Image-Based Volatile Organic Compound Identification Using the Cosine Similarity Method

Accurate detection and identification of volatile organic compounds (VOCs) is critical for environmental monitoring, industrial safety, and human health. Currently, microsensor-based VOC detection techniques mainly rely on electrical or frequency-based readout signals, which cannot be used for the identification of chemicals. Herein, we present a visual and intuitive method for identifying VOCs. Our technique is based on the analysis of images generated by photoionization of gas samples using a helium-based dielectric plasma detector. The helium plasma color change upon sample injection is recorded by a standard phone camera. The specific plasma images corresponding to VOC injections are extracted from the recorded videos. VOCs (n-pentane (C5), benzene, acetone) from three different classes (alkanes, polar organic compounds, aromatics) and their mixtures (C5-benzene, C5-acetone, benzene-acetone) are tested at four different injection pressures (17, 20, 24, 27 Psi). All the obtained plasma images are divided into test and training data sets, and the cosine similarity method is then used to identify the test data set images. The image feature vectors for cosine similarity calculation are obtained by direct image segmentation. Moreover, the impact on identification accuracy, including different regions of the image is also studied, revealing a sensitivity to image regions. Results show that VOC identification accuracy is close to 100% when using the cosine similarity method based on direct image segmentation. This work demonstrates a new approach for VOC identification through plasma image analysis and has the potential to facilitate the development of low-cost VOC sensors.