A data labeling and data set preparation method supporting spatial science experiment analysis

Abstract

The embodiment of the application provides a data labeling and dataset preparation method supporting spatial science experiment analysis, and solves the technical problem that a spatial science experiment data labeling method is difficult to meet high-quality labeling requirements. The method comprises: in response to an input completion operation on a label name item, jumping to a labeling interface associated with a task type item; based on a task type item and a preset experiment condition, generating a category labeling box of a to-be-labeled object according to a selection operation of any to-be-labeled object input in a selection area of a category item; labeling the to-be-labeled object on an image frame according to the category labeling box to obtain a labeling result; and in the case of selecting an image labeling dataset, marking the to-be-labeled object on the image frame through a first labeling box. In the case of selecting an image labeling dataset, the labeling result is associated with the time of a time axis labeling area. Through dynamic interface adaptation, intelligent labeling assistance and space-time association storage, the high-quality labeling requirements are met.

Description

Technical Field

[0001] This application relates to the field of space science data processing technology, and in particular to a data annotation and dataset preparation method that supports space science experimental analysis. Background Technology

[0002] As a space laboratory, the space station possesses the capability to cover all disciplines of space science. It boasts unique advantages such as strong on-orbit support capabilities, support for astronaut participation, and round-trip transportation between Earth and space, enabling large-scale, multidisciplinary space science experiments. During the planning period, thousands of scientific experiments will be implemented, continuously acquiring massive amounts of scientific data and driving major scientific breakthroughs in fields such as space life sciences, microgravity physics, and space materials. Statistics show that over 80% of the data generated by on-orbit scientific experiments consists of images and videos. This data contains a wealth of crucial scientific information, covering multiple key research directions including space life sciences and biotechnology, space materials science, space fluid physics and combustion, and fundamental space physics. It forms the core foundation for monitoring experimental processes, discovering scientific laws, data mining, and intelligent analysis. Taking space life sciences as an example, the behavior, development, and neural function changes of model animals (such as fruit flies, nematodes, and zebrafish) in microgravity environments are a key research focus. During experiments, high-resolution cameras are typically used to continuously record their behavior, forming large-scale spatiotemporal sequence image data. Researchers need to perform individual attitude estimation, behavior recognition, and group interaction modeling to reveal the impact of the space environment on life activities. For example, in space materials science and fluid physics experiments, high-speed imaging is often used to record the solidification process of containerless molten metal, the flow morphology of two-phase flow, and the structural evolution of combustion flames, generating multi-channel, high-frame-rate, microscale image or video sequences to assist in the modeling of interface changes and the analysis of microscopic mechanisms.

[0003] Currently, artificial intelligence (AI) technology enables the automatic identification, parameter extraction, and modeling analysis of behavioral features in experimental images. For example, after image preprocessing (such as noise reduction and background segmentation) of model animal movement videos obtained in ground experiments or simulated microgravity experiments, convolutional neural networks (such as ResNet-50) can be used to extract frame-level image features. Subsequently, combined with time series modeling methods (such as LSTM or Support Vector Regression (SVR), the dynamic changes in behavior are modeled, and representative parameters such as trajectory curvature, angular velocity, and segmental oscillation frequency are extracted. A feature fusion module can encode spatial and temporal information into low-dimensional feature vectors. SHAP values ​​are used to evaluate the model's sensitivity to various behavioral variables, and dimensionality reduction clustering methods such as t-SNE are used to visualize, classify, and analyze behavioral patterns, thereby assisting in the identification of behavioral features significantly affected by microgravity. In specific implementations, DeepLabCut can be used to automatically detect and locate key parts in video frames (such as fruit fly wings, nematode segments, and zebrafish spines), generating time series key point coordinate information (such as head orientation angle and trunk curvature). For specific behavioral events (such as the C-shaped turn of nematodes and the spiral swimming of zebrafish), frame-level event annotation is performed using tools such as BORIS, combined with expert-defined binary behavioral labels. In complex behavioral modeling, a sliding window strategy can be used to segment the video and fuse multimodal labels (such as velocity and pose categories) to construct a multi-task supervised learning model. This model utilizes LSTM to handle temporal dependencies while simultaneously predicting behavioral categories and dynamic parameters in parallel. Classification and regression performance is improved through joint optimization (cross-entropy loss and mean squared error). Finally, the consistency of the labeling system is evaluated using a confusion matrix and Kappa coefficients, and the contribution of key behavioral features is analyzed using SHAP to ensure the interpretability and reliability of the results.

[0004] However, space experimental data generally exhibits characteristics such as high dimensionality, multimodality, strong temporal sequence, and low signal-to-noise ratio. Traditional data annotation methods typically suffer from poor adaptability, cumbersome operation, and mixed label usage in terms of task configuration, label management, and interaction efficiency, making it difficult to meet the actual needs of space science experiments for high-quality data annotation. Therefore, there is an urgent need for an efficient, intelligent, and customizable data annotation and analysis system to improve the efficiency and annotation quality of space experimental data processing and promote the development of complex scientific experiments towards refinement and intelligence. Summary of the Invention

[0005] This application provides a data annotation and dataset preparation method that supports space science experiment analysis, which can solve the technical problem that space science experiment data annotation methods are difficult to meet the requirements of high-quality annotation.

[0006] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0007] In a first aspect, embodiments of this application provide a data annotation and dataset preparation method supporting space science experimental analysis. This method includes: responding to an input completion operation for a label name item, navigating to an annotation interface associated with a task type item; the annotation interface includes a display area for a dataset file and a selection area for a category item; the display area is used to display at least one image frame of the dataset file; the dataset file includes an image annotation dataset and a video annotation dataset of space animal behavior; the category item is a category name characterizing the features to be labeled; based on an input operation to select any object to be labeled in the selection area of ​​the category item, and based on the task type item and preset experimental conditions, generating a category label box for the object to be labeled; labeling the object to be labeled on the image frame according to the category label box, obtaining a labeling result; the labeling result is at least one of a pose estimation result, a tracking result, and a behavior labeling result; wherein, when an image annotation dataset is selected, the category label box includes a first label box; the first label box includes multiple label points connected sequentially to form a closed label box; the annotation interface also includes a selection area for a category item, confirming the object to be labeled based on an input operation to select any object to be labeled in the selection area of ​​the category item; labeling the object to be labeled on the image frame using the first label box, obtaining a labeling result. When an image annotation dataset is selected, the annotation interface also includes a category selection area and a time axis annotation area; the annotation results are correlated with the time of the time axis annotation area.

[0008] Based on the above description of the data annotation and dataset preparation method for supporting space science experimental analysis provided in the embodiments of this application, it can be seen that this data annotation and dataset preparation method for supporting space science experimental analysis includes automatically jumping to the corresponding annotation interface according to the user-selected task type (such as posture estimation, behavior tracking), avoiding manual switching and realizing responsive interface jumping. The annotation interface is adapted to image annotation datasets and video annotation datasets, and the display area dynamically loads image frames or video streams to achieve multi-dataset support. The first annotation box (closed polygon) is used for image datasets, and animal outlines (such as mouse joints, head) are accurately marked by sequentially connected annotation points. The closed polygon annotation box can capture animal posture details. The time axis annotation area is used for video datasets, and the annotation results (such as "chasing behavior") are associated with the time axis, supporting frame-level behavior segmentation. Time axis associated annotation supports the statistics of behavior duration. It supports mixed output of posture estimation (annotation point coordinates), tracking results (ID + trajectory), and behavior annotation (such as "grooming"). By using dynamic interface adaptation, intelligent annotation assistance, and spatiotemporal related storage, the problems of low efficiency, poor consistency, and spatiotemporal information fragmentation in animal behavior annotation are solved, thereby meeting the requirements for high-quality annotation.

[0009] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis further includes: when selecting an image annotation dataset, the category annotation box includes a second annotation box; the second annotation box includes a rectangular selection box; the annotation interface also includes a category item selection area and a time axis annotation area; based on the operation of selecting any object to be annotated entered in the category item selection area, the object to be annotated and the annotation time period of the object to be annotated are confirmed; the object to be annotated is marked on any image frame within the annotation time period using the second annotation box to obtain the annotation result.

[0010] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis also includes: when selecting an image annotation dataset, the category annotation box includes a time period selector; the annotation interface also includes a category item selection area and a time axis annotation area; based on the operation of selecting any object to be annotated in the category item selection area, the object to be annotated and the annotation time period of the object to be annotated are confirmed; within the annotation time period, the object to be annotated is marked by selecting sub-time periods through the time period selector, and the annotation result is obtained.

[0011] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experimental analysis further includes: when selecting an image annotation dataset, the category annotation box includes a third annotation box; the third annotation box includes multiple annotation points connected sequentially to form a closed annotation box; multiple category items include connection relationships, and the annotation interface also includes a time axis annotation area; confirm the object to be annotated and the annotation time period of the object to be annotated; within the annotation time period, mark the object to be annotated on any image frame through the third annotation box to obtain a first annotation result; in response to the generation of the first annotation result, generate and display multiple category items and connection relationships; adjust the position of the category items on the first annotation result to obtain the annotation result.

[0012] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experimental analysis further includes: responding to the input operation of creating a dataset, selecting a dataset type on a first interface to jump to a second interface; the dataset types include image annotation datasets and video annotation datasets of space animal behavior; responding to the input operation of uploading data on the second interface, uploading a dataset file according to the dataset type; responding to the input operation of creating a labeling task on the second interface, jumping to a third interface to perform the operation of inputting a task name item and selecting a task type item on the third interface; the task type item includes image annotation selection items and video annotation selection items; the task type item corresponds to the dataset type; responding to the input confirmation operation on the third interface, jumping to the second interface associated with the task type item. The system consists of four interfaces for editing label name items in the label library. Upon confirmation on the fourth interface, a fifth interface is invoked. This fifth interface is associated with editing label name items in the label library, including tag name items. Each tag name item includes multiple category items. Upon completion of input on a tag name item, a labeling interface associated with the task type item is invoked. The labeling interface includes a display area for the dataset file and a selection area for category items. The display area shows at least one image frame from the dataset file. Based on the selection of any object to be labeled entered in the category item selection area, a category label box for the object to be labeled is generated, based on the task type item and preset experimental conditions. The object to be labeled is then labeled on the image frame according to the category label box, yielding the labeling result. The labeling result is associated with the label name items in the label library.

[0013] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis further includes: responding to a confirmation operation entered on the annotation interface, redirecting to an annotation review interface; the annotation review interface is generated based on pre-set review conditions associated with the task type item; responding to a confirmation operation entered on the annotation review interface, redirecting to a file storage interface; the file storage interface includes multiple file storage format options. In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis further includes: inputting the annotation results as a dataset file on a second interface.

[0014] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experimental analysis also includes: when the task type item is an image annotation option, the annotation type item includes instance segmentation option, object detection option, and image classification option; when the task type item is a video annotation option, the annotation type item includes process localization option, behavior detection option, and pose tracking option.

[0015] In the feasible implementation of the first aspect, in response to the input of a "create dataset" operation, a dataset type is selected on the first interface to jump to the second interface; the dataset types include image annotation datasets and video annotation datasets; both image annotation datasets and video annotation datasets are space science experiment datasets; in response to the input of a "upload data" operation on the second interface, the dataset file is uploaded according to the dataset type; in response to the input of a "create labeling task" operation on the second interface, the user jumps to the third interface to perform the operations of inputting a task name item and selecting a task type item; the task type items include image annotation selection items and video annotation selection items; the task type items correspond to the dataset types; in response to the input of a "confirm" operation on the third interface, the user jumps to the fourth interface associated with the task type item. The system allows users to edit the label name entries in the label library. Upon confirmation on the fourth interface, the system navigates to the fifth interface. The fifth interface is associated with the editable label name entries, including the tag name entries. Each tag name entry includes multiple category entries. Upon completion of input on the tag name entries, the system navigates to the annotation interface associated with the task type. The annotation interface includes a display area for the dataset file and a selection area for the category entries. The display area shows at least one image frame from the dataset file. Based on the selection of any object to be annotated in the category selection area, and considering the task type and preset experimental conditions, a category annotation box for the object to be annotated is generated. The object to be annotated is then annotated on the image frame based on the category annotation box, yielding the annotation result. The annotation result is associated with the label name entries in the label library.

[0016] Based on the above description of the data annotation and dataset preparation method for supporting space science experimental analysis provided in the embodiments of this application, it can be seen that the data annotation and dataset preparation method for supporting space science experimental analysis includes an initial diversion mechanism based on "selecting the dataset type on the first interface in response to the input dataset creation operation," which forces users to complete the necessary configuration according to preset experimental conditions and reduces redundant operations. Secondly, through the automated tag library binding rule of "corresponding task type item to dataset type," it ensures that the annotation system and data modality are strictly matched (e.g., image annotation tasks only call the image tag library), avoiding the risk of cross-modal tag mixing. Finally, relying on "dynamically adapting components to the annotation interface according to the task type" (e.g., video annotation loading timeline control, image annotation enhanced region selection tool), while ensuring the consistency of basic functions, the interaction path is optimized for different scenarios such as image frame annotation and video time-series annotation. Finally, the structured storage of annotation results is achieved through "real-time association between category annotation boxes and tag libraries," forming a closed-loop efficiency improvement from data upload, task configuration to annotation execution. This allows for the construction of an on-orbit experimental data processing system that meets the diverse and personalized analytical needs of space science experiments. This system includes efficient annotation methods, standardized dataset construction, intelligent analysis, and customized applications. By providing a flexible and configurable annotation workflow, personalized dataset management, and intelligent analysis technologies, the system ensures high-quality, highly applicable, and scalable data to support in-depth research and intelligent development in space science experiments. Ultimately, this enhances the data annotation capabilities for space science experimental analysis.

[0017] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis also includes: in response to a confirmation operation entered in the annotation interface, jumping to the annotation review interface; the annotation review interface is generated based on pre-set review conditions associated with the task type item; in response to a confirmation operation entered in the annotation review interface, jumping to the file storage interface; the file storage interface includes multiple file storage format options.

[0018] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis also includes: when selecting image annotation options, the category annotation box includes a first annotation box; the first annotation box includes multiple annotation points connected in sequence to form a closed annotation box; the annotation interface also includes a category selection area, and the object to be annotated is confirmed by selecting any object to be annotated according to the operation entered in the category selection area; the object to be annotated is marked on the image frame through the first annotation box to obtain the annotation result.

[0019] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis further includes: when selecting video annotation options, the category annotation box includes a second annotation box; the second annotation box includes a rectangular selection box; the annotation interface also includes a category selection area and a time axis annotation area; based on the operation of selecting any object to be annotated entered in the category selection area, the object to be annotated and the annotation time period of the object to be annotated are confirmed; the object to be annotated is marked on any image frame within the annotation time period using the second annotation box to obtain the annotation result.

[0020] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis also includes: when selecting video annotation options, the category annotation box includes a time period selector; the annotation interface also includes a category selection area and a time axis annotation area; based on the operation of selecting any object to be annotated in the category selection area, the object to be annotated and the annotation time period of the object to be annotated are confirmed; within the annotation time period, the object to be annotated is marked by selecting sub-time periods through the time period selector, and the annotation result is obtained.

[0021] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experiment analysis further includes: when selecting the video annotation option, the category annotation box includes a third annotation box; the third annotation box includes multiple annotation points connected sequentially to form a closed annotation box; multiple category items include connection relationships, and the annotation interface also includes a time axis annotation area; confirm the object to be annotated and the annotation time period of the object to be annotated; within the annotation time period, mark the object to be annotated on any image frame through the third annotation box to obtain the first annotation result; in response to the generation of the first annotation result, generate and display multiple category items and connection relationships; adjust the position of the category items on the first annotation result to obtain the annotation result.

[0022] In the feasible implementation of the first aspect, the data annotation and dataset preparation method that supports space science experiment analysis also includes: inputting the annotation results as a dataset file on the second interface.

[0023] In the feasible implementation of the first aspect, the data annotation and dataset preparation method supporting space science experimental analysis also includes: when the task type item is an image annotation option, the annotation type item includes instance segmentation option, object detection option, and image classification option; when the task type item is a video annotation option, the annotation type item includes process localization option, behavior detection option, and pose tracking option.

[0024] Secondly, embodiments of this application provide a data annotation and dataset preparation system for supporting space science experimental analysis. The data annotation and dataset preparation system for supporting space science experimental analysis includes: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method provided in the first aspect.

[0025] In this way, based on the initial triage mechanism of "reacting to the input of the dataset creation operation and selecting the dataset type on the first interface," users are forced to complete the necessary configuration according to the preset experimental conditions, reducing redundant operations. Secondly, through the automated tag library binding rule of "corresponding task type items with dataset type," the annotation system and data modality are strictly matched (e.g., image annotation tasks only call the image tag library), avoiding the risk of cross-modal tag mixing. Finally, relying on "the annotation interface dynamically adapts components according to the task type" (e.g., video annotation loading timeline control, image annotation enhancement region selection tool), while ensuring the consistency of basic functions, the interaction path is optimized for different scenarios such as image frame annotation and video time-series annotation. Finally, the structured storage of annotation results is achieved through "real-time association between category annotation boxes and tag libraries," forming a closed-loop efficiency improvement from data upload, task configuration to annotation execution.

[0026] Thirdly, embodiments of this application provide a computer-readable medium having computer program instructions stored thereon, which can be executed by a processor to implement the method provided in the first aspect.

[0027] In this way, based on the initial triage mechanism of "reacting to the input of the dataset creation operation and selecting the dataset type on the first interface," users are forced to complete the necessary configuration according to the preset experimental conditions, reducing redundant operations. Secondly, through the automated tag library binding rule of "corresponding task type items with dataset type," the annotation system and data modality are strictly matched (e.g., image annotation tasks only call the image tag library), avoiding the risk of cross-modal tag mixing. Finally, relying on "the annotation interface dynamically adapts components according to the task type" (e.g., video annotation loading timeline control, image annotation enhancement region selection tool), while ensuring the consistency of basic functions, the interaction path is optimized for different scenarios such as image frame annotation and video time-series annotation. Finally, the structured storage of annotation results is achieved through "real-time association between category annotation boxes and tag libraries," forming a closed-loop efficiency improvement from data upload, task configuration to annotation execution. Attached Figure Description

[0028] Figure 1a A schematic diagram of a data annotation and dataset preparation system for supporting space science experimental analysis provided in this application embodiment;

[0029] Figure 1b A schematic diagram of a data annotation and dataset preparation system for supporting space science experimental analysis provided in this application embodiment;

[0030] Figure 2 A flowchart illustrating a data annotation and dataset preparation method for supporting space science experimental analysis, provided as an embodiment of this application;

[0031] Figure 3a A flowchart illustrating a data annotation and dataset preparation method for supporting space science experimental analysis, provided as an embodiment of this application;

[0032] Figure 3b A flowchart illustrating a data annotation and dataset preparation method for supporting space science experimental analysis, provided as an embodiment of this application;

[0033] Figure 4 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0034] Figure 5 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0035] Figure 6 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0036] Figure 7 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0037] Figure 8 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0038] Figure 9 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0039] Figure 10 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0040] Figure 11 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0041] Figure 12A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0042] Figure 13 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0043] Figure 14 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0044] Figure 15 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0045] Figure 16 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0046] Figure 17 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0047] Figure 18 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0048] Figure 19 A schematic diagram of an interface for a data annotation and dataset preparation method supporting space science experimental analysis provided in an embodiment of this application;

[0049] Figure 20 A flowchart illustrating a data annotation and dataset preparation method for supporting space science experimental analysis, provided as an embodiment of this application;

[0050] Figure 21 This is a flowchart illustrating a method for data annotation and dataset preparation to support space science experimental analysis, provided in an embodiment of this application. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. In the description of the embodiments of the present invention, unless otherwise stated, "multiple" refers to two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0052] Furthermore, to facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" are not necessarily different. Meanwhile, in the embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is being used as an example, illustration, or description. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner for ease of understanding.

[0053] The principles and features of this application are described below. The examples given are only for explaining this application and are not intended to limit the scope of this application.

[0054] Currently, data annotation in space science experiments still heavily relies on a combination of manual operation and limited automation. Although deep learning technology has achieved significant results in general tasks such as image recognition, space experimental data (such as behavioral videos in space life experiments, phase transition images in containerless material experiments, flame evolution sequences in combustion experiments, and dynamic images of two-phase flows in fluid experiments) often have characteristics such as low signal-to-noise ratio, high-dimensional temporal sequence, and modal complexity, making it difficult for general algorithms to accurately model their space characteristics and physical processes. For example, the solidification process of materials under microgravity conditions on the space station requires manual identification of crystal nucleation and growth interfaces from a large number of image sequences captured by high-speed cameras. Existing automated tools are prone to misjudgment under conditions such as phase transition blurring and strong background noise. In addition, as the scale of experiments expands and the sampling frequency increases, the amount of data grows exponentially. For example, a single high-throughput life experiment can generate tens of TB of video and multimodal sensor data, while the data annotation speed lags far behind the acquisition capacity, forming a "data dam".

[0055] Meanwhile, a knowledge gap exists between domain experts and algorithm developers, further exacerbating the annotation challenge. Annotating the physical meaning of space science experiments (such as determining the behavioral state of space animals, identifying material evolution stages, and extracting flame structural features) requires reliance on experimental background, physical models, and professional experience, making it difficult to describe using general rules. Algorithm engineers often lack an understanding of space experiment design and scientific objectives, making it difficult to systematically transform domain knowledge into trainable labels. Conversely, while scientists are proficient in the experimental content, they are unfamiliar with standardized data annotation processes, tool usage, and label system construction, leading to annotation results that are highly subjective, inconsistent across tasks, and difficult to reuse. This "two-way cognitive barrier" hinders the continuous optimization and widespread application of intelligent annotation systems, preventing the efficient transformation of massive amounts of space experimental data into scientific research value.

[0056] To meet the diverse and customized data analysis needs of space science experiments, this application proposes a data annotation and dataset preparation method and system applicable to the field of space science experiments. This method is applicable to multiple experimental directions such as space physics, life sciences, materials science, and fluid mechanics, supporting multi-level applications from experimental process monitoring and scientific phenomenon identification to high-dimensional data mining and intelligent analysis. It can effectively improve the usability and intelligent processing capabilities of space science experimental data. The system features a highly configurable annotation workflow, a task-driven multimodal adaptation mechanism, and a standardized label management and storage module. It can automatically match annotation templates according to experimental tasks, supporting the annotation needs of various data types such as image frames, video sequences, and spatiotemporal annotations, significantly improving data annotation efficiency and consistency, and providing a high-quality data foundation for space science research.

[0057] This application provides a data annotation and dataset preparation system that supports space science experiment analysis, and can execute the data annotation and dataset preparation method for supporting space science experiment analysis provided in this application. Figure 1a This is a schematic diagram of a data annotation and dataset preparation system that supports space science experimental analysis, provided as an embodiment of this application.

[0058] like Figure 1a As shown, the data annotation and dataset preparation system 001 supporting space science experiment analysis includes at least one processor 011 and a memory 012 communicatively connected to the at least one processor; wherein, the memory 012 stores instructions that can be executed by the at least one processor 011, and the instructions are executed by the at least one processor 011 to enable the at least one processor 011 to execute the data annotation and dataset preparation method supporting space science experiment analysis provided in the embodiments of this application.

[0059] like Figure 1bAs shown, in some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes a 6-layer structure, from bottom to top: infrastructure layer, data storage layer, container platform layer, service layer, application layer, and presentation layer (client).

[0060] The infrastructure layer, the lowest layer of this platform, consists of cluster nodes, including cluster management nodes, compute worker nodes, data storage nodes, and other nodes. This layer, based on interface encapsulation, primarily interacts with the cloud unified service management system to achieve unified management and operation of the cluster nodes.

[0061] The data storage layer is used for data storage, which includes user information, platform configuration information, log information, user data, file data, etc.

[0062] The container platform layer includes Kubernetes clusters, a managed converged cloud container control panel, and Kubernetes CRD orchestration tools. The Kubernetes cluster serves as the control plane, containing load balancers, access brokers, the control plane, and internal Kubernetes interfaces.

[0063] The service layer is deployed and configured based on the container platform layer to support the functional applications of the upper application layer. For example, the service layer includes five software components: data management service, multi-source data access service, organization service, user service, and statistics service. Each component has its own independent service layer, as well as a unified front-end access service layer, which includes proxy services and data access services.

[0064] The application layer, the core functional layer of the public cloud data ecosystem service platform, provides the functionality of key components. For example, the application layer includes data deployment services, data visualization, deep learning model development, and data annotation. Data visualization includes image data visualization, video data visualization, and 2D or 3D icon visualization. Deep learning model development includes Notebook, Canvas, IDE, and task management and collaboration. Data annotation includes image annotation, video annotation, a label library, and annotation tasks.

[0065] The presentation layer (client-side) is the top layer of this platform, a browser / server (B / S) architecture software based on web services. Users can access and use the platform from the front end through any system's browser, WeChat client, WeChat official account, etc.

[0066] In some embodiments, the data annotation and dataset preparation system 001 supporting space science experiment analysis is built based on Vue3 + TypeScript. For example, an annotation task module is built based on Vue3 + TypeScript.

[0067] In some embodiments, the data annotation and dataset preparation system 001 supporting space science experimental analysis adopts a modular and responsive front-end architecture. Pinia is used for global state management to ensure the consistency and sharing of labeled data. Axios is used to enable front-end and back-end data interaction, supporting CRUD operations for annotation tasks.

[0068] To enable efficient uploading and management of large files, in some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes an object storage SDK for managing raw annotation materials.

[0069] In some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes a Pinia store for maintaining the global state of annotation tasks, including task progress, user annotation information, etc.

[0070] In some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes an HTML5 video player for providing precise video playback control.

[0071] In some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes an integrated view-ui-plus for providing a user-friendly interface.

[0072] In some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes Axios, which is used to synchronize annotation results to the backend in real time to ensure data consistency and security.

[0073] To provide a fast development server and efficient packaging performance, in some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes Vite as a development build tool.

[0074] To enhance code quality and development experience, in some embodiments, the data annotation and dataset preparation system 001 that supports space science experiment analysis includes TypeScript for strong type checking.

[0075] In this way, by using the aforementioned technology stack, a data annotation and dataset preparation system that supports space science experimental analysis can be built, capable of performing high-performance and user-friendly annotation task solutions.

[0076] In some contexts, different experimental research objectives differ, and the data annotation and analysis needs vary due to differences in experimental content, research direction, and data characteristics. Traditional fixed-process analysis methods are difficult to meet the personalized needs of scientists.

[0077] Figure 2 This is a flowchart illustrating a data annotation and dataset preparation method for supporting space science experimental analysis, provided as an embodiment of this application. Figure 2 As shown, in some embodiments, the data annotation and dataset preparation method supporting space science experiment analysis includes the following steps:

[0078] S01, in response to the completion of the input for the label name item, navigates to the annotation interface associated with the task type item. The annotation interface includes a display area for the dataset file and a selection area for the category item. The display area is used to display at least one image frame from the dataset file. The dataset file includes an image annotation dataset and a video annotation dataset of spatial animal behavior. The category item is used to characterize the category name of the feature to be labeled.

[0079] S02, based on the operation of selecting any object to be labeled entered in the category selection area, generate the category label box of the object to be labeled according to the task type item and the preset experimental conditions.

[0080] S03, based on the category bounding boxes, label the objects to be labeled on the image frame to obtain the labeling results. The labeling results are at least one of the pose estimation results, tracking results, and behavior labeling results.

[0081] When an image annotation dataset is selected, the category annotation box includes a first annotation box. The first annotation box consists of multiple annotation points connected sequentially to form a closed annotation box. The annotation interface also includes a category selection area. By selecting any object to be annotated in the category selection area, the object is confirmed. The object to be annotated is marked on the image frame using the first annotation box, resulting in the annotation result. When an image annotation dataset is selected, the annotation interface also includes a category selection area and a timeline annotation area. The annotation result is temporally correlated with the timeline annotation area.

[0082] Figure 3a This is a flowchart illustrating a data annotation and dataset preparation method for supporting space science experimental analysis, provided as an embodiment of this application. Figure 3a As shown, in some embodiments, the data annotation and dataset preparation method supporting space science experiment analysis includes the following steps:

[0083] S1, in response to the input of the dataset creation operation, select the dataset type on the first interface to jump to the second interface.

[0084] Dataset types include image annotation datasets and video annotation datasets. In some embodiments, dataset types may also include text datasets.

[0085] like Figure 4 As shown, in one implementation, step S1 can begin with the user clicking the "Start" node button and then clicking the "Create" labeled dataset node button.

[0086] S2, in response to the data upload operation entered on the second interface, uploads the dataset file according to the dataset type.

[0087] In some embodiments, raw data is acquired from on-orbit scientific experimental platforms such as space stations and probes. This raw data includes image data and video data. Image data serves as the foundational data for image annotation tasks, and video data serves as the foundational data for video annotation tasks. In one implementation, based on task definition and task partitioning, the raw data is partitioned to obtain multiple subsets of data to be annotated.

[0088] For example, the task definition could be "location of the experimental process of containerless electrostatic levitation materials." The raw data includes video data of "a high-definition camera continuously recording the evolution of the material in an electromagnetic field." The task is divided into multiple experimental stages such as "sample ball release, sample heating, sample melting, deep supercooling and re-ignition, sample recovery, and abnormal sample displacement." The system divides the raw video data into the "containerless material levitation process" subset, and then completes the annotation process by combining it with specific annotation tasks (such as key process start and end time markers, key process category labels, and abnormal event annotations).

[0089] The labeled data is integrated into a structured, semantically clear, and task-ready AI-ready dataset for subsequent AI algorithm training and inference applications, supporting intelligent analysis tasks such as state recognition, process prediction, and anomaly detection in space science experiments.

[0090] like Figure 5 As shown, in one implementation, step S2 can be performed by the user clicking the "Save Uploaded Data" node button to save the file to be labeled. Understandably, if the file upload is successful, the process will proceed to "Generate Labeled Dataset," then to the "Processing Results" stage, and finally complete the entire process when the user clicks the "End" node button. However, if an upload fails due to a problem, the process will jump to the "Error Reporting" stage, requiring error handling and a retry until the labeled dataset is successfully created.

[0091] In some embodiments, users can create image / video annotation tasks within the platform. The data to be annotated in the task can use pre-created annotation dataset data sources, or it can be read directly from object storage or NAS space. In one implementation, the annotation task creation process starts when the user clicks the "Start" button. First, the user needs to "Select Data Source," i.e., determine the data source required for the annotation task. Then, the user proceeds to the "Create Annotation Task" step, which is a decision-making step. If creation fails, the process will proceed to the "Error Report" step, and then to the "Process Result" step; if creation is successful, the user will directly proceed to the "Process Result" step. Finally, the entire process is completed in the "End" step.

[0092] S3, in response to the "Create Mark Task" operation entered on the second interface, jumps to the third interface to perform the operations of entering the task name field and selecting the task type field on the third interface.

[0093] In some embodiments, the task type item includes image annotation options and video annotation options. The task type item corresponds to the dataset type. In some embodiments, the task type item may also include a text annotation option.

[0094] In some embodiments, when the task type item is an image annotation option, the annotation type item includes an instance segmentation option, an object detection option, and an image classification option.

[0095] When the task type option is video annotation, the annotation type options include process localization, behavior detection, and pose tracking.

[0096] like Figure 3b As shown, in some embodiments, by supporting multiple annotation tasks, including instance segmentation, object detection, process localization, behavior detection, and pose tracking, it is possible to combine with model-assisted annotation to improve annotation efficiency and accuracy. By supporting intelligent data processing in-orbit, the data transmission burden is reduced, and data utilization is improved.

[0097] Understandably, data annotation is not a one-time process, but rather an iterative one. Through model training and inference results, problems or deficiencies in the original annotations may be discovered, prompting a return to the annotation stage for correction and improvement. This feedback mechanism helps to continuously improve annotation quality and model performance, driving the continuous optimization of the entire system.

[0098] S4, in response to the confirmation operation entered on the third interface, jumps to the fourth interface associated with the task type item to edit the tag name item in the tag library.

[0099] In this way, a standardized labeled dataset is built, including a label library and label configuration, ensuring data consistency and reusability.

[0100] The labeling feature supports batch management at the "library" level, allowing the creation of label libraries for each labeling task. When creating a new labeling task, users can select a label library to apply a specific group of labels. Label libraries support multi-user editing and maintenance.

[0101] like Figure 6 As shown, in one implementation, the Label library creation process begins with the user clicking the "Start" button, first executing the "Create Label Library" operation. Then, the user enters the "Fill in Information" stage. Understandably, after completing the information, the system checks for "Successful Creation." If creation is successful, the process generates the "Created Label Library," processes it, and finally reaches the "End" node. If creation fails, the system displays an "Error" message and returns to the "Fill in Information" stage to repeat the process until the Label library is successfully created and the processing is complete.

[0102] S5, in response to the confirmation input on the fourth screen, jumps to the fifth screen.

[0103] The fifth interface links to the tag name item in the tag library, including the tag name item.

[0104] like Figure 8 As shown, the tag name item includes multiple category items.

[0105] S6, in response to the completion of input for the label name field, jumps to the labeling interface associated with the task type field.

[0106] The annotation interface includes a display area for the dataset file and a selection area for category items.

[0107] The display area is used to display at least one image frame from the dataset file.

[0108] S7. Based on the operation of selecting any object to be labeled entered in the selection area of ​​the category item, and based on the task type item and preset experimental conditions, generate the category label box of the object to be labeled.

[0109] Category annotation boxes include closed annotation boxes formed by connecting multiple annotation points in sequence, rectangular selection boxes, etc. For example, some category annotation boxes support image data annotation, point-line region delimitation, multi-person annotation, and automatic model annotation.

[0110] The details of step S8 are explained in detail and will not be repeated here.

[0111] S8: Based on the category labeling box, label the objects to be labeled on the image frame to obtain the labeling results.

[0112] The tagging results are associated with the tag name entries in the tag library.

[0113] like Figure 12 or Figure 13 As shown, in some embodiments, when an image annotation option is selected, the category annotation box includes a first annotation box. The first annotation box includes multiple annotation points connected sequentially to form a closed annotation box. The annotation interface also includes a category selection area, and the object to be annotated is confirmed based on the operation of selecting any object entered in the category selection area. The data annotation and dataset preparation method supporting space science experiment analysis also includes:

[0114] S811, mark the object to be annotated on the image frame using the first annotation box to obtain the annotation result.

[0115] like Figure 12 As shown, in one implementation, the boundaries of different object instances in an image are identified through instance segmentation, for example, different IDs and colors are assigned to different instances of the same class. The platform allows annotators to view images and perform instance segmentation annotation. For example, the instance segmentation process in image annotation begins when the user clicks the "Start" button. The user clicks the target instance segmentation annotation task and then selects "My Tasks - Manual Annotation" on the annotation task overview page. The system then displays the image to be annotated, and the user selects an appropriate label from preset labels and accurately segments the boundary of the target region by connecting lines. After completing the annotation, the user clicks submit, and the annotation data is stored in the metadata database. Then, the system automatically displays the next image to be annotated, repeating the above steps until all images are annotated, at which point the process finally enters the "End" node.

[0116] like Figure 13As shown, in another implementation, image data is labeled using object detection annotation, with the goal of identifying specific objects in the image. Besides marking the labeled areas with rectangles, object detection also involves indicating the object's category and location. Commonly used annotation methods include bounding box annotation, which uses rectangles to outline the object's location and indicates its category. The platform allows annotators to view images and perform object detection annotation. It supports image annotation, rectangle and category annotation, multi-user annotation, and automatic model annotation. The object detection annotation process begins when the user clicks "Start." First, the user selects an object detection annotation task and enters the task overview page, then clicks "My Tasks - Manual Annotation" to enter the annotation interface. After the system displays the image to be annotated, the user selects appropriate labels for the targets in the image and precisely annotates the target areas by drawing rectangles. After completing the annotation, the user submits the data, and the system stores it in the metadata database. If there are still unannotated images, the system automatically displays the next one, and the user continues annotating until all tasks are completed. Finally, the user clicks "End," marking the completion of the entire object detection annotation process.

[0117] like Figure 15 As shown, in some embodiments, when a video annotation option is selected, the category annotation box includes a second annotation box. The second annotation box includes a rectangular selection box. The annotation interface also includes a selection area for the category item and a timeline annotation area. The data annotation and dataset preparation method supporting space science experiment analysis further includes:

[0118] S821, based on the operation of selecting any object to be labeled entered in the selection area of ​​the category item, confirm the object to be labeled and the labeling time period of the object to be labeled.

[0119] S822: Mark the object to be annotated on any image frame within the annotation time period using the second annotation box to obtain the annotation result.

[0120] In one implementation, the user clicks on the target process to locate the annotation task, and then selects "My Tasks - Manual Annotation" on the annotation task overview page. The system then displays the videos to be annotated, from which the user must select appropriate tags and determine the start and end times and category of the video process. After confirming that everything is correct, the user clicks submit, and the data is stored in the metadata database. The system then automatically displays the next video to be annotated, repeating the above steps until all tasks are completed, at which point the process ends.

[0121] This allows for the identification and localization of experimental processes within a given timeframe in long videos, achieving video process localization while simultaneously labeling the temporal boundaries and category tags of process instances. The platform supports annotators playing the video, locating video segments, and labeling each segment with its type. It supports video data annotation, including annotation of video start and end times and categories, multi-person annotation, and automatic model annotation.

[0122] like Figure 14 As shown, in some embodiments, when a video annotation option is selected, the category annotation box includes a time period selector. The annotation interface also includes a selection area for category items and a timeline annotation area. Data annotation and dataset preparation methods supporting space science experiment analysis also include:

[0123] S831, based on the operation of selecting any object to be labeled entered in the selection area of ​​the category item, confirm the object to be labeled and the labeling time period of the object to be labeled.

[0124] S832: Within the labeled time period, the object to be labeled is marked by selecting a sub-time period using a time period selector, and the labeling result is obtained.

[0125] In some embodiments, by annotating specific targets in a video sequence and continuously tracking their position, state, and motion trajectory, real-time identification, behavior analysis, and cross-frame consistent modeling of the target can be achieved. The platform supports segmenting the video to be annotated according to a certain frame rate. Users can annotate some or all frames, and the platform will then combine them sequentially to generate the final annotation result. It supports video data annotation, including rectangle and entity labeling, and rectangle and category annotation. It supports multi-user annotation and automatic model annotation. From the start of the task, the user selects the target in the annotation interface and manually annotates the target region, assigning a unique ID to each target and ensuring cross-frame consistency. After completing the annotation, the user submits the data to the metadata database, and the system then pushes the next video segment to be annotated, repeating this process until the task ends.

[0126] To address the issue of excessively long sub-time periods, in some embodiments, the sub-time periods can be shortened by selecting a speed adjustment option.

[0127] like Figure 16 and Figure 17 As shown, in some embodiments, when the video annotation option is selected, the category annotation box includes a third annotation box. The third annotation box includes multiple annotation points connected sequentially to form a closed annotation box. Multiple category items include connection relationships, and the annotation interface also includes a timeline annotation area. The data annotation and dataset preparation method supporting space science experiment analysis further includes:

[0128] S841, confirm the object to be labeled and the labeling time period for the object to be labeled.

[0129] S842, within the annotation time period, mark the object to be annotated on any image frame using the third annotation box to obtain the first annotation result.

[0130] S843, in response to the generation of the first annotation result, generates and displays multiple category items and connection relationships.

[0131] S844, adjust the position of the category item in the first annotation result to obtain the annotation result.

[0132] For a specific object (such as a nematode), the keypoint types to be labeled for this object are set, and skeletons (such as head, body, and tail) are defined for the keypoints. The platform supports segmenting the video to be labeled according to a certain frame rate. Users can annotate some or all frames, and the platform will then combine them sequentially to generate the final annotation results. It supports video data annotation, including single-point labeling, point-line region delineation, rectangles, entity labeling, video start and end times, and rectangle and category annotation. It supports multi-user annotation and automatic model annotation. It supports dual-channel stereo rendering with an output frame rate ≥30 frames / second. The video annotation-pose tracking processing flow begins with the user clicking on the target video. The system then assigns a pose tracking annotation task. The user clicks "My Tasks" on the annotation task interface to perform manual annotation. After annotation, the system displays the annotated video, and the user selects a suitable label. The user then segments the target region using selection boxes, and the system displays keypoints on the video. The user adjusts their positions and submits, completing metadata collection. The system then displays the next video to be labeled, and this cycle repeats until the process ends.

[0133] like Figure 9 , Figure 10 and Figure 11 As shown, in some embodiments, before executing step S8, the Label library creator can add other members within the organization as users and collaborators of the Label library by clicking on their names, supporting multi-user editing and maintenance. Users and collaborators can complete labeling by performing input operations. Figure 7 As shown, in one implementation, the user clicks the "Start" button to begin, first entering the "Select Label Library" stage. Next, depending on whether the user clicks "Add Collaborator," the process splits into two directions. If a collaborator is added, the process proceeds smoothly, the shared Label library is processed, and finally the "Processing Result" stage ends. If no collaborator is added, an error will occur, requiring error handling, before the process also enters the "Processing Result" stage and ends.

[0134] In some embodiments, the creator and collaborators of the Label library can edit the Label dictionary within the library. Manual Label configuration is supported, as is batch configuration of existing Label dictionaries. Dictionary files support csV, xlsx, and txt formats. Label shortcuts can be configured for use during annotation tasks. The Label configuration process begins when the user clicks the "Start" node button. First, selections are made from the Label library, which is the foundation of the configuration. Next, the "Add Label Configuration" step begins. If any problems occur or the configuration does not meet requirements during this process, the system will trigger an "Error" node, displaying error messages. If the configuration is successful, the generated labels will proceed to the "Processing Result" node for further processing. Regardless of whether errors occur, the final process leads to the "End" node, completing the entire Label configuration process.

[0135] like Figure 9 and Figure 10 As shown, the annotation task creator can configure the annotation workflow and personnel, including the Labels used, whether to use model services for annotation, and assigning annotators and reviewers. After configuration, annotators and reviewers can view the tasks to be executed. The annotation workflow and personnel configuration process starts when the user clicks the "Start" node button. First, the specific annotation task type is selected, followed by the "Set Workflow and Personnel" section, where the process is planned and responsible personnel are configured. Then, the "Annotation Workflow Configuration" is checked. If the configuration is successful, the "Workflow Creation Successful" step is taken, and the process proceeds to the "Processing Results" step, ultimately ending the process. If problems occur during configuration, an "Error Report" will be triggered. The error must be resolved, and the configuration must be repeated until successful before continuing the subsequent processes.

[0136] By executing steps S7 and S8, structured data with annotations is obtained, which can be better used for training algorithm models. Users can efficiently complete multiple functions such as data extraction, task allocation, data annotation, annotation review, and progress statistics through the data annotation function. It supports Label library management, allowing users to annotate image and video data on the platform. It also supports data annotation task allocation as manual annotation or model service annotation, and supports exporting data annotation results for direct analysis, modeling, and training within the platform.

[0137] As demonstrated by the above embodiments, optimizing data annotation and analysis methods for on-orbit experiments in manned spaceflight missions is crucial. Integrating artificial intelligence and manual annotation improves the processing efficiency of on-orbit experimental data. Supporting personalized and differentiated analysis needs meets the refined data processing requirements of different experimental directions. Establishing standardized datasets enhances data consistency, reusability, and model training effectiveness. Supporting on-orbit intelligent analysis reduces data transmission volume and improves data utilization efficiency. Supporting visualization analysis provides efficient experimental data exploration tools, helping scientists uncover the potential value of experimental data. Furthermore, it enables efficient and accurate annotation of space science experimental images and video data, improving the data's structure and providing high-quality training data for subsequent scientific research and artificial intelligence analysis.

[0138] In summary, data annotation plays a crucial role in the intelligent analysis process of space science experiments, bridging the raw data and model training. It has an irreplaceable and vital function in improving the accuracy, reliability, and practicality of the model.

[0139] In some embodiments, after performing step S8, the data annotation and dataset preparation method supporting space science experiment analysis further includes:

[0140] S9, in response to the confirmation operation entered on the annotation interface, redirects to the annotation review interface.

[0141] The annotation review interface is generated based on pre-set review conditions associated with the task type item.

[0142] In this way, annotation task reviewers can review annotation results one by one or process them in batches. When errors or inaccuracies are found, they can be directly modified, and the modified results are immediately updated and injected into the annotation process to ensure the accuracy of the data annotation. The manual review process for model annotation begins when the user clicks on the target annotation task. Then, the user clicks "My Tasks - Manual Review" on the annotation task overview page, at which point the system displays the already annotated results. Subsequently, the reviewer can operate on the annotation results, clicking "Approve," "Skip," or modifying the annotation results. Finally, the relevant information is stored in the metadata database, and the process ends.

[0143] In some embodiments, an automatic review mechanism can be pre-configured for the model service annotation process. When the annotation score is greater than or equal to the set "approval score," the platform will automatically mark the result as approved, eliminating the need for further manual confirmation and improving annotation efficiency. The process begins with the reviewer clicking on the target annotation task. Next, when configuring the workflow, the user selects annotation with a model service. Then, the user selects automatic review, enters the approval score, and clicks confirm. The system then runs the model service to begin annotation. When the model's output score reaches the set threshold, the annotation result is automatically approved and saved to the metadata database. Finally, the user views the model annotation results, and the process ends.

[0144] like Figure 11 As shown, in some embodiments, the creator, annotator, and reviewer of the annotation task can view the status of the annotation task, including annotation progress, data sources to be annotated, personnel information, Label information, workflow overview, and my tasks (annotation or review). First, the user clicks the "Start" button to begin, and then selects a specific annotation task. Next, they proceed to the "View Annotation Task Status" step to confirm the task status. If the task status is correctly identified (i.e., the task completion rate is not zero), detailed information is obtained and the results are processed, ultimately ending the process. If the task status is incorrectly identified (i.e., the task completion rate is zero), an error reporting operation must be performed first, followed by result processing, again ultimately ending the process.

[0145] like Figure 18 As shown, in some embodiments, the label processing result is reviewed on the label review interface to obtain a verification label. In one implementation, if the task status is correctly identified (i.e., the task completion amount is not zero), detailed information is obtained and the result is processed, and the process ends. If the task status is incorrectly identified (i.e., the task completion amount is zero), an error reporting operation is performed first, and then the result is processed, and the process ends again. In one implementation,

[0146] In some embodiments, based on the validation labels, the data in each partitioned dataset is re-partitioned to obtain multiple partitioned datasets. Each partitioned dataset includes data and labels. Each partitioned dataset is then input into a training model. The training model is obtained by selecting an original model based on a model library, considering the training environment and computing resources, and then performing hyperparameter settings and model training. The trained model can be used for validation of large numbers of labels.

[0147] This provides both manual and automated review mechanisms to ensure data quality.

[0148] In some embodiments, after performing step S9, the data annotation and dataset preparation method supporting space science experiment analysis further includes:

[0149] S10, in response to the confirmation operation entered on the annotation review interface, redirects to the file storage interface.

[0150] The file storage interface includes multiple file storage format options. In some embodiments, when the data source for the annotation task is an annotation dataset or object storage, the user can click the "Generate Dataset" button on the annotation task overview page to export the currently reviewed annotation results. When the source data for the annotation task is NAS, the annotation results will be automatically written back to the NAS without manual export. The process begins by the user clicking the target annotation task, then clicking "Export Annotation Results" on the annotation task overview page. The system then prompts the user to select the annotation result format. After the user selects the format and clicks "Confirm," the system generates the dataset, which is then stored in the metadata database, ending the process.

[0151] like Figure 19 As shown, it supports exporting annotation results in multiple formats for subsequent scientific analysis and model training.

[0152] In some situations, researchers label a dataset for one experimental purpose, and then need to label the same dataset for another experimental purpose. For example, fruit fly data was first labeled for pose tracking, and after generating a pose dataset, scientists wanted to further analyze fruit fly behavior.

[0153] To enable annotation for different experimental purposes on the same dataset, some embodiments of the data annotation and dataset preparation methods supporting space science experiment analysis further include:

[0154] S11, input the annotation results as a dataset file on the second interface.

[0155] In this way, when performing behavior detection on labeled fruit fly data, the data can be selected from the already labeled pose dataset, and the pose annotation results can be imported.

[0156] In some embodiments, the data annotation and dataset preparation method for supporting space science experimental analysis provided in this application covers data annotation, task management, dataset preparation and intelligent analysis, and supports flexible configuration to adapt to different experimental needs, including: data annotation, annotation task setting and execution, annotation task management and intelligent analysis steps.

[0157] The data annotation process includes: supporting the annotation of various on-orbit scientific experimental data, including image and video data; combining manual annotation with model-assisted annotation to improve annotation efficiency and accuracy; and constructing standardized annotation datasets, including a label library and label configurations, to ensure data consistency and reusability.

[0158] The annotation task setup and execution steps include: supporting various annotation tasks, including instance segmentation, object detection, process localization, behavior detection, and pose tracking. It provides multiple annotation modes, including automatic annotation, semi-automatic annotation, and manual annotation, to adapt to different accuracy requirements and data characteristics.

[0159] The annotation task management steps include: providing both manual and automated review mechanisms to ensure data quality; supporting the export of annotation results for subsequent scientific analysis and model training; and featuring duplicate annotation and data optimization functions to improve data reliability and annotation consistency.

[0160] The intelligent analysis steps include: combining deep learning and machine learning methods to perform intelligent analysis on labeled data; optimizing tasks such as object detection, instance segmentation, and video pose tracking to improve model accuracy; and supporting intelligent data processing in orbit to reduce data transmission burden and improve data utilization.

[0161] In some embodiments, when the amount of data to be labeled is large and manual labeling is inefficient, model services can be used for labeling during workflow configuration. Pre-labeling using a model followed by manual review significantly improves labeling efficiency. Image and video data labeling is supported, as is automatic model labeling. The automatic model service labeling process starts with "Start." First, the user clicks on the target labeling task, then selects to use model services for labeling during workflow configuration and clicks "Confirm." The system then runs the model service to begin labeling, storing the labeling information in the metadata database. Finally, the user views the pre-labeled model results, and the process reaches the "End" stage.

[0162] In some situations, users find that the annotation results need to be optimized during algorithm training.

[0163] To optimize the annotation results, in some embodiments, the annotation task can be returned to for manual review of the results. Existing annotations can be viewed and adjusted. For example, starting from "Start," the user first clicks the target annotation task, then clicks "My Tasks - Manual Review" on the annotation task overview page. The system then displays the existing annotation results. The reviewer then modifies the annotation results, and the modified results are stored in the metadata database. Finally, the process ends at "End."

[0164] Data annotation is crucial in the development of deep learning models.

[0165] like Figure 20As shown, in some embodiments, the system first accesses experimental data from multiple sources (space experimental data analysis system, experimental data storage and management system) and of multiple types (structured and unstructured). Next, it enters the data processing and model building phase, developing using methods such as visual modeling, Notebook modeling, and Cloud IDE. The development results are encapsulated into code snippets, visual components, flowcharts, and templates for reuse. During this process, model design is performed, and through model training and optimization, research reports and algorithm models are output. Model evaluation and management tools, such as model performance tracking and evaluation tools and multi-model performance comparison tools, are used to evaluate the model.

[0166] In this way, the raw data can be classified, labeled, and processed to transform unstructured or raw experimental data into structured data that the model can understand. This provides accurate samples for model training, enabling the model to learn the features and patterns in the data, thereby improving the model's accuracy, generalization ability, and other performance. This is the foundation for the model to learn effectively and make correct predictions.

[0167] like Figure 21 As shown, in some embodiments, starting from data processing and model building (through visualization / notebook / scripts, etc.), algorithm resources in the model repository are called upon and personal tool libraries are used to assist development. After the model is visualized, explained, and optimized, it is exported and deployed, ultimately forming an API service or web application to realize inference functionality. In this way, the data annotation function, as the underlying support, provides structured training data for model development, directly affecting the model training quality and inference accuracy. The model development function realizes knowledge reuse through the accumulation mechanism of the model repository and tool library, ultimately ensuring that the deployed model has reliable inference capabilities. These three together constitute a value loop from data to application.

[0168] Based on the same concept, this application also provides a data annotation and dataset preparation system for supporting space science experiment analysis. The method corresponding to this system can be the same as the method described in the foregoing embodiments, and its problem-solving principle is similar. The data annotation and dataset preparation system for supporting space science experiment analysis provided in this application includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the methods and / or technical solutions of the various embodiments of this application.

[0169] Another embodiment of this application provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application described above.

[0170] Specifically, this embodiment may employ any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. Computer-readable storage media may include, for example, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0171] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0172] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0173] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Small Language, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0174] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0175] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0176] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or page components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0177] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0178] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0179] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0181] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for data annotation and dataset preparation to support space science experimental analysis, characterized in that, include: In response to the completion of input for the tag name item, the user is redirected to the tagging interface associated with the task type item. The annotation interface includes a display area for the dataset file and a selection area for category items; the display area is used to display at least one image frame of the dataset file; the dataset file includes an image annotation dataset and a video annotation dataset of spatial animal behavior; the category items are used to characterize the category name of the feature to be labeled; Based on the operation of selecting any of the objects to be labeled entered in the selection area of ​​the category item, a category labeling box for the object to be labeled is generated based on the task type item and preset experimental conditions; Based on the category labeling box, the object to be labeled is labeled on the image frame to obtain the labeling result; The annotation result is at least one of the pose estimation result, tracking result, and behavior annotation result; When the image annotation dataset is selected, the category annotation box includes a first annotation box; the first annotation box includes multiple annotation points connected in sequence to form a closed annotation box; the annotation interface also includes a selection area for the category item, and the object to be annotated is confirmed by selecting any of the objects to be annotated according to the operation input in the selection area of ​​the category item; the object to be annotated is marked on the image frame through the first annotation box to obtain the annotation result; When the image annotation dataset is selected, the annotation interface also includes a selection area for the category items and a time axis annotation area; the annotation results are associated with the time of the time axis annotation area.

2. The data annotation and dataset preparation method for supporting space science experiment analysis according to claim 1, characterized in that, The data annotation and dataset preparation methods supporting space science experiment analysis also include: When the image annotation dataset is selected, the category annotation box includes a second annotation box; the second annotation box includes a rectangular selection box; The annotation interface also includes a selection area for the category item and a time axis annotation area; based on the operation of selecting any object to be annotated entered in the selection area of ​​the category item, the object to be annotated and the annotation time period of the object to be annotated are confirmed; the object to be annotated is marked on any image frame within the annotation time period by the second annotation box to obtain the annotation result.

3. The data annotation and dataset preparation method for supporting space science experiment analysis according to claim 1, characterized in that, The data annotation and dataset preparation methods supporting space science experiment analysis also include: When the image annotation dataset is selected, the category annotation box includes a time period selector; The annotation interface also includes a selection area for the category items and a timeline annotation area; based on the operation of selecting any object to be annotated entered in the selection area of ​​the category items, the object to be annotated and the annotation time period of the object to be annotated are confirmed; within the annotation time period, the object to be annotated is marked by selecting sub-time periods through the time period selector, and the annotation result is obtained.

4. The data annotation and dataset preparation method for supporting space science experiment analysis according to claim 1, characterized in that, The data annotation and dataset preparation methods supporting space science experiment analysis also include: When the image annotation dataset is selected, the category annotation box includes a third annotation box; the third annotation box includes multiple annotation points connected in sequence to form a closed annotation box; The multiple category items include connection relationships, and the annotation interface also includes a time axis annotation area; confirm the object to be annotated and the annotation time period of the object to be annotated; within the annotation time period, mark the object to be annotated on any image frame using the third annotation box to obtain a first annotation result; in response to the generation of the first annotation result, generate and display the multiple category items and the connection relationships; adjust the position of the category items on the first annotation result to obtain the annotation result.

5. The data annotation and dataset preparation method for supporting space science experimental analysis according to any one of claims 1-4, characterized in that, The data annotation and dataset preparation methods supporting space science experiment analysis also include: In response to the input of a dataset creation operation, the user selects a dataset type on the first interface to jump to the second interface; the dataset types include image-annotated datasets and video-annotated datasets of spatial animal behavior. In response to the data upload operation entered on the second interface, upload the dataset file according to the dataset type; In response to the "Create Labeling Task" operation entered on the second interface, the system navigates to a third interface to perform the operations of entering a task name and selecting a task type; the task type includes image annotation and video annotation options; the task type corresponds to the dataset type. In response to a confirmation operation entered on the third interface, the user is redirected to a fourth interface associated with the task type item to edit the tag name item in the tag library. In response to a confirmation input on the fourth interface, the user is redirected to the fifth interface; the fifth interface is associated with the tag name item in the edit tag library, including the tag name item; the tag name item includes multiple category items; In response to the completion of the input for the label name item, the user is redirected to the annotation interface associated with the task type item; the annotation interface includes a display area for the dataset file and a selection area for the category item; the display area is used to display at least one image frame of the dataset file; Based on the operation of selecting any of the objects to be labeled entered in the selection area of ​​the category item, a category labeling box for the object to be labeled is generated based on the task type item and preset experimental conditions; Based on the category labeling box, the object to be labeled is labeled on the image frame to obtain the labeling result; The tagging result is associated with the tag name item in the tag library.

6. The data annotation and dataset preparation method for supporting space science experiment analysis according to claim 5, characterized in that, The data annotation and dataset preparation methods supporting space science experiment analysis also include: In response to a confirmation input on the annotation interface, the user is redirected to the annotation review interface; the annotation review interface is generated based on pre-set review conditions associated with the task type item. In response to the confirmation operation entered on the annotation review interface, the user is redirected to the file storage interface, which includes multiple file storage format options.

7. The data annotation and dataset preparation method for supporting space science experiment analysis according to claim 5, characterized in that, The data annotation and dataset preparation methods supporting space science experiment analysis also include: The annotation results are entered as a dataset file on the second interface.

8. The data annotation and dataset preparation method for supporting space science experiment analysis according to claim 1 or 2, characterized in that, The data annotation and dataset preparation methods supporting space science experiment analysis also include: When the task type item is an image annotation option, the annotation type item includes instance segmentation option, object detection option, and image classification option; when the task type item is a video annotation option, the annotation type item includes process localization option, behavior detection option, and pose tracking option.

9. A data annotation and dataset preparation system supporting space science experimental analysis, characterized in that, include: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 8.

10. A computer-readable medium having computer program instructions stored thereon, characterized in that, The computer program instructions can be executed by a processor to implement the method as described in any one of claims 1 to 8.

Images