Towards practical object detection for weed spraying in precision agriculture

Area of Science:

• Agricultural Engineering
• Computer Vision
• Environmental Science

Background:

• Weed control is crucial for crop yields, but conventional herbicides harm ecosystems.
• Precision spraying offers an eco-friendly alternative by targeting weeds specifically.
• Current weed detection methods often lack comprehensive real-world performance evaluations.

Purpose of the Study:

• To evaluate the feasibility of precision spraying through weed detection and spraying accuracy.
• To assess state-of-the-art object detection algorithms for weed identification.
• To introduce and utilize novel metrics for real-world precision spraying performance.

Main Methods:

• Utilized two distinct datasets and multiple image resolutions for algorithm testing.
• Employed several state-of-the-art object detection algorithms for weed identification.
• Developed a simplified precision spraying model to evaluate algorithm performance under varying nozzle precision.

Main Results:

• Achieved 93% weed coverage while spraying only 30% of the total area.
• Introduced and validated 'weed coverage rate' and 'area sprayed' metrics.
• Demonstrated the effectiveness of vision-based weed detection in reducing herbicide application.

Conclusions:

• Precision spraying, powered by advanced weed detection, significantly reduces herbicide usage.
• The proposed metrics effectively capture real-world performance for precision spraying systems.
• State-of-the-art vision methods enable highly efficient and targeted weed control in agriculture.