Research Article: Automatic UAV-based detection of Cynodon dactylon for site-specific vineyard management

Date Published: June 11, 2019

Publisher: Public Library of Science

Author(s): Francisco Manuel Jiménez-Brenes, Francisca López-Granados, Jorge Torres-Sánchez, José Manuel Peña, Pilar Ramírez, Isabel Luisa Castillejo-González, Ana Isabel de Castro, Anil Shrestha.


The perennial and stoloniferous weed, Cynodon dactylon (L.) Pers. (bermudagrass), is a serious problem in vineyards. The spectral similarity between bermudagrass and grapevines makes discrimination of the two species, based solely on spectral information from multi-band imaging sensor, unfeasible. However, that challenge can be overcome by use of object-based image analysis (OBIA) and ultra-high spatial resolution Unmanned Aerial Vehicle (UAV) images. This research aimed to automatically, accurately, and rapidly map bermudagrass and design maps for its management. Aerial images of two vineyards were captured using two multispectral cameras (RGB and RGNIR) attached to a UAV. First, spectral analysis was performed to select the optimum vegetation index (VI) for bermudagrass discrimination from bare soil. Then, the VI-based OBIA algorithm developed for each camera automatically mapped the grapevines, bermudagrass, and bare soil (accuracies greater than 97.7%). Finally, site-specific management maps were generated. Combining UAV imagery and a robust OBIA algorithm allowed the automatic mapping of bermudagrass. Analysis of the classified area made it possible to quantify grapevine growth and revealed expansion of bermudagrass infested areas. The generated bermudagrass maps could help farmers improve weed control through a well-programmed strategy. Therefore, the developed OBIA algorithm offers valuable geo-spatial information for designing site-specific bermudagrass management strategies leading farmers to potentially reduce herbicide use as well as optimize fuel, field operating time, and costs.

Partial Text

Vineyard yield and grape quality are variable as a consequence of intrinsic factors related to the crop and the field [1]. However, most vineyards have been managed as homogenous parcels of land due to the absence of methods that accurately analyze variability [2]. Therefore, analysis of the influence and spatial distribution of variability will allow grape growers to manage vineyards more efficiently for production and grape quality [3]. This approach is the agronomic basis of precision viticulture (PV), which assesses within-field spatial variability (e.g., soil characteristics, weed patches, fungi infection, insect pest attack, grape quality or maturation, production, balance between vegetative growth, and reproductive growth, among others) [4]. Implementation of PV, for either targeted management of inputs and/or selective harvesting at vintage, begins with monitoring vineyard performance and associated attributes, followed by interpretation and evaluation of the collected data [5]. PV is mainly focused on optimizing crop production and profitability by reducing production inputs;, therefore, its main objective is to diminish the potential damage to the environment and unnecessary costs due to over-application of inputs. Besides these economic and environmental benefits, PV practices comply with the European Policy to regulate a sustainable and rational use of agricultural products and pesticides at a farm level to lead current climatic, socio-economic, and environmental changes while ensuring feasibility and profitability [6].

Based on the high competition caused by bermudagrass infestation in the inter-row of vineyards, the possibility of mapping this weed using UAV-imagery was evaluated to facilitate site-specific weed management in the context of PV. Aerial images of several fields were captured using two sensors (RGB and RGNIR) attached to the UAV that allowed us to obtain ultra-high spatial resolution imagery and operate on demand according to the necessities of the grapevines. First, the spectral data analyses showed significant differences between the bare soil and bermudagrass, then ExGR and GNDVI were the optimum VIs selected to carry out the discrimination between both classes for the RGB- and RGNIR-orthomosaic, respectively. Second, an accurate and fully automatic VI-based OBIA algorithm was developed to map bermudagrass infesting the inter-row of vineyards, where the optimum VI for each camera was implemented. Grapevines were mapped using photogrammetric-based DSMs, thus avoiding misclassification due to the spectral similarity between the vines and bermudagrass. High values of map classification accuracy (>97.7%) were achieved with each of the cameras, proving that it is possible to map bare soil, grapevines, and bermudagrass at the vegetative stage based on RGB- and RGNIR-imagery. Thus, due to the similar results and handling and cheaper cost of the conventional camera, the use of an RGB sensor was recommended for that objective.




Leave a Reply

Your email address will not be published.