CN111382706A - Prediction method and device based on federal learning, storage medium and remote sensing equipment - Google Patents
Prediction method and device based on federal learning, storage medium and remote sensing equipment Download PDFInfo
- Publication number
- CN111382706A CN111382706A CN202010164191.0A CN202010164191A CN111382706A CN 111382706 A CN111382706 A CN 111382706A CN 202010164191 A CN202010164191 A CN 202010164191A CN 111382706 A CN111382706 A CN 111382706A
- Authority
- CN
- China
- Prior art keywords
- federal
- prediction
- local
- data
- remote sensing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 84
- 230000005540 biological transmission Effects 0.000 claims abstract description 11
- 230000004931 aggregating effect Effects 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 22
- 238000007781 pre-processing Methods 0.000 claims description 21
- 230000008569 process Effects 0.000 description 9
- 238000004891 communication Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 238000012937 correction Methods 0.000 description 4
- 239000003208 petroleum Substances 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a prediction method, a prediction device, a storage medium and remote sensing equipment based on federal learning, wherein the method comprises the following steps: training the initial model according to local training data to obtain a unilateral prediction model; transmitting the unilateral prediction model to a ground information center, and aggregating a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices by the ground information center to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels; and receiving a corresponding federal submodel, and predicting local data to be predicted based on the federal submodel to obtain a prediction label. Therefore, model training is performed locally on the remote sensing equipment, comprehensive utilization of high-altitude data is achieved based on federal learning, and the technical bottleneck of data transmission is broken.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a prediction method and device based on federal learning, a storage medium and remote sensing equipment.
Background
With the development of computer technology, more and more technologies (big data, distributed, Blockchain, artificial intelligence, etc.) are applied to the financial field, and the traditional financial industry is gradually changing to financial technology (Fintech), but higher requirements are also put forward on the technologies due to the requirements of security and real-time performance of the financial industry.
Currently, more and more remote sensing devices such as artificial satellites operate in high altitude, and perform respective tasks to obtain corresponding data. As is well known, the remote sensing technology is difficult and expensive, the data obtained by the remote sensing equipment is precious, and if the data of each remote sensing equipment is to be comprehensively utilized, the data needs to be summarized to a ground information center, and the ground information center performs data analysis and processing. However, network delay exists between the remote sensing equipment and the ground information center, the data volume of high-altitude data obtained by the remote sensing equipment is huge, the requirement on network broadband is high, and further data transmission becomes a technical bottleneck.
Disclosure of Invention
The invention provides a prediction method, a prediction device, a storage medium and remote sensing equipment based on federal learning, and aims to realize comprehensive utilization of high-altitude data and break through the technical bottleneck of data transmission.
In order to achieve the above object, the present invention provides a prediction method based on federal learning, which is applied to a remote sensing system, and the method comprises:
training the initial model according to local training data to obtain a unilateral prediction model;
transmitting the unilateral prediction model to a ground information center, and aggregating a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices by the ground information center to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels;
and receiving a corresponding federal submodel, and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
Preferably, the step of training the initial model according to the local training data to obtain the unilateral prediction model includes:
randomly obtaining initial model parameters, obtaining initial prediction labels of the local training data by using the initial model parameters, and calculating a loss function based on the initial prediction labels and actual labels of the local training data;
updating model parameters in a gradient descending mode based on the loss function;
and if the convergence condition is reached, stopping updating, and storing the corresponding model parameters as final model parameters so as to obtain the unilateral prediction model.
Preferably, the remote sensing devices have different inclination angles, the local data to be predicted is predicted based on the federal submodel, and the step of obtaining the prediction label further includes:
judging whether the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality;
if the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality, executing the following steps: and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
Preferably, the step of training the initial model according to the local training data to obtain the unilateral prediction model further includes:
receiving an instruction sent by the ground information center through a control system, executing a task according to the instruction, and obtaining local data, wherein the local data comprises image data;
and preprocessing the local data to obtain local training data.
Preferably, the step of judging whether the local data to be predicted of each remote sensing device includes a plurality of target data with the same dimensionality includes:
determining a plurality of dimensions to be judged based on the prediction task;
obtaining local dimensions in local data to be predicted of each remote sensing device, and judging whether the local dimensions comprise a plurality of same dimensions or not based on the dimensions to be judged;
if the local dimensions comprise the dimensions to be judged, judging that the local dimensions of the remote sensing equipment comprise a plurality of same dimensions;
if the local dimensions of each remote sensing device comprise a plurality of same dimensions, searching target data corresponding to the plurality of dimensions to be judged from the local data to be predicted;
and if the local data of each remote sensing device has target data corresponding to the plurality of dimensions to be judged, judging that the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensions.
Preferably, the step of preprocessing the local data to obtain local training data includes:
and preprocessing the local data according to a flow, and storing the preprocessed local data as the local training data.
Preferably, the step of receiving the corresponding federal submodel, predicting local data to be predicted based on the federal submodel, and obtaining the prediction tag includes:
receiving the federal submodel correspondingly returned by the ground information center;
and inputting the local data to be predicted into the federal submodel, and outputting a corresponding prediction label by the federal submodel.
In addition, to achieve the above object, the present invention further provides a prediction device based on federal learning, including:
the training module is used for training the initial model according to local training data to obtain a unilateral prediction model;
the transmission module is used for transmitting the unilateral prediction model to a ground information center, and the ground information center aggregates a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels;
and the prediction module is used for receiving the corresponding federal submodel and predicting the local data to be predicted based on the federal submodel to obtain a prediction label.
In addition, in order to achieve the above object, the present invention further provides a remote sensing device, which includes a processor, a memory and a prediction program based on federated learning stored in the memory, wherein when the prediction program based on federated learning is executed by the processor, the steps of the prediction method based on federated learning as described above are implemented.
In addition, to achieve the above object, the present invention further provides a computer storage medium, on which a prediction program based on federal learning is stored, and the prediction program based on federal learning implements the steps of the prediction method based on federal learning as described above when being executed by a processor.
Compared with the prior art, the invention discloses a prediction method, a prediction device, a storage medium and remote sensing equipment based on federal learning, wherein an initial model is trained according to local training data to obtain a single-side prediction model; transmitting the unilateral prediction model to a ground information center, and aggregating a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices by the ground information center to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels; and receiving a corresponding federal submodel, and predicting local data to be predicted based on the federal submodel to obtain a prediction label. Therefore, model training is performed locally on the remote sensing equipment, comprehensive utilization of high-altitude data is achieved based on federal learning, and the technical bottleneck of data transmission is broken.
Drawings
FIG. 1 is a schematic diagram of a hardware configuration of a remote sensing device according to embodiments of the present invention;
FIG. 2 is a schematic flow chart diagram of a first embodiment of the prediction method based on federated learning of the present invention;
FIG. 3 is a schematic diagram of a scenario of an embodiment of the prediction method based on federated learning according to the present invention;
FIG. 4 is a flow chart illustrating a second embodiment of the federated learning-based prediction method of the present invention;
fig. 5 is a functional module diagram of a first embodiment of the prediction device based on federal learning according to the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The remote sensing equipment mainly related to the embodiment of the invention is network connection equipment capable of realizing network connection, and the remote sensing equipment can be a server, a cloud platform and the like.
Referring to FIG. 1, FIG. 1 is a schematic diagram of a hardware configuration of a remote sensing device according to embodiments of the present invention. In the embodiment of the present invention, the remote sensing device may include a processor 1001 (e.g., a Central Processing Unit, CPU), a communication bus 1002, an input port 1003, an output port 1004, and a memory 1005. The communication bus 1002 is used for realizing connection communication among the components; the input port 1003 is used for data input; the output port 1004 is used for data output, the memory 1005 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as a magnetic disk memory, and the memory 1005 may optionally be a storage device independent of the processor 1001. Those skilled in the art will appreciate that the hardware configuration depicted in FIG. 1 is not intended to be limiting of the present invention, and may include more or less components than those shown, or some components in combination, or a different arrangement of components.
With continued reference to FIG. 1, the memory 1005 of FIG. 1, which is one type of readable storage medium, may include an operating system, a network communication module, an application program module, and a prediction program based on federated learning. In fig. 1, the network communication module is mainly used for connecting to a server and performing data communication with the server; and the processor 1001 may invoke the prediction program based on federal learning stored in the memory 1005 and perform the following operations:
training the initial model according to local training data to obtain a unilateral prediction model;
transmitting the unilateral prediction model to a ground information center, and aggregating a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices by the ground information center to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels;
and receiving a corresponding federal submodel, and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
Further, the processor 1001 may be further configured to call a prediction program based on federal learning stored in the memory 1005, and execute the following steps:
randomly obtaining initial model parameters, obtaining initial prediction labels of the local training data by using the initial model parameters, and calculating a loss function based on the initial prediction labels and actual labels of the local training data;
updating model parameters in a gradient descending mode based on the loss function; and if the convergence condition is reached, stopping updating, and storing the corresponding model parameters as final model parameters so as to obtain the unilateral prediction model.
Further, the processor 1001 may be further configured to call a prediction program based on federal learning stored in the memory 1005, and execute the following steps:
judging whether the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality;
if the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality, executing the following steps: and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
Further, the processor 1001 may be further configured to call a prediction program based on federal learning stored in the memory 1005, and execute the following steps:
receiving an instruction sent by the ground information center through a control system, executing a task according to the instruction, and obtaining local data, wherein the local data comprises image data;
and preprocessing the local data to obtain local training data.
Further, the processor 1001 may be further configured to call a prediction program based on federal learning stored in the memory 1005, and execute the following steps:
determining a plurality of dimensions to be judged based on the prediction task;
obtaining local dimensions in local data to be predicted of each remote sensing device, and judging whether the local dimensions comprise a plurality of same dimensions or not based on the dimensions to be judged;
if the local dimensions comprise the dimensions to be judged, judging that the local dimensions of the remote sensing equipment comprise a plurality of same dimensions;
if the local dimensions of each remote sensing device comprise a plurality of same dimensions, searching target data corresponding to the plurality of dimensions to be judged from the local data to be predicted;
and if the local data of each remote sensing device has target data corresponding to the plurality of dimensions to be judged, judging that the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensions.
Further, the processor 1001 may be further configured to call a prediction program based on federal learning stored in the memory 1005, and execute the following steps:
and preprocessing the local data according to a flow, and storing the preprocessed local data as the local training data.
Further, the processor 1001 may be further configured to call a prediction program based on federal learning stored in the memory 1005, and execute the following steps:
receiving the federal submodel correspondingly returned by the ground information center;
and inputting the local data to be predicted into the federal submodel, and outputting a corresponding prediction label by the federal submodel.
Based on the structure, the invention provides various embodiments of the prediction method based on the federal learning.
Referring to fig. 2, fig. 2 is a schematic flow chart of a first embodiment of the prediction method based on federal learning according to the present invention.
In this embodiment, the prediction method based on federated learning is applied to a remote sensing device, and the method includes:
step S101: training the initial model according to local training data to obtain a unilateral prediction model;
in this embodiment, the remote sensing device generally refers to a high-altitude device that works by using a remote sensing technology, such as a high tower, a balloon, an airplane, a rocket, an artificial earth satellite, a spacecraft, and a space shuttle. The remote sensing technology is a comprehensive technology for detecting and monitoring the resources and the environment of the earth by detecting the electromagnetic wave radiation information on the earth surface through a sensor from a remote sensing device far away from the ground and then transmitting, processing, interpreting and analyzing the information. The aerial view is detected from a long distance in a high-altitude aerial view mode, the remote sensing images comprise multipoint remote sensing images, multispectral remote sensing images, multiple time periods and multiple heights, and multiple times of enhanced remote sensing information, continuous regional synchronous information of comprehensive systematicness, instantaneity or synchronism can be provided, and the aerial view is applied to the field of environmental science with great superiority. Remote sensing devices have many uses, such as terrain detection, oil detection, atmospheric environment, and terrain detection, among others.
Usually, each remote sensing device has its inclination angle (the included angle between the orbit plane of the remote sensing device and the earth equator plane is called the orbit inclination angle), therefore, the images of these remote sensing devices are photographed according to a certain angle to obtain images, and in this embodiment, the image data is marked as local data.
In this embodiment, in step S101, the step of training the initial model according to the local training data to obtain the unilateral prediction model further includes:
step S1001: receiving an instruction sent by the ground information center through a control system, executing a task according to the instruction, and obtaining local data, wherein the local data comprises image data;
generally, the remote sensing device needs to establish a communication connection with a ground information center. Therefore, the remote sensing equipment can carry out data transmission with the ground information center.
The remote sensing equipment is pre-loaded with a control system, the control system comprises a plurality of preset instructions and can also receive the control instructions sent by a ground information center and execute corresponding tasks according to the preset instructions and/or the control instructions. The task may be mapping, terrain surveying, etc.
And after the remote sensing equipment receives the instruction sent by the ground information center through the control system, executing a task according to the instruction to obtain corresponding local data. Typically, the local data is image data.
Step S1002: and preprocessing the local data to obtain local training data.
And after the local data are obtained, preprocessing the local data. Specifically, the step of preprocessing the local data to obtain local training data includes:
and preprocessing the local data according to a flow, and storing the preprocessed local data as the local training data.
And preprocessing the local initial data according to a flow to obtain local training data capable of performing model training. The preprocessing comprises noise reduction, geometric correction, image enhancement, image cutting and the like, and the preprocessed local data is stored as the local training data. The local initial data may be image data. Generally, the local training data may be denoised by a filter. In the remote sensing imaging process, under the comprehensive influence of a plurality of factors, the geometric position, shape, size, dimension, orientation and other characteristics of the ground object on the original image are often inconsistent with the characteristics of the ground object corresponding to the geometric position, shape, size, dimension, orientation and the like, and the inconsistency is geometric deformation, also called geometric distortion, so that correction processing is needed. The process refers to the order of pretreatment, and is generally set empirically, for example, as follows: noise reduction, geometric correction and image cutting.
In this embodiment, the local data is preprocessed to obtain the local training data, machine learning training is performed on model parameters obtained at random based on the local training data, iteration updating is continuously performed on the model parameters until convergence, and final model parameters are stored, so that a one-way prediction model is obtained.
In this embodiment, a plurality of remote sensing devices are selected, the plurality of remote sensing devices have different tilt angles, and a panoramic remote sensing system is formed based on the tilt angles. For example, a 360 ° panoramic remote sensing system, a 180 ° remote sensing system or a 270 ° remote sensing system. For a 360-degree panoramic remote sensing system, a plurality of remote sensing devices with different inclination angles are selected, and the selected remote sensing devices can obtain 360-degree panoramic data.
Specifically, referring to fig. 3, fig. 3 is a scene diagram of an embodiment of the prediction method based on federal learning according to the present invention. In FIG. 3a, 6 telemetry devices with different inclinations are selected, the 6 telemetry devices having respective inclinations θ1、θ2、θ3、θ4、θ5、θ6The 6 remote sensing devices form a 360-degree panoramic remote sensing system. In FIG. 3b, 4 remote sensing devices C are selected1、C2、C3And C4Wherein, C1、C2、C3And C4The inclination angles are different, data of different angles and different directions can be obtained, and therefore a remote sensing system can be formed and is in communication connection with the ground information center S.
In this embodiment, the plurality of remote sensing devices respectively perform model training to obtain a plurality of unilateral prediction models, and the unilateral prediction models are expressed as:
where M represents the model and i represents the remote sensing device compilationThe number t indicates the current number of iterations. If the iteration times of the current model are t, the initial model is subjected to iteration according to the local training data
Training is carried out until the initial model
Converge to obtain
Step S102: transmitting the unilateral prediction model to a ground information center, and aggregating a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices by the ground information center to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels;
and after the unilateral prediction model is obtained, transmitting the unilateral prediction model to the ground information center through the connection which is established with the ground information center in advance. The ground information center receives a plurality of unilateral prediction models uploaded by a plurality of remote sensing devices, and aggregates the unilateral prediction models to obtain a federal prediction model.
The ground information center aggregates a plurality of unilateral prediction models to obtain a federal prediction model Mt+1：
Further, the federal submodel is expressed as
And sending the federal submodel to the remote sensing devices.
Step S103: and receiving a corresponding federal submodel, and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
And the remote sensing equipment receives the corresponding federal submodel returned by the ground information center, and predicts local data to be predicted based on the federal submodel to obtain a prediction label.
Specifically, the step of receiving the corresponding federal submodel, predicting local data to be predicted based on the federal submodel, and obtaining the prediction tag includes:
step S103 a: receiving the federal submodel correspondingly returned by the ground information center;
and receiving the federal submodel correspondingly returned by the ground information center. And storing the federal submodel to a preset storage path, and updating the federal submodel of the remote sensing equipment. In particular, the federal submodel may be saved directly. If an original federal submodel exists on the remote sensing equipment, the original federal submodel is updated, similarities and differences between the stored federal submodel and the original federal submodel can also be compared, different difference parameters are obtained, the original federal submodel is updated based on the difference, generally, after the federal submodel correspondingly returned by a ground information center is received, the federal submodel is marked as a latest model, and the latest model is used for prediction. In other embodiments, the original federal submodel may be deleted or disabled.
Step S103 b: and inputting the local data to be predicted into the federal submodel, and outputting a corresponding prediction label by the federal submodel.
In this embodiment, the local data to be predicted is input into the federal submodel, and the local data to be predicted is processed by the federal submodel according to model parameters to obtain a prediction tag. The prediction tags may be 0, 1, and others. The prediction tag may also include a prediction outcome and a probability. It should be noted that, the local data obtained by the remote sensing device according to the task is preprocessed in advance to obtain the local data to be predicted, and the preprocessing process is consistent with the process of processing the local initial data to obtain the local training data. And the federal submodel predicts the local data to be predicted and outputs a prediction label, wherein the prediction label is related to a specific task. For example for a terrain survey, the prediction tag may be the terrain type of the local data to be predicted, and the probability of belonging to that type. For another example, for a petroleum survey, the prediction tag may be yes or no, which may be represented by the corresponding numbers 1, 0.
The step S103, predicting local data to be predicted based on the federal submodel, and obtaining a prediction label, wherein the step before:
step S103-1: judging whether the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality;
in this embodiment, the local data acquired by the plurality of remote sensing devices has a plurality of same dimensions. Therefore, in order to obtain the data to be predicted which meets the requirements, whether the local data to be predicted of each remote sensing device comprises the target data with the same dimensionality or not can be judged in advance.
Specifically, the step S103-1: the step of judging whether the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality comprises the following steps:
step S103-1 a: determining a plurality of dimensions to be judged based on the prediction task;
as can be understood, the local data to be predicted includes data of multiple dimensions. Based on different tasks, the local data obtained by the remote sensing devices of different tasks include one or more pieces of data with different dimensions, and the local data obtained by the remote sensing devices of different tasks also can include one or more pieces of data with the same dimensions. The number of the same dimensions and the corresponding dimensions to be judged may be specifically set according to a task, for example, if the prediction method based on federal learning needs to predict ocean terrain, height measurement data, sea level data, and the like are needed, and therefore, the height measurement data and the sea level data may be determined as the dimensions to be judged.
Step S103-1 b: obtaining local dimensions in local data to be predicted of each remote sensing device, and judging whether the local dimensions comprise a plurality of same dimensions or not based on the dimensions to be judged;
and analyzing the local data to be predicted to obtain the local dimension in the local data to be predicted of each remote sensing device. And comparing the local dimension with the dimension to be judged. Specifically, whether the number of the local dimensions is greater than or equal to the number of the dimensions to be judged is judged; if the number of the local dimensions is larger than or equal to the number of the dimensions to be judged, further comparing the content of the local dimensions with the content of the specific dimensions of the dimensions to be judged, and if the local dimensions comprise the content of the specific dimensions of the dimensions to be judged, judging that the local dimensions comprise the plurality of dimensions to be judged.
Step S103-1 c: if the local dimensions comprise the dimensions to be judged, judging that the local dimensions of the remote sensing equipment comprise a plurality of same dimensions;
on the contrary, if the local dimension does not completely include the plurality of dimensions to be judged, the local dimension of each remote sensing device is judged not to include the plurality of same dimensions.
Step S103-1 d: if the local dimensions of each remote sensing device comprise a plurality of same dimensions, searching target data corresponding to the plurality of dimensions to be judged from the local data to be predicted;
on the contrary, if the local dimensions of each remote sensing device do not comprise a plurality of same dimensions, the corresponding local data to be predicted are judged to be invalid data, and the invalid data are ignored. It can be understood that if the local data of each remote sensing device does not include target data of several same dimensions, the prediction result may be biased due to incomplete information of the local data.
Step S103-1 e: and if the local data of each remote sensing device has target data corresponding to the plurality of dimensions to be judged, judging that the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensions.
If the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality, executing the following steps: and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
According to the scheme, the initial model is trained according to local training data to obtain a single-side prediction model; transmitting the unilateral prediction model to a ground information center, and aggregating a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices by the ground information center to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels; and receiving a corresponding federal submodel, and predicting local data to be predicted based on the federal submodel to obtain a prediction label. Therefore, model training is performed locally on the remote sensing equipment, comprehensive utilization of high-altitude data is achieved based on federal learning, and the technical bottleneck of data transmission is broken.
A third embodiment of the present invention is proposed based on the second embodiment shown in fig. 2 described above. As shown in fig. 4, fig. 4 is a flowchart illustrating a second embodiment of the prediction method based on federal learning according to the present invention.
The step of training the initial model according to the local training data to obtain the unilateral prediction model comprises the following steps:
step S101 a: randomly obtaining initial model parameters, obtaining initial prediction labels of the local training data by using the initial model parameters, and calculating a loss function based on the initial prediction labels and actual labels of the local training data;
the initial model may be a convolutional neural network model, a deep learning model, a decision tree, etc. And randomly obtaining model parameters of each layer of the initial model, and storing the model parameters as the initial model parameters. The initial model parameters may also be determined empirically.
In this embodiment, a large amount of local data is acquired, and an actual tag of the local data is marked, and the actual tag is determined according to a result of the local data. For example, for a petroleum surveying task, if the local data results in the presence of petroleum in the region corresponding to the picture, the actual tag may be 1; if the result of the local data is that no petroleum exists in the region corresponding to the picture, the actual tag may be 0; if the result of the local data is that whether the region corresponding to the picture has oil or not cannot be determined, the actual tag may be other.
And preprocessing the local data to obtain corresponding local training data. The local data is image data, and the preprocessing comprises operations of noise reduction, geometric correction, image enhancement, image cropping and the like.
And processing the local training data based on the initial model parameters to obtain the initial prediction label. A loss function is calculated from the initial predicted tag and the actual tag. The loss function may be a common loss function for a task, and in this embodiment, the cross entropy loss function is calculated based on a mini-batch.
Step S101 b: updating model parameters in a gradient descending mode based on the loss function;
calculating gradients corresponding to the parameters in the initial model according to the cross entropy loss function, and correspondingly updating the parameters according to the gradients of the parameters, namely adjusting the parameters of the initial model. Here, the process of updating the model parameters according to the cross entropy loss function is similar to the existing model parameter updating process, and is not described in detail here.
Step S101 c: and if the convergence condition is reached, stopping updating, and storing the corresponding model parameters as final model parameters so as to obtain the unilateral prediction model.
And judging whether the cross entropy loss function is converged, and if the cross entropy loss function is converged, judging that the corresponding quality inspection model is converged. The convergence condition is that the cross entropy loss function obtains a minimum value.
And if the initial model is in a convergence state, stopping training, saving the last training parameter as a final model parameter, and obtaining the unilateral prediction model based on the final model parameter.
Otherwise, if the unilateral prediction model does not reach the convergence state, continuing training: the iterative update is continued until convergence. And finally obtaining the unilateral prediction model.
Based on the scheme, the embodiment randomly obtains initial model parameters, randomly obtains the initial model parameters, obtains the initial prediction labels of the local training data by using the initial model parameters, and calculates the loss function based on the initial prediction labels and the actual labels of the local training data; updating model parameters in a gradient descending mode based on the loss function; and if the convergence condition is reached, stopping updating, and storing the corresponding model parameters as final model parameters so as to obtain the unilateral prediction model. Therefore, model training is carried out through local training data, comprehensive utilization of high-altitude data is achieved, and the technical bottleneck of data transmission is broken.
In addition, the embodiment also provides a prediction device based on federal learning. Referring to fig. 5, fig. 5 is a functional module diagram of a first embodiment of the prediction apparatus based on federal learning according to the present invention.
In this embodiment, the prediction device based on federal learning is a virtual device, and is stored in the memory 1005 of the remote sensing device shown in fig. 1, so as to realize all functions of the prediction program based on federal learning: the system comprises a local training data acquisition module, a single-side prediction module and a single-side prediction module, wherein the local training data acquisition module is used for acquiring local training data; the system comprises a ground information center, a single-party prediction model, a federal prediction model and a remote sensing device, wherein the single-party prediction model is used for being transmitted to the ground information center, the ground information center aggregates a plurality of single-party prediction models corresponding to a plurality of remote sensing devices to obtain the federal prediction model, and the federal prediction model comprises a plurality of federal submodels; and the prediction module is used for receiving the corresponding federal submodel and predicting the local data to be predicted based on the federal submodel to obtain a prediction label.
Specifically, the prediction device based on federal learning includes:
the training module 10 is used for training the initial model according to local training data to obtain a single-party prediction model;
the transmission module 20 is configured to transmit the unilateral prediction model to a ground information center, and the ground information center aggregates the unilateral prediction models corresponding to the remote sensing devices to obtain a federal prediction model, where the federal prediction model includes several federal submodels;
and the prediction module 30 is configured to receive the corresponding federal submodel, and predict local data to be predicted based on the federal submodel to obtain a prediction tag.
Further, the training module comprises:
a calculating unit, configured to randomly obtain initial model parameters, obtain initial prediction labels of the local training data using the initial model parameters, and calculate a loss function based on the initial prediction labels and actual labels of the local training data;
the updating unit is used for updating the model parameters in a gradient descending mode based on the loss function;
and the storage unit is used for stopping updating if the convergence condition is reached and storing the corresponding model parameters as final model parameters so as to obtain the unilateral prediction model.
Further, the prediction module further comprises:
the judging unit is used for judging whether the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality;
an execution unit, configured to execute the following steps if the local data to be predicted of each remote sensing device includes a plurality of target data with the same dimensionality: and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
Further, the training module further comprises
The receiving unit is used for receiving an instruction sent by the ground information center through a control system, executing a task according to the instruction and acquiring the local data, wherein the local data comprises image data;
and the preprocessing unit is used for preprocessing the local data to obtain local training data.
Further, the judging unit includes:
the determining subunit is used for determining a plurality of dimensions to be judged based on the prediction task;
the acquisition subunit is used for acquiring local dimensions in the local data to be predicted of each remote sensing device and judging whether the local dimensions comprise a plurality of same dimensions or not based on the dimensions to be judged;
the first judging subunit is configured to judge that the local dimension of each remote sensing device includes a plurality of same dimensions if the local dimensions all include the plurality of dimensions to be judged;
the searching unit is used for searching target data corresponding to the dimensions to be judged from the local data to be predicted if the local dimensions of the remote sensing equipment comprise a plurality of same dimensions;
and the second judging subunit is used for judging that the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality if the local data of each remote sensing device comprises the target data corresponding to the dimensionalities to be judged.
Further, the preprocessing unit includes:
a pretreatment subunit: and preprocessing the local data according to a flow, and storing the preprocessed local data as the local training data.
Further, the prediction module comprises:
the receiving unit is used for receiving the federal submodel correspondingly returned by the ground information center;
and the input unit is used for inputting the local data to be predicted into the federal submodel and outputting a corresponding prediction label by the federal submodel.
In addition, an embodiment of the present invention further provides a computer storage medium, where a prediction program based on federal learning is stored in the computer storage medium, and when the prediction program based on federal learning is executed by a processor, the steps of the prediction method based on federal learning are implemented, which are not described herein again.
Compared with the prior art, the prediction method, the prediction device, the storage medium and the remote sensing equipment based on the federal learning provided by the invention comprise the following steps: receiving the federal submodel sent by a ground information center; and inputting local data into the federal submodel, and outputting a corresponding prediction label by the federal submodel. Therefore, based on federal learning, the local data of the remote sensing equipment system is used for prediction, the technical bottleneck of data transmission is broken, and comprehensive utilization of high-altitude data is realized.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) as described above and includes instructions for causing a terminal device to execute the method according to the embodiments of the present invention.
The above description is only for the preferred embodiment of the present invention and is not intended to limit the scope of the present invention, and all equivalent structures or flow transformations made by the present specification and drawings, or applied directly or indirectly to other related arts, are included in the scope of the present invention.
Claims (10)
1. A prediction method based on federal learning, the method comprising:
training the initial model according to local training data to obtain a unilateral prediction model;
transmitting the unilateral prediction model to a ground information center, and aggregating a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices by the ground information center to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels;
and receiving a corresponding federal submodel, and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
2. The method of claim 1, wherein the training the initial model based on local training data to obtain the unilateral predictive model comprises:
randomly obtaining initial model parameters, obtaining initial prediction labels of the local training data by using the initial model parameters, and calculating a loss function based on the initial prediction labels and actual labels of the local training data;
updating model parameters in a gradient descending mode based on the loss function;
and if the convergence condition is reached, stopping updating, and storing the corresponding model parameters as final model parameters so as to obtain the unilateral prediction model.
3. The method of claim 1, wherein the plurality of remote sensing devices have different tilt angles, the predicting local data to be predicted based on the federal submodel, and the obtaining a prediction label further comprises:
judging whether the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality;
if the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensionality, executing the following steps: and predicting local data to be predicted based on the federal submodel to obtain a prediction label.
4. The method of claim 1, wherein the step of training the initial model based on local training data to obtain the unilateral predictive model is preceded by the step of:
receiving an instruction sent by the ground information center through a control system, executing a task according to the instruction, and obtaining local data, wherein the local data comprises image data;
and preprocessing the local data to obtain local training data.
5. The method of claim 3, wherein the step of determining whether the local data to be predicted of each remote sensing device comprises a plurality of target data of the same dimension comprises:
determining a plurality of dimensions to be judged based on the prediction task;
obtaining local dimensions in local data to be predicted of each remote sensing device, and judging whether the local dimensions comprise a plurality of same dimensions or not based on the dimensions to be judged;
if the local dimensions comprise the dimensions to be judged, judging that the local dimensions of the remote sensing equipment comprise a plurality of same dimensions;
if the local dimensions of each remote sensing device comprise a plurality of same dimensions, searching target data corresponding to the plurality of dimensions to be judged from the local data to be predicted;
and if the local data of each remote sensing device has target data corresponding to the plurality of dimensions to be judged, judging that the local data to be predicted of each remote sensing device comprises a plurality of target data with the same dimensions.
6. The method of claim 4, wherein the step of preprocessing the local data to obtain local training data comprises:
and preprocessing the local data according to a flow, and storing the preprocessed local data as the local training data.
7. The method according to claim 1, wherein the step of receiving the corresponding federal submodel and predicting the local data to be predicted based on the federal submodel to obtain the prediction tag comprises the steps of:
receiving the federal submodel correspondingly returned by the ground information center;
and inputting the local data to be predicted into the federal submodel, and outputting a corresponding prediction label by the federal submodel.
8. A federal learning-based prediction apparatus, comprising:
the training module is used for training the initial model according to local training data to obtain a unilateral prediction model;
the transmission module is used for transmitting the unilateral prediction model to a ground information center, and the ground information center aggregates a plurality of unilateral prediction models corresponding to a plurality of remote sensing devices to obtain a federal prediction model, wherein the federal prediction model comprises a plurality of federal submodels;
and the prediction module is used for receiving the corresponding federal submodel and predicting the local data to be predicted based on the federal submodel to obtain a prediction label.
9. A remote sensing device comprising a processor, a memory, and a federal learning based prognostics program stored in the memory, the federal learning based prognostics program, when executed by the processor, implementing the steps of the federal learning based prognostics method as claimed in any of claims 1-7.
10. A computer storage medium having stored thereon a federal learning based prediction program for implementing the steps of a federal learning based prediction method as claimed in any one of claims 1-7 when executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010164191.0A CN111382706A (en) | 2020-03-10 | 2020-03-10 | Prediction method and device based on federal learning, storage medium and remote sensing equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010164191.0A CN111382706A (en) | 2020-03-10 | 2020-03-10 | Prediction method and device based on federal learning, storage medium and remote sensing equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111382706A true CN111382706A (en) | 2020-07-07 |
Family
ID=71221632
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010164191.0A Pending CN111382706A (en) | 2020-03-10 | 2020-03-10 | Prediction method and device based on federal learning, storage medium and remote sensing equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111382706A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112001502A (en) * | 2020-08-24 | 2020-11-27 | 平安科技（深圳）有限公司 | Federal learning training method and device for high-delay network environment robustness |
| CN112862347A (en) * | 2021-03-02 | 2021-05-28 | 同济大学 | Equipment abnormity monitoring method and system based on federal learning, storage medium and terminal |
| CN113313268A (en) * | 2021-06-11 | 2021-08-27 | 杭州煋辰数智科技有限公司 | Federal learning-based prediction method and device, storage medium and remote sensing equipment |
| WO2021189906A1 (en) * | 2020-10-20 | 2021-09-30 | 平安科技（深圳）有限公司 | Target detection method and apparatus based on federated learning, and device and storage medium |
| WO2022111068A1 (en) * | 2020-11-27 | 2022-06-02 | 中兴通讯股份有限公司 | Rru undervoltage risk prediction method, apparatus, and system, device, and medium |
| WO2022130098A1 (en) * | 2020-12-15 | 2022-06-23 | International Business Machines Corporation | Federated learning for multi-label classification model for oil pump management |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108985238A (en) * | 2018-07-23 | 2018-12-11 | 武汉大学 | The high-resolution remote sensing image impervious surface extracting method and system of combined depth study and semantic probability |
| CN110428058A (en) * | 2019-08-08 | 2019-11-08 | 深圳前海微众银行股份有限公司 | Federal learning model training method, device, terminal device and storage medium |
| CN110444263A (en) * | 2019-08-21 | 2019-11-12 | 深圳前海微众银行股份有限公司 | Disease data processing method, device, equipment and medium based on federation's study |
| US20190354644A1 (en) * | 2018-05-18 | 2019-11-21 | Honeywell International Inc. | Apparatuses and methods for detecting anomalous aircraft behavior using machine learning applications |
| CN110601814A (en) * | 2019-09-24 | 2019-12-20 | 深圳前海微众银行股份有限公司 | Federal learning data encryption method, device, equipment and readable storage medium |
| CN110633806A (en) * | 2019-10-21 | 2019-12-31 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning system optimization method, device, equipment and readable storage medium |
-
2020
- 2020-03-10 CN CN202010164191.0A patent/CN111382706A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190354644A1 (en) * | 2018-05-18 | 2019-11-21 | Honeywell International Inc. | Apparatuses and methods for detecting anomalous aircraft behavior using machine learning applications |
| CN108985238A (en) * | 2018-07-23 | 2018-12-11 | 武汉大学 | The high-resolution remote sensing image impervious surface extracting method and system of combined depth study and semantic probability |
| CN110428058A (en) * | 2019-08-08 | 2019-11-08 | 深圳前海微众银行股份有限公司 | Federal learning model training method, device, terminal device and storage medium |
| CN110444263A (en) * | 2019-08-21 | 2019-11-12 | 深圳前海微众银行股份有限公司 | Disease data processing method, device, equipment and medium based on federation's study |
| CN110601814A (en) * | 2019-09-24 | 2019-12-20 | 深圳前海微众银行股份有限公司 | Federal learning data encryption method, device, equipment and readable storage medium |
| CN110633806A (en) * | 2019-10-21 | 2019-12-31 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning system optimization method, device, equipment and readable storage medium |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112001502A (en) * | 2020-08-24 | 2020-11-27 | 平安科技（深圳）有限公司 | Federal learning training method and device for high-delay network environment robustness |
| WO2021189906A1 (en) * | 2020-10-20 | 2021-09-30 | 平安科技（深圳）有限公司 | Target detection method and apparatus based on federated learning, and device and storage medium |
| WO2022111068A1 (en) * | 2020-11-27 | 2022-06-02 | 中兴通讯股份有限公司 | Rru undervoltage risk prediction method, apparatus, and system, device, and medium |
| WO2022130098A1 (en) * | 2020-12-15 | 2022-06-23 | International Business Machines Corporation | Federated learning for multi-label classification model for oil pump management |
| CN112862347A (en) * | 2021-03-02 | 2021-05-28 | 同济大学 | Equipment abnormity monitoring method and system based on federal learning, storage medium and terminal |
| CN113313268A (en) * | 2021-06-11 | 2021-08-27 | 杭州煋辰数智科技有限公司 | Federal learning-based prediction method and device, storage medium and remote sensing equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111382706A (en) | Prediction method and device based on federal learning, storage medium and remote sensing equipment | |
| KR102358554B1 (en) | Training method and apparatus, electronic device, and storage medium for semantic segmentation model | |
| CN109683699B (en) | Method and device for realizing augmented reality based on deep learning and mobile terminal | |
| CN111401621B (en) | Prediction method, device, equipment and storage medium based on federal learning | |
| US9386266B2 (en) | Method and apparatus for increasing frame rate of an image stream using at least one higher frame rate image stream | |
| CN111310770A (en) | Target detection method and device | |
| US20220366646A1 (en) | Computer Vision Systems and Methods for Determining Structure Features from Point Cloud Data Using Neural Networks | |
| CN114648709A (en) | Method and equipment for determining image difference information | |
| CN112013820B (en) | Real-time target detection method and device for deployment of airborne platform of unmanned aerial vehicle | |
| US20220215645A1 (en) | Computer Vision Systems and Methods for Determining Roof Conditions from Imagery Using Segmentation Networks | |
| US10346717B1 (en) | System and method for thresholding of local image descriptors | |
| CN115861755A (en) | Feature fusion method and device, electronic equipment and automatic driving vehicle | |
| CN115731451A (en) | Model training method and device, electronic equipment and storage medium | |
| CN115063826A (en) | Mobile terminal driver license identification method and system based on deep learning | |
| CN110321794B (en) | Remote sensing image oil tank detection method integrated with semantic model | |
| Christian | Optical attitude determination from horizon orientation using image segmentation | |
| CN117036982B (en) | Method and device for processing optical satellite image of mariculture area, equipment and medium | |
| CN115240171B (en) | Road structure sensing method and device | |
| CN117152577B (en) | Casting defect detection method, casting defect detection device, electronic equipment and storage medium | |
| US20220383515A1 (en) | Digital video computing system for vehicle | |
| CN117788302B (en) | Mapping graphic processing system | |
| KR102406600B1 (en) | Method of photorealistic image transformation for marine images, apparatus for photorealistic image transformation for marine images, and computer program for the method | |
| Singh et al. | Performance Assessment of Deep Learning Algorithm Using xView Data Set | |
| CN114821068B (en) | Panorama segmentation and depth determination processing method, device, equipment and medium | |
| JP7425406B2 (en) | Image inspection device, learned model generation device, image inspection system, image inspection program, learned model generation program, and learned model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |