POP-CNN: Predicting Odor Pleasantness with Convolutional Neural Network

Predicting odor's pleasantness with electronic nose can simplify the evaluation process of odors, and it has potential applications in the perfumes and environmental monitoring industry. Classical algorithms for predicting odor's pleasantness generally use a manual feature extractor and an independent classifier. The feature extractor is the key to developing accurate algorithms. However, its design requires expertise and experience. In order to circumvent this difficulty, we propose a model for predicting odor's pleasantness by using convolutional neural network. It was found that our model, which uses convolutional neural layers, outperforms manual feature extractor. Experiment results showed that the correlation between our model and human was over 90% on pleasantness rating. Our model also achieved an accuracy of 99.9% in distinguishing between absolutely pleasant and unpleasant odors.