Quantitative Detection Of Upconversion Luminescence Based On Machine Learning

Upconversion luminescent nanomaterials have received much attention in recent years due to their excellent optical properties and promising biosensing applications.Focusing on the field of rapid quantitative drug detection,this thesis provides insight into the applications of upconversion nanomaterials-based lateral flow chromatography biosensors in four areas: nanomaterials,signal processing,machine learning and data privacy protection.For the superior optical properties of upconversion luminescent nanomaterials,this study combines them with the lateral flow immunochromatography technique to design a portable and ultra-sensitive biosensor for drug detection.The main contributions of this thesis are as follows.(1)In this study,the fluorescence signal of the upconverted luminescent nanomaterial on the lateral flow immunochromatography test strip was analysed to improve the specificity and sensitivity using a computer vision image denoising model,and the sensitivity and robustness of the test results were improved using a waveform reconstruction method based on Gaussian distribution.This enables the sensor to achieve instant inspection with high accuracy and repeatability in a field environment.(2)To further improve the performance of the detection system,this study employs migration learning techniques to construct a simple,accurate and robust detection system.A migration learning-based solution strategy was designed to experiment with the accuracy of multiple training models in quantitative detection under small datasets,and compared with traditional classification algorithms.In addition,the effect of image noise on quantitative detection results is investigated in depth.(3)Considering the potential data privacy and data silo problems in the data transmission process of this quantitative detection platform,this study addresses the problems of sensitive medical data,small data volume and highly dispersed data distribution by building a drug detection system based on cross-regional collaboration and federated transfer learning,using techniques such as transfer learning,federated learning and edge computing.This allows the system to have high accuracy and robustness while protecting data privacy.