Filtro de dados com redes neurais multinível para remoção deruídos na resposta de sensores baseados na ressonância Deplasmon de superfície

Abstract

Surface plasmon resonance (SPR) sensor devices are conventionally used for molecular interaction detection from features extracted from the SPR curve, which represents the resonance position and reflects the quality of the provided response. The instruments used in the SPR sensors require very specific coupling conditions for optical-electronic, mechanical, and fluidic components in order to have the correct excitation of surface plasmons obtained. Inherent noise from the sensor and in these instruments is reflected on the SPR curve and has an impact on the quality of the extracted features. Smart data processing is used to eliminate the main general noise inherent to the analysis process of the SPR sensors. A multilayer artificial neural networkbased smart filter (NNSF) was designed for this purpose and coupled to SPR sensors running in spectral/wavelength interrogation (WIM) and angular interrogation (AIM) modes. To increase the filter’s application capacity, a committee machine was implemented to reduce variance, being an alternative for a system with a high level of noise. NNSF correlated the experimental and theoretical responses, with a mean square error of 10􀀀5 and tested model correlation indeces over 90%, consequently draining most of the noise throughout the experiments. The use of NNSF led to better Width, Energy, Phase, Asymmetry, and Position (minimum wavelength and angle) resonance monitoring performance of the SPR curves. Filter noise bypassing occurs due to the following aspects: metallic film layer roughness and imprecise incident light angle and/or wavelength. The filter also eliminates amplitude and phase electromagnetic oscillations caused by defective temperature, luminous source current, and experimental environment control. For this reason, NNSF has been a less onerous alternative, avoiding the replacement or incorporation of new components