The Oral Odor Detection Technology Based On Sensor Selection And Fusion

Although the existing electronic nose system has a certain accuracy to thedetection of the disease, it cannot give a high accuracy rate because of itslimitations. It does not take the specificity and particularity into account, so it ispossible to improve the accuracy.This dissertation proposes a sensor selection algorithm and sensor fusionalgorithm, leveraging existing electronic nose systems that can provide valuableinformation of the sensor, removing the irrelevant sensors. With the purpose toselect those existing collection system on classification results which has beenraised and can achieve the best effect sensor combination. With the selectedsubset of these best sensor fusion, diabetes ultimately can be detected quicklyand accurately through the fusion of the results.The main task of this project includes the following three parts: oral odorsignal data preprocessing and feature extraction, sensor selection algorithm, andthe sensor fusion algorithm. The main contents include:(1)Use an existing gas collection device for sample collection to build thesample library; and then analyze the collected samples, divide the whole samplesinto two categories: people with diabetes samples and samples of healthyindividuals. At last preprocess the sample data and extract the feature of thesamples;(2)The sensor selection algorithm. A sensor selection algorithm based onlinear discriminant analysis, with the obtained weights of each sensor to elect anoptimal combination of sensors, and then analyze the experimental results andsummarize;(3)The sensor fusion algorithm. A method based decision level sensor fusion,with the classification results of logistic regression method to do the sensorfusion, and finally analyze the final results of the sensor fusion algorithm andconclude;The sensor selection and fusion of oral odor detection technology plays animportant role in promoting the respiratory diagnosis, especially in the diabetesdetection’s application and development. This issue demonstrates the feasibilityof the proposed method, and improves the results for diabetes accuracy of theoriginal equipment.