Browsing by Person "Heinrich, Stefan"

Now showing 1 - 1 of 1

Entwicklung eines neuartigen Selektivhacksystems zur Unkrautkontrolle in Zuckerrüben
(2022) Heinrich, Stefan; Köller, Karlheinz
The production of organic sugar beets is one of the biggest challenges in organic farming. Due to its slow youth development, it is less competitive with seed weeds like white goosefoot, knotweed, or cockspur. The control in organic farming is mainly done mechanically with the help of hoeing systems between the rows and with harrow and finger weeders within the rows. The brittleness in the youth stages doesn’t allow an aggressive deployment of the in-row methods mentioned before, so the results stay far behind the expectations. To raise the yields and to make a harvest at least possible a manual removal of the weeds is essential. The effort on manual labor is settled between 50 to 250 hours per hectare. To reduce the costs for man-ual weeding some attempts were made to use automatic weeding systems from the vegetable production. Due to the small distances between the plants and the missing head start of the crop, the losses of the used systems were higher than the generated benefits. The main rea-sons for the fail belong to the plant detection system and the tool design. State-of-the-art tech-nology uses mainly color information and geometrical measurements to separate the plants from each other. Most of the tools use simple systems which open in front of a plant and close behind it. The driving speed is limited throughout the swing-in process. None of the tools that have been used so far have no “zero-intervention”, so the risk of the plants being buried in-creases in proportion to the driving speed. In the present dissertation, new concepts for online plant recognition are first developed. The detection of sugar beet is one of the most complex tasks in mechanical weed control. Due to the small distance between the plants and the special growth habits a precise detection of the center point is challenging. Common systems out of the industrial image processing field can-not be used directly, because of the small similarities between the plants. To get successful detection results important attributes like the distinctive leaf blade must be worked out. The main task is to develop a fully practical proofed selective weeding system. Therefore an evalu-ation between an autonomous platform and a tractor-mounted system has to be made. Due to the fact of the low power supply, it was decided to develop the tractor-mounted system. For the start, a common front weeding system is used to build the prototype. It is first used to collect image data. In the following steps, the platform is required to integrate the first in-row weeding tools. Starting with the results from the previous master thesis classical image recognition methods are used to develop different approaches for the detection of the center points. The approaches mainly differ in the plant growth stages. Until the two-leaf stage, binary operators deliver good results. Starting with the four-leaf stage, the new edge detection system for local-izing the centerline of the leaf reaches an accuracy of up to 20 mm. The single plant detection system achieves recognition rates from 50 % to 98 % depending on different growth stages and lighting conditions. Thanks to the grid seeding system it was possible to reach excellent detection rates even under a high degree of weed cover and harsh conditions. By sowing the plants in a triangular arrangement, it was possible to reduce the plant losses below 1 %. To bring the detection results to the ground the currently developed rotor weeding system from the University of Bonn was used as a starting point for developing new kinematics. It achieves throughout the hydromechanical contact pressure control a better ground contour following. It is also significantly less susceptible to difficult working conditions. Due to the inclination of the rotor, the so-called “zero-intervention” could be developed. Due to the adapted cutting angle to the row, the driving speed is completely compensated. This prevents the crops from covering with soil. The tool is hydraulically driven to create a robust and inexpensive platform. It should also be possible to use the components in future serial production. The cascade controller from the rotor achieved an angular accuracy below 0,7°. The plant positions primarily detected for the in-row system could also be used for the realization of a row guidance system. For this purpose, an active implement steering was designed in addition to the automatic tractor steer-ing system. This improved the guiding behavior of the front weeding system. In order to achieve the highest possible regulating effect for every growth stage, disc colters are used in combination with optional angled shares. In later stages goosefoot shares with optional wings can be used to reach a higher pile effect. Also, the in-row system is fully integrated into the tine carrier to obtain a compact unit. By linking the information from the sowing technology and the image processing, it has been possible to develop a highly robust and fail-safe system for mechanical weed control in organic sugar beet.