Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Digital image analysis to detect weeds in crops of sugarcane aiming at a localized treatment
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: Paper number 141899586, 2014 Montreal, Quebec Canada July 13 – July 16, 2014. (doi: 10.13031/aim.20141899586) @2014
Authors: Wesley Esdras Santiago, Antonio Ruby Barreto, Danilo Galdino Figueredo, Barbara Janet Teruel, Neucimar Jeronimo Leite, Guilherme Alonso Martins
Keywords: Precision agriculture, weed detection, machine vision, image processing, sugarcane tillage.
Abstract. Agriculture plays a major role in the Brazilian economy so that actions such as reducing costs and improving the productive system are sought continuously. In the cultivation of sugarcane, the control of weeds is among the factors that increase the cost of production, since the weeds compete with the crop for water, nutrients and sunlight, and can compromise harvesting procedures. An automated system capable of distinguishing weed culture is a potential tool in the search for a more economical and sustainable agriculture. Such a system can be associated to equipment for spraying, or even for mapping the field infestation. In this paper, we propose a new approach for automatic classification of weeds and crop digital images. The aim is to assess whether a method based on the visual bag of words can properly discriminate six weeds and sugarcane, rate the classification and the time suitable for implementing real-time operations. In our experiments, we took into account only one descriptor for describing characteristics of the concerned plants. Analyses the results show that the method achieved 94 % hit rate, using a validation set of 149 test images. The processing time for each image is about 34 seconds, indicating the need the adjustment for consider its use in real-time operations. Nevertheless, the method proposed here constitutes an interesting alternative to the construction of functional maps of infestation which, eventually, can be considered in localized agricultural treatment.
(Download PDF) (Export to EndNotes)