The work of this workpackage is divided in two parts: close range precision spraying and canopy optimized spraying. Main partners involved in this work package are University of Ljubljana and University of Milano.
Application of agrochemicals is at present, the primary method used to protect plants from pests (diseases, fungi, insects, weeds). To do this, active ingredients formulations are diluted in water and distributed to vegetation in form of spray droplets. To protect plants from diseases and pests, agrochemicals have to be sprayed to ensure a uniform coverage of susceptible targets at the appropriate time in the season. In orchard crops and grapevine susceptible targets (fruits, bunches, new sprouts, younger leaves etc.) can be located anywhere in the vegetation, consequently current spraying techniques aims to cover all parts of plants, front and behind, top and bottom, as well as within the canopy. Consequently, a high volume air-carrier flow assists the transport of high volume of pesticides to deposit the spray droplets to internal and hidden parts of plants inside canopy.
A significant amount of pesticide can be lost as drift during such type of application or it can be just unnecessary sprayed on canopy regions where no protection is actually needed. All this has negative effects on quality of execution of agro-technical operations, in term of production costs, environment impact, operator and population exposure and possible residues on products.
In workpackage precision spraying the topic of intelligent control of spraying in orchard and vineyard is addressed by two main activities, close range precision spraying and canopy optimized spraying.
With close range precision spraying we want to selectively and precisely spray disease patches or limited portions of the canopy which may represent susceptible targets of specific diseases or pests. These selectively treated areas mainly include initial discrete foci, which are the predominant source of inoculum for new infections, rapidly causing the spread of pathogens to the field.
In canopy optimised spraying the sprayer the sprayer automatically and continuously adapts it's properties to changing characteristics of the vegetation while driving the sprayer along the row. Using this approach 30% pesticides may be saved under favourable conditions.
Both the canopy optimised and close range precision spraying approaches have the common objective of reducing the amount of chemicals applied, while increasing the protection efficiency through an improved targeting on the vegetation. Due to the high number of protection treatments on grapevine and orchard crops, an optimised spraying approach and improvement of targeting can strongly impact the amount of used pesticides.
Close range precision spraying
The possibility to optically detect disease symptoms relies on the modifications induced by the pathogen in the plant tissue and, in turn, in the way how light interacts with it. Beside disease-specific pigmentation, main optical effects of plant diseases are associated to spectral absorption bands of chlorophyll, where tissue degradation induced by pathogens is especially emphasised.
Within this framework, we have investigated disease the detection capability by two different concurring methods (Figure 1). On one hand we studied the application of point-wise spectrophotometry at leaf and sub-leaf scale to characterize diseased tissue by reflectance-transflectance properties in the spectral range 390-850 nm and UV-induced fluorescence properties in the range 400-750 nm. A second approach was based instead on imaging techniques, by using a multicamera system allowing the acquisition of multispectral images (red, green, NIR channels), hyperspectral line-images (400-900 nm), RGB color images, and monochromatic images (at selectable wavelength). The second sensing approach, based on imaging systems, which was oriented to final application. In particular the multispectral R-G-NIR camera was selected as the only sensor on which the disease detection capability of the system relies on.
Figure 1. CROPS diseases sensing approaches experiments in vineyard; left: static point-wise spectrophotometry at leaf and sub-leaf level; middle: multicamera imaging system for dynamic measurements at canopy scale; right: disease detection algorithm block diagram.
Disease symptoms identification in R-G-NIR images is based on the algorithmic combination of two approaches: one based on the value of two spectral indexes calculated at pixel level, and the other based on relative variations (local gradients) of grey level intensity in the normalised red channel. The disease detection algorithm block diagram is shown in Figure 1.
Close range precision sprayer integration was performed at University of Milano in spring 2013. System integration consisted in the operational interfacing of the main components (Figure 2):
- prime mover with an encoder,
- R-G-NIR multispectral camera and a RGB high resolution camera on a position adjustable holding frame ,
- diffuse illumination panels,
- personal computer for analysis of sensing data for real time disease detection,
- manipulator control PC and power case,
- CROPS manipulator with waterproof protecting case,
- pesticide control circuit with a pump,
- precision spraying end effector.
The accuracy of the complete prototype sprayer (manipulator + spraying end effector) in delivering precision spray on a identified target at a known position was assessed with specific tests. Registration of multispectral camera for disease detection indicated an average offset in the order of 1 cm between image and real world coordinates. Hence in current configuration we achieved overall precision spraying inaccuracies within 3 cm or less if the fuzziness of spraying diameter is taken into account.
Figure 2. Left: Main components of prototype sprayer for precision close-range target spraying. Right: prototype sprayer for precision close-range target spraying in operation.
Plant material preparation for laboratory experiments was carefully executed. 180 plants of Vitis vinifera L. cv Cabernet Sauvignon were propagated from wood and nursed in pots in controlled greenhouse environment. Grapevine plants were pruned in order to reach a new full development stage around the date of experiment. A subset of grown grape plants was inoculated to induce powdery mildew infection, and nursed in a separated greenhouse room with favourable conditions for disease development. Healthy and diseased plants were used to recreate indoor vineyard canopy conditions by aligning grapevine plants in pots on greenhouse nursing tables. To simulate localised disease foci within healthy vegetation, diseased plants with different levels of symptoms were positioned within the recreated canopy wall. These foci were the actual targets of close range selective spraying performed by CROPS robot (Figure 3).
Figure 3. Left: schematics of spraying procedure adopted during the greenhouse experiment and right: Visualization of a sprayed leaves with fluorescent dye when illuminated by blue light
Results are shown in Figure 4. Based on experimental results we expect to treat 80-95% of the targets (depending on the disease stage) and to limit to the 5-15% of the healthy area the portion sprayed unnecessarily. As a specific and illustrative result, in first run according to the plant pathologist survey the canopy needed 10 spot sprayings to cover all the disease foci. The robot sprayed 25 spots which actually covered all the disease foci. The pesticide used was reduced from 100% (conventional homogeneous canopy treatment) to 16% of the canopy area. According to plant pathologist survey the minimum necessary would have been only 6%. In this specific run the CROPS robot was able to automatically detect the disease foci in the canopy and selectively treat them, reducing the pesticide used by 84% compared to conventional spraying, although there was the possibility for further 10% reduction.
Figure 4. Precision spraying experimental results. Left: red dots mark positions for spot sprayings as estimated by plant pathologist survey, while targets identified by the detection system and the associated spot sprayings performed by a CROPS robotic system are shown in blue circles. Right: Number of spots sprayed and pesticide reduction achieved with CROPS robotics close range precision sprayer.
Please follow also the link below to read about the
including a video clip.
The canopy optimised sprayer is based on a conventional trailed sprayer with one axle as shown in Figure 5. Eight degrees of freedom were selected for three spraying arms of the canopy optimized sprayer. Every spraying arm carries a spraying end effector. End effector consists of two types of nozzles, pesticide nozzles and airflow nozzles. The canopy optimized sprayer is designed to position end effectors at a selected distance and perpendicular to the canopy outline. The canopy is divided into three parts, low, middle and high. The most important control strategy for positioning of spraying arms is requirement to spray perpendicular to the canopy outline from selected distance, in our case 500 mm.
Figure 5. Canopy optimised sprayer in operation.
To faithfully capture canopy properties as canopy optimized sprayer moves along the row of trees in an orchard, sensing must be performed in short distance intervals or at high speed, resulting in high data flow rate from sensors and among software modules during analysis. All software modules runs in parallel and data is among them exchanged using global variables, preventing queuing of processes due to waiting on another process to finish. With high number of computation intensive processes running, sensing, analysis and spraying arms control modules run with loop times below 2 ms, creating a system able to respond to continuously changing properties of rows of trees in an orchard.
To solve the difficulties with canopy characterization and positioning of the sprayer arms, that they do not get in contact with the canopy, suitable spraying arm positioning algorithm was developed. It was based exclusively on LIDAR measurements (Figure 6) and provides locations and directions for spraying nozzles located at the end of the spraying arms. At a given position along the row, measured points are sent from LIDAR and a contour of the canopy is determined. It is then approximated with three linear segments selected for optimum coverage of each canopy segment. In its final step the algorithm deals with smoothing of consecutive calculated positions to facilitate movements of the spraying arms by hydraulic cylinders.
Spraying experiments with canopy optimised sprayer in orchards were performed during vegetation season in 2013. Five test runs were performed: (1) canopy optimised spraying arms, nozzles open all the time, (2) canopy adapting spraying arms, nozzles open according to density (ver. 1), 3) canopy adapting spraying arms, nozzles open according to density (ver. 2), (4) comparison with a classical sprayer, and (5) canopy adapting spraying arms, nozzles open all the time (exp. no. 1 repeated). Results are shown in Figure 6.
Figure 6: Left: LIDAR-based positioning algorithm provides good spraying results due to adequate nozzles positions. Right: Spray coverage as a function of tree height.