Remote sensing image segmentation method based on global feature enhancement and fourier detail adjustment

Abstract

The application discloses a remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment, and belongs to the technical field of remote sensing image processing. The method comprises the following steps: building an image segmentation model comprising a wavelet-Mamba global feature enhancement module, a fast Fourier detail adjustment unit and a decoding and segmentation prediction module; performing remote sensing image segmentation training on the built image segmentation model; and performing image segmentation on a target remote sensing image by using the trained image segmentation model. By means of the wavelet-Mamba global feature enhancement module and the fast Fourier detail adjustment unit, the expression capability of ground object structure, texture and edge information in the remote sensing image can be effectively improved, and high-precision segmentation of small targets and fuzzy boundaries in a complex scene can be realized. The application is especially suitable for accurate identification of targets such as buildings, roads and water bodies in high-resolution remote sensing images, and has high practical value and popularization prospect.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment. Background Technology

[0002] With the rapid development of remote sensing technology and the widespread adoption of high-resolution imaging systems, remote sensing image segmentation, as a crucial step in extracting ground feature information, plays a vital role in fields such as urban planning, land use monitoring, disaster assessment, and agricultural monitoring. The goal of remote sensing image segmentation is to divide pixels in an image into semantically meaningful regions, thereby enabling the automatic identification and extraction of ground features such as buildings, water bodies, roads, and farmland. However, due to the characteristics of remote sensing images—high resolution, complex content, blurred feature boundaries, and dramatic scale variations—traditional image segmentation methods face significant challenges in this field.

[0003] Currently, most mainstream remote sensing image segmentation methods are based on deep convolutional neural networks (CNNs) or Transformer architectures, leveraging their powerful representation capabilities to perform semantic modeling and pixel classification. However, while CNNs excel at extracting local spatial features, they still have shortcomings in capturing large-scale structures, long-distance dependencies, and the continuity of ground feature boundaries in remote sensing images. While Transformers possess global modeling capabilities, their high computational cost and weak ability to depict spatial details limit their practical application in high-resolution remote sensing image segmentation.

[0004] Furthermore, existing methods are prone to problems such as blurred boundaries, missed detection of small targets, and lack of texture information when dealing with fine-grained features in remote sensing images (such as narrow roads, rivers, canals, and rooftops). This is especially true in areas with complex scenes and small inter-class differences, where segmentation accuracy is difficult to improve further. Therefore, how to balance global semantic modeling and local detail perception in remote sensing images has become a key issue that current remote sensing image segmentation technology urgently needs to overcome. Summary of the Invention

[0005] In view of this, the present invention provides a remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment, which aims to integrate multi-level information enhancement mechanisms in the frequency domain and spatial domain. Through the wavelet-Mamba global feature enhancement module and the fast Fourier detail adjustment unit, the method can effectively improve the ability to express the structure, texture and edge information of ground objects in remote sensing images, and achieve high-precision segmentation of small targets and blurred boundaries in complex scenes.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, embodiments of the present invention provide a remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment, the method mainly including the following steps:

[0008] S1. Construct an image segmentation model that includes a wavelet-Mamba global feature enhancement module, a fast Fourier detail adjustment unit, and a decoding and segmentation prediction module;

[0009] S2. Perform remote sensing image segmentation training on the constructed image segmentation model;

[0010] S3. Use the trained image segmentation model to perform image segmentation on the target remote sensing image.

[0011] Furthermore, in the wavelet-Mamba global feature enhancement module, the input features are decomposed into multiple frequencies using discrete wavelet transform to model the image's structure, edges, and texture information. A channel-level Mamba mechanism is introduced to achieve global dependency modeling. The specific process includes:

[0012] The input features are normalized to obtain normalized features.

[0013] Normalized features are obtained using two-dimensional discrete wavelet transform. Decomposed into four frequency domain sub-bands, the expression is:

[0014]

[0015] in, It is a low-frequency-low-frequency sub-band, containing overall structural and texture information; The low-frequency to high-frequency sub-bands characterize the horizontal edges; The high-frequency-low-frequency sub-band represents the vertical edge; High-frequency sub-bands represent diagonal edges; DWT represents two-dimensional discrete wavelet transform.

[0016] The low-frequency-low-frequency subband is enhanced by applying convolutional layers, activation functions, convolutional layers, activation functions, normalization layers, Mamba modules, and convolutional layers; the low-frequency-high-frequency subband, high-frequency-low-frequency subband, and high-frequency-high-frequency subband are enhanced only by convolutional layers.

[0017] The four subbands are merged and restored to the enhanced features using a two-dimensional discrete wavelet inverse transform.

[0018] Enhanced features With normalization characteristics The output features are obtained by fusing through residual connections.

[0019] Furthermore, the Mamba module is used to perform global dependency modeling on the feature map along the channel dimension. Key operations include:

[0020] Normalization: Used to stabilize training and prevent gradient vanishing or exploding;

[0021] State-space modeling: Enhancing channel information flow using state-space equations;

[0022] Linear mapping and gating mechanisms: Enhance selective control over features, making inter-class differences in remote sensing images more apparent.

[0023] Furthermore, in the image segmentation model, multi-level wavelet-Mamba global feature enhancement modules are stacked, and after multiple downsampling and enhancement processes, deep semantic features are extracted.

[0024] Furthermore, in the Fast Fourier Transform detail adjustment unit, a frequency domain amplitude-phase dual-channel enhancement strategy is adopted, introducing dynamic convolution and dilated convolution to enhance the structure and edges separately, and feature reconstruction is achieved through frequency-space interaction modeling. The specific process includes:

[0025] Deep semantic features As input, a two-dimensional fast Fourier transform is applied to map it to the frequency domain to obtain the amplitude spectrum and phase spectrum;

[0026] The amplitude spectrum is enhanced by applying full-dimensional dynamic convolution, activation functions, and convolutional layers, while the phase spectrum is enhanced by applying convolutional layers, activation functions, and dilated convolutional layers.

[0027] The enhanced amplitude spectrum and phase spectrum are fused, and then the inverse fast Fourier transform is applied to restore them to the spatial domain to obtain the enhanced detail feature map:

[0028] Applying a depthwise separable convolutional layer to the enhanced detail feature map yields detail refinement features:

[0029] Integrating detailed adjustment features with deep semantic features By performing addition and fusion, the fusion characteristics are obtained. Fusion features The features are obtained through two branches, with the upper branch containing linear layers, convolutional layers, and a sigmoid activation function. The lower branch obtains features through a linear layer and the ReLU activation function. The two branches are then merged, and the output features with frequency domain detail adjustment are obtained through a linear layer.

[0030] Furthermore, in the decoding and segmentation prediction module, the output features are adjusted based on frequency domain details. Upsampling is performed through a series of transposed convolutions or bilinear interpolation and convolutions to restore the spatial resolution, and semantic information is restored through a decoder to generate the final remote sensing image segmentation result.

[0031] Furthermore, when training the constructed image segmentation model for remote sensing image segmentation, a joint loss function is used for training optimization. The joint loss function includes cross-entropy loss and boundary-aware loss, and its expression is:

[0032]

[0033] In the formula, λ1 and λ2 represent the combined loss; λ1 and λ2 represent the weighting coefficients. Represents cross-entropy loss; This represents boundary-aware loss; This represents the true label of the (i,j)th pixel in class c; This represents the corresponding predicted probability value; H and W represent the height and width of the image, respectively, and C represents the number of channels; Represents a binary edge map; This represents the predicted boundary map.

[0034] Secondly, the present invention also provides an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the above-described remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment.

[0035] Compared with the prior art, the present invention has at least the following beneficial effects:

[0036] 1) This invention provides a remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment. By applying wavelet-Mamba global feature enhancement module and fast Fourier detail adjustment unit, it can effectively improve the ability to express the structure, texture and edge information of ground objects in remote sensing images, and achieve high-precision segmentation of small targets and blurred boundaries in complex scenes.

[0037] 2) The wavelet-Mamba global feature enhancement module proposed in this invention combines discrete wavelet transform to perform multi-frequency domain decomposition of features, which can model the structure, edge and texture information of the image respectively, and introduces a channel-level Mamba mechanism to realize global dependency modeling, which significantly improves the ability to express complex land cover morphology and spatial hierarchical relationship in remote sensing images, and effectively makes up for the shortcomings of traditional convolutional structure in long-distance dependency and detailed feature extraction.

[0038] 3) The Fast Fourier Detail Adjustment Unit constructed in this invention adopts a frequency domain amplitude-phase dual-channel enhancement strategy, introduces dynamic convolution and dilated convolution to enhance the structure and edges respectively, and realizes feature reconstruction through frequency-space interaction modeling, thereby effectively improving the segmentation model's ability to identify small targets and blurred edges, and enhancing the fine-grained segmentation accuracy of key features such as buildings, roads, and water bodies in remote sensing images.

[0039] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0040] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0043] Figure 1 This is a schematic diagram of the remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment provided in an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram illustrating the working principle of the feature extraction and wavelet Mamba global enhancement module provided in the embodiments of the present invention.

[0045] Figure 3 This is a schematic diagram illustrating the working principle of the Fast Fourier Transform detail adjustment unit provided in an embodiment of the present invention.

[0046] Figure 4 A schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0048] In the description of this invention, it should be noted that some processes described in this application specification and drawings include multiple operations that appear in a specific order. However, it should be clearly understood that these operations may be performed in any order or in parallel. Furthermore, various numbers are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0049] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0050] See Figure 1 As shown, this invention provides a remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment. This method mainly includes the following steps:

[0051] S1. Construct an image segmentation model that includes a wavelet-Mamba global feature enhancement module, a fast Fourier detail adjustment unit, and a decoding and segmentation prediction module;

[0052] S2. Train the constructed image segmentation model for remote sensing image segmentation;

[0053] S3. Use the trained image segmentation model to perform image segmentation on the target remote sensing image.

[0054] The following is combined Figures 2-3 The working principle and specific implementation methods of the present invention are described in detail below:

[0055] I. Feature Extraction and Wavelet Mamba Global Feature Enhancement:

[0056] Module structure and workflow as follows Figure 2 As shown, it specifically includes:

[0057] 1.1: Remote Sensing Image Preprocessing and Feature Extraction:

[0058] First, the remote sensing image undergoes a series of preprocessing steps (including noise suppression, geometric correction, and image registration) to obtain image I, which is then input into the neural network. To effectively extract multi-scale spatial features, image I is processed sequentially through two depthwise separable convolutional layers: a 3×3 depthwise separable convolutional layer to capture local spatial structure information, and a 1×1 depthwise separable convolutional layer for channel compression and fusion. The output features of the two convolutional layers are denoted as follows:

[0059]

[0060] Here, DSConv represents a depthwise separable convolutional layer. The two are then element-wise summed and fused to enhance feature representation capabilities:

[0061]

[0062] The fused features are then spatially downsampled (e.g., using max pooling or convolution with a stride of 2) to obtain higher-level semantic representations, resulting in preliminary feature maps.

[0063] 1.2: Wavelet-Mamba Global Feature Enhancement Module:

[0064] To enhance the model's ability to perceive global details such as texture and edges in remote sensing images, a global feature enhancement module based on wavelet transform is introduced, and channel enhancement is performed in conjunction with the Mamba mechanism.

[0065] First, the input features Normalization is performed to obtain features.

[0066] Subsequently, the feature map was processed using the two-dimensional discrete wavelet transform (DWT). It is decomposed into four frequency domain sub-bands. This invention employs the Haar wavelet, which is computationally simple and suitable for high-resolution image processing. Its low-pass filter L and high-pass filter H are defined as follows:

[0067]

[0068] The input feature map is decomposed into four sub-bands through horizontal and vertical filtering operations:

[0069] Low-low frequency, containing overall structure and texture information;

[0070] Low-high frequency (Low-High) characterizes horizontal edges;

[0071] High-low frequency characterizes vertical edges;

[0072] High-high frequency (High-High) characterizes diagonal edges.

[0073] The wavelet transform process can be represented as:

[0074]

[0075] Next, the low-frequency sub-band... The application uses convolutional layers, activation functions, normalization layers, channel Mamba modules (described in detail in Section 1.3 below), and convolutional layers; while low-frequency to high-frequency subbands, high-frequency to low-frequency subbands, and high-frequency to high-frequency subbands are processed only through convolutional layers.

[0076] Then, the four sub-bands are fused and restored to the enhanced feature map using the inverse two-dimensional discrete wavelet transform (Inverse DWT, IDWT).

[0077]

[0078] Finally, the enhanced features are fused with the original normalized features through residual connections to obtain the final output.

[0079] 1.3: Channel-level Mamba module:

[0080] To further enhance semantic modeling capabilities in remote sensing images, a channel-level Mamba module is introduced to perform global dependency modeling of feature maps along the channel dimension. Traditional convolution primarily focuses on local regions, while the Mamba module, as a Transformer-like efficient modeling method, can effectively capture long-range dependencies and is suitable for the identification and segmentation of complex terrain boundaries in remote sensing images.

[0081] Assume the input is a low-frequency subband feature map processed by wavelet. This invention first performs channel-dimensional rearrangement and normalization, and then expresses it through the Mamba module as follows:

[0082]

[0083] This module mainly includes the following key operations:

[0084] Normalization layer: Used to stabilize training and prevent gradient vanishing or exploding;

[0085] State-space modeling: Enhancing channel information flow using state-space equations;

[0086] Linear mapping and gating mechanisms: Enhance selective control over features, making inter-class differences in remote sensing images more apparent.

[0087] The final output features are then passed through a convolutional layer (Conv) and a normalization layer (Norm) to be fused with other subbands, expressed as:

[0088] 1.4: Stacked Multi-Level Wavelet-Mamba Feature Enhancement Module:

[0089] To progressively extract hierarchical semantic information from remote sensing images, this method cascades the aforementioned wavelet-Mamba global feature enhancement modules. Through continuous downsampling, a deep multi-scale feature abstraction capability is achieved.

[0090] Specifically, feature maps The first-level module takes input as input, and its output serves as input for the next-level module. Each module contains one downsampling step and a set of wavelet-Mamba enhancement procedures. This process iterates four times.

[0091]

[0092] in, WaveletMamba(·) represents the complete augmentation module structure. After four rounds of downsampling and augmentation, the final extracted deep semantic features (multi-scale semantic features) are:

[0093]

[0094] This feature will serve as input to the subsequent Fast Fourier Details Adjustment Unit for accurate segmentation of land cover categories in remote sensing images.

[0095] II. Fast Fourier Transform Detail Adjustment Unit:

[0096] To further enhance the representation of edge and structural details of ground features in remote sensing images, a Fast Fourier Transform (FFT) detail adjustment unit based on frequency domain processing is proposed. This unit effectively improves the model's detail perception capability in high-resolution remote sensing image segmentation by combining frequency domain enhancement and spatial reconstruction. The structure and workflow of the FFT detail adjustment unit are as follows: Figure 3 As shown, the details are as follows:

[0097] 2.1: Obtaining frequency domain information using Fast Fourier Transform:

[0098] The deep semantic features output in step one above are used to... As input, a two-dimensional Fast Fourier Transform (FFT) is applied to map it to the frequency domain, yielding the amplitude spectrum and phase spectrum:

[0099]

[0100] In the Fast Fourier Transform (FFT), an image is transformed from the spatial domain (usually row and column pixel coordinates (x, y)) to the frequency domain. The transformed image coordinates are the frequency coordinates (u, v): where u represents the frequency component in the horizontal direction; v represents the frequency component in the vertical direction. These two coordinates represent different frequency "components" in the frequency domain, that is, the periodic changing patterns in the image; FFT2(·) represents the two-dimensional Fast Fourier Transform. Represents the complex coefficients on the frequency coordinate (u,v); Represents Euler's formula, and represents the polar coordinate form of a complex number; It represents frequency domain amplitude information, frequency component intensity, and mainly includes the structure and texture of the image; It represents the phase information in the frequency domain, the positional offset of frequency components, and mainly preserves the edges and positions of the image.

[0101] 2.2: Amplitude Channel Enhancement Path:

[0102] amplitude The branch first enhances its adaptive expressiveness through a set of Omni-Dimensional Dynamic Convolution (ODConv) modules. Compared to ordinary convolution, ODConv can dynamically adjust the convolution kernel parameters in the spatial dimension, channel dimension, and kernel size dimension, making it suitable for processing complex remote sensing image structures. The amplitude processing procedure is as follows:

[0103]

[0104] Here, ReLU represents the ReLU activation function, and Conv represents a convolutional layer. This path enhances image texture and structural information, improving the model's ability to perceive ground features.

[0105] 2.3: Phase Channel Enhancement Path:

[0106] phase The branch is used to preserve image edge and feature location information. This path is designed as follows: first, a standard convolutional layer is passed, followed by an activation function and a dilation convolution (DConv) layer to expand the receptive field, thereby obtaining richer edge contextual relationships. The phase processing is as follows:

[0107]

[0108] Dilated convolution can effectively capture sparse but important boundary features in remote sensing images, such as the edges of roads and water bodies, and the outlines of buildings.

[0109] 2.4: Frequency Domain Fusion and Spatial Domain Reconstruction

[0110] Enhanced amplitude spectrum Phase spectrum After fusion, the inverse fast Fourier transform (IFFT) is applied to restore the spatial domain, yielding the enhanced detail feature map:

[0111]

[0112] Subsequently, to further improve spatial feature consistency and edge continuity, a 3×3 depthwise separable convolutional layer (DWConv) is applied to obtain the final detail adjustment features:

[0113]

[0114] in, This indicates the final detailed adjustment features.

[0115] 2.5: and By performing addition and fusion, the features are obtained. feature The process involves two branches. The upper branch contains a linear layer, a convolutional layer, and a sigmoid activation function. The lower branch is obtained by passing through a linear layer and the ReLU activation function. The two branches are merged, and the output features with frequency domain detail adjustment are obtained through a linear layer.

[0116] This invention effectively solves the problems of difficult identification of small targets and blurred edges in remote sensing images by using amplitude-phase dual-channel enhancement and frequency-spatial domain interactive reconstruction, providing high-precision and fine-grained feature support for subsequent segmentation tasks. This module effectively compensates for the insufficient ability of conventional spatial convolution to characterize local details. Through the modeling and reconstruction of frequency domain information, it significantly improves the segmentation accuracy of complex terrain features in remote sensing images, and is particularly suitable for the identification of fine-grained targets such as buildings, roads, and water systems in high-resolution remote sensing images.

[0117] III. Decoding and Segmentation Prediction Module:

[0118] Output features after frequency domain detail enhancement Subsequently, a lightweight decoder structure is constructed to restore high-level semantic features to spatial resolution and generate the final remote sensing image segmentation result. This module includes sub-processes such as stepwise upsampling, edge-guided fusion, semantic restoration, and supervised optimization, aiming to achieve accurate segmentation of ground objects in remote sensing images.

[0119] 3.1: Feature Upsampling and Edge Information Guidance

[0120] In order to gradually restore spatial details, features First, upsampling operations are performed using a series of transposed convolutions or bilinear interpolation and convolutions to gradually restore the original image resolution. Let the output of the upsampling layer be... Considering that edge information in remote sensing images is crucial for the identification of small targets and complex structures, an auxiliary edge-guided branch is introduced. This branch uses an edge detection module (such as Sobel convolution or Laplacian convolution) to process low-level features, obtains an edge response map, and then concatenates and fuses it with the upsampled features to obtain the final edge response map. This fusion operation helps improve the positioning accuracy of the segmentation boundary, especially showing a significant improvement in areas such as building edges and road intersections.

[0121] 3.2: Semantic Prediction and Multi-Scale Fusion

[0122] Fusion features The data enters the decoder backbone and undergoes semantic information reconstruction through several convolutional layers (such as 3×3 standard convolution + BN + ReLU) to obtain the class prediction probability map for each pixel:

[0123]

[0124] in, This is the predicted output for the final segmentation result.

[0125] 3.3: Constructing a joint loss function for training optimization:

[0126] To fully consider issues such as inter-class imbalance and boundary blurring in remote sensing images, this invention designs a combined loss function for end-to-end optimization, including cross-entropy loss and boundary-aware loss:

[0127] ① Cross-entropy loss:

[0128] For pixel-level classification optimization, it is defined as follows:

[0129]

[0130] in, The true label of the (i,j)th pixel in class c; The corresponding predicted probability value.

[0131] ② Boundary-aware loss:

[0132] Guide the model to focus on feature boundaries and improve edge accuracy. Binary edge maps can be used. With the predicted boundary map Binary cross-entropy (BCE) loss between:

[0133]

[0134] ③ Total loss function:

[0135] Combining the two sub-losses, the total loss function is constructed as follows:

[0136]

[0137] Wherein, λ1 and λ2 are weighting coefficients, balancing the contributions of the main segmentation task and the boundary augmentation task.

[0138] Finally, the trained image segmentation model can be used to efficiently and accurately segment the target remote sensing image.

[0139] 3.4: Segmentation Result Output:

[0140] Finally, during the inference phase, the category corresponding to the highest predicted probability is taken as the final segmentation category label, generating a pixel-level remote sensing ground feature segmentation map.

[0141]

[0142] Where H and W are the height and width of the input image.

[0143] This module achieves high-precision, fine-grained segmentation results for complex features (such as buildings, roads, water bodies, and farmland) in remote sensing images through the combined effects of semantic reconstruction, edge guidance, and loss constraints. Combined with the wavelet-Mamba global feature enhancement module in step one and the Fast Fourier detail adjustment unit (frequency domain enhancement) in step two, it effectively balances global semantic understanding and local detail recognition, significantly improving the overall performance of remote sensing image segmentation.

[0144] As described in the above embodiments, those skilled in the art will understand that this invention proposes a remote sensing image segmentation method based on wavelet-Mamba global feature enhancement and Fourier detail adjustment. This method comprehensively utilizes the advantages of frequency domain analysis and spatial modeling, optimizing feature representation capabilities from multiple scales, channels, and branches, thereby improving the model's accuracy in analyzing and recognizing complex land cover structures in remote sensing images. Specifically, this invention first designs a wavelet-Mamba global enhancement module, combining two-dimensional discrete wavelet transform (DWT) with a lightweight Mamba mechanism to achieve joint modeling of edge, texture, and semantic information in remote sensing images during the feature extraction stage. Different frequency sub-bands obtained through wavelet transform can respectively enhance the contour boundaries and overall structural features of land cover, while the Mamba module introduces long-distance dependency modeling capabilities in the channel dimension, further improving the discriminativeness and robustness of feature representation.

[0145] To further enhance the model's ability to perceive spatial details, this invention proposes a Fast Fourier Detail Adjustment Unit. Based on frequency domain information, this unit performs a Fourier Transform (FFT) on deep semantic features to obtain amplitude and phase spectra. Omni-Dimensional Dynamic Convolution (ODConv) and dilated convolution (DConv) are then used to enhance the amplitude and phase paths, respectively. Finally, an Inverse Fourier Transform (IFFT) is used to restore the enhanced detail map. This process effectively recovers ground object edges, fine-grained targets, and texture structures, compensating for the limitations of spatial convolution in representing small targets.

[0146] In the model decoding stage, this invention introduces a lightweight semantic reconstruction structure and edge-guided branches. Through multi-level upsampling and edge enhancement fusion, it achieves progressive reconstruction of high-level semantic features, further improving the accuracy and boundary coherence of land cover category segmentation. Furthermore, to address the issues of inter-class imbalance and boundary blurring in remote sensing images, this invention also designs a joint optimization strategy combining pixel-level cross-entropy loss and boundary-aware loss, effectively promoting collaborative learning between semantic representation and edge perception in the model.

[0147] In summary, this invention introduces innovative designs in key aspects of remote sensing image segmentation (feature extraction, detail enhancement, and decoding prediction), constructing a unified processing framework that takes into account both spatial structure and frequency domain information. This significantly improves the segmentation accuracy of fine-grained features in remote sensing images, and is particularly suitable for the accurate identification of targets such as buildings, roads, and water bodies in high-resolution remote sensing images. It has high practical value and promising prospects for widespread application.

[0148] Additionally, refer to Figure 4 As shown, this embodiment of the invention also provides an electronic device, which may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10. The processor executes the computer program to implement a remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment in the above method embodiment.

[0149] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units, microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control core of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 and calls data stored in the memory 11 to perform various functions and process data within the electronic device.

[0150] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, electronic devices, or computer program products, etc. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0151] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The words "a" or "an" preceding a component do not exclude the presence of a plurality of such components. This invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer.

[0152] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0153] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment, characterized in that, The method includes the following steps: S1. Construct an image segmentation model that includes a wavelet-Mamba global feature enhancement module, a fast Fourier detail adjustment unit, and a decoding and segmentation prediction module; S2. Perform remote sensing image segmentation training on the constructed image segmentation model; S3. Use the trained image segmentation model to segment the target remote sensing image; In the wavelet-Mamba global feature enhancement module, the input features are decomposed into multiple frequencies using discrete wavelet transform to model the image's structure, edges, and texture information. A channel-level Mamba mechanism is introduced to achieve global dependency modeling. The specific process includes: The input features are normalized to obtain normalized features. Normalized features are obtained using two-dimensional discrete wavelet transform. Decomposed into four frequency domain sub-bands, the expression is: in, It is a low-frequency-low-frequency sub-band, containing overall structural and texture information; The low-frequency to high-frequency sub-bands characterize the horizontal edges; The high-frequency-low-frequency sub-band represents the vertical edge; High-frequency sub-bands represent diagonal edges; DWT represents two-dimensional discrete wavelet transform. The low-frequency-low-frequency subband is enhanced by applying convolutional layers, activation functions, convolutional layers, activation functions, normalization layers, Mamba modules, and convolutional layers; the low-frequency-high-frequency subband, high-frequency-low-frequency subband, and high-frequency-high-frequency subband are enhanced only by convolutional layers. The four subbands are merged and restored to the enhanced features using a two-dimensional discrete wavelet inverse transform. Enhanced features With normalization characteristics The output features are obtained by fusing through residual connections.

2. The remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment according to claim 1, characterized in that, The Mamba module is used to perform global dependency modeling on feature maps along the channel dimension. Key operations include: Normalization: Used to stabilize training and prevent gradient vanishing or exploding; State-space modeling: Enhancing channel information flow using state-space equations; Linear mapping and gating mechanisms: Enhance selective control over features, making inter-class differences in remote sensing images more apparent.

3. The remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment according to claim 1, characterized in that, In the image segmentation model, multiple wavelet-Mamba global feature enhancement modules are stacked, and deep semantic features are extracted after multiple downsampling and enhancement processes.

4. The remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment according to claim 3, characterized in that, In the Fast Fourier Transform detail adjustment unit, a frequency domain amplitude-phase dual-channel enhancement strategy is adopted, introducing dynamic convolution and dilated convolution to enhance the structure and edges respectively, and feature reconstruction is achieved through frequency-space interaction modeling. The specific process includes: Deep semantic features As input, a two-dimensional fast Fourier transform is applied to map it to the frequency domain to obtain the amplitude spectrum and phase spectrum; The amplitude spectrum is enhanced by applying full-dimensional dynamic convolution, activation functions, and convolutional layers, while the phase spectrum is enhanced by applying convolutional layers, activation functions, and dilated convolutional layers. The enhanced amplitude spectrum and phase spectrum are fused, and then the inverse fast Fourier transform is applied to restore them to the spatial domain to obtain the enhanced detail feature map: Applying a depthwise separable convolutional layer to the enhanced detail feature map yields detail refinement features: Integrating detailed adjustment features with deep semantic features By performing addition and fusion, the fusion characteristics are obtained. Fusion features The features are obtained through two branches, with the upper branch containing linear layers, convolutional layers, and a sigmoid activation function. The lower branch obtains features through a linear layer and the ReLU activation function. The two branches are then merged, and the output features with frequency domain detail adjustment are obtained through a linear layer.

5. The remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment according to claim 4, characterized in that, In the decoding and segmentation prediction module, the output features are adjusted for frequency domain details. Upsampling is performed through a series of transposed convolutions or bilinear interpolation and convolutions to restore the spatial resolution, and semantic information is restored through a decoder to generate the final remote sensing image segmentation result.

6. The remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment according to claim 1, characterized in that, When training the constructed image segmentation model for remote sensing image segmentation, a joint loss function is used for training optimization. The joint loss function includes cross-entropy loss and boundary-aware loss, and its expression is: In the formula, λ1 and λ2 represent the combined loss; λ1 and λ2 represent the weighting coefficients. Represents cross-entropy loss; This represents boundary-aware loss; This represents the true label of the (i,j)th pixel in class c; This represents the corresponding predicted probability value; H and W represent the height and width of the image, respectively, and C represents the number of channels; Represents a binary edge map; This represents the predicted boundary map.

7. An electronic device, characterized in that, The system includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the remote sensing image segmentation method based on global feature enhancement and Fourier detail adjustment as described in any one of claims 1-6.

Images