Special Electronic Nose System For Agriculture Products Quality Detection

Electronic nose is an artificial olfactory devices for analysis, identification and detection of complex odors and volatile components. Based on the simulation of human olfactory, e-nose uses an array of gas sensors to obtain the signals of different smell, and then identifies the gas type by pattern recognition technology.During the processing of corruption, agricultural products will release gases with different features. E-nose can identify their quality by detecting the characteristics of these gases. We used a metal oxide sensors array, designed a whole set of portable electronic nose system and applied to quality detection of agricultural products.In this paper, a small sensor array was designed. With the smaller sensor array, we greatly lower the volume of the detection chamber. Therefore, the sample gas concentration decrease caused by the chamber volume can be reduced, and the portable of the electronic nose system has achieved, which increases the usefulness of electronic nose.The metal oxide sensors are sensitive to the changes of temperature and humidity of the work environment. In order to stabilize the sensors'work situation and increase the measurement precision and stability, we designed a special temperature and humidity control system.By using the semiconductor cooling pieces, we designed a PID temperature control system, which can stabilize the sensors'work environment with temperature at±1℃, and the control range at 10-60℃.Combination of the dynamic headspace and gas drier, we get the humidity control system which can successfully control the sensor working environment humidity below 10%.We use the e-nose to identify the quality of agricultural products, such as banana, loquat, peach, etc. Three different algorithms (BP, PCA+BP and LDA+BP) was used in the experiment, and the recognition accuracy was compared. Overall, LDA+BP performs best, the recognition accuracy is more than 90%; followed by the PCA+BP neural network; when directly use of BP neural network, the recognition accuracy fluctuates severely, up to 100%, while the minimum is only 76.9%.