensing image ocean and non-sea area segmentation method based on a pyramid mechanism. The method comprises the following steps: obtaining a high-resolution remote sensing image, clipping the high-resolution remote sensing image, and drawing a corresponding sea-land segmentation true value map; sequentially performing block division and image rigid transformation on the clipped high-resolution remote sensing image, and dividing a training set and a test set based on the transformation result; establishing a pyramid mechanism fusion neural network, inputting data in the training set into the pyramid mechanism fusion neural network for learning and training to obtain an ocean and non-sea area segmentation model; and inputting data in the test set into the pyramid mechanism fusion neural network to obtain a remote sensing image sea-land segmentation result.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image segmentation, and in particular to a remote sensing image ocean and non-sea area segmentation method based on a pyramid mechanism. Background Art

[0002] my country is a maritime country. The coastline is one of the most important landmarks, which is the boundary between the sea and the land. With the rapid development of the marine economy, the southern coastal areas of my country have gradually become the main areas for people's activities due to their superior geographical location. The coastline will change accordingly due to external factors and human factors. For example, seawater erosion, silt discharge, beach reclamation and sea sand collection will lead to the expansion and contraction of the coastline. In recent years, my country's remote sensing technology has made progress with the rapid development of remote sensing satellites. The advantage of remote sensing technology is that it will not be affected by surface changes, weather differences and geographical environment, so it has been widely used in marine development. High-resolution remote sensing images are conducive to obtaining image information, extracting image features and interpreting images due to their high clarity. Image semantic segmentation plays a key role in remote sensing image applications. In particular, the segmentation of ocean and land can accurately obtain coastline information, which plays an important role in the dynamic changes of the coast and the extraction of important information. In addition, there are interference areas such as land, clouds, and flying objects in marine remote sensing images, which bring great challenges to marine target detection. Therefore, the segmentation of marine and non-sea areas in remote sensing images is of great significance.

[0003] With the continuous development of artificial intelligence technology, machine learning methods have been widely used in various fields and are the research focus and hot issues of image semantic segmentation. Convolutional neural network (CNN) has achieved remarkable results in the field of remote sensing image processing with its great advantages. High-resolution remote sensing images have good imaging quality and high clarity, which is of great significance for detecting coastline changes and macro-change trends of shores and beaches. Extracting coastline information and segmenting interference targets in remote sensing images are of great significance to the development of the ocean periphery. Usually, the coastline is extracted by segmenting ocean and land images, and cloud targets and flying objects are extracted by sea-cloud segmentation. However, traditional methods tend to ignore the semantic relationship between the upper and lower coastlines when segmenting sea and non-sea areas of remote sensing images, and then obtain incorrect feature discrimination mechanisms, which makes it difficult to distinguish seawater with high sediment concentration from other coastal waters, aquaculture ponds, etc. The existing methods for high-resolution remote sensing image segmentation mainly include threshold segmentation method, edge detection method, wavelet transform method, region growing method and machine learning algorithm. Most traditional algorithms perform image segmentation based on the principle of pixel value difference of remote sensing images, but it is easy to make mis-segmentation based on the theory of pixels alone. Traditional machine learning algorithms distinguish between ocean and land in the form of features, but for remote sensing images where the semantic features of the context are not obvious, it is also difficult to obtain more accurate coastline information. Summary of the invention

[0004] The purpose of the present invention is to provide a method for segmenting ocean and non-sea areas of remote sensing images based on a pyramid mechanism. The pyramid mechanism is added to the encoder and decoder of ResNet34 to learn the characteristics of ocean and non-sea areas in multi-scale remote sensing images, solve the problem of blurred boundaries, and improve the accuracy of sea and non-sea segmentation of high-resolution remote sensing images, so as to solve the problems raised in the above background technology.

[0005] The present invention is realized by the following technical scheme: The present invention discloses a method for segmenting ocean and non-ocean areas of remote sensing images based on a pyramid mechanism, and the method comprises the following steps:

[0006] Acquire a high-resolution remote sensing image, crop the high-resolution remote sensing image, and draw a corresponding land-sea segmentation true value map;

[0007] The cropped high-resolution remote sensing images are divided into blocks and image rigid transformations in turn, and the training set and test set are divided based on the transformation results;

[0008] Establish a pyramid mechanism fusion neural network, input the data in the training set into the pyramid mechanism fusion neural network for learning and training to obtain the ocean and non-sea area segmentation model;

[0009] The data in the test set are input into the pyramid mechanism fusion neural network to obtain the sea and land segmentation results of the remote sensing image.

[0010] Optionally, the high-resolution remote sensing image is cropped, and the cropped image contains all information about the area near the coastline and the land.

[0011] Optionally, when drawing the corresponding land-sea segmentation true map, the process includes: based on ArcGIS 10.2 tools, manually drawing the ocean and land areas in the cropped high-resolution remote sensing image, and obtaining a vector file in shp format composed of points, lines and surfaces as the true map.

[0012] Optionally, when the cropped high-resolution remote sensing image is sequentially divided into blocks and subjected to image rigid transformation, the process includes:

[0013] The cropped high-resolution remote sensing image is divided into blocks with a block size of N×N, where N is a natural number not exceeding 256;

[0014] The divided images are flipped up and down, left and right, and rotated at a certain angle to expand the sample size.

[0015] Optionally, when the data in the training set is input into the pyramid mechanism fusion neural network for learning and training to obtain the ocean and non-sea area segmentation probability map, the process includes:

[0016] Set A = {A1, A2, ..., A i}Contains all high-resolution remote sensing image training datasets where d m ,d n Indicates sample A i size;

[0017] The training set A i The first layer of the input pyramid fusion neural network performs residual convolution to obtain the feature map The feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0018] The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F0;

[0019] For the feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0020] The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F1;

[0021] For the feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0022] The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F2;

[0023] The feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0024] The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F3;

[0025] The feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0026] The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F4;

[0027] The feature map Input the pyramid mechanism to fuse the pyramid fusion module of the neural network to obtain features

[0028] The characteristics Feature F4, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0029] The characteristics Feature F3, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0030] The characteristics Feature F2, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0031] The characteristics Feature F1, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0032] The characteristics Feature F0, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0033] For the characteristics Perform convolution of size m×m, upsampling, and 1×1 convolution to obtain the final probability map of ocean and non-sea area segmentation.

[0034] Optionally, four different rates of dilated convolution are performed in the pyramid mechanism module, the convolution size is m×m, and the rates are R1, R2, R3 and R4, respectively, to finally obtain multi-scale features.

[0035] Optionally, the training process of the pyramid fusion module includes:

[0036] Set the pyramid fusion module to perform four different rates of dilated convolutions, with a convolution size of m×m, and rates of R1, R2, R3, and R4, respectively. The features obtained are recorded as A, B, C, and D, respectively.

[0037] Connect feature A and feature B to perform m×m convolution, and perform Softmax function to obtain the features, then multiply them element-by-element with feature A and feature B, and then add them element-by-element to obtain feature E.

[0038] Connect feature C and feature E to perform m×m convolution operation, and perform Softmax function to obtain the features, then multiply them element-by-element with feature C and feature E respectively, and then add them element-by-element to obtain feature F;

[0039] Connect feature D and feature F to perform m×m convolution operation, and perform Softmax function to obtain the features, then multiply them element by element with feature D and feature F, and add them element by element to obtain feature G, which is recorded as

[0040] Optionally, the method further includes:

[0041] The ocean and non-sea area segmentation probability map is compared with the true value map. If the similarity is high, it means that the ocean and non-sea area segmentation probability map is correct and the pyramid mechanism fusion neural network training is completed. Otherwise, the data in the training set is re-input into the pyramid mechanism fusion neural network for learning and training.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The invention provides a method for sea and land segmentation of ocean remote sensing images using a pyramid pooling U-shaped network. The method can achieve a high degree of consistency with manual segmentation by experts in the segmentation of ocean and non-ocean areas of high-resolution remote sensing images through the pyramid U-shaped network. The pyramid pooling structure is embedded in a ResNet network to perform a multi-scale feature fusion operation, thereby improving the segmentation accuracy of high-resolution remote sensing images at different scales. The pyramid fusion structure is added in the transition stage from the encoder to the decoder, thereby learning hierarchical representation in aggregated features and improving the accuracy of extracting ocean and non-ocean target information. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 A flowchart of a method for segmenting ocean and non-sea areas of remote sensing images based on a pyramid mechanism provided by the present invention;

[0046] Figure 2 A diagram of a pyramid mechanism fusion neural network structure provided by an embodiment of the present invention;

[0047] Figure 3 It is a structural schematic diagram of a pyramid fusion module provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme and advantages of the present invention more obvious, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described in the present invention, all other embodiments obtained by those skilled in the art without creative work should fall within the protection scope of the present invention.

[0049] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features well known in the art are not described.

[0050] It should be understood that the present invention can be implemented in different forms and should not be interpreted as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to make the disclosure thorough and complete and to fully convey the scope of the present invention to those skilled in the art.

[0051] The purpose of the terms used herein is only to describe specific embodiments and is not intended to be limiting of the present invention. When used herein, the singular forms "one", "an" and "said / the" are also intended to include plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "consisting of" and / or "comprising", when used in this specification, determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0052] In order to fully understand the present invention, a detailed structure will be proposed in the following description to illustrate the technical solution proposed by the present invention. The optional embodiments of the present invention are described in detail as follows, but in addition to these detailed descriptions, the present invention may also have other implementations.

[0053] The purpose of this invention is to propose a method for segmenting sea and non-sea areas in marine remote sensing images based on a pyramid mechanism fusion network, that is, to add the pyramid mechanism to the encoder and decoder of ResNet34 to learn the characteristics of sea and non-sea areas in multi-scale remote sensing images, solve the boundary fuzzy problem, and improve the accuracy of sea and non-sea segmentation in high-resolution remote sensing images. Figure 1-Figure 3 , which comprises the following steps:

[0054] S1. Acquire a high-resolution remote sensing image, crop the high-resolution remote sensing image, and draw a corresponding land-sea segmentation true value map;

[0055] S2, dividing the cropped high-resolution remote sensing image into blocks and performing image rigid transformation in turn, and dividing the training set and the test set based on the transformation results;

[0056] S3, establish a pyramid mechanism fusion neural network, input the data in the training set into the pyramid mechanism fusion neural network for learning and training to obtain the ocean and non-sea area segmentation model;

[0057] S4. Input the data in the test set into the pyramid mechanism fusion neural network to obtain the sea and land segmentation result of the remote sensing image.

[0058] The high-resolution remote sensing image collected in this embodiment is derived from a false color image of the coastal area within the South China Sea taken by the Gaofen-1 (GF-1) satellite. In step S1, since the high-resolution remote sensing image taken by the satellite has an irregular shape, which increases the difficulty of coastline extraction, the collected remote sensing image is cropped. The cropped image contains all information about the area near the coastline and the land.

[0059] Furthermore, based on ArcGIS 10.2 tools, the ocean and land areas were manually drawn in the cropped high-resolution remote sensing images, and a vector file in shp format consisting of points, lines and surfaces was obtained as the true value map.

[0060] In step S2, since the high-resolution remote sensing image has a high resolution and the image size is too large, the cropped high-resolution remote sensing image is divided into blocks, and the block size is N×N, where N is a natural number not exceeding 256, for example, N=256;

[0061] Deep learning requires a large number of training samples, so the divided images are flipped upside down, left to right, and rotated at a certain angle to expand the sample size;

[0062] Finally, the divided and expanded sample data is divided into training set and test set according to a certain ratio, where the ratio of training set to test set is 4:1.

[0063] In step S3, the present invention further discloses a training method for inputting the data in the training set into a pyramid mechanism fusion neural network for learning and training to obtain a probability map of ocean and non-sea area segmentation, the process of which includes:

[0064] S301, set A = {A1, A2, ..., A i}Contains all high-resolution remote sensing image training datasets where d m ,d n Indicates sample A i size;

[0065] S302, the training set A i The first layer of the input pyramid fusion neural network performs m×m residual convolution and follows a ReLU function with a decay rate of 0.85 to obtain the feature map The feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0066] S303: the feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network, perform four different rate dilated convolutions, the convolution size is m×m, and the rates are R1, R2, R3 and R4 respectively, to obtain the multi-scale feature F0;

[0067] S304: the feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0068] S305: the feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network, perform four different rate dilated convolutions, the convolution size is m×m, and the rates are R1, R2, R3 and R4 respectively, to obtain the multi-scale feature F1;

[0069] S306: the feature map Perform an m×m residual convolution and follow it with a ReLU function with a decay rate of 0.85 to obtain the feature map At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0070] S307: The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network, and perform four different rate dilated convolutions through the pyramid mechanism module. The convolution size is m×m, and the rates are R1, R2, R3 and R4 respectively, to obtain the multi-scale feature F2;

[0071] S308: The feature map Perform an m×m residual convolution and follow it with a ReLU function with a decay rate of 0.85 to obtain the feature map At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0072] S309: the feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network, and perform four different rate dilated convolutions through the pyramid mechanism module. The convolution size is m×m, and the rates are R1, R2, R3 and R4 respectively, to obtain the multi-scale feature F3;

[0073] S310: the feature map Perform m×m residual convolution to obtain the feature map At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map

[0074] S311, the feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network, and perform four different rate dilated convolutions through the pyramid mechanism module. The convolution size is m×m, and the rates are R1, R2, R3 and R4 respectively, to obtain the multi-scale feature F4;

[0075] S312: the feature map Input the pyramid mechanism to fuse the pyramid fusion module of the neural network to obtain features

[0076] S313, the characteristics Feature F4, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0077] S314, the features Feature F3, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0078] S315, the features Feature F2, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0079] S316, the features Feature F1, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0080] S317, the characteristics Feature F0, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features

[0081] S318, the characteristics Perform convolution of size m×m, upsampling, and 1×1 convolution to obtain the final probability map of ocean and non-sea area segmentation.

[0082] Furthermore, the present invention also discloses a training method for the pyramid fusion module, the process of which includes:

[0083] S312-1. Set the pyramid fusion module to perform four different rate dilated convolutions. The convolution size is m×m. The rates are R1, R2, R3, and R4, respectively. The obtained features are recorded as A, B, C, and D, respectively.

[0084] S312-2, connect feature A and feature B to perform m×m convolution, and perform Softmax function to obtain the features, then multiply them element-by-element with feature A and feature B respectively, and then add them element-by-element to obtain feature E;

[0085] S312-3, connect feature C and feature E to perform m×m convolution operation, and perform Softmax function to obtain the features, then multiply them element by element with feature C and feature E respectively, and then add them element by element to obtain feature F;

[0086] S312-4, connect feature D and feature F to perform m×m convolution operation, and perform Softmax function to obtain the features, then multiply them element by element with feature D and feature F, and then add them element by element to obtain feature G, which is recorded as

[0087] Furthermore, in the network training process disclosed in this embodiment, a hybrid loss function consisting of two functions is designed. The first loss function is the Dice loss directly aimed at optimizing the segmentation performance evaluation index, which is defined as:

[0088]

[0089] Among them, p(x i )Voxel x i The predicted probability, and g(x i ) is the corresponding gold standard at the same voxel. X represents the training image, and ε prevents the loss function from being divided by 0.

[0090] The second loss function is Focal loss, which is improved by log loss to solve the problem of imbalanced government samples and is defined as:

[0091]

[0092] in represents the balance factor of focal loss, which is set to 0.2, and γ represents the focusing parameter for smooth adjustment of the weight rate, which is set to 1.

[0093] To sum up, the total loss function is expressed as:

[0094]

[0095] Among them, η1 and η2 represent the weight factors of the two loss functions, which are set to 0.8 and 0.2 respectively.

[0096] Optionally, the method further includes:

[0097] S319, comparing the ocean and non-sea area segmentation probability map with the true value map. If the similarity is high, it means that the ocean and non-sea area segmentation probability map is correct and the pyramid mechanism fusion neural network training is completed. Otherwise, the data in the training set is re-input into the pyramid mechanism fusion neural network for learning and training.

[0098] In step S4, the data in the test set is input into the trained pyramid mechanism fusion neural network to obtain the final ocean and non-sea area segmentation probability map.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for segmenting ocean and non-ocean areas of remote sensing images based on a pyramid mechanism, characterized in that: The method comprises the following steps: Acquire a high-resolution remote sensing image, crop the high-resolution remote sensing image, and draw a corresponding land-sea segmentation true value map; The cropped high-resolution remote sensing images are divided into blocks and image rigid transformations in turn, and the training set and test set are divided based on the transformation results; Establish a pyramid mechanism fusion neural network, input the data in the training set into the pyramid mechanism fusion neural network for learning and training to obtain the ocean and non-sea area segmentation model; Inputting the data in the test set into the pyramid mechanism fusion neural network to obtain the sea and land segmentation result of the remote sensing image; The high-resolution remote sensing image is cropped, and the cropped image contains all information of the area near the coastline and the land; When drawing the corresponding land-sea segmentation true value map, the process includes: manually drawing the ocean and land areas in the cropped high-resolution remote sensing image based on ArcGIS 10.2 tools, and obtaining a vector file in shp format composed of points, lines and surfaces as the true value map; The process of sequentially dividing the cropped high-resolution remote sensing image into blocks and performing image rigid transformation includes: The cropped high-resolution remote sensing image is divided into blocks with a block size of N×N, where N is a natural number not exceeding 256; Flip the divided images upside down and left to right, and rotate them at a certain angle to expand the sample size; When the data in the training set is input into the pyramid mechanism and fused into the neural network for learning and training to obtain the probability map of ocean and non-sea area segmentation, the process includes: Let A = {A1, A2, ..., A i }Contains all high-resolution remote sensing image training datasets i∈{1,2,…,N}, where d m ,d n Indicates sample A i size; The training set A i The first layer of the input pyramid fusion neural network performs residual convolution to obtain the feature map The feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F0; For the feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F1; For the feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F2; The feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F3; The feature map Perform residual convolution to obtain feature maps At the same time, the feature map The input pyramid mechanism is fused with the pooling layer of the neural network for downsampling to obtain the feature map The feature map Input the pyramid mechanism module of the pyramid mechanism fusion neural network to obtain the multi-scale feature F4; The feature map Input the pyramid mechanism to fuse the pyramid fusion module of the neural network to obtain features The characteristics Feature F4, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features The characteristics Feature F3, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features The characteristics Feature F2, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features The characteristics Feature F1, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features The characteristics Feature F0, feature map Perform element-by-element superposition, and perform convolution of size m×m, upsampling, and 1×1 convolution on the superposition results to obtain features For the characteristics Perform convolution of size m×m, upsampling, and 1×1 convolution to obtain the final probability map of ocean and non-sea area segmentation.

2. The method for segmenting ocean and non-ocean areas of remote sensing images based on a pyramid mechanism according to claim 1, characterized in that: In the pyramid mechanism module, four different rates of dilated convolution are performed, the convolution size is m×m, and the rates are R1, R2, R3 and R4 respectively, and finally multi-scale features are obtained.

3. The method for segmenting ocean and non-ocean areas of remote sensing images based on a pyramid mechanism according to claim 2, characterized in that: The training process of the pyramid fusion module includes: Set the pyramid fusion module to perform four different rates of dilated convolutions, with a convolution size of m×m, and rates of R1, R2, R3, and R4, respectively. The features obtained are recorded as A, B, C, and D, respectively. Connect feature A and feature B to perform m×m convolution, and perform Softmax function to obtain the features, then multiply them element-by-element with feature A and feature B, and then add them element-by-element to obtain feature E. Connect feature C and feature E to perform m×m convolution operation, and perform Softmax function to obtain the features, then multiply them element-by-element with feature C and feature E respectively, and then add them element-by-element to obtain feature F; Connect feature D and feature F to perform m×m convolution operation, and perform Softmax function to obtain the features, then multiply them element by element with feature D and feature F, and add them element by element to obtain feature G, which is recorded as 4. The method for segmenting ocean and non-ocean areas of remote sensing images based on a pyramid mechanism according to claim 3, characterized in that: The method further comprises: The ocean and non-sea area segmentation probability map is compared with the true value map. If the similarity is high, it means that the ocean and non-sea area segmentation probability map is correct and the pyramid mechanism fusion neural network training is completed. Otherwise, the data in the training set is re-input into the pyramid mechanism fusion neural network for learning and training.

Images