An urban functional area automatic extraction and classification method, device and electronic equipment

Abstract

The present application provides a kind of urban functional area automatic extraction and classification method, device and electronic equipment, wherein the urban functional area automatic extraction and classification method comprises: obtaining image data of target area, and the image data is preprocessed;The image data includes RGB remote sensing image and night light remote sensing image;Interest point data in the target area is obtained, and space distribution heat map is generated based on the interest point data;Call pre-constructed and trained multi-input head instance segmentation network, and obtain the functional area boundary and category information of the target area by the image data and the space distribution heat map;The accuracy of the functional area boundary and category information of the target area is evaluated, and the urban functional area data of the target area is integrated to generate.The present application improves overall accuracy and processing efficiency, and solves the problem of poor classification effect caused by single data source in the prior art.

Description

Technical Field

[0001] This invention relates to the field of urban planning technology, and in particular to a method, apparatus and electronic device for automatic extraction and classification of urban functional areas. Background Technology

[0002] Accurate identification and delineation of urban functional zones are of great significance for urban planning, land use management, and urban development decision-making. The delineation of urban functional zones reflects the spatial structure and land use characteristics of the city, providing important decision-making basis for urban managers. Traditional methods of urban functional zone delineation mainly rely on manual field surveys and experience-based judgment, which suffers from low efficiency, strong subjectivity, and long update cycles. With the development of remote sensing technology and the accumulation of urban big data, it has become possible to automatically identify and extract urban functional zones using multi-source data.

[0003] Existing methods for identifying urban functional zones mainly suffer from the following problems: First, traditional two-stage methods require region segmentation before functional classification, resulting in complex processing procedures, severe error accumulation, and difficulty in ensuring consistency between segmentation boundaries and functional classifications. Second, existing methods primarily rely on a single remote sensing image data source, failing to fully utilize the complementary advantages of multi-source heterogeneous urban data, leading to insufficient accuracy and robustness in functional zone identification. Third, the lack of an end-to-end deep learning framework makes it impossible to simultaneously optimize segmentation accuracy and classification accuracy, affecting the overall performance of urban functional zone extraction.

[0004] There is currently no effective solution to the problem of poor classification results caused by the single data source in existing related technologies. Summary of the Invention

[0005] This invention provides a method, apparatus, and electronic device for automatic extraction and classification of urban functional areas, which solves the problem of poor classification effect caused by the single data source in existing related technologies, and realizes end-to-end processing of functional area boundary extraction and classification.

[0006] In a first aspect, the present invention provides a method for automatically extracting and classifying urban functional areas, comprising: Acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images; Acquire point-of-interest (POI) data within the target area, and generate a spatial distribution heatmap based on the POI data; The pre-built and trained multi-input head instance segmentation network is invoked, and the functional area boundaries and category information of the target region are obtained through the image data and the spatial distribution heatmap; The accuracy of the functional zone boundaries and category information of the target area is evaluated, and the data of urban functional zones of the target area is integrated to generate such data.

[0007] According to a method for automatic extraction and classification of urban functional areas provided by the present invention, image data of a target area is acquired, and the image data is preprocessed, including: Radiometric correction, geometric correction, and image enhancement processing are performed on the RGB remote sensing images with a resolution higher than the preset standard. The nighttime light remote sensing images are subjected to noise removal, outlier detection, and spatial resampling.

[0008] According to a method for automatic extraction and classification of urban functional areas provided by the present invention, the method acquires point-of-interest (POI) data within the target area and generates a spatial distribution heatmap based on the POI data, comprising: Obtain the coordinates and category information of points of interest within the target area; The spatial distribution heatmap is generated using a kernel density estimation method; the bandwidth parameter of the kernel density estimation method is adjusted according to the spatial distribution density of the points of interest.

[0009] According to the present invention, an automatic extraction and classification method for urban functional areas is provided, wherein the multi-input head instance segmentation network includes a multi-input head feature extraction module, a multi-modal feature fusion module, and an enhanced classification head module; The multi-input head feature extraction module is used to extract data features from the image data and the spatial distribution heatmap; The multimodal feature fusion module is used to fuse the data features of the image data and the spatial distribution heatmap to generate multimodal features; The enhanced classification head module is used to predict the functional area category of the target region.

[0010] According to the present invention, an automatic extraction and classification method for urban functional areas is provided, wherein the multi-input head feature extraction module includes: RGB input head: The visual features of the RGB remote sensing images are extracted using a ResNet50 backbone network; Interest point heatmap input header: A lightweight convolutional neural network is used to extract the interest point distribution features of the spatial distribution heatmap; Nighttime light input head: A single-branch convolutional neural network is used to extract nighttime activity features from the nighttime light remote sensing image.

[0011] According to the present invention, an automatic extraction and classification method for urban functional areas is provided, which fuses the data features of the image data and the spatial distribution heatmap to generate multimodal features, including: The weights of the visual features, the point of interest distribution features, and the nighttime activity features are determined respectively; The visual features, interest point distribution features, and nighttime activity features are weighted and fused based on an attention mechanism, and residual connection processing is performed to generate the multimodal features.

[0012] According to the present invention, an automatic extraction and classification method for urban functional areas is provided, which trains the multi-input head instance segmentation network, including: Based on a multi-task learning strategy, the loss function of the multi-input head instance segmentation network is determined; The input head instance segmentation network is trained using a transfer learning method combined with the Adam optimizer and a learning rate decay strategy.

[0013] According to the present invention, an automatic extraction and classification method for urban functional zones is provided, which obtains the functional zone boundaries and category information of the target area through the image data and the spatial distribution heat map, including: The image data and the spatial distribution heatmap are input into the multi-input head instance segmentation network to obtain prediction results; the prediction results include the functional area boundaries and category information of the target region. The nonmaximum suppression algorithm is used to remove overlapping data from the prediction results; Filter out segmented regions in the prediction results whose area is lower than a preset area threshold; The prediction results are then subjected to boundary smoothing.

[0014] Secondly, the present invention also provides an automatic extraction and classification device for urban functional areas, comprising: The acquisition module is used to acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images. The processing module is used to acquire point-of-interest (POI) data within the target area and generate a spatial distribution heatmap based on the POI data. The prediction module is used to call a pre-built and trained multi-input head instance segmentation network and obtain the functional area boundaries and category information of the target area through the image data and the spatial distribution heatmap; The integration module is used to perform accuracy assessment on the functional zone boundaries and category information of the target area, and to integrate and generate urban functional zone data of the target area.

[0015] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the automatic extraction and classification method for urban functional areas as described in the first aspect above.

[0016] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for automatic extraction and classification of urban functional areas as described in the first aspect above.

[0017] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method for automatic extraction and classification of urban functional areas as described in the first aspect above.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The automatic extraction and classification method for urban functional areas provided by this invention can integrate three input data sources through a multi-input head instance segmentation network. It utilizes the complementarity of visual features, facility distribution features, and human activity features, which significantly improves the accuracy and robustness of functional area identification. It realizes end-to-end processing of functional area boundary extraction and classification, avoids the error accumulation problem of traditional two-stage methods, improves overall accuracy and processing efficiency, and solves the problem of poor classification effect caused by a single data source in existing related technologies. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the method for automatic extraction and classification of urban functional areas provided by the present invention; Figure 2 This is a schematic diagram illustrating the process of extracting urban functional zones in an embodiment of the present invention; Figure 3 This is a schematic diagram of the extraction results of urban functional areas in an example of the present invention; Figure 4 This is a structural block diagram of the urban functional area automatic extraction and classification device provided by the present invention; Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] This invention provides a method for automatically extracting and classifying urban functional areas. Figure 1 This is a flowchart of the automatic extraction and classification method for urban functional areas provided by the present invention, such as... Figure 1 As shown, the method includes the following steps: Step S101: Acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images. Step S102: Obtain point of interest data within the target area and generate a spatial distribution heatmap based on the point of interest data; Step S103: Invoke the pre-built and trained multi-input head instance segmentation network, and obtain the functional area boundaries and category information of the target area through image data and spatial distribution heatmap; Step S104: Accuracy assessment of the functional zone boundaries and category information of the target area is performed, and urban functional zone data of the target area is generated by integration.

[0023] In this method, firstly, RGB remote sensing images and nighttime light remote sensing images of the target area are acquired, and the acquired image data is preprocessed to improve data quality, thereby improving the accuracy of subsequent area identification and classification. Then, Point of Interest (POI) data within the target area is acquired, and a spatial distribution heatmap of POIs is generated. Next, the RGB remote sensing images, nighttime light remote sensing images, and spatial distribution heatmap are input into a multi-input head instance segmentation network to fully acquire the functional zone boundaries and category information of the target area. Finally, the accuracy of the functional zone boundaries and category information is evaluated, and the data is integrated to generate urban functional zone data, achieving the extraction and classification of urban functional zones. The functional zone classification system includes five categories: residential areas, entertainment areas, transportation areas, industrial areas, and office areas, each with clear discrimination criteria and feature descriptions. In the above process, the multi-input head instance segmentation network can integrate three input data sources, and utilize the complementarity of visual features, facility distribution features and human activity features to significantly improve the accuracy and robustness of functional area recognition. It realizes end-to-end processing of functional area boundary extraction and classification, avoids the error accumulation problem of traditional two-stage methods, improves overall accuracy and processing efficiency, and solves the problem of poor classification effect caused by a single data source in existing related technologies.

[0024] Figure 2 This is a schematic diagram illustrating the process of extracting urban functional areas in an embodiment of the present invention, such as... Figure 2 As shown, in some embodiments, step S101 involves acquiring image data of the target area and preprocessing the image data, including: performing radiometric correction, geometric correction, and image enhancement processing on RGB remote sensing images with resolutions higher than a preset standard; and performing noise removal, outlier detection, and spatial resampling processing on nighttime light remote sensing images.

[0025] For example, radiometric correction, geometric correction, and image enhancement are performed on high-resolution RGB remote sensing images above 3m to ensure that the image quality meets the input requirements of deep learning models. Noise removal, outlier detection, and spatial resampling are performed on nighttime light remote sensing image data to make it have the same spatial resolution and coverage as the RGB remote sensing images.

[0026] In some embodiments, step S102, acquiring point-of-interest (POI) data within the target area and generating a spatial distribution heatmap based on the POI data, includes: acquiring the coordinate positions and category information of POIs within the target area; generating a spatial distribution heatmap using a kernel density estimation method; and adjusting the bandwidth parameter of the kernel density estimation method according to the spatial distribution density of the POIs.

[0027] For example, data on points of interest within the target area are collected, including coordinate location and category information. A spatial distribution heatmap of the points of interest is generated using the kernel density estimation method, as shown in the following formula:

[0028] in, express The density value of points of interest at the location, Indicates the total number of points of interest. Indicates bandwidth parameter, express Position to number i The Euclidean distance between points of interest The kernel function is represented here. In this embodiment, the spatial distribution heatmap is generated using an adaptive bandwidth selection method, which automatically adjusts the bandwidth parameter of the kernel density estimation based on the spatial distribution density of the points of interest, ensuring that the spatial distribution heatmap accurately reflects the spatial clustering characteristics of the points of interest.

[0029] In some embodiments, the multi-input head instance segmentation network based on the improved Mask R-CNN includes a multi-input head feature extraction module, a multi-modal feature fusion module, and an enhanced classification head module; the multi-input head feature extraction module is used to extract data features from image data and spatial distribution heatmaps; the multi-modal feature fusion module is used to fuse the data features from image data and spatial distribution heatmaps to generate multi-modal features; and the enhanced classification head module is used to predict the functional area category of the target region.

[0030] Specifically, the multi-input head feature extraction module includes: RGB input head: using a ResNet50 backbone network to extract visual features from RGB remote sensing images; Interest point heatmap input head: using a lightweight convolutional neural network (CNN) to extract interest point distribution features from the spatial distribution heatmap; Night light input head: using a single-branch convolutional neural network to extract nighttime activity features from night light remote sensing images.

[0031] More specifically, the backbone network of the RGB input head architecture is ResNet50 + Feature Pyramid Networks (FPN); the input scale of the feature maps is 1 / 4, 1 / 8, 1 / 16, 1 / 32, and 1 / 64; the number of feature channels is 256. The interest point heatmap input head architecture consists of a three-layer convolutional network: Conv2d(1, 64, 3, 1, 1) → Conv2d(64, 128, 3, 1, 1) → Conv2d(128, 256, 3, 1, 1); the activation function is the Rectified Linear Unit (ReLU). During pooling, adaptive average pooling is performed to the same size as the visual features of the RGB remote sensing image. The nighttime light input head consists of a two-layer convolutional network: Conv2d(1, 64, 3, 1, 1) → Conv2d(64, 256, 3, 1, 1); the activation function is the Rectified Linear Unit (ReLU). During the upsampling operation, bilinear interpolation is performed to achieve the same size as the visual features of the RGB remote sensing image.

[0032] Based on this, the data features of image data and spatial distribution heatmap are fused to generate multimodal features, including: determining the weights of visual features, point of interest distribution features and nighttime activity features respectively; and performing weighted fusion of visual features, point of interest distribution features and nighttime activity features based on attention mechanism and residual connection processing to generate multimodal features.

[0033] For example, an attention mechanism is used to fuse features from three modalities, with the specific formula as follows:

[0034]

[0035]

[0036]

[0037] in, This represents the attention weight of visual features on the interest distribution features. This represents the attention weight of the interest point distribution features on the nighttime activity features. This represents the attentional weight of nighttime activity features on visual features. Indicates visual characteristics, Indicates the distribution characteristics of points of interest. Indicates characteristics of nighttime activity. and All are learning parameters. Representing feature dimension, This represents multimodal features. During feature fusion, the attention mechanisms employed include channel attention and spatial attention. The channel attention mechanism uses the Squeeze-and-Excitation (SE) module, while the spatial attention mechanism uses the Convolutional Block Attention Module (CBAM).

[0038] In some embodiments, training the multi-input head instance segmentation network includes: determining the loss function of the multi-input head instance segmentation network based on a multi-task learning strategy; and training the input head instance segmentation network using a transfer learning method combined with an Adam optimizer and a learning rate decay strategy.

[0039] For example, using a multi-task learning strategy, the total loss function includes segmentation loss, detection loss, and classification loss, as shown in the following formula:

[0040] in, Represents the loss function. Indicates the loss from partitioning. Indicates detection loss, Represents classification loss. , and All parameters are weights. Transfer learning is used, and the RGB input head is initialized with pre-trained ResNet50 weights. Optimization is performed using the Adam optimizer with an initial learning rate of 0.001. A learning rate decay strategy is employed, decreasing the learning rate by a factor of 0.1 every 50 epochs.

[0041] In some embodiments, step S103, obtaining the functional area boundaries and category information of the target area through image data and spatial distribution heatmap, includes: inputting the image data and spatial distribution heatmap into a multi-input head instance segmentation network to obtain prediction results; the prediction results include the functional area boundaries and category information of the target area; using a non-maximum suppression algorithm to remove overlapping data in the prediction results; filtering segmentation regions in the prediction results whose area is lower than a preset area threshold; and performing boundary smoothing processing on the prediction results.

[0042] In this embodiment, based on the original Mask R-CNN classification head, a multimodal feature input is added, and a fully connected layer is used to predict the functional region category. The specific formula is as follows:

[0043] in, Indicates the predicted probability of functional area categories. This indicates a fully connected layer operation. This represents multimodal features.

[0044] Then, the non-maximum suppression (NMS) algorithm is used to remove overlapping prediction results, and excessively small segmentation regions are filtered based on area thresholds. Finally, boundary smoothing is performed to improve the visual quality of the prediction results.

[0045] Based on this, the accuracy of the prediction results is evaluated. Specifically, the accuracy evaluation metrics include segmentation accuracy and classification accuracy. Segmentation accuracy is evaluated using IoU (Intersection over Union) and pixel accuracy, while classification accuracy is evaluated using overall accuracy, Kappa coefficient, and F1 score. After the evaluation, vectorized urban functional area data containing spatial boundaries and functional category attributes are output.

[0046] The following specific examples illustrate the above method: 1. Based on Google imagery data with a resolution of 1 meter, perform preprocessing operations such as image enhancement, noise reduction, and spatial registration.

[0047] 2. Collect Gaode POI data covering the study area and generate a spatial distribution heat map of points of interest.

[0048] 3. Using the 130-meter resolution Luojia-1 nighttime light remote sensing data, noise removal, outlier detection, and spatial resampling were performed to make it have the same spatial resolution and coverage as the RGB image.

[0049] 4. 500 functional area samples were labeled, including 5 functional categories, for training the multi-input head instance segmentation network, and the trained network model was saved.

[0050] 5. Input the data from the study area into the network described above, and extract all functional area instances within the study area, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the extraction results of urban functional areas in an example of the present invention.

[0051] In summary, this method, through a multi-input instance segmentation network, achieves end-to-end processing of functional area boundary extraction and classification, avoiding the error accumulation problem of traditional two-stage methods and improving overall accuracy and processing efficiency. This method integrates RGB remote sensing imagery, POI heatmaps, and nighttime light data. Figure 3 This method utilizes multiple data sources, fully leveraging the complementarity of visual features, facility distribution features, and human activity features, significantly improving the accuracy and robustness of functional zone identification. Furthermore, the method employs a specialized multi-input head architecture and feature fusion mechanism, effectively integrating spatial information from different modalities and providing a new technical approach for urban functional zone extraction. In addition, this method is highly automated, possesses good scalability and adaptability, can be rapidly updated, and can be applied to cities of different sizes and types, making it suitable for large-scale dynamic monitoring of urban functional zones.

[0052] The present invention also provides an automatic extraction and classification device for urban functional areas. The automatic extraction and classification device for urban functional areas provided by the present invention is described below. The automatic extraction and classification device for urban functional areas described below can be referred to in correspondence with the automatic extraction and classification method for urban functional areas described above. Figure 4 This is a structural block diagram of the urban functional area automatic extraction and classification device provided by the present invention, as shown below. Figure 4 As shown, the device includes: The acquisition module 401 is used to acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images. Processing module 402 is used to acquire point-of-interest (POI) data within the target area and generate a spatial distribution heatmap based on the POI data; Prediction module 403 is used to call a pre-built and trained multi-input head instance segmentation network and obtain the functional area boundaries and category information of the target area through image data and spatial distribution heatmaps; The integration module 404 is used to perform accuracy assessment on the functional zone boundaries and category information of the target area, and to integrate and generate urban functional zone data of the target area.

[0053] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include: a processor 501, a communication interface 502, a memory 503, and a communication bus 504. The processor 501, communication interface 502, and memory 503 communicate with each other via the communication bus 504. The processor 501 can call logical instructions from the memory 503 to execute an automatic extraction and classification method for urban functional areas. This method includes: Acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images; Acquire point-of-interest (POI) data within the target area and generate a spatial distribution heatmap based on the POI data; The pre-built and trained multi-input head instance segmentation network is invoked, and the functional area boundaries and category information of the target area are obtained through image data and spatial distribution heatmaps; The accuracy of the functional zone boundaries and category information of the target area is evaluated, and the urban functional zone data of the target area is generated by integration.

[0054] Furthermore, the logical instructions in the aforementioned memory 503 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0055] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the urban functional area automatic extraction and classification method provided by the above methods, the method including: Acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images; Acquire point-of-interest (POI) data within the target area and generate a spatial distribution heatmap based on the POI data; The pre-built and trained multi-input head instance segmentation network is invoked, and the functional area boundaries and category information of the target area are obtained through image data and spatial distribution heatmaps; The accuracy of the functional zone boundaries and category information of the target area is evaluated, and the urban functional zone data of the target area is generated by integration.

[0056] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the automatic extraction and classification method for urban functional areas provided by the methods described above, the method comprising: Acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images; Acquire point-of-interest (POI) data within the target area and generate a spatial distribution heatmap based on the POI data; The pre-built and trained multi-input head instance segmentation network is invoked, and the functional area boundaries and category information of the target area are obtained through image data and spatial distribution heatmaps; The accuracy of the functional zone boundaries and category information of the target area is evaluated, and the urban functional zone data of the target area is generated by integration.

[0057] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0058] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically extracting and classifying urban functional zones, characterized in that, include: Acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images; Acquire point-of-interest (POI) data within the target area, and generate a spatial distribution heatmap based on the POI data; The pre-built and trained multi-input head instance segmentation network is invoked, and the functional area boundaries and category information of the target area are obtained through the image data and the spatial distribution heatmap; The accuracy of the functional zone boundaries and category information of the target area is evaluated, and the urban functional zone data of the target area is generated by integration. The multi-input head instance segmentation network includes a multi-input head feature extraction module, a multi-modal feature fusion module, and an enhanced classification head module; The multi-input head feature extraction module is used to extract data features from the image data and the spatial distribution heatmap; The multimodal feature fusion module is used to fuse the data features of the image data and the spatial distribution heatmap to generate multimodal features; The enhanced classification head module is used to predict the functional area category of the target region. The multi-input head feature extraction module includes: RGB input head: The visual features of the RGB remote sensing images are extracted using a ResNet50 backbone network; the backbone network of the RGB input head architecture is ResNet50 + feature pyramid network; Interest point heatmap input head: A lightweight convolutional neural network is used to extract the interest point distribution features of the spatial distribution heatmap; the interest point heatmap input head architecture includes a three-layer convolutional network, and the activation function is the linear rectified unit activation function; Nighttime light input head: A single-branch convolutional neural network is used to extract nighttime activity features from the nighttime light remote sensing image; the nighttime light input head includes two layers of convolutional network, and the activation function is the linear correction unit activation function; The data features of the image data and the spatial distribution heatmap are fused to generate multimodal features, including: The weights of the visual features, the point of interest distribution features, and the nighttime activity features are determined respectively; The visual features, interest point distribution features, and nighttime activity features are weighted and fused based on an attention mechanism, and residual connection processing is performed to generate the multimodal features. An attention mechanism is used to fuse features from three modalities, as shown in the following formula: in, This represents the attention weight of visual features on the interest distribution features. This represents the attention weight of the interest point distribution features on the nighttime activity features. This represents the attentional weight of nighttime activity features on visual features. Indicates visual characteristics, Indicates the distribution characteristics of points of interest. Indicates characteristics of nighttime activity. and All are learning parameters. Representing feature dimension, It represents multimodal features; when performing feature fusion, the attention mechanisms used include channel attention mechanism and spatial attention mechanism. The channel attention mechanism uses the SE module, and the spatial attention mechanism uses the convolutional block attention module.

2. The method for automatic extraction and classification of urban functional areas according to claim 1, characterized in that, Acquire image data of the target area and preprocess the image data, including: Radiometric correction, geometric correction, and image enhancement processing are performed on the RGB remote sensing images with a resolution higher than the preset standard. The nighttime light remote sensing images are subjected to noise removal, outlier detection, and spatial resampling.

3. The method for automatic extraction and classification of urban functional areas according to claim 1, characterized in that, Acquiring point-of-interest (POI) data within the target area and generating a spatial distribution heatmap based on the POI data, including: Obtain the coordinates and category information of points of interest within the target area; The spatial distribution heatmap is generated using a kernel density estimation method; the bandwidth parameter of the kernel density estimation method is adjusted according to the spatial distribution density of the points of interest.

4. The method for automatic extraction and classification of urban functional areas according to claim 1, characterized in that, Training the multi-input head instance segmentation network includes: Based on a multi-task learning strategy, the loss function of the multi-input head instance segmentation network is determined; The input head instance segmentation network is trained using a transfer learning method combined with the Adam optimizer and a learning rate decay strategy.

5. The method for automatic extraction and classification of urban functional areas according to claim 1, characterized in that, The functional zone boundaries and category information of the target area are obtained through the image data and the spatial distribution heatmap, including: The image data and the spatial distribution heatmap are input into the multi-input head instance segmentation network to obtain prediction results; the prediction results include the functional area boundaries and category information of the target region. The nonmaximum suppression algorithm is used to remove overlapping data from the prediction results; Filter out segmented regions in the prediction results whose area is lower than a preset area threshold; The prediction results are then subjected to boundary smoothing.

6. An automatic extraction and classification device for urban functional areas, used to implement the automatic extraction and classification method for urban functional areas as described in any one of claims 1-5, characterized in that, include: The acquisition module is used to acquire image data of the target area and preprocess the image data; the image data includes RGB remote sensing images and nighttime light remote sensing images. The processing module is used to acquire point-of-interest (POI) data within the target area and generate a spatial distribution heatmap based on the POI data. The prediction module is used to call a pre-built and trained multi-input head instance segmentation network and obtain the functional area boundaries and category information of the target area through the image data and the spatial distribution heatmap; The integration module is used to perform accuracy assessment on the functional zone boundaries and category information of the target area, and to integrate and generate urban functional zone data of the target area.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for automatic extraction and classification of urban functional areas as described in any one of claims 1 to 5.

Images