Weed classification by active shape models

The objective is to present a new method for classification of weed species by image processing based on active shape models. Young weed seedlings with up to two true leaves and without mutual overlapping with other leaves are to be identified. A database containing image examples of 19 of the most important weed species in Danish agricultural fields has been established. The images have been used as training data for the construction of an active shape model for each species. On the basis of these models, an algorithm for the identification of weed species in digital images has been developed. Results obtained by classifying a test set of weed seedlings have shown that the performance rate (rate of correctly identified weed seedlings) may vary from 65% to above 90%, depending on the weed species. (c) 2005 Silsoe Research Institute. All rights reserved Published by Elsevier Ltd.