Electromyography (EMG) Spectrograms Database for Hand Gesture Classification

This dataset presents a collection of electromyography (EMG) spectrograms derived from three hand gestures: open hand, closed hand, and pinch. The signals were acquired using a BITalino device with two surface electrodes placed on the forearm of workers in a Colombian private company, primarily women engaged in repetitive assembly tasks. The raw EMG signals were collected via OpenSignals software, stored in .h5 format, and subsequently processed in Python using the biosignalsnotebooks library. Preprocessing steps included band-pass filtering (25\u2013300 Hz), rectification, and envelope extraction. Spectrograms were generated using Hanning windows and converted into .jpg images. The final dataset consists of 408 spectrograms distributed across three gesture classes. This dataset is intended primarily for Convolutional Neural Networks (CNNs) in hand gesture classification tasks but can be extended to prosthesis control, rehabilitation, and other biomedical applications.