Brushlet transform for hyperspectral feature extraction in automated detection of nutsedge presence in soybean

To detect any differences in the leaf reflectance of weed-free soybean versus soybean in the presence of purple nutsedge, a brushlet-based automated classification system is presented in this paper. For a given hyperspectral reflectance curve, the automated system completes the following: the brushlet transform is computed; energy feature vectors are extracted from the brushlet coefficients; Fisher´s linear discriminant analysis is used to reduce the dimensionality of the feature vector; and a minimum distance classifier is used to assign the curve to its appropriate class. Cross-validation testing is used to evaluate the classification performance. The experimental results show that with the use of the brushlet transform, the classification accuracy can be increased by as much as 38% when compared to more traditional data reduction methods, such as principal component analysis