Validation of commercial precision spraying technology

McCarthy, William (2016) Validation of commercial precision spraying technology. [USQ Project]

[img]
Preview
Text ( Main Project)
McCarthy_W_McCarthy.pdf

Download (29Mb) | Preview

Abstract

The agricultural industry is a leading contributor to the Australian economy with an approximated revenue of $40 Billion in the 09/10 financial year. Weeds are portrayed as costing the agricultural industry up to $4 Billion per annum in herbicide use and lost production. Research into proposed methods to reduce this cost whilst maximising agricultural produce with sustainable practices will benefit the Australian economy.

Blanket spraying is a procedure utilised by all broadacre farmers to manage and control on-farm weeds. During fallow, paddocks typically have only a 20% coverage of weeds, therefore, a blanket spray could unnecessarily spray as much as 80% of the field. Not only is this an expensive waste of herbicide, it also has negative impacts on the environment and possible accumulation in food products through residue buildup and runoff.

Precision spraying platforms are available in agriculture which detect weeds in real time and activate nozzle solenoids to deliver chemicals to the weed. Precision spraying, therefore, targets only weeds and results in herbicide saving and a decrease in herbicide resistance. In theory this is impressive, however, adoption of this technology has been poor throughout the agricultural industry due to the large capital expenses required to purchase the systems and fear of change with no guarantee of the kill rate. There is no quantitative data that provides proof on the accuracy of any weed detection system commercially available. Therefore, this project aims to develop hardware and associated software to form the basis of a standardised test procedure for evaluating weed detection systems.

Initially an assessment of two commercial weed detection systems were undertaken, the WeedSeeker and WEEDit platforms, to determine interfacing methods to recognise when the systems detect weeds. This assessment led to the development of two separate Weed-Check modules which could interface to the di↵erent platforms and capture a signal when a weed was detected. When this signal is recognised by the WeedCheck module, a camera is triggered which captures an image of the weed, whilst also geotagging the image with GPS position information.

Field trials were designed to test the accuracy of the weed detection platforms. These trials were performed to gather information on three attributes. The first being the accuracy of the weed detection platform. This included determining the hit and miss rates of the technologies through taking images of the weeds detected and post analysing them. The second interest was the spray footprint of the weed detection platforms, which is important to ensure chemical is delivered to the weed. This test clearly showed the WEEDit had a better spray footprint of approximately 500mm whereas the WeedSeeker platform only had 150mm. The third stage of the trial involved using the GPS positions to create weed maps. This, however, proved to be inaccurate as a GPS error of up to 4.3 metres was observed. The images of the weeds were then analysed to match identical weeds captured within and between trials which formed the basis of the weed detection accuracy assessment.

The outcomes of the trials proved a feasible method was developed for determining the accuracy of weed detection platforms. Through matching weeds within the image frame, it allowed an assessment of hit and miss rates of the two technologies. Unfortunately, due to unforeseen GPS error the weed map comparison was deemed unreliable. Further future development of the computer vision algorithm to automatically sort and match weeds within frames would be an excellent method of validation. The final outcome of the project found the WEEDit was better at detecting weeds under di↵erent conditions, whereas the WeedSeeker platform regularly missed smaller weeds, however, this comparison was only undertaken with a 0.16ha trial block due to GPS position errors. Further testing would need to be conducted to verify these findings.

These findings and the software developed in this project have industry benefits as the result sets a standard of comparison of new developments in weed detection platforms against current commercial systems. This is of high significance to industry as there is no advantage in developing agricultural robots to change farming systems if the attachments and sensors available have not yet been validated. Robotics in agriculture can only be as effective as the sensors and implements available. To encourage adoption, farmers need to clearly see the benefits of the change and know they are moving forward not backward in weed control, sustainability and profitability.


Statistics for USQ ePrint 31439
Statistics for this ePrint Item
Item Type: USQ Project
Item Status: Live Archive
Additional Information: Bachelor of Engineering (Honours) Major Agricultural Engineering project
Faculty / Department / School: Current - Faculty of Health, Engineering and Sciences - School of Civil Engineering and Surveying
Supervisors: McCarthy, Cheryl
Date Deposited: 20 Jul 2017 23:47
Last Modified: 20 Jul 2017 23:47
Uncontrolled Keywords: commercial precision spraying technology; agricultural industry; sustainable practices; nozzle solenoids; GPS error
Fields of Research : 09 Engineering > 0907 Environmental Engineering > 090703 Environmental Technologies
URI: http://eprints.usq.edu.au/id/eprint/31439

Actions (login required)

View Item Archive Repository Staff Only