J. M. Peña-Barragán

Configuration and Specifications of an Unmanned Aerial Vehicle (UAV) for Early Site Specific Weed Management

A new aerial platform has risen recently for image acquisition, the Unmanned Aerial Vehicle (UAV). This article describes the technical specifications and configuration of a UAV used to capture remote images for early season site- specific weed management (ESSWM). Image spatial and spectral properties required for weed seedling discrimination were also evaluated. Two different sensors, a still visible camera and a six-band multispectral camera, and three flight altitudes (30, 60 and 100 m) were tested over a naturally infested sunflower field. The main phases of the UAV workflow were the following: 1) mission planning, 2) UAV flight and image acquisition, and 3) image pre-processing. Three different aspects were needed to plan the route: flight area, camera specifications and UAV tasks. The pre-processing phase included the correct alignment of the six bands of the multispectral imagery and the orthorectification and mosaicking of the individual images captured in each flight. The image pixel size, area covered by each image and flight timing were very sensitive to flight altitude. At a lower altitude, the UAV captured images of finer spatial resolution, although the number of images needed to cover the whole field may be a limiting factor due to the energy required for a greater flight length and computational requirements for the further mosaicking process. Spectral differences between weeds, crop and bare soil were significant in the vegetation indices studied (Excess Green Index, Normalised Green-Red Difference Index and Normalised Difference Vegetation Index), mainly at a 30 m altitude. However, greater spectral separability was obtained between vegetation and bare soil with the index NDVI. These results suggest that an agreement among spectral and spatial resolutions is needed to optimise the flight mission according to every agronomical objective as affected by the size of the smaller object to be discriminated (weed plants or weed patches).

Airborne multi-spectral imagery for mapping cruciferous weeds in cereal and legume crops

Cruciferous weeds are competitive broad-leaved species that cause losses in winter crops. In the present study, research on remote sensing was conducted on seven naturally infested fields located in Córdoba and Seville, southern Spain. Multi-spectral aerial images (four bands, including blue (B), green (G), red (R) and near-infrared bands) taken in April 2007 were used to evaluate the feasibility of mapping cruciferous patches (Diplotaxis spp. and Sinapis spp.) in winter crops (wheat, broad bean and pea) and compare the accuracy of different supervised classification methods (vegetation indices, maximum likelihood and spectral angle mapper). The best classification method was selected to develop site-specific cruciferous treatment maps. Cruciferous patches were efficiently discriminated with red/blue (R/B) and blue/green (B/G) vegetation indices and the maximum likelihood classifier. At all of the locations, the accuracy of the results obtained from the spectral angler mapper was relatively low. The cruciferous weed-classified imagery of each location were created according to the method that provided the best discrimination results and were used to obtain site-specific treatment maps for in-season post-emergence control measures or herbicide applications for subsequent years. By applying the site-specific treatment maps, herbicide savings from 71.7 to 95.4% for the no-treatment areas and from 4.3 to 12% for the low-dose herbicide were obtained.

Spectral discrimination of wild oat and canary grass in wheat fields for less herbicide application

Wheat, Triticum durum L, is a major cereal crop in Spain with over five million ha grown annually. Wild oat, Avena sterilis L., and canary grass, Phalaris spp., are distributed only in patches in wheat fields but herbicides are applied over entire fields, thus leading to over-application and unnecessary pollution. To reduce herbicide application, site-specific management techniques based on weed maps are being developed to treat only weed patches. Intensive weed scouting from the ground is time-consuming and expensive, and it relies on estimates of weeds at unsampled points. Remote sensing of weed canopies has been shown to be a more efficient alternative. The principle of weed remote sensing is that there are differences in the spectral reflectance between weeds and crops. To test this principle, we studied spectral signatures taken on the ground in the visible and near-infrared windows for discriminating wheat, wild oat and canary grass at their last phenological stages. Late-season phenological stages included initial seed maturation through advanced maturation for weeds, and initial senescence to senescent for wheat. Spectral signatures were collected on eight sampling dates from April 28 through May 26 using a handheld field spectroradiometer. A stepwise discriminant analysis was used to detect differences in reflectance and to determine the accuracy performance for a species classification as affected by their phenological stage. Four scenarios or classification sets were considered: wheat-wild oat-canary grass, with each species represented by a different group of spectra; wheat and grass weeds, combining the two weed species into one spectral group; wheat and wild oat with each represented as a single group, and finally, wheat and canary grass. Our analysis achieved 100% classification accuracy at the phenological stages of initial seed maturation, and green and advanced seed maturation and partly green for weeds and wheat, respectively, between the dates of April 28 and May 6. Furthermore, we reduced the number of hyperspectral wavelengths to thirteen out of 50. Multispectral analysis also showed that broad wavebands corresponding to those of QuickBird satellite imagery discriminated wild oat, canary grass and wheat at the same phenological stages and dates. Our findings are very useful for determining the timeframe during which future multispectral QuickBird satellite images will be obtained and the concrete wavelengths that should be used in case of using airborne hyperspectral imaging. Accurate and timely mapping of the spatial distribution of weeds is a key element in achieving site-specific herbicide applications for reducing spraying volume of herbicides and costs.

Parameter estimation of q-Gaussian Radial Basis Functions Neural Networks with a Hybrid Algorithm for binary classification

A classification problem is a decision-making task that many researchers have studied. A number of techniques have been proposed to perform binary classification. Neural networks are one of the artificial intelligence techniques that has had the most successful results when applied to this problem. Our proposal is the use of q-Gaussian Radial Basis Function Neural Networks (q-Gaussian RBFNNs). This basis function includes a supplementary degree of freedom in order to adapt the model to the distribution of data. A Hybrid Algorithm (HA) is used to search for a suitable architecture for the q-Gaussian RBFNN. The use of this type of more flexible kernel could greatly improve the discriminative power of RBFNNs. In order to test performance, the RBFNN with the q-Gaussian basis functions is compared to RBFNNs with Gaussian, Cauchy and Inverse Multiquadratic RBFs, and to other recent neural networks approaches. An experimental study is presented on 11 binary-classification datasets taken from the UCI repository. Moreover, aerial imagery taken in mid-May, mid-June and mid-July was used to evaluate the potential of the methodology proposed for discriminating Ridolfia segetum patches (one of the most dominant and harmful weeds in sunflower crops) in two naturally infested fields in southern Spain.

Discriminating cropping systems and agro-environmental measures by remote sensing

The agrarian policy of the European Union tends to support sustainable agriculture, subsidising only cropping systems that are implemented with specific agro-environmental measures. These actions require a precise follow-up of the crops and of the agricultural practices over a large surface. To that end, remote-sensing techniques are unique and cost-effective. We developed here a digital land cover classification in the Mediterranean dryland, mapping and assessing the main cropping systems and some agro-environmental measures such as cover crops in olive orchards and crop stubble for reducing soil erosion. We analysed a high spatial resolution satellite image (QuickBird) taken in early summer around Montilla, southern Spain. Images of the four broad wavebands, six band ratios and three vegetation indices were extracted from the satellite image and studied for the discrimination of nine land covers. The classified regions were determined by applying adequate boundary digital values to the selected images. Our results show that the land covers were discriminated with an overall accuracy of about 90%. Images of the normalised difference vegetation index and the ratio vegetation index discriminated between vegetation and non-vegetation zones. The visible wavebands discriminated roadside trees and herbaceous crops, and the near-infrared waveband highways and urban soil plus bare soil. The ratios blue/green and red/green were useful for distinguishing non-burnt stubble. The burnt stubble area was discriminated through the adapted burnt area index. Olive orchards were classified once the regions of vegetation, non-vegetation and non-burnt stubble were extracted. This technology will be a useful tool of agroecology control for the administration and will be a substitute for the current follow-up of cropping systems by ground visits. It can also be used on a farm level in order to help farmers and technicians to make decisions about the management of sustainable agricultural practices.

A digital elevation model to aid geostatistical mapping of weeds in sunflower crops.

A major concern in landscape management and precision agriculture is the variable-rate application of herbicides in order to reduce herbicide treatment load. These applications require a correct assessment and knowledge of the density and potential spatial variability of weed species within fields. This article addresses the issue of incorporating a digital elevation model as secondary spatial information into the mapping of main weed species present in two sunflower crops in Andalusia, Spain. Two prediction methods were used and compared for mapping weed density for precision agriculture. The primary information was obtained from an intensive grid weed density sampling and the secondary spatial information, e.g., elevation from a digital elevation model. The prediction methods were two geostatistical algorithms: ordinary kriging and kriging with an external drift, which takes into account the influence of landscape. Mean squared error was used to evaluate the performance of the map prediction quality. The best prediction method for mapping most of the weed species was kriging with an external drift, with the smallest mean squared error, indicating the highest accuracy. The results showed that kriging with an external drift with elevation reduced the prediction variance compared with ordinary kriging. Maps obtained from these kriged estimates showed that the incorporation of a digital elevation model as secondary exhaustive information can improve the accuracy of predicted weed densities within fields. These results suggest that kriging with an external drift of weed density data with elevation as a secondary exhaustive variable could be used in such situations, and in this way, the accuracy of maps for precision agriculture, which is the preliminary step in a precision agricultural management program, could be improved with little or no additional cost, since a digital elevation model could be obtained as part of other analyses.

Imagery from unmanned aerial vehicles for early site specific weed management

A new aerial platform has become available in recent years for image acquisition, the unmanned aerial vehicle (UAV). This article defines and evaluates the technical specifications of a UAV and the spatial and spectral properties of the images captured by two different sensors, a still visible band camera and a six-band multispectral camera, for early site-specific weed management applications (ESSWM). The tested UAV has shown promising ability for remote sensing for agricultural data collection. It enables the acquisition of ultra-high spatial resolution imagery (pixel size 1–5 cm). Moreover, the analysis of the spectral information acquired showed promising results for ESSWM.

Semiautomatic Detection of Artificial Terrestrial Targets for Remotely Sensed Image Georeferencing

Georeferencing of remote imagery with high spatial resolution can be achieved using the semiAUtomatic GEOreferencing (AUGEO) system which is based on artificial terrestrial targets (ATTs) and software AUGEO-2.0 for location and georeferencing. The aim of this letter is to describe the system and validate it. The ATTs consist of colored hexagonal tarps 0.25–1.0 m in diameter, placed on the ground and georeferenced. The proposed software works as an add-on of Environment for Visualizing Images and was able to locate the ATTs (isolated or disposed in associated couples) in remote images based on its spectral band specificity. To validate the AUGEO system, ATTs were placed on the ground, and remote images were taken from airplanes and unmanned aerial vehicles several times throughout the year at two locations in Southern Spain in 2008. Three variables were considered to study ATT detection accuracy: 1) ATT size; 2) ATT color; and 3) distance between ATTs when they were coupled in pairs. The averaged accuracy for the coupled 1-m red ATTs (separated by 2.5 m) was 95.9%. As the ATT size decreased, the accuracy generally decreased, regardless of the color of the ATTs. Results from coupled analysis show that ATT detection increased as the distance between the ATTs decreased. The proposed system required less time than conventional georeferencing work and allowed the georeferencing of images that do not contain recognizable ground control points. This also contributed to the site-specific management of agricultural plots through remote sensing, which required high-spatial-resolution and accurate georeferenced images.