A Seal-Whisker-Inspired AI Smart Sensing System for Under-Water Object Detection

Seal whiskers are known to be sensitive to external signals. For instance, when animals or underwater objects move nearby, a seal’s whiskers could detect details like size, shape, locations and movement trajectory, guiding the seal to make precise reactions. Our goal is to decode the relationship between external stimuli and the signals received through the whiskers by using AI models and algorithms. Specifically, given the time-series nature of the input data, we use architectures such as LSTM, GRU, and RNN to extract information, as recurrent neural networks are effective at capturing both long- and short-term dependencies in time-series data. These models generate an embedding that encodes all information from previous time steps, with a stronger focus on recent data. This embedding is then passed to either a classification head or a regression head to produce the final predictions. For evaluation, we use classification accuracy and F1 score (to address potential class imbalance) for classification tasks. For regression, we assess performance using normalized mean residual error (MRE) for x, y, and z coordinates, as well as L1 and L2 distances. Refer to Figure 1 for a visual representation.