Relevance of phase information for object classification in automotive ultrasonic sensing using convolutional neural networks

Jona Eisele,André Gerlach,Marcus Maeder,Steffen Marburg

DOI: https://doi.org/10.1121/10.0024753

2024-02-01

The Journal of the Acoustical Society of America

Abstract:Automotive ultrasonic sensors come into play for close-range surround sensing in parking and maneuvering situations. In addition to ultrasonic ranging, classifying obstacles based on ultrasonic echoes to improve environmental perception for advanced driver-assistance systems is an ongoing research topic. Related studies consider only magnitude-based features for classification. However, the phase of an echo signal contains relevant information for target discrimination. This study discusses and evaluates the relevance of the target phase in echo signals for object classification in automotive ultrasonic sensing based on lab and field measurements. Several phase-aware features in the time domain and time-frequency features based on the continuous wavelet transform are proposed and processed using a convolutional neural network. Indeed, phase features are found to contain relevant information, producing only 4% less classification accuracy than magnitude features when the phase is appropriately processed. The investigation reveals high redundancy when magnitude and phase features are jointly fed into the neural network, especially when dealing with time-frequency features. However, incorporating the target phase information facilitates the identification quality in high clutter environments, increasing the model's robustness against signals with low signal-to-noise ratios. Ultimately, the presented work takes one further step toward enhanced object discrimination in advanced driver-assistance systems.

acoustics,audiology & speech-language pathology

What problem does this paper attempt to address?

This paper attempts to address the issue of evaluating the relevance of target phase information in object classification using Convolutional Neural Networks (CNN) in automotive ultrasonic sensing. Specifically, existing research typically considers only amplitude features for object classification, while this paper explores whether phase information can provide additional useful information to improve classification performance. Using both laboratory and field measurement data, the authors propose and process several time-domain and time-frequency domain phase-aware features and use CNN for classification tasks.

The main objectives include:
1. **Evaluating phase information**: Determine whether the target phase of the echo signal contains relevant information that aids in object classification.
2. **Comparing different features**: Compare the performance of amplitude features with phase features, as well as the effectiveness of their combined use.
3. **Enhancing robustness**: Explore whether incorporating phase information can enhance the robustness of the classification model in high-clutter environments, especially under low signal-to-noise ratio conditions.
4. **Verifying redundancy**: Analyze whether there is a high degree of redundancy when both amplitude and phase features are input into the neural network, particularly when dealing with time-frequency features.

In summary, this study aims to improve object recognition capabilities in automotive driver assistance systems by introducing phase information.