Faster Sample Collection Method for Ultrasound Gesture Recognition Modeling Using Direct Sequence Spread Spectrum System

Collecting training samples for ultrasound-based gesture recognition systems incurs a high cost. This paper proposes a method enabling simultaneous sample collection using two devices to address this issue. This method takes advantage of Direct Sequence Spread Spectrum (DSSS) technology, in which interference from similar devices can be suppressed through correlation processing. This approach generates two distinct genuine gesture samples from a single action, improving sample collection efficiency and diversity. A prototype system was constructed, and experiments were conducted to validate the proposed method. The results demonstrate that the model trained on samples collected using this method, where 400 actions have been done, reached an accuracy of 96.7%. In contrast, if the traditional method, the single system acquisition method, is used, the same number of actions can only get half of the samples, and the accuracy of the trained model is 0.9% lower.