Precision Robotic Spot-Spraying: Reducing Herbicide Use and Enhancing Environmental Outcomes in Sugarcane

TL;DR

This study demonstrates that a ground-based, computer-vision–driven spot-spraying system (AutoWeed) can substantially reduce herbicide use in sugarcane farming while maintaining high weed knockdown efficacy. By training site-specific DL weed classifiers (MobileNetV2) on a large, field-collected RGB dataset and retrofitting spray hardware onto existing equipment, the authors achieve an average $97\%$ of blanket-spray efficacy with a $35\%$ reduction in herbicide usage across six field trials covering $\approx$25 hectares, and they report notable improvements in irrigation runoff water quality ($39\%$ lower mean concentration and $54\%$ lower mean loads). The work highlights the value of site-specific datasets, real-time embedded inference (≈$21.9$ ms per image on Jetson Nano), and carefully designed timing to ensure proper coverage, while acknowledging limitations such as occasional misclassification and the need for site-specific retraining. Overall, AutoWeed offers a practical, scalable pathway to reduce chemical inputs and enhance environmental outcomes in sugarcane production, with potential applicability to other crops and regions through targeted dataset expansion and semi-supervised training approaches.

Abstract

Precise robotic weed control plays an essential role in precision agriculture. It can help significantly reduce the environmental impact of herbicides while reducing weed management costs for farmers. In this paper, we demonstrate that a custom-designed robotic spot spraying tool based on computer vision and deep learning can significantly reduce herbicide usage on sugarcane farms. We present results from field trials that compare robotic spot spraying against industry-standard broadcast spraying, by measuring the weed control efficacy, the reduction in herbicide usage, and the water quality improvements in irrigation runoff. The average results across 25 hectares of field trials show that spot spraying on sugarcane farms is 97\% as effective as broadcast spraying and reduces herbicide usage by 35\%, proportionally to the weed density. For specific trial strips with lower weed pressure, spot spraying reduced herbicide usage by up to 65\%. Water quality measurements of irrigation-induced runoff, three to six days after spraying, showed reductions in the mean concentration and mean load of herbicides of 39\% and 54\%, respectively, compared to broadcast spraying. These promising results reveal the capability of spot spraying technology to reduce herbicide usage on sugarcane farms without impacting weed control and potentially providing sustained water quality benefits.

Figures (15)

• Figure 1: (a) An illustration of the AutoWeed system where the AutoWeed unit detects the weed (1) and activates the sprayer (2). (b) A photo of the AutoWeed system mounted to a broadcast spraying boom.
• Figure 2: The AutoWeed weed detection and spraying boom mount design for a 1-metre boom section, including (a) the AutoWeed detection unit including a machine vision camera, an NVIDIA Jetson GPU, and a custom solenoid sprayer board that can individually control up to four solenoids per camera; (b) a protective sun shade, (c) a 1" wet boom, (d) a 40 $\times$ 40 mm steel hollow section frame, and (e) TeeJet solenoids and nozzle body adaptors.
• Figure 3: Illustration of the AutoWeed system retrofitted on different sprayers: (a) a 13-row high-rise John Deere R4720 self-propelled sprayer for Irvin leg spraying, with (b) detection units mounted in front of the Irvin legs between cane row centres; (c) a 4-row sprayer fitted for broadcast spraying, with (d) detection units mounted to the spray boom frame.
• Figure 4: The classification approach for sugarcane where the camera is centred on the interrow and (a) images are split into four tiles for annotation and (b) the field of view is cropped to have two tiles of equal size in the centre of the frame for inference. For each tile, if a weed is detected the corresponding spray nozzle is activated.
• Figure 5: Sample annotated images from the training datasets for (a) trials 1 & 2, (b) trial 3, (c) trials 4 & 5, and (d) trial 6. Red borders indicate a tile contains a target weed and green borders do not. The blue border for (c) indicates a second target weed.