Browsing by Subject "Bildverarbeitung"

Now showing 1 - 3 of 3

An image analysis and classification system for automatic weed species identification in different crops for precision weed management
(2010) Weis, Martin; Gerhards, Roland
A system for the automatic weed detection in arable fields was developed in this thesis. With the resulting maps, weeds in fields can be controlled on a sub-field level, according to their abundance. The system contributes to the emerging field of Precision Farming technologies. Precision Farming technologies have been developed during the last two decades to refine the agricultural management practise. The goal of Precision Farming is to vary treatments within fields, according to the local situation. These techniques lead to an optimisation of the management practice, thereby saving resources, increasing the farmers outcome, reducing the overall management costs and the environmental impact. A successful introduction of Precision Farming involves the development of application equipment capable of varying treatments and sensor technology to measure the spatial heterogeneity of important growth factors. Such systems are able to record, store and use large amounts of data gathered by the sensors. Decision components are needed to transform the measurements into practical management decisions. Since the treatments are varied spatially, positional data, usually measured using GPS technology, has to be processed. The located measurements lead to a delineation of management zones within a field and are represented by geo-data and can be visualised in maps. The improved, detailed knowledge of the situation within the field leads to new and extended scopes of applications and allows to document the management practices more precisely. In this work, parts of Precision Farming technology were developed for site-specific weed management. Five selected publications are presented, covering the technological prerequisites and details of the developed system.
Entwicklung einer selektiv arbeitenden Reihenhackmaschine mit elektrisch angetriebenem Werkzeug zur Unkrautregulierung im ökologischen Zuckerrübenanbau
(2018) Bucher, Ulrich Paul; Köller, Karlheinz
Weed control within the planted rows continues to present a major challenge to organic sugar beet farmers. As sugar beet is very susceptible to competition from weeds during its early development, it is essential that farmers ensure that the soil is kept weed-free until row closure. Mechanical hoeing is available for the soil between the rows, but often only manual hoeing can be used for weed control within the row, with the exception of a small number of non-selective row hoeing techniques. Depending on the level of weed infestation in the field, its use can fluctuate within a range of 60 to 340 Akh/ha. A project for the development of a selectively working row hoeing machine was jointly started in 2009 at Universität Hohenheim with the Baden-Wuerttemberg Ministry for Nutrition and Rural Affairs, the Association of Baden-Wuerttemberg Sugar Beet Growers and Schmotzer, based in Bad Windsheim, to support organic sugar beet cultivation. The project was based on a single-row hoeing machine prototype from a previous project, which already had an image processing algorithm for the selection and positioning of sugar beets. There were also two tool shapes for weed control within a row of sugar beets. These hoeing tools were powered by a hydraulic motor, the speed of which was regulated by a PWM solenoid valve based on driving speed and image processing. Field tests were carried out with both at the start of the follow-on project, and a new design of hoeing tool was also tested. The quality of work performed by the three tools was then compared. The extent of the worked and unworked area within a simulated row of sugar beets was examined, among other aspects, and the method of operation and susceptibility of the tool to becoming blocked when it met larger weeds were also assessed. A mobile electric high-voltage drive was also designed with Ludwigsburg-based Jetter AG, in parallel to the field experiments, in view of problems experienced with the control dynamics of the hydraulic drive. Field experiments were then conducted with the advanced prototype under practical conditions on two test sites in Hohenheim during the following vegetation period. The row hoeing machine was compared with various methods, including manual hoeing combined with a standard hoeing machine for the soil between the rows, and also compared to the results achieved by using conventional chemical weed control. Following the initial practical experiments and the findings obtained from them, the row hoeing machine was further revised and a second row was added. The field experiments were then repeated in the same arrangement as in the previous year, again at two sites, using this two-row hoeing machine. After a two-year long test phase, it is clear that the use of a selectively working row hoeing machine reduces manual work by up to 40 %, depending on the extent of the weeds. Manual weed control continues to be indispensable for removing weeds in the immediate vicinity of the sugar beet plants, which can have an adverse impact on the yield if not removed. Furthermore, the use of the row hoeing machine leads to an unavoidable loss of plants, which, to a certain extent, neither affects the technical quality nor overall harvest of the sugar beets. In contrast, later weeding after row closure can cause significant loss of yield. In conclusion, it is worth mentioning that image processing reaches its limits under difficult conditions of up to 400 weeds per m2. The image processing algorithm also requires further improvement. By contrast, the mobile electrical high-voltage drive and the continuously moving shape of the hoeing tool both fulfill all requirements. Future developments should focus on the further improvement and optimisation of plant recognition and thus the differentiation between cultivated plants and weeds and their position in front of the hoeing machine. The working speed could be increased to more than 3.6 km/h with a faster and more precise image processing method, and damage to plants, or even loss of sugar beet plants, could then be prevented or minimised.
Teilschlagspezifische Unkrautbekämpfung durch raumbezogene Bildverarbeitung im Offline- und (Online-) Verfahren (TURBO)
(2006) Oebel, Horst; Gerhards, Roland
Georeferenced application maps (TURBO) is presented. The system was applied and analysed on agricultural fields from 2004 to 2005. The results can be summarized as followed: For online image acquisition bi-spectral cameras were developed which took homogeneous grey scale pictures with a strong contrast using a combination of two spectral channels in the near infrared and the visible spectrum. Three bi-spectral cameras were mounted in front of a prototype carrier vehicle. Using an automatic control of the exposure time, well focused pictures of weeds in cereals, maize, sugar beets, peas and oil seed rape were taken at a speed up to 10 km/h and stored together with their GPS coordinates. Under changing light conditions, bi-spectral images were free of faults. Stones, mulch and soil were not illustrated. The picture quality showed a clear improvement compared to current image analysis technologies using colour and infrared cameras in plant production. The geometric resolution of the cameras was sufficient for creating application maps. With a size of 0.014 m² per picture weed seedlings were representatively assessed. The dense grid of 3.500 sampling points per hectare allowed an efficient detection of weed distribution within agricultural fields. The procedure of shape analysis allowed precise identification of weed species in a speed of 20 images per second. The classification rate of unclassified plants using Fuzzy Logic or the principle of minimum distance was between 73 % (malt barley) and 85 % (oil seed rape). The calculation of discrimination functions to separate crops and weed classes by shape parameters allowed a better classification of unknown plants and increased the classification rate to 88.4 % (sugar beets) and 94 % (malt barley). Characteristic shape features of 45 weed species in the growth stages BBCH 10 to BBCH 14 were stored in a database and the classification of weed species in malt barley, maize and sugar beets was studied using discrimination analysis. In growth stage BBCH 10 weed species could be differentiated on average by 70 %. Crops were accurately differentiated from broadleaved weeds and grass weeds. Joining weeds species (BBCH 10) in the classes broadleaved weed species, grass weeds, Galium aparine and crop resulted in correct classification of 83 % in malt barley to 96 % in maize. With manual, GIS-based and image analysis sampling methods treatment maps for three weed species classes were created for site-specific weed control in cereals, sugar beet, maize, oil seed rape and peas on a total of 138 ha. Economic weed threshold were used as a decision rule for chemical weed control. Herbicides were only applied when the economic weed threshold was exceeded. Above the economic weed threshold the herbicide dosage was varied from 70 % to 100 % depending on the density of weed species. Herbicide application was performed with a newly developed multiple sprayer. The sprayer integrates three conventional sprayers on one machine including three separated hydraulic circuits, boom section control (3 m), dGPS for real time location and a central control unit. During application the on-board computer loading a georeferenced application maps was linked to the spray control system for precise application of up to three different herbicide mixtures. Herbicide savings using site-specific weed control depended on the cultivated crop, weed species composition and weed infestation levels. On average 47 % of herbicides for grass weeds and 35 % for broad-leaved weeds were saved. Herbicide use to control Galium aparine and Cirsium arvense was reduced by 71 %. The efficacy of site-specific weed control was documented by manual weed sampling before and after post emergent herbicide treatments. It ranged from 71.8 % to 98.8 %. Weed infestation level did not increase in the following crops. First results with yield mapping of experimental fields revealed that site-specific weed control did not cause yield reduction. On contrary, in cereals higher yields were observed at locations where no herbicides were applied. However, further studies are needed to prove this hypothesis. The economic evaluation of site-specific weed control over two years on practical farm sites showed that site-specific weed control was profitable. The average savings for herbicides were 27.61 ?/ha. This resulted in an average profit of 11.35 ?/ha using the system for site-specific weed control.