CN111695488A - Interest plane identification method, device, equipment and storage medium - Google Patents

Abstract

The present application discloses a method, apparatus, device, and storage medium for identifying a face of interest, and relates to artificial intelligence and deep learning. The specific implementation scheme is as follows: by obtaining the target function category information corresponding to the target interest surface in the target area to be identified, and the geographic attribute related data of the target area; according to the target function category information and the geographic attribute related data of the target area, obtaining Input data of the corresponding preset computer vision model; according to the input data and the corresponding preset computer vision model, identify the contour of the target surface of interest in the target area. The application uses computer vision technology to realize the identification of the interest surface in the target area, and does not require manual annotation based on street view images, which can reduce costs, improve efficiency, and has a high accuracy rate. It can be applied to different scenarios and has a wider application range. It can be applied to the identification of massive interest surfaces, which makes it possible to describe the surface information of massive interest points.

Description

Interest plane identification method, device, device and storage medium

technical field

The embodiments of the present application relate to artificial intelligence and deep learning in computer technology, and in particular, to a method, apparatus, device, and storage medium for identifying a face of interest.

Background technique

Area of Interest (AOI) refers to an area on an electronic map that actually exists in the physical world, and is used to identify the urban functions (such as schools, residential areas, hospitals, shopping centers, etc.) represented by the geographic location. ) and its geographic boundaries. The interest surface can vividly describe the geographic boundary attributes of urban functions; in addition, in the label completion task of the urban point of interest (POI), when the geographic boundary of the interest surface is known, the interest surface can be The data such as pictures, texts, trajectories and other data generated by users within the covered geographic range are associated with relevant POIs to enrich the semantic label information of POIs; the interest surface is also very helpful for judging whether users have visited POIs, which can then be accurately shop recommendation.

The existing technology requires professional collection equipment to collect street view images, such as collection vehicles, etc., and then manual annotation by professional collection personnel, so as to mark the contour of the surface of interest; The algorithm clusters the points of interest of the same functional category to determine the contour of the surface of interest.

The manual collection method in the prior art has high equipment and labor costs and low efficiency, and cannot be applied to the identification of a large number of interest surfaces; while the bottom-up clustering method has a complicated process and cannot guarantee the accuracy.

SUMMARY OF THE INVENTION

The present application provides a method, apparatus, device and storage medium for identifying a face of interest, so as to reduce the cost of identifying a face of interest and improve the efficiency and accuracy of the identification.

According to a first aspect of the present application, a method for identifying a face of interest is provided, comprising:

Obtain target function category information corresponding to the target interest surface in the target area to be identified, and geographic attribute-related data of the target area;

According to the target function category information and the geographic attribute related data of the target area, obtain the input data of the corresponding preset computer vision model;

According to the input data and the corresponding preset computer vision model, the contour of the target surface of interest in the target area is identified.

According to a second aspect of the present application, a device for identifying a face of interest is provided, comprising:

an acquisition module, configured to acquire target function category information corresponding to the target surface of interest in the target area to be identified, and geographic attribute-related data of the target area;

a processing module, configured to obtain input data of the corresponding preset computer vision model according to the target function category information and the geographic attribute-related data of the target area;

The identification module is configured to identify the contour of the target surface of interest in the target area according to the input data and the corresponding preset computer vision model.

According to a third aspect of the present application, an electronic device is provided, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of the first aspect.

According to a fourth aspect of the present application, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method of the first aspect.

According to a fifth aspect of the present application, there is provided a computer program, including program code, when the computer runs the computer program, the program code executes the rearview mirror adjustment method according to the first aspect.

The method, device, device, and storage medium for identifying a face of interest provided by the present application obtain target functional category information corresponding to the target face of interest in the target area to be identified, and data related to the geographic attributes of the target area; according to the target function category information , and the geographic attribute related data of the target area, to obtain the input data of the corresponding preset computer vision model; according to the input data and the corresponding preset computer vision model, identify the contour of the target surface of interest in the target area. The application uses computer vision technology to realize the identification of the interest surface in the target area, and does not require manual annotation based on street view images, which can reduce costs, improve efficiency, and has a high accuracy rate. It can be applied to different scenarios and has a wider range of applications It can be applied to the identification of a large number of interest surfaces, which makes it possible to describe the surface information of a large number of interest points.

It should be understood that the content described in this section is not intended to identify key or critical features of the embodiments of the application, nor is it intended to limit the scope of the application. Other features of the present application will become readily understood from the following description.

Description of drawings

The accompanying drawings are used for better understanding of the present solution, and do not constitute a limitation to the present application. in:

1 is a schematic diagram of a scene of a method for recognizing a face of interest provided according to an embodiment of the present application;

2 is a flowchart of a method for identifying a face of interest provided according to an embodiment of the present application;

3 is a flowchart of a method for identifying a face of interest provided according to an embodiment of the present application;

4A is a satellite image of a target area provided according to an embodiment of the present application;

4B is a satellite image marked with a point of interest identifier provided according to an embodiment of the present application;

4C is a satellite image marked with a point of interest identifier provided according to another embodiment of the present application;

4D is a schematic diagram of a result obtained by an image instance segmentation model provided according to an embodiment of the present application;

5 is a flowchart of a method for identifying a face of interest provided according to an embodiment of the present application;

6 is a schematic diagram of a result obtained by an image semantic segmentation model provided according to an embodiment of the present application;

7 is a schematic diagram of an image semantic segmentation model provided according to an embodiment of the present application;

8 is a flowchart of a method for identifying a face of interest provided according to an embodiment of the present application;

9 is a block diagram of an apparatus for identifying a face of interest provided according to an embodiment of the present application;

FIG. 10 is a block diagram of an electronic device used to implement the method for recognizing a face of interest according to an embodiment of the present application.

Detailed ways

Exemplary embodiments of the present application are described below with reference to the accompanying drawings, which include various details of the embodiments of the present application to facilitate understanding, and should be considered as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.

In the prior art, manual annotation based on street view images to obtain the contour of the surface of interest requires professional equipment and professionals, and the equipment and labor costs are high and inefficient. The bottom-up clustering method uses the clustering of interest points Obtaining the contour of the surface of interest is a complicated process, and the accuracy cannot be guaranteed. That is, the interest surface identification method in the prior art has high cost, low identification efficiency and low accuracy, and cannot be applied to the identification of massive interest surfaces.

In view of the problems existing in the prior art, in this application, computer vision technology is considered to realize the identification of the surface of interest in the target area, and the target function category information corresponding to the target surface of interest in the target area to be identified, and The geographic attribute-related data of the region does not require manual annotation based on street view images, which can reduce costs, improve efficiency, and has high accuracy. It makes it possible to characterize the planar information of massive interest points.

The face of interest identification method provided by this application can be applied to the scenario shown in FIG. 1 , where the database 11 is used to store geographic attribute related data of different regions, and the electronic device 10 can obtain the geographic attribute of the target region to be identified from the database 11 . related data, in addition, the electronic device 10 can also obtain the target function category information corresponding to the target interest surface in the target area. For example, the target function category information can be preset by the user, and then the target function category information and the geographical attributes of the target area are determined according to the target function category information. According to the relevant data, the input data of the corresponding preset computer vision model is obtained, and then the contour of the target surface of interest in the target area is identified according to the input data and the corresponding preset computer vision model.

Optionally, after the outline of the target surface of interest is identified, the outline of the target surface of interest may be output through the display screen 13 . Then, the geographic boundary attributes of urban functions can be described through the interest surface; in addition, in the label completion task of the urban interest point, when the geographic boundary of the interest surface is known, the user can use the interest surface to cover the geographic range. The generated data such as pictures, text, and trajectories are associated with relevant points of interest, enriching the semantic label information of the points of interest; the interest surface is also very helpful for judging whether users have visited the points of interest, and then can make accurate store recommendations.

The process of identifying the face of interest provided by the present application will be described in detail below with reference to specific embodiments.

An embodiment of the present application provides a method for identifying a face of interest, and FIG. 2 is a flowchart of the method for identifying a face of interest according to an embodiment of the present invention. The executive body may be the electronic device 10 in FIG. 1 , as shown in FIG. 2 , and the specific steps of the method are as follows:

S201. Acquire target function category information corresponding to a target surface of interest in a target area to be identified, and data related to geographic attributes of the target area.

In this embodiment, the target city can be divided into multiple areas according to the road network information. Optionally, the areas can be divided according to the road network of preset road classifications, for example, the seven-level roads that can be used by motor vehicles are used to divide the areas. The city is divided into blocks, so that the semantic integrity of each area can be guaranteed.

When it is necessary to identify one or more areas of the target city, the target area can be selected by the user, and after receiving the user's target area selection instruction, the target area can be selected from the multiple areas according to the target area selection instruction. Determine the target area to be identified.

Further, after the target area to be identified is determined, target function category information corresponding to the target surface of interest in the target area and data related to geographic attributes of the target area may be acquired.

Among them, the target functional category information can be preset by the user, for example, the target area may include geographical entities of various functional categories, such as schools, residential areas, hospitals, etc., and when the user needs to identify the functional category in the target area as the interest of the school When the outline of the face is displayed, the user can set the target function category information corresponding to the target interest face as school.

The geographic attribute-related data of the target area is the data used to characterize the geographic attributes of the target area, which may specifically include, but not limited to, satellite images of the target area, Point of Information (POI) information in the target area, and roads in the target area. Network information, user behavior information in the target area, etc. The specific required data related to the geographic attributes of the target area can be determined according to the difficulty of data acquisition. For example, some underdeveloped areas or emerging areas do not have satellite images, so they can be used. Other geographic attribute related data. Of course, according to different data related to the geographic attributes of the target area, a corresponding preset computer vision model can be selected to perform subsequent contour recognition processing of the target surface of interest.

For example, the geographic attribute-related data of the target area may include satellite images of the target area and point-of-interest information of the target area, and the preset computer vision model is an image instance segmentation model; for another example, the geographic attribute-related data of the target area may include the target area The feature information of interest points in the region and the feature information of the road network in the target region can also include user behavior information in the target region, etc. The preset computer vision model is an image semantic segmentation model.

Of course, the preset computer vision model in this embodiment is not limited to the models listed above, the preset computer vision model may also be other models, and the geographic attribute related data of the target area may also be determined according to model requirements. In addition, when determining a preset computer vision model for subsequent contour recognition processing of the target surface of interest, the geographic attribute-related data of the target area may also be determined according to the preset computer vision model.

S202. Acquire input data of a corresponding preset computer vision model according to the target function category information and the geographic attribute related data of the target area.

In this embodiment, since the subsequent contour recognition processing of the target surface of interest is performed by using the preset computer vision model, it is necessary to obtain the preset computer vision model according to the acquired target functional category information and the geographic attribute related data of the target area. Input data.

For example, for example, the geographic attribute-related data of the target area may include satellite images of the target area and point-of-interest information of the target area, the preset computer vision model is an image instance segmentation model, and the input data may be satellites marked with point-of-interest identifiers For another example, the geographic attribute-related data of the target area may include the feature information of points of interest in the target area and the road network feature information of the target area, and may also include user behavior information in the target area, etc. The preset computer vision model is an image. In the semantic segmentation model, the input data may be a semantic feature map representing the target area obtained according to the above-mentioned geographic attribute related data. The preset computer vision model in this embodiment is not limited to the models listed above, so the input data and the process of acquiring the input data can be determined according to the preset computer vision model, which will not be repeated here.

S203. Identify the contour of the target surface of interest in the target area according to the input data and the corresponding preset computer vision model.

In this embodiment, input data is input into a preset computer vision model, and the preset computer vision model can obtain the contour of the target surface of interest in the target area after processing the input data by using the corresponding computer vision technology.

It can be understood that, after the contour of the target surface of interest is identified, the contour of the target surface of interest can be output through the display screen.

The method for identifying a face of interest provided by this embodiment obtains target function category information corresponding to a target face of interest in a target area to be identified, and data related to geographic attributes of the target area; according to the target function category information, and For the geographic attribute related data of the target area, the input data of the corresponding preset computer vision model is obtained; according to the input data and the corresponding preset computer vision model, the contour of the target surface of interest in the target area is identified. In this embodiment, the computer vision technology is used to realize the identification of the interest surface in the target area, and manual annotation based on the street view image is not required, which can reduce the cost, improve the efficiency, and has a high accuracy rate, which can be applied to different scenarios. The scope of application is large, and it can be applied to the identification of massive interest surfaces, thereby making it possible to describe the surface information of massive interest points.

FIG. 3 is a flowchart of a method for recognizing a face of interest provided by another embodiment of the present invention. On the basis of the above-mentioned embodiment, this embodiment provides a method for recognizing a face of interest, which is aimed at a situation where satellite images of a target area are available.

In this embodiment, since satellite images contain rich physical spatial information, such as buildings, streets, vegetation, etc., these spatial information is exactly what computer vision technology is good at, so the problem of interest surface recognition and image instances can be segmented The problem is analogous. Image instance segmentation is to identify objects in the image and outline the boundaries of the objects. In this embodiment, the image instance segmentation technology is used to identify the surface of interest in the satellite image and outline the boundary of the surface of interest. Considering that the surface of interest is closely related to urban functions, and this semantic information about urban functions cannot be reflected from satellite images, it is not possible to effectively identify the surface of interest only by relying on satellite images. Therefore, in this embodiment, the point of interest (POI ) semantic information is introduced into the satellite image, and the satellite image is enhanced as the input of the image instance segmentation model.

Therefore, the geographic attribute-related data of the target area in this embodiment may include satellite images of the target area and point-of-interest information of the target area; the preset computer vision model is an image instance segmentation model.

The specific steps of the method for identifying a face of interest provided by this embodiment are as follows:

S301. Acquire target function category information corresponding to a target surface of interest in a target area to be identified, and geographic attribute-related data of the target area.

The geographic attribute-related data of the target area may include satellite images of the target area and point-of-interest information of the target area.

In this embodiment, the principle and implementation manner of S301 are similar to the above-mentioned S201, and details are not repeated here.

Further, S202 in the above embodiment may specifically include the following S302-S303, and the specific process is as follows:

S302 . According to the target function category information and the interest point information of the target area, filter a set of first target interest points in the target area with the same target function category.

In this embodiment, for the points of interest in the target area, screening can be performed according to the target function category information corresponding to the target interest surface, and the first target point of interest that is the same as the target function category is selected from the points of interest in the target area. , to obtain the set of first target interest points.

S303. According to the set of the first target interest points, mark the identifier of the first target interest point on the corresponding position of the satellite image of the target area, and obtain the satellite image marked with the interest point identifier as the image Input data for the instance segmentation model.

In this embodiment, since the satellite image does not have the semantic information of the city function, but the POI has the semantic information of the city function, marking the first target interest point in the satellite image can make the satellite image have the semantic information of the city function. That is, it is possible to know how many first target interest points exist in the target area, as well as the location, distribution, and the like of the first target interest points by marking the satellite image with the interest point identifier. The satellite image of the target area shown in FIG. 4A can be marked with a satellite image marked with a point of interest as shown in FIG. 4B , where the dot is the identifier of the first target point of interest.

Optionally, when marking the identifier of the first target interest point on the corresponding position of the satellite image of the target area, the first target interest point may be selected in random order from the set of the first target interest point, if If there is no occlusion with the marked interest point identifier in the satellite image, the marker is marked on the corresponding position of the satellite image of the target area.

In this embodiment, considering that the POI information has great differences in different regions, how to selectively and reasonably mark POI identifiers on the satellite image will determine whether the target surface of interest can be accurately and effectively identified. important. Therefore, in this embodiment, a predetermined rule is used to select the first target interest point from the set of first target interest points for marking. Specifically, the predetermined rule may be: each point of interest corresponds to a logo (icon); The points of interest use different identifiers; the category of the points of interest marked on the satellite image is the same as the target function category corresponding to the target surface of interest, and the identifiers (icons) of each point of interest do not block each other; the points of interest are selected in random order for marking, until there are no marked points of interest. In this embodiment, a heuristic algorithm may be used to mark the point of interest identifier on the satellite image according to the above rules. Specifically, first randomly select a first target interest point from the set of first target interest points, mark a mark on the satellite image according to its position information, and then randomly select the next first target interest point to determine whether it is the same as the existing one. The marked interest point identifiers are occluded. If no occlusion occurs, it is marked, and if occlusion occurs, it is discarded, and so on, until there is no first target interest point that can continue to be marked on the satellite image in the set of first target interest points. In this embodiment, the random order can be used to avoid deviations in the identification results of the interest surfaces caused by the uneven distribution of the interest points. For the satellite image shown in FIG. 4A , through the above-mentioned selective marking process, as shown in FIG. 4C , it can be seen that the distribution of the marks of the first target interest point is relatively uniform and there is no occlusion.

Further, S203 in the above embodiment may specifically include the following S304, and the specific process is as follows:

S304. Perform image instance segmentation on the satellite image marked with the point of interest identifier by the image instance segmentation model, and identify the contour of the target surface of interest in the target area.

In this embodiment, the image instance segmentation model is an existing neural network model, such as MaskR-CNN. The image instance segmentation model is used to segment the satellite image marked with the point of interest. The process of image instance segmentation is different from the existing image The instance segmentation model performs image instance segmentation in a similar process, except that there are differences in the input data, and the specific process will not be repeated here. As an example, the output result of the image instance segmentation model is shown in FIG. 4D , and the outline of the target surface of interest in the target area (the white thick line range) is the white thin line range.

It should be noted that the image instance segmentation model in this embodiment can be trained by training data in advance, and the training data can be satellite images marked with point of interest identifiers, and the contour of the target surface of interest has been identified. The training process is as follows. It is not repeated here.

FIG. 5 is a flowchart of a method for recognizing a face of interest provided by another embodiment of the present invention. On the basis of the above-mentioned embodiment, this embodiment provides a method for identifying a face of interest, which is aimed at the situation where satellite images of the target area are unavailable, such as underdeveloped areas, emerging areas, or other areas where it is difficult to obtain satellite images .

In this embodiment, since it is difficult to obtain satellite images of the target area, the cheaper and easier to obtain point of interest feature information of the target area and the road network feature information of the target area are used as the geographic location of the target area. Attribute related data. The preset computer vision model in this embodiment is an image semantic segmentation model.

As shown in FIG. 5 , the specific steps of the method for identifying a face of interest provided by this embodiment are as follows:

S401. Obtain target function category information corresponding to a target surface of interest in a target area to be identified, and data related to geographic attributes of the target area.

The geographic attribute-related data of the target area may include feature information of points of interest of the target area and road network feature information of the target area.

Optionally, the feature information of points of interest in the target area may include feature information in the following three dimensions: distribution of functional categories of points of interest in the target area, distribution of functional categories of points of interest adjacent to each point of interest, and distribution of functional categories of points of interest adjacent to each point of interest. The location of the point of interest, of course, may also include other characteristic information such as the name of the point of interest. The road network feature information of the target area may include, but is not limited to, the location of the road network, the start and end points of the road, the length of the road, the density of the road network, the distance between the point of interest and the road, and the like.

In this embodiment, the principle and implementation manner of S401 are similar to the above-mentioned S201, and details are not repeated here.

Further, S202 in the above embodiment may specifically include the following S402-S404, and the specific process is as follows:

S402. Determine the center position of the target surface of interest according to the feature information of the point of interest of the target area and the feature information of the road network of the target area.

In this embodiment, according to the target function category information, a set of second target interest points with the same target function category can be selected from the interest points in the target area; further, according to the second target interest Point-of-interest feature information of each second target point of interest in the set of points, road network feature information of the target area, and a preset classification model, and a representative point of interest is determined from the set of second target points of interest, and take its position as the center position of the target surface of interest. Specifically, the point of interest feature information of each second target interest point and the road network feature information of the target area are sequentially input into the preset classification model, and it is sequentially determined whether the interest point of each second target interest point is the target interest surface until the point of interest of a certain second target interest point is determined as the center position of the target interest surface. The preset classification model in this embodiment may be any binary classification model.

S403. Perform gridding on the target area, and for each grid, construct multiple grids of the grid according to the feature information of the points of interest falling in the grid and the road network feature information in the grid. feature map.

In this embodiment, the target area is first gridded, that is, the target area is divided into multiple grids, and then for any grid, the points of interest that fall within the grid are determined, and then the points of interest falling into the grid are obtained. The feature information of the points of interest and the feature information of the road network in the grid, and multiple feature maps of the grid are constructed, such as the feature map of the road network, the feature map of the point of interest, and the feature map of the point of interest can specifically include the feature map of the point of interest function category, Interest point name feature map, adjacent interest point information feature map, and so on.

S404. Use the center position of the target surface of interest and multiple feature maps of each grid as input data of the image semantic segmentation model.

In this embodiment, the center position of the target surface of interest obtained in the above steps and the multiple feature maps of each grid are input into the image semantic segmentation model, and the subsequent recognition of the target surface of interest is carried out. It can perform image semantic segmentation processing on each grid separately, which can improve the accuracy and efficiency of image semantic segmentation processing, and can also make the contour of the determined target interest surface more accurate.

Further, S203 in the above embodiment may specifically include the following S405-S406, and the specific process is as follows:

S405, taking the center position of the target surface of interest as the center, and sequentially determining whether each grid around the center position of the target surface of interest belongs to the target surface of interest through the image semantic segmentation model;

S406. Determine the contour of the target surface of interest in the target area according to the grid position belonging to the target surface of interest.

In this embodiment, the center position of the target surface of interest is used as the center to determine whether the surrounding grids belong to the target surface of interest. On the contour of the target surface of interest, and so on, the contour of the target surface of interest can be obtained. In this embodiment, the image semantic segmentation is performed on the grid around the center of the target surface of interest in turn, which can be quickly located on the contour. It is not necessary to perform image semantic segmentation processing on all the grids in the entire target area, so that the efficiency of identifying the contour of the target surface of interest can be improved. In addition, the image semantic segmentation model is used to sequentially determine whether the grids around the center of the target surface of interest belong to the target surface of interest. Whether the grid belongs to the representative point of interest at the center of the target surface of interest, and if so, it is determined that the grid belongs to the target surface of interest. The output result of S406 is shown in FIG. 6 , in which the target area is the range delineated by the thick gray line, and the gray shaded part is the range of the target surface of interest.

More specifically, the image semantic segmentation model may specifically include a fully convolutional network (Fully Convolutional Network, FCN) and a convolutional network (Convolutional Neural Networks, CNN). Further, S405 may specifically include:

For any grid, the multiple feature maps of the grid are spliced and input into the fully convolutional network of the image semantic segmentation model, respectively, to obtain high-dimensional features; after splicing each of the high-dimensional features Input into the convolutional network of the image semantic segmentation model to obtain the probability that the grid belongs to the target surface of interest; if the probability is greater than a preset threshold, it is determined that the grid belongs to the target surface of interest.

In this embodiment, considering that a single feature map is used, the features are relatively thin and contain less semantic information. Therefore, the global semantic information is fused together by splicing multiple feature maps to increase the semantic information of the feature map. , so that the image semantic segmentation model can better determine whether the grid belongs to the target interest surface according to the feature map, and improve the accuracy of image semantic segmentation.

Optionally, since the feature maps related to interest points are rich in semantic information and can represent local semantic information, while the semantic information of road network feature maps is relatively thin, the feature maps related to interest points can be spliced into the road network feature map. The global feature map is obtained from , and input into a fully convolutional network, and the feature map related to the interest point includes the feature map of the function category of the point of interest, the feature map of the name of the point of interest, the feature map of the information of the adjacent interest points, etc. Each interest point The relevant feature maps are respectively input into a fully convolutional network and processed in parallel through multiple fully convolutional networks. Each fully convolutional network can obtain a high-dimensional feature. After splicing the high-dimensional features Input into the convolutional network of the image semantic segmentation model. It should be noted that the feature map input by the fully convolutional network usually has multiple channels. When a feature map has only one channel, the convolutional network can be used instead of the fully convolutional network.

Optionally, the image semantic segmentation model of this embodiment is shown in FIG. 7 , the feature maps related to the interest points are spliced into the road network feature map to obtain a global feature map and input to the fully convolutional network 410 , and the feature map of the function category of the interest points is obtained. Input to the fully convolutional network 411, the point of interest name feature map is input to the fully convolutional network 412, the adjacent interest point information feature map is a channel, input to the convolutional network 413, the fully convolutional network 410, 411, 412 and the volume The high-dimensional features output by the product network 413 are spliced and input into the convolution network 410 for convolution processing. Finally, the probability that the grid belongs to the target surface of interest is obtained through the activation layer, and the result of whether the grid belongs to the target surface of interest is finally output.

It should be noted that, the image semantic segmentation model in this embodiment may be trained by training data in advance, and the training data may be marked feature map data, and the training process will not be repeated here.

In the above-mentioned embodiment, the feature information of interest points of the target area and the road network feature information of the target area, which are cheaper and easier to obtain, are used as the geographic attribute related data of the target area, and based on the image semantic segmentation model, the satellite image can be analyzed. The target area of interest that is not available in the image does not need to be manually labeled based on the street view image to identify the target area of interest, which can reduce costs, improve efficiency, and have a high accuracy rate.

On the basis of the above embodiment, optionally, the geographic attribute-related data of the target area also includes user behavior information in the target area, such as user comments, check-ins, uploading pictures, etc. in the target area. Some semantic features that can characterize the target area are obtained after extraction and quantification.

In S403, according to the feature information of the points of interest falling in the grid and the road network feature information in the grid, constructing multiple feature maps of the grid may include:

A plurality of feature maps of the grid are constructed according to the feature information of the points of interest falling within the grid, the road network feature information within the grid, and the user behavior information within the grid.

In this embodiment, the geographic attribute-related data of the target area increases the user behavior information in the target area, so that more semantic features can be added to the target area, which is convenient to further improve the accuracy in the process of semantic segmentation by the image semantic segmentation model , which can improve the accuracy of identifying the contour of the target surface of interest.

On the basis of any of the foregoing embodiments, as shown in FIG. 8 , the method for identifying a face of interest may further include:

S501. Divide the target city into multiple regions according to the road network information;

S502. Receive a target area selection instruction, and determine the to-be-identified target area from the multiple areas according to the target area selection instruction.

In this embodiment, by dividing a plurality of regions and selecting a target region to be identified by the user, the interest surface identification of one or more regions in the multiple regions can be realized to meet user requirements.

Further, after determining the target area to be identified, it can be judged whether the satellite image of the target area is available, if the satellite image of the target area is available, then adopt S301-S304, if the satellite image of the target area is unavailable, then S401-406 are used.

An embodiment of the present application provides an apparatus for identifying a face of interest. FIG. 9 is a structural diagram of the apparatus for identifying a face of interest according to an embodiment of the present invention. As shown in FIG. 9 , the apparatus 600 for identifying a face of interest specifically includes: an acquisition module 601 , a processing module 602 and an identification module 603 .

An acquisition module 601, configured to acquire target function category information corresponding to a target surface of interest in a target area to be identified, and geographic attribute-related data of the target area;

A processing module 602, configured to obtain input data of a corresponding preset computer vision model according to the target function category information and the geographic attribute-related data of the target area;

The identification module 603 is configured to identify the contour of the target surface of interest in the target area according to the input data and the corresponding preset computer vision model.

In an optional embodiment, the geographic attribute-related data of the target area includes a satellite image of the target area and point-of-interest information of the target area; the preset computer vision model is an image instance segmentation model;

The processing module 602 is used for:

According to the target function category information and the interest point information of the target area, filter the set of first target interest points in the target area that are the same as the target function category;

According to the set of the first target interest points, mark the identifier of the first target interest point on the corresponding position of the satellite image of the target area, and obtain the satellite image marked with the interest point identifier, which is used as the image instance segmentation input data to the model;

The identification module 603 is used for:

The image instance segmentation model is used to segment the satellite image marked with the point of interest identification to identify the contour of the target surface of interest in the target area.

On the basis of the above embodiment, when the processing module 602 marks the identifier of the first target interest point on the corresponding position of the satellite image of the target area according to the set of the first target interest point, use At:

The first target interest points are selected in random order from the set of the first target interest points, and if there is no occlusion with the marked interest point identifiers in the satellite image, the corresponding position in the satellite image of the target area is marked. superior.

In another optional embodiment, the geographic attribute-related data of the target area includes feature information of points of interest of the target area and road network feature information of the target area; the preset computer vision model is image semantics segmentation model;

The processing module 602 is used for:

Determine the center position of the target surface of interest according to the point of interest feature information of the target area and the road network feature information of the target area;

The target area is gridded, and for each grid, a plurality of features of the grid are constructed according to the feature information of the points of interest falling within the grid and the road network feature information within the grid picture;

The center position of the target interest surface and multiple feature maps of each grid are used as input data of the image semantic segmentation model.

On the basis of the above embodiment, the identification module 603 is used for:

Taking the center position of the target surface of interest as the center, the image semantic segmentation model is used to sequentially determine whether each grid around the center position of the target surface of interest belongs to the target surface of interest;

The contour of the target surface of interest in the target area is determined according to the grid positions belonging to the target surface of interest.

On the basis of the above embodiment, when the recognition module 603 sequentially determines whether each grid around the center of the target surface of interest belongs to the target surface of interest through the image semantic segmentation model, it is used for:

For any grid, the multiple feature maps of the grid are spliced and input into the fully convolutional network of the image semantic segmentation model, respectively, to obtain high-dimensional features;

After splicing each of the high-dimensional features, they are input into the convolutional network of the image semantic segmentation model, and the probability that the grid belongs to the target interest surface is obtained;

If the probability is greater than a preset threshold, it is determined that the grid belongs to the target surface of interest.

On the basis of the above embodiment, when determining the center position of the target surface of interest according to the feature information of the point of interest of the target area and the feature information of the road network of the target area, the processing module 602 is used to:

According to the target function category information, filter out a set of second target interest points with the same target function category from the interest points in the target area;

According to the interest point feature information of each second target interest point in the second target interest point set, the road network feature information of the target area, and the preset classification model, from the second target interest point set Identify a representative point of interest in

The position of the representative interest point is taken as the center position of the target interest surface.

On the basis of the above embodiment, the feature information of the points of interest in the target area includes the distribution of functional categories of the points of interest in the target area, the distribution of functional categories of the points of interest adjacent to each point of interest, and the location of each point of interest. .

On the basis of the above embodiment, the geographic attribute-related data of the target area further includes user behavior information in the target area;

When the processing module 602 constructs a plurality of feature maps of the grid according to the feature information of the points of interest falling in the grid and the road network feature information in the grid:

A plurality of feature maps of the grid are constructed according to the feature information of the points of interest falling within the grid, the road network feature information within the grid, and the user behavior information within the grid.

On the basis of any of the foregoing embodiments, the obtaining module 601 is further configured to:

Divide the target city into multiple regions according to the road network information;

A target area selection instruction is received, and the target area to be identified is determined from the plurality of areas according to the target area selection instruction.

The apparatus for recognizing a face of interest provided in this embodiment may be specifically configured to execute the method embodiments provided in the preceding figures, and the specific functions will not be repeated here.

The device for recognizing interest faces provided in this embodiment acquires target function category information corresponding to a target interest facet in a target area to be identified, and data related to geographic attributes of the target area; according to the target function category information, and For the geographic attribute related data of the target area, the input data of the corresponding preset computer vision model is obtained; according to the input data and the corresponding preset computer vision model, the contour of the target surface of interest in the target area is identified. In this embodiment, the computer vision technology is used to realize the identification of the interest surface in the target area, and manual annotation based on the street view image is not required, which can reduce the cost, improve the efficiency, and has a high accuracy rate, which can be applied to different scenarios. The scope of application is large, and it can be applied to the identification of massive interest surfaces, thereby making it possible to describe the surface information of massive interest points.

According to the embodiments of the present application, the present application further provides an electronic device and a readable storage medium.

As shown in FIG. 10 , it is a block diagram of an electronic device according to the method for identifying a face of interest in an embodiment of the present application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are by way of example only, and are not intended to limit implementations of the application described and/or claimed herein.

As shown in FIG. 10, the electronic device includes: one or more processors 701, a memory 702, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. The various components are interconnected using different buses and may be mounted on a common motherboard or otherwise as desired. The processor may process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of the GUI on an external input/output device, such as a display device coupled to the interface. In other embodiments, multiple processors and/or multiple buses may be used with multiple memories and multiple memories, if desired. Likewise, multiple electronic devices may be connected, each providing some of the necessary operations (eg, as a server array, a group of blade servers, or a multiprocessor system). In FIG. 10, a processor 701 is used as an example.

The memory 702 is the non-transitory computer-readable storage medium provided by the present application. Wherein, the memory stores instructions executable by at least one processor, so that the at least one processor executes the method for recognizing the face of interest provided by the present application. The non-transitory computer-readable storage medium of the present application stores computer instructions, and the computer instructions are used to cause the computer to execute the method for recognizing the face of interest provided by the present application.

As a non-transitory computer-readable storage medium, the memory 702 can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as program instructions/modules (for example, The acquisition module 601, the processing module 602 and the identification module 603 shown in FIG. 9). The processor 701 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions and modules stored in the memory 702 , that is, to implement the interest surface identification method in the above method embodiments.

The memory 702 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the electronic device of the interest surface identification method Wait. Additionally, memory 702 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 702 may optionally include memory located remotely relative to the processor 701, and these remote memories may be connected to the electronic device of the face-of-interest recognition method via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The electronic device of the method for recognizing the face of interest may further include: an input device 703 and an output device 704 . The processor 701 , the memory 702 , the input device 703 and the output device 704 may be connected by a bus or in other ways, and the connection by a bus is taken as an example in FIG. 10 .

The input device 703 can receive input numerical or character information, and generate key signal input related to user settings and function control of the electronic device for the face-of-interest recognition method, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, and a pointing stick , one or more mouse buttons, trackballs, joysticks and other input devices. Output devices 704 may include display devices, auxiliary lighting devices (eg, LEDs), haptic feedback devices (eg, vibration motors), and the like. The display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some implementations, the display device may be a touch screen.

Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuit systems, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpretable on a programmable system including at least one programmable processor that The processor, which may be a special purpose or general-purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device an output device.

These computational programs (also referred to as programs, software, software applications, or codes) include machine instructions for programmable processors, and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine languages calculation program. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or apparatus for providing machine instructions and/or data to a programmable processor ( For example, magnetic disks, optical disks, memories, programmable logic devices (PLDs), including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.

To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and can be in any form (including acoustic input, voice input, or tactile input) to receive input from the user.

The systems and techniques described herein may be implemented on a computing system that includes back-end components (eg, as a data server), or a computing system that includes middleware components (eg, an application server), or a computing system that includes front-end components (eg, a user's computer having a graphical user interface or web browser through which a user may interact with implementations of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

A computer system can include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or a cloud host. It is a host product in the cloud computing service system to solve the traditional physical host and VPS services, which are difficult to manage and weak in business scalability. defect.

According to the technical solutions of the embodiments of the present application, by acquiring the target function category information corresponding to the target surface of interest in the target area to be identified, and the geographic attribute related data of the target area; According to the input data and the corresponding preset computer vision model, the outline of the target surface of interest in the target area is identified. In this embodiment, the computer vision technology is used to realize the identification of the interest surface in the target area, and manual annotation based on the street view image is not required, which can reduce the cost, improve the efficiency, and has a high accuracy rate, which can be applied to different scenarios. The scope of application is large, and it can be applied to the identification of massive interest surfaces, thereby making it possible to describe the surface information of massive interest points.

The present application also provides a computer program, including program code, when the computer runs the computer program, the program code executes the method for recognizing a face of interest as described in the foregoing embodiments.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, the steps described in the present application can be performed in parallel, sequentially or in different orders, and as long as the desired results of the technical solutions disclosed in the present application can be achieved, no limitation is imposed herein.

The above-mentioned specific embodiments do not constitute a limitation on the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims (22)

1. A method for identifying a face of interest, comprising: Obtain target function category information corresponding to the target interest surface in the target area to be identified, and geographic attribute-related data of the target area; According to the target function category information and the geographic attribute related data of the target area, obtain the input data of the corresponding preset computer vision model; According to the input data and the corresponding preset computer vision model, the contour of the target surface of interest in the target area is identified. 2. The method according to claim 1, wherein the geographic attribute-related data of the target area includes satellite images of the target area and point-of-interest information of the target area; the preset computer vision model is an image instance segmentation model; The obtaining the input data of the corresponding preset computer vision model according to the target function category information and the geographic attribute related data of the target area, including: According to the target function category information and the interest point information of the target area, filter the set of first target interest points in the target area that are the same as the target function category; According to the set of the first target interest points, mark the identifier of the first target interest point on the corresponding position of the satellite image of the target area, and obtain the satellite image marked with the interest point identifier, which is used as the image instance segmentation input data to the model; The identifying the contour of the target surface of interest in the target area according to the input data and the corresponding preset computer vision model, including: The image instance segmentation model is used to segment the satellite image marked with the point of interest identification to identify the contour of the target surface of interest in the target area. 3. The method according to claim 2, wherein marking the identifier of the first target interest point on the corresponding position of the satellite image of the target area according to the set of the first target interest point, comprising: The first target interest points are selected in random order from the set of the first target interest points, and if there is no occlusion with the marked interest point identifiers in the satellite image, the corresponding position in the satellite image of the target area is marked. superior. 4. The method according to claim 1, wherein the geographic attribute-related data of the target area includes the feature information of points of interest of the target area and the road network feature information of the target area; the preset computer vision model is an image semantic segmentation model; The obtaining the input data of the corresponding preset computer vision model according to the target function category information and the geographic attribute related data of the target area, including: Determine the center position of the target surface of interest according to the point of interest feature information of the target area and the road network feature information of the target area; The target area is gridded, and for each grid, a plurality of features of the grid are constructed according to the feature information of the points of interest falling within the grid and the road network feature information within the grid picture; The center position of the target interest surface and multiple feature maps of each grid are used as input data of the image semantic segmentation model. 5. The method according to claim 4, wherein the identifying the contour of the target surface of interest in the target area according to the input data and the corresponding preset computer vision model comprises: Taking the center position of the target surface of interest as the center, the image semantic segmentation model is used to sequentially determine whether each grid around the center position of the target surface of interest belongs to the target surface of interest; The contour of the target surface of interest in the target area is determined according to the grid positions belonging to the target surface of interest. 6. The method according to claim 5, wherein the step of sequentially judging whether each grid around the center position of the target surface of interest belongs to the target surface of interest through the image semantic segmentation model comprises: For any grid, the multiple feature maps of the grid are spliced and input into the fully convolutional network of the image semantic segmentation model, respectively, to obtain high-dimensional features; After splicing each of the high-dimensional features, they are input into the convolutional network of the image semantic segmentation model, and the probability that the grid belongs to the target interest surface is obtained; If the probability is greater than a preset threshold, it is determined that the grid belongs to the target surface of interest. 7. The method according to claim 4, wherein the determining the center position of the target surface of interest according to the feature information of points of interest of the target area and the road network feature information of the target area comprises: According to the target function category information, filter out a set of second target interest points with the same target function category from the interest points in the target area; According to the interest point feature information of each second target interest point in the second target interest point set, the road network feature information of the target area, and the preset classification model, from the second target interest point set Identify a representative point of interest in The position of the representative interest point is taken as the center position of the target interest surface. 8. The method according to any one of claims 4-7, wherein the feature information of the points of interest of the target area includes the distribution of function categories of the points of interest in the target area, the functions of the points of interest adjacent to each point of interest Category distribution, location of each point of interest. 9. The method according to any one of claims 4-7, wherein the geographic attribute-related data of the target area further includes user behavior information in the target area; The constructing a plurality of feature maps of the grid according to the feature information of the points of interest falling in the grid and the road network feature information in the grid, including: A plurality of feature maps of the grid are constructed according to the feature information of the points of interest falling within the grid, the road network feature information within the grid, and the user behavior information within the grid. 10. The method of claim 1, further comprising: Divide the target city into multiple regions according to the road network information; A target area selection instruction is received, and the target area to be identified is determined from the plurality of areas according to the target area selection instruction. 11. A device for face-of-interest recognition, comprising: an acquisition module, configured to acquire target function category information corresponding to the target surface of interest in the target area to be identified, and geographic attribute-related data of the target area; a processing module, configured to obtain input data of the corresponding preset computer vision model according to the target function category information and the geographic attribute-related data of the target area; The identification module is configured to identify the contour of the target surface of interest in the target area according to the input data and the corresponding preset computer vision model. 12. The apparatus according to claim 11, wherein the geographic attribute-related data of the target area includes satellite images of the target area and point-of-interest information of the target area; the preset computer vision model is an image instance segmentation model; The processing module is used for: According to the target function category information and the interest point information of the target area, filter the set of first target interest points in the target area that are the same as the target function category; According to the set of the first target interest points, mark the identifier of the first target interest point on the corresponding position of the satellite image of the target area, and obtain the satellite image marked with the interest point identifier, which is used as the image instance segmentation input data to the model; The identification module is used for: The image instance segmentation model is used to segment the satellite image marked with the point of interest identification to identify the contour of the target surface of interest in the target area. 13. The apparatus according to claim 12, when the processing module marks the identifier of the first target interest point on the corresponding position of the satellite image of the target area according to the set of the first target interest point , for: The first target interest points are selected in random order from the set of the first target interest points, and if there is no occlusion with the marked interest point identifiers in the satellite image, the corresponding position in the satellite image of the target area is marked. superior. 14. The apparatus according to claim 11, wherein the geographic attribute-related data of the target area includes feature information of points of interest of the target area and road network feature information of the target area; the preset computer vision model is an image semantic segmentation model; The processing module is used for: Determine the center position of the target surface of interest according to the point of interest feature information of the target area and the road network feature information of the target area; The target area is gridded, and for each grid, a plurality of features of the grid are constructed according to the feature information of the points of interest falling within the grid and the road network feature information within the grid picture; The center position of the target interest surface and multiple feature maps of each grid are used as input data of the image semantic segmentation model. 15. The apparatus of claim 14, wherein the identification module is used to: Taking the center position of the target surface of interest as the center, the image semantic segmentation model is used to sequentially determine whether each grid around the center position of the target surface of interest belongs to the target surface of interest; The contour of the target surface of interest in the target area is determined according to the grid positions belonging to the target surface of interest. 16. The device according to claim 15, wherein when the recognition module sequentially determines whether each grid around the center position of the target surface of interest belongs to the target surface of interest by using the image semantic segmentation model, it is used for : For any grid, the multiple feature maps of the grid are spliced and input into the fully convolutional network of the image semantic segmentation model, respectively, to obtain high-dimensional features; After splicing each of the high-dimensional features, they are input into the convolutional network of the image semantic segmentation model, and the probability that the grid belongs to the target interest surface is obtained; If the probability is greater than a preset threshold, it is determined that the grid belongs to the target surface of interest. 17. The device according to claim 14, wherein, when the processing module determines the center position of the target surface of interest according to the feature information of the point of interest of the target area and the feature information of the road network of the target area, Used for: According to the target function category information, filter out a set of second target interest points with the same target function category from the interest points in the target area; According to the interest point feature information of each second target interest point in the second target interest point set, the road network feature information of the target area, and the preset classification model, from the second target interest point set Identify a representative point of interest in The position of the representative interest point is taken as the center position of the target interest surface. 18. The apparatus according to any one of claims 14-17, wherein the feature information of the points of interest of the target area includes the distribution of function categories of the points of interest in the target area, the functions of the points of interest adjacent to each point of interest Category distribution, location of each point of interest. 19. The apparatus according to any one of claims 14-17, wherein the geographic attribute-related data of the target area further includes user behavior information in the target area; When the processing module constructs a plurality of feature maps of the grid according to the feature information of the points of interest falling in the grid and the road network feature information in the grid, it is used for: A plurality of feature maps of the grid are constructed according to the feature information of the points of interest falling within the grid, the road network feature information within the grid, and the user behavior information within the grid. 20. The apparatus of claim 11, wherein the obtaining module is further configured to: Divide the target city into multiple regions according to the road network information; A target area selection instruction is received, and the target area to be identified is determined from the plurality of areas according to the target area selection instruction. 21. An electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the execution of any of claims 1-10 Methods. 22. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method of any of claims 1-10.

Images