| Bionic olfactory technology is a detection technology that actively imitates the human olfactory nervous system.It has been widely used in food safety,environmental monitoring and biomedical,etc.due to rapid detection response,broad detection range and intelligence.Detection field.However,with the expansion of bionic olfactory technology applications,the perceived information also exhibits high dimension,strong noise,non-linear characteristics,and then there is a problem of measurement data distortion,and the characteristics of difficulty extraction and reduction in recognition performance.Specifically,in the following:(1)Effective acquisition and processing of odor information under extreme poor conditions.The method and manufacturing process of the material,the method of designing,the current bionic olfactory system is measured when measuring the gas signal,and there is a problem of detection response signal crossover,affected by environmental factors(temperature and humidity etc.)and easy drift.(2)It is difficult to extract and identify complex odor map information characteristics.Since the data perceived by bionic olfactory system presents high dimensional linear characteristics,and the characteristic space after data drop is highly associated with the original data.Therefore,it is necessary to develop efficient reduction and identification methods to address data processing and feature extraction of complex data.This thesis analyzes the technical challenges facing the application expansion of bionic olfactory system,starting from the basic principles of bionic olfactory systems.In particular,the data is high dimension,nonlinearity,etc.when the complex object is measured,and the odor spectrum information is effectively acquired for non-ideal cases,and proposes a resistive drift distortion problem based on statistical characteristics.It has been proposed in a variety of effective data reduction and feature algorithms for a wide variety of effective data design,comparison,comparison,and eventually compare,and finally give a comprehensive comparison of a large number of sample data.The research work completed by this thesis is as follows:(1)In order to solve the problem of continuous measurement of data volatility due to factors such as sensitive materials and manufacturing processes in the bionic olfactory system.This paper has proposed a signal preprocessing method based on statistical score,to improve the overall regulation of measurement data samples,and reduce the influence of measurement data generated by sensor drift.The experimental results show that the measured data noise generated by the statistical score of the sensor drift distortion can be effectively reduced by the measurement data noise generated by the sensor drift distortion,providing a valid data support for subsequent data.(2)Effective data feature selection can reduce the complexity of data processing.In this paper,the data map characteristics of the measurement responses of the PEN3 electronic nose system are selected,and the two types of mathematical statistical characteristics based on statistical characteristics and measurement curves are selected,which is used to characterize the internal features of the measurement response map,thereby reducing the processes of the original data.Dimension.The experimental results show that the complexity of the reduction processing can be effectively improved by feature data selection,and it is also optimized that the desired feature spatial distribution is optimized.(3)In order to enhance the number of bionic olfactory system to measure the number of gases,the effectiveness of the measurement data is usually used to measure the gas array to measure the gas,and the sample data is presented on the space while the amount of data and the amount of data and feature information are significantly improved.High dimensions and cross-overlapping distribution.Especially when processing large samples,it is easy to initiate data dimensional disasters,resulting in the characteristics that cannot be subjected to feature extraction or extraction,and this paper proposes a PCA+FLDA linear cascade process algorithm for processing high dimensional data.DEVICE and feature extraction,through testing of four industrial virus gas data samples,experimental results show that under equivalent conditions,superiority is PCA+FLDA>FLDA>PCA.The average recognition of PCA+FLDA is 84.80%,the best recognition accuracy is 86.41%.(4)In order to solve the problem of difficulty damage to the characteristics extraction of complex data measurements measured by bionic olfactory systems.In particular,in the proposed PCA+FLDA cascading process algorithm,the matrixS?is not a real number of solutions due to the sample class,resulting in the problem of optimal mapping(non-solid space)due to the sample class.This paper has proposed a linear reduction method based on embedded nuclear functions,to solve the matrixS?optimized solution.The experimental results showed that the experimental results showed that the average recognition rate of KLDA was89.06%,the best recognition rate was 92.18%significantly higher than FLDA and PCA.For test identification rates of three types of algorithms,the priority level is KLDA>FLDA>PCA,but due to the higher complexity of the algorithm,the calculation is time consuming,especially for high-dimensional sample data operations,algorithm integration Performance indicators will be affected seriously.(5)In order to further explore the highly complex data map designation and feature reconstruction of the bionic olfactory system,this paper has proposed a SMA algorithm to detail the geometric logic,mathematical principle and implementation steps of the SMA algorithm.The data processing process visualization is visually visualized by using PEN3 E-nose measuring Aucklandia lappa incense.The SMA and PCA,PCA+FLDA,and KLDA are contrasted in detail in terms of feature mapping space,calculation time,and identification accuracy.The experimental results show that the SMA-based non-linear data processing method is prominent in designing and recognition accuracy of the design,the average recognition accuracy is 88%,and the best recognition accuracy is 94.37%,at time It also performs well in terms of consumption,more suitable for signal processing and feature extraction of PEN3 electronic nose. |
PDF Full Text Request