Research On Method Of Apple Gases Recognition Based On ANN

Electronic nose technique has become an international research focus and has made rapid progress during the latest years. And practical electronic nose products have been extensively applied in many fields.In this thesis, it's introduced the principle and construction of gas sensor array, our attention is paid to the constitution of gas sensor array which is sensitive to apple's gas and the characteristic of gas sensors. The emphasis on BP network and RBF network is the application in the electronic nose. On the study of the extensive nose reference in the field of electronic nose, we have designed an electronic nose equipment. Such equipment can classify the apples by odor. The main contents and methods of the thesis are:1. The electronic nose equipment is designed on the basic form of gas sensor array which is composed of four basic cells: chamber to produce gas, measure chamber, A/D conversion module and a computer seeing to sample and dispose data.2. The sample of experimental data to pinklady apples is proceed using principal component analysis(PCA), from which it can be seen the reflection degree of every sensor to apple's gas and the correlation between sensor array. It provides a rapid and exact identification measure to distinguish different mature apples using two principal components instead of primary eight sensors.3. It's programmed by Visual C language that provides a friendly interface. The Windows API function realizes the communication between the computer's serial port and gas sensor array, which can modify the communication's parameters, sampling interval and whole sampling time On-line.4. The pattern recognition models of BP network and RBF network are established. The identification of different mature apples(good, touch, bad) and two different states(good, bad) are performed. At first, sampling data is preprocessed. Improved BP algorithm and nearest neighbor-K means clustering RBF algorithm are put forward. And the test result proves that the two algorithms can identify the sampling data well.