Synthetic benchmarks for machine olfaction: Classification, segmentation and sensor damage

Abstract

The design of the signal and data processing algorithms requires a validation stage and some data relevant for a validation procedure. While the practice to share public data sets and make use of them is a recent and still on-going activity in the community, the synthetic benchmarks presented here are an option for the researches, who need data for testing and comparing the algorithms under development. The collection of synthetic benchmark data sets were generated for classification, segmentation and sensor damage scenarios, each defined at 5 difficulty levels. The published data are related to the data simulation tool, which was used to create a virtual array of 1020 sensors with a default set of parameters