Neural networks for crop/weed discrimination in different cabbage trials

Crop/weed/soil discrimination using optical reflectance from cabbage fields is feasible. An intelligent crop/weed/soil classifier algorithm capable of identifying different spectral samples into three groups: crop, weed and soil was developed and could be used by an automatic spot spraying machine to optimize herbicide application in the field. The best discriminant wavelengths for a cabbage crop showed high crop/weed/soil discrimination accuracies although reflectance differences caused by crop varieties and weed species incidence were present. The selected wavelengths obtained from discriminant analyses were fed to the neural network classifier. The classifier was able to indicate with high success rates whether the sample was a crop, a weed or soil. Neural network performance for crop/weed/soil discrimination showed better results than traditional discriminant analysis providing an unique algorithm capable of discriminating crops from weeds for six different trials non-dependant on cabbage variety (savoy, spring or autumn), growth stage and weed species population