Detection of cowpea weevil (Callosobruchus maculatus (F.)) in soybean with hyperspectral spectrometry and a backpropagation neural network

To improve stored legumes protection, and to implement timely targeted insect pest control measures, it is essential to have better tools for accurate early detection and quantification of damage caused by the cowpea weevils. The hyperspectral spectrometry and a backpropagation neural network (BPNN) model were used to detect the cowpea weevils (Callosobruchus maculatus (F.)) in soybean. Spectrum of each sample was measured using a ASD FieldSpec® 3 Spectroradiometer fitted with a High Intensity Contact Probe. Spectra data were processed by ANOVA (Analysis of variance) and BPNN using MATLAB. After the optimum eigenvalues were determined based on the spectral curves, they are used as input vectors to create the BPNN model. Results showed that: The sensitive bands 780-900nm, 920-1000nm, and 1205-1560nm, have the potential to detect the infestation caused by cowpea weevils in soybeans; The eigenvalues, such as the crest or trough positions of the spectral curves, and the slope degree of the edges of the first derivative spectrum, were found to be useful and optimum eigenvalues for differentiating the infested soybean samples caused by cowpea weevils from non-infested soybean samples; The correct classification of the obtained BPNN model arrived 87.5% for the testing samples set and 93.5% for the total samples set. It can be concluded that cowpea weevils in soybean could be detected using hyperspectral spectrometry and a BPNN model.