Detection of Insect Infestations in Paddy Field using an Electronic Nose

An electronic nose was used to predict the number of infesting insects and the storage time of paddy rice. The multivariate statistical techniques such as principal component analysis (PCA), linear discriminant analysis (LDA), principle component regression (PCR), partial least square (PLS) and back-propagation neural networks (BPNN) were used to evaluate the electronic nose data, respectively. The PCA and LDA results showed that the electronic nose can distinguish paddy rice with different storage time (ST) and different number of infesting insects (NI). After employing PCR, PLS and BPNN, respectively to predict the infestation index (NI&ST), the three methods all had good prediction performances. The correlation coefficient between the NI real and the three predicted values was 0.955, 0.864, and 0.996 for the PCR, the PLS and the BPNN, respectively. The correlation coefficient between the ST real and the three predicted values was 0.992, 0.852 and 0.998. BPNN model had the highest prediction accuracy. The results implied that it is possible to predict the characteristics of insect infestation in stored paddy rice from signal of electronic nose. (C) 2011 Friends Science Publishers