Contents lists available at ScienceDirect Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag An automatic method for weed mapping in oat felds based on UAV imagery Mateo Gašparović a, ⁎ , Mladen Zrinjski b , Đuro Barković c , Dorijan Radočaj d a University of Zagreb, Faculty of Geodesy, Chair of Photogrammetry and Remote Sensing, 10000 Zagreb, Croatia b University of Zagreb, Faculty of Geodesy, Chair of Instrumental Techniques, Kačićeva 26, 10000 Zagreb, Croatia c University of Zagreb, Faculty of Geodesy, Chair of Land Surveying, Kačićeva 26, 10000 Zagreb, Croatia d Josip Juraj Strossmayer University of Osijek, Faculty of Agrobiotechnical Sciences Osijek, Chair of Geoinformation Technology and GIS, Vladimira Preloga 1, 31000 Osijek, Croatia ARTICLE INFO Keywords: UAV Imagery classifcation Weed mapping Oats Precision agriculture ABSTRACT The accurate detection and treatment of weeds in agricultural felds is a necessary procedure for managing crop yield and avoiding herbicide pollution. With the emergence of unmanned aerial vehicles (UAV), the ability to acquire spatial data at the desired spatial and temporal resolution became available, and the resulting input data met high standards for weed management. In this paper, we tested four independent classifcation algorithms for the creation of weed maps, combining automatic and manual methods, as well as object-based and pixel-based classifcation approaches, which were used separately on two subsets. Input UAV data were collected using a low-cost RGB camera due to its afordability compared to multispectral cameras. Classifcation algorithms were based on the random forest machine learning algorithm for weed and bare soil extraction, following an un- supervised classifcation with the K-means algorithm for further estimation of weeds and bare soil presence in non-weed and non-soil areas. Of the four classifcation algorithms tested, the automatic object-based classif- cation method achieved the highest classifcation accuracy, resulting in an overall accuracy of 89.0% for subset A and 87.1% for subset B. Automatic classifcation methods were robustly developed, using at least 0.25% of the scene size as the training data set in all circumstances anticipated for the random forest classifcation algorithm to operate. The use of the algorithm resulted in weed maps consisting of zoned classes and covering areas with similar biological properties, making them ready for use as inputs in weed treatments that use agricultural machinery. 1. Introduction The development of unmanned aerial vehicles (UAVs) has pro- gressed precision agriculture, forestry, geodesy, architecture, cultural heritage protection and other related areas (Colomina and Molina, 2014). Spatial data collection in precision agriculture is now accessible to farmers by the use of UAVs, which can track changes in agricultural felds with high spatial and temporal resolution (Zhang and Kovacs, 2012) and are economical compared to the use of satellite and aerial imagery (Matese et al., 2015). An important component of a UAV survey is the possibility to observe crops from an aerial perspective, allowing detection of many vegetation properties that are difcult to observe from the ground (Candiago et al., 2015). Because of the many possibilities for processing and analysing the collected imagery (Zhang and Kovacs, 2012), the use of UAVs in precision agriculture allows for increased crop yields and lowered treatment costs of agricultural felds through more efcient use of fertilisers and herbicides. By detecting anomalies in agricultural parcels, farmers are given the opportunity to prevent negative efects by appropriate agrotechnical operations and to improve sowing quality in the subsequent seasons. Even minor anomalies in agricultural felds can be efectively identifed by UAVs (Seelan et al., 2003). Such methods for collecting spatial data are a cheaper and more time-consuming process compared to classical ter- restrial methods and thus more efcient (Gašparović et al., 2017). Weeds are responsible for a signifcant reduction in potential crop yield (up to 35%), according to Pérez-Ortiz et al. (2016). Pérez-Ortiz et al. (2016) also stated that the most common practice for weed treatment is to apply herbicide to the whole agricultural feld, resulting with inappropriate and excessive herbicide application. This leads to unnecessary pollution and negative impacts on the environment (López-Granados et al., 2016). UAV surveys present a backbone in map- based site-specifc weed management, which includes spraying weed https://doi.org/10.1016/j.compag.2020.105385 Received 25 February 2019; Received in revised form 17 March 2020; Accepted 26 March 2020 ⁎ Corresponding author. E-mail addresses: mgasparovic@geof.unizg.hr (M. Gašparović), mzrinski@geof.unizg.hr (M. Zrinjski), barkovic@geof.unizg.hr (Đ. Barković), dradocaj@fazos.hr (D. Radočaj). Computers and Electronics in Agriculture 173 (2020) 105385 0168-1699/ © 2020 Elsevier B.V. All rights reserved. T