A method and device for enhancing spatiotemporal trajectory data quality based on deep learning
Through deep learning technology, image super-resolution reconstruction and convolutional neural networks are used to process low-quality trajectory data, solving the problem of inconsistent sampling rate and noise in spatiotemporal trajectory data, improving data quality, and providing high-quality trajectory data mining enter.
Patent Information
- Application Number
- CN202110631405.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-06-07
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2041-06-07
AI Technical Summary
The prior art is susceptible to inconsistent sampling rates and noise when processing spatiotemporal trajectory data, resulting in low data quality. The RNN-based methods have problems such as error accumulation and inability to consider future inputs during the prediction process.
Using a deep learning-based approach, by constructing trajectory image training pairs, convert low-resolution trajectory images into high-resolution images using image super-resolution reconstruction technology and convolutional neural networks, and embed them into low-dimensional vector space to improve data quality.
It effectively overcomes the low-quality trajectory problems caused by inconsistent sampling rates and noise, generates high-quality trajectory representations, and provides good input data for spatiotemporal trajectory data mining.
Smart Images

Figure CN113592710B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of spatiotemporal data mining, and in particular to a method and device for enhancing the quality of spatiotemporal trajectory data based on deep learning. Background Art
[0002] With the rise of mobile terminals and the increasing maturity of positioning technology, applications based on location information have been used more widely than ever before in people's production and life. However, the trajectory data collected in reality generally has a certain degree of low quality, which will significantly affect the effect of trajectory data mining. This low quality is mainly reflected in two aspects:
[0003] 1. Inconsistent sampling rate: Due to different sensors, sensor settings, and communication conditions, different trajectories often have different sampling rates. Even for the same trajectory, there will be different sampling rates on different trajectory segments;
[0004] 2. Noise: In reality, trajectory data is usually noisy, that is, the coordinate values of trajectory points will deviate from their true values to a certain extent. These noises are usually caused by signal blocking, atmospheric conditions, and the design of the sensor itself.
[0005] In the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art:
[0006] 1. Trajectory data mining using trajectory sequences as trajectory representation is easily affected by inconsistent sampling rates and noise.
[0007] 2. The trajectory representation method based on recurrent neural network (RNN) uses sequence prediction to repair the missing or noise-interfered trajectory points. However, the RNN-based method will encounter the problem of error accumulation in the testing phase, that is, the model relies on the results of the previous step prediction to predict the next trajectory point, and the previously predicted trajectory point may not exist in the training data, so the prediction error will accumulate in the process of generating the trajectory sequence. In addition, RNN's prediction of trajectory points can only be based on previous inputs and cannot take into account subsequent inputs. But in fact, the next step of the trajectory position point depends not only on the position of the previous step, but also on the subsequent position.
[0008] In summary, current spatiotemporal trajectory mining encounters the problem of low-quality trajectories caused by inconsistent sampling rates and noise, and existing methods have the shortcomings listed above. How to overcome the low-quality problem of trajectories and thereby improve the quality of trajectory data is the problem to be solved by the present invention. Summary of the invention
[0009] The purpose of the embodiments of the present invention is to provide a method and device for enhancing the quality of spatiotemporal trajectory data based on deep learning, so as to solve the problem that inconsistent sampling rate and noise affect the quality of trajectory data.
[0010] According to a first aspect of an embodiment of the present application, a method for enhancing the quality of spatiotemporal trajectory data based on deep learning is provided, comprising:
[0011] Obtaining original spatiotemporal trajectory data within a predetermined space;
[0012] constructing a trajectory image training pair according to the original spatiotemporal trajectory data, the trajectory image training pair comprising a low-resolution trajectory image and a corresponding high-resolution trajectory image;
[0013] Taking the low-resolution trajectory image as input, using image super-resolution reconstruction technology to generate a super-resolution trajectory image;
[0014] The super-resolution trajectory image and the corresponding high-resolution trajectory image are embedded in a low-dimensional vector space using an image embedding model based on a convolutional neural network to obtain a low-dimensional trajectory vector, thereby enhancing the quality of spatiotemporal trajectory data.
[0015] Furthermore, obtaining the original spatiotemporal trajectory data in the predetermined space includes:
[0016] A predetermined area is selected from the map, wherein there is more than a predetermined amount of original spatiotemporal trajectory data in the area.
[0017] Furthermore, after obtaining the original spatiotemporal trajectory data in the predetermined space, the method further includes:
[0018] The region is scaled to obtain an image of a predetermined size.
[0019] Furthermore, constructing a trajectory image training pair according to the original spatiotemporal trajectory data includes:
[0020] Performing a downsampling operation on the trajectory points in the original spatiotemporal trajectory data;
[0021] After the downsampling operation, performing noise addition processing to obtain low-quality trajectory data;
[0022] generating a low-resolution trajectory image according to the low-quality trajectory data;
[0023] Using the original trajectory data as corresponding high-quality trajectory data to generate a high-resolution trajectory image;
[0024] The low-resolution trajectory image and the high-resolution trajectory image are combined into a trajectory image training pair.
[0025] According to a second aspect of an embodiment of the present application, a device for enhancing spatiotemporal trajectory data quality based on deep learning is provided, comprising:
[0026] An acquisition module, used to acquire original spatiotemporal trajectory data within a predetermined space;
[0027] A construction module, configured to construct a trajectory image training pair according to the original spatiotemporal trajectory data, wherein the trajectory image training pair includes a low-resolution trajectory image and a corresponding high-resolution trajectory image;
[0028] A generating module, configured to use the low-resolution trajectory image as input and generate a super-resolution trajectory image using an image super-resolution reconstruction technique;
[0029] The quality enhancement module is used to use an image embedding model based on a convolutional neural network to embed the super-resolution trajectory image and the corresponding high-resolution trajectory image into a low-dimensional vector space to obtain a low-dimensional trajectory vector, thereby enhancing the quality of the spatiotemporal trajectory data.
[0030] Furthermore, the acquisition module includes:
[0031] A predetermined area is selected from the map, wherein there is more than a predetermined amount of original spatiotemporal trajectory data in the area.
[0032] Furthermore, after obtaining the module, it also includes:
[0033] The scaling module is used to scale the area to obtain an image of a predetermined size.
[0034] Furthermore, the building blocks include:
[0035] A downsampling submodule, used for performing a downsampling operation on the trajectory points in the original spatiotemporal trajectory data;
[0036] A noise processing submodule, used for performing noise processing after the downsampling operation to obtain low-quality trajectory data;
[0037] A first generating submodule, configured to generate a low-resolution trajectory image according to the low-quality trajectory data;
[0038] A second generating submodule, configured to use the original trajectory data as corresponding high-quality trajectory data to generate a high-resolution trajectory image;
[0039] The combining submodule is used to combine the low-resolution trajectory image and the high-resolution trajectory image into a trajectory image training pair.
[0040] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in the first aspect.
[0041] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method described in the first aspect are implemented.
[0042] The technical solution provided by the embodiments of the present application may have the following beneficial effects:
[0043] It can be seen from the above embodiments that the present application enhances the quality of trajectory data based on deep learning, and uses image super-resolution reconstruction technology to reconstruct lost details and inaccurate parts in low-quality trajectories, effectively overcoming the problem of low-quality trajectory data caused by inconsistent trajectory sampling rate and noise, thereby obtaining high-quality trajectory representation and providing high-quality input data for trajectory data mining tasks.
[0044] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0046] Figure 1 The present invention is a flowchart of a method for enhancing the quality of spatiotemporal trajectory data based on deep learning according to an exemplary embodiment.
[0047] Figure 2 is a schematic diagram of a selected predetermined area according to an exemplary embodiment.
[0048] Figure 3 is a flowchart of step S12 according to an exemplary embodiment.
[0049] Figure 4 1 is a flowchart of converting trajectory data into a trajectory image according to an exemplary embodiment, wherein (a) is the mapping of trajectory coordinates onto a regional image, (b) is the division of the image into a grid, (c) is the assignment of pixels within the grid, and (d) is the acquisition of a single-channel grayscale image.
[0050] Figure 5 is a schematic diagram of a trajectory image training pair according to an exemplary embodiment, wherein (a) is an LR image and (b) is an HR image;
[0051] Figure 6 It is a model framework diagram shown according to an exemplary embodiment.
[0052] Figure 7 It is a block diagram of a device for enhancing spatiotemporal trajectory data quality based on deep learning according to an exemplary embodiment.
[0053] Figure 8 is a block diagram of building modules according to an exemplary embodiment. DETAILED DESCRIPTION
[0054] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.
[0055] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0056] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0057] Figure 1 is a flowchart of a method for enhancing the quality of spatiotemporal trajectory data based on deep learning according to an exemplary embodiment. Figure 1 , an embodiment of the present invention provides a method for enhancing the quality of spatiotemporal trajectory data based on deep learning, and the method may include:
[0058] Step S11, obtaining original spatiotemporal trajectory data in a predetermined space;
[0059] Step S12, constructing a trajectory image training pair according to the original spatiotemporal trajectory data, wherein the trajectory image training pair includes a low-resolution trajectory image and a corresponding high-resolution trajectory image;
[0060] Step S13, taking the low-resolution trajectory image as input, and using image super-resolution reconstruction technology to generate a super-resolution trajectory image;
[0061] Step S14, using a convolutional neural network-based image embedding model to embed the super-resolution trajectory image and the corresponding high-resolution trajectory image into a low-dimensional vector space to obtain a low-dimensional trajectory vector, thereby enhancing the quality of the spatiotemporal trajectory data.
[0062] It can be seen from the above embodiments that the present application enhances the quality of trajectory data based on deep learning, and uses image super-resolution reconstruction technology to reconstruct lost details and inaccurate parts in low-quality trajectories, effectively overcoming the problem of low-quality trajectory data caused by inconsistent trajectory sampling rate and noise, thereby obtaining high-quality trajectory representation and providing high-quality input data for trajectory data mining tasks.
[0063] In the specific implementation of step S11, obtaining the original spatiotemporal trajectory data in the predetermined space includes:
[0064] A predetermined area is selected from the map, wherein there is more than a predetermined amount of original spatiotemporal trajectory data in the area.
[0065] Specifically, for a given original spatiotemporal trajectory data set, the spatial range of the trajectory is determined according to the coordinate information of the original spatiotemporal trajectory data, and a predetermined area is selected in the two-dimensional map corresponding to the spatial range, such as Figure 2 As shown, a predetermined amount of original spatiotemporal trajectory data is required to exist in the area.
[0066] After obtaining the original spatiotemporal trajectory data in the predetermined space, the following steps may also be performed:
[0067] Specifically, in order to obtain a trajectory image, it is necessary to determine the area represented by the image and the size of the image. The predetermined area is taken as the area represented by the image, and according to the predetermined image size, the predetermined area is scaled to an image of a predetermined size. The image is used as a base, and all trajectory data generated in the predetermined area can be mapped to corresponding pixels of the image, so that the trajectory data can be converted into a trajectory image.
[0068] In the specific implementation of step S12, a trajectory image training pair is constructed based on the original spatiotemporal trajectory data. Figure 3 , the step S12 may include the following sub-steps:
[0069] Step S21, performing a downsampling operation on the trajectory points in the original spatiotemporal trajectory data;
[0070] Specifically, the trajectory points in the trajectory are randomly deleted according to a certain probability to obtain a trajectory data with a lower sampling rate; specifically, given the original trajectory sequence Where k represents the number of sampling points in the trajectory. We consider high sampling rate trajectories And downsample it to get a trajectory with a relatively low sampling rate Where k'≤k. and Both are sampled from the same real path R, and the only difference between them is The sampling rate of is higher and therefore closer to the true path R.
[0071] Step S22, after the downsampling operation, performing noise addition processing to obtain low-quality trajectory data;
[0072] Specifically, to simulate the noise in trajectory data, we Each trajectory point in Artificially add Gaussian noise with a radius of 50 meters:
[0073] x=x+50·d x ,d x ~N(0,1)
[0074] y=y+50·d y ,d y ~N(0,1)
[0075] Step S23, generating a low-resolution trajectory image according to the low-quality trajectory data;
[0076] Specifically, the image base is first divided into grids of equal size, and then the coordinates of the trajectory points of the spatiotemporal trajectory data are mapped to the corresponding grids of the image base. This operation ensures that the coordinates of the trajectory points are mapped to a spatial area of a predetermined size rather than a point in space, thereby alleviating the sparseness problem of the trajectory image; finally, the pixel value is set for the grid area corresponding to the trajectory point to obtain a single-channel grayscale trajectory image. The pixel value depends on the number of trajectory points in the grid. The more trajectory points there are, the larger the pixel value. The specific process is as follows: Figure 4 As shown. According to the above process, it is converted into a low-resolution trajectory image. Where W and H are the width and height of the image respectively, and G represents the generation process of the image, such as Figure 5 (a) in .
[0077] Step S24, using the original trajectory data as corresponding high-quality trajectory data to generate a high-resolution trajectory image;
[0078] Specifically, the original trajectory data is converted into a high-resolution trajectory image according to the process described in step S23 using a 2-fold magnification factor as the corresponding high-quality trajectory data. like Figure 5 (b) in .
[0079] Step S25 , combining the low-resolution trajectory image and the high-resolution trajectory image into a trajectory image training pair.
[0080] In the specific implementation of step S13, the low-resolution trajectory image is used as input, and an image super-resolution reconstruction technology is used to generate a super-resolution trajectory image;
[0081] Specifically, the super-resolution generation of a single image is realized based on the FSRCNN super-resolution model, such as Figure 6 As shown in the FSRCNN in , Conv represents the convolution layer, BN represents the batch normalization operation, ReLU is the activation function, and DeConv represents the deconvolution layer. The model is mainly divided into five parts: feature extraction, contraction, nonlinear mapping, expansion and deconvolution. In the feature extraction part, the model uses a 5×5 convolution kernel to extract the features of the low-resolution image; the contraction part uses a 1×1 convolution kernel to reduce the network parameters and reduce the computational complexity; the nonlinear mapping part uses multiple layers of 3×3 convolution layers to expand the receptive field; the expansion part uses a 1×1 convolution kernel to expand the dimension; and finally the deconvolution layer is used to achieve upsampling of the image. The advantage of using this model is that it enlarges the image size through the final deconvolution layer, so the low-resolution image can be directly input into the input of the model.
[0082] The model uses image mean square error as the loss function:
[0083]
[0084] In the specific implementation of step S14, the super-resolution trajectory image and the corresponding high-resolution trajectory image are embedded in a low-dimensional vector space using an image embedding model based on a convolutional neural network to obtain a low-dimensional trajectory vector, thereby enhancing the quality of the spatiotemporal trajectory data.
[0085] Specifically, since there are a large number of zero-value pixels in the trajectory image, the output of the super-resolution model, i.e., the trajectory image, is not suitable for directly representing the trajectory. Therefore, the image embedding model based on convolutional neural network is used to further embed the super-resolution image I SR and the target image I HR Embedded into a low-dimensional vector space, such as Figure 6 As shown in the Embed Block in . Maxpooling represents the pooling operation, which is used to reduce the size of the image; LeakyRelu is the activation function; reshape means expanding the image after the size reduction into a one-dimensional vector as the final trajectory representation vector. This not only can obtain a more concise trajectory representation, but also can eliminate redundant information in the trajectory image, which is helpful for subsequent trajectory data mining tasks.
[0086] It should be noted here that Figure 6 The model shown consists of two parts: 1) a super-resolution model based on FSRCNN, and 2) an image embedding model based on convolutional neural network. The two parts are cascaded together to achieve the purpose of enhancing trajectory quality.
[0087] The optimization goal of the image embedding model based on convolutional neural network is to make the trajectory representation vector after the low-quality trajectory data is embedded as similar as possible to the trajectory representation vector after the high-quality trajectory data is embedded. SR and I HR The embedded vector v SR and v HR The similarity error between:
[0088]
[0089] Where |v| represents the dimension of vector v.
[0090] In order to achieve super-resolution image I SR Can be visually perceived with the target image I HR Similar, and I SR Can be embedded into a better low-dimensional vector representation, using the weighted sum of the similarity error and the image mean square error as the perceptual loss to optimize the super-resolution model:
[0091] l SR =λ·l MSE +μ·l vec
[0092] Where λ and μ are weight coefficients.
[0093] The optimization process of the image embedding model based on convolutional neural network is as follows:
[0094]
[0095] Corresponding to the aforementioned embodiment of a method for enhancing the quality of spatiotemporal trajectory data based on deep learning, the present application also provides an embodiment of a device for enhancing the quality of spatiotemporal trajectory data based on deep learning.
[0096] Figure 7 is a block diagram of a device for enhancing the quality of spatiotemporal trajectory data based on deep learning according to an exemplary embodiment. Figure 7 , the device comprises:
[0097] An acquisition module 31 is used to acquire original spatiotemporal trajectory data in a predetermined space;
[0098] A construction module 32, configured to construct a trajectory image training pair according to the original spatiotemporal trajectory data, wherein the trajectory image training pair includes a low-resolution trajectory image and a corresponding high-resolution trajectory image;
[0099] A generating module 33, configured to use the low-resolution trajectory image as input and generate a super-resolution trajectory image using an image super-resolution reconstruction technique;
[0100] The quality enhancement module 34 is used to use the image embedding model based on the convolutional neural network to embed the super-resolution trajectory image and the corresponding high-resolution trajectory image into a low-dimensional vector space to obtain a low-dimensional trajectory vector, thereby enhancing the quality of the spatiotemporal trajectory data.
[0101] In the embodiment of the present invention, the acquisition module 31 includes:
[0102] A predetermined area is selected from the map, wherein there is more than a predetermined amount of original spatiotemporal trajectory data in the area.
[0103] In the embodiment of the present invention, after obtaining the module, the following is further included:
[0104] The scaling module is used to scale the area to obtain an image of a predetermined size.
[0105] In the embodiment of the present invention, reference Figure 8 , building block 32, the following sub-modules include:
[0106] A downsampling submodule 41 is used to perform a downsampling operation on the trajectory points in the original spatiotemporal trajectory data;
[0107] A noise processing submodule 42, configured to perform noise processing after the downsampling operation to obtain low-quality trajectory data;
[0108] A first generating submodule 43, configured to generate a low-resolution trajectory image according to the low-quality trajectory data;
[0109] A second generating submodule 44 is used to generate a high-resolution trajectory image by using the original trajectory data as corresponding high-quality trajectory data;
[0110] The combining submodule 45 is used to combine the low-resolution trajectory image and the high-resolution trajectory image into a trajectory image training pair.
[0111] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0112] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.
[0113] Correspondingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement a method for enhancing the quality of spatiotemporal trajectory data based on deep learning as described above.
[0114] Correspondingly, the present application also provides a computer-readable storage medium on which computer instructions are stored, characterized in that when the instructions are executed by a processor, a method for enhancing the quality of spatiotemporal trajectory data based on deep learning as described above is implemented.
[0115] Those skilled in the art will readily appreciate other embodiments of the present application after considering the description and practicing the contents disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary techniques in the art that are not disclosed in the present application. The description and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.
[0116] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.
Claims
1. A method for enhancing the quality of spatiotemporal trajectory data based on deep learning, characterized in that: include: Obtaining original spatiotemporal trajectory data within a predetermined space; constructing a trajectory image training pair according to the original spatiotemporal trajectory data, the trajectory image training pair comprising a low-resolution trajectory image and a corresponding high-resolution trajectory image; The low-resolution trajectory image is used as input, and a super-resolution trajectory image is generated using an image super-resolution reconstruction technology; specifically: the image super-resolution reconstruction technology is based on the FSRCNN super-resolution model to achieve super-resolution generation of a single image; the model is divided into five parts: feature extraction, contraction, nonlinear mapping, expansion and deconvolution; the model uses a convolution kernel of size 5×5 in the feature extraction part to extract features of the low-resolution image; The shrinking part uses a 1×1 convolution kernel; The nonlinear mapping part uses multiple 3×3 convolution layers; the expansion part uses a 1×1 convolution kernel; and finally the image is upsampled through the deconvolution layer. The super-resolution trajectory image and the corresponding high-resolution trajectory image are embedded into a low-dimensional vector space using an image embedding model based on a convolutional neural network to obtain a low-dimensional trajectory vector, thereby enhancing the quality of spatiotemporal trajectory data.
2. The method according to claim 1, characterized in that Get the original spatiotemporal trajectory data within the predetermined space, including: A predetermined area is selected from the map, wherein there is more than a predetermined amount of original spatiotemporal trajectory data in the area.
3. The method according to claim 2, characterized in that After obtaining the original spatiotemporal trajectory data in the predetermined space, it also includes: The region is scaled to obtain an image of a predetermined size.
4. The method according to claim 1, characterized in that: According to the original spatiotemporal trajectory data, a trajectory image training pair is constructed, including: Performing a downsampling operation on the trajectory points in the original spatiotemporal trajectory data; After the downsampling operation, performing noise addition processing to obtain low-quality trajectory data; generating a low-resolution trajectory image according to the low-quality trajectory data; Using the original trajectory data as corresponding high-quality trajectory data to generate a high-resolution trajectory image; The low-resolution trajectory image and the high-resolution trajectory image are combined into a trajectory image training pair.
5. A device for enhancing the quality of spatiotemporal trajectory data based on deep learning, characterized in that: include: An acquisition module, used to acquire original spatiotemporal trajectory data within a predetermined space; A construction module, configured to construct a trajectory image training pair according to the original spatiotemporal trajectory data, wherein the trajectory image training pair includes a low-resolution trajectory image and a corresponding high-resolution trajectory image; A generation module, used to use the low-resolution trajectory image as input and generate a super-resolution trajectory image using an image super-resolution reconstruction technology; specifically: the image super-resolution reconstruction technology is based on the FSRCNN super-resolution model to achieve super-resolution generation of a single image; the model is divided into five parts: feature extraction, contraction, nonlinear mapping, expansion and deconvolution; the model uses a convolution kernel of size 5×5 in the feature extraction part to extract features of the low-resolution image; The shrinking part uses a 1×1 convolution kernel; The nonlinear mapping part uses multiple 3×3 convolution layers; the expansion part uses a 1×1 convolution kernel; and finally the image is upsampled through the deconvolution layer. The quality enhancement module is used to embed the super-resolution trajectory image and the corresponding high-resolution trajectory image into a low-dimensional vector space by using an image embedding model based on a convolutional neural network to obtain a low-dimensional trajectory vector, thereby enhancing the quality of the spatiotemporal trajectory data.
6. The device according to claim 5, characterized in that Get modules, including: A predetermined area is selected from the map, wherein there is more than a predetermined amount of original spatiotemporal trajectory data in the area.
7. The device according to claim 6, characterized in that After getting the module, it also includes: The scaling module is used to scale the area to obtain an image of a predetermined size.
8. The device according to claim 5, characterized in that Building blocks, including: A downsampling submodule, used for performing a downsampling operation on the trajectory points in the original spatiotemporal trajectory data; A noise processing submodule, used for performing noise processing after the downsampling operation to obtain low-quality trajectory data; A first generating submodule, configured to generate a low-resolution trajectory image according to the low-quality trajectory data; A second generating submodule, configured to use the original trajectory data as corresponding high-quality trajectory data to generate a high-resolution trajectory image; The combining submodule is used to combine the low-resolution trajectory image and the high-resolution trajectory image into a trajectory image training pair.
9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4.
10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.
