Evaluating Synthetic Data: A Comparison with Real | 您所在的位置:网站首页 › spaceclaim graphics is not available › Evaluating Synthetic Data: A Comparison with Real |
In our previous blog in this series, we learned what synthetic data is and its importance in training artificial intelligence/machine learning (AI/ML) systems for radar applications. To recap, synthetic data is information that is artificially generated rather than collected from real-world observations. It can be a valuable tool for training ML models with its ability to overcome data scarcity issues, simulate rare or hard-to-observe scenarios, and augment real-world data. However, there is a concern that synthetic data may not accurately represent real-world systems and phenomena. Coupled with that, radar systems are complex and require high-quality data. To remedy these concerns, Ansys has developed a simulation workflow that enables you to model complex radar scenarios in real time using an electromagnetic (EM) simulation technique based on the shooting and bouncing rays (SBR) method. This solver is based on the same SBR solver found within Ansys HFSS and has been graphics processing unit (GPU)-accelerated to perform simulation in real time. This radar sensor simulation capability is available within Ansys AVxcelerate Sensors add-ons and provides the core toolset needed to generate high-quality synthetic data for radar applications. Still, to encourage widespread adoption of synthetic data, we need to build confidence in artificially generated data, validate it, and prove we can acquire it at scale with accuracy. Let’s take a closer look at how Ansys’ solution tackles these challenges. Building Confidence: Validating Synthetic DataThere are several approaches that can be used to generate synthetic radar data for use in AI/ML applications. These approaches can be generally classified into three categories: reduced-order models (ROMs), data augmentation, and physics-based simulation. ROMs involve using a simplified model to approximate the radar returns of complex scenes, while data augmentation includes modifying existing data to create a new, unique dataset by using AI/ML algorithms or other methods. Physics-based simulation requires using the fundamental laws of physics to describe wave propagation and interaction in a complex scene. However, if the synthetic data does not accurately represent the physical behavior of real-world systems, it will not be useful for real-world applications. Ansys’ solution builds confidence in synthetic data by capturing the radar returns for a complex scene, including real material properties, animations, and antenna patterns. Therefore, Ansys’ solution delivers high accuracy, not approximations. Further, the resulting synthetic data can be created at scale, which is most useful and essential for effective ML training. |
CopyRight 2018-2019 实验室设备网 版权所有 |