Parameter Optimization Framework for Enhancing Radar-Based Material Recognition

Radar-based material recognition requires the use of radar sensor parameters optimized for specific applications, along with machine learning (ML) models trained on data collected using these parameters. While hyperparameter optimization for ML models has been well studied, little attention has been given to optimizing radar sensor parameters, which are critical for enhancing material recognition accuracy. To achieve high performance, it is essential to select sensor parameters that effectively capture the features most relevant for distinguishing different materials. This study presents a method to optimize radar sensor parameters for improved performance in radar-based material recognition models. A key challenge is that changes in sensor parameters alter the characteristics of the collected data, necessitating reoptimization of the ML model. To address this, we introduce SimOpt, a framework that rapidly identifies the optimal combination of sensor parameters and ML hyperparameters. Using this framework, we developed SimOpt-MR, a system that simultaneously optimizes radar sensor parameters and material recognition model hyperparameters, leading to enhanced accuracy in radar-based material recognition. We validated the improvements achieved by SimOpt-MR by comparing its model performance with previous studies. In addition, we demonstrated the necessity of simultaneous optimization by comparing models generated by this approach with those independently optimized for hyperparameters and sensor parameters. The results showed that the SimOpt-MR-based system achieved superior material recognition accuracy with faster inference speed, confirming the effectiveness of the proposed method.