A Cascaded Vernier Sensing Structure Combined With Ensemble Learning for High-Sensitivity and Large-Range Bovine Serum Albumin Detection

In this work, we establish a bovine serum albumin (BSA) concentration measurement model based on a cascaded Vernier sensing structure combined with ensemble learning algorithm, which combines adaptive feature selection based on random forest (RF), anomaly diagnosis by kernel-ridge regression (KRR), and the least absolute shrinkage and selection operator (LASSO) regression algorithm to achieve BSA concentration detection with high sensitivity, large range, and high accuracy. The refractive index measurement sensitivity can reach 11951.24 nm/refractive index unit (RIU), with good repeatability and stability (0.22 nm). The sensitivity of BSA measurement is 2.03 nm (mg/mL). In the detection range of 0–122mg/mL, its average absolute percentage error is 2.05%, Compared with traditional wavelength demodulation algorithm, its measurement accuracy has been improved 9.29%, and more importantly, its detection range has been expanded by eight times (16 mg/mL). The method achieves the intelligent detection of BSA concentration. It provides a new intelligent demodulation approach for fiber optic sensing demodulation methods.