Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
, 20 (2)

High Speed Crop and Weed Identification in Lettuce Fields for Precision Weeding

Affiliations

High Speed Crop and Weed Identification in Lettuce Fields for Precision Weeding

Lydia Elstone et al. Sensors (Basel).

Abstract

Precision weeding can significantly reduce or even eliminate the use of herbicides in farming. To achieve high-precision, individual targeting of weeds, high-speed, low-cost plant identification is essential. Our system using the red, green, and near-infrared reflectance, combined with a size differentiation method, is used to identify crops and weeds in lettuce fields. Illumination is provided by LED arrays at 525, 650, and 850 nm, and images are captured in a single-shot using a modified RGB camera. A kinematic stereo method is utilised to compensate for parallax error in images and provide accurate location data of plants. The system was verified in field trials across three lettuce fields at varying growth stages from 0.5 to 10 km/h. In-field results showed weed and crop identification rates of 56% and 69%, respectively. Post-trial processing resulted in average weed and crop identifications of 81% and 88%, respectively.

Keywords: kinetic stereo imaging; multispectral imaging; plant detection; precision weeding.

Conflict of interest statement

No conflict of interest to declare.

Figures

Figure 1
Figure 1
The system in field trials. (a) The system mounted to the rear of the tractor. (b) The sensor system in-field.
Figure 2
Figure 2
Illumination system containing nine LED arrays per lamp, positioned at regular intervals.
Figure 3
Figure 3
Responses of the red (a), green (b), and blue (c) camera channels, in controlled illumination.
Figure 4
Figure 4
Image processing steps. (a) Greyscale output of the RGNIR function. (b) Image following thresholding.
Figure 5
Figure 5
Heat map of the normalised position error across a frame.
Figure 6
Figure 6
Height estimation using kinematic stereo method—preliminary test. (a) The image prior to processing containing two objects of different heights. (b) The height disparity map produced using the optical flow algorithm on the first attempt.
Figure 7
Figure 7
Height estimation using the kinematic stereo method with the addition of a Laplacian filtered image. (a) Sum of the original and Laplacian filtered images. (b) The height disparity map using the optical flow algorithm with addition of laplacian filtered image to reduce error.
Figure 8
Figure 8
Screenshot of the system during calibration process for plant identification, showing a desirable output following thresholding.
Figure 9
Figure 9
Images following processing (left) using in-field (top) and post-trial (bottom) calibration. Blue areas were designated as crops, and weeds are shown in green, with the centre of each target identified by a star. Unprocessed images are labelled by hand (right) according to the accuracy of the system identification using the key as defined in Table 3.
Figure 10
Figure 10
Proportion of weeds correctly identified for each field (F1–3) and calibration (C1, C2).
Figure 11
Figure 11
Proportion of weeds identified as multiple targets for each field (F1–3) and calibration (C1, C2).
Figure 12
Figure 12
Proportion of weeds identified for each field (F1–3) and calibration (C1, C2).
Figure 13
Figure 13
Proportion of objects identified which are debris for each field (F1–3) and calibration (C1, C2).
Figure 14
Figure 14
Percentage of crops misidentified for each field (F1-3) and calibration (C1, C2).
Figure 15
Figure 15
Proportion of objects identified which are debris versus weed count.
Figure 16
Figure 16
System weed identification capability with varying speed and weed count. (a) Percentage of correctly identified weeds versus total weed count. (b) Percentage of correctly identified weeds versus system speed in-field.

Similar articles

See all similar articles

References

    1. Pimentel D. Amounts of pesticides reaching target pests: Environmental impacts and ethics. J. Agric. Environ. Ethics. 1995;8:17–29. doi: 10.1007/BF02286399. - DOI
    1. Rüegg W.T., Quadranti M., Zoschke A. Herbicide research and development: Challenges and opportunities. Weed Res. 2007;47:271–275. doi: 10.1111/j.1365-3180.2007.00572.x. - DOI
    1. Slaughter D., Giles D., Downey D. Autonomous robotic weed control systems: A review. Comput. Electron. Agric. 2008;61:63–78. doi: 10.1016/j.compag.2007.05.008. - DOI
    1. Heap I. Herbicide Resistant Weeds. In: Pimentel D., Peshin R., editors. Integrated Pest Management: Pesticide Problems. Volume 3. Springer; Dordrecht, The Netherlands: 2014. pp. 281–301. - DOI
    1. Oerke E.C., Dehne H.W. Safeguarding production—Losses in major crops and the role of crop protection. Crop Prot. 2004;23:275–285. doi: 10.1016/j.cropro.2003.10.001. - DOI
Feedback