Spectral channels selection of weed identification in the field

A Weed identification system in the field was investigated by hyperspectral channel selection and wide-band channel selection using ground imaging spectrometer data. Sensitive hyperspectral channels to biochemistry character were selected by using the Bhattacharyya distance method. The classification accuracy can achieve 96% by utilizing 344 dimension spectrometer data with 10 selected hyperspectral channels. The classification accuracy can achieve around 80% by using 3～4 stimulated wide-band channels and can achieve more than 85% by using 5～7 stimulated wide-band channels. These results provide references for sensor and channel selection in weed identification in the field.