Remote sensing basic model rapid training method and system based on template editing

Abstract

The present invention provides a remote sensing basic model rapid training method and system based on template editing, which relates to the field of remote sensing image processing technology and is used to solve the technical problem of algorithm integration difficulties in the existing remote sensing basic model training process. The method includes: obtaining a pre-labeled remote sensing sample data set; determining a remote sensing basic model architecture corresponding to the task requirements according to the user's task requirements; editing a corresponding algorithm framework and mirror environment for the remote sensing basic model architecture based on the task requirements; inputting the remote sensing sample data set into the remote sensing basic model architecture, and training the remote sensing basic model architecture using the edited algorithm framework and mirror environment; visually monitoring the progress, parameter changes and performance indicators of the remote sensing basic model architecture during the training process.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image processing, and in particular to a remote sensing basic model rapid training method and system based on template editing. Background Art

[0002] At present, remote sensing image processing has been widely used in various fields. Among them, model-based algorithms have become one of the key technologies in remote sensing image processing. However, the current basic model training methods have the problem of difficulty in model algorithm integration, which limits their efficiency and accuracy in practical applications.

[0003] In the existing technology, the rapid training method based on template editing, on the one hand, needs to integrate multiple algorithms and technologies. However, the integration and coordination between multiple algorithms and technologies is not easy, especially when they come from different research fields, which may lead to the difficulty and complexity of algorithm integration, limiting the practical application and promotion of the method. On the other hand, it is necessary to go through multiple steps and operations to select, edit, train, etc. templates. However, the current technology often lacks visual support for the entire training process, making it difficult for users to understand and master the specific steps and operation procedures of the method, and it is also difficult to monitor and adjust each link in the training process. Therefore, the difficulty of integrating algorithms and the lack of process visualization are the problems that the current rapid training method and system of remote sensing basic models based on template editing need to solve. Summary of the invention

[0004] In view of the above problems, the present invention provides a method and system for rapid training of remote sensing basic models based on template editing.

[0005] According to a first aspect of the present invention, a method for rapid training of a remote sensing basic model based on template editing is provided, comprising: obtaining a pre-labeled remote sensing sample data set; determining a remote sensing basic model architecture corresponding to the task requirements according to the user's task requirements; editing a corresponding algorithm framework and mirror environment for the remote sensing basic model architecture based on the task requirements; inputting the remote sensing sample data set into the remote sensing basic model architecture, and training the remote sensing basic model architecture using the edited algorithm framework and mirror environment; and visually monitoring the progress, parameter changes, and performance indicators of the remote sensing basic model architecture during the training process.

[0006] According to an embodiment of the present invention, before a remote sensing sample data set is input into a remote sensing basic model architecture and the remote sensing basic model architecture is trained using an edited algorithm framework and a mirror environment, it also includes: initializing parameters of the remote sensing basic model architecture; and creating a training task based on the remote sensing sample data set, the algorithm framework, and the mirror environment.

[0007] According to an embodiment of the present invention, a remote sensing sample data set is input into a remote sensing basic model framework, and the remote sensing basic model framework is trained using an edited algorithm framework and a mirror environment, including: according to a training task, the remote sensing sample data set is input into the remote sensing basic model framework for training; and the parameters of the algorithm framework are iteratively updated using an intelligent parameter adjustment and optimization method.

[0008] According to an embodiment of the present invention, the use of an intelligent parameter optimization method to iteratively update the parameters of an algorithm framework includes: determining the hyperparameters of the algorithm framework; iteratively optimizing the hyperparameters according to the intelligent parameter optimization algorithm; obtaining evaluation indicators of the remote sensing basic model architecture on a validation set; and according to the evaluation indicators, taking the part of the hyperparameter that reaches the indicator threshold during the iterative optimization process as the starting point for the next iteration, until the iteration termination condition is reached.

[0009] According to an embodiment of the present invention, the progress, parameter changes and performance indicators of the remote sensing basic model architecture during the training process are visually monitored: a visualization dynamic library matching the remote sensing basic model architecture is determined; the parameters and loss function of the remote sensing basic model architecture during the training process are passed to the visualization dynamic library by dynamic mounting; and the progress, parameter changes and performance indicators of the training process are monitored in real time according to the visualization dynamic library.

[0010] According to an embodiment of the present invention, editing the corresponding algorithm framework and mirror environment for the remote sensing basic model architecture based on task requirements includes: creating a template, wherein the template is pre-integrated with at least one of the tensor flow framework, Pytorch, Pulse framework, MXNet, MindSpore, Caffe, Caffe2 and just-in-time compilation framework; selecting the algorithm framework corresponding to the task requirements in the template and configuring it into the remote sensing basic model architecture.

[0011] According to an embodiment of the present invention, the template also integrates at least one of the X86_64 image architecture, the MIPS_64 image architecture and the ARM_64 image architecture.

[0012] According to an embodiment of the present invention, creating a training task according to a remote sensing sample data set, an algorithm framework, and a mirror environment includes: creating a single-machine training task or a distributed training task according to the remote sensing sample data set, an algorithm framework, and a mirror environment.

[0013] According to an embodiment of the present invention, the remote sensing sample data set includes: one of a remote sensing image data set, a remote sensing text data set, a remote sensing voice data set, a remote sensing video data set, a remote sensing electronic data set, and a remote sensing situation data set.

[0014] The second aspect of the present invention provides a remote sensing basic model rapid training system based on template editing, including: a sample data management module for obtaining a pre-labeled remote sensing sample data set; a template management module for determining a remote sensing basic model architecture corresponding to the task requirements according to the user's task requirements; and editing the corresponding algorithm framework and mirror environment for the remote sensing basic model architecture based on the task requirements; a training management module for inputting the remote sensing sample data set into the remote sensing basic model architecture, and training the remote sensing basic model architecture using the edited algorithm framework and mirror environment; and visually monitoring the progress, parameter changes and performance indicators of the remote sensing basic model architecture during the training process.

[0015] According to the remote sensing basic model rapid training method and system based on template editing provided by the present invention, by editing the algorithm framework and mirror environment on the remote sensing basic model architecture template according to task requirements, the integration between algorithms and technologies is realized, and the problem of difficulty in integration and coordination between algorithms and technologies corresponding to different fields is solved. At the same time, the training process of the remote sensing basic model architecture is visually monitored, so that users can understand and grasp the progress, parameter changes and performance indicators in the training process in real time, which is conducive to the rapid training and optimization of the remote sensing basic model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above contents and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0017] Figure 1 A flowchart of a remote sensing basic model rapid training method based on template editing according to an embodiment of the present invention is schematically shown;

[0018] Figure 2 The structural block diagram of the remote sensing basic model rapid training system based on template editing according to an embodiment of the present invention is schematically shown. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "including", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0020] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, or can communicate with each other; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0021] In the description of the present invention, it should be understood that the terms "longitudinal", "length", "circumferential", "front", "rear", "left", "right", "top", "bottom", "inside", "outside", etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the subsystem or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0022] Throughout the drawings, the same elements are represented by the same or similar reference numerals. Conventional structures or configurations will be omitted when they may cause confusion in the understanding of the present invention. The shapes, sizes, and positional relationships of the components in the drawings do not reflect the actual size, proportion, and actual positional relationship.

[0023] Similarly, in order to simplify the present invention and help understand one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure or description thereof. The description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0024] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0025] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0026] In the technical solution of the present invention, the collection, storage, use, processing, transmission, provision, disclosure and application of the data involved (including but not limited to user personal information) comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0027] Figure 1 The flowchart of the rapid training method of the remote sensing basic model based on template editing according to an embodiment of the present invention is schematically shown.

[0028] like Figure 1 As shown, the remote sensing basic model rapid training method based on template editing in this embodiment includes operations S1 to S5.

[0029] In operation S1, a pre-labeled remote sensing sample data set is obtained.

[0030] In this embodiment, the remote sensing sample data set may be, for example, one of a remote sensing image data set, a remote sensing text data set, a remote sensing voice data set, a remote sensing video data set, a remote sensing electronic data set, and a remote sensing situation data set.

[0031] In operation S2, a remote sensing basic model architecture corresponding to the task requirement is determined according to the task requirement of the user.

[0032] In this embodiment, a suitable basic model architecture is selected according to task requirements, for example, a basic model architecture such as a convolutional neural network (CNN) or a deep residual network (ResNet).

[0033] In operation S3, the corresponding algorithm framework and mirror environment are edited for the remote sensing basic model architecture based on task requirements.

[0034] In this embodiment, the template structure of the remote sensing basic model is defined and edited according to the selected remote sensing basic model framework. The algorithm framework and version are selected according to the task requirements, the mirror environment required for algorithm training is added, the training command is entered and the algorithm code file is uploaded, and the template integration can be completed after adjusting the resource information.

[0035] In operation S4, the remote sensing sample data set is input into the remote sensing basic model architecture, and the remote sensing basic model architecture is trained using the edited algorithm framework and mirror environment.

[0036] According to an embodiment of the present disclosure, before operation S4, operations S400 to S401 are also included:

[0037] In operation S400, parameters of the remote sensing basic model framework are initialized and set.

[0038] In this embodiment, template editing can be used to set default parameters for the initial model, or a pre-trained model can be uploaded to initialize the model to shorten the training time.

[0039] In operation S401, a training task is created according to a remote sensing sample data set, an algorithm framework, and a mirror environment.

[0040] In this embodiment, a prepared remote sensing sample data set, algorithm framework, and mirror environment are used to create a single-machine or distributed training task.

[0041] According to an embodiment of the present disclosure, operation S4 may further include operations S402 to S403:

[0042] In operation S402, according to the training task, the remote sensing sample data set is input into the remote sensing basic model framework for training.

[0043] In this embodiment, according to the created stand-alone or distributed training task, the remote sensing sample data set is input into the remote sensing basic model architecture for training, and an appropriate training batch size and number of training rounds are set.

[0044] In operation S403, the parameters of the algorithm framework are iteratively updated using an intelligent parameter optimization method.

[0045] According to an embodiment of the present disclosure, operation S403 may further include the following operations:

[0046] Determine the hyperparameters of the algorithm framework, where the hyperparameters include learning rate, number of training rounds, etc.

[0047] Select a suitable intelligent parameter optimization algorithm, such as grid search, random search, Bayesian optimization search, genetic algorithm, etc. Different algorithms have different search strategies and optimization performance. Select a suitable algorithm according to the specific situation and iteratively optimize the hyperparameters according to the intelligent parameter optimization algorithm.

[0048] Get the evaluation metrics of the remote sensing base model architecture on the validation set.

[0049] According to the evaluation indicators, the part of the hyperparameter that reaches the indicator threshold during the iterative optimization process is used as the starting point of the next iteration until the iteration termination condition is reached. That is, according to the performance of the model on the validation set, the hyperparameters of the training process are adjusted to further improve the model performance.

[0050] In operation S5, the progress, parameter changes and performance indicators of the remote sensing basic model architecture during the training process are visually monitored.

[0051] According to an embodiment of the present disclosure, operation S5 may further include operations S500 to S502:

[0052] In operation S500 , a visualization dynamic library matching the remote sensing basic model architecture is determined.

[0053] In this embodiment, a suitable visualization dynamic library is selected, such as TensorBoard, Matplotlib, etc. These visualization dynamics provide visualization tools and functions that can be used to monitor and display the progress, parameter changes and performance indicators of the training process in real time.

[0054] In operation S501, the parameters and loss function of the remote sensing basic model architecture during the training process are transferred to the visualization dynamic library by dynamic mounting.

[0055] In operation S502, the progress, parameter changes, and performance indicators during the training process are monitored in real time according to the visualization dynamic library.

[0056] During the training process, use the functions and tools provided by the visualization dynamic library to monitor the training progress, parameter changes, and performance indicators in real time. For example, you can draw loss function curves, display training accuracy, and visualize parameter changes.

[0057] In another embodiment, operation S3 may further include the following operations:

[0058] Create a template, wherein the template is pre-integrated with at least one of the tensor flow framework (TensorFlow), Pytorch, Feipang framework, MXNet, MindSpore, Caffe, Caffe2 and the just-in-time compilation framework (Jittor), and the template also integrates at least one of the X86_64 image architecture, the MIPS_64 image architecture and the ARM_64 image architecture.

[0059] Select the algorithm framework corresponding to the task requirements in the template and configure it into the remote sensing basic model architecture.

[0060] Since mainstream frameworks such as TensorFlow, Pytorch, Feipian Framework, MXNet, MindSpore, Caffe, Caffe2 and Just-in-Time Compilation Framework (Jittor), as well as X86_64 image architecture, MIPS_64 image architecture and ARM_64 image architecture are pre-integrated into the template, users no longer need to spend time editing the corresponding algorithms and environments themselves. Instead, they can directly select the corresponding algorithms and environments in the template according to task requirements, which integrates the algorithms and brings great convenience to users.

[0061] Figure 2 The structural block diagram of the remote sensing basic model rapid training system based on template editing according to an embodiment of the present invention is schematically shown.

[0062] like Figure 2 As shown, the remote sensing basic model rapid training system based on template editing of this embodiment includes: a sample data management module 301, a template management module 302, and a training management module 303.

[0063] The sample data management module 301 is used to obtain a pre-labeled remote sensing sample data set. The sample data management module 301 has sample integration and labeling capabilities, wherein the remote sensing sample data set includes: one of a remote sensing image data set, a remote sensing text data set, a remote sensing voice data set, a remote sensing video data set, a remote sensing electronic data set, and a remote sensing situation data set.

[0064] The template management module 302 is used to determine the remote sensing basic model architecture corresponding to the task requirements according to the user's task requirements; and edit the corresponding algorithm framework and mirror environment for the remote sensing basic model architecture based on the task requirements.

[0065] The training management module 303 is used to input the remote sensing sample data set into the remote sensing basic model architecture, train the remote sensing basic model architecture using the edited algorithm framework and mirror environment; and visually monitor the progress, parameter changes and performance indicators of the remote sensing basic model architecture during the training process.

[0066] In another embodiment, the remote sensing basic model rapid training system based on template editing further includes: a task scheduling module 304 .

[0067] The task scheduling module 304 is used to provide resource load balancing and task tracing capabilities.

[0068] In the training of remote sensing basic models, on the one hand, a large amount of computing resources are needed to process large-scale data sets and complex model structures. The goal of resource load balancing is to ensure that these resources can be fully utilized and avoid excessive or insufficient use of resources to improve the efficiency and speed of training. On the other hand, the training of remote sensing basic models involves multiple tasks, such as sample data preprocessing, model training, verification and evaluation. The purpose of task tracing is to track and record the start time, completion time, execution progress, execution results and other information of each task. Through task tracing, you can understand the execution status of the task and promptly discover and solve problems in task execution, such as errors, anomalies or performance degradation. Task tracing can also be used to analyze and optimize the training process to help improve the training effect and performance of the model.

[0069] According to an embodiment of the present invention, any multiple modules of the sample data management module 301, the template management module 302, and the training management module 303 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present invention, at least one of the sample data management module 301, the template management module 302, and the training management module 303 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, at least one of the sample data management module 301, the template management module 302, and the training management module 303 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0070] It should be noted that the remote sensing basic model rapid training system based on template editing in the embodiment of the present invention is partially corresponding to the remote sensing basic model rapid training method based on template editing in the embodiment of the present invention, and their specific implementation details and the technical effects brought about are also the same, which will not be repeated here.

[0071] The flow chart and block diagram in the accompanying drawings illustrate the system architecture, functions and operations that may be implemented according to the system and method of various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0072] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present invention may be combined and / or combined in various ways, even if such combinations and / or combinations are not explicitly described in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features described in the various embodiments of the present invention may be combined and / or combined in various ways. All of these combinations and / or combinations fall within the scope of the present invention.

[0073] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although each embodiment is described above, it does not mean that the measures in each embodiment cannot be used in combination advantageously. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A remote sensing basic model rapid training method based on template editing, characterized in that: include: Obtain pre-labeled remote sensing sample datasets; Determine the remote sensing basic model architecture corresponding to the task requirements according to the user's task requirements; Based on the task requirements, the corresponding algorithm framework and mirror environment are edited for the remote sensing basic model architecture, including: Creating a template, wherein the template is pre-integrated with at least one of a tensorflow framework, Pytorch, a paddle framework, MXNet, MindSpore, Caffe, Caffe2, and a just-in-time compilation framework; The template also integrates at least one of the X86_64 image architecture, the MIPS_64 image architecture, and the ARM_64 image architecture; Selecting an algorithm framework corresponding to the task requirement in the template and configuring it into the remote sensing basic model architecture; Initializing parameters of the remote sensing basic model architecture; Creating a training task based on the remote sensing sample data set and the algorithm framework and mirror environment; Inputting the remote sensing sample data set into the remote sensing basic model framework, and training the remote sensing basic model framework using the edited algorithm framework and mirror environment, including: According to the training task, the remote sensing sample data set is input into the remote sensing basic model framework for training; Iteratively update the parameters of the algorithm framework using an intelligent parameter optimization method; Visually monitor the progress, parameter changes and performance indicators of the remote sensing basic model architecture during training.

2. The remote sensing basic model rapid training method based on template editing according to claim 1 is characterized in that: The method of iteratively updating the parameters of the algorithm framework using the intelligent parameter adjustment optimization method includes: Determining hyperparameters of the algorithm framework; Iteratively optimize the hyperparameters according to an intelligent parameter optimization algorithm; Obtaining evaluation indicators of the remote sensing basic model architecture on a validation set; According to the evaluation index, the part of the hyperparameter that reaches the index threshold during the iterative optimization process is used as the starting point of the next iteration until the iteration termination condition is reached.

3. The remote sensing basic model rapid training method based on template editing according to claim 1 is characterized in that: The visual monitoring of the progress, parameter changes and performance indicators of the remote sensing basic model architecture during the training process includes: Determining a visualization dynamic library that matches the remote sensing basic model architecture; The parameters and loss function of the remote sensing basic model architecture during the training process are transferred to the visualization dynamic library by means of dynamic mounting; The progress, parameter changes and performance indicators during the training process are monitored in real time according to the visual dynamic library.

4. The remote sensing basic model rapid training method based on template editing according to claim 1 is characterized in that: The creating of a training task according to the remote sensing sample data set and the algorithm framework and the mirror environment comprises: A single-machine training task or a distributed training task is created according to the remote sensing sample data set and the algorithm framework and the mirror environment.

5. The remote sensing basic model rapid training method based on template editing according to claim 1 is characterized in that: The remote sensing sample data set includes: one of a remote sensing image data set, a remote sensing text data set, a remote sensing voice data set, a remote sensing video data set, a remote sensing electronic data set, and a remote sensing situation data set.

6. A remote sensing basic model rapid training system based on template editing, comprising: The sample data management module is used to obtain pre-labeled remote sensing sample data sets; A template management module, used to determine a remote sensing basic model architecture corresponding to the task requirements according to the user's task requirements; And based on the task requirements, the corresponding algorithm framework and mirror environment are edited for the remote sensing basic model architecture, including: Creating a template, wherein the template is pre-integrated with at least one of a tensorflow framework, Pytorch, a paddle framework, MXNet, MindSpore, Caffe, Caffe2, and a just-in-time compilation framework; The template also integrates at least one of the X86_64 image architecture, the MIPS_64 image architecture, and the ARM_64 image architecture; Selecting an algorithm framework corresponding to the task requirement in the template and configuring it into the remote sensing basic model architecture; And initializing parameters of the remote sensing basic model architecture; Creating a training task based on the remote sensing sample data set and the algorithm framework and mirror environment; The training management module is used to input the remote sensing sample data set into the remote sensing basic model framework, and train the remote sensing basic model framework using the edited algorithm framework and mirror environment, including: According to the training task, the remote sensing sample data set is input into the remote sensing basic model framework for training; Iteratively update the parameters of the algorithm framework using an intelligent parameter optimization method; And visually monitor the progress, parameter changes and performance indicators of the remote sensing basic model architecture during the training process.

Images