Time-Series Data Augmentation via Segment Manipulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data augmentation techniques for time-series data primarily focus on class identification across entire video inputs, lacking a specific method for augmenting time-series data that includes frames belonging to different classes.
Innovation Solution
A data augmentation apparatus and method that acquire source time-series data and generate augmented data by performing deletion, length change, or position change processes on target time-series data within the source data, ensuring the augmented data includes frames from different classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is performed on entire video data for class identification, then the number of training data pieces increases, but the technique cannot handle time-series data with multiple classes effectively
Solution Approach 1:
The patent segments the video data into multiple time-series segments, each corresponding to different classes. Instead of treating the entire video as a single unit, the method divides it into discrete temporal segments that can be independently augmented. This segmentation enables the data augmentation technique to handle multi-class time-series data effectively while still increasing the overall quantity of training data.
Solution Approach 2:
The patent applies different data augmentation operations to different local segments of the time-series data based on their class characteristics. By treating each time-series segment with appropriate augmentation techniques suited to its specific class, the method maintains class-specific properties while expanding the training dataset. This local quality approach resolves the contradiction by making the technique adaptable to multi-class data while preserving the benefits of data augmentation.
2Measurement precision
If more training data is prepared manually, then the accuracy of classification models improves, but the time and effort required for data preparation increases
Solution Approach 1:
The patent uses data augmentation to create synthetic copies of existing time-series data through various transformations. Instead of manually preparing new training data, the system generates additional training samples by applying deletion, length change, and position change operations to existing data. This copying approach increases the quantity and diversity of training data needed for high classification accuracy while avoiding the time-consuming manual data preparation process.
3Adaptability or versatility
If data augmentation operations are applied to time-series data, then the variety of training data increases, but the complexity of the augmentation process increases
Solution Approach 1:
The patent implements dynamic data augmentation by allowing different augmentation operations to be applied selectively to different time-series segments based on their characteristics. The system dynamically chooses which operations (deletion, length change, position change) to apply to which segments, creating diverse training data without requiring a fixed complex augmentation pipeline. This dynamic approach increases data variety while managing process complexity through flexible, adaptive operations.
Data Source
AI summary
A data augmentation apparatus acquires source time-series data. The data augmentation apparatus generates augmented time-series data by performing a modification process on one or more pieces of target time-series data included in the source time-series data. The target time-series data includes a plurality of frames belonging to the same class. The modification process includes a deletion process of deleting the target time-series data, a length change process of changing a length of the target time-series data, or a position change process of changing a position of the target time-series data on a time axis.