Journal of Food, Agriculture and Environment

Vol 8, Issue 1,2010
Online ISSN: 1459-0263
Print ISSN: 1459-0255

Support vector data description for weed/corn image recognition


Xien Liu, Mengjun Li, Yuanlun Sun, Xiaoyan Deng *

Recieved Date: 2009-10-02, Accepted Date: 2009-11-05


This paper focused on support vector data description (SVDD) technique for the weed/corn recognition based on imbalanced weed/corn image samples. The original images were taken from corn fields under natural lighting and various field backgrounds, and converted into gray level images by a vegetation index: Excess Green minus Excess Red (ExG-ExR). The 2-level wavelet transform was employed to decompose grey image into the approximate component and the detail components. The features of weed/corn images, including the morphological features and the wavelet-based energy features, were extracted. The morphological features were obtained from binary weed/corn images. The various combinations of these features were used as input vector to construct support vector data description models. The experiment results showed that the SVDD model could provide better recognition effect than support vector machine (SVM) and Fisher linear discrimination algorithm (FLDA) based upon the same imbalanced training data set. The two “best” SVDD models were obtained, which achieved a weed/corn recognition rate of 95.59%.


Weed/crop recognition, wavelet transform, feature extraction, support vector data description

Journal: Journal of Food, Agriculture and Environment
Year: 2010
Volume: 8
Issue: 1
Category: Agriculture
Pages: 214-219

Full text for Subscribers

Note to users

The requested document is freely available only to subscribers/registered users with an online subscription to the Journal of Food, Agriculture & Environment. If you have set up a personal subscription to this title please enter your user name and password. All abstracts are available for free.

Article purchasing

If you like to purchase this specific document such as article, review or this journal issue, contact us. Specify the title of the article or review, issue, number, volume and date of the publication. Software and compilation, Science & Technology, all rights reserved. Your use of this website details or service is governed by terms of use. Authors are invited to check from time to time news or information.

Purchase this Article:   20 Purchase PDF Order Reprints for 15

Share this article :