Dataset for "Hardware-encoded modality separation for multimodal sensing and artificial perception"

This dataset supports the article “Hardware-encoded modality separation for multimodal sensing and artificial perception.” The dataset contains raw sensor measurements and processed continuous wavelet transform (CWT) tensors used to evaluate hardware-encoded separation of strain and temperature signals and deep learning-based object classification. The main classification dataset consists of 2,400 raw measurements collected from eight object classes using three independently fabricated sensor devices over ten measurement days. The record also includes raw and processed data from 150 motorized external-test measurements and 90 processed samples from the independent contact-position variation test. The one-, two-, and three-channel CWT tensors used for the main classification and channel-ablation analyses can be generated from the raw CSV files using the provided preprocessing scripts. The generated NumPy arrays have a spatial size of 96 × 256, with one, two, or three sensor channels. The sensor values are digitized sensor outputs reported in arbitrary units. Detailed descriptions of the archives, folder structure, filenames, preprocessing procedure, and data usage are provided in DATASET_README.md. A versioned snapshot of the preprocessing, training, evaluation, benchmarking, and real-time inference code is included in this Zenodo record. The code is also available at: https://github.com/FabioCannavaro/Piezo_Decoupling The dataset is distributed under the Creative Commons Attribution 4.0 International license (CC BY 4.0).