CN-122024051-A - Building change detection method and device based on double-time-phase multi-scale edge driving

CN122024051ACN 122024051 ACN122024051 ACN 122024051ACN-122024051-A

Abstract

The invention discloses a building change detection method and device based on double-time-phase multi-scale edge driving, the method comprises the steps of inputting an image to be detected into a building change detection model, obtaining a building change detection result of the image to be detected output by the building change detection model, wherein the building change detection model comprises an initial feature extractor based on a transform model, an edge perception module EPM and a global and local exploration module GLEM, the EPM is focused on capturing local feature information of the edge of a building so as to enhance extraction of a building monomerized global feature, and a GLEM module performs progressive exploration of the local feature, so that the problem that the building can not be accurately segmented and the change condition can not be accurately identified under a dense building scene is solved, and meanwhile, the problems of discontinuity and incompleteness existing in the process of extracting the change feature are solved.

Inventors

LIU CHUNYANG
MAO BO
LIU CHAO

Assignees

安徽理工大学

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. The building change detection method based on double-time-phase multi-scale edge driving is characterized by comprising the following steps of: Acquiring an image to be detected, wherein the image to be detected comprises a front time-phase remote sensing image of a target area and a rear time-phase remote sensing image of the target area; inputting the image to be detected into a building change detection model, and obtaining a building change detection result of the image to be detected output by the building change detection model; The building change detection model comprises a coding network and a decoding network, wherein the coding network comprises one coding branch used for carrying out feature analysis on a front-time-phase remote sensing image and another coding branch used for carrying out feature analysis on a rear-time-phase remote sensing image, the coding branch sequentially comprises an initial feature extractor based on a transducer model, an edge perception module EPM and a global and local exploration module GLEM, the initial feature extractor of the transducer model is used for extracting feature images of different scales of the remote sensing image, the edge perception module EPM is used for carrying out edge feature enhancement on the feature images of each scale respectively, the global and local exploration module GLEM is used for supplementing local feature information of a building to the feature images of each scale after the edge feature enhancement, and the decoding network is used for carrying out channel splicing after the feature images of different scales are respectively sampled to the same scale and carrying out building change detection by utilizing a pre-measurement head to output a building change detection result.
2. The method for detecting building change based on double-phase multi-scale edge driving according to claim 1, wherein the edge perception module EPM comprises a first branch, a second branch and a third branch which are parallel; the first branch comprises an edge convolution processing layer, a1 multiplied by 1 convolution layer and a space attention layer which are connected in sequence; The system comprises a first branch, a second branch and a residual module, wherein the first branch comprises a main branch and a shortcut branch, the main branch comprises a first convolution layer, a batch normalization layer BN and a first ReLU activation function layer which are sequentially connected, the shortcut branch adds the input of the first convolution layer of the main branch and the output of the first ReLU activation function layer, and the added input is activated through the first ReLU activation function layer and then is output as the output of the residual module; The third branch is realized by adopting a residual shortcut branch which directly adds the input of the edge sensing module EPM to the output of the edge sensing module EPM; The output of the first branch, the second branch and the third branch are added and fused, and the result is sequentially processed through the second convolution layer and the convolution block attention layer and then is used as the output of the edge perception module EPM.
3. The method for detecting the building change based on the double-time-phase multi-scale edge driving according to claim 2, wherein the edge convolution processing layer adopts a plurality of Sobel operators in different directions and at least 1 Laplacian operator together as edge convolution to obtain edge characteristics in different directions of a building, and performs channel fusion on the edge characteristics obtained by the plurality of edge convolution to obtain building contour edge characteristic information with richer semantics.
4. A method for detecting a change in a building based on a double-phase multi-scale edge driving according to claim 3, wherein the process of obtaining edge features by each edge convolution in the edge convolution processing layer is as follows: Wherein, the A convolution layer with a convolution kernel of 1 x1 is shown, An input feature map representing an edge convolution processing layer, Representing a depth convolution of the image with the image, Representing the edge detection operator(s), Represents the scaling parameters of the corresponding edge convolution layer, Represents the bias term for the corresponding edge convolution layer, And the channel broadcast multiplication is represented, and the edge features extracted by the edge detection operators are respectively represented.
5. The method for detecting building changes based on dual-temporal multi-scale edge driving according to claim 1, wherein the global and local exploration module GLEM comprises a fourth branch, a fifth branch and a sixth branch which are parallel, and outputs of the fourth branch, the fifth branch and the sixth branch are added and fused to be used as outputs of the global and local exploration module GLEM; The fourth branch comprises a BN layer and a long-distance feature capturing layer which are sequentially connected, and the long-distance feature capturing layer is used for extracting long-distance feature information of an input feature map of the GLEM module; The fifth branch comprises a spatial attention module and a1 multiplied by 1 convolution layer which are sequentially connected, wherein the output of the 1 multiplied by 1 convolution layer is subjected to element multiplication with the output of the BN layer of the fourth branch to be used as the output of the fifth branch; The sixth branch comprises a channel attention module and a ReLU activation function layer which are sequentially connected, wherein the output of the ReLU activation function layer is multiplied by the output of the BN layer of the fourth branch to be used as the output of the sixth branch.
6. The method for detecting building changes based on double-phase multi-scale edge driving according to claim 5, wherein said long-distance feature capturing layer comprises: The device comprises a third convolution layer and two parallel convolution branches behind the third convolution layer, wherein the two parallel convolution branches are respectively marked as a first convolution branch and a second convolution branch, the first convolution branch comprises m layers of second convolution units which are sequentially connected, the output of each layer of second convolution units is added and fused to be used as the output of the first convolution branch, the second convolution branch comprises m layers of cavity convolution units which are sequentially connected, the output of each layer of cavity convolution units is added and fused to be used as the output of the second convolution branch, the cavity rate of each layer of cavity convolution units is different and is increased layer by layer, and the output of the first convolution branch and the output of the second convolution branch are subjected to feature fusion to be used as the output of a long-distance feature capturing layer.
7. The method for detecting building changes based on double-phase multi-scale edge driving according to claim 6, wherein, The number m of layers of convolution units in the first convolution branch and the second convolution branch is 4; each layer of cavity convolution unit comprises a cavity convolution layer with a 3 multiplied by 3 convolution kernel, a batch normalization layer and a ReLU activation function layer which are connected in sequence; the void ratio of the void convolution units connected in sequence of 4 layers is 1,2,3 and 5 in sequence; The second convolution unit comprises a convolution layer with a convolution kernel of 3 multiplied by 3, a batch normalization layer and an activation function ReLU layer which are connected in sequence, and the third convolution layer adopts a standard convolution kernel of 3 multiplied by 3.
8. The method for detecting the change of the building based on the double-time-phase multi-scale edge driving according to claim 6, wherein the feature fusion comprises the following steps: Adding the output characteristic diagram of the first convolution branch and the output characteristic diagram of the second convolution branch to obtain an addition result; carrying out convolution treatment on the addition result through a layer of convolution layer to obtain an addition convolution result; And multiplying the input feature map of the long-distance feature capturing layer with the addition convolution result and outputting the result to serve as an output result of feature fusion, namely the output of the long-distance feature capturing layer.
9. A building change detection device based on double-phase multi-scale edge drive, comprising: The device comprises an image acquisition unit to be detected, a detection unit and a detection unit, wherein the image acquisition unit to be detected is used for acquiring an image to be detected, and the image to be detected comprises a front time-phase remote sensing image of a target area and a rear time-phase remote sensing image of the target area; The building change detection unit is used for inputting the image to be detected into a building change detection model to obtain a building change detection result of the image to be detected output by the building change detection model, the building change detection model comprises a coding network and a decoding network, the coding network comprises one coding branch used for carrying out characteristic analysis on a front-time-phase remote sensing image and the other coding branch used for carrying out characteristic analysis on a rear-time-phase remote sensing image, the coding branches sequentially comprise an initial characteristic extractor based on a Transformer model, an edge perception module EPM and a global and local exploration module GLEM, the initial characteristic extractor of the Transformer model is used for extracting different-scale characteristic diagrams of the remote sensing image, the edge perception module EPM is used for carrying out edge characteristic enhancement on the characteristic diagrams of each scale respectively, the global and local exploration module GLEM is used for supplementing local characteristic information of the building on the characteristic diagrams of each scale after the edge characteristic enhancement, and the decoding network is used for carrying out channel splicing after the characteristic diagrams of different scales are respectively sampled to be the same, and the building change detection result is output by utilizing a preshaping head.
10. A computer readable storage medium storing executable instructions which when executed by a processor implement the dual phase multiscale edge driven based building change detection method of any one of claims 1 to 8.

Description

Building change detection method and device based on double-time-phase multi-scale edge driving Technical Field The invention relates to the technical field of remote sensing image processing, in particular to a building change detection method and device based on double-time-phase multi-scale edge driving. Background With the continuous growth of socioeconomic performance, modern cities are expanding continuously. Meanwhile, the rapid development of the remote sensing earth observation technology provides more effective technical means for the change detection of the multi-temporal remote sensing building. At present, the building change detection technology is widely applied to the fields of urban management, land resource utilization, disaster assessment and the like. Therefore, the intensive research on the building change detection technology has extremely important practical significance for promoting the urban process. In recent years, significant progress has been made in the remote sensing field due to the introduction of convolutional neural networks (Convolutional Neural Networks, CNNs). For example ,"Hou B, Wang Y, Liu Q. Change Detection Based on Deep Features and Low Rank[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(12): 2418-2422." proposes a change detection technique ."Zhan Y, Fu K, Yan M et al. Change Detection Based on Deep Siamese Convolutional Network for Optical Aerial Images[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(10): 1845-1849." that combines CNNs features with low-rank decomposition, directly extracts features in an image pair by using depth conjoined CNN, introduces a weighted contrast loss function to improve reliability of the features, and detects changes in the image pair by the distance between feature vectors. While these CNN-based methods are more accurate in feature extraction than traditional methods, they tend to ignore global information in changing features. The transducer architecture effectively relieves the limit caused by CNNs due to the inherent global attention mechanism, the first change detection network of the pure transducer architecture is widely applied ."Zhang C, Wang L, Cheng S et al. SwinSUNet: Pure Transformer Network for Remote Sensing Image Change Detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-13." in the field of remote sensing change detection, and the long-term global information can be better extracted in the space-time dimension by utilizing Swim Transformer. However, there are still disadvantages in processing background irrelevant feature information. However, the global feature capture by the transducer alone is not capable of accurately capturing the edge feature information between buildings. The latest research introduces a state space model into the remote sensing field, and the unique scanning mechanism of the state space model effectively enhances the attention capability to image features. The Mamba model of the double-flow parallel omnibearing scanning fully verifies the effectiveness of the state space model in a change detection task. However, although the Mamba-based change detection method has been preliminarily verified in the remote sensing field, the research for realizing high-precision segmentation of the building change detection task in a more complex scene is still insufficient. Disclosure of Invention Aiming at the problems in the prior art, the invention provides a building change detection method and device based on double-time-phase multi-scale edge driving, which effectively improve the building change detection precision under a complex scene. The technical scheme is as follows: In a first aspect, a method for detecting a building change based on dual-phase multi-scale edge driving is provided, including: Acquiring an image to be detected, wherein the image to be detected comprises a front time-phase remote sensing image of a target area and a rear time-phase remote sensing image of the target area; inputting the image to be detected into a building change detection model, and obtaining a building change detection result of the image to be detected output by the building change detection model; The building change detection model comprises a coding network and a decoding network, wherein the coding network comprises one coding branch used for carrying out feature analysis on a front-time-phase remote sensing image and another coding branch used for carrying out feature analysis on a rear-time-phase remote sensing image, the coding branch sequentially comprises an initial feature extractor based on a transducer model, an edge perception module EPM and a global and local exploration module GLEM, the initial feature extractor of the transducer model is used for extracting feature images of different scales of the remote sensing image, the edge perception module EPM is used for carrying out edge feature enhancement on the feature images of each scale respectively