A temperature-decoupled impedance-based mass sensing using CBAM-CNN and adaptive weighted average preprocessing with high accuracy

Microcantilevers are widely adopted in various fields for micro-mass sensing due to their simple structure and high sensitivity. The conventional microcantilever-based measurement methods required ultra-precision temperature control. However, such a requirement could hardly be satisfied and the temperature shifting would deteriorate the accuracy of mass sensing. In this research, a new temperature decoupling method for mass sensing is proposed by taking advantage of the relative relationship of the peaks in the impedance signals. Besides, a CBAM-CNN network with an adaptive weighted average preprocessing method is developed for the evaluation of the relative relationship of the peaks. This method can enhance the features of the impedance signal and highlight critical information with limited samples. Experiments showed that robust mass sensing with accuracy up to 99.70% can be achieved over a wide temperature range from 25 to 55 ℃. In addition, the performance of the proposed network was compared to SVM, DNN, and CNN. This method offers a new mechanism for eliminating the influences of temperatures for precision cantilever-based mass sensing.