SmellNet is a novel dataset that contains data on real-world smells from 50 different food and plant-based substances, gathered through portable gas sensors. The dataset comprises over 180,000 time steps and 50 hours of recorded data, which enables the training of deep learning models to classify substances by smell. It employs advanced techniques like First-Order Temporal Differences for preprocessing and utilizes high-resolution GC-MS data for cross-modal learning. The system has been validated for practical applications, including detecting allergens in food products.
Our research introduces SmellNet, the first large-scale dataset of real-world smells collected using portable gas sensors across 50 food and plant-based substances. With over 180,000 time steps and 50 hours of data, we train deep learning models that classify substances based on smell alone.
Our approach includes novel temporal preprocessing, such as First-Order Temporal Differences, and cross-modal learning using high-resolution GC-MS data. We demonstrate real-time substance classification.
The system was validated with a practical case study that focuses on detecting allergens, such as gluten or peanuts, in food.
Smell is hard, genetic, and complex, but researchers at MIT Media Labs' Multisensory Intelligence Group are training AI to perceive smell like humans.
Collection
[
|
...
]