{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/deepbreath-deep-learning-of-breathing","title":"DeepBreath: Deep Learning of Breathing Patterns for Automatic Stress Recognition using Low-Cost Thermal Imaging in Unconstrained Settings","arxiv_id":"1708.06026","date":"2017-08-20","proceeding":null,"authors":["Youngjun Cho","Nadia Bianchi-Berthouze","Simon J. Julier"],"abstract":"We propose DeepBreath, a deep learning model which automatically recognises\npeople's psychological stress level (mental overload) from their breathing\npatterns. Using a low cost thermal camera, we track a person's breathing\npatterns as temperature changes around his/her nostril. The paper's technical\ncontribution is threefold. First of all, instead of creating hand-crafted\nfeatures to capture aspects of the breathing patterns, we transform the\nuni-dimensional breathing signals into two dimensional respiration variability\nspectrogram (RVS) sequences. The spectrograms easily capture the complexity of\nthe breathing dynamics. Second, a spatial pattern analysis based on a deep\nConvolutional Neural Network (CNN) is directly applied to the spectrogram\nsequences without the need of hand-crafting features. Finally, a data\naugmentation technique, inspired from solutions for over-fitting problems in\ndeep learning, is applied to allow the CNN to learn with a small-scale dataset\nfrom short-term measurements (e.g., up to a few hours). The model is trained\nand tested with data collected from people exposed to two types of cognitive\ntasks (Stroop Colour Word Test, Mental Computation test) with sessions of\ndifferent difficulty levels. Using normalised self-report as ground truth, the\nCNN reaches 84.59% accuracy in discriminating between two levels of stress and\n56.52% in discriminating between three levels. In addition, the CNN\noutperformed powerful shallow learning methods based on a single layer neural\nnetwork. Finally, the dataset of labelled thermal images will be open to the\ncommunity.","url_abs":"http://arxiv.org/abs/1708.06026v1","url_pdf":"http://arxiv.org/pdf/1708.06026v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"deepbreath-deep-learning-of-breathing","repo_url":"https://github.com/deepneuroscience/Paced-Math-Test","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"deepbreath-deep-learning-of-breathing","repo_url":"https://github.com/deepneuroscience/Mental-Overload-Detection-RVS","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"data-augmentation","task_name":"Data Augmentation"},{"task_slug":null,"task_name":"Mental Stress Detection"},{"task_slug":"physiological-computing","task_name":"Physiological Computing"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1708.06026","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}