Study On The Wound Infection Detection Algorithm For Electronic Nose Based On Quantum-behaved Particle Swarm

Medical electronic nose is the electronic nose system for medical diagnosisaccording to the features of patients’ breath gas or wound headspace gases to implementthe diagnosis of disease and wound infection. Traditional methods of bacteria detectionin pathology consume large amounts of energy and time (usually takes more than48hours), and the detection needs professional person to operate. In recent years,electronic nose as a new attractive method received a widespread attention andapplication, and it has the advantages of short response time, fast detection speed, lowcost, convenient and easy in combination with artificial intelligence. Male rats are usedas experimental subjects and animal bacteria wound infection detection serves asapplication background, then further research is realized according to the establishedexperiment system of medical electronic nose for animal wound detection. The mainwork includes:Feature extraction is an important part of the electronic nose data processing, thedimension of original feature can be very large, which greatly affects the subsequentclassification ability of electronic nose. In view of the traditional feature extraction, onlythe time domain or frequency domain response signal is extracted as feature, whichcannot fully describe the complete information of electronic nose data. In order toestablish the data which can contain most useful information of the wound infection,two kinds of commonly used feature extraction methods in electronic nose areintroduced to extract the key information, reduce the data dimension and improve therecognition accuracy of subsequent classifier. They are feature extraction methodsbased on time domain and transform domain respectively, where the time domainextracts the maximum value of the steady-state responses and the transfer domain useswavelet transform and Fourier transform to extract the wavelet and Fourier coefficient.Finally experiment results show that different feature extraction methods have a greatimpact on the subsequent recognition of classifier and the method of hybrid featurematrix construction in this thesis can contain more useful information and greatlyimprove the recognition ability of classifier.Optimization algorithm has been widely used in function optimization, data mining,pattern recognition and other fields in recent years, for the research of wound infectiondetection algorithm of electronic nose in the thesis, optimization algorithm will play great role in the aspects of feature selection, classifier parameters and feature subsetoptimization, so we focused on the research for one optimization algorithm with goodperformance, which is quantum-behaved particle swarm optimization (QPSO). First thealgorithm is introduced in theory and mathematical analysis, and compared with otheroptimization algorithms such as traditional particle swarm optimization algorithm andgenetic algorithm in performance, it is concluded that QPSO algorithm has greatadvantages in the aspect of searching ability, convergence speed, solution accuracy andsolving robustness and so on. Then for the characteristics of hybrid feature matrix withhigh dimension, and exiting some redundancy in the selection of initial sensor array atthe early stage, we can only proceed the optimizing configuration for the sensors laterwith the selection from the hybrid feature. So a feature selection algorithm based onquantum-behaved particle swarm is proposed, it adopts the two classification modelwhich can extract useful sensors from hybrid feature matrix effectively and accurately,also overcomes the shortcomings of traditional feature selection method that is ofcomplex operation, large amount of calculation, poor for wound detection data etc.In addition, in view of the great influence to the classifier recognition effect causedby different classifier parameters as well as the importance of each sensor, the quantumparticle swarm optimization algorithm is introduced into the synchronous optimizationof classifier parameters and feature subset. By finding the classifier parameters andimportance weighting coefficient of hybrid feature subset by QPSO algorithm, we canreach the synchronous optimization and finally achieve a good performance.According to the construction of feature matrix, feature selection and synchronousoptimization of classifier parameters and feature subset, a complete signal processingand recognition algorithm of wound infection detection system was established. Theexperimental results show that these methods can significantly improve the recognitionability of wound infection detection.