Plant discrimination by Support Vector Machine classifier based on spectral reflectance

Abstract

Support Vector Machine (SVM) algorithms are developed for weed-crop discrimination and their accuracies are compared with a conventional data-aggregation method based on the evaluation of discrete Normalised Difference Vegetation Indices (NDVIs) at two different wavelengths. A testbed is especially built to collect the spectral reflectance properties of corn (as a crop) and silver beet (as a weed) at 635 nm, 685 nm, and 785 nm, at a speed of 7.2 km/h. Results show that the use of the Gaussian-kernel SVM method, in conjunction with either raw reflected intensities or NDVI values as inputs, provides better discrimination accuracy than that attained using the discrete NDVI-based aggregation algorithm. Experimental results carried out in laboratory conditions demonstrate that the developed Gaussian SVM algorithms can classify corn and silver beet with corn/silver-beet discrimination accuracies of 97%, whereas the maximum accuracy attained using the conventional NDVI-based method does not exceed 70%.