A function area mapping method and device, a terminal device, and a storage medium

Abstract

The application discloses a function area mapping method and device, a terminal equipment and a storage medium. The method comprises the following steps: acquiring a region image of a region to be mapped; inputting the region image into a function area identification model, so that the function area identification model identifies each function area category in the region image; identifying each function area category in the region image after the identification, and generating a function area image corresponding to the region to be mapped; wherein the construction of the function area identification model comprises the following steps: acquiring region sample images of each function sub-region of the region to be mapped; taking each region sample image as an input and taking a function area category corresponding to each region sample image as an output, training a preset neural network model, and generating the function area identification model. Through the implementation of the application, the mapping efficiency and accuracy can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and in particular to a functional area mapping method and device, a terminal device and a storage medium. BACKGROUND

[0002] In urban areas, the same type of land use often has the same demand for land space, location and function, which leads to the aggregation of the same type of activities in urban space, thereby generating various functional areas, such as residential areas, industrial areas, commercial areas, etc. The classification of urban functional areas is conducive to the rational layout of the city and has important significance for promoting production, facilitating the life of residents and protecting the urban environment. The traditional method of classifying urban functional areas usually involves manually drawing a distribution map of urban functional areas by technical personnel in the relevant field after questionnaire survey or on-site investigation. This method is time-consuming and labor-intensive, and the standards are not uniform, and it is difficult to map large areas. SUMMARY

[0003] The present application provides a functional area mapping method, device, terminal device and storage medium, which can improve the mapping efficiency and accuracy.

[0004] An embodiment of the present application provides a functional area mapping method, comprising:

[0005] obtaining a region image of a region to be mapped;

[0006] inputting the region image into a functional area recognition model to enable the functional area recognition model to recognize each functional area category in the region image;

[0007] identifying each functional area category in the region image after recognition to generate a functional area image corresponding to the region to be mapped;

[0008] The construction of the functional area recognition model comprises:

[0009] obtaining region sample images of each functional sub-area of the region to be mapped;

[0010] training a preset neural network model by taking each region sample image as input and taking the functional area category corresponding to each region sample image as output to generate the functional area recognition model.

[0011] Further, the region sample image comprises a functional area polygon for identifying the region where the functional area is located.

[0012] The neural network model comprises a multi-scale attention feature extractor, a mask layer and a classifier.

[0013] In the process of training the neural network model, the multi-scale attention feature extractor performs feature extraction on the regional sample image to generate a first feature map;

[0014] The mask layer generates a second feature map according to the first feature map and the functional area map spot of the corresponding regional sample image;

[0015] The classifier identifies the second feature map to determine the functional area category of the regional sample image.

[0016] Further, the regional sample image of each functional subarea of the region to be mapped is obtained, comprising:

[0017] Obtain road network data of the region to be mapped; generate a road network image of the region to be mapped according to the road network data of the region to be mapped;

[0018] Obtain high-resolution images of the region to be mapped and POI data of each functional area in the region to be mapped;

[0019] Generate a heat map of each functional area in the region to be mapped according to the POI data of each functional area;

[0020] Superimpose the heat maps of all functional areas to generate a multi-channel heat map;

[0021] Superimpose the multi-channel heat map and the high-resolution image of the region to be mapped to generate a fusion image of the region to be mapped;

[0022] Segment the fusion image according to the road network image to generate a regional sample image of each functional subarea of the region to be mapped.

[0023] Further, the road network image of the region to be mapped is generated according to the road network data of the region to be mapped, comprising:

[0024] Select roads in the road network data that meet a predetermined level to generate a first road network image;

[0025] For each road in the first road network image, a buffer area is established around the road as the center line, and the buffer areas are fused to generate a second road network image;

[0026] Extract the center line of the fused buffer area in the second road network image, and perform topological error correction on the extracted center line to generate the road network image of the region to be mapped.

[0027] Further, the road network image comprises a plurality of closed subareas;

[0028] The segmentation of the fusion image according to the road network image comprises:

[0029] According to the shape of each closed sub-region in the road network image, a minimum circumscribed square of the corresponding closed sub-region is determined, and a region to be segmented of the corresponding closed sub-region is generated according to the minimum circumscribed square;

[0030] The fusion image is segmented according to the region to be segmented.

[0031] Further, the multi-scale attention feature extractor comprises:

[0032] The multi-scale attention feature extractor comprises five stages, the first stage comprises one convolution layer and one pooling layer, the second stage comprises three multi-scale attention residual blocks, the third stage comprises four multi-scale attention residual blocks, the fourth stage comprises six multi-scale attention residual blocks, and the fifth stage comprises three multi-scale attention residual blocks; wherein each multi-scale attention residual block comprises a hierarchical convolution module and an attention mechanism module.

[0033] Further, the loss function of the functional area recognition model is:

[0034]

[0035] wherein, is the loss function of the functional area recognition model, y i,k and p i,k are respectively the real label and the predicted probability value of the kth class of the ith functional area in the region to be mapped, and C is the total number of functional area classes.

[0036] On the basis of the above method embodiment, the application correspondingly provides a device embodiment;

[0037] An embodiment of the application correspondingly provides a functional area mapping device, comprising:

[0038] a data acquisition module, a functional area classification module, a mapping module and a functional area recognition model construction module;

[0039] The data acquisition module is configured to acquire a region image of a region to be mapped.

[0040] The functional area recognition module is configured to input the region image into a functional area recognition model, so that the functional area recognition model recognizes each functional area class in the region image.

[0041] The mapping module is configured to identify each functional area class in the region image after recognition, and generate a functional area image corresponding to the region to be mapped.

[0042] The function area recognition model construction module is configured to acquire area sample images of each function subarea of a region to be mapped; train a preset neural network model by taking each area sample image as input and taking a function area category corresponding to each area sample image as output, and generate the function area recognition model.

[0043] Another embodiment of the present application provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the functional area mapping method according to the above-mentioned embodiments of the present application when executing the computer program.

[0044] Another embodiment of the present application provides a storage medium, comprising a stored computer program, wherein the storage medium controls a device where the storage medium is located to execute the functional area mapping method according to the above-mentioned embodiments of the present application when the computer program is running.

[0045] The present application has the following beneficial effects: the present application provides a functional area mapping method, device, terminal device and storage medium; the method identifies each function area category contained in a region image of a region to be mapped by constructing a function area recognition model, represents the function area in the region image to be mapped in a category after identifying the function area category, and then generates a corresponding function area image; the function area recognition model is constructed to replace the manual examination mode with the model identification mode, thereby improving the mapping efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 Fig. 1 is a flowchart of a functional area mapping method according to an embodiment of the present application.

[0047] Figure 2 Fig. 2 is a structural diagram of an SF-net neural network model according to an embodiment of the present application.

[0048] Figure 3 Fig. 3 is a flowchart of a road network image generation process according to an embodiment of the present application.

[0049] Figure 4 Fig. 4 is a flowchart of a region sample image data preprocessing process according to an embodiment of the present application.

[0050] Figure 5 Fig. 5 is a structural diagram of a multi-scale attention feature extractor according to an embodiment of the present application.

[0051] Figure 6 Fig. 6 is a structural diagram of an sCBAM module according to an embodiment of the present application.

[0052] Figure 7is a plurality of Bottleneck structure schematic diagram provided by an embodiment of the application.

[0053] Figure 8 is a mask layer structure schematic diagram provided by an embodiment of the application.

[0054] Figure 9 is a global average pooling layer, a fully connected layer and a Softmax classifier structure schematic diagram provided by an embodiment of the application.

[0055] Figure 10 is a functional area mapping device structure schematic diagram provided by an embodiment of the application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0057] As shown in the drawings, Figure 1 a functional area mapping method provided by an embodiment of the application comprises the following steps.

[0058] Step S1: acquiring a region image of a region to be mapped;

[0059] Step S2: inputting the region image into a functional area recognition model, so that the functional area recognition model recognizes each functional area category in the region image;

[0060] Step S3: identifying each functional area category in the region image after recognition, and generating a functional area image corresponding to the region to be mapped;

[0061] For step S2, the region image of the region to be mapped is input into a pre-trained functional area recognition model, so that the functional area recognition model recognizes each functional area category contained in the region image; wherein the construction of the functional area recognition model comprises: acquiring region sample images of each functional sub-area of the region to be mapped; taking each region sample image as input and taking the functional area category corresponding to each region sample image as output, training a preset neural network model to generate the functional area recognition model.

[0062] In a preferred embodiment, the region sample image includes: functional area patches for identifying the region where the functional area is located; the neural network model includes a multi-scale attention feature extractor, a mask layer, and a classifier; during the training of the neural network model, the multi-scale attention feature extractor extracts features from the region sample image to generate a first feature map; the mask layer generates a region of interest based on the first feature map and the corresponding functional area patches of the region sample image; the classifier identifies the region of interest to determine the functional area category of the region sample image.

[0063] Specifically, such as Figure 2 The diagram shown is a schematic representation of a neural network model according to an embodiment of the present invention. Preferably, in one embodiment of the present invention, the neural network model is an SF-net structure, which includes a multi-scale attention feature extractor, a mask layer, and a classifier. The classifier is composed of two fully connected layers and a Softmax function. When training the above neural network model, the multi-scale attention feature extractor extracts features from the region sample image to generate a first feature map, and transmits the generated first feature map to the mask layer. The region sample image includes functional area patches used to identify the functional area. The shape information of the corresponding functional area patches is obtained from the region sample image, and the shape information of the functional area patches is downsampled and transmitted to the mask layer. The mask layer multiplies the first feature map and the corresponding pixels of the downsampled functional area patch shape information to extract the region of interest. The feature map of the region of interest extracted by the mask layer is subjected to global average pooling to obtain a one-dimensional feature vector. Finally, the classifier identifies the region of interest and determines the functional area category of the region sample image. In this embodiment, the multi-scale attention feature extractor uses multi-scale training to reduce the scale effect caused by the different sizes of urban functional areas, and promotes the fusion of high-resolution images and point density heatmaps through an attention mechanism. In the mask layer, the image features extracted by the multi-scale attention feature extractor are masked by the stored irregular functional area shape information, thereby obtaining the features of the location of the functional area and shielding other irrelevant information.

[0064] It should be noted that the functional area categories in this embodiment of the invention include: residential area, public service area, green space, industrial area, commercial area, educational area, and agricultural area; wherein, a residential area is defined as: an area where the function of the houses is mainly for living, such as a residential community or a large self-built house; a public service area is defined as: an area that provides public services, such as government departments, hospitals, and stations; a green space is defined as: an area with a lot of green plants, such as wetland parks and lawns; an industrial area is defined as: an area for the production or storage of goods, such as factories and logistics warehouses; a commercial area is defined as: a commercial, trade, or sales area, such as a financial center, shopping center, or market; an educational area is defined as: schools and some educational and training institutions; and an agricultural area is defined as: an agricultural production area, such as farmland and fishponds.

[0065] In a preferred embodiment, acquiring regional sample images of each functional zone of the area to be mapped includes: acquiring road network data of the area to be mapped; generating a road network image of the area to be mapped based on the road network data; acquiring high-resolution imagery of the area to be mapped and POI data of various functional zones in the area to be mapped; generating a heat map of each type of functional zone in the area to be mapped based on the POI data of various functional zones; overlaying the heat maps of all functional zones to generate a multi-channel heat map; overlaying the multi-channel heat map and the high-resolution image of the area to be mapped to generate a fused image of the area to be mapped; and segmenting the fused image based on the road network image to generate regional sample images of each functional zone of the area to be mapped.

[0066] Specifically, the road network data of the area to be mapped is obtained, and the corresponding road network image is generated based on the road network data of the area to be mapped; the road network data can be obtained through the open-source map service OpenStreetMap (OSM).

[0067] In a preferred embodiment, generating a road network image of the area to be mapped based on road network data of the area to be mapped includes: selecting roads in the road network data that meet a preset level to generate a first road network image; establishing a buffer zone with the road as the center line for each road in the first road network image, and merging the buffer zones to generate a second road network image; extracting the center line of the merged buffer zone in the second road network image, and correcting topological errors on the extracted center line to generate a road network image of the area to be mapped.

[0068] Specifically, such as Figure 3 The diagram shown is a flowchart of a road network image generation process for a region to be mapped, provided by an embodiment of the present invention. Figure (a) is the original OSM road network image, Figure (b) is the first road network image, Figure (c) is the second road network image, and Figure (d) is the road network image for the region to be mapped. In this embodiment, generating the road network image for the region to be mapped includes the following steps:

[0069] Step S101: Extracting roads meeting the preset level from the original OSM road network image of the region to be mapped, preferably, the roads meeting the preset level are high-level roads of level three or above, and after selection, low-level dense roads can be removed, to generate a first road network image containing high-level roads of level three or above (Fig. (b));

[0070] Step S102: Merging lanes in the first road network image; since each road in the road network often has multiple lanes, a road is composed of multiple line elements. In this case, a fifty-meter buffer zone is established for each line element (i.e. the above-mentioned road), and the resulting buffer polygon elements are fused to generate a second road network image (Fig. (c));

[0071] Step S103: Extracting the center line of the buffer polygon element in the second road network image (Fig. (c)) as a new road, and correcting some topological errors in the road network, such as hanging lines. The resulting road network image of the region to be mapped, in which each road contains only one line element (Fig. (d)).

[0072] After the above three steps, the road network image of the region to be mapped (Fig. (d)) obtained can be used as the basis for dividing blocks, and the polygon surrounded by each road is a block, which is regarded as a basic classification unit of urban functional areas.

[0073] After generating the road network image of the region to be mapped, high-resolution images of the region to be mapped and POI (Point of Interest) data of various functional areas in the region to be mapped are obtained; preferably, in the embodiment of the present application, the data obtained are high-resolution remote sensing image data and POI data with a spatial resolution of 1 m, which respectively provide spatial structure information and social and economic information. For example, Figure 4 Fig. 1 is a flow chart of the pre-processing process of the regional sample image data provided by an embodiment of the present application; the POI data is discrete vector points, in order to enable it to be processed by a convolutional neural network CNN (Convolutional Neural Network) together with the high-resolution image, it needs to be converted into a continuous image. The present application converts the discrete POI data into a multi-channel point density heat map, in which each channel is the spatial distribution of the density of a certain type of POI point aligned with the spatial coordinates of the high-resolution image. The processing process includes:

[0074] Step S201: obtaining high-resolution images of the region to be mapped and POI data of various functional areas in the region to be mapped; in this embodiment, as shown in Figure 4 Fig. 2, three categories of POI data are selected, which are commercial POI, educational POI and public service POI;

[0075] Step S202: Calculate the heatmaps corresponding to the three types of POI data in step S201 respectively. For each grid pixel in the heatmap corresponding to a certain type of POI, it is obtained by dividing the number of points of that type of POI in its 300*300 neighborhood by the area of ​​the 300*300 neighborhood.

[0076] Step S203: Overlay the three heatmaps generated from the business POI, education POI, and public service POI data to form a three-channel heatmap. Where C, H, and W represent the number of bands, height, and width of the heatmap, respectively;

[0077] Step S204: Overlay the three-channel density heatmap obtained in step S203 with the high-resolution image of the region to be mapped to generate a fused image of the region to be mapped;

[0078] Step S205: Segment the fused image according to the road network image to generate regional sample images of each functional partition of the area to be mapped.

[0079] In a preferred embodiment, the road network image includes several closed sub-regions; the segmentation of the fused image based on the road network image includes: determining the minimum bounding square of each closed sub-region based on the shape of the closed sub-region in the road network image; generating a region to be segmented for the corresponding closed sub-region based on the minimum bounding square; and segmenting the fused image based on the region to be segmented.

[0080] Specifically, such as Figure 3 As shown in Figure (d), the road network image includes several closed sub-regions. A minimum bounding square is determined for each smallest closed sub-region based on its shape. Then, the region to be segmented for each closed sub-region is determined based on the size of the minimum bounding square. The fused image is then segmented based on the region to be segmented. Specifically, in step S205, the fused image is cropped using the minimum bounding square of the previously obtained basic classification unit (i.e., the closed sub-region) to obtain region sample images of each functional partition of the region to be mapped. Among them, I j C and H represent sample images generated for the j-th urban functional area. j and W j represents the number of channels, height, and width of the smallest bounding square of the j-th urban functional area, respectively, and n represents the number of functional areas within the mapping area.

[0081] In a preferred embodiment, the multi-scale attention feature extractor comprises five stages, the first stage comprises a convolution layer and a pooling layer, the second stage comprises three multi-scale attention residual blocks, the third stage comprises four multi-scale attention residual blocks, the fourth stage comprises six multi-scale attention residual blocks, and the fifth stage comprises three multi-scale attention residual blocks; wherein each multi-scale attention residual block comprises a hierarchical convolution module and an attention mechanism module.

[0082] Specifically, as shown in the accompanying drawings, Figure 5 The multi-scale attention feature extractor is a convolutional neural network (CNN) with residual characteristics, and has attention mechanism and multi-scale feature extraction capability.

[0083] Each regional sample image is generated by dividing the minimum circumscribed square of the corresponding closed sub-region in the road network image according to the shape of each closed sub-region, and each closed sub-region corresponds to a functional area. The area difference between the functional areas in the city is large, resulting in different sizes of sample images. For sample images of different sizes, the existing convolutional neural network CNN cannot utilize them for batch training, and training each sample image separately will result in a decrease in computational efficiency and accuracy. However, if all sample images are scaled to the same size, it will cause a scale effect and cause information loss. To this end, the regional sample images are resampled to four different sizes (192x192, 256x256, 512x512 and 800x800), and these images of different sizes will be used as the input of the multi-scale attention feature extractor, so as to minimize the loss of scale information. The output feature map of the multi-scale attention feature extractor will be 1 / 32 of the input image size, and the dimension will be 2048. Therefore, the four kinds of input 6x192x192, 6x256x256, 6x512x512 and 6x800x800 correspond to four kinds of output 2048x6x6, 2048x8x8, 2048x16x16 and 2048x25x25.

[0084] The multi-scale attention feature extractor is generally divided into five stages. The first stage (stage 0) contains a 7*7 convolutional layer and a 3*3 max-pooling layer, and the following four stages contain 3, 4, 6 and 3 multi-scale attention residual blocks respectively. Inside each residual block, there is a 1*1 convolutional layer, a hierarchical group convolution module (HGC), a small convolutional block attention module (sCBAM), and finally another 1*1 convolutional layer.

[0085] Specifically, as shown in Figure 6 (a), it is a structure diagram of a multi-scale attention residual block provided by an embodiment of the present application. Compared with an ordinary residual block (as shown in Figure 6 (b)), it has multi-scale feature extraction capability and attention mechanism. After the input image is subjected to 1*1 convolution, it is first subjected to hierarchical group convolution layer HGC. Specifically, the feature map is divided into 4 parts x n (n = 1, 2, 3, 4) according to the channel dimension. The first part x1 is not processed and is directly output as y1; the second part x2 is subjected to 3*3 convolution and is divided into two lines, one of which is directly output as y2, and the other of which is transmitted to x3, so that the third line obtains the information of the second line; the third line and the fourth line are similar. Thus, the outputs of different levels will have different sizes of receptive field, so that multi-scale information can be extracted. Finally, the obtained y n (n = 1, 2, 3, 4) are channel spliced to fuse their information at different levels.

[0086] The spliced image can be considered to have two dimensions: channel dimension and spatial dimension. In the channel dimension, different channels of the image have different contributions to the classification of urban functional areas, and for this purpose, the channel attention mechanism can give different weights to different channels, thereby quantifying this contribution. In the spatial dimension, urban functional areas are usually composed of different urban land uses in different regions, and the spatial attention mechanism can adaptively focus on the features of different regions in the spatial dimension. In order to achieve the above two purposes, the present application introduces a lightweight attention module CBAM (Convolutional Block Attention Module) and improves it to obtain the sCBAM (small-CBAM) module as shown in Figure 7 . The lightweight attention module sCBAM first realizes its channel attention, and then realizes its spatial attention. Globe Average Pooling (GAP) is performed on each channel of the channel weight redistribution feature map to obtain a one-dimensional feature vector capable of representing the information of each channel; then the one-dimensional convolution is used to learn the relationship between each channel, and finally the Sigmoid function is used to obtain the weight of each channel The weight is multiplied by the input image to obtain a channel weight redistribution feature map (C, h and w represent the number of channels, height and width, respectively), that is, the effect of strengthening important channels and inhibiting unimportant channels is achieved; then spatial attention is realized, average value pooling and maximum value pooling are performed on the channel weight redistribution feature map to generate two feature maps, the two feature maps are stacked in the channel and processed by two-dimensional convolution, and then the new feature map after two-dimensional convolution is normalized by using the activation function Sigmoid to obtain a two-dimensional spatial attention map The two-dimensional spatial attention map is multiplied by the channel weight redistribution feature map , that is, spatial attention correction is realized on the input image, and finally a feature map on which attention mechanisms are performed in the channel and in the space is obtained

[0087] After the region sample image is input into the multi-scale feature extractor for feature extraction, a first feature map containing the depth features of the region sample image is obtained. However, the extracted features are the features of the minimum circumscribed square region of the irregular functional area, and there is a lot of irrelevant information. The present application adds a mask layer to the neural network model, so that it can save the original shape information of the functional area patch; as shown in Figure 8 , specifically, the region sample image generated by using the minimum circumscribed square of each functional area is input into the convolutional neural network CNN, while saving the functional area patch shape information, the saving method is: the original functional area part is filled with 1, and the other parts of the minimum circumscribed square are filled with 0, and the filled information is converted into an image Mask, then in the mask layer, the first feature map obtained by the multi-scale attention feature extractor is multiplied by the image corresponding to the shape information to remove the features of the irrelevant region, and a second feature map is obtained.

[0088] The mask layer does not change the size of the feature map, so the multi-scale attention feature extractor still has four sizes of output, that is, 2048x6x6, 2048x8x8, 2048x16x16 and 2048x25x25. In order to unify their sizes for classification, the present application introduces a global average pooling layer after the mask layer, as shown in Figure 9 ​As shown, different size feature maps will be unified into 2048x1x1 one-dimensional feature vector, and then input into two fully connected layers, the first layer fully connected layer is 2048, the second layer is 1024, and finally the prediction probability value of each functional area category is output by using the Softmax function. In the process of constructing the functional area recognition model, images of the same size will be trained under the same epoch; and in the process of functional area recognition, the original size of the region sample image is directly used for recognition.

[0089] In a preferred embodiment, the loss function of the functional area recognition model is:

[0090] wherein, is the loss function of the functional area recognition model, y i,k and p i,k are the true label and the prediction probability value of the k-th functional area in the i-th region to be mapped, and C is the total number of functional area categories. Based on the loss function, the SF-net is trained by the back propagation algorithm, the parameters of the network are learned and optimized, and in the training process, the Adam optimizer is used, and the learning rate is set to 0.001.

[0091] For step S3, after the functional area category recognition of the region image of the region to be mapped is performed, the functional area image corresponding to the region to be mapped is generated.

[0092] On the basis of the above-mentioned method embodiment, the present application correspondingly provides a device embodiment.

[0093] As Figure 10 shown, an embodiment of the present application provides a functional area mapping device, comprising: a data acquisition module, a functional area classification module, a mapping module and a functional area recognition model construction module;

[0094] The data acquisition module is used to acquire the region image of the region to be mapped;

[0095] The functional area recognition module is used to input the region image into the functional area recognition model, so that the functional area recognition model recognizes each functional area category in the region image;

[0096] The mapping module is used to identify each functional area category in the region image after recognition, and generate the functional area image corresponding to the region to be mapped;

[0097] The functional area recognition model construction module is used to acquire the region sample image of each functional area of the region to be mapped; each region sample image is used as input, and the functional area category corresponding to each region sample image is used as output, a preset neural network model is trained, and the functional area recognition model is generated.

[0098] It should be noted that the apparatus embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiments provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0099] Those skilled in the art can clearly understand that, for the convenience and brevity, the specific working process of the apparatus described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0100] On the basis of the foregoing method embodiments, the present application correspondingly provides terminal device embodiments.

[0101] An embodiment of the present application provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, when the computer program is executed by the processor, a function area mapping method according to any one of the embodiments of the present application is realized.

[0102] The terminal device can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The terminal device can include, but is not limited to, a processor and a memory.

[0103] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like. The processor is the control center of the terminal device, and connects all parts of the terminal device through various interfaces and lines.

[0104] The memory can be used to store the computer program, and the processor realizes various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, and the like; and the data storage area can store data created according to the use of the mobile phone and the like. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0105] On the basis of the above-mentioned method embodiment, the application provides a storage medium embodiment.

[0106] An embodiment of the application provides a storage medium, which comprises a stored computer program, wherein the computer program controls a device where the storage medium is located to perform a function area mapping method according to any one of the embodiments of the application when the computer program is run.

[0107] The storage medium is a computer readable storage medium, and the computer program is stored in the computer readable storage medium. The computer program can realize the steps of the above-mentioned various method embodiments when the computer program is executed by a processor. The computer program comprises computer program code, which can be in a form of source code, object code, an executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal and a software distribution medium, etc.

[0108] The above-mentioned is a preferred embodiment of the application. It should be noted that, for those skilled in the art, without departing from the principle of the application, some improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the application.

Claims

1. A method of mapping a functional region, characterized by, The method comprises: obtaining a region image of a region to be mapped; inputting the region image into a functional area identification model to enable the functional area identification model to identify each functional area category in the region image; identifying each functional area category in the region image after identification to generate a functional area image corresponding to the region to be mapped; wherein the construction of the functional area identification model comprises: obtaining region sample images of each functional sub-region of the region to be mapped; training a preset neural network model with each region sample image as input and the functional area category corresponding to each region sample image as output to generate the functional area identification model; the obtaining of the region sample images of each functional sub-region of the region to be mapped comprises: obtaining road network data of the region to be mapped; generating a road network image of the region to be mapped according to the road network data of the region to be mapped; obtaining high-resolution images of the region to be mapped and POI data of each functional area in the region to be mapped; generating a heat map of each type of functional area in the region to be mapped according to the POI data of each type of functional area; superimposing the heat maps of all functional areas to generate a multi-channel heat map; superimposing the multi-channel heat map and the high-resolution images of the region to be mapped to generate a fusion image of the region to be mapped; segmenting the fusion image according to the road network image to generate region sample images of each functional sub-region of the region to be mapped.

2. The method of claim 1, wherein, The region sample image comprises a functional area polygon for identifying the region where the functional area is located. The neural network model comprises a multi-scale attention feature extractor, a mask layer and a classifier. In the process of training the neural network model, the multi-scale attention feature extractor extracts features from the region sample image to generate a first feature map; the mask layer generates a second feature map according to the first feature map and the functional area polygon of the corresponding region sample image; the classifier identifies the second feature map to determine the functional area category of the region sample image.

3. A method for charting a function region as recited in claim 2, wherein, The generation of the road network image of the region to be mapped according to the road network data of the region to be mapped comprises: selecting roads in the road network data that meet a preset level to generate a first road network image; establishing a buffer zone centered on each road in the first road network image, and fusing the buffer zones to generate a second road network image; extracting the center lines of the fused buffer zones in the second road network image, and performing topological error correction on the extracted center lines to generate the road network image of the region to be mapped.

4. The method of claim 2, wherein, The road network image comprises a plurality of closed sub-regions. The segmentation of the fusion image according to the road network image comprises: determining the minimum circumscribed square of each closed sub-region in the road network image according to the shape of the corresponding closed sub-region, and generating a region to be segmented for the corresponding closed sub-region according to the minimum circumscribed square; segmenting the fusion image according to the region to be segmented.

5. The method of claim 2, wherein, The multi-scale attention feature extractor comprises: The multi-scale attention feature extractor comprises five stages, the first stage comprises a convolution layer and a pooling layer, the second stage comprises three multi-scale attention residual blocks, the third stage comprises four multi-scale attention residual blocks, the fourth stage comprises six multi-scale attention residual blocks, and the fifth stage comprises three multi-scale attention residual blocks; wherein each multi-scale attention residual block comprises a hierarchical convolution module and an attention mechanism module.

6. The method of claim 1, wherein, The loss function of the functional area recognition model is: wherein, is a loss function of the functional area recognition model, and are respectively a true label and a predicted probability value of a class of the i-th functional area in the region to be mapped, is a total number of functional area classes.​​ 7. A functional region mapping device, characterized by comprises: a data acquisition module, a functional area classification module, a mapping module, and a functional area recognition model construction module; The data acquisition module is configured to acquire a regional image of a region to be mapped. The functional area classification module is configured to input the regional image into a functional area recognition model, so that the functional area recognition model identifies each functional area category in the regional image. The mapping module is configured to identify each functional area category in the regional image after identification, and generate a functional area image corresponding to the region to be mapped. The functional area recognition model construction module is configured to acquire regional sample images of each functional subarea of the region to be mapped; take each regional sample image as input and the functional area category corresponding to each regional sample image as output, train a preset neural network model, and generate the functional area recognition model. The acquisition of the regional sample images of each functional subarea of the region to be mapped comprises: acquiring road network data of the region to be mapped; generating a road network image of the region to be mapped according to the road network data of the region to be mapped; acquiring high-resolution images of the region to be mapped and POI data of each functional area in the region to be mapped; generating a heat map of each functional area in the region to be mapped according to the POI data of each functional area; superimposing the heat maps of all functional areas to generate a multi-channel heat map; superimposing the multi-channel heat map and the high-resolution images of the region to be mapped to generate a fusion image of the region to be mapped; segmenting the fusion image according to the road network image to generate regional sample images of each functional subarea of the region to be mapped.

8. A terminal device, comprising: The storage medium comprises a stored computer program, wherein the computer program controls a device in which the storage medium is located to perform a functional area mapping method as claimed in any one of claims 1 to 6 when the computer program is running.

9. A storage medium, characterized by The storage medium comprises a stored computer program, wherein the computer program controls a device in which the storage medium is located to perform a functional area mapping method as claimed in any one of claims 1 to 6 when the computer program is running.

Images