SmellNet: A Large-scale Dataset for Real-world Smell Recognition
Summary: arXiv:2506.00239v5 Announce Type: replace
Abstract: The ability of AI to sense and identify various substances based on their smell alone can have profound impacts on allergen detection (e.g. smelling gluten or peanuts in a cake), monitoring the manufacturing process, and sensing hormones that indicate emotional states, stress levels, and diseases.
Despite these broad impacts, there are few standardized datasets, and therefore little progress, for training and evaluating AI systems’ ability to ‘smell’ in the real-world. In this paper, we use small gas and chemical sensors to create SmellNet, a comparatively large dataset for sensor-based machine olfaction that digitizes a diverse range of smells in the natural world.
Dataset Overview
SmellNet contains approximately 828,000 time-series data points across 50 substances, which include:
- Nuts
- Spices
- Herbs
- Fruits
- Vegetables
Additionally, the dataset comprises 43 mixtures among these substances with fixed ingredient volumetric ratios. A total of 68 hours of data was collected to ensure robustness and diversity in smell representation.
Model Development
Using SmellNet, we developed ScentFormer, a Transformer-based architecture that incorporates:
- Temporal differencing
- Sliding-window augmentation
These methodologies enhance the model’s capability to process and understand smell data effectively. For the SmellNet-Base classification tasks, ScentFormer achieves a notable 63.3% Top-1 accuracy with GC-MS supervision. In addition, for the SmellNet-Mixture distribution prediction tasks, ScentFormer attains a performance of 50.2% [email protected] on the test-seen split.
Implications and Applications
ScentFormer’s ability to generalize across various conditions and capture transient chemical dynamics illustrates the potential of temporal modeling in sensor-based olfactory AI. The implications of this research extend to several critical domains, including:
- Healthcare: Early detection of diseases through scent analysis.
- Food and Beverage: Ensuring quality control and allergen detection.
- Environmental Monitoring: Tracking pollution and hazardous substances.
- Manufacturing: Enhancing process monitoring for safety and efficiency.
- Entertainment: Creating immersive experiences through scent integration.
Conclusion
SmellNet and ScentFormer provide a foundational framework for advancing sensor-based olfactory applications. As the field of AI continues to evolve, the ability to recognize and interpret smells opens new doors for innovation across various industries. This research not only paves the way for enhanced olfactory AI systems but also addresses a critical gap in dataset availability, facilitating future studies and applications in this promising area of technology.