Weed map generation from UAV image mosaics based on crop row detection

Publikation: Konferencebidrag uden forlag/tidsskriftKonferenceabstrakt til konferenceForskningpeer review


To control weed in a field effectively with a minimum of herbicides, knowledge about the weed patches is required. Based on images acquired by Unmanned Aerial Vehicles (UAVs), a vegetation map of the entire field can be generated. Manual analysis, which is often required, to detect weed patches in this vegetation map is a major obstacle to site specific weed management based on the vegetation map. A semiautomatic method for detecting weed patches based on crop row detection is described in this study. Vegetation outside the crop rows is considered as weeds.

A color image (RGB) mosaic of the entire field is used as input for the method. Issues related to perspective distortion are reduced by using an orthomosaic, which is a high resolution image of the entire field, built from hundreds of images taken by a UAV. A vegetation map is generated from the orthomosaic by calculating the excess green color index and then comparing with a threshold value.

Location of crop rows in the vegetation map is determined by analysing smaller regions of the vegetation map one at a time, the smaller regions are extracted using a sliding window that ensures a significant overlap between adjacent regions. The crop rows in each region are located by first calculating the Hough Transform, then determining the dominant orientation of crop rows in the region and finally searching for crop rows with that orientation.

A crop row is detected in the orthomosaic, if the crop row is detected in at least two different analysis regions. The detected crop rows are marked in the orthomosaic. The weed map is the vegetation map, from which all vegetation near detected crop rows have been removed. The suggested analysis of the orthomosaic is considerably faster (10 min) than the generation of the orthomosaic itself (hours) and provides an accurate weed map.
Publikationsdato28. jun. 2016
StatusUdgivet - 28. jun. 2016
BegivenhedInternational Conference on Agricultural Engineering: Automation, Environment and Food Safety - Aarhus University, Aarhus, Danmark
Varighed: 26. jun. 201629. jun. 2016


KonferenceInternational Conference on Agricultural Engineering
LokationAarhus University


Dyk ned i forskningsemnerne om 'Weed map generation from UAV image mosaics based on crop row detection'. Sammen danner de et unikt fingeraftryk.