A Novel Detection Framework Via Drift Compensation for Inter Board Differences

This paper designs a multi-sensor odor detection system for lung cancer detection, which can be used to collect exhaled gas and non-invasive predict lung cancer diseases. In response to the widespread drift problem in multi-sensor odor detection systems in the medical context, we have added constraints that can represent inter class differences in the improved differential empirical distance and proposed a new formulation. Inspired by the principles of machine learning, we consider the source domain data as non-drift data, while the target domain data as cross-domain data. The derived enhanced category discrepancy domain adaption (ECDDA) framework considers the consistency between statistical and geometric distributions. Thereby improving the compensation performance of sensor drift by combining domain adaptation to reduce category distribution differences and Bayesian probability to extract category information, establish an unsupervised cross-domain category difference maximization model for drift compensation, overcoming inter-board differences on different machines, and increasing the sample size to a certain extent when used for medical data consolidation. The results show the effectiveness of the proposed design.