Improving Multisensor Detection System Performance via Subspace Projection and Embedded Hilbert Framework

In recent years, multi-sensor detection systems have gained widespread attention in the fields of environmental monitoring and food quality assessment, etc. However, the problems of sensor drift, redundant data, and high cost of the system have seriously restricted their long-term stability and detection accuracy. To cope with these challenges, this paper designs a low-cost multi-sensor detection system for the identification of different kinds of herbal medicines. On the basis of the system, considering the similarity between drift-free data and drift data, we derive a drift calibration framework via subspace projection to improve the long-term stability of the designed system. Furthermore, considering the redundancy feature in the source domain may be highly possibility to be redundant in the target domain, a sensor array optimization named embedded conditional Hilbert scheme is developed to reduce the number of sensors, which considers the redundancy between features and features, features and labels. Finally, the experimental results show that the proposed calibration framework can achieve the best performance by using the least sensors.