Domain Adaptation Design for Multisensor Detection System via a Novel Weighted Framework Against Board Differences

This article develops a domain adaptation design via a novel weighted framework against board differences for multisensor detection that can diagnose lung cancer disease through patient exhalation. Due to the scarcity of medical data and the differences in cross-machine experimental data, we compensate for the differences between sensor boards through transfer learning. However, the existing domain adaptation methods often overlook differences in class prior distributions, namely, class weight deviations between domains. To address this problem, this article proposes a weighted-enhanced class difference domain adaptation (WECDDA) algorithm model, which takes into account interdomain class information and utilizes class prior probabilities in the source and target domains for weighting. The influence of features on the discrimination results is considered in the experiment. Numerical results prove the advantages of this algorithm.