Department of Information Engineering and Research Center E. Piaggio, Faculty of Engineering, University of Pisa, Via G Caruso 16, I-56122 Pisa, Italy.
Physiol Meas. 2013 Apr;34(4):449-64. doi: 10.1088/0967-3334/34/4/449. Epub 2013 Mar 22.
This work aims at showing improved performances of an emotion recognition system embedding information gathered from cardiorespiratory (CR) coupling. Here, we propose a novel methodology able to robustly identify up to 25 regions of a two-dimensional space model, namely the well-known circumplex model of affect (CMA). The novelty of embedding CR coupling information in an autonomic nervous system-based feature space better reveals the sympathetic activations upon emotional stimuli. A CR synchrogram analysis was used to quantify such a coupling in terms of number of heartbeats per respiratory period. Physiological data were gathered from 35 healthy subjects emotionally elicited by means of affective pictures of the international affective picture system database. In this study, we finely detected five levels of arousal and five levels of valence as well as the neutral state, whose combinations were used for identifying 25 different affective states in the CMA plane. We show that the inclusion of the bivariate CR measures in a previously developed system based only on monovariate measures of heart rate variability, respiration dynamics and electrodermal response dramatically increases the recognition accuracy of a quadratic discriminant classifier, obtaining more than 90% of correct classification per class. Finally, we propose a comprehensive description of the CR coupling during sympathetic elicitation adapting an existing theoretical nonlinear model with external driving. The theoretical idea behind this model is that the CR system is comprised of weakly coupled self-sustained oscillators that, when exposed to an external perturbation (i.e. sympathetic activity), becomes synchronized and less sensible to input variations. Given the demonstrated role of the CR coupling, this model can constitute a general tool which is easily embedded in other model-based emotion recognition systems.
本研究旨在展示嵌入心呼吸（CR）耦合信息的情绪识别系统的改进性能。在此，我们提出了一种新的方法，能够稳健地识别二维空间模型（即著名的情感环模型）中的多达 25 个区域。在基于自主神经系统的特征空间中嵌入 CR 耦合信息的新颖性能够更好地揭示情绪刺激下的交感神经激活。使用 CR 同步图分析来量化这种耦合，以每呼吸周期的心跳数表示。生理数据是通过国际情感图片系统数据库中的情感图片对 35 名健康受试者进行情感诱发而收集的。在这项研究中，我们精细地检测了五个唤醒水平和五个效价水平以及中性状态，这些组合用于在 CMA 平面上识别 25 种不同的情感状态。我们表明，将双变量 CR 测量值包含在仅基于心率变异性、呼吸动力学和皮肤电反应的单变量测量值的先前开发的系统中，可极大地提高二次判别分类器的识别精度，每个类别的正确分类率超过 90%。最后，我们提出了一种适应于已有理论非线性模型的外部驱动的交感神经激发下的 CR 耦合的综合描述。该模型背后的理论思想是，CR 系统由弱耦合的自维持振荡器组成，当暴露于外部干扰（即交感神经活动）时，它会同步并且对输入变化的敏感性降低。鉴于 CR 耦合的作用，该模型可以构成一个通用工具，很容易嵌入其他基于模型的情绪识别系统中。
