Multi-Layer Time Series Images for High-Detail Vision Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time series analysis and forecasting techniques face challenges with large datasets, as they often result in low resolution representations when trying to maintain high detail, leading to increased storage and processing overhead, making them impractical for large datasets.
Innovation Solution
The method involves encoding time series data into multi-layer images, such as RGB images, where each dataset is scaled to maintain high resolution across multiple layers, allowing for high detail while keeping existing image resolution requirements intact, enabling efficient storage and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If time series data is represented in graphical images to enable visual assessment and machine learning, then the data can be processed using computer vision techniques, but the storage requirements and processing overhead increase significantly for large datasets
Solution Approach 1:
The patent divides the time series data representation into multiple layers (e.g., RGB layers), where each layer contains a portion of the data. This segmentation allows the data to be stored efficiently across multiple channels while maintaining the ability to process the complete dataset through computer vision techniques.
Solution Approach 2:
The patent transitions from representing all time series data in a single 2D image to distributing data across multiple layers, effectively adding a dimensional aspect to the data structure. This allows more data to be encoded without proportionally increasing the spatial footprint of each individual layer.
2Measurement precision
If the image resolution is increased to maintain high detail for large datasets, then the data representation accuracy improves, but the storage requirements and processing overhead increase significantly
Solution Approach 1:
By segmenting the data across multiple layers, each layer can maintain a manageable resolution while collectively representing the complete high-detail dataset. This avoids the need to increase the resolution of a single image to accommodate all data points.
Solution Approach 2:
The patent uses multiple layers to provide an additional dimension for data storage, allowing high detail to be achieved without increasing the in-plane resolution of individual images. The detail is distributed across layers rather than compressed into a single high-resolution image.
3Quantity of substance
If multiple time series datasets are encoded into a single image to reduce storage overhead, then the storage efficiency improves, but the resolution and detail of individual datasets decrease
Solution Approach 1:
The patent segments multiple time series datasets into different layers, allowing each dataset to occupy its own layer with full resolution. This maintains high detail for individual datasets while achieving storage efficiency by utilizing the multi-layer structure.
Solution Approach 2:
By using multiple layers as an additional dimension, the patent can store multiple complete-time-series datasets simultaneously without forcing them into a single 2D plane. Each layer maintains full resolution while the collective storage remains efficient.
Data Source
AI summary
Provided are embodiments for obtaining time series datasets, determining resolution and section layout for a multi-layer encoded data image, determining, based on the image resolution and the section layout, a section resolution, generating image section arrays for the datasets that includes scaling the time series datasets to the section resolution to generate scaled time series datasets, generating an image section array for the time series datasets that include a first pixel value at points in the image section array corresponding to a data point of the scaled time series dataset and a second pixel value at points in the image section array that do not correspond to a data point of the scaled time series dataset, and generating, based on the image section arrays, the multi-layer encoded data image where each image section array is populated into a respective section of a respective layer of the image.