A method of building damage assessment

Abstract

The application provides a building damage assessment method, which can be applied to the technical field of remote sensing image processing. The method comprises the following steps: performing wavelet transform processing and multi-round adaptive global feature extraction on a pre-disaster image of a target building group to obtain a multi-scale first feature map set; performing mask processing on the multi-scale first feature map set to obtain a first image segmentation result with mask information; performing wavelet transform processing, multi-round adaptive global feature extraction and multi-round same-dimension feature map fusion on the pre-disaster image and a post-disaster image of the target building group in parallel to obtain a multi-scale second feature map set; performing convolution processing and activation processing on the multi-scale second feature map set to obtain a second image segmentation result; and performing image fusion on the first image segmentation result and the second image segmentation result to obtain a damage assessment result of the target building group. The application further provides a building damage assessment device, an electronic device and a storage medium.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to the field of remote sensing image processing technology, and more specifically to a method for assessing building damage. Background Technology

[0002] Buildings, as indispensable spaces for human habitation and work, play a central role in urban and rural development. Rapid and accurate assessment of large-scale buildings is crucial for obtaining disaster information and assisting relevant personnel in formulating post-disaster relief and reconstruction strategies. Since buildings are often located in densely populated areas, real-time damage assessment can predict collapse risks and evaluate potential threats to surrounding residents. Therefore, in-depth research and development of rapid building assessment technologies, especially post-disaster building damage assessment technologies, is particularly important and urgent.

[0003] Existing post-disaster building damage assessment technologies primarily rely on machine learning methods and convolutional neural networks to extract building appearance information for evaluation. These methods perform reasonably well when buildings have clear outlines and significantly rise above their surroundings. However, real-world situations are complex and varied, especially in disaster-stricken areas. Furthermore, building appearance and characteristics are influenced by factors such as location, scene, and lighting. Some buildings are difficult to distinguish due to similar materials and colors to the background, while others are difficult to extract due to irregular size and shape, obstruction, or damage. In addition, existing methods have limited generalization capabilities; obstructions such as trees, other buildings, or debris can interfere with building identification and assessment. Moreover, information processing speed and large-scale data processing capabilities are also bottlenecks, hindering rapid response to the needs of emergency disaster situations. In conclusion, existing building damage assessment technologies generally lack broad applicability, practicality, and high accuracy, failing to meet the demands of large-scale post-disaster building damage assessment. Summary of the Invention

[0004] In view of the above problems, the present invention provides a building damage assessment method that improves generalizability and assessment accuracy.

[0005] According to a first aspect of the present invention, a method for assessing building damage is provided, comprising:

[0006] The first image processing branch of the building damage assessment model is used to perform wavelet transform processing and multi-round adaptive global feature extraction on the pre-disaster images of the target building complex to obtain a multi-scale first feature map set;

[0007] The first image processing branch is used to perform masking on the multi-scale first feature map set to obtain the first image segmentation result with mask information;

[0008] The second image processing branch of the building damage assessment model is used to perform wavelet transform processing, multi-round adaptive global feature extraction, and multi-round same-dimensional feature map fusion on the pre-disaster and post-disaster images of the target building complex in parallel to obtain a multi-scale second feature map set.

[0009] The second image processing branch is used to perform convolution and activation processing on the multi-scale second feature map to obtain the second image segmentation result.

[0010] The damage assessment results of the target building complex are obtained by fusing the first image segmentation result and the second image segmentation result using a building damage assessment model.

[0011] According to an embodiment of the present invention, the above-mentioned wavelet transform processing and multi-round adaptive global feature extraction of the pre-disaster image of the target building complex using the first image processing branch of the building damage assessment model to obtain a multi-scale first feature map set includes:

[0012] The pre-disaster images of the target building complex were normalized and standardized using a building damage assessment model to obtain pre-processed pre-disaster images.

[0013] The multi-scale wavelet fusion module of the first image processing branch is used to perform split wavelet transform processing on the preprocessed pre-disaster image, and the multiple initial wavelet transform feature maps are stitched together and convolutionally reduced to obtain wavelet transform feature maps.

[0014] The first image processing branch is used to divide the wavelet transform feature map into striped regions in the vertical and horizontal directions. Then, the wavelet transform feature map with multiple striped regions is applied to the neighborhood attention mechanism to obtain a wavelet transform feature map with neighborhood information.

[0015] Multiple adaptive cross-attention modules in the first image processing branch are used to perform multiple rounds of adaptive global feature extraction on wavelet transform feature maps with neighborhood information to obtain a multi-scale first feature map set.

[0016] According to an embodiment of the present invention, the multi-scale wavelet fusion module of the first image processing branch performs split wavelet transform processing on the preprocessed pre-disaster image, and performs concatenation and dimensionality reduction processing on the obtained multiple initial wavelet transform feature maps to obtain wavelet transform feature maps including:

[0017] The feature map generation stream of the multi-scale wavelet fusion module is used to perform convolution processing on the preprocessed pre-disaster image, and the convolution processing result is batch normalized to generate the first initial wavelet transform feature map.

[0018] Based on Haar wavelet transform, the image decomposition stream of the multi-scale wavelet fusion module is used to decompose the pre-processed pre-disaster image into high-frequency and low-frequency components, and obtain multiple second initial wavelet transform feature maps.

[0019] The first initial wavelet transform feature map and multiple second initial wavelet transform feature maps are concatenated along the channel dimension using a multi-scale wavelet fusion module, and the resulting feature map is then subjected to convolutional dimensionality reduction to obtain the wavelet transform feature map.

[0020] According to an embodiment of the present invention, the above-mentioned method of dividing the wavelet transform feature map into striped regions in the vertical and horizontal directions using a first image processing branch, and applying a neighborhood attention mechanism to the wavelet transform feature map with multiple striped regions to obtain a wavelet transform feature map with neighborhood information includes:

[0021] The first image processing branch is used to divide the wavelet transform feature map into striped regions in the vertical and horizontal directions to obtain a wavelet transform feature map with multiple striped regions.

[0022] Based on the preset window size information, the first image processing branch is used to divide each stripe region in the wavelet transform feature map with multiple stripe regions into multiple cross-attention window regions.

[0023] Based on the neighborhood attention mechanism, the first image processing branch is used to fuse neighborhood features of all pixels in each cross-attention window region to obtain a wavelet transform feature map with neighborhood information.

[0024] According to an embodiment of the present invention, the above-mentioned neighborhood attention mechanism, which utilizes the first image processing branch to perform neighborhood feature fusion on all pixels in each cross-attention window region to obtain a wavelet transform feature map with neighborhood information, includes:

[0025] The local features of each pixel in each cross-attention window region are extracted using the first image processing branch to generate a local feature matrix for each pixel;

[0026] The first image processing branch is used to perform a learnable linear transformation on the local feature matrix of each pixel to obtain the query vector, key vector and value vector of each pixel;

[0027] The attention weight of each pixel is calculated using the query vector, key vector, and value vector of each pixel, and the relative positional deviation obtained by the convolution of the first image processing branch is embedded into the attention weight.

[0028] The attention weights and local feature matrices of each pixel are weighted and summed using the first image processing branch to obtain a wavelet transform feature map with neighborhood information.

[0029] According to an embodiment of the present invention, the above-mentioned multi-round adaptive global feature extraction of wavelet transform feature maps with neighborhood information using multiple adaptive cross-attention modules of the first image processing branch to obtain a multi-scale first feature map set includes:

[0030] A compression- and incentive-based attention mechanism is introduced into multiple adaptive cross-attention modules, and a hierarchical structure is set for each adaptive cross-attention module;

[0031] The first adaptive cross-attention module is used to perform global average pooling on the wavelet transform feature map with neighborhood information to obtain the first pooling feature map with context information.

[0032] The first adaptive cross-attention module is used to perform fully connected processing and nonlinear activation processing on the wavelet transform feature map with neighborhood information to obtain the first activated feature map with channel attention weights.

[0033] The first attention feature map is obtained by upsampling the first pooling feature map with contextual information and the first activation feature map with channel attention weights using the first adaptive cross-attention module, based on the same channel dimension.

[0034] According to an embodiment of the present invention, the above-mentioned multi-round adaptive global feature extraction of wavelet transform feature maps with neighborhood information using multiple adaptive cross-attention modules of the first image processing branch to obtain a multi-scale first feature map set further includes:

[0035] The first attention feature map is processed by global average pooling using the second adaptive cross-attention module to obtain a second pooled feature map with contextual information.

[0036] The first attention feature map is processed by fully connected processing and nonlinear activation processing using the second adaptive cross-attention module to obtain a second activation feature map with channel attention weights.

[0037] The second adaptive cross-attention module is used to fuse the second pooling feature map with contextual information and the second activation feature map with channel attention weights by upsampling features based on the same channel dimension to obtain the second attention feature map.

[0038] According to an embodiment of the present invention, the above-mentioned multi-round adaptive global feature extraction of wavelet transform feature maps with neighborhood information using multiple adaptive cross-attention modules of the first image processing branch to obtain a multi-scale first feature map set further includes:

[0039] The second attention feature map is processed by global average pooling using the third adaptive cross-attention module to obtain a third pooled feature map with contextual information.

[0040] The second attention feature map is processed by fully connected processing and nonlinear activation processing using the third adaptive cross-attention module to obtain the third activation feature map with channel attention weights.

[0041] According to an embodiment of the present invention, the above-mentioned masking process of the multi-scale first feature map set using the first image processing branch to obtain a first image segmentation result with mask information includes:

[0042] The first image processing branch is used to upsample the feature maps in the multi-scale first feature map set, and the upsampled feature maps are then spliced ​​and fused with the same channel dimension to obtain the spliced ​​and fused first feature map.

[0043] The first feature map after splicing and fusion is activated by the preset activation function of the first image processing branch to obtain the first image segmentation result of the mask information.

[0044] According to an embodiment of the present invention, the second image processing branch of the building damage assessment model performs wavelet transform processing, multi-round adaptive global feature extraction, and multi-round same-dimensional feature map fusion on the pre-disaster and post-disaster images of the target building complex in parallel, resulting in a multi-scale second feature map set including:

[0045] Wavelet transforms are performed on pre-disaster and post-disaster images using different processing units in the second image processing branch. The wavelet transform results of the pre-disaster and post-disaster images are then fused with features of the same dimension to obtain a wavelet transform fused feature map.

[0046] By using different processing units in the second image processing branch, multiple rounds of adaptive global feature extraction are performed on the wavelet transform results of the pre-disaster image and the wavelet transform results of the post-disaster image, resulting in multiple pre-disaster global feature maps and multiple post-disaster global feature maps at different scales.

[0047] The global feature maps before and after the disaster in each round are fused with features of the same dimension to obtain multiple global feature maps of different scales.

[0048] The building damage assessment method provided by this invention, through multi-scale wavelet fusion of images of a target building complex, can learn robust features of buildings at different scales, thereby improving the assessment accuracy and generalization ability of the building damage assessment model. Simultaneously, by applying a multi-round adaptive global attention mechanism to the images of the target building complex, it can efficiently extract key features of the target building complex, significantly enhancing the feature representation capability of the building damage assessment model. Furthermore, the building damage assessment method provided by this invention, through deep image fusion technology, further improves the overall performance and generalization ability of the building damage assessment model, enabling it to perform building damage assessment in complex scenarios and further improve the detailed recognition of target building complexes. Attached Figure Description

[0049] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0050] Figure 1 This is an application scenario diagram of the building damage assessment method according to an embodiment of the present invention;

[0051] Figure 2 This is a flowchart of a building damage assessment method according to an embodiment of the present invention;

[0052] Figure 3 This is a damage assessment architecture diagram of a target building complex according to an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of data processing in the first image processing branch according to an embodiment of the present invention;

[0054] Figure 5 This is a schematic diagram of data processing in the second image processing branch according to an embodiment of the present invention;

[0055] Figure 6 This is a structural diagram of a multi-scale wavelet fusion module according to an embodiment of the present invention;

[0056] Figure 7 This is a structural diagram of the adaptive attention mechanism module according to an embodiment of the present invention;

[0057] Figure 8 This is a structural diagram of a feedforward neural network with a gating mechanism according to an embodiment of the present invention;

[0058] Figure 9 This is a schematic diagram illustrating the training of a building damage assessment model according to an embodiment of the present invention;

[0059] Figure 10This is a schematic diagram of the results of post-disaster building damage assessment in a hybrid xBD and xFBD dataset according to an embodiment of the present invention;

[0060] Figure 11 This is a schematic diagram illustrating the results of building damage assessment in the Ida-BD dataset according to an embodiment of the present invention;

[0061] Figure 12 A schematic diagram illustrating the structure of a building damage assessment device according to an embodiment of the present invention is shown.

[0062] Figure 13 This is a block diagram of an electronic device suitable for implementing a building damage assessment method and a training method for a building damage assessment model according to an embodiment of the present invention. Detailed Implementation

[0063] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0064] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components. All terms used herein (including technical and scientific terms) have the meaning commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner. In the use of expressions such as “at least one of A, B, and C,” the expression should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (e.g., “a system having at least one of A, B, and C” should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0065] Building damage assessment has important practical value, especially post-disaster building damage assessment, which is of great significance for post-disaster relief strategies and post-disaster reconstruction plans.

[0066] However, existing building damage assessment technologies suffer from poor generalization and accuracy, failing to meet the requirements of large-scale building damage assessments, especially in complex post-disaster scenarios. Therefore, it is necessary to provide a building damage assessment technology solution with high accuracy, broad applicability, and practicality.

[0067] The embodiments of the present invention provide a building damage assessment method, which applies the field of remote sensing image processing technology and the scenario of large-scale building damage assessment after a disaster.

[0068] This invention innovatively proposes a building damage assessment method based on the Damage Assessment Fusion Network with Adaptive Attention Mechanism (DAFN-FAAM). The core of the DAFN-FAAM model lies in its introduction of an Adaptive Attention Mechanism (AAM) module. This AAM module utilizes an adaptive attention mechanism to efficiently extract key features of damaged buildings, significantly enhancing the model's feature representation capabilities. While traditional Transformers excel in handling global features, they still have limitations in capturing local features and spatial information. To overcome this deficiency, this invention further proposes a feedforward neural network (GMFFN) with enhanced squeeze and excitation (SE) gating. By combining 3×3 depthwise convolutions with SE attention gating, GMFFN significantly improves the model's overall performance and generalization ability, making it outstanding in handling complex scenes and detail recognition. Furthermore, to learn robust features of buildings at different scales, the DAFN-FAAM model also introduces a multi-scale wavelet fusion (MWF) module. This module further enhances the model's evaluation accuracy and generalization ability by fusing multi-scale information. To verify the effectiveness and generalization ability of the DAFN-FAAM model, this invention underwent rigorous testing on the large-scale disaster datasets xBD and xFBD. Test results show that the DAFN-FAAM model achieves state-of-the-art performance in accuracy and demonstrates significant advantages in handling complex scenes and detail recognition. Furthermore, to further verify the model's generalization ability, this invention also introduced the high-resolution satellite image dataset Ida-BD, focusing on a hurricane in a certain location in 2021. Test results confirm the excellent performance of the DAFN-FAAM model in high-precision damage assessment. The DAFN-FAAM framework not only achieves significant improvements in evaluation accuracy but also demonstrates excellent multi-scale feature fusion capabilities and cross-dataset transfer performance. In practical applications, this framework exhibits good adaptability and robustness, effectively addressing the complex and diverse terrain environments in remote sensing imagery. Therefore, the DAFN-FAAM framework has high potential for promotion and application, and is expected to provide stronger technical support and decision-making basis for emergency response and disaster management.

[0069] The following detailed embodiments illustrate the application scenarios, specific technical solutions, training methods for building damage assessment models, and assessment and verification methods of the building damage assessment method provided by the present invention.

[0070] Figure 1This is an application scenario diagram of the building damage assessment method according to an embodiment of the present invention.

[0071] like Figure 1 As shown, application scenario 100 according to this embodiment may include remote sensing image processing and post-disaster large-scale building damage assessment. Network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0072] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0073] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0074] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0075] It should be noted that the building damage assessment method provided in the embodiments of the present invention can generally be executed by server 105. Correspondingly, the building damage assessment device provided in the embodiments of the present invention can generally be located in server 105. The building damage assessment method provided in the embodiments of the present invention can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the building damage assessment device provided in the embodiments of the present invention can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0076] It should be understood that Figure 1The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0077] The following will be based on Figure 1 The described scene, through Figures 2-8 The building damage assessment method of the disclosed embodiments is described in detail.

[0078] Figure 2 This is a flowchart of a building damage assessment method according to an embodiment of the present invention.

[0079] like Figure 2 As shown, the building damage assessment method of this embodiment includes operations S210 to S250.

[0080] In operation S210, the first image processing branch of the building damage assessment model is used to perform wavelet transform processing and multi-round adaptive global feature extraction on the pre-disaster images of the target building complex to obtain a multi-scale first feature map set.

[0081] The pre-disaster and post-disaster images of the target building complex involved in this invention may optionally be remote sensing images.

[0082] The aforementioned building damage assessment model is constructed based on DAFN-FAAM. The building damage assessment model is divided into two image branches. The first image processing branch is used to process the pre-disaster images of the target building complex. Wavelet transform processing and multi-round adaptive global feature extraction are performed on the pre-disaster images. The multi-scale wavelet fusion module (MWF) of the DAFN-FAAM model is used for wavelet transform, and the adaptive attention module (AAM) of the DAFN-FAAM model is used for multi-round adaptive global feature extraction.

[0083] The first image processing branch of the aforementioned building damage assessment model consists of four stages: one stage of wavelet transform processing and three stages of adaptive global feature extraction.

[0084] In operation S220, the first image processing branch is used to perform masking processing on the multi-scale first feature map set to obtain the first image segmentation result with mask information.

[0085] In operation S230, the second image processing branch of the building damage assessment model is used to perform wavelet transform processing, multi-round adaptive global feature extraction, and multi-round same-dimensional feature map fusion on the pre-disaster and post-disaster images of the target building complex in parallel, so as to obtain a multi-scale second feature map set.

[0086] Similar to the pre-disaster image processing workflow, the second image processing branch also includes a wavelet transform process and a three-stage adaptive global feature extraction process for processing pre-disaster and post-disaster images.

[0087] The second image processing branch of the building damage assessment model is used to process the pre-disaster and post-disaster images of the target building complex in parallel. During wavelet transform processing and multi-round adaptive global feature extraction, the pre-disaster and post-disaster images are fused at the same scale and then processed in subsequent stages of image processing. For example, the wavelet transform results of the pre-disaster images and the wavelet transform results of the post-disaster images are fused at multiple resolutions and then processed in subsequent stages.

[0088] In operation S240, the second image processing branch is used to perform convolution and activation processing on the multi-scale second feature map set to obtain the second image segmentation result.

[0089] In operation S250, the building damage assessment model is used to fuse the first image segmentation result and the second image segmentation result to obtain the damage assessment result of the target building complex.

[0090] The damage assessment results for the aforementioned target building complex include the background portion of the image, the undamaged portion, the slightly damaged portion, the severely damaged portion, and the completely destroyed portion.

[0091] The building damage assessment method provided by this invention, through multi-scale wavelet fusion of images of a target building complex, can learn robust features of buildings at different scales, thereby improving the assessment accuracy and generalization ability of the building damage assessment model. Simultaneously, by applying a multi-round adaptive global attention mechanism to the images of the target building complex, it can efficiently extract key features of the target building complex, significantly enhancing the feature representation capability of the building damage assessment model. Furthermore, the building damage assessment method provided by this invention, through deep image fusion technology, further improves the overall performance and generalization ability of the building damage assessment model, enabling it to perform building damage assessment in complex scenarios and further improve the detailed recognition of target building complexes.

[0092] The following specific embodiments and appendices Figures 3-5 The method for assessing building damage involved in this invention will be described in further detail.

[0093] Figure 3 This is a damage assessment architecture diagram of a target building complex according to an embodiment of the present invention.

[0094] Figure 4 This is a schematic diagram of data processing in the first image processing branch according to an embodiment of the present invention.

[0095] Figure 5This is a schematic diagram of data processing in the second image processing branch according to an embodiment of the present invention.

[0096] Satellite remote sensing images from the xBD and xFBD datasets are randomly cropped into 512×512 pixel patches. The images are in RGB three-channel format. A test set is determined from the cropped dataset, and the test set is used to illustrate the building damage assessment method provided by this invention.

[0097] Based on existing high-resolution semantic segmentation structures, this invention proposes a building damage assessment method based on the deep learning network DAFN-FAAM. The data processing architecture of the aforementioned DAFN-FAAM is as follows: Figure 2 As shown, DAFN-FAAM consists of two image processing branches: the first image processing branch and the second image processing branch. Both branches are image segmentation networks. The first image processing branch uses a Transformer-based network to segment buildings in the pre-disaster image and generate building location masks. The pre-disaster image is randomly cropped to 512×512 pixels and fed into the first image processing branch for building segmentation.

[0098] The first image processing branch processes the pre-disaster images of the target building complex as follows: Figure 4As shown: First, the image is normalized and standardized. In the first stage, the 512×512 image is downsampled to a 128×128 feature map using MWF, providing a foundation for subsequent stages. In the second stage, a four-layer AAM is used. The network not only maintains the high-resolution feature map from the first stage but also generates a lower-resolution (64×64) feature map through downsampling, forming two parallel feature map sets with different resolutions. These two feature map sets interact through a specific fusion strategy (e.g., adding the restored resolutions after upsampling), achieving preliminary multi-scale information fusion. The third stage also uses four layers of AAM to further increase the resolution diversity of the feature maps, generating three types of feature maps: the original high-resolution (128×128), intermediate resolution (64×64), and lower resolution (32×32). These feature maps also interact through a fusion strategy, enabling the network to capture richer contextual information. In the fourth stage, using only two layers of AAM, the first image processing branch ultimately generated feature maps at four resolutions: highest resolution (128×128), higher resolution (64×64), intermediate resolution (32×32), and lowest resolution (32×32). These feature maps were upsampled to the same size and then fused through a channel-dimensional concatenation mechanism, fully utilizing information at different scales. Finally, Softmax activation and a four-fold upsampling were applied to generate accurate and detailed semantic segmentation results. Throughout the entire process, the first image processing branch maintained a high-resolution feature representation.

[0099] The second image processing branch processes the pre-disaster and post-disaster images of the target building complex as follows: Figure 5 As shown, using two images, one before and one after a disaster, a Transformer-based fusion network (the second image processing branch) is used to classify the damage to buildings after the disaster. This second image processing branch has two sub-branches, each with the same structure as the main image processing branch. The difference is that in stages one through four, all feature maps from the two sub-branches are obtained. Each feature map is then concatenated according to its channel dimension, and a 3×3 convolution is used to halve the channel dimension before feeding it into the next stage of the two sub-branches. Finally, the resulting feature map undergoes a 3×3 convolution followed by a Softmax operation. This result is combined with a mask generated by the first image processing branch, and after post-processing, the final assessment of building damage is achieved.

[0100] According to an embodiment of the present invention, the above-mentioned wavelet transform processing and multi-round adaptive global feature extraction of the pre-disaster image of the target building complex using the first image processing branch of the building damage assessment model to obtain a multi-scale first feature map set includes: normalizing and standardizing the pre-disaster image of the target building complex using the building damage assessment model to obtain a preprocessed pre-disaster image; performing split wavelet transform processing on the preprocessed pre-disaster image using the multi-scale wavelet fusion module of the first image processing branch, and performing splicing and convolutional dimensionality reduction processing on the multiple initial wavelet transform feature maps obtained to obtain wavelet transform feature maps; dividing the wavelet transform feature maps into striped regions in the vertical and horizontal directions using the first image processing branch, and applying a neighborhood attention mechanism to the wavelet transform feature maps with multiple striped regions to obtain wavelet transform feature maps with neighborhood information; and performing multi-round adaptive global feature extraction on the wavelet transform feature maps with neighborhood information using multiple adaptive cross-attention modules of the first image processing branch to obtain a multi-scale first feature map set.

[0101] The following describes specific implementation methods in conjunction with appendices. Figures 6-7 The process of obtaining the multi-scale first feature map set according to the present invention will be described in further detail.

[0102] Figure 6 This is a structural diagram of a multi-scale wavelet fusion module according to an embodiment of the present invention.

[0103] Figure 7 This is a structural diagram of the adaptive attention mechanism module according to an embodiment of the present invention.

[0104] Figure 8 This is a structural diagram of a feedforward neural network with a gating mechanism according to an embodiment of the present invention.

[0105] The acquisition process of the above-mentioned multi-scale first feature map mainly involves wavelet transform and adaptive global feature extraction, and utilizes the multi-scale wavelet fusion (MWF) module and adaptive attention (AAM) module of DAFN-FAAM.

[0106] Because buildings vary in scale across images from different disasters and regions, it is necessary to effectively capture detailed damage features and background information at different scales. Therefore, a multi-scale wavelet fusion (MWF) module is introduced into the building damage assessment model. The structure of MWF is as follows: Figure 6 As shown, the MWF module uses wavelet transform to decompose the input image into high-frequency and low-frequency components, thereby capturing buildings at different damage scales. This enables the model to capture key, fine-grained damage features and broader contextual information, such as the surrounding environment and spatial relationships between buildings, which are crucial for accurate building damage assessment.

[0107] According to an embodiment of the present invention, the above-mentioned multi-scale wavelet fusion module of the first image processing branch performs split wavelet transform processing on the preprocessed pre-disaster image, and performs convolutional dimensionality reduction processing on the obtained multiple initial wavelet transform feature maps to obtain wavelet transform feature maps includes: performing convolution processing on the preprocessed pre-disaster image using the feature map generation stream of the multi-scale wavelet fusion module, and performing batch normalization processing on the result of the convolution processing to generate a first initial wavelet transform feature map; based on Haar wavelet transform, decomposing the preprocessed pre-disaster image into high-frequency components and low-frequency components using the image decomposition stream of the multi-scale wavelet fusion module to obtain multiple second initial wavelet transform feature maps; and performing channel-dimensional concatenation processing on the first initial wavelet transform feature map and multiple second initial wavelet transform feature maps using the multi-scale wavelet fusion module, and performing convolutional dimensionality reduction processing on the concatenated feature map to obtain the wavelet transform feature map.

[0108] Although the first and second image processing branches are structurally similar, the pre-disaster and post-disaster images are transformed at two different scales.

[0109] First, the original image has a resolution of (512×512), and this image yields a (128×128) feature map through MWF. Features are then extracted using wavelet transform and embedded into the network encoder, as follows: Figure 5 As shown, the MWF module comprises two streams. The first stream operates on an image at its original resolution with a size of (512×512), generating feature maps using convolutional layers and batch normalization. The second stream processes four feature maps (256×256) obtained through Haar wavelet transform. The Haar wavelet decomposes the image into low-frequency components (YL) and high-frequency components (YH). The low-frequency components (YL) preserve the overall structure and main features of the image, while the high-frequency components (YH) capture finer details and edge information. The feature maps from these two streams are then concatenated for further integration by stitching them together along the channel dimension. To ensure consistency with the number of channels introduced in the first stage of the second image processing branch, convolution is applied to the concatenated feature maps to reduce the dimensionality, resulting in a final feature map size of (128×128). By embedding image features at different scales into the network, the MWF module can effectively capture subtle damage features of buildings while enhancing broader contextual information.

[0110] According to an embodiment of the present invention, the above-mentioned method of dividing the wavelet transform feature map into striped regions in the vertical and horizontal directions using a first image processing branch, and applying a neighborhood attention mechanism to the wavelet transform feature map with multiple striped regions to obtain a wavelet transform feature map with neighborhood information includes: dividing the wavelet transform feature map into striped regions in the vertical and horizontal directions using a first image processing branch to obtain a wavelet transform feature map with multiple striped regions; dividing each striped region in the wavelet transform feature map with multiple striped regions into multiple cross-attention window regions using a first image processing branch based on preset window size information; and performing neighborhood feature fusion on all pixels in each cross-attention window region using a first image processing branch based on the neighborhood attention mechanism to obtain a wavelet transform feature map with neighborhood information.

[0111] The above-mentioned striped region division helps to capture the characteristics of building damage, which facilitates subsequent classification tasks.

[0112] According to an embodiment of the present invention, the above-described method of fusing neighborhood features of all pixels in each cross-attention window region using a first image processing branch to obtain a wavelet transform feature map with neighborhood information includes: extracting local features of each pixel in each cross-attention window region using the first image processing branch to generate a local feature matrix for each pixel; performing a learnable linear transformation on the local feature matrix of each pixel using the first image processing branch to obtain a query vector, key vector, and value vector for each pixel; calculating the attention weight of each pixel using the query vector, key vector, and value vector, and embedding the relative position deviation obtained by convolution of the first image processing branch into the attention weight; and performing a weighted summation of the attention weight and local feature matrix of each pixel using the first image processing branch to obtain a wavelet transform feature map with neighborhood information.

[0113] According to an embodiment of the present invention, the above-mentioned multi-round adaptive global feature extraction of wavelet transform feature maps with neighborhood information using multiple adaptive cross-attention modules of the first image processing branch to obtain a multi-scale first feature map set includes: introducing an attention mechanism based on compression and activation into multiple adaptive cross-attention modules and setting a hierarchical structure for each adaptive cross-attention module; using the first adaptive cross-attention module to perform global average pooling processing on the wavelet transform feature maps with neighborhood information to obtain a first pooled feature map with context information; using the first adaptive cross-attention module to perform fully connected processing and nonlinear activation processing on the wavelet transform feature maps with neighborhood information to obtain a first activated feature map with channel attention weights; and using the first adaptive cross-attention module to perform upsampling feature fusion based on the same channel dimension on the first pooled feature map with context information and the first activated feature map with channel attention weights to obtain a first attention feature map.

[0114] According to an embodiment of the present invention, the above-mentioned multi-round adaptive global feature extraction of wavelet transform feature maps with neighborhood information using multiple adaptive cross-attention modules of the first image processing branch to obtain a multi-scale first feature map set further includes: performing global average pooling processing on the first attention feature map using a second adaptive cross-attention module to obtain a second pooling feature map with context information; performing fully connected processing and nonlinear activation processing on the first attention feature map using the second adaptive cross-attention module to obtain a second activated feature map with channel attention weights; and performing upsampling feature fusion based on the same channel dimension on the second pooling feature map with context information and the second activated feature map with channel attention weights using the second adaptive cross-attention module to obtain a second attention feature map.

[0115] According to an embodiment of the present invention, the above-mentioned multi-round adaptive global feature extraction of wavelet transform feature maps with neighborhood information by using multiple adaptive cross-attention modules of the first image processing branch to obtain a multi-scale first feature map set further includes: using a third adaptive cross-attention module to perform global average pooling processing on the second attention feature map to obtain a third pooling feature map with context information; and using the third adaptive cross-attention module to perform fully connected processing and nonlinear activation processing on the second attention feature map to obtain a third activated feature map with channel attention weights.

[0116] According to an embodiment of the present invention, the above-mentioned masking process of the multi-scale first feature map set using the first image processing branch to obtain the first image segmentation result with mask information includes: upsampling the feature map in the multi-scale first feature map set using the first image processing branch, and splicing and fusing the upsampled feature map with the same channel dimension to obtain the spliced ​​and fused first feature map; and activating the spliced ​​and fused first feature map using a preset activation function of the first image processing branch to obtain the first image segmentation result with mask information.

[0117] According to an embodiment of the present invention, the second image processing branch of the building damage assessment model described above performs wavelet transform processing, multi-round adaptive global feature extraction, and multi-round same-dimensional feature map fusion on the pre-disaster and post-disaster images of the target building complex in parallel to obtain a multi-scale second feature map set. This includes: performing wavelet transform on the pre-disaster and post-disaster images respectively using different processing units of the second image processing branch, and fusing the wavelet transform results of the pre-disaster and post-disaster images with features of the same dimension to obtain a wavelet transform fused feature map; performing multi-round adaptive global feature extraction on the wavelet transform results of the pre-disaster and post-disaster images respectively using different processing units of the second image processing branch to obtain multiple pre-disaster global feature maps and multiple post-disaster global feature maps of different scales; and fusing the pre-disaster global feature maps and post-disaster global feature maps of the same dimension in each round to obtain multiple global feature maps of different scales.

[0118] The multi-scale first feature map set provided by the present invention will be further described in detail below through specific embodiments. Since the second image processing branch is structurally similar to the first image processing branch, obtaining the multi-scale second feature map set is similar to obtaining the multi-scale first feature map set, and will be described together here.

[0119] To enhance the robustness of building damage assessment models, this invention proposes an innovative Adaptive Attention (AAM) module, such as... Figure 7As shown, this AAM module combines the advantages of CNN and Transformer, introducing an adaptive cross-attention mechanism to reduce computational complexity while achieving efficient global feature extraction. The AAM module employs a cross-window-based self-attention mechanism, dynamically adjusting the attention range to achieve a balance between efficient computation and flexibility in adapting to different task requirements and input features. The feature map processing flow is similar to that used when processing the input feature map I∈R2C×H×W. This invention employs an innovative method: first, the feature map is divided into striped regions along its horizontal and vertical directions, with the width and height of these stripes set to h1 and w1, respectively, ensuring they are larger than the size of the local attention window k. Within each h1×w1 striped region, this invention further divides the region into several k×k windows and applies a neighborhood attention mechanism within them. Specifically, for each pixel coordinate (i,j) within the k×k window, this invention extracts its local features to form a matrix Xp(i,j). Then, this matrix is ​​transformed through a learnable transformation to generate Query, Key, and Value matrices Q, K, and V. Next, combining the relative positional deviation B(i,j), this invention calculates the attention weight Aij, where dk represents the dimension of the Key. According to formula (1), the weighted sum of Aij and Vp(i,j) yields Z(i,j), representing the fused features of pixel (i,j) and its neighborhood. After iterating over all pixels within the stripe region, this invention concatenates the output features Z(i,j) to obtain a feature map Z(i,j), which reflects the results of the local attention operation. To further enhance the feature representation, this invention applies global attention within each h1×w1 stripe region. At this stage, the feature map within the window is reshaped into a two-dimensional matrix, where each row represents a feature channel and each column represents a pixel position within the window. Then, a standard self-attention mechanism is applied by performing a linear transformation to obtain new Query, Key, and Value matrices. By combining the relative positional encoding BE(i,j), attention weights are calculated and used to perform a weighted sum on the Value matrix, thereby obtaining a feature representation rich in global information. Finally, the weighted sum is combined with the convolution position bias Cij, as shown in Equation (2), and reshaped to its original spatial dimensions to obtain the global attention feature map for each stripe region. By connecting the global attention feature maps of all stripe regions, the present invention obtains the final feature map Yij. This feature map helps to capture building damage features, providing a richer and more comprehensive representation for subsequent classification tasks.

[0120] (1),

[0121] (2).

[0122] To address the limitations of Transformers in capturing local features and spatial information and to improve the model's generalization ability, this invention proposes an innovative feedforward neural network (GMFFN) that draws on the SE attention mechanism. GMFFN introduces the SE mechanism to enhance feature representation capabilities along the channel dimension. Figure 8 As shown, this invention divides the original feature map into two parts along the channel dimension. The first part uses global average pooling to efficiently condense the input feature map and extract global contextual information. The second part processes the condensed information through a fully connected layer and a non-linear activation function (ReLU) to generate precise channel attention weights. These weights reflect the importance of different channel features, enabling dynamic and precise control of channel characteristics. Finally, the weights are multiplied element-wise with the original input feature map to fine-tune the channel features. The final feature map size is similar to the original input feature map. Figure 1 This design, while retaining the global information processing capabilities of the Transformer, introduces a refined attention mechanism for the channel dimension. Therefore, GMFFN can effectively capture and utilize the potential relationships and differences between channel features, thereby improving the accuracy and efficiency of the model in building damage assessment.

[0123] Figure 9 This is a schematic diagram illustrating the training of a building damage assessment model according to an embodiment of the present invention.

[0124] like Figure 9 As shown, to obtain the aforementioned building damage assessment model, satellite remote sensing imagery from the xBD and xFBD datasets is needed to train the model. Due to limitations in computer computing power, the images are randomly cropped into 512×512 pixel tiles, and the images are in RGB three-channel format. The dataset is divided into training, validation, and test sets, and validation is performed on the Ida-BD dataset.

[0125] This invention uses Focal Loss and Dice Loss functions to address the challenge of class imbalance in datasets, thereby optimizing the model's learning process. The specific loss functions are shown in formulas (3) to (5). Focal Loss is a loss function specifically designed to solve the class imbalance problem. It adds an adjustment factor to the traditional cross-entropy loss, allowing the model to focus more on samples that are difficult to distinguish. This adjustment factor is dynamic and gradually decreases to zero as the model's confidence in the correct class prediction increases. This means that during training, the model will pay less attention to samples that can be accurately classified, and more energy will be focused on samples that are difficult to classify correctly. Specifically, y represents the true label of the sample, and p represents the model's prediction probability. This formula essentially measures the difference between the prediction result and the true label. The final loss value is obtained by subtracting this difference from 1. Therefore, when the prediction result is completely consistent with the true label, the loss value is zero; while when the prediction result is completely inconsistent with the true label, the loss value reaches its maximum value, i.e., 1:

[0126] (3),

[0127] (4),

[0128] (5).

[0129] During model training, the input data was first cropped to 512×512 images. Experiments were conducted using the PyTorch 2.2.0 platform, with computational support provided by an NVIDIA RTX-3090 GPU. To enhance data diversity in the mixed xBD and xFBD datasets, various data augmentation strategies were employed, including rotation, cropping, adding noise, and flipping. For gradient updates, the AdamW optimizer was chosen due to its advantages in improving the training process. AdamW enhances the traditional Adam optimizer by incorporating more complex weight decay adjustment methods, thereby promoting better generalization and faster convergence. The learning rate was consistently set to 1e-4 for both building localization (stage 1) and damage classification (stage 2) stages. The model in stage 1 achieved its best performance at 150 epochs, while the model in stage 2 achieved its best performance at 300 epochs, utilizing the pre-trained weights from stage 1. The Ida-BD dataset was not included in the training process but was tested separately. This approach ensures unbiased validation of the model's performance on high-resolution satellite imagery.

[0130] In accordance with standard practices in building damage assessment research, this invention selects precision and recall as the main indicators for evaluating model performance. This selection ensures that the assessment method is consistent with established benchmarks in the research community. In this invention, precision is defined as the ratio of true positives (TP) to the sum of true positives (TP) and false positives (FP), representing the accuracy of positive predictions. Recall, also known as sensitivity, is calculated as the ratio of true positives (TP) to the sum of true positives (TP) and false negatives (FP), showing the model's ability to accurately identify all actual building damage instances. This invention utilizes building damage assessment indicators proposed in the xView2 computer vision challenge, as shown in Equation (6). These indicators include the weighted average F1 score (F1loc) for building segmentation and the uniform average F1 score (F1cls) for damage classification, where F1Ci is the F1 score for each damage category, and Ci is the i-th damage level. Specifically, C1 to C4 correspond to no damage, minor damage, severe damage, and damage, respectively. The definitions of F1loc and F1cls are given in Equations (7) and (8).

[0131] (6),

[0132] (7),

[0133] (8).

[0134] To better illustrate the advantages of the trained DAFN-FAAM-based building damage assessment method, the following section uses Tables 1 and 2, along with appendices, to demonstrate its effectiveness. Figure 10 and appendix Figure 11 To verify the effectiveness of the building damage assessment method provided by this invention.

[0135] Figure 10 This is a schematic diagram of the results of post-disaster building damage assessment in a hybrid dataset of xBD and xFBD according to an embodiment of the present invention.

[0136] Figure 11 This is a schematic diagram of the results of building damage assessment in the Ida-BD dataset according to an embodiment of the present invention.

[0137] Table 1 - Statistical Table of Comprehensive Indicators for Post-Disaster Building Damage Assessment

[0138] (A mix of xBD and XFBD datasets)

[0139]

[0140] Table 2 - Statistics of Comprehensive Validation Indicators for the Ida-BD Dataset

[0141]

[0142] Through Tables 1 and 2 and Figure 10 and Figure 11 As shown, the building damage assessment method based on DAFN-FAAM of this invention is far superior to existing building damage assessment methods in terms of actual performance and verification indicators.

[0143] Based on the above-described building damage assessment method, this invention also provides a building damage assessment device. The following will be combined with... Figure 12 The device is described in detail.

[0144] Figure 12 A schematic diagram of a building damage assessment device according to an embodiment of the present invention is shown.

[0145] like Figure 12 As shown, the building damage assessment device 1200 of this embodiment includes a first acquisition module 1210, a mask processing module 1220, a second acquisition module 1230, a convolution and activation module 1240, and an image fusion module 1250.

[0146] The first acquisition module 1210 is used to perform wavelet transform processing and multi-round adaptive global feature extraction on the pre-disaster image of the target building complex using the first image processing branch of the building damage assessment model to obtain a multi-scale first feature map set. In one embodiment, the first acquisition module 1210 can be used to perform the operation S210 described above, which will not be repeated here.

[0147] The mask processing module 1220 is used to perform mask processing on the multi-scale first feature map set using the first image processing branch to obtain a first image segmentation result with mask information. In one embodiment, the mask processing module 1220 can be used to perform the operation S220 described above, which will not be repeated here.

[0148] The second acquisition module 1230 is used to perform wavelet transform processing, multi-round adaptive global feature extraction, and multi-round same-dimensional feature map fusion on the pre-disaster and post-disaster images of the target building complex in parallel using the second image processing branch of the building damage assessment model to obtain a multi-scale second feature map set. In one embodiment, the second acquisition module 1230 can be used to perform the operation S230 described above, which will not be repeated here.

[0149] The convolution and activation module 1240 is used to perform convolution and activation processing on the multi-scale second feature map using the second image processing branch to obtain the second image segmentation result; in one embodiment, the convolution and activation module 1240 can be used to perform the operation S240 described above, which will not be repeated here.

[0150] The image fusion module 1250 is used to fuse the first image segmentation result and the second image segmentation result using the building damage assessment model to obtain the damage assessment result of the target building group. In one embodiment, the image fusion module 1250 can be used to perform the operation S250 described above, which will not be repeated here.

[0151] According to embodiments of the present invention, any plurality of modules among the first acquisition module 1210, mask processing module 1220, second acquisition module 1230, convolution and activation module 1240, and image fusion module 1250 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the first acquisition module 1210, mask processing module 1220, second acquisition module 1230, convolution and activation module 1240, and image fusion module 1250 can be at least partially implemented as hardware circuitry, such as field-programmable gate array (FPGA), programmable logic array (PLA), system-on-a-chip, system-on-a-substrate, system-on-package, application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any one of the three implementation methods or a suitable combination of any of them. Alternatively, at least one of the first acquisition module 1210, mask processing module 1220, second acquisition module 1230, convolution and activation module 1240, and image fusion module 1250 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0152] Figure 13 This is a block diagram of an electronic device suitable for implementing a building damage assessment method and a training method for a building damage assessment model according to an embodiment of the present invention.

[0153] like Figure 13 As shown, an electronic device 1300 according to an embodiment of the present invention includes a processor 1301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1302 or a program loaded from a storage portion 1308 into a random access memory (RAM) 1303. The processor 1301 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1301 may also include onboard memory for caching purposes. The processor 1301 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0154] RAM 1303 stores various programs and data required for the operation of electronic device 1300. Processor 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. Processor 1301 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 1302 and / or RAM 1303. It should be noted that the programs may also be stored in one or more memories other than ROM 1302 and RAM 1303. Processor 1301 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0155] According to an embodiment of the present invention, the electronic device 1300 may further include an input / output (I / O) interface 1305, which is also connected to the bus 1304. The electronic device 1300 may also include one or more of the following components connected to the input / output (I / O) interface 1305: an input section 1306 including a keyboard, mouse, etc.; an output section 1307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1308 including a hard disk, etc.; and a communication section 1309 including a network interface card such as a LAN card, modem, etc. The communication section 1309 performs communication processing via a network such as the Internet. A drive 1310 is also connected to the input / output (I / O) interface 1305 as needed. A removable medium 1311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1310 as needed so that computer programs read from it can be installed into the storage section 1308 as needed.

[0156] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0157] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 1302 and / or RAM 1303 and / or one or more memories other than ROM 1302 and RAM 1303 described above.

[0158] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or pairings fall within the scope of this invention.

[0159] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A method for assessing building damage, characterized in that, The method includes: The first image processing branch of the building damage assessment model is used to perform wavelet transform processing and multi-round adaptive global feature extraction on the pre-disaster images of the target building complex to obtain a multi-scale first feature map set; The first image processing branch is used to perform masking on the multi-scale first feature map set to obtain a first image segmentation result with mask information; The second image processing branch of the building damage assessment model is used to perform wavelet transform processing, multi-round adaptive global feature extraction, and multi-round same-dimensional feature map fusion on the pre-disaster and post-disaster images of the target building complex in parallel to obtain a multi-scale second feature map set. The second image processing branch is used to perform convolution and activation processing on the multi-scale second feature map to obtain the second image segmentation result. The damage assessment results of the target building complex are obtained by fusing the first image segmentation result and the second image segmentation result using the building damage assessment model. Specifically, the first image processing branch of the building damage assessment model is used to perform wavelet transform processing and multi-round adaptive global feature extraction on the pre-disaster images of the target building complex, resulting in a multi-scale first feature map set including: The pre-disaster images of the target building complex are normalized and standardized using the building damage assessment model to obtain pre-processed pre-disaster images. The multi-scale wavelet fusion module of the first image processing branch is used to perform split wavelet transform processing on the preprocessed pre-disaster image, and the multiple initial wavelet transform feature maps are spliced ​​and convolutionally reduced to obtain wavelet transform feature maps. The first image processing branch is used to divide the wavelet transform feature map into striped regions in the vertical and horizontal directions, and a neighborhood attention mechanism is applied to the wavelet transform feature map with multiple striped regions to obtain a wavelet transform feature map with neighborhood information. The first feature map set is obtained by using multiple adaptive cross-attention modules of the first image processing branch to perform multiple rounds of adaptive global feature extraction on the wavelet transform feature map with neighborhood information.

2. The method according to claim 1, characterized in that, The preprocessed pre-disaster image is subjected to split wavelet transform processing using the multi-scale wavelet fusion module of the first image processing branch. The resulting multiple initial wavelet transform feature maps are then stitched together and subjected to convolutional dimensionality reduction processing to obtain wavelet transform feature maps including: The feature map generation stream of the multi-scale wavelet fusion module is used to perform convolution processing on the preprocessed pre-disaster image, and the result of the convolution processing is batch normalized to generate the first initial wavelet transform feature map. Based on Haar wavelet transform, the image decomposition stream of the multi-scale wavelet fusion module is used to decompose the pre-processed pre-disaster image into high-frequency components and low-frequency components to obtain multiple second initial wavelet transform feature maps. The first initial wavelet transform feature map and multiple second initial wavelet transform feature maps are concatenated along the channel dimension using the multi-scale wavelet fusion module, and the resulting feature map is then subjected to convolutional dimensionality reduction processing to obtain the wavelet transform feature map.

3. The method according to claim 1, characterized in that, The first image processing branch is used to divide the wavelet transform feature map into striped regions in the vertical and horizontal directions. Then, a neighborhood attention mechanism is applied to the wavelet transform feature map with multiple striped regions to obtain a wavelet transform feature map containing neighborhood information, including: The wavelet transform feature map is divided into stripe regions in the vertical and horizontal directions using the first image processing branch to obtain a wavelet transform feature map with multiple stripe regions. Based on the preset window size information, the first image processing branch is used to divide each of the stripe regions in the wavelet transform feature map with multiple stripe regions into multiple cross-attention window regions. Based on the neighborhood attention mechanism, the first image processing branch is used to perform neighborhood feature fusion on all pixels in each of the cross attention window regions to obtain the wavelet transform feature map with neighborhood information.

4. The method according to claim 3, characterized in that, Based on the neighborhood attention mechanism, the first image processing branch performs neighborhood feature fusion on all pixels in each of the cross-attention window regions to obtain the wavelet transform feature map with neighborhood information, including: The local features of each pixel in each of the cross-attention window regions are extracted using the first image processing branch to generate a local feature matrix for each pixel; The first image processing branch is used to perform a learnable linear transformation on the local feature matrix of each pixel to obtain the query vector, key vector and value vector of each pixel; The attention weight of each pixel is calculated using the query vector, key vector, and value vector of each pixel, and the relative positional deviation obtained by the convolution of the first image processing branch is embedded into the attention weight. The attention weight and local feature matrix of each pixel are weighted and summed using the first image processing branch to obtain the wavelet transform feature map with neighborhood information.

5. The method according to claim 1, characterized in that, The first image processing branch utilizes multiple adaptive cross-attention modules to perform multi-round adaptive global feature extraction on the wavelet transform feature map with neighborhood information, resulting in the multi-scale first feature map set comprising: A compression- and incentive-based attention mechanism is introduced into multiple adaptive cross-attention modules, and a hierarchical structure is set for each adaptive cross-attention module; The wavelet transform feature map with neighborhood information is subjected to global average pooling using the first adaptive cross-attention module to obtain a first pooled feature map with context information. The first adaptive cross-attention module is used to perform fully connected processing and nonlinear activation processing on the wavelet transform feature map with neighborhood information to obtain a first activated feature map with channel attention weights. The first adaptive cross-attention module is used to perform upsampling feature fusion of the first pooling feature map with contextual information and the first activation feature map with channel attention weights based on the same channel dimension to obtain the first attention feature map.

6. The method according to claim 5, characterized in that, Also includes: The first attention feature map is subjected to global average pooling using the second adaptive cross-attention module to obtain a second pooled feature map with contextual information. The first attention feature map is processed by fully connected processing and nonlinear activation processing using the second adaptive cross-attention module to obtain a second activation feature map with channel attention weights. The second adaptive cross-attention module is used to perform upsampling feature fusion of the second pooling feature map with contextual information and the second activation feature map with channel attention weights based on the same channel dimension to obtain the second attention feature map.

7. The method according to claim 6, characterized in that, Also includes: The second attention feature map is processed by global average pooling using the third adaptive cross attention module to obtain a third pooled feature map with contextual information. The second attention feature map is processed by fully connected processing and nonlinear activation processing using the third adaptive cross-attention module to obtain a third activation feature map with channel attention weights.

8. The method according to claim 1, characterized in that, The first image processing branch is used to perform masking processing on the multi-scale first feature map set to obtain a first image segmentation result with mask information, including: The first image processing branch is used to upsample the feature maps in the multi-scale first feature map set, and the upsampled feature maps are spliced ​​and fused with the same channel dimension to obtain the spliced ​​and fused first feature map; The first image segmentation result of the mask information is obtained by using the preset activation function of the first image processing branch to activate the first feature map after splicing and fusion.

9. The method according to claim 1, characterized in that, The second image processing branch of the building damage assessment model is used to perform wavelet transform processing, multi-round adaptive global feature extraction, and multi-round same-dimensional feature map fusion on the pre-disaster and post-disaster images of the target building complex in parallel, resulting in a multi-scale second feature map set including: Wavelet transform is performed on the pre-disaster image and the post-disaster image using different processing units of the second image processing branch, and the wavelet transform results of the pre-disaster image and the post-disaster image are fused with features of the same dimension to obtain a wavelet transform fused feature map. The wavelet transform results of the pre-disaster image and the wavelet transform results of the post-disaster image are subjected to multiple rounds of adaptive global feature extraction using different processing units of the second image processing branch, resulting in multiple pre-disaster global feature maps and multiple post-disaster global feature maps of different scales. The global feature maps before and after the disaster in each round are fused with features of the same dimension to obtain multiple global feature maps of different scales.

Images