Agriculture Featured

An Automatic Random Forest-OBIA Algorithm for Early Weed Mapping between and within Crop Rows Using UAV Imagery

Paper Open access published in Remote Sensing on February 12, 2018. Authors: Ana I. de Castro, Jorge Torres-Sánchez, José M. Peña, Francisco M. Jiménez-Brenes, Ovidiu Csillik and Francisca López-Granados

This study was carried out by researchers from the imaPing group ( led by Dr. Francisca López Granados of the Institute of Sustainable Agriculture of Córdoba (IAS-CSIC), the Institute of Agricultural Sciences of Madrid (ICA-CSIC) and the University of Salzburg.
This research was catalogued of special interest by those responsible for the Open Access Remote Sensing Journal. For this reason, its editorial committee has selected it to be included in the following links on LinkedIn and Twitter in order to give it greater visibility.

The detection of weed seedlings within crop rows at the initial growth stage is considered one of the main challenges in Site-Specific Weed Management (SSWM). In this context, a robust and automatic object-based image analysis algorithm (OBIA, Object-Based Image Analysis) has been developed to solve the problem of spectral similarity between weeds and the crop in the early growth phase. To this end, images were taken with an unmanned aerial vehicle (UAV) on several cotton and sunflower plots on different dates (2016 and 2017) shortly after the emergence of the crop, an effective time for post-emergence application of the herbicide.

The algorithm developed consists of several phases:
(i) photogrammetric structure-from-motion techniques in the images with a high overlap to generate the three-dimensional crop and weed model (CHM, Crop Height Model) and the orthoimage;
(ii) automatic sample selection using Machine Learning (Random Forest) techniques for algorithm training using plant height as a discriminating variable;
(iii) automatic classification and generation of the herbicide treatment map. The high precision achieved in the classification of the maps which were validated spatially and thematically (through an index elaborated in this study, Wda-Weed detection Accuracy) with truth-field data taken in the field, demonstrates the validity and robustness of the automatic algorithm developed and allows its adaptation to other wide row arable crops such as corn.

These maps are a breakthrough in weed discrimination between and within crop rows and represent significant savings in herbicide applications. Definitely, they are a useful tool to assist farmers or technicians in making decisions to improve crop management by localizing the application of the herbicide at the optimal phenological time.

Similar Posts