This news article was originally written in Spanish. It has been automatically translated for your convenience. Reasonable efforts have been made to provide an accurate translation, however, no automated translation is perfect nor is it intended to replace a human translator. The original article in Spanish can be viewed at Uso de imágenes procedentes de un vehículo aéreo no tripulado (UAV) para cartografía de objetivos agronómicos

The flow of work and the here presented algorithm constitute an important innovation in the field of the agriculture of precision, and more specifically inside the control located of bad grasses

Use of pertinent images of an aerial vehicle no tripulado (UAV) for cartography of aims agronómicos

Jorge Torres Sánchez, José Manuel Crag Barragán and Francisca López Granados (Group of Teledetection and Agriculture of Precision of the IAS-CSIC)20/09/2013

The technology based in an Aerial Vehicle No Tripulado that it describes to continuation is extraordinarily versatile what allows that it was extrapolable to different aims agronómicos, environingingmental or of another type that require a cartography of the terrain. This work was rewarded in the 9th European Conference on Precision Agriculture-2013 where gathered more than 300 researchers of the 5 continents and exposed more than 100 works R&D.

Introduction

The agriculture of precision bases in the analysis of the variability intraparcelaria of abiotic factors (floor, drainages, stress hídrico) and biotic (bad grasses, plagues, funguses, harvest) existent in the fields of crop. For said analysis is necessary to locate the variable object of interest by means of the utilisation of GPS, diverse sensors and the support of computers. The aim is to differentiate subparcelas or zones of handle that they incorporate to the equipment for the application located variable (different dose or different products). This allows to increase the profits agroeconómicos and improve the sustainability by means of the application of agroquímicos directed and adjusted to the real requests of the crop.

Inside the agriculture of precision freamers the ‘control located of bad grasses in state fenológico early', since it is the moment in which they are used to apply the treatments herbicides of post-emergency. To realizar this control is necessary previously develop methods to detect and mapear the bad grasses. One of the main sources of information used in the agriculture of precision is the analysis of images taken from aerial platforms. The aerial vehicles no tripulados (UAV, by the acronyms in English of ‘Unmanned Aerial Vehicle') have developed in the last years like an aerial platform for the acquisition of images with crowd of applications, some of them related with the agriculture of precision and the control located of bad grasses (Zhang & Kovacs, 2012). The main advantages of the employment of the UAV to reach objective agronómicos in front of other aerial platforms used up to now (and.G., satellites, aeroplanes tripulados) are the ease to realizar just flights in the moment wished, his lower cost and the lower risk that suppose in comparison with the aeroplanes tripulados.

In the context of the control located of bad grasses in early phase, one of the most notable profits is the possibility to fly to low heights, what allows the capture of images of very high space resolution in which they can detect and classify objects of few centimetres. This makes possible the use of pertinent images of UAV for the discrimination and cartography of bad grasses in early phase (2-6 true leaves) with the end to design maps of treatment located in post-early emergency. The greater difficulty of this aim roots in that in this state fenológico the plants of crop and the bad grasses are similar in appearance and espectralmente (López-Granados, 2011) and thus they require images with lower pixels of 2-5 cm. At present it is not possible to obtain a space resolution so elevated with the sensors used in satellites and aeroplanes tripulados.

It appears 1: UAV taking height on a crop of corn.

On board of the UAV have installed two different cameras separately. A conventional camera Olympus PEN And-PM1 with a resolution of 12 MP, and a camera multiespectral Tetracam Mini-MCA of 1,3 MP of resolution and that it is able to capture information in the infrared, a zone of the spectrum employed usually in the characterisation of the vigour of the vegetation.

Team of work

The UAV used in our works is a multirrotor MD4-1000 (Figure 1) with capacity to carry joined up a sensor of until 1,25 kg and an autonomy of flight of 45 minutes (Microdrones, 2012). This model, with capacity of takeoff and vertical landing, is endowed of a GPS that allows him fly of automatic way following a previously programmed route. The system of handle of the UAV includes an emisora of radiocontrol, a basic station for reception of data of telemetry and a software for design of routes, configuration of the vehicle and interpretation of the telemetry. The capture of the images is accionada automatically by the UAV according to the configuration of flight preestablecida.

Flow of work

To continuation describes the methodology developed for the generation of the maps of infestación of bad grasses.

1. Design of the flight

This phase, in which it decides the height of flight and the camera used, is of vital importance to achieve images with the suitable resolution. It is necessary to take into account the aim that pursues and the crop in that it works , since this influences on the height to which will fly and to his time this determines the space resolution of the images, the number of necessary images to cover the crop and the length of the flight, appearance very important to avoid problems with the autonomy of the vehicle (Torres-Sánchez et al., 2013). If it looks for the detection of individual plants in crops of narrow row like the wheat, is necessary to work with space resolutions of around 1 cm, what forces to fly to a height of some 30 m with the camera Olympus. Instead, if they want to mapear rodales of bad grasses in crops in wide row like corn or sunflower, can work with resolutions of some 5 cm that achieve flying to some 100 m with the camera multiespectral and to 130 m with the conventional.

Once selected the height of flight adapted for the development of our aims, designs the plan of flight with the software typical of the UAV and finally implements in the vehicle.

2. Execution of the flight

Already in the field of crop, installs the camera on board of the UAV and, after the manual takeoff, activates the route of flight programmed so that the vehicle begin automatically to visit the field of crop taking numerous images until it has it sobrevolado completely. In this moment happens to manual control to proceed to the landing. During all the time of flight the UAV sends to the basic station (Figure 2) information on different appearances eat: position, been of the batteries or power of the motors.

It appears 2: basic Station of the UAV.

3. Preprocesado Of the images

The images taken during the flight are moved from the camera to a computer. The archives generated by the conventional camera can be used such cual, however the ones of the camera multiespectral need some previous treatment to be able to be processed.

Before the analysis of the images that will generate the map of infestación, is necessary a process of mosaicado. Said process consists in combining and give coordinates to all the images taken in flight so that at the end it obtain an only image (designated ortoimagen) that show the field of crop in his whole.

4. Generation of the map of infestación

The process of generation of the map of infestación of bad grasses carries out automatically by means of an algorithm of analysis of image oriented to objects developed in our group of investigation. Each object is a group of pixels homogéneos adjacent and allows a more precise analysis that the based only in pixels (Blaschke, 2010; Crag et al., 2013). This is due to that has the advantage to incorporate in the algorithms of classification, in addition to the spectral information as it is usual in the analyses of image based in pixels, the position of the bad grasses regarding the lines of crop and other additional parameters like the form and size of the plants or parameters of texture of the present objects in the image.

After the analysis of the image obtains a map with the location of the bad grasses. This is divided forming a structure of mesh adapted to the dimensions of the machinery of treatment (p.ej., separation of filters of application of herbicide). In each one of the boxes of said mesh calculates the coverage of bad grasses, and finally export the results in formats of image and of table for his back analysis and integration in the machinery of treatment.

Example of application: crop of corn

In this section describes a case the one who has used the methodology exposed previously. In the case presented treats of a crop of corn, although also it is applying in sunflower and wheat.

The study carried out with a payment of images taken in May of 2011 in a plot of corn situated in Arganda del Rey (Madrid) infested of course of several species of bad grasses of wide leaf (Amaranthus blitoides, Xanthium strumarium) and of narrow leaf (Sorghum halepense). The state fenológico of the crop and the bad grasses was of 4-6 true leaves (Figure 3). The UAV was programmed to fly to 30 m of height on the field of crop with the camera multiespectral.

It appears 3: Field of corn studied.

The algorithm of analysis of image evaluated comparing the results of the classifications obtained (percentage of infestación of bad grasses) with a series of data truth-terrain purchased the same day in that they took the images with the UAV. The sampling realizar for the truth-terrain obtained determining the infestación of bad grasses in a series of freamers of willing aluminium by all the plot of study and that they were identified in the image of the crop generated after the mosaicado.

The resultant boxes (Figure 4) of the final segmentation of the map created classified in four categories of infestación according to the coverage of bad grasses: Without infestación; Drop (<5%); Average (5-20%); Alta (>20%). The number of categories considered and the thresholds are configurables by the user and adaptable to the requests of the machinery or system of treatment that use for the control of the bad grasses. The accuracy of the method of classification and the busy surface by each category of coverage of infestación indicate in the Table 1.

Table 1: Accuracy of classification and busy surface for each category of coverage of bad grasses considered.

It appears 4: Map of coverage of bad grasses after his segmentation.

Of the data showed can conclude that the procedure of work described and the algorithm employed are adapted for the generation of maps of infestación of bad grasses in crops in early phase, just the most adapted moment for the application of measures of control located of bad grasses.

Final considerations

The flow of work and the here presented algorithm constitute an important innovation in the field of the agriculture of precision, and more specifically inside the control located of bad grasses. The final cost that the realisation of a map of infestación could have for an agriculturalist that hired this service will depend mainly of the surface to cover and of the aim looked for (rodales or bad individual grasses). The flexibility in the programming of the flight allowed by the UAV and the fact that the flow of complete work can be carried out in 1 or 2 days, allow that the detection and mapping of bad grasses was realizar in the most convenient moment for the crop, without need of waits that cause that the treatment herbicide finish applying in a moment little suitable.

The methodology described is extrapolable to different aims agronómicos, environingingmental or of another type that require a cartography of the terrain since the possibilities of this technology for the cartography of variables agronómicas are innumerable.

Agredecimientos

This work was partially funded by the projects RHEA (ref.: NMP-CP-IP-245986-2, 7º Program Mark of the EU), TOAS (Program Marie Curie, ref.: FP7-PEOPLE-2011-CIG-293991, 7º Program Mark of the EU) and TELEPLAM (AGL2011-30442-CO2-02, MINECO-FEDER). The investigation of Jorge Torres Sánchez and José Manuel Crag Barragán was funded by the programs FPI-MINECO, and JAE-Doc-CSIC, respectively.

Bibliographic references

Blaschke T (2010). Object based image analysis for remote sensing. ISPRS Journal of Photogrammetry and Remote Sensing, 65, 2-16.
López-Granados F (2011). Weed detection for Site-specific weed management: mapping and real-time approaches. Weed Research, 51, 1–11.
Crag JM, Torres-Sánchez J, of Castro AI, Kelly M, López-Grandos F. (2013). Weed mapping in early-season maize fields using object-based analysis of unmanned aerial vehicle (UAV) images. PLoS ONE, in presses
Torres-Sánchez J, López-Granados F, of Castro AI and Crag-Barragán JM (2013). Configuration and specifications of an unmanned aerial vehicle (UAV) for early Site specific weed management. PLoS ONE, 8, and58210.
Zhang C & Kovacs J (2012). The application of small unmanned aerial systems for precision agriculture: To review. Precision Agriculture, 13, 693-712.