Deep learning coordinate prediction using satellite and service data
View Patent ↗Systems and methods of deep learning coordinate prediction using satellite and service data are disclosed herein. In some example embodiments, for each one of a plurality of places, a computer system trains a deep learning model based on training data of the plurality of places. The deep leaning model is configured to generate a predicted geographical location of a place based on satellite image data and service data associated with the place. The training data for each place comprises satellite image data of the place, service data, and a ground truth geographical location of the place. The service data comprises at least one of pick-up data indicating a geographical location at which a provider started transporting a requester in servicing a request associated with the place or drop-off data indicating a geographical location at which the provider completed transporting the requester in servicing the request associated with the place.
1. A computer-implemented method comprising:
for each one of a plurality of places, receiving, by a computer system having at least one hardware processor, training data associated with the place of the plurality of places, the training data comprising satellite image data of the place, service data, and a ground truth geographical location of the place, the service data comprising at least one of pick-up data or drop-off data, the pick-up data indicating a geographical location at which a provider started transporting a requester in servicing a request associated with the place, the drop-off data indicating a geographical location at which the provider completed transporting the requester in servicing the request associated with the place;
training, by the computer system, a deep learning model based on the training data of the plurality of places, the deep learning model configured to generate a predicted geographical location of another place based on satellite image data and service data associated with the other place;
receiving, by the computer system, satellite image data and service data for the other place not included in the plurality of places, the service data for the other place comprising at least one of pick-up data or drop-off data of the other place, the pick-up data of the other place indicating a geographical location at which another provider started transporting another requester in servicing another request associated with the other place, the drop-off data of the other place indicating a geographical location at which the other provider completed transporting the other requester in servicing the other request associated with the other place; and
generating, by the computer system, the predicted geographical location of the other place using the deep learning model based on the received satellite image data and the service data for the other place.
2. The computer-implemented method of claim 1 , further comprising:
storing, by the computer system, the predicted geographical location of the other place in a database in association with an identification of the other place;
receiving, by the computer system, a third request for a service associated with the other place from a computing device of a third requester, the third request comprising the identification of the other place;
identifying, by the computer system, the predicted geographical location of the other place based on a search of the database using the identification of the other place; and
transmitting, by the computer system, the predicted geographical location of the other place to a computing device of a third provider of the service of the third request.
3. The computer-implemented method of claim 1 , wherein the service data of the training data comprises a geocode, the service data for the other place comprises a geocode, and the predicted geographical location of the other place comprises a geocode.
4. The computer-implemented method of claim 1 , wherein the deep learning model is trained using a convolutional neural network.
5. The computer-implemented method of claim 1 , wherein the service data of the training data comprises global positioning system (GPS) data of a mobile device of the provider corresponding to the request corresponding to the service data of the training data; and the service data for the other place comprises GPS data of a mobile device of the other provider corresponding to the other request corresponding to the service data for the other place.
6. The computer-implemented method of claim 1 , wherein the predicted geographical location of the other place comprises a subsection of the satellite image data of the other place.
7. The computer-implemented method of claim 1 , wherein the training of the deep learning model comprises:
for each one of the plurality of places, generating integrated data by superimposing indications of the corresponding service data on the corresponding satellite image data; and
training the deep learning model using the integrated data.
8. The computer-implemented method of claim 1 , wherein the training of the deep learning model comprises, for each one of the plurality of places, passing the corresponding satellite image data to the deep learning model separately from the corresponding service data.
9. The computer-implemented method of claim 1 , wherein the training of the deep learning model comprises training a plurality of deep learning models based on the training data of the plurality of places, and the generating the predicted geographical location of the other place comprises:
for each one of the plurality of deep learning models, generating a preliminary predicted geographical location of the other place using the one of the deep learning models based on the received satellite image data and the service data for the other place; and
determining the predicted geographical location based on the preliminary predicted geographical locations generated using the plurality of deep learning models.
10. The computer-implemented method of claim 9 , wherein the determining the predicted geographical location comprises selecting the preliminary predicted geographical location that is most common among the preliminary predicted geographical locations generated using the plurality of deep learning models.
11. The computer-implemented method of claim 9 , wherein the determining the predicted geographical location comprises determining an average of the preliminary predicted geographical locations generated using the plurality of deep learning models.
12. The computer-implemented method of claim 1 , further comprising:
identifying a single category of places to which all of the plurality of places belong; and
selecting the deep learning model from a plurality of deep learning models for use in the training of the deep learning model based on the identifying of the single category.
13. The computer-implemented method of claim 1 , further comprising:
identifying a category of places to which the other place belongs; and
selecting the deep learning model for use in the generating of the predicted geographical location of the other place based on the identifying of the category.
14. The computer-implemented method of claim 1 , wherein, for each one of the plurality of places, the use of the training data of the place in the training of the deep learning model is based on a determination that an amount of the service data of the training data of the place satisfies a minimum threshold amount.
15. A system comprising:
at least one hardware processor; and
a non-transitory machine-storage medium embodying a set of instructions that, when executed by the at least one hardware processor, cause the at least one hardware processor to perform operations, the operations comprising:
for each one of a plurality of places, receiving training data associated with the place of the plurality of places, the training data comprising satellite image data of the place, service data, and a ground truth geographical location of the place, the service data comprising at least one of pick-up data or drop-off data, the pick-up data indicating a geographical location at which a provider started transporting a requester in servicing a request associated with the place, the drop-off data indicating a geographical location at which the provider completed transporting the requester in servicing the request associated with the place;
training a deep learning model based on the training data of the plurality of places, the deep learning model configured to generate a predicted geographical location of another place based on satellite image data and service data associated with the other place;
receiving satellite image data and service data for the other place not included in the plurality of places, the service data for the other place comprising at least one of pick-up data or drop-off data of the other place, the pick-up data of the other place indicating a geographical location at which another provider started transporting a another requester in servicing another request associated with the other place, the drop-off data of the other place indicating a geographical location at which the other provider completed transporting the other requester in servicing the other request associated with the other place; and
generating the predicted geographical location of the other place using the deep learning model based on the received satellite image data and the service data for the other place.
16. The system of claim 15 , wherein the operations further comprise storing the predicted geographical location of the other place in a database in association with an identification of the other place.
17. The system of claim 16 , wherein the operations further comprise:
receiving a third request for a service associated with the other place from a computing device of a third requester, the third request comprising the identification of the other place;
identifying the predicted geographical location of the other place based on a search of the database using the identification of the other place; and
transmitting the predicted geographical location of the other place to a computing device of a third provider of the service of the third request.
18. The system of claim 15 , wherein the service data of the training data comprises a geocode, the service data for the other place comprises a geocode, and the predicted geographical location of the other place comprises a geocode.
19. The system of claim 15 , wherein the deep learning model is trained using a convolutional neural network.
20. A non-transitory machine-storage medium embodying a set of instructions that, when executed by at least one hardware processor, cause the at least one hardware processor to perform operations, the operations comprising:
for each one of a plurality of places, receiving training data associated with the place, the training data comprising satellite image data of the place, service data, and a ground truth geographical location of the place, the service data comprising at least one of pick-up data and drop-off data, the pick-up data indicating a geographical location at which a provider started transporting a requester in servicing a request associated with the place, the drop-off data indicating a geographical location at which the provider completed transporting the requester in servicing the request associated with the place;
training a deep learning model based on the training data of the plurality of places, the deep learning model configured to generate a predicted geographical location of a place based on satellite image data and service data associated with the place;
receiving satellite image data and service data for another place not included in the plurality of places, the service data for the other place comprising at least one of pick-up data and drop-off data, the pick-up data of the other place indicating a geographical location at which a provider started transporting a requester in servicing a request associated with the other place, the drop-off data of the other place indicating a geographical location at which the provider completed transporting the requester in servicing the request associated with the other place; and
generating a predicted geographical location of the other place using the deep learning model based on the received satellite image data and the service data for the other place.