3D Power Device Dataset Augmentation Using Physical Scene Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The lack of a suitable dataset for robustness testing of power vision identification models in the professional field leads to inadequate model verification and non-objective reflection of model robustness, due to the unavailability of flexible and standardized data acquisition methods that align with physical system characteristics.
Innovation Solution
A power vision dataset augmentation method that involves acquiring an initial three-dimensional structure point cloud, preprocessing it to determine vertices and edges, partitioning non-significant points with geometric plane primitives, extracting light and shadow information from captured images, and constructing a dataset based on three-dimensional true-color structures to align with physical system characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a physical engine is used for visual data generation, then the data complies with physical system characteristics, but the computing capability requirement and technical level requirement increase
Solution Approach 1:
The patent segments the complex physical engine into a pre-trained generation model that can be deployed on terminals. The model is trained offline on substantial datasets to learn physical system characteristics, then deployed as a compact, pre-processed solution that doesn't require heavy computing resources during actual use.
Solution Approach 2:
The patent performs preliminary training of the generation model using large amounts of data and complex computations during the offline pre-training phase. This preliminary action transfers the computational burden to the training stage, allowing the deployed model to operate efficiently on terminals with limited computing resources while still complying with physical system characteristics.
2Reliability
If a physical engine is used for visual data generation, then the data complies with physical system characteristics, but the ease of operation deteriorates due to dynamic parameter setting requirements
Solution Approach 1:
The generation model is designed to autonomously generate augmented data based on input images without requiring users to manually set physical parameters. The model self-adjusts to comply with physical system characteristics through its pre-trained understanding, eliminating the need for users to understand or configure complex physical engine parameters.
Solution Approach 2:
The patent creates a simplified copy of the physical engine's functionality within the generation model. Instead of requiring users to operate the full physical engine with its complex parameter settings, the model captures and replicates the essential physical characteristics in a user-friendly form that automatically generates physically compliant augmented data.
3Measurement precision
If public large datasets are used for model testing, then the test capability can be estimated, but the model robustness test reflection is not objective enough for professional fields
Solution Approach 1:
The patent transitions from using general public datasets to creating specialized augmented datasets tailored to specific professional fields (power systems). The generation model produces domain-specific data that reflects local characteristics and challenges of the target application area, enabling more objective and relevant robustness testing for professional field models.
Solution Approach 2:
The patent uses parameter changes in the generation model to create diverse augmented data variations that specifically target robustness testing. By adjusting the model's generation parameters, the system can produce data with different characteristics, conditions, and scenarios relevant to the professional field, providing comprehensive and objective robustness evaluation.
Data Source
AI summary
A power vision dataset augmentation method includes acquiring an initial three-dimensional structure point cloud of a power device, preprocessing the initial three-dimensional structure point cloud to obtain a three-dimensional structure point cloud, and determining point clouds of power device vertices and edges based on the three-dimensional structure point cloud; partitioning point clouds of the non-significant vertices and edges in the three-dimensional structure point cloud and replacing partitions with geometric plane primitives to obtain a three-dimensional structure represented by the geometric plane primitives; determining light and shadow information of the power device based on a captured image; matching the three-dimensional structure with the captured image; determining a three-dimensional true-color structure of the power device based on the light and shadow information of the power device after the three-dimensional structure matches the captured image; and constructing a dataset based on the three-dimensional true-color structure and the captured image.