IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. XX, NO. YY, MARCH 2016 1 A Reliable and Reconfigurable Signal Processing Framework for Estimation of Metabolic Equivalent of Task in Wearable Sensors Parastoo Alinia, Student Member, IEEE, Ramyar Saeedi, Student Member, IEEE, Ramin Fallahzadeh , Student Member, IEEE, Ali Rokni , Student Member, IEEE, Hassan Ghasemzadeh, Member, IEEE Abstract—Wearable motion sensors are widely used to estimate Metabolic Equivalent of Task (MET) values associated with physical activities. However, one major obstacle in widespread adoption of current wearables is that any changes in config- uration of the network requires new data collection and re- training of the underlying signal processing algorithms. For any wearable-based MET estimation framework to be considered a viable platform, it needs to be reconfigurable, reliable, and power-efficient. In this paper, aim to address the issues of sensor misplacement, power efficiency, and new sensor addition and propose a reliable and reconfigurable MET estimation framework. We introduce a power-aware sensor localization approach that allows users to wear the sensors on different body locations without need for adhering to a specific installation protocol. Furthermore, we propose a novel transductive transfer learning approach, which gives end-users the ability to add new sensors to the network without need for collecting new training data. This is accomplished by transferring the knowledge of already trained sensors to the untrained sensors in real-time. Our experiments demonstrate that our sensor localization algorithm achieves an accuracy of 90.8% in detecting location of the wearable sensors. The integrated model of sensor localization and MET calculation achieves an R 2 of 0.8 in estimating MET values using a regression-based model. Furthermore, our transfer learning algorithm improves the R 2 value of MET estimation up to 60%. Index Terms—Metabolic equivalent of task, Physical activities, Transfer learning, Node localization, Sensor misplacement, Mo- tion sensors. I. I NTRODUCTION M ETABOLIC equivalent of task (MET) is an approxi- mation of energy expenditure and an indicator of the intensity of physical activities. This measurement is commonly used to asses performance of physical activity interventions associated with many chronic illnesses such as coronary heart disease, type-2 diabetes, and cancer [1]. Healthy lifestyle changes such as diet control and exercise, which maintain a balance between dietary intake and calories burned, are key approaches in reducing complications due to these diseases [2]. This requires real-time tracking of physical activities that individuals at high risk of chronic diseases perform daily [3]. There are several approaches to calculate food intake and level of physical activity, including traditional self-reported questionnaires, indirect calorie meters, doubly labeled water techniques, and electrocardiographs [3], [4]. In recent years, Authors are with the School of Electrical Engineering and Computer Science, Washington State University, Pullman, WA 99164–2752 USA e-mail: (visit http://www.epsl.eecs.edu/). however, accelerometers, gyroscopes, pressure sensors, and heart rate monitors have been used for physical activity detection and energy expenditure calculation [5]–[7] due to their small size, portability, low-power consumption, and low cost [3], [4]. Accelerometers have been widely used to estimate energy expenditure and MET of physical activities [3]–[5], [8]. Al- though, the current approaches for estimation of MET values using wearable sensors have proven be to accurate [5], [8], they do not take two important issues into consideration for deployment in real-world settings. First, users would naturally tend to add new wearable sensors to the network as new sensors such as smart watches, ankle bracelets and necklaces, become available. Current research requires collection of new labeled training data for the purpose of algorithm development when a new sensor is added to the system. Second, users prefer to carry their mobile devices on various body locations, resulting in a displacement of the sensor [9]. Therefore, claimed accuracy of current MET estimation systems (MES) is dependent on adhering to the deployment protocols; for ex- ample, users must wear sensors on predefined body locations. One issue with real time monitoring of the location of the wearable sensors is excessive power usage and frequent need to charge the multiple sensor nodes. To address this issue, power optimization should be considered in different design levels [10]. An important aspect of the low-power system level design and optimization in wearables is to develop efficient signal processing and data reduction algorithms that reduce computation load of the processing units, allowing low- cost processors to be embedded with the wearable device. These requirements are limiting practical use and potentially imposing discomfort for end-users. In order to make wearables of the futures more reliable and reconfigurable, the underlying MET estimation model needs to be updated upon changes in the wearable network [11]. In this paper, we propose a framework to address reliability and reconfiguration challenges of wearable sensor networks. First, an approach is proposed to compensate unreliability due to change of on-body sensor location while taking into account the computation complexity of the devised sensor localization algorithms. Second, a novel transfer learning algorithm is developed to adopt the knowledge of existing nodes in a new configuration of the network. The result is a reliable, power-efficient, and reconfigurable MET calculation system that allows users to change the location of the sensors or