A remote sensing surveying data enhancement method and system for real estate surveying

Abstract

The application discloses a remote sensing surveying and mapping data enhancement method and system for real estate surveying and mapping, and relates to the technical field of image enhancement; an initial remote sensing image is preprocessed to obtain a preprocessed remote sensing image; multi-scale sampling is performed on the preprocessed remote sensing image to obtain a multi-scale remote sensing image; the multi-scale remote sensing image is substituted into an enhancement model to obtain a multi-scale enhanced remote sensing image; each remote sensing image and the enhanced remote sensing image of the same scale are substituted into an image fusion model to obtain a high-scale fusion image, a medium-scale fusion image and a low-scale fusion image; the high-scale fusion image, the medium-scale fusion image and the low-scale fusion image are fused to obtain an enhanced remote sensing image; through the synergistic effect of preprocessing, multi-scale sampling, enhancement and fusion, the detail definition, contrast and overall visual quality of the image under poor light conditions can be significantly improved under the premise of preserving the geometric precision and spectral authenticity of the original image, so that the accuracy and efficiency of real estate surveying and mapping are improved.

Description

Technical Field

[0001] This invention belongs to the field of image enhancement technology, specifically relating to a method and system for enhancing remote sensing mapping data used in real estate surveying. Background Technology

[0002] Remote sensing technology, as a crucial support for modern real estate surveying, can efficiently and over large areas acquire information on land cover and buildings, providing a key data foundation for property rights delineation, change monitoring, and spatial planning. However, in actual remote sensing data acquisition, it is highly dependent on airborne or spaceborne sensor equipment and external environmental conditions, resulting in inherent limitations in image quality. Among these, variable lighting conditions are a particularly prominent and difficult-to-control factor. Due to the influence of imaging time, weather conditions, and building shadows, the initial remote sensing images often exhibit problems such as insufficient overall illumination, local overexposure or underexposure, heavy shadows, and significantly reduced contrast.

[0003] Image quality degradation caused by poor lighting can severely limit the performance of subsequent computer vision algorithms. Key tasks in real estate surveying, such as high-precision building outline extraction, feature classification and segmentation, and change detection, rely heavily on the quality of the input image. Low-light, high-noise, and low-contrast images make feature extraction unstable, leading to a large number of missed and false detections.

[0004] Patent CN119359555A discloses a remote sensing image enhancement processing method, which includes remote sensing image dataset creation, initialization module construction, optimization sampling module one construction, optimization sampling module two construction, feature processing module construction, remote sensing image enhancement model construction, training and use of remote sensing image enhancement model; at the same time, it proposes a remote sensing image enhancement model, wherein optimization sampling module one calculates the compensation feature map through convolution operation and compensation feature map calculation, and optimization sampling module two calculates the compensation feature map through weighted average.

[0005] Existing image enhancement methods, such as histogram equalization and gamma correction, while improving visual effects to some extent, are often global adjustments. This can easily lead to overexposure and loss of information in bright areas while enhancing dark regions, and they cannot adaptively handle the complex illumination unevenness in remote sensing images. Some deep learning-based enhancement methods are designed for general scenarios and fail to fully consider the geometric accuracy requirements of real estate surveying. While the enhanced image may look better visually, it introduces spectral distortion or edge artifacts, ultimately destroying the spatial information used for accurate measurements and resulting in low efficiency in real estate surveying. Summary of the Invention

[0006] The purpose of this invention is to solve the problem of low efficiency in real estate surveying due to the inability to adaptively handle complex illumination unevenness in remote sensing images during remote sensing image enhancement, and to propose a remote sensing surveying data enhancement method and system for real estate surveying.

[0007] In a first aspect of this invention, a method for enhancing remote sensing mapping data for real estate surveying is first proposed, the method comprising: Acquire an initial remote sensing image, and preprocess the initial remote sensing image to obtain a preprocessed remote sensing image; The preprocessed remote sensing image is sampled at multiple scales to obtain a multi-scale remote sensing image; the multi-scale remote sensing image includes a high-scale remote sensing image, a medium-scale remote sensing image, and a low-scale remote sensing image; The multi-scale remote sensing images are substituted into the enhancement model to obtain multi-scale enhanced remote sensing images; the multi-scale enhanced remote sensing images include high-scale enhanced remote sensing images, medium-scale enhanced remote sensing images, and low-scale enhanced remote sensing images; Substituting each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model yields high-scale fused images, medium-scale fused images, and low-scale fused images. The high-scale fused image, the medium-scale fused image, and the low-scale fused image are fused to obtain an enhanced remote sensing image.

[0008] Optionally, preprocessing the initial remote sensing image to obtain a preprocessed remote sensing image includes: The first convolutional feature is obtained by performing a 3×3 convolution operation on the initial remote sensing image; Substitute the first convolutional feature into the attention block to obtain the first attention feature; The first convolutional feature and the first attention feature are concatenated and then subjected to a 1×1 convolution to obtain the first concatenated feature; The first spliced ​​feature is subjected to a two-dimensional fast Fourier transform and then two 1×1 convolutions to obtain the second convolutional feature; The second convolutional feature is subjected to a two-dimensional inverse fast Fourier transform to obtain the third convolutional feature; The first concatenated feature is concatenated with the third concatenated feature after performing two 3×3 convolution operations on the first concatenated feature; Substitute the second splicing feature into the channel attention layer to obtain the second attention feature; The initial remote sensing image and the second attention feature are fused to obtain a preprocessed remote sensing image.

[0009] Optionally, substituting the multi-scale remote sensing images into the enhancement model to obtain multi-scale enhanced remote sensing images includes: The target-scale remote sensing image is processed by average pooling and max pooling to obtain the average pooling image and the max pooling image; the target-scale remote sensing image is any one of the multi-scale remote sensing images; After substituting the average pooling image and the max pooling image into the fully connected layer respectively, the average pooling weight and the max pooling weight are obtained by using the Sigmoid activation function. Substituting the average pooled image into the channel attention branch yields the channel attention features; Substituting the max-pooled image into the spatial attention branch yields spatial attention features; The target-scale enhanced remote sensing image is obtained by fusing the target-scale remote sensing image, the channel attention feature, and the spatial attention feature according to the average pooling weight and the max pooling weight; and a multi-scale enhanced remote sensing image is obtained by fusing the target-scale enhanced remote sensing images at all scales.

[0010] Optionally, fusing the target-scale remote sensing image, the channel attention feature, and the spatial attention feature based on the average pooling weight and the max pooling weight to obtain the target-scale enhanced remote sensing image includes: Through formula Obtain the initial target scale-enhanced remote sensing image; in, Enhance the remote sensing image at the initial target scale. The average pooling weight, The channel attention feature, The maximum pooling weight, For the spatial attention features, The target-scale remote sensing image; The initial target-scale enhanced remote sensing image is normalized to obtain the target-scale enhanced remote sensing image.

[0011] Optionally, by substituting each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model, high-scale fused images, medium-scale fused images, and low-scale fused images are obtained, including: Perform 1×1 convolution, ReLU activation, 3×3 convolution, and ReLU activation operations on remote sensing images and enhanced remote sensing images of the same scale to obtain the first convolutional remote sensing image and the first convolutional enhanced remote sensing image; The first convolutional remote sensing image and the first convolutional enhanced remote sensing image are stitched together to obtain a stitched remote sensing image. The stitched remote sensing image is subjected to average pooling along the channel dimension to obtain the channel average remote sensing features; the stitched remote sensing image is subjected to max pooling along the channel dimension to obtain the channel maximum remote sensing features; The average remote sensing features and the maximum remote sensing features of the channel are concatenated and then subjected to a 7×7 convolution, and a weight map is obtained by activating with Sigmoid. Based on the weighted map, remote sensing images and enhanced remote sensing images of the same scale are fused to obtain high-scale fused images, medium-scale fused images, and low-scale fused images.

[0012] In a second aspect of this invention, a remote sensing mapping data enhancement system for real estate surveying is provided, comprising: The preprocessing module is used to acquire an initial remote sensing image and preprocess the initial remote sensing image to obtain a preprocessed remote sensing image. A multi-scale sampling module is used to perform multi-scale sampling on the preprocessed remote sensing image to obtain a multi-scale remote sensing image; the multi-scale remote sensing image includes a high-scale remote sensing image, a medium-scale remote sensing image, and a low-scale remote sensing image. The remote sensing image enhancement module is used to substitute the multi-scale remote sensing image into the enhancement model to obtain a multi-scale enhanced remote sensing image; the multi-scale enhanced remote sensing image includes a high-scale enhanced remote sensing image, a medium-scale enhanced remote sensing image, and a low-scale enhanced remote sensing image; The remote sensing image fusion module is used to input each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model to obtain high-scale fused images, medium-scale fused images and low-scale fused images; A multi-scale image fusion module is used to fuse the high-scale fused image, the medium-scale fused image, and the low-scale fused image to obtain an enhanced remote sensing image.

[0013] Optionally, the preprocessing module includes: The first convolutional feature generation module is used to perform a 3×3 convolution operation on the initial remote sensing image to obtain the first convolutional feature. The first attention feature generation module is used to substitute the first convolutional features into the attention block to obtain the first attention features; The first splicing feature generation module is used to splice the first convolutional feature and the first attention feature and then perform a 1×1 convolution to obtain the first splicing feature; The second convolutional feature generation module is used to perform a two-dimensional fast Fourier transform on the first concatenated feature and then perform two 1×1 convolutions to obtain the second convolutional feature. The third convolutional feature generation module is used to perform a two-dimensional fast Fourier inverse transform on the second convolutional feature to obtain the third convolutional feature; The second splicing feature generation module is used to perform two 3×3 convolution operations on the first splicing feature and then splice it with the third convolution feature to obtain the second splicing feature; The second attention feature generation module is used to substitute the second splicing feature into the channel attention layer to obtain the second attention feature; The preprocessed remote sensing image generation module is used to fuse the initial remote sensing image and the second attention feature to obtain a preprocessed remote sensing image.

[0014] Optionally, the remote sensing image enhancement module includes: The pooling processing module is used to perform average pooling and max pooling processing on the target-scale remote sensing image to obtain the average pooling image and the max pooling image; the target-scale remote sensing image is any one of the multi-scale remote sensing images; The pooling weight generation module is used to substitute the average pooling image and the max pooling image into the fully connected layer respectively, and then obtain the average pooling weight and the max pooling weight through the Sigmoid activation function. The channel attention feature generation module is used to substitute the average pooling image into the channel attention branch to obtain channel attention features; A spatial attention feature generation module is used to substitute the max-pooled image into a spatial attention branch to obtain spatial attention features; The feature fusion module is used to fuse the target-scale remote sensing image, the channel attention feature, and the spatial attention feature according to the average pooling weight and the max pooling weight to obtain a target-scale enhanced remote sensing image; and to obtain a multi-scale enhanced remote sensing image based on the target-scale enhanced remote sensing images at all scales.

[0015] Optionally, the feature fusion module includes: The initial target scale enhancement remote sensing image generation module is used to generate remote sensing images using the formula Obtain the initial target scale-enhanced remote sensing image; in, Enhance the remote sensing image at the initial target scale. The average pooling weight, The channel attention feature, The maximum pooling weight, For the spatial attention features, The target-scale remote sensing image; The target scale enhanced remote sensing image generation module is used to normalize the initial target scale enhanced remote sensing image to obtain the target scale enhanced remote sensing image.

[0016] Optionally, the remote sensing image fusion module includes: The convolutional remote sensing image generation module is used to perform 1×1 convolution, ReLU activation, 3×3 convolution, and ReLU activation operations on remote sensing images and enhanced remote sensing images of the same scale to obtain the first convolutional remote sensing image and the first convolutional enhanced remote sensing image. The convolutional remote sensing image stitching module is used to stitch the first convolutional remote sensing image and the first convolutional enhanced remote sensing image to obtain a stitched remote sensing image; The remote sensing feature extraction module is used to perform average pooling processing on the stitched remote sensing image along the channel dimension to obtain the channel average remote sensing feature; and to perform max pooling processing on the stitched remote sensing image along the channel dimension to obtain the channel maximum remote sensing feature; The weight map generation module is used to concatenate the average remote sensing features of the channel and the maximum remote sensing features of the channel, perform a 7×7 convolution, and obtain a weight map by activating it with Sigmoid. The fused image generation module is used to fuse remote sensing images and enhanced remote sensing images of the same scale according to the weight map to obtain high-scale fused images, medium-scale fused images and low-scale fused images.

[0017] The beneficial effects of this invention are: This invention proposes a remote sensing mapping data augmentation method for real estate surveying. First, the initial remote sensing image is preprocessed, then multi-scale sampling is performed. An augmentation model is used to obtain multi-scale enhanced images. Next, the original image and the enhanced image at the same scale are fused. Finally, the fused images at all scales are summarized to obtain the final enhanced image. This method can finely handle illumination issues, avoid information loss and distortion, ensure geometric accuracy, make subsequent visual algorithms more stable, reduce missed detections and false detections, and improve the accuracy and efficiency of real estate surveying. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 A flowchart of a remote sensing mapping data enhancement method for real estate surveying provided in an embodiment of the present invention; Figure 2 A schematic diagram of the framework of an enhancement model provided in an embodiment of the present invention; Figure 3 This is a framework diagram of a remote sensing mapping data enhancement system for real estate surveying, provided as an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention provides a method for enhancing remote sensing mapping data used in real estate surveying. See also... Figure 1 , Figure 1 A flowchart illustrating a remote sensing mapping data enhancement method for real estate surveying, provided as an embodiment of the present invention. The method includes the following steps: S101, acquire the initial remote sensing image, and preprocess the initial remote sensing image to obtain the preprocessed remote sensing image; S102, Multi-scale sampling is performed on the preprocessed remote sensing image to obtain a multi-scale remote sensing image; S103, Substitute the multi-scale remote sensing image into the enhancement model to obtain the multi-scale enhanced remote sensing image; S104, Substitute each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model to obtain high-scale fused image, medium-scale fused image and low-scale fused image; S105, fusing high-scale fused images, medium-scale fused images and low-scale fused images to obtain enhanced remote sensing images; Among them, multi-scale remote sensing images include high-scale remote sensing images, meso-scale remote sensing images, and low-scale remote sensing images; multi-scale enhanced remote sensing images include high-scale enhanced remote sensing images, meso-scale enhanced remote sensing images, and low-scale enhanced remote sensing images. Based on the remote sensing mapping data enhancement method for real estate surveying provided in this embodiment of the invention, through the synergistic effect of preprocessing, multi-scale sampling, enhancement and fusion, it can significantly improve the detail clarity, contrast and overall visual quality of the image under poor lighting conditions while preserving the geometric accuracy and spectral authenticity of the original image, thereby improving the accuracy and efficiency of real estate surveying.

[0023] In one implementation, high-scale images focus on local details, such as building textures and road markings; this is key to recovering subtle features lost in shadows or blur. Low-scale images preserve global structure and contextual information, such as the layout of an entire block and the outline of large plots; this helps maintain the overall contrast and lighting consistency of the image. Medium-scale images strike a balance between detail and the global picture, capturing the complete outline of individual buildings, etc.

[0024] In one implementation, the enhanced remote sensing image is obtained by fusing high-scale fused images, mesoscale fused images, and low-scale fused images by upsampling the mesoscale fused images and low-scale fused images to make their scale the same as that of the high-scale fused image, and then averaging all the images.

[0025] In one implementation, the enhanced image may introduce subtle geometric deformations or spectral shifts. By fusing it with the original image, the irreplaceable geometric and spectral fidelity of the original image can be used to anchor and correct the enhancement result.

[0026] In one embodiment, preprocessing the initial remote sensing image to obtain a preprocessed remote sensing image includes: The first convolutional feature is obtained by performing a 3×3 convolution operation on the initial remote sensing image; Substitute the first convolutional feature into the attention block to obtain the first attention feature; The first convolutional feature and the first attention feature are concatenated and then subjected to a 1×1 convolution to obtain the first concatenated feature; The second convolutional feature is obtained by performing a two-dimensional fast Fourier transform on the first concatenated feature and then performing two 1×1 convolutions. The third convolutional feature is obtained by performing a two-dimensional inverse fast Fourier transform on the second convolutional feature; The first concatenated feature is concatenated with the third concatenated feature after performing two 3×3 convolution operations on the first concatenated feature; Substitute the second splicing feature into the channel attention layer to obtain the second attention feature; The initial remote sensing image and the second attention feature are fused to obtain the preprocessed remote sensing image.

[0027] In one implementation, the information value of dark and bright areas is different in an image with uneven lighting; the attention mechanism enables the model to learn to ignore areas with redundant information or a lot of noise, and concentrate computational resources on areas that are key to real estate mapping; the attention block is the attention module in the Dual Attention Network (DANet).

[0028] In one implementation, concatenating the first convolutional feature with the first attention feature and then performing a 1×1 convolution can combine the original convolutional feature with the attention-weighted feature, combining the original information and importance assessment, so that the feature map contains richer semantics; the subsequent 1×1 convolution is used to fuse these channels and adjust the dimensions, which integrates information and controls computational complexity.

[0029] In one implementation, the image is converted from the spatial domain to the frequency domain. In the frequency domain, the overall illumination and slowly changing shadows (low-frequency information) of the image are separated from the sharp edges and fine textures (high-frequency information). Convolution operations in the frequency domain can process these different frequency components very precisely, suppressing low-frequency interference caused by uneven illumination while enhancing the high-frequency components representing the boundaries of ground features. The processed frequency information is then restored to a spatial image, thereby correcting the illumination and contrast at a global level, preparing for subsequent remote sensing image enhancement.

[0030] In one implementation, two 3×3 convolutions further extract deeper and more complex features in the spatial domain; the third convolution provides features that have been globally optimized in the frequency domain; concatenating the two is equivalent to combining the deep local features in the spatial domain and the global structural features in the frequency domain, thus achieving information complementarity.

[0031] In one implementation, the initial remote sensing image and the second attention feature are fused by averaging. The processed features are directly fused with the original image, ensuring that the preprocessed image does not lose the most fundamental geometric information and spectral authenticity of the original image. At the same time, the entire preprocessing module only needs to learn the parts of the image that need to be enhanced and corrected, which greatly reduces the learning difficulty of the model and improves training efficiency and stability.

[0032] In one embodiment, see Figure 2 , Figure 2 This is a schematic diagram of an enhancement model provided in an embodiment of the present invention. Substituting multi-scale remote sensing images into the enhancement model to obtain multi-scale enhanced remote sensing images includes: Average pooling and max pooling are performed on the target-scale remote sensing image to obtain the average pooling image and the max pooling image; the target-scale remote sensing image can be any one of the multi-scale remote sensing images; After substituting the average pooling image and the max pooling image into the fully connected layer respectively, the average pooling weight and the max pooling weight are obtained by using the Sigmoid activation function. Substituting the average pooled image into the channel attention branch yields the channel attention features; Substituting the max-pooled image into the spatial attention branch yields spatial attention features; The target-scale enhanced remote sensing image is obtained by fusing the target-scale remote sensing image, channel attention features, and spatial attention features based on the average pooling weight and the max pooling weight; and a multi-scale enhanced remote sensing image is obtained by fusing the target-scale enhanced remote sensing image at all scales.

[0033] In one implementation, average pooling and max pooling are performed simultaneously on the target-scale image to capture two different and complementary global information: average pooling, which obtains the average value of the entire feature map, represents the overall context of the image and the statistical information of the uniformly illuminated area, and helps to stabilize the enhancement effect; max pooling, which captures the most significant and active features in the feature map, represents the sharpest edges, the brightest points or the most unique textures in the image, and helps to highlight key details; the combined use provides a more comprehensive information basis for subsequent attention enhancement.

[0034] In one implementation, the result of double pooling is passed through a fully connected layer and a sigmoid function to generate learnable weights. Max pooling weights The average pooling weights approach no longer simply adds or concatenates the features of the two paths, but instead allows the model to automatically determine which path's enhancement information should be relied upon more based on the specific content of the current input image.

[0035] In one implementation, the channel attention branch and the spatial attention branch are two components of the convolutional block attention module (CBAM), which is in the prior art.

[0036] In one implementation, the core advantage of average pooling is that it preserves global statistical information of the image, while the role of channel attention is to filter and enhance key channels (such as channels that distinguish buildings from vegetation in remote sensing images). Based on global statistical information, the importance of each channel to the mapping task is judged, avoiding channel misjudgment caused by relying on local information, and ensuring that information on key land feature attributes (such as building materials and land cover types) is not lost and is more prominent.

[0037] In one implementation, the core advantage of max pooling is to highlight local salient features of the image, while the role of spatial attention is to locate and repair local defect areas. The combination of the two allows spatial attention to be precisely focused on the key spatial locations captured by max pooling, to optimize the clarity of details in a targeted manner, and to avoid edge blurring or artifacts caused by global spatial adjustment, which meets the requirements of geometric accuracy for real estate surveying.

[0038] In one embodiment, fusing the target-scale remote sensing image, channel attention features, and spatial attention features based on average pooling weights and max pooling weights to obtain the target-scale enhanced remote sensing image includes: Through formula Obtain the initial target scale-enhanced remote sensing image; in, Enhance remote sensing images to the initial target scale. For average pooling weights, For channel attention features, For maximum pooling weights, Spatial attention features For target-scale remote sensing images; The target-scale enhanced remote sensing image is obtained by normalizing the initial target-scale enhanced remote sensing image.

[0039] In one implementation, channel attention features and spatial attention features are fused with the original target-scale image according to weights based on the average pooling weights and max pooling weights learned dynamically, avoiding single feature dominance or interference between features; normalization can adjust the pixel values ​​and feature distribution of the initial fused image to a uniform and reasonable range, avoiding overexposure, underexposure or pixel value overflow caused by excessive differences in the values ​​of multiple features, and ensuring stable visual effect of the output target-scale enhanced image.

[0040] In one embodiment, substituting each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model to obtain a high-scale fused image, a medium-scale fused image, and a low-scale fused image includes: Perform 1×1 convolution, ReLU activation, 3×3 convolution, and ReLU activation operations on remote sensing images and enhanced remote sensing images of the same scale to obtain the first convolutional remote sensing image and the first convolutional enhanced remote sensing image; The first convolutional remote sensing image and the first convolutional enhanced remote sensing image are stitched together to obtain a stitched remote sensing image; Average pooling is performed along the channel dimension of the stitched remote sensing image to obtain the channel average remote sensing features; max pooling is performed along the channel dimension of the stitched remote sensing image to obtain the channel maximum remote sensing features; The average remote sensing features and the maximum remote sensing features of each channel are concatenated and then subjected to a 7×7 convolution, followed by Sigmoid activation to obtain a weight map. Based on the weighted map, remote sensing images and enhanced remote sensing images of the same scale are fused to obtain high-scale fused images, medium-scale fused images, and low-scale fused images.

[0041] In one implementation, 1×1 convolution, 3×3 convolution, and ReLU activation are performed on the original remote sensing image and the enhanced remote sensing image, respectively. This can filter out noise and redundant information in the two types of images, highlight key features (building edges, ground feature outlines), avoid the introduction of noise artifacts from the original image into the result during direct fusion, provide a high-quality feature base for subsequent stitching and weighting, and ensure the purity of the fused image.

[0042] In one implementation, the stitched image is subjected to average pooling (preserving the global channel distribution, corresponding to the spatial accuracy advantage of the original image) and max pooling (highlighting local salient features, corresponding to the enhancement of image detail quality) along the channel dimension. Then, it is subjected to a 7×7 large convolutional kernel to capture a wider range of contextual information. Finally, a dynamic weight map is generated through Sigmoid. This weight map can automatically allocate weights according to the characteristics of local regions of the image, replacing fixed-ratio fusion and solving the problem of difficulty in balancing enhancement effect and mapping accuracy.

[0043] In one implementation, the fusion of remote sensing images and enhanced remote sensing images of the same scale based on the weight map is explained by the formula R=a*A+(1-a)B, where R is the fused image (high-scale fused image, medium-scale fused image, and low-scale fused image), a is the weight map, A is the enhanced remote sensing image, and B is the remote sensing image.

[0044] In one implementation, the fusion result avoids both missed and false detections in subsequent algorithms caused by poor illumination of the original image, and prevents the damage to surveying accuracy caused by spectral distortion and edge artifacts in the enhanced image, thereby improving the accuracy and efficiency of the final real estate surveying.

[0045] Based on the same inventive concept, this invention also provides a remote sensing mapping data enhancement system for real estate surveying. See also Figure 3 , Figure 3 A framework diagram of a remote sensing mapping data augmentation system for real estate surveying provided in an embodiment of the present invention includes: The preprocessing module is used to acquire the initial remote sensing image and preprocess it to obtain the preprocessed remote sensing image. The multi-scale sampling module is used to perform multi-scale sampling on the preprocessed remote sensing image to obtain multi-scale remote sensing images; the multi-scale remote sensing images include high-scale remote sensing images, medium-scale remote sensing images, and low-scale remote sensing images. The remote sensing image enhancement module is used to input multi-scale remote sensing images into the enhancement model to obtain multi-scale enhanced remote sensing images; the multi-scale enhanced remote sensing images include high-scale enhanced remote sensing images, meso-scale enhanced remote sensing images, and low-scale enhanced remote sensing images; The remote sensing image fusion module is used to input each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model to obtain high-scale fused images, medium-scale fused images and low-scale fused images; The multi-scale image fusion module is used to fuse high-scale, medium-scale, and low-scale fused images to obtain enhanced remote sensing images.

[0046] Based on the remote sensing mapping data enhancement system for real estate surveying provided in this embodiment of the invention, through the synergistic effect of preprocessing, multi-scale sampling, enhancement and fusion, it can significantly improve the detail clarity, contrast and overall visual quality of the image under poor lighting conditions while preserving the geometric accuracy and spectral authenticity of the original image, thereby improving the accuracy and efficiency of real estate surveying.

[0047] In one embodiment, the preprocessing module includes: The first convolutional feature generation module is used to perform a 3×3 convolution operation on the initial remote sensing image to obtain the first convolutional feature; The first attention feature generation module is used to substitute the first convolutional features into the attention block to obtain the first attention features; The first concatenation feature generation module is used to concatenate the first convolutional feature and the first attention feature and then perform a 1×1 convolution to obtain the first concatenation feature; The second convolutional feature generation module is used to perform a two-dimensional fast Fourier transform on the first concatenated feature and then perform two 1×1 convolutions to obtain the second convolutional feature. The third convolutional feature generation module is used to perform a two-dimensional fast Fourier inverse transform on the second convolutional features to obtain the third convolutional features. The second splicing feature generation module is used to perform two 3×3 convolution operations on the first splicing feature and then splice it with the third convolution feature to obtain the second splicing feature. The second attention feature generation module is used to substitute the second splicing feature into the channel attention layer to obtain the second attention feature; The preprocessed remote sensing image generation module is used to fuse the initial remote sensing image and the second attention feature to obtain the preprocessed remote sensing image.

[0048] In one embodiment, the remote sensing image enhancement module includes: The pooling processing module is used to perform average pooling and max pooling processing on the target-scale remote sensing image to obtain the average pooling image and the max pooling image; the target-scale remote sensing image can be any one of the multi-scale remote sensing images; The pooling weight generation module is used to substitute the average pooling image and the max pooling image into the fully connected layer respectively, and then obtain the average pooling weight and the max pooling weight through the Sigmoid activation function. The channel attention feature generation module is used to substitute the average pooled image into the channel attention branch to obtain channel attention features; The spatial attention feature generation module is used to substitute the max-pooled image into the spatial attention branch to obtain spatial attention features; The feature fusion module is used to fuse the target-scale remote sensing image, channel attention features, and spatial attention features according to the average pooling weight and the max pooling weight to obtain the target-scale enhanced remote sensing image; and to obtain the multi-scale enhanced remote sensing image based on the target-scale enhanced remote sensing images at all scales.

[0049] In one embodiment, the feature fusion module includes: The initial target scale enhancement remote sensing image generation module is used to generate remote sensing images using the formula Obtain the initial target scale-enhanced remote sensing image; in, Enhance remote sensing images to the initial target scale. For average pooling weights, For channel attention features, For maximum pooling weights, Spatial attention features For target-scale remote sensing images; The target scale enhanced remote sensing image generation module is used to normalize the initial target scale enhanced remote sensing image to obtain the target scale enhanced remote sensing image.

[0050] In one embodiment, the remote sensing image fusion module includes: The convolutional remote sensing image generation module is used to perform 1×1 convolution, ReLU activation, 3×3 convolution, and ReLU activation operations on remote sensing images and enhanced remote sensing images of the same scale to obtain the first convolutional remote sensing image and the first convolutional enhanced remote sensing image. The convolutional remote sensing image stitching module is used to stitch together the first convolutional remote sensing image and the first convolutional enhanced remote sensing image to obtain a stitched remote sensing image; The remote sensing feature extraction module is used to perform average pooling processing on the stitched remote sensing image along the channel dimension to obtain the channel average remote sensing feature; and to perform max pooling processing on the stitched remote sensing image along the channel dimension to obtain the channel maximum remote sensing feature. The weight map generation module is used to concatenate the channel average remote sensing features and the channel maximum remote sensing features, perform a 7×7 convolution, and then activate the sigmoid to obtain the weight map. The fused image generation module is used to fuse remote sensing images and enhanced remote sensing images of the same scale according to the weight map to obtain high-scale fused images, medium-scale fused images, and low-scale fused images.

[0051] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for enhancing remote sensing mapping data for real estate surveying, characterized in that, The method includes: Acquire an initial remote sensing image, and preprocess the initial remote sensing image to obtain a preprocessed remote sensing image; The preprocessed remote sensing image is sampled at multiple scales to obtain a multi-scale remote sensing image; the multi-scale remote sensing image includes a high-scale remote sensing image, a medium-scale remote sensing image, and a low-scale remote sensing image; The multi-scale remote sensing images are substituted into the enhancement model to obtain multi-scale enhanced remote sensing images; the multi-scale enhanced remote sensing images include high-scale enhanced remote sensing images, medium-scale enhanced remote sensing images, and low-scale enhanced remote sensing images; Substituting each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model yields high-scale fused images, medium-scale fused images, and low-scale fused images. The high-scale fused image, the medium-scale fused image, and the low-scale fused image are fused to obtain an enhanced remote sensing image.

2. The remote sensing mapping data enhancement method for real estate surveying according to claim 1, characterized in that, Preprocessing the initial remote sensing image to obtain a preprocessed remote sensing image includes: The first convolutional feature is obtained by performing a 3×3 convolution operation on the initial remote sensing image; Substitute the first convolutional feature into the attention block to obtain the first attention feature; The first convolutional feature and the first attention feature are concatenated and then subjected to a 1×1 convolution to obtain the first concatenated feature; The first spliced ​​feature is subjected to a two-dimensional fast Fourier transform and then two 1×1 convolutions to obtain the second convolutional feature; The second convolutional feature is subjected to a two-dimensional inverse fast Fourier transform to obtain the third convolutional feature; The first concatenated feature is concatenated with the third concatenated feature after performing two 3×3 convolution operations on the first concatenated feature; Substitute the second splicing feature into the channel attention layer to obtain the second attention feature; The initial remote sensing image and the second attention feature are fused to obtain a preprocessed remote sensing image.

3. The remote sensing mapping data enhancement method for real estate surveying according to claim 1, characterized in that, Substituting the multi-scale remote sensing images into the enhancement model yields multi-scale enhanced remote sensing images, including: Average pooling and max pooling are performed on the target-scale remote sensing image to obtain the average pooling image and the max pooling image; the target-scale remote sensing image is any one of the multi-scale remote sensing images; After substituting the average pooling image and the max pooling image into the fully connected layer respectively, the average pooling weight and the max pooling weight are obtained by using the Sigmoid activation function. Substituting the average pooled image into the channel attention branch yields the channel attention features; Substituting the max-pooled image into the spatial attention branch yields spatial attention features; The target-scale enhanced remote sensing image is obtained by fusing the target-scale remote sensing image, the channel attention feature, and the spatial attention feature according to the average pooling weight and the max pooling weight; and a multi-scale enhanced remote sensing image is obtained by fusing the target-scale enhanced remote sensing images at all scales.

4. The remote sensing mapping data enhancement method for real estate surveying according to claim 3, characterized in that, The target-scale enhanced remote sensing image is obtained by fusing the target-scale remote sensing image, the channel attention feature, and the spatial attention feature based on the average pooling weight and the max pooling weight, including: Through formula Obtain the initial target scale-enhanced remote sensing image; in, Enhance the remote sensing image at the initial target scale. The average pooling weight, The channel attention feature, The maximum pooling weight, For the spatial attention features, The target-scale remote sensing image; The initial target-scale enhanced remote sensing image is normalized to obtain the target-scale enhanced remote sensing image.

5. A method for enhancing remote sensing mapping data for real estate surveying according to claim 1, characterized in that, Substituting each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model yields high-scale fused images, medium-scale fused images, and low-scale fused images, including: Perform 1×1 convolution, ReLU activation, 3×3 convolution, and ReLU activation operations on remote sensing images and enhanced remote sensing images of the same scale to obtain the first convolutional remote sensing image and the first convolutional enhanced remote sensing image; The first convolutional remote sensing image and the first convolutional enhanced remote sensing image are stitched together to obtain a stitched remote sensing image. The stitched remote sensing image is subjected to average pooling along the channel dimension to obtain the channel average remote sensing features; the stitched remote sensing image is subjected to max pooling along the channel dimension to obtain the channel maximum remote sensing features; The average remote sensing features and the maximum remote sensing features of the channel are concatenated and then subjected to a 7×7 convolution, and a weight map is obtained by activating with Sigmoid. Based on the weighted map, remote sensing images and enhanced remote sensing images of the same scale are fused to obtain high-scale fused images, medium-scale fused images, and low-scale fused images.

6. A remote sensing mapping data augmentation system for real estate surveying, characterized in that, The system includes: The preprocessing module is used to acquire an initial remote sensing image and preprocess the initial remote sensing image to obtain a preprocessed remote sensing image. A multi-scale sampling module is used to perform multi-scale sampling on the preprocessed remote sensing image to obtain a multi-scale remote sensing image; the multi-scale remote sensing image includes a high-scale remote sensing image, a medium-scale remote sensing image, and a low-scale remote sensing image. The remote sensing image enhancement module is used to substitute the multi-scale remote sensing image into the enhancement model to obtain a multi-scale enhanced remote sensing image; the multi-scale enhanced remote sensing image includes a high-scale enhanced remote sensing image, a medium-scale enhanced remote sensing image, and a low-scale enhanced remote sensing image; The remote sensing image fusion module is used to input each remote sensing image and enhanced remote sensing image of the same scale into the image fusion model to obtain high-scale fused images, medium-scale fused images and low-scale fused images; A multi-scale image fusion module is used to fuse the high-scale fused image, the medium-scale fused image, and the low-scale fused image to obtain an enhanced remote sensing image.

7. A remote sensing mapping data enhancement system for real estate surveying according to claim 6, characterized in that, The preprocessing module includes: The first convolutional feature generation module is used to perform a 3×3 convolution operation on the initial remote sensing image to obtain the first convolutional feature. The first attention feature generation module is used to substitute the first convolutional features into the attention block to obtain the first attention features; The first splicing feature generation module is used to splice the first convolutional feature and the first attention feature and then perform a 1×1 convolution to obtain the first splicing feature; The second convolutional feature generation module is used to perform a two-dimensional fast Fourier transform on the first concatenated feature and then perform two 1×1 convolutions to obtain the second convolutional feature. The third convolutional feature generation module is used to perform a two-dimensional fast Fourier inverse transform on the second convolutional feature to obtain the third convolutional feature; The second splicing feature generation module is used to perform two 3×3 convolution operations on the first splicing feature and then splice it with the third convolution feature to obtain the second splicing feature; The second attention feature generation module is used to substitute the second splicing feature into the channel attention layer to obtain the second attention feature; The preprocessed remote sensing image generation module is used to fuse the initial remote sensing image and the second attention feature to obtain a preprocessed remote sensing image.

8. A remote sensing mapping data enhancement system for real estate surveying according to claim 6, characterized in that, The remote sensing image enhancement module includes: The pooling processing module is used to perform average pooling and max pooling processing on the target-scale remote sensing image to obtain the average pooling image and the max pooling image; the target-scale remote sensing image is any one of the multi-scale remote sensing images; The pooling weight generation module is used to substitute the average pooling image and the max pooling image into the fully connected layer respectively, and then obtain the average pooling weight and the max pooling weight through the Sigmoid activation function. The channel attention feature generation module is used to substitute the average pooling image into the channel attention branch to obtain channel attention features; A spatial attention feature generation module is used to substitute the max-pooled image into a spatial attention branch to obtain spatial attention features; The feature fusion module is used to fuse the target-scale remote sensing image, the channel attention feature, and the spatial attention feature according to the average pooling weight and the max pooling weight to obtain a target-scale enhanced remote sensing image; and to obtain a multi-scale enhanced remote sensing image based on the target-scale enhanced remote sensing images at all scales.

9. A remote sensing mapping data enhancement system for real estate surveying according to claim 8, characterized in that, The feature fusion module includes: The initial target scale enhancement remote sensing image generation module is used to generate remote sensing images using the formula Obtain the initial target scale-enhanced remote sensing image; in, Enhance the remote sensing image at the initial target scale. The average pooling weight, The channel attention feature, The maximum pooling weight, For the spatial attention features, The target-scale remote sensing image; The target scale enhanced remote sensing image generation module is used to normalize the initial target scale enhanced remote sensing image to obtain the target scale enhanced remote sensing image.

10. A remote sensing mapping data enhancement system for real estate surveying according to claim 6, characterized in that, The remote sensing image fusion module includes: The convolutional remote sensing image generation module is used to perform 1×1 convolution, ReLU activation, 3×3 convolution, and ReLU activation operations on remote sensing images and enhanced remote sensing images of the same scale to obtain the first convolutional remote sensing image and the first convolutional enhanced remote sensing image. The convolutional remote sensing image stitching module is used to stitch the first convolutional remote sensing image and the first convolutional enhanced remote sensing image to obtain a stitched remote sensing image; The remote sensing feature extraction module is used to perform average pooling processing on the stitched remote sensing image along the channel dimension to obtain the channel average remote sensing feature; and to perform max pooling processing on the stitched remote sensing image along the channel dimension to obtain the channel maximum remote sensing feature; The weight map generation module is used to concatenate the average remote sensing features of the channel and the maximum remote sensing features of the channel, perform a 7×7 convolution, and obtain a weight map by activating it with Sigmoid. The fused image generation module is used to fuse remote sensing images and enhanced remote sensing images of the same scale according to the weight map to obtain high-scale fused images, medium-scale fused images and low-scale fused images.

Images