| With the development of the computer's technology, people have earned great production in the research of computer-based vision,hearing and feeling, some of them even exceed the people's sense organ in application. Nevertheless, the research on olfaction develops relatively more slow than others, it mainly because of the difficulty in researching it and the complexity in realizing it.In recent years, the artificial olfactory technology based on an array of gas sensors and a pattern recognition technology has been a pop focus researched by people all over the world. It shows a wide outlook in application of detecting odors in many fields, such as inspecting environment,medical diagnoses,medical and pharmaceutical industry,food industry,chemical industry and army affairs etc. Presently, some people have attempted to use it in detecting agricultural products. Overseas, it has been used to detect the quality of foodstuff in storage, classify corn and judge the degree of fruits' maturity. It appears an excellent situation while the artificial olfactory technology is used in detecting agricultural products.This article starts with the software and algorithm of the artificial olfactory system (AOS), has researched the algorithm and principle of olfaction recognition, developed an AOS and used it in detecting wines and oils to quicken the march of using it in detecting agricultural products. The article approaches the following points:The author Researched and analyzed the main theories and methods used in the field of artificial olfaction inside and outside. AOS 's sensor array composed of several gas sensors with different sensitivities, it makes use of the cross sensitivity in gas sensors to reduce the demand of the sensor's selectivity, consequently, it enlarges the system's range of application. When collecting data, the curves of the response of sensors changed uninterruptedly, after a period of undulation, they get to steady-stage response. The curves contain a large number of data. It's unsuitable to input all of the data to classifier for recognition. Therefore, it's significant to extract the more effective message of data from the array's instantaneous response for the system's ability of classification and recognition. Thus, people bring forward many algorithms for feature extraction,reduction of data's dimension and normalization. For example, using the array normalization algorithm:normalize the vector quantities of the array's response, make the data being in the area (0,1), this can reduce the effect of the concentration of sample on system's recognition. Of the studies on this all over the world, the universal method for feature extraction and reduction of data's dimension include: extracting several feature from the response curves, principal-component analysis (PCA) and the Karhunen -Loeve transform etc.In aspects of pattern recognition, the artificial neural network is widely used in AOS all over the world because of it' highness of nonlinear-in-the-parameter, excellent fault-tolerance and robustness. Furthermore, the BP algorithm of the 3-layers feed forward neural networks is used most widely of all. Recently, with the fuzzy theory,genetic algorithm and wavelet theory soaking into ANN each other , they have been partly used in AOS. On account of the olfaction recognition is a fuzzy conception, the fuzzy neural network would have a wide range of application in the development of AOS in the future.In this article, six SnO2 gas sensors of MOS are used to compose the sensor array, the cross sensitivity in gas sensors is adequately used which turns the inferior position to predominance and reduces the demand of the sensors' selectivity, moreover, it extends the application's range of the AOS.For the first time, like the method used in the process of image, this article attempts to divide the response data of the sensor array into some pieces, choose 6 continuous instantaneous response of sensor array to be a "piece", then use the wavelet theory compress...
|
PDF Full Text Request