Design of Multiphase Sensor System Detection Based on Transfer Learning and Array Optimization

This article considers a multiphase sensor system detection scheme via joint design of the domain adaptive transfer learning and the mutual information (MI) array optimization. We develop three novel frameworks for the designed sensor detection system. Specifically, a weighted feature extraction design is first derived for the multiphase detection data, Then the improved modified class difference joint domain adaptation (MCDJ-DA) framework synchronously minimizes the conditional distribution difference and maximizes the interclass discrimination through the Bayesian divergence index to achieve unsupervised domain adaptation. The weighted AUC optimization algorithm driven by MI weighted area under curve (WAUC) establishes a triple evaluation criterion of the feature correlation, complementarity, and redundancy. Finally, the reported results confirm the effectiveness of the weighted feature extraction, the transfer learning, and the information theory optimization. Compared with existing methods, the proposed approach significantly improves accuracy, stability, and feature selection. In particular, the experiments show that the optimized four-sensor array achieves a recognition accuracy rate of 86.21% in the lung cancer screening task, with an 8.2% improvement in accuracy compared with traditional methods and a 78.9% reduction in the number of sensors.